ccd.c revision 1.171.4.2
1 /* $NetBSD: ccd.c,v 1.171.4.2 2017/04/29 10:50:46 pgoyette 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 1.171.4.2 2017/04/29 10:50:46 pgoyette 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 *); 180 181 static dev_type_open(ccdopen); 182 static dev_type_close(ccdclose); 183 static dev_type_read(ccdread); 184 static dev_type_write(ccdwrite); 185 static dev_type_ioctl(ccdioctl); 186 static dev_type_strategy(ccdstrategy); 187 static dev_type_size(ccdsize); 188 189 const struct bdevsw ccd_bdevsw = { 190 DEVSW_MODULE_INIT 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 DEVSW_MODULE_INIT 203 .d_open = ccdopen, 204 .d_close = ccdclose, 205 .d_read = ccdread, 206 .d_write = ccdwrite, 207 .d_ioctl = ccdioctl, 208 .d_stop = nostop, 209 .d_tty = notty, 210 .d_poll = nopoll, 211 .d_mmap = nommap, 212 .d_kqfilter = nokqfilter, 213 .d_discard = nodiscard, 214 .d_flag = D_DISK | D_MPSAFE 215 }; 216 217 #ifdef DEBUG 218 static void printiinfo(struct ccdiinfo *); 219 #endif 220 221 static LIST_HEAD(, ccd_softc) ccds = LIST_HEAD_INITIALIZER(ccds); 222 static kmutex_t ccd_lock; 223 static size_t ccd_nactive = 0; 224 225 #ifdef _MODULE 226 static struct sysctllog *ccd_clog; 227 #endif 228 229 SYSCTL_SETUP_PROTO(sysctl_kern_ccd_setup); 230 231 static struct ccd_softc * 232 ccdcreate(int unit) { 233 struct ccd_softc *sc = kmem_zalloc(sizeof(*sc), KM_SLEEP); 234 if (sc == NULL) { 235 #ifdef DIAGNOSTIC 236 printf("%s: out of memory\n", __func__); 237 #endif 238 return NULL; 239 } 240 /* Initialize per-softc structures. */ 241 snprintf(sc->sc_xname, sizeof(sc->sc_xname), "ccd%d", unit); 242 sc->sc_unit = unit; 243 mutex_init(&sc->sc_dvlock, MUTEX_DEFAULT, IPL_NONE); 244 sc->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 245 cv_init(&sc->sc_stop, "ccdstop"); 246 cv_init(&sc->sc_push, "ccdthr"); 247 disk_init(&sc->sc_dkdev, sc->sc_xname, NULL); /* XXX */ 248 return sc; 249 } 250 251 static void 252 ccddestroy(struct ccd_softc *sc) { 253 mutex_obj_free(sc->sc_iolock); 254 mutex_exit(&sc->sc_dvlock); 255 mutex_destroy(&sc->sc_dvlock); 256 cv_destroy(&sc->sc_stop); 257 cv_destroy(&sc->sc_push); 258 disk_destroy(&sc->sc_dkdev); 259 kmem_free(sc, sizeof(*sc)); 260 } 261 262 static struct ccd_softc * 263 ccdget(int unit, int make) { 264 struct ccd_softc *sc; 265 if (unit < 0) { 266 #ifdef DIAGNOSTIC 267 panic("%s: unit %d!", __func__, unit); 268 #endif 269 return NULL; 270 } 271 mutex_enter(&ccd_lock); 272 LIST_FOREACH(sc, &ccds, sc_link) { 273 if (sc->sc_unit == unit) { 274 mutex_exit(&ccd_lock); 275 return sc; 276 } 277 } 278 mutex_exit(&ccd_lock); 279 if (!make) 280 return NULL; 281 if ((sc = ccdcreate(unit)) == NULL) 282 return NULL; 283 mutex_enter(&ccd_lock); 284 LIST_INSERT_HEAD(&ccds, sc, sc_link); 285 ccd_nactive++; 286 mutex_exit(&ccd_lock); 287 return sc; 288 } 289 290 static void 291 ccdput(struct ccd_softc *sc) { 292 mutex_enter(&ccd_lock); 293 LIST_REMOVE(sc, sc_link); 294 ccd_nactive--; 295 mutex_exit(&ccd_lock); 296 ccddestroy(sc); 297 } 298 299 /* 300 * Called by main() during pseudo-device attachment. All we need 301 * to do is allocate enough space for devices to be configured later. 302 */ 303 void 304 ccdattach(int num) 305 { 306 mutex_init(&ccd_lock, MUTEX_DEFAULT, IPL_NONE); 307 308 /* Initialize the component buffer pool. */ 309 ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0, 310 0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL); 311 } 312 313 void 314 ccddetach(void) 315 { 316 pool_cache_destroy(ccd_cache); 317 mutex_destroy(&ccd_lock); 318 } 319 320 static int 321 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp, 322 struct lwp *l) 323 { 324 struct ccdcinfo *ci = NULL; 325 int ix; 326 struct ccdgeom *ccg = &cs->sc_geom; 327 char *tmppath; 328 