ccd.c revision 1.143.10.4
1 /* $NetBSD: ccd.c,v 1.143.10.4 2017/12/03 11:36:58 jdolecek Exp $ */ 2 3 /*- 4 * Copyright (c) 1996, 1997, 1998, 1999, 2007, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe, and by Andrew Doran. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 1988 University of Utah. 34 * Copyright (c) 1990, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * This code is derived from software contributed to Berkeley by 38 * the Systems Programming Group of the University of Utah Computer 39 * Science Department. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * from: Utah $Hdr: cd.c 1.6 90/11/28$ 66 * 67 * @(#)cd.c 8.2 (Berkeley) 11/16/93 68 */ 69 70 /* 71 * "Concatenated" disk driver. 72 * 73 * Notes on concurrency: 74 * 75 * => sc_dvlock serializes access to the device nodes, excluding block I/O. 76 * 77 * => sc_iolock serializes access to (sc_flags & CCDF_INITED), disk stats, 78 * sc_stop, sc_bufq and b_resid from master buffers. 79 * 80 * => a combination of CCDF_INITED, sc_inflight, and sc_iolock is used to 81 * serialize I/O and configuration changes. 82 * 83 * => the in-core disk label does not change while the device is open. 84 * 85 * On memory consumption: ccd fans out I/O requests and so needs to 86 * allocate memory. If the system is desperately low on memory, we 87 * single thread I/O. 88 */ 89 90 #include <sys/cdefs.h> 91 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.143.10.4 2017/12/03 11:36:58 jdolecek Exp $"); 92 93 #if defined(_KERNEL_OPT) 94 #include "opt_compat_netbsd.h" 95 #endif 96 97 #include <sys/param.h> 98 #include <sys/systm.h> 99 #include <sys/kernel.h> 100 #include <sys/proc.h> 101 #include <sys/errno.h> 102 #include <sys/buf.h> 103 #include <sys/kmem.h> 104 #include <sys/pool.h> 105 #include <sys/module.h> 106 #include <sys/namei.h> 107 #include <sys/stat.h> 108 #include <sys/ioctl.h> 109 #include <sys/disklabel.h> 110 #include <sys/device.h> 111 #include <sys/disk.h> 112 #include <sys/syslog.h> 113 #include <sys/fcntl.h> 114 #include <sys/vnode.h> 115 #include <sys/conf.h> 116 #include <sys/mutex.h> 117 #include <sys/queue.h> 118 #include <sys/kauth.h> 119 #include <sys/kthread.h> 120 #include <sys/bufq.h> 121 #include <sys/sysctl.h> 122 123 #include <uvm/uvm_extern.h> 124 125 #include <dev/ccdvar.h> 126 #include <dev/dkvar.h> 127 128 #include <miscfs/specfs/specdev.h> /* for v_rdev */ 129 130 #include "ioconf.h" 131 132 #if defined(CCDDEBUG) && !defined(DEBUG) 133 #define DEBUG 134 #endif 135 136 #ifdef DEBUG 137 #define CCDB_FOLLOW 0x01 138 #define CCDB_INIT 0x02 139 #define CCDB_IO 0x04 140 #define CCDB_LABEL 0x08 141 #define CCDB_VNODE 0x10 142 int ccddebug = 0x00; 143 #endif 144 145 #define ccdunit(x) DISKUNIT(x) 146 147 struct ccdbuf { 148 struct buf cb_buf; /* new I/O buf */ 149 struct buf *cb_obp; /* ptr. to original I/O buf */ 150 struct ccd_softc *cb_sc; /* pointer to ccd softc */ 151 int cb_comp; /* target component */ 152 SIMPLEQ_ENTRY(ccdbuf) cb_q; /* fifo of component buffers */ 153 }; 154 155 /* component buffer pool */ 156 static pool_cache_t ccd_cache; 157 158 #define CCD_GETBUF() pool_cache_get(ccd_cache, PR_WAITOK) 159 #define CCD_PUTBUF(cbp) pool_cache_put(ccd_cache, cbp) 160 161 #define CCDLABELDEV(dev) \ 162 (MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART)) 163 164 /* called by main() at boot time */ 165 void ccddetach(void); 166 167 /* called by biodone() at interrupt time */ 168 static void ccdiodone(struct buf *); 169 170 static void ccdinterleave(struct ccd_softc *); 171 static int ccdinit(struct ccd_softc *, char **, struct vnode **, 172 struct lwp *); 173 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *, 174 daddr_t, void *, long); 175 static void ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *); 176 static void ccdgetdisklabel(dev_t); 177 static void ccdmakedisklabel(struct ccd_softc *); 178 static void ccdstart(struct ccd_softc *); 179 static void ccdthread(void *); 180 static struct ccd_softc *ccdget(int, int); 181 182 static dev_type_open(ccdopen); 183 static dev_type_close(ccdclose); 184 static dev_type_read(ccdread); 185 static dev_type_write(ccdwrite); 186 static dev_type_ioctl(ccdioctl); 187 static dev_type_strategy(ccdstrategy); 188 static dev_type_size(ccdsize); 189 190 const struct bdevsw ccd_bdevsw = { 191 .d_open = ccdopen, 192 .d_close = ccdclose, 193 .d_strategy = ccdstrategy, 194 .d_ioctl = ccdioctl, 195 .d_dump = nodump, 196 .d_psize = ccdsize, 197 .d_discard = nodiscard, 198 .d_flag = D_DISK | D_MPSAFE 199 }; 200 201 const struct cdevsw ccd_cdevsw = { 202 .d_open = ccdopen, 203 .d_close = ccdclose, 204 .d_read = ccdread, 205 .d_write = ccdwrite, 206 .d_ioctl = ccdioctl, 207 .d_stop = nostop, 208 .d_tty = notty, 209 .d_poll = nopoll, 210 .d_mmap = nommap, 211 .d_kqfilter = nokqfilter, 212 .d_discard = nodiscard, 213 .d_flag = D_DISK | D_MPSAFE 214 }; 215 216 #ifdef DEBUG 217 static void printiinfo(struct ccdiinfo *); 218 #endif 219 220 static LIST_HEAD(, ccd_softc) ccds = LIST_HEAD_INITIALIZER(ccds); 221 static kmutex_t ccd_lock; 222 static size_t ccd_nactive = 0; 223 224 #ifdef _MODULE 225 static struct sysctllog *ccd_clog; 226 #endif 227 228 SYSCTL_SETUP_PROTO(sysctl_kern_ccd_setup); 229 230 static void 231 ccdminphys(struct buf *bp) 232 { 233 struct ccd_softc *cs; 234 long xmax; 235 int unit = ccdunit(bp->b_dev); 236 237 cs = ccdget(unit, 0); 238 239 if (__predict_false(cs == NULL)) { 240 panic("minphys called on missing ccd unit %d", unit); 241 } 242 xmax = cs->sc_maxphys; 243 244 if (bp->b_bcount > xmax) 245 bp->b_bcount = xmax; 246 } 247 248 const struct dkdriver ccd_dkdriver = { 249 .d_strategy = ccdstrategy, 250 .d_minphys = ccdminphys, 251 }; 252 253 static struct ccd_softc * 254 ccdcreate(int unit) { 255 struct ccd_softc *sc = kmem_zalloc(sizeof(*sc), KM_SLEEP); 256 257 /* Initialize per-softc structures. */ 258 snprintf(sc->sc_xname, sizeof(sc->sc_xname), "ccd%d", unit); 259 sc->sc_unit = unit; 260 mutex_init(&sc->sc_dvlock, MUTEX_DEFAULT, IPL_NONE); 261 sc->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 262 cv_init(&sc->sc_stop, "ccdstop"); 263 cv_init(&sc->sc_push, "ccdthr"); 264 disk_init(&sc->sc_dkdev, sc->sc_xname, &ccd_dkdriver); /* XXX */ 265 return sc; 