int error, path_alloced; 329 uint64_t psize, minsize; 330 unsigned secsize, maxsecsize; 331 struct disk_geom *dg; 332 333 #ifdef DEBUG 334 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 335 printf("%s: ccdinit\n", cs->sc_xname); 336 #endif 337 338 /* Allocate space for the component info. */ 339 cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo), 340 KM_SLEEP); 341 tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 342 343 cs->sc_size = 0; 344 345 /* 346 * Verify that each component piece exists and record 347 * relevant information about it. 348 */ 349 maxsecsize = 0; 350 minsize = 0; 351 for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) { 352 ci = &cs->sc_cinfo[ix]; 353 ci->ci_vp = vpp[ix]; 354 355 /* 356 * Copy in the pathname of the component. 357 */ 358 memset(tmppath, 0, MAXPATHLEN); /* sanity */ 359 error = copyinstr(cpaths[ix], tmppath, 360 MAXPATHLEN, &ci->ci_pathlen); 361 if (ci->ci_pathlen == 0) 362 error = EINVAL; 363 if (error) { 364 #ifdef DEBUG 365 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 366 printf("%s: can't copy path, error = %d\n", 367 cs->sc_xname, error); 368 #endif 369 goto out; 370 } 371 ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP); 372 memcpy(ci->ci_path, tmppath, ci->ci_pathlen); 373 path_alloced++; 374 375 /* 376 * XXX: Cache the component's dev_t. 377 */ 378 ci->ci_dev = vpp[ix]->v_rdev; 379 380 /* 381 * Get partition information for the component. 382 */ 383 error = getdisksize(vpp[ix], &psize, &secsize); 384 if (error) { 385 #ifdef DEBUG 386 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 387 printf("%s: %s: disksize failed, error = %d\n", 388 cs->sc_xname, ci->ci_path, error); 389 #endif 390 goto out; 391 } 392 393 /* 394 * Calculate the size, truncating to an interleave 395 * boundary if necessary. 396 */ 397 maxsecsize = secsize > maxsecsize ? secsize : maxsecsize; 398 if (cs->sc_ileave > 1) 399 psize -= psize % cs->sc_ileave; 400 401 if (psize == 0) { 402 #ifdef DEBUG 403 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 404 printf("%s: %s: size == 0\n", 405 cs->sc_xname, ci->ci_path); 406 #endif 407 error = ENODEV; 408 goto out; 409 } 410 411 if (minsize == 0 || psize < minsize) 412 minsize = psize; 413 ci->ci_size = psize; 414 cs->sc_size += psize; 415 } 416 417 /* 418 * Don't allow the interleave to be smaller than 419 * the biggest component sector. 420 */ 421 if ((cs->sc_ileave > 0) && 422 (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) { 423 #ifdef DEBUG 424 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 425 printf("%s: interleave must be at least %d\n", 426 cs->sc_xname, (maxsecsize / DEV_BSIZE)); 427 #endif 428 error = EINVAL; 429 goto out; 430 } 431 432 /* 433 * If uniform interleave is desired set all sizes to that of 434 * the smallest component. 435 */ 436 if (cs->sc_flags & CCDF_UNIFORM) { 437 for (ci = cs->sc_cinfo; 438 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 439 ci->ci_size = minsize; 440 441 cs->sc_size = cs->sc_nccdisks * minsize; 442 } 443 444 /* 445 * Construct the interleave table. 446 */ 447 ccdinterleave(cs); 448 449 /* 450 * Create pseudo-geometry based on 1MB cylinders. It's 451 * pretty close. 452 */ 453 ccg->ccg_secsize = DEV_BSIZE; 454 ccg->ccg_ntracks = 1; 455 ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize); 456 ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors; 457 458 dg = &cs->sc_dkdev.dk_geom; 459 memset(dg, 0, sizeof(*dg)); 460 dg->dg_secperunit = cs->sc_size; 461 dg->dg_secsize = ccg->ccg_secsize; 462 dg->dg_nsectors = ccg->ccg_nsectors; 463 dg->dg_ntracks = ccg->ccg_ntracks; 464 dg->dg_ncylinders = ccg->ccg_ncylinders; 465 466 if (cs->sc_ileave > 0) 467 aprint_normal("%s: Interleaving %d component%s " 468 "(%d block interleave)\n", cs->sc_xname, 469 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""), 470 cs->sc_ileave); 471 else 472 aprint_normal("%s: Concatenating %d component%s\n", 473 cs->sc_xname, 474 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : "")); 475 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 476 ci = &cs->sc_cinfo[ix]; 477 aprint_normal("%s: %s (%ju blocks)\n", cs->sc_xname, 478 ci->ci_path, (uintmax_t)ci->ci_size); 479 } 480 aprint_normal("%s: total %ju blocks\n", cs->sc_xname, cs->sc_size); 481 482 /* 483 * Create thread to handle deferred I/O. 484 */ 485 cs->sc_zap = false; 486 error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread, 487 cs, &cs->sc_thread, "%s", cs->sc_xname); 488 if (error) { 489 printf("ccdinit: can't create thread: %d\n", error); 490 goto out; 491 } 492 493 /* 494 * Only now that everything is set up can we enable the device. 495 */ 496 mutex_enter(cs->sc_iolock); 497 cs->sc_flags |= CCDF_INITED; 498 mutex_exit(cs->sc_iolock); 499 kmem_free(tmppath, MAXPATHLEN); 500 return (0); 501 502 out: 503 for (ix = 0; ix < path_alloced; ix++) { 504 kmem_free(cs->sc_cinfo[ix].ci_path, 505 cs->sc_cinfo[ix].ci_pathlen); 506 } 507 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo)); 508 kmem_free(tmppath, MAXPATHLEN); 509 return (error); 510 } 511 512 static void 513 ccdinterleave(struct ccd_softc *cs) 514 { 515 struct ccdcinfo *ci, *smallci; 516 struct ccdiinfo *ii; 517 daddr_t bn, lbn; 518 int ix; 519 u_long size; 520 521 #ifdef DEBUG 522 if (ccddebug & CCDB_INIT) 523 printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave); 524 #endif 525 /* 526 * Allocate an interleave table. 527 * Chances are this is too big, but we don't care. 528 */ 529 size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo); 530 cs->sc_itable = kmem_zalloc(size, KM_SLEEP); 531 532 /* 533 * Trivial case: no interleave (actually interleave of disk size). 534 * Each table entry represents a single component in its entirety. 535 */ 536 if (cs->sc_ileave == 0) { 537 bn = 0; 538 ii = cs->sc_itable; 539 540 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 541 /* Allocate space for ii_index. */ 542 ii->ii_indexsz = sizeof(int); 543 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 544 ii->ii_ndisk = 1; 545 ii->ii_startblk = bn; 546 ii->ii_startoff = 0; 547 ii->ii_index[0] = ix; 548 bn += cs->sc_cinfo[ix].ci_size; 549 ii++; 550 } 551 ii->ii_ndisk = 0; 552 #ifdef DEBUG 553 if (ccddebug & CCDB_INIT) 554 printiinfo(cs->sc_itable); 555 #endif 556 return; 557 } 558 559 /* 560 * The following isn't fast or pretty; it doesn't have to be. 561 */ 562 size = 0; 563 bn = lbn = 0; 564 for (ii = cs->sc_itable; ; ii++) { 565 /* Allocate space for ii_index. */ 566 ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks; 567 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 568 569 /* 570 * Locate the smallest of the remaining components 571 */ 572 smallci = NULL; 573 for (ci = cs->sc_cinfo; 574 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 575 if (ci->ci_size > size && 576 (smallci == NULL || 577 ci->ci_size < smallci->ci_size)) 578 smallci = ci; 579 580 /* 581 * Nobody left, all done 582 */ 583 if (smallci == NULL) { 584 ii->ii_ndisk = 0; 585 break; 586 } 587 588 /* 589 * Record starting logical block and component offset 590 */ 591 ii->ii_startblk = bn / cs->sc_ileave; 592 ii->ii_startoff = lbn; 593 594 /* 595 * Determine how many disks take part in this interleave 596 * and record their indices. 