266 } 267 268 static void 269 ccddestroy(struct ccd_softc *sc) { 270 mutex_obj_free(sc->sc_iolock); 271 mutex_exit(&sc->sc_dvlock); 272 mutex_destroy(&sc->sc_dvlock); 273 cv_destroy(&sc->sc_stop); 274 cv_destroy(&sc->sc_push); 275 disk_destroy(&sc->sc_dkdev); 276 kmem_free(sc, sizeof(*sc)); 277 } 278 279 static struct ccd_softc * 280 ccdget(int unit, int make) { 281 struct ccd_softc *sc; 282 if (unit < 0) { 283 #ifdef DIAGNOSTIC 284 panic("%s: unit %d!", __func__, unit); 285 #endif 286 return NULL; 287 } 288 mutex_enter(&ccd_lock); 289 LIST_FOREACH(sc, &ccds, sc_link) { 290 if (sc->sc_unit == unit) { 291 mutex_exit(&ccd_lock); 292 return sc; 293 } 294 } 295 mutex_exit(&ccd_lock); 296 if (!make) 297 return NULL; 298 if ((sc = ccdcreate(unit)) == NULL) 299 return NULL; 300 mutex_enter(&ccd_lock); 301 LIST_INSERT_HEAD(&ccds, sc, sc_link); 302 ccd_nactive++; 303 mutex_exit(&ccd_lock); 304 return sc; 305 } 306 307 static void 308 ccdput(struct ccd_softc *sc) { 309 mutex_enter(&ccd_lock); 310 LIST_REMOVE(sc, sc_link); 311 ccd_nactive--; 312 mutex_exit(&ccd_lock); 313 ccddestroy(sc); 314 } 315 316 /* 317 * Called by main() during pseudo-device attachment. All we need 318 * to do is allocate enough space for devices to be configured later. 319 */ 320 void 321 ccdattach(int num) 322 { 323 mutex_init(&ccd_lock, MUTEX_DEFAULT, IPL_NONE); 324 325 /* Initialize the component buffer pool. */ 326 ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0, 327 0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL); 328 } 329 330 void 331 ccddetach(void) 332 { 333 pool_cache_destroy(ccd_cache); 334 mutex_destroy(&ccd_lock); 335 } 336 337 static int 338 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp, 339 struct lwp *l) 340 { 341 struct ccdcinfo *ci = NULL; 342 int ix; 343 struct ccdgeom *ccg = &cs->sc_geom; 344 char *tmppath; 345 int error, path_alloced; 346 uint64_t psize, minsize; 347 unsigned secsize, maxsecsize; 348 struct disk_geom *dg; 349 350 #ifdef DEBUG 351 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 352 printf("%s: ccdinit\n", cs->sc_xname); 353 #endif 354 355 /* Allocate space for the component info. */ 356 cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo), 357 KM_SLEEP); 358 tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 359 360 cs->sc_size = 0; 361 cs->sc_maxphys = MACHINE_MAXPHYS; 362 363 /* 364 * Verify that each component piece exists and record 365 * relevant information about it. 366 */ 367 maxsecsize = 0; 368 minsize = 0; 369 for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) { 370 ci = &cs->sc_cinfo[ix]; 371 ci->ci_vp = vpp[ix]; 372 struct disk *diskp; 373 374 /* 375 * Copy in the pathname of the component. 376 */ 377 memset(tmppath, 0, MAXPATHLEN); /* sanity */ 378 error = copyinstr(cpaths[ix], tmppath, 379 MAXPATHLEN, &ci->ci_pathlen); 380 if (ci->ci_pathlen == 0) 381 error = EINVAL; 382 if (error) { 383 #ifdef DEBUG 384 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 385 printf("%s: can't copy path, error = %d\n", 386 cs->sc_xname, error); 387 #endif 388 goto out; 389 } 390 ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP); 391 memcpy(ci->ci_path, tmppath, ci->ci_pathlen); 392 path_alloced++; 393 394 /* 395 * XXX: Cache the component's dev_t. 396 */ 397 ci->ci_dev = vpp[ix]->v_rdev; 398 if ((diskp = disk_find_blk(ci->ci_dev)) == NULL) { 399 panic("no disk for device %d %d", major(ci->ci_dev), 400 DISKUNIT(ci->ci_dev)); 401 } 402 cs->sc_maxphys = MIN(cs->sc_maxphys, disk_maxphys(diskp)); 403 404 /* 405 * Get partition information for the component. 406 */ 407 error = getdisksize(vpp[ix], &psize, &secsize); 408 if (error) { 409 #ifdef DEBUG 410 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 411 printf("%s: %s: disksize failed, error = %d\n", 412 cs->sc_xname, ci->ci_path, error); 413 #endif 414 goto out; 415 } 416 417 /* 418 * Calculate the size, truncating to an interleave 419 * boundary if necessary. 420 */ 421 maxsecsize = secsize > maxsecsize ? secsize : maxsecsize; 422 if (cs->sc_ileave > 1) 423 psize -= psize % cs->sc_ileave; 424 425 if (psize == 0) { 426 #ifdef DEBUG 427 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 428 printf("%s: %s: size == 0\n", 429 cs->sc_xname, ci->ci_path); 430 #endif 431 error = ENODEV; 432 goto out; 433 } 434 435 if (minsize == 0 || psize < minsize) 436 minsize = psize; 437 ci->ci_size = psize; 438 cs->sc_size += psize; 439 } 440 441 /* 442 * Don't allow the interleave to be smaller than 443 * the biggest component sector. 444 */ 445 if ((cs->sc_ileave > 0) && 446 (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) { 447 #ifdef DEBUG 448 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 449 printf("%s: interleave must be at least %d\n", 450 cs->sc_xname, (maxsecsize / DEV_BSIZE)); 451 #endif 452 error = EINVAL; 453 goto out; 454 } 455 456 /* 457 * If uniform interleave is desired set all sizes to that of 458 * the smallest component. 459 */ 460 if (cs->sc_flags & CCDF_UNIFORM) { 461 for (ci = cs->sc_cinfo; 462 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 463 ci->ci_size = minsize; 464 465 cs->sc_size = cs->sc_nccdisks * minsize; 466 } 467 468 /* 469 * Construct the interleave table. 470 */ 471 ccdinterleave(cs); 472 473 /* 474 * Create pseudo-geometry based on 1MB cylinders. It's 475 * pretty close. 476 */ 477 ccg->ccg_secsize = DEV_BSIZE; 478 ccg->ccg_ntracks = 1; 479 ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize); 480 ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors; 481 482 dg = &cs->sc_dkdev.dk_geom; 483 memset(dg, 0, sizeof(*dg)); 484 dg->dg_secperunit = cs->sc_size; 485 dg->dg_secsize = ccg->ccg_secsize; 486 dg->dg_nsectors = ccg->ccg_nsectors; 487 dg->dg_ntracks = ccg->ccg_ntracks; 488 dg->dg_ncylinders = ccg->ccg_ncylinders; 489 490 if (cs->sc_ileave > 0) 491 aprint_normal("%s: Interleaving %d component%s " 492 "(%d block interleave)\n", cs->sc_xname, 493 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""), 494 cs->sc_ileave); 495 else 496 aprint_normal("%s: Concatenating %d component%s\n", 497 cs->sc_xname, 498 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : "")); 499 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 500 ci = &cs->sc_cinfo[ix]; 501 aprint_normal("%s: %s (%ju blocks)\n", cs->sc_xname, 502 ci->ci_path, (uintmax_t)ci->ci_size); 503 } 504 aprint_normal("%s: total %ju blocks\n", cs->sc_xname, cs->sc_size); 505 506 /* 507 * Create thread to handle deferred I/O. 508 */ 509 cs->sc_zap = false; 510 error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread, 511 cs, &cs->sc_thread, "%s", cs->sc_xname); 512 if (error) { 513 printf("ccdinit: can't create thread: %d\n", error); 514 goto out; 515 } 516 517 /* 518 * Only now that everything is set up can we enable the device. 