597 */ 598 ix = 0; 599 for (ci = cs->sc_cinfo; 600 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 601 if (ci->ci_size >= smallci->ci_size) 602 ii->ii_index[ix++] = ci - cs->sc_cinfo; 603 ii->ii_ndisk = ix; 604 bn += ix * (smallci->ci_size - size); 605 lbn = smallci->ci_size / cs->sc_ileave; 606 size = smallci->ci_size; 607 } 608 #ifdef DEBUG 609 if (ccddebug & CCDB_INIT) 610 printiinfo(cs->sc_itable); 611 #endif 612 } 613 614 /* ARGSUSED */ 615 static int 616 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l) 617 { 618 int unit = ccdunit(dev); 619 struct ccd_softc *cs; 620 struct disklabel *lp; 621 int error = 0, part, pmask; 622 623 #ifdef DEBUG 624 if (ccddebug & CCDB_FOLLOW) 625 printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags); 626 #endif 627 if ((cs = ccdget(unit, 1)) == NULL) 628 return ENXIO; 629 630 mutex_enter(&cs->sc_dvlock); 631 632 lp = cs->sc_dkdev.dk_label; 633 634 part = DISKPART(dev); 635 pmask = (1 << part); 636 637 /* 638 * If we're initialized, check to see if there are any other 639 * open partitions. If not, then it's safe to update 640 * the in-core disklabel. Only read the disklabel if it is 641 * not already valid. 642 */ 643 if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED && 644 cs->sc_dkdev.dk_openmask == 0) 645 ccdgetdisklabel(dev); 646 647 /* Check that the partition exists. */ 648 if (part != RAW_PART) { 649 if (((cs->sc_flags & CCDF_INITED) == 0) || 650 ((part >= lp->d_npartitions) || 651 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 652 error = ENXIO; 653 goto done; 654 } 655 } 656 657 /* Prevent our unit from being unconfigured while open. */ 658 switch (fmt) { 659 case S_IFCHR: 660 cs->sc_dkdev.dk_copenmask |= pmask; 661 break; 662 663 case S_IFBLK: 664 cs->sc_dkdev.dk_bopenmask |= pmask; 665 break; 666 } 667 cs->sc_dkdev.dk_openmask = 668 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 669 670 done: 671 mutex_exit(&cs->sc_dvlock); 672 return (error); 673 } 674 675 /* ARGSUSED */ 676 static int 677 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l) 678 { 679 int unit = ccdunit(dev); 680 struct ccd_softc *cs; 681 int part; 682 683 #ifdef DEBUG 684 if (ccddebug & CCDB_FOLLOW) 685 printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags); 686 #endif 687 688 if ((cs = ccdget(unit, 0)) == NULL) 689 return ENXIO; 690 691 mutex_enter(&cs->sc_dvlock); 692 693 part = DISKPART(dev); 694 695 /* ...that much closer to allowing unconfiguration... */ 696 switch (fmt) { 697 case S_IFCHR: 698 cs->sc_dkdev.dk_copenmask &= ~(1 << part); 699 break; 700 701 case S_IFBLK: 702 cs->sc_dkdev.dk_bopenmask &= ~(1 << part); 703 break; 704 } 705 cs->sc_dkdev.dk_openmask = 706 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 707 708 if (cs->sc_dkdev.dk_openmask == 0) { 709 if ((cs->sc_flags & CCDF_KLABEL) == 0) 710 cs->sc_flags &= ~CCDF_VLABEL; 711 } 712 713 mutex_exit(&cs->sc_dvlock); 714 return (0); 715 } 716 717 static bool 718 ccdbackoff(struct ccd_softc *cs) 719 { 720 721 /* XXX Arbitrary, should be a uvm call. */ 722 return uvmexp.free < (uvmexp.freemin >> 1) && 723 disk_isbusy(&cs->sc_dkdev); 724 } 725 726 static void 727 ccdthread(void *cookie) 728 { 729 struct ccd_softc *cs; 730 731 cs = cookie; 732 733 #ifdef DEBUG 734 if (ccddebug & CCDB_FOLLOW) 735 printf("ccdthread: hello\n"); 736 #endif 737 738 mutex_enter(cs->sc_iolock); 739 while (__predict_true(!cs->sc_zap)) { 740 if (bufq_peek(cs->sc_bufq) == NULL) { 741 /* Nothing to do. */ 742 cv_wait(&cs->sc_push, cs->sc_iolock); 743 continue; 744 } 745 if (ccdbackoff(cs)) { 746 /* Wait for memory to become available. */ 747 (void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1); 748 continue; 749 } 750 #ifdef DEBUG 751 if (ccddebug & CCDB_FOLLOW) 752 printf("ccdthread: dispatching I/O\n"); 753 #endif 754 ccdstart(cs); 755 mutex_enter(cs->sc_iolock); 756 } 757 cs->sc_thread = NULL; 758 mutex_exit(cs->sc_iolock); 759 #ifdef DEBUG 760 if (ccddebug & CCDB_FOLLOW) 761 printf("ccdthread: goodbye\n"); 762 #endif 763 kthread_exit(0); 764 } 765 766 static void 767 ccdstrategy(struct buf *bp) 768 { 769 int unit = ccdunit(bp->b_dev); 770 struct ccd_softc *cs; 771 if ((cs = ccdget(unit, 0)) == NULL) 772 return; 773 774 /* Must be open or reading label. */ 775 KASSERT(cs->sc_dkdev.dk_openmask != 0 || 776 (cs->sc_flags & CCDF_RLABEL) != 0); 777 778 mutex_enter(cs->sc_iolock); 779 /* Synchronize with device init/uninit. */ 780 if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) { 781 mutex_exit(cs->sc_iolock); 782 #ifdef DEBUG 783 if (ccddebug & CCDB_FOLLOW) 784 printf("ccdstrategy: unit %d: not inited\n", unit); 785 #endif 786 bp->b_error = ENXIO; 787 bp->b_resid = bp->b_bcount; 788 biodone(bp); 789 return; 790 } 791 792 /* Defer to thread if system is low on memory. */ 793 bufq_put(cs->sc_bufq, bp); 794 if (__predict_false(ccdbackoff(cs))) { 795 mutex_exit(cs->sc_iolock); 796 #ifdef DEBUG 797 if (ccddebug & CCDB_FOLLOW) 798 printf("ccdstrategy: holding off on I/O\n"); 799 #endif 800 return; 801 } 802 ccdstart(cs); 803 } 804 805 static void 806 ccdstart(struct ccd_softc *cs) 807 { 808 daddr_t blkno; 809 int wlabel; 810 struct disklabel *lp; 811 long bcount, rcount; 812 struct ccdbuf *cbp; 813 char *addr; 814 daddr_t bn; 815 vnode_t *vp; 816 buf_t *bp; 817 818 KASSERT(mutex_owned(cs->sc_iolock)); 819 820 bp = bufq_get(cs->sc_bufq); 821 KASSERT(bp != NULL); 822 823 disk_busy(&cs->sc_dkdev); 824 825 #ifdef DEBUG 826 if (ccddebug & CCDB_FOLLOW) 827 printf("ccdstart(%s, %p)\n", cs->sc_xname, bp); 828 #endif 829 830 /* If it's a nil transfer, wake up the top half now. */ 831 if (bp->b_bcount == 0) 832 goto done; 833 834 lp = cs->sc_dkdev.dk_label; 835 836 /* 837 * Do bounds checking and adjust transfer. If there's an 838 * error, the bounds check will flag that for us. Convert 839 * the partition relative block number to an absolute. 840 */ 841 blkno = bp->b_blkno; 842 wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING); 843 if (DISKPART(bp->b_dev) != RAW_PART) { 844 if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0) 845 goto done; 846 blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset; 847 } 848 mutex_exit(cs->sc_iolock); 849 bp->b_rawblkno = blkno; 850 851 /* Allocate the component buffers and start I/O! */ 852 bp->b_resid = bp->b_bcount; 853 bn = bp->b_rawblkno; 854 addr = bp->b_data; 855 for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) { 856 cbp = ccdbuffer(cs, bp, bn, addr, bcount); 857 rcount = cbp->cb_buf.b_bcount; 858 bn += btodb(rcount); 859 addr += rcount; 860 vp = cbp->cb_buf.b_vp; 861 if ((cbp->cb_buf.b_flags & B_READ) == 0) { 862 mutex_enter(vp->v_interlock); 863 vp->v_numoutput++; 864 mutex_exit(vp->v_interlock); 865 } 866 (void)VOP_STRATEGY(vp, &cbp->cb_buf); 867 } 868 return; 869 870 done: 871 disk_unbusy(&cs->sc_dkdev, 0, 0); 872 cv_broadcast(&cs->sc_stop); 873 cv_broadcast(&cs->sc_push); 874 mutex_exit(cs->sc_iolock); 875 bp->b_resid = bp->b_bcount; 876 biodone(bp); 877 } 878 879 /* 880 * Build a component buffer header. 881 */ 882 static struct ccdbuf * 883 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr, 884 long bcount) 885 { 886 struct ccdcinfo *ci; 887 struct ccdbuf *cbp; 888 daddr_t cbn, cboff; 889 u_int64_t cbc; 890 int ccdisk; 891 892 #ifdef DEBUG 893 if (ccddebug & CCDB_IO) 894 printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n", 895 cs, bp, bn, addr, bcount); 896 #endif 897 /* 898 * Determine which component bn falls in. 899 */ 900 cbn = bn; 901 cboff = 0; 902 903 /* 904 * Serially concatenated 905 */ 906 if (cs->sc_ileave == 0) { 907 daddr_t sblk; 908 909 sblk = 0; 910 for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk]; 911 cbn >= sblk + ci->ci_size; 912 ccdisk++, ci = &cs->sc_cinfo[ccdisk]) 913 sblk += ci->ci_size; 914 cbn -= sblk; 915 } 916 /* 917 * Interleaved 918 */ 919 else { 920 struct ccdiinfo *ii; 921 int off; 922 923 cboff = cbn % cs->sc_ileave; 924 cbn /= cs->sc_ileave; 925 for (ii = cs->sc_itable; ii->ii_ndisk; ii++) 926 if (ii->ii_startblk > cbn) 927 break; 928 ii--; 929 off = cbn - ii->ii_startblk; 930 if (ii->ii_ndisk == 1) { 931 ccdisk = ii->ii_index[0]; 932 cbn = ii->ii_startoff + off; 933 } else { 934 ccdisk = ii->ii_index[off % ii->ii_ndisk]; 935 cbn = ii->ii_startoff + off / ii->ii_ndisk; 936 } 937 cbn *= cs->sc_ileave; 938 ci = &cs->sc_cinfo[ccdisk]; 939 } 940 941 /* 942 * Fill in the component buf structure. 