519 */ 520 mutex_enter(cs->sc_iolock); 521 cs->sc_flags |= CCDF_INITED; 522 mutex_exit(cs->sc_iolock); 523 kmem_free(tmppath, MAXPATHLEN); 524 return (0); 525 526 out: 527 for (ix = 0; ix < path_alloced; ix++) { 528 kmem_free(cs->sc_cinfo[ix].ci_path, 529 cs->sc_cinfo[ix].ci_pathlen); 530 } 531 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo)); 532 kmem_free(tmppath, MAXPATHLEN); 533 return (error); 534 } 535 536 static void 537 ccdinterleave(struct ccd_softc *cs) 538 { 539 struct ccdcinfo *ci, *smallci; 540 struct ccdiinfo *ii; 541 daddr_t bn, lbn; 542 int ix; 543 u_long size; 544 545 #ifdef DEBUG 546 if (ccddebug & CCDB_INIT) 547 printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave); 548 #endif 549 /* 550 * Allocate an interleave table. 551 * Chances are this is too big, but we don't care. 552 */ 553 size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo); 554 cs->sc_itable = kmem_zalloc(size, KM_SLEEP); 555 556 /* 557 * Trivial case: no interleave (actually interleave of disk size). 558 * Each table entry represents a single component in its entirety. 559 */ 560 if (cs->sc_ileave == 0) { 561 bn = 0; 562 ii = cs->sc_itable; 563 564 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 565 /* Allocate space for ii_index. */ 566 ii->ii_indexsz = sizeof(int); 567 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 568 ii->ii_ndisk = 1; 569 ii->ii_startblk = bn; 570 ii->ii_startoff = 0; 571 ii->ii_index[0] = ix; 572 bn += cs->sc_cinfo[ix].ci_size; 573 ii++; 574 } 575 ii->ii_ndisk = 0; 576 #ifdef DEBUG 577 if (ccddebug & CCDB_INIT) 578 printiinfo(cs->sc_itable); 579 #endif 580 return; 581 } 582 583 /* 584 * The following isn't fast or pretty; it doesn't have to be. 585 */ 586 size = 0; 587 bn = lbn = 0; 588 for (ii = cs->sc_itable; ; ii++) { 589 /* Allocate space for ii_index. */ 590 ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks; 591 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 592 593 /* 594 * Locate the smallest of the remaining components 595 */ 596 smallci = NULL; 597 for (ci = cs->sc_cinfo; 598 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 599 if (ci->ci_size > size && 600 (smallci == NULL || 601 ci->ci_size < smallci->ci_size)) 602 smallci = ci; 603 604 /* 605 * Nobody left, all done 606 */ 607 if (smallci == NULL) { 608 ii->ii_ndisk = 0; 609 break; 610 } 611 612 /* 613 * Record starting logical block and component offset 614 */ 615 ii->ii_startblk = bn / cs->sc_ileave; 616 ii->ii_startoff = lbn; 617 618 /* 619 * Determine how many disks take part in this interleave 620 * and record their indices. 621 */ 622 ix = 0; 623 for (ci = cs->sc_cinfo; 624 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 625 if (ci->ci_size >= smallci->ci_size) 626 ii->ii_index[ix++] = ci - cs->sc_cinfo; 627 ii->ii_ndisk = ix; 628 bn += ix * (smallci->ci_size - size); 629 lbn = smallci->ci_size / cs->sc_ileave; 630 size = smallci->ci_size; 631 } 632 #ifdef DEBUG 633 if (ccddebug & CCDB_INIT) 634 printiinfo(cs->sc_itable); 635 #endif 636 } 637 638 /* ARGSUSED */ 639 static int 640 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l) 641 { 642 int unit = ccdunit(dev); 643 struct ccd_softc *cs; 644 struct disklabel *lp; 645 int error = 0, part, pmask; 646 647 #ifdef DEBUG 648 if (ccddebug & CCDB_FOLLOW) 649 printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags); 650 #endif 651 if ((cs = ccdget(unit, 1)) == NULL) 652 return ENXIO; 653 654 mutex_enter(&cs->sc_dvlock); 655 656 lp = cs->sc_dkdev.dk_label; 657 658 part = DISKPART(dev); 659 pmask = (1 << part); 660 661 /* 662 * If we're initialized, check to see if there are any other 663 * open partitions. If not, then it's safe to update 664 * the in-core disklabel. Only read the disklabel if it is 665 * not already valid. 666 */ 667 if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED && 668 cs->sc_dkdev.dk_openmask == 0) 669 ccdgetdisklabel(dev); 670 671 /* Check that the partition exists. */ 672 if (part != RAW_PART) { 673 if (((cs->sc_flags & CCDF_INITED) == 0) || 674 ((part >= lp->d_npartitions) || 675 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 676 error = ENXIO; 677 goto done; 678 } 679 } 680 681 /* Prevent our unit from being unconfigured while open. */ 682 switch (fmt) { 683 case S_IFCHR: 684 cs->sc_dkdev.dk_copenmask |= pmask; 685 break; 686 687 case S_IFBLK: 688 cs->sc_dkdev.dk_bopenmask |= pmask; 689 break; 690 } 691 cs->sc_dkdev.dk_openmask = 692 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 693 694 done: 695 mutex_exit(&cs->sc_dvlock); 696 return (error); 697 } 698 699 /* ARGSUSED */ 700 static int 701 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l) 702 { 703 int unit = ccdunit(dev); 704 struct ccd_softc *cs; 705 int part; 706 707 #ifdef DEBUG 708 if (ccddebug & CCDB_FOLLOW) 709 printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags); 710 #endif 711 712 if ((cs = ccdget(unit, 0)) == NULL) 713 return ENXIO; 714 715 mutex_enter(&cs->sc_dvlock); 716 717 part = DISKPART(dev); 718 719 /* ...that much closer to allowing unconfiguration... */ 720 switch (fmt) { 721 case S_IFCHR: 722 cs->sc_dkdev.dk_copenmask &= ~(1 << part); 723 break; 724 725 case S_IFBLK: 726 cs->sc_dkdev.dk_bopenmask &= ~(1 << part); 727 break; 728 } 729 cs->sc_dkdev.dk_openmask = 730 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 731 732 if (cs->sc_dkdev.dk_openmask == 0) { 733 if ((cs->sc_flags & CCDF_KLABEL) == 0) 734 cs->sc_flags &= ~CCDF_VLABEL; 735 } 736 737 mutex_exit(&cs->sc_dvlock); 738 return (0); 739 } 740 741 static bool 742 ccdbackoff(struct ccd_softc *cs) 743 { 744 745 /* XXX Arbitrary, should be a uvm call. */ 746 return uvmexp.free < (uvmexp.freemin >> 1) && 747 disk_isbusy(&cs->sc_dkdev); 748 } 749 750 static void 751 ccdthread(void *cookie) 752 { 753 struct ccd_softc *cs; 754 755 cs = cookie; 756 757 #ifdef DEBUG 758 if (ccddebug & CCDB_FOLLOW) 759 printf("ccdthread: hello\n"); 760 #endif 761 762 mutex_enter(cs->sc_iolock); 763 while (__predict_true(!cs->sc_zap)) { 764 if (bufq_peek(cs->sc_bufq) == NULL) { 765 /* Nothing to do. */ 766 cv_wait(&cs->sc_push, cs->sc_iolock); 767 continue; 768 } 769 if (ccdbackoff(cs)) { 770 /* Wait for memory to become available. */ 771 (void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1); 772 continue; 773 } 774 #ifdef DEBUG 775 if (ccddebug & CCDB_FOLLOW) 776 printf("ccdthread: dispatching I/O\n"); 777 #endif 778 ccdstart(cs); 779 mutex_enter(cs->sc_iolock); 780 } 781 cs->sc_thread = NULL; 782 mutex_exit(cs->sc_iolock); 783 #ifdef DEBUG 784 if (ccddebug & CCDB_FOLLOW) 785 printf("ccdthread: goodbye\n"); 786 #endif 787 kthread_exit(0); 788 } 789 790 static void 791 ccdstrategy(struct buf *bp) 792 { 793 int unit = ccdunit(bp->b_dev); 794 struct ccd_softc *cs; 795 if ((cs = ccdget(unit, 0)) == NULL) 796 return; 797 798 /* Must be open or reading label. */ 799 KASSERT(cs->sc_dkdev.dk_openmask != 0 || 800 (cs->sc_flags & CCDF_RLABEL) != 0); 801 802 mutex_enter(cs->sc_iolock); 803 /* Synchronize with device init/uninit. */ 804 if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) { 805 mutex_exit(cs->sc_iolock); 806 #ifdef DEBUG 807 if (ccddebug & CCDB_FOLLOW) 808 printf("ccdstrategy: unit %d: not inited\n", unit); 809 #endif 810 bp->b_error = ENXIO; 811 bp->b_resid = bp->b_bcount; 812 biodone(bp); 813 return; 814 } 815 816 /* Defer to thread if system is low on memory. */ 817 bufq_put(cs->sc_bufq, bp); 818 if (__predict_false(ccdbackoff(cs))) { 819 mutex_exit(cs->sc_iolock); 820 #ifdef DEBUG 821 if (ccddebug & CCDB_FOLLOW) 822 printf("ccdstrategy: holding off on I/O\n"); 823 #endif 824 return; 825 } 826 ccdstart(cs); 827 } 828 829 static void 830 ccdstart(struct ccd_softc *cs) 831 { 832 daddr_t blkno; 833 int wlabel; 834 struct disklabel *lp; 835 long bcount, rcount; 836 struct ccdbuf *cbp; 837 char *addr; 838 daddr_t bn; 839 vnode_t *vp; 840 buf_t *bp; 841 842 KASSERT(mutex_owned(cs->sc_iolock)); 843 844 bp = bufq_get(cs->sc_bufq); 845 KASSERT(bp != NULL); 846 847 disk_busy(&cs->sc_dkdev); 848 849 #ifdef DEBUG 850 if (ccddebug & CCDB_FOLLOW) 851 printf("ccdstart(%s, %p)\n", cs->sc_xname, bp); 852 #endif 853 854 /* If it's a nil transfer, wake up the top half now. */ 855 if (bp->b_bcount == 0) 856 goto done; 857 858 lp = cs->sc_dkdev.dk_label; 859 860 /* 861 * Do bounds checking and adjust transfer. If there's an 862 * error, the bounds check will flag that for us. Convert 863 * the partition relative block number to an absolute. 864 */ 865 blkno = bp->b_blkno; 866 wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING); 867 if (DISKPART(bp->b_dev) != RAW_PART) { 868 if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0) 869 goto done; 870 blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset; 871 } 872 mutex_exit(cs->sc_iolock); 873 bp->b_rawblkno = blkno; 874 875 /* Allocate the component buffers and start I/O! */ 876 bp->b_resid = bp->b_bcount; 877 bn = bp->b_rawblkno; 878 addr = bp->b_data; 879 for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) { 880 cbp = ccdbuffer(cs, bp, bn, addr, bcount); 881 rcount = cbp->cb_buf.b_bcount; 882 bn += btodb(rcount); 883 addr += rcount; 884 vp = cbp->cb_buf.b_vp; 885 if ((cbp->cb_buf.b_flags & B_READ) == 0) { 886 mutex_enter(vp->v_interlock); 887 vp->v_numoutput++; 888 mutex_exit(vp->v_interlock); 889 } 890 (void)VOP_STRATEGY(vp, &cbp->cb_buf); 891 } 892 return; 893 894 done: 895 disk_unbusy(&cs->sc_dkdev, 0, 0); 896 cv_broadcast(&cs->sc_stop); 897 cv_broadcast(&cs->sc_push); 898 mutex_exit(cs->sc_iolock); 899 bp->b_resid = bp->b_bcount; 900 biodone(bp); 901 } 902 903 /* 904 * Build a component buffer header. 905 */ 906 static struct ccdbuf * 907 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr, 908 long bcount) 909 { 910 struct ccdcinfo *ci; 911 struct ccdbuf *cbp; 912 daddr_t cbn, cboff; 913 u_int64_t cbc; 914 int ccdisk; 915 916 #ifdef DEBUG 917 if (ccddebug & CCDB_IO) 918 printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n", 919 cs, bp, bn, addr, bcount); 920 #endif 921 /* 922 * Determine which component bn falls in. 923 */ 924 cbn = bn; 925 cboff = 0; 926 927 /* 928 * Serially concatenated 929 */ 930 if (cs->sc_ileave == 0) { 931 daddr_t sblk; 932 933 sblk = 0; 934 for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk]; 935 cbn >= sblk + ci->ci_size; 936 ccdisk++, ci = &cs->sc_cinfo[ccdisk]) 937 sblk += ci->ci_size; 938 cbn -= sblk; 939 } 940 /* 941 * Interleaved 942 */ 943 else { 944 struct ccdiinfo *ii; 945 int off; 946 947 cboff = cbn % cs->sc_ileave; 948 cbn /= cs->sc_ileave; 949 for (ii = cs->sc_itable; ii->ii_ndisk; ii++) 950 if (ii->ii_startblk > cbn) 951 break; 952 ii--; 953 off = cbn - ii->ii_startblk; 954 if (ii->ii_ndisk == 1) { 955 ccdisk = ii->ii_index[0]; 956 cbn = ii->ii_startoff + off; 957 } else { 958 ccdisk = ii->ii_index[off % ii->ii_ndisk]; 959 cbn = ii->ii_startoff + off / ii->ii_ndisk; 960 } 961 cbn *= cs->sc_ileave; 962 ci = &cs->sc_cinfo[ccdisk]; 963 } 964 965 /* 966 * Fill in the component buf structure. 967 */ 968 cbp = CCD_GETBUF(); 969 KASSERT(cbp != NULL); 970 buf_init(&cbp->cb_buf); 971 cbp->cb_buf.b_flags = bp->b_flags; 972 cbp->cb_buf.b_oflags = bp->b_oflags; 973 cbp->cb_buf.b_cflags = bp->b_cflags; 974 cbp->cb_buf.b_iodone = ccdiodone; 975 cbp->cb_buf.b_proc = bp->b_proc; 976 cbp->cb_buf.b_dev = ci->ci_dev; 977 cbp->cb_buf.b_blkno = cbn + cboff; 978 cbp->cb_buf.b_data = addr; 979 cbp->cb_buf.b_vp = ci->ci_vp; 980 cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock; 981 if (cs->sc_ileave == 0) 982 cbc = dbtob((u_int64_t)(ci->ci_size - cbn)); 983 else 984 cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff)); 985 cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount; 986 987 /* 988 * context for ccdiodone 989 */ 990 cbp->cb_obp = bp; 991 cbp->cb_sc = cs; 992 cbp->cb_comp = ccdisk; 993 994 BIO_COPYPRIO(&cbp->cb_buf, bp); 995 996 #ifdef DEBUG 997 if (ccddebug & CCDB_IO) 998 printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p" 999 " bcnt %d\n", 1000 ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp, 1001 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 1002 cbp->cb_buf.b_bcount); 1003 #endif 1004 1005 return (cbp); 1006 } 1007 1008 /* 1009 * Called at interrupt time. 1010 * Mark the component as done and if all components are done, 1011 * take a ccd interrupt. 1012 */ 1013 static void 1014 ccdiodone(struct buf *vbp) 1015 { 1016 struct ccdbuf *cbp = (struct ccdbuf *) vbp; 1017 struct buf *bp = cbp->cb_obp; 1018 struct ccd_softc *cs = cbp->cb_sc; 1019 int count; 1020 1021 #ifdef DEBUG 1022 if (ccddebug & CCDB_FOLLOW) 1023 printf("ccdiodone(%p)\n", cbp); 1024 if (ccddebug & CCDB_IO) { 1025 printf("ccdiodone: bp %p bcount %d resid %d\n", 1026 bp, bp->b_bcount, bp->b_resid); 1027 printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p" 1028 " bcnt %d\n", 1029 cbp->cb_buf.b_dev, cbp->cb_comp, cbp, 1030 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 1031 cbp->cb_buf.b_bcount); 1032 } 1033 #endif 1034 1035 if (cbp->cb_buf.b_error != 0) { 1036 bp->b_error = cbp->cb_buf.b_error; 1037 printf("%s: error %d on component %d\n", 1038 cs->sc_xname, bp->b_error, cbp->cb_comp); 1039 } 1040 count = cbp->cb_buf.b_bcount; 1041 buf_destroy(&cbp->cb_buf); 1042 CCD_PUTBUF(cbp); 1043 1044 /* 1045 * If all done, "interrupt". 1046 */ 1047 mutex_enter(cs->sc_iolock); 1048 bp->b_resid -= count; 1049 if (bp->b_resid < 0) 1050 panic("ccdiodone: count"); 1051 if (bp->b_resid == 0) { 1052 /* 1053 * Request is done for better or worse, wakeup the top half. 1054 */ 1055 if (bp->b_error != 0) 1056 bp->b_resid = bp->b_bcount; 1057 disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid), 1058 (bp->b_flags & B_READ)); 1059 if (!disk_isbusy(&cs->sc_dkdev)) { 1060 if (bufq_peek(cs->sc_bufq) != NULL) { 1061 cv_broadcast(&cs->sc_push); 1062 } 1063 cv_broadcast(&cs->sc_stop); 1064 } 1065 mutex_exit(cs->sc_iolock); 1066 biodone(bp); 1067 } else 1068 mutex_exit(cs->sc_iolock); 1069 } 1070 1071 /* ARGSUSED */ 1072 static int 1073 ccdread(dev_t dev, struct uio *uio, int flags) 1074 { 1075 int unit = ccdunit(dev); 1076 struct ccd_softc *cs; 1077 1078 #ifdef DEBUG 1079 if (ccddebug & CCDB_FOLLOW) 1080 printf("ccdread(0x%"PRIx64", %p)\n", dev, uio); 1081 #endif 1082 if ((cs = ccdget(unit, 0)) == NULL) 1083 return 0; 1084 1085 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1086 if ((cs->sc_flags & CCDF_INITED) == 0) 1087 return (ENXIO); 1088 1089 return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio)); 1090 } 1091 1092 /* ARGSUSED */ 1093 static int 1094 ccdwrite(dev_t dev, struct uio *uio, int flags) 1095 { 1096 int unit = ccdunit(dev); 1097 struct ccd_softc *cs; 1098 1099 #ifdef DEBUG 1100 if (ccddebug & CCDB_FOLLOW) 1101 printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio); 1102 #endif 1103 if ((cs = ccdget(unit, 0)) == NULL) 1104 return ENOENT; 1105 1106 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1107 if ((cs->sc_flags & CCDF_INITED) == 0) 1108 return (ENXIO); 1109 1110 return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio)); 1111 } 1112 1113 static int 1114 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1115 { 1116 int unit = ccdunit(dev); 1117 int i, j, lookedup = 0, error = 0; 1118 int part, pmask, make; 1119 struct ccd_softc *cs; 1120 struct ccd_ioctl *ccio = (struct ccd_ioctl *)data; 1121 kauth_cred_t uc; 1122 char **cpp; 1123 struct pathbuf *pb; 1124 struct vnode **vpp; 1125 #ifdef __HAVE_OLD_DISKLABEL 1126 struct disklabel newlabel; 1127 #endif 1128 1129 switch (cmd) { 1130 #if defined(COMPAT_60) && !defined(_LP64) 1131 case CCDIOCSET_60: 1132 #endif 1133 case CCDIOCSET: 1134 make = 1; 1135 break; 1136 default: 1137 make = 0; 1138 break; 1139 } 1140 1141 if ((cs = ccdget(unit, make)) == NULL) 1142 return ENOENT; 1143 uc = kauth_cred_get(); 1144 1145 /* 1146 * Compat code must not be called if on a platform where 1147 * sizeof (size_t) == sizeof (uint64_t) as CCDIOCSET will 1148 * be the same as CCDIOCSET_60 1149 */ 1150 #if defined(COMPAT_60) && !defined(_LP64) 1151 switch (cmd) { 1152 case CCDIOCSET_60: { 1153 struct ccd_ioctl ccionew; 1154 struct ccd_ioctl_60 *ccio60 = 1155 (struct ccd_ioctl_60 *)data; 1156 ccionew.ccio_disks = ccio->ccio_disks; 1157 ccionew.ccio_ndisks = ccio->ccio_ndisks; 1158 ccionew.ccio_ileave = ccio->ccio_ileave; 1159 ccionew.ccio_flags = ccio->ccio_flags; 1160 ccionew.ccio_unit = ccio->ccio_unit; 1161 error = ccdioctl(dev, CCDIOCSET, &ccionew, flag, l); 1162 if (!error) { 1163 /* Copy data back, adjust types if necessary */ 1164 ccio60->ccio_disks = ccionew.ccio_disks; 1165 ccio60->ccio_ndisks = ccionew.ccio_ndisks; 1166 ccio60->ccio_ileave = ccionew.ccio_ileave; 1167 ccio60->ccio_flags = ccionew.ccio_flags; 1168 ccio60->ccio_unit = ccionew.ccio_unit; 1169 ccio60->ccio_size = (size_t)ccionew.ccio_size; 1170 } 1171 return error; 1172 } 1173 break; 1174 1175 case CCDIOCCLR_60: 1176 /* 1177 * ccio_size member not used, so existing struct OK 1178 * drop through to existing non-compat version 1179 */ 1180 cmd = CCDIOCCLR; 1181 break; 1182 } 1183 #endif /* COMPAT_60 && !_LP64*/ 1184 1185 /* Must be open for writes for these commands... */ 1186 switch (cmd) { 1187 case CCDIOCSET: 1188 case CCDIOCCLR: 1189 case DIOCSDINFO: 1190 case DIOCWDINFO: 1191 case DIOCCACHESYNC: 1192 case DIOCAWEDGE: 1193 case DIOCDWEDGE: 1194 case DIOCMWEDGES: 1195 #ifdef __HAVE_OLD_DISKLABEL 1196 case ODIOCSDINFO: 1197 case ODIOCWDINFO: 1198 #endif 1199 case DIOCKLABEL: 1200 case DIOCWLABEL: 1201 if ((flag & FWRITE) == 0) 1202 return (EBADF); 1203 } 1204 1205 mutex_enter(&cs->sc_dvlock); 1206 1207 /* Must be initialized for these... */ 1208 switch (cmd) { 1209 case CCDIOCCLR: 1210 case DIOCGDINFO: 1211 case DIOCGSTRATEGY: 1212 case DIOCGCACHE: 1213 case DIOCCACHESYNC: 1214 case DIOCAWEDGE: 1215 case DIOCDWEDGE: 1216 case DIOCLWEDGES: 1217 case DIOCMWEDGES: 1218 case DIOCSDINFO: 1219 case DIOCWDINFO: 1220 case DIOCGPARTINFO: 1221 case DIOCWLABEL: 1222 case DIOCKLABEL: 1223 case DIOCGDEFLABEL: 1224 #ifdef __HAVE_OLD_DISKLABEL 1225 case ODIOCGDINFO: 1226 case ODIOCSDINFO: 1227 case ODIOCWDINFO: 1228 case ODIOCGDEFLABEL: 1229 #endif 1230 if ((cs->sc_flags & CCDF_INITED) == 0) { 1231 error = ENXIO; 1232 goto out; 1233 } 1234 } 1235 1236 error = disk_ioctl(&cs->sc_dkdev, dev, cmd, data, flag, l); 1237 if (error != EPASSTHROUGH) 1238 goto out; 1239 1240 error = 0; 1241 switch (cmd) { 1242 case CCDIOCSET: 1243 if (cs->sc_flags & CCDF_INITED) { 1244 error = EBUSY; 1245 goto out; 1246 } 1247 1248 /* Validate the flags. */ 1249 if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) { 1250 error = EINVAL; 1251 goto out; 1252 } 1253 1254 if (ccio->ccio_ndisks > CCD_MAXNDISKS || 1255 ccio->ccio_ndisks == 0) { 1256 error = EINVAL; 1257 goto out; 1258 } 1259 1260 /* Fill in some important bits. */ 1261 cs->sc_ileave = ccio->ccio_ileave; 1262 cs->sc_nccdisks = ccio->ccio_ndisks; 1263 cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK; 1264 1265 /* 1266 * Allocate space for and copy in the array of 1267 * component pathnames and device numbers. 1268 */ 1269 cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP); 1270 vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP); 1271 error = copyin(ccio->ccio_disks, cpp, 1272 ccio->ccio_ndisks * sizeof(*cpp)); 1273 if (error) { 1274 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1275 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1276 goto out; 1277 } 1278 1279 #ifdef DEBUG 1280 if (ccddebug & CCDB_INIT) 1281 for (i = 0; i < ccio->ccio_ndisks; ++i) 1282 printf("ccdioctl: component %d: %p\n", 1283 i, cpp[i]); 1284 #endif 1285 1286 for (i = 0; i < ccio->ccio_ndisks; ++i) { 1287 #ifdef DEBUG 1288 if (ccddebug & CCDB_INIT) 1289 printf("ccdioctl: lookedup = %d\n", lookedup); 1290 #endif 1291 error = pathbuf_copyin(cpp[i], &pb); 1292 if (error == 0) { 1293 error = dk_lookup(pb, l, &vpp[i]); 1294 } 1295 pathbuf_destroy(pb); 1296 if (error != 0) { 1297 for (j = 0; j < lookedup; ++j) 1298 (void)vn_close(vpp[j], FREAD|FWRITE, 1299 uc); 1300 kmem_free(vpp, ccio->ccio_ndisks * 1301 sizeof(*vpp)); 1302 kmem_free(cpp, ccio->ccio_ndisks * 1303 sizeof(*cpp)); 1304 goto out; 1305 } 1306 ++lookedup; 1307 } 1308 1309 /* Attach the disk. */ 1310 disk_attach(&cs->sc_dkdev); 1311 bufq_alloc(&cs->sc_bufq, "fcfs", 0); 1312 1313 /* 1314 * Initialize the ccd. Fills in the softc for us. 1315 */ 1316 if ((error = ccdinit(cs, cpp, vpp, l)) != 0) { 1317 for (j = 0; j < lookedup; ++j) 1318 (void)vn_close(vpp[j], FREAD|FWRITE, 1319 uc); 1320 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1321 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1322 disk_detach(&cs->sc_dkdev); 1323 mutex_exit(&cs->sc_dvlock); 1324 bufq_free(cs->sc_bufq); 1325 return error; 1326 } 1327 1328 /* We can free the temporary variables now. */ 1329 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1330 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1331 1332 /* 1333 * The ccd has been successfully initialized, so 1334 * we can place it into the array. Don't try to 1335 * read the disklabel until the disk has been attached, 1336 * because space for the disklabel is allocated 1337 * in disk_attach(); 1338 */ 1339 ccio->ccio_unit = unit; 1340 ccio->ccio_size = cs->sc_size; 1341 1342 /* Try and read the disklabel. */ 1343 ccdgetdisklabel(dev); 1344 disk_set_info(NULL, &cs->sc_dkdev, NULL); 1345 1346 /* discover wedges */ 1347 mutex_exit(&cs->sc_dvlock); 1348 dkwedge_discover(&cs->sc_dkdev); 1349 return 0; 1350 1351 case CCDIOCCLR: 1352 /* 1353 * Don't unconfigure if any other partitions are open 1354 * or if both the character and block flavors of this 1355 * partition are open. 1356 */ 1357 part = DISKPART(dev); 1358 pmask = (1 << part); 1359 if ((cs->sc_dkdev.dk_openmask & ~pmask) || 1360 ((cs->sc_dkdev.dk_bopenmask & pmask) && 1361 (cs->sc_dkdev.dk_copenmask & pmask))) { 1362 error = EBUSY; 1363 goto out; 1364 } 1365 1366 /* Delete all of our wedges. */ 1367 dkwedge_delall(&cs->sc_dkdev); 1368 1369 /* Stop new I/O, wait for in-flight I/O to complete. */ 1370 mutex_enter(cs->sc_iolock); 1371 cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL); 1372 cs->sc_zap = true; 1373 while (disk_isbusy(&cs->sc_dkdev) || 1374 bufq_peek(cs->sc_bufq) != NULL || 1375 cs->sc_thread != NULL) { 1376 cv_broadcast(&cs->sc_push); 1377 (void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz); 1378 } 1379 mutex_exit(cs->sc_iolock); 1380 1381 /* 1382 * Free ccd_softc information and clear entry. 1383 */ 1384 1385 /* Close the components and free their pathnames. */ 1386 for (i = 0; i < cs->sc_nccdisks; ++i) { 1387 /* 1388 * XXX: this close could potentially fail and 1389 * cause Bad Things. Maybe we need to force 1390 * the close to happen? 1391 */ 1392 #ifdef DEBUG 1393 if (ccddebug & CCDB_VNODE) 1394 vprint("CCDIOCCLR: vnode info", 1395 cs->sc_cinfo[i].ci_vp); 1396 #endif 1397 (void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE, 1398 uc); 1399 kmem_free(cs->sc_cinfo[i].ci_path, 1400 cs->sc_cinfo[i].ci_pathlen); 1401 } 1402 1403 /* Free interleave index. */ 1404 for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) { 1405 kmem_free(cs->sc_itable[i].ii_index, 1406 cs->sc_itable[i].ii_indexsz); 1407 } 1408 1409 /* Free component info and interleave table. */ 1410 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * 1411 sizeof(struct ccdcinfo)); 1412 kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) * 1413 sizeof(struct ccdiinfo)); 1414 1415 aprint_normal("%s: detached\n", cs->sc_xname); 1416 1417 /* Detach the disk. */ 1418 disk_detach(&cs->sc_dkdev); 1419 bufq_free(cs->sc_bufq); 1420 ccdput(cs); 1421 /* Don't break, otherwise cs is read again. */ 1422 return 0; 1423 1424 case DIOCGSTRATEGY: 1425 { 1426 struct disk_strategy *dks = (void *)data; 1427 1428 mutex_enter(cs->sc_iolock); 1429 if (cs->sc_bufq != NULL) 1430 strlcpy(dks->dks_name, 1431 bufq_getstrategyname(cs->sc_bufq), 1432 sizeof(dks->dks_name)); 1433 else 1434 error = EINVAL; 1435 mutex_exit(cs->sc_iolock); 1436 dks->dks_paramlen = 0; 1437 break; 1438 } 1439 1440 case DIOCGCACHE: 1441 { 1442 int dkcache = 0; 1443 1444 /* 1445 * We pass this call down to all components and report 1446 * intersection of the flags returned by the components. 1447 * If any errors out, we return error. CCD components 1448 * can not change unless the device is unconfigured, so 1449 * device feature flags will remain static. RCE/WCE can change 1450 * of course, if set directly on underlying device. 1451 */ 1452 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1453 error = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, &j, 1454 flag, uc); 1455 if (error) 1456 break; 1457 1458 if (i == 0) 1459 dkcache = j; 1460 else 1461 dkcache = DKCACHE_COMBINE(dkcache, j); 1462 } 1463 1464 *((int *)data) = dkcache; 1465 break; 1466 } 1467 1468 case DIOCCACHESYNC: 1469 /* 1470 * We pass this call down to all components and report 1471 * the first error we encounter. 1472 */ 1473 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1474 j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data, 1475 flag, uc); 1476 if (j != 0 && error == 0) 1477 error = j; 1478 } 1479 break; 1480 1481 case DIOCWDINFO: 1482 case DIOCSDINFO: 1483 #ifdef __HAVE_OLD_DISKLABEL 1484 case ODIOCWDINFO: 1485 case ODIOCSDINFO: 1486 #endif 1487 { 1488 struct disklabel *lp; 1489 #ifdef __HAVE_OLD_DISKLABEL 1490 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1491 memset(&newlabel, 0, sizeof newlabel); 1492 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1493 lp = &newlabel; 1494 } else 1495 #endif 1496 lp = (struct disklabel *)data; 1497 1498 cs->sc_flags |= CCDF_LABELLING; 1499 1500 error = setdisklabel(cs->sc_dkdev.dk_label, 1501 lp, 0, cs->sc_dkdev.dk_cpulabel); 1502 if (error == 0) { 1503 if (cmd == DIOCWDINFO 1504 #ifdef __HAVE_OLD_DISKLABEL 1505 || cmd == ODIOCWDINFO 1506 #endif 1507 ) 1508 error = writedisklabel(CCDLABELDEV(dev), 1509 ccdstrategy, cs->sc_dkdev.dk_label, 1510 cs->sc_dkdev.dk_cpulabel); 1511 } 1512 1513 cs->sc_flags &= ~CCDF_LABELLING; 1514 break; 1515 } 1516 1517 case DIOCKLABEL: 1518 if (*(int *)data != 0) 1519 cs->sc_flags |= CCDF_KLABEL; 1520 else 1521 cs->sc_flags &= ~CCDF_KLABEL; 1522 break; 1523 1524 case DIOCWLABEL: 1525 if (*(int *)data != 0) 1526 cs->sc_flags |= CCDF_WLABEL; 1527 else 1528 cs->sc_flags &= ~CCDF_WLABEL; 1529 break; 1530 1531 case DIOCGDEFLABEL: 1532 ccdgetdefaultlabel(cs, (struct disklabel *)data); 1533 break; 1534 1535 #ifdef __HAVE_OLD_DISKLABEL 1536 case ODIOCGDEFLABEL: 1537 ccdgetdefaultlabel(cs, &newlabel); 1538 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1539 return ENOTTY; 1540 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1541 break; 1542 #endif 1543 1544 default: 1545 error = ENOTTY; 1546 } 1547 1548 out: 1549 mutex_exit(&cs->sc_dvlock); 1550 return (error); 1551 } 1552 1553 static int 1554 ccdsize(dev_t dev) 1555 { 1556 struct ccd_softc *cs; 1557 struct disklabel *lp; 1558 int part, unit, omask, size; 1559 1560 unit = ccdunit(dev); 1561 if ((cs = ccdget(unit, 0)) == NULL) 1562 return -1; 1563 1564 if ((cs->sc_flags & CCDF_INITED) == 0) 1565 return (-1); 1566 1567 part = DISKPART(dev); 1568 omask = cs->sc_dkdev.dk_openmask & (1 << part); 1569 lp = cs->sc_dkdev.dk_label; 1570 1571 if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp)) 1572 return (-1); 1573 1574 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1575 size = -1; 1576 else 1577 size = lp->d_partitions[part].p_size * 1578 (lp->d_secsize / DEV_BSIZE); 1579 1580 if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp)) 1581 return (-1); 1582 1583 return (size); 1584 } 1585 1586 static void 1587 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp) 1588 { 1589 struct ccdgeom *ccg = &cs->sc_geom; 1590 1591 memset(lp, 0, sizeof(*lp)); 1592 1593 if (cs->sc_size > UINT32_MAX) 1594 lp->d_secperunit = UINT32_MAX; 1595 else 1596 lp->d_secperunit = cs->sc_size; 1597 lp->d_secsize = ccg->ccg_secsize; 1598 lp->d_nsectors = ccg->ccg_nsectors; 1599 lp->d_ntracks = ccg->ccg_ntracks; 1600 lp->d_ncylinders = ccg->ccg_ncylinders; 1601 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1602 1603 strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename)); 1604 lp->d_type = DKTYPE_CCD; 1605 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1606 lp->d_rpm = 3600; 1607 lp->d_interleave = 1; 1608 lp->d_flags = 0; 1609 1610 lp->d_partitions[RAW_PART].p_offset = 0; 1611 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit; 1612 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 1613 lp->d_npartitions = RAW_PART + 1; 1614 1615 lp->d_magic = DISKMAGIC; 1616 lp->d_magic2 = DISKMAGIC; 1617 lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label); 1618 } 1619 1620 /* 1621 * Read the disklabel from the ccd. If one is not present, fake one 1622 * up. 1623 */ 1624 static void 1625 ccdgetdisklabel(dev_t dev) 1626 { 1627 int unit = ccdunit(dev); 1628 struct ccd_softc *cs; 1629 const char *errstring; 1630 struct disklabel *lp; 1631 struct cpu_disklabel *clp; 1632 1633 if ((cs = ccdget(unit, 0)) == NULL) 1634 return; 1635 lp = cs->sc_dkdev.dk_label; 1636 clp = cs->sc_dkdev.dk_cpulabel; 1637 KASSERT(mutex_owned(&cs->sc_dvlock)); 1638 1639 memset(clp, 0, sizeof(*clp)); 1640 1641 ccdgetdefaultlabel(cs, lp); 1642 1643 /* 1644 * Call the generic disklabel extraction routine. 1645 */ 1646 cs->sc_flags |= CCDF_RLABEL; 1647 if ((cs->sc_flags & CCDF_NOLABEL) != 0) 1648 errstring = "CCDF_NOLABEL set; ignoring on-disk label"; 1649 else 1650 errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy, 1651 cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel); 1652 if (errstring) 1653 ccdmakedisklabel(cs); 1654 else { 1655 int i; 1656 struct partition *pp; 1657 1658 /* 1659 * Sanity check whether the found disklabel is valid. 1660 * 1661 * This is necessary since total size of ccd may vary 1662 * when an interleave is changed even though exactly 1663 * same componets are used, and old disklabel may used 1664 * if that is found. 1665 */ 1666 if (lp->d_secperunit < UINT32_MAX ? 1667 lp->d_secperunit != cs->sc_size : 1668 lp->d_secperunit > cs->sc_size) 1669 printf("WARNING: %s: " 1670 "total sector size in disklabel (%ju) != " 1671 "the size of ccd (%ju)\n", cs->sc_xname, 1672 (uintmax_t)lp->d_secperunit, 1673 (uintmax_t)cs->sc_size); 1674 for (i = 0; i < lp->d_npartitions; i++) { 1675 pp = &lp->d_partitions[i]; 1676 if (pp->p_offset + pp->p_size > cs->sc_size) 1677 printf("WARNING: %s: end of partition `%c' " 1678 "exceeds the size of ccd (%ju)\n", 1679 cs->sc_xname, 'a' + i, (uintmax_t)cs->sc_size); 1680 } 1681 } 1682 1683 #ifdef DEBUG 1684 /* It's actually extremely common to have unlabeled ccds. */ 1685 if (ccddebug & CCDB_LABEL) 1686 if (errstring != NULL) 1687 printf("%s: %s\n", cs->sc_xname, errstring); 1688 #endif 1689 1690 /* In-core label now valid. */ 1691 cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL; 1692 } 1693 1694 /* 1695 * Take care of things one might want to take care of in the event 1696 * that a disklabel isn't present. 1697 */ 1698 static void 1699 ccdmakedisklabel(struct ccd_softc *cs) 1700 { 1701 struct disklabel *lp = cs->sc_dkdev.dk_label; 1702 1703 /* 1704 * For historical reasons, if there's no disklabel present 1705 * the raw partition must be marked FS_BSDFFS. 1706 */ 1707 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 1708 1709 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 1710 1711 lp->d_checksum = dkcksum(lp); 1712 } 1713 1714 #ifdef DEBUG 1715 static void 1716 printiinfo(struct ccdiinfo *ii) 1717 { 1718 int ix, i; 1719 1720 for (ix = 0; ii->ii_ndisk; ix++, ii++) { 1721 printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64, 1722 ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff); 1723 for (i = 0; i < ii->ii_ndisk; i++) 1724 printf(" %d", ii->ii_index[i]); 1725 printf("\n"); 1726 } 1727 } 1728 #endif 1729 1730 MODULE(MODULE_CLASS_DRIVER, ccd, "dk_subr"); 1731 1732 static int 1733 ccd_modcmd(modcmd_t cmd, void *arg) 1734 { 1735 int error = 0; 1736 #ifdef _MODULE 1737 int bmajor = -1, cmajor = -1; 1738 #endif 1739 1740 1741 switch (cmd) { 1742 case MODULE_CMD_INIT: 1743 #ifdef _MODULE 1744 ccdattach(0); 1745 1746 error = devsw_attach("ccd", &ccd_bdevsw, &bmajor, 1747 &ccd_cdevsw, &cmajor); 1748 sysctl_kern_ccd_setup(&ccd_clog); 1749 #endif 1750 break; 1751 1752 case MODULE_CMD_FINI: 1753 #ifdef _MODULE 1754 mutex_enter(&ccd_lock); 1755 if (ccd_nactive) { 1756 mutex_exit(&ccd_lock); 1757 error = EBUSY; 1758 } else { 1759 mutex_exit(&ccd_lock); 1760 error = devsw_detach(&ccd_bdevsw, &ccd_cdevsw); 1761 ccddetach(); 1762 } 1763 sysctl_teardown(&ccd_clog); 1764 #endif 1765 break; 1766 1767 case MODULE_CMD_STAT: 1768 return ENOTTY; 1769 1770 default: 1771 return ENOTTY; 1772 } 1773 1774 return error; 1775 } 1776 1777 static int 1778 ccd_units_sysctl(SYSCTLFN_ARGS) 1779 { 1780 struct sysctlnode node; 1781 struct ccd_softc *sc; 1782 int error, i, nccd, *units; 1783 size_t size; 1784 1785 nccd = 0; 1786 mutex_enter(&ccd_lock); 1787 LIST_FOREACH(sc, &ccds, sc_link) 1788 nccd++; 1789 mutex_exit(&ccd_lock); 1790 1791 if (nccd != 0) { 1792 size = nccd * sizeof(*units); 1793 units = kmem_zalloc(size, KM_SLEEP); 1794 i = 0; 1795 mutex_enter(&ccd_lock); 1796 LIST_FOREACH(sc, &ccds, sc_link) { 1797 if (i >= nccd) 1798 break; 1799 units[i] = sc->sc_unit; 1800 } 1801 mutex_exit(&ccd_lock); 1802 } else { 1803 units = NULL; 1804 size = 0; 1805 } 1806 1807 node = *rnode; 1808 node.sysctl_data = units; 1809 node.sysctl_size = size; 1810 1811 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1812 if (units) 1813 kmem_free(units, size); 1814 return error; 1815 } 1816 1817 static int 1818 ccd_info_sysctl(SYSCTLFN_ARGS) 1819 { 1820 struct sysctlnode node; 1821 struct ccddiskinfo ccd; 1822 struct ccd_softc *sc; 1823 int unit; 1824 1825 if (newp == NULL || newlen != sizeof(int)) 1826 return EINVAL; 1827 1828 unit = *(const int *)newp; 1829 newp = NULL; 1830 newlen = 0; 1831 ccd.ccd_ndisks = ~0; 1832 mutex_enter(&ccd_lock); 1833 LIST_FOREACH(sc, &ccds, sc_link) { 1834 if (sc->sc_unit == unit) { 1835 ccd.ccd_ileave = sc->sc_ileave; 1836 ccd.ccd_size = sc->sc_size; 1837 ccd.ccd_ndisks = sc->sc_nccdisks; 1838 ccd.ccd_flags = sc->sc_flags; 1839 break; 1840 } 1841 } 1842 mutex_exit(&ccd_lock); 1843 1844 if (ccd.ccd_ndisks == ~0) 1845 return ENOENT; 1846 1847 node = *rnode; 1848 node.sysctl_data = &ccd; 1849 node.sysctl_size = sizeof(ccd); 1850 1851 return sysctl_lookup(SYSCTLFN_CALL(&node)); 1852 } 1853 1854 static int 1855 ccd_components_sysctl(SYSCTLFN_ARGS) 1856 { 1857 struct sysctlnode node; 1858 int error, unit; 1859 size_t size; 1860 char *names, *p, *ep; 1861 struct ccd_softc *sc; 1862 1863 if (newp == NULL || newlen != sizeof(int)) 1864 return EINVAL; 1865 1866 size = 0; 1867 unit = *(const int *)newp; 1868 newp = NULL; 1869 newlen = 0; 1870 mutex_enter(&ccd_lock); 1871 LIST_FOREACH(sc, &ccds, sc_link) 1872 if (sc->sc_unit == unit) { 1873 for (size_t i = 0; i < sc->sc_nccdisks; i++) 1874 size += strlen(sc->sc_cinfo[i].ci_path) + 1; 1875 break; 1876 } 1877 mutex_exit(&ccd_lock); 1878 1879 if (size == 0) 1880 return ENOENT; 1881 names = kmem_zalloc(size, KM_SLEEP); 1882 p = names; 1883 ep = names + size; 1884 mutex_enter(&ccd_lock); 1885 LIST_FOREACH(sc, &ccds, sc_link) 1886 if (sc->sc_unit == unit) { 1887 for (size_t i = 0; i < sc->sc_nccdisks; i++) { 1888 char *d = sc->sc_cinfo[i].ci_path; 1889 while (p < ep && (*p++ = *d++) != '\0') 1890 continue; 1891 } 1892 break; 1893 } 1894 mutex_exit(&ccd_lock); 1895 1896 node = *rnode; 1897 node.sysctl_data = names; 1898 node.sysctl_size = ep - names; 1899 1900 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1901 kmem_free(names, size); 1902 return error; 1903 } 1904 1905 SYSCTL_SETUP(sysctl_kern_ccd_setup, "sysctl kern.ccd subtree setup") 1906 { 1907 const struct sysctlnode *node = NULL; 1908 1909 sysctl_createv(clog, 0, NULL, &node, 1910 CTLFLAG_PERMANENT, 1911 CTLTYPE_NODE, "ccd", 1912 SYSCTL_DESCR("ConCatenated Disk state"), 1913 NULL, 0, NULL, 0, 1914 CTL_KERN, CTL_CREATE, CTL_EOL); 1915 1916 if (node == NULL) 1917 return; 1918 1919 sysctl_createv(clog, 0, &node, NULL, 1920 CTLFLAG_PERMANENT | CTLFLAG_READONLY, 1921 CTLTYPE_STRUCT, "units", 1922 SYSCTL_DESCR("List of ccd unit numbers"), 1923 ccd_units_sysctl, 0, NULL, 0, 1924 CTL_CREATE, CTL_EOL); 1925 sysctl_createv(clog, 0, &node, NULL, 1926 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1927 CTLTYPE_STRUCT, "info", 1928 SYSCTL_DESCR("Information about a CCD unit"), 1929 ccd_info_sysctl, 0, NULL, 0, 1930 CTL_CREATE, CTL_EOL); 1931 sysctl_createv(clog, 0, &node, NULL, 1932 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1933 CTLTYPE_STRUCT, "components", 1934 SYSCTL_DESCR("Information about CCD components"), 1935 ccd_components_sysctl, 0, NULL, 0, 1936 CTL_CREATE, CTL_EOL); 1937 } 1938
Indexes created Thu Oct 02 14:10:14 GMT 2025