943 */ 944 cbp = CCD_GETBUF(); 945 KASSERT(cbp != NULL); 946 buf_init(&cbp->cb_buf); 947 cbp->cb_buf.b_flags = bp->b_flags; 948 cbp->cb_buf.b_oflags = bp->b_oflags; 949 cbp->cb_buf.b_cflags = bp->b_cflags; 950 cbp->cb_buf.b_iodone = ccdiodone; 951 cbp->cb_buf.b_proc = bp->b_proc; 952 cbp->cb_buf.b_dev = ci->ci_dev; 953 cbp->cb_buf.b_blkno = cbn + cboff; 954 cbp->cb_buf.b_data = addr; 955 cbp->cb_buf.b_vp = ci->ci_vp; 956 cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock; 957 if (cs->sc_ileave == 0) 958 cbc = dbtob((u_int64_t)(ci->ci_size - cbn)); 959 else 960 cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff)); 961 cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount; 962 963 /* 964 * context for ccdiodone 965 */ 966 cbp->cb_obp = bp; 967 cbp->cb_sc = cs; 968 cbp->cb_comp = ccdisk; 969 970 BIO_COPYPRIO(&cbp->cb_buf, bp); 971 972 #ifdef DEBUG 973 if (ccddebug & CCDB_IO) 974 printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p" 975 " bcnt %d\n", 976 ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp, 977 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 978 cbp->cb_buf.b_bcount); 979 #endif 980 981 return (cbp); 982 } 983 984 /* 985 * Called at interrupt time. 986 * Mark the component as done and if all components are done, 987 * take a ccd interrupt. 988 */ 989 static void 990 ccdiodone(struct buf *vbp) 991 { 992 struct ccdbuf *cbp = (struct ccdbuf *) vbp; 993 struct buf *bp = cbp->cb_obp; 994 struct ccd_softc *cs = cbp->cb_sc; 995 int count; 996 997 #ifdef DEBUG 998 if (ccddebug & CCDB_FOLLOW) 999 printf("ccdiodone(%p)\n", cbp); 1000 if (ccddebug & CCDB_IO) { 1001 printf("ccdiodone: bp %p bcount %d resid %d\n", 1002 bp, bp->b_bcount, bp->b_resid); 1003 printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p" 1004 " bcnt %d\n", 1005 cbp->cb_buf.b_dev, cbp->cb_comp, cbp, 1006 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 1007 cbp->cb_buf.b_bcount); 1008 } 1009 #endif 1010 1011 if (cbp->cb_buf.b_error != 0) { 1012 bp->b_error = cbp->cb_buf.b_error; 1013 printf("%s: error %d on component %d\n", 1014 cs->sc_xname, bp->b_error, cbp->cb_comp); 1015 } 1016 count = cbp->cb_buf.b_bcount; 1017 buf_destroy(&cbp->cb_buf); 1018 CCD_PUTBUF(cbp); 1019 1020 /* 1021 * If all done, "interrupt". 1022 */ 1023 mutex_enter(cs->sc_iolock); 1024 bp->b_resid -= count; 1025 if (bp->b_resid < 0) 1026 panic("ccdiodone: count"); 1027 if (bp->b_resid == 0) { 1028 /* 1029 * Request is done for better or worse, wakeup the top half. 1030 */ 1031 if (bp->b_error != 0) 1032 bp->b_resid = bp->b_bcount; 1033 disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid), 1034 (bp->b_flags & B_READ)); 1035 if (!disk_isbusy(&cs->sc_dkdev)) { 1036 if (bufq_peek(cs->sc_bufq) != NULL) { 1037 cv_broadcast(&cs->sc_push); 1038 } 1039 cv_broadcast(&cs->sc_stop); 1040 } 1041 mutex_exit(cs->sc_iolock); 1042 biodone(bp); 1043 } else 1044 mutex_exit(cs->sc_iolock); 1045 } 1046 1047 /* ARGSUSED */ 1048 static int 1049 ccdread(dev_t dev, struct uio *uio, int flags) 1050 { 1051 int unit = ccdunit(dev); 1052 struct ccd_softc *cs; 1053 1054 #ifdef DEBUG 1055 if (ccddebug & CCDB_FOLLOW) 1056 printf("ccdread(0x%"PRIx64", %p)\n", dev, uio); 1057 #endif 1058 if ((cs = ccdget(unit, 0)) == NULL) 1059 return 0; 1060 1061 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1062 if ((cs->sc_flags & CCDF_INITED) == 0) 1063 return (ENXIO); 1064 1065 return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio)); 1066 } 1067 1068 /* ARGSUSED */ 1069 static int 1070 ccdwrite(dev_t dev, struct uio *uio, int flags) 1071 { 1072 int unit = ccdunit(dev); 1073 struct ccd_softc *cs; 1074 1075 #ifdef DEBUG 1076 if (ccddebug & CCDB_FOLLOW) 1077 printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio); 1078 #endif 1079 if ((cs = ccdget(unit, 0)) == NULL) 1080 return ENOENT; 1081 1082 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1083 if ((cs->sc_flags & CCDF_INITED) == 0) 1084 return (ENXIO); 1085 1086 return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio)); 1087 } 1088 1089 static int 1090 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1091 { 1092 int unit = ccdunit(dev); 1093 int i, j, lookedup = 0, error = 0; 1094 int part, pmask, make; 1095 struct ccd_softc *cs; 1096 struct ccd_ioctl *ccio = (struct ccd_ioctl *)data; 1097 kauth_cred_t uc; 1098 char **cpp; 1099 struct pathbuf *pb; 1100 struct vnode **vpp; 1101 #ifdef __HAVE_OLD_DISKLABEL 1102 struct disklabel newlabel; 1103 #endif 1104 1105 switch (cmd) { 1106 #if defined(COMPAT_60) && !defined(_LP64) 1107 case CCDIOCSET_60: 1108 #endif 1109 case CCDIOCSET: 1110 make = 1; 1111 break; 1112 default: 1113 make = 0; 1114 break; 1115 } 1116 1117 if ((cs = ccdget(unit, make)) == NULL) 1118 return ENOENT; 1119 uc = kauth_cred_get(); 1120 1121 /* 1122 * Compat code must not be called if on a platform where 1123 * sizeof (size_t) == sizeof (uint64_t) as CCDIOCSET will 1124 * be the same as CCDIOCSET_60 1125 */ 1126 #if defined(COMPAT_60) && !defined(_LP64) 1127 switch (cmd) { 1128 case CCDIOCSET_60: { 1129 struct ccd_ioctl ccionew; 1130 struct ccd_ioctl_60 *ccio60 = 1131 (struct ccd_ioctl_60 *)data; 1132 ccionew.ccio_disks = ccio->ccio_disks; 1133 ccionew.ccio_ndisks = ccio->ccio_ndisks; 1134 ccionew.ccio_ileave = ccio->ccio_ileave; 1135 ccionew.ccio_flags = ccio->ccio_flags; 1136 ccionew.ccio_unit = ccio->ccio_unit; 1137 error = ccdioctl(dev, CCDIOCSET, &ccionew, flag, l); 1138 if (!error) { 1139 /* Copy data back, adjust types if necessary */ 1140 ccio60->ccio_disks = ccionew.ccio_disks; 1141 ccio60->ccio_ndisks = ccionew.ccio_ndisks; 1142 ccio60->ccio_ileave = ccionew.ccio_ileave; 1143 ccio60->ccio_flags = ccionew.ccio_flags; 1144 ccio60->ccio_unit = ccionew.ccio_unit; 1145 ccio60->ccio_size = (size_t)ccionew.ccio_size; 1146 } 1147 return error; 1148 } 1149 break; 1150 1151 case CCDIOCCLR_60: 1152 /* 1153 * ccio_size member not used, so existing struct OK 1154 * drop through to existing non-compat version 1155 */ 1156 cmd = CCDIOCCLR; 1157 break; 1158 } 1159 #endif /* COMPAT_60 && !_LP64*/ 1160 1161 /* Must be open for writes for these commands... */ 1162 switch (cmd) { 1163 case CCDIOCSET: 1164 case CCDIOCCLR: 1165 case DIOCSDINFO: 1166 case DIOCWDINFO: 1167 case DIOCCACHESYNC: 1168 case DIOCAWEDGE: 1169 case DIOCDWEDGE: 1170 case DIOCMWEDGES: 1171 #ifdef __HAVE_OLD_DISKLABEL 1172 case ODIOCSDINFO: 1173 case ODIOCWDINFO: 1174 #endif 1175 case DIOCKLABEL: 1176 case DIOCWLABEL: 1177 if ((flag & FWRITE) == 0) 1178 return (EBADF); 1179 } 1180 1181 mutex_enter(&cs->sc_dvlock); 1182 1183 /* Must be initialized for these... */ 1184 switch (cmd) { 1185 case CCDIOCCLR: 1186 case DIOCGDINFO: 1187 case DIOCGSTRATEGY: 1188 case DIOCGCACHE: 1189 case DIOCCACHESYNC: 1190 case DIOCAWEDGE: 1191 case DIOCDWEDGE: 1192 case DIOCLWEDGES: 1193 case DIOCMWEDGES: 1194 case DIOCSDINFO: 1195 case DIOCWDINFO: 1196 case DIOCGPARTINFO: 1197 case DIOCWLABEL: 1198 case DIOCKLABEL: 1199 case DIOCGDEFLABEL: 1200 #ifdef __HAVE_OLD_DISKLABEL 1201 case ODIOCGDINFO: 1202 case ODIOCSDINFO: 1203 case ODIOCWDINFO: 1204 case ODIOCGDEFLABEL: 1205 #endif 1206 if ((cs->sc_flags & CCDF_INITED) == 0) { 1207 error = ENXIO; 1208 goto out; 1209 } 1210 } 1211 1212 error = disk_ioctl(&cs->sc_dkdev, dev, cmd, data, flag, l); 1213 if (error != EPASSTHROUGH) 1214 goto out; 1215 1216 error = 0; 1217 switch (cmd) { 1218 case CCDIOCSET: 1219 if (cs->sc_flags & CCDF_INITED) { 1220 error = EBUSY; 1221 goto out; 1222 } 1223 1224 /* Validate the flags. */ 1225 if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) { 1226 error = EINVAL; 1227 goto out; 1228 } 1229 1230 if (ccio->ccio_ndisks > CCD_MAXNDISKS || 1231 ccio->ccio_ndisks == 0) { 1232 error = EINVAL; 1233 goto out; 1234 } 1235 1236 /* Fill in some important bits. */ 1237 cs->sc_ileave = ccio->ccio_ileave; 1238 cs->sc_nccdisks = ccio->ccio_ndisks; 1239 cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK; 1240 1241 /* 1242 * Allocate space for and copy in the array of 1243 * component pathnames and device numbers. 1244 */ 1245 cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP); 1246 vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP); 1247 error = copyin(ccio->ccio_disks, cpp, 1248 ccio->ccio_ndisks * sizeof(*cpp)); 1249 if (error) { 1250 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1251 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1252 goto out; 1253 } 1254 1255 #ifdef DEBUG 1256 if (ccddebug & CCDB_INIT) 1257 for (i = 0; i < ccio->ccio_ndisks; ++i) 1258 printf("ccdioctl: component %d: %p\n", 1259 i, cpp[i]); 1260 #endif 1261 1262 for (i = 0; i < ccio->ccio_ndisks; ++i) { 1263 #ifdef DEBUG 1264 if (ccddebug & CCDB_INIT) 1265 printf("ccdioctl: lookedup = %d\n", lookedup); 1266 #endif 1267 error = pathbuf_copyin(cpp[i], &pb); 1268 if (error == 0) { 1269 error = dk_lookup(pb, l, &vpp[i]); 1270 } 1271 pathbuf_destroy(pb); 1272 if (error != 0) { 1273 for (j = 0; j < lookedup; ++j) 1274 (void)vn_close(vpp[j], FREAD|FWRITE, 1275 uc); 1276 kmem_free(vpp, ccio->ccio_ndisks * 1277 sizeof(*vpp)); 1278 kmem_free(cpp, ccio->ccio_ndisks * 1279 sizeof(*cpp)); 1280 goto out; 1281 } 1282 ++lookedup; 1283 } 1284 1285 /* Attach the disk. */ 1286 disk_attach(&cs->sc_dkdev); 1287 bufq_alloc(&cs->sc_bufq, "fcfs", 0); 1288 1289 /* 1290 * Initialize the ccd. Fills in the softc for us. 1291 */ 1292 if ((error = ccdinit(cs, cpp, vpp, l)) != 0) { 1293 for (j = 0; j < lookedup; ++j) 1294 (void)vn_close(vpp[j], FREAD|FWRITE, 1295 uc); 1296 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1297 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1298 disk_detach(&cs->sc_dkdev); 1299 mutex_exit(&cs->sc_dvlock); 1300 bufq_free(cs->sc_bufq); 1301 return error; 1302 } 1303 1304 /* We can free the temporary variables now. */ 1305 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1306 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1307 1308 /* 1309 * The ccd has been successfully initialized, so 1310 * we can place it into the array. Don't try to 1311 * read the disklabel until the disk has been attached, 1312 * because space for the disklabel is allocated 1313 * in disk_attach(); 1314 */ 1315 ccio->ccio_unit = unit; 1316 ccio->ccio_size = cs->sc_size; 1317 1318 /* Try and read the disklabel. */ 1319 ccdgetdisklabel(dev); 1320 disk_set_info(NULL, &cs->sc_dkdev, NULL); 1321 1322 /* discover wedges */ 1323 mutex_exit(&cs->sc_dvlock); 1324 dkwedge_discover(&cs->sc_dkdev); 1325 return 0; 1326 1327 case CCDIOCCLR: 1328 /* 1329 * Don't unconfigure if any other partitions are open 1330 * or if both the character and block flavors of this 1331 * partition are open. 1332 */ 1333 part = DISKPART(dev); 1334 pmask = (1 << part); 1335 if ((cs->sc_dkdev.dk_openmask & ~pmask) || 1336 ((cs->sc_dkdev.dk_bopenmask & pmask) && 1337 (cs->sc_dkdev.dk_copenmask & pmask))) { 1338 error = EBUSY; 1339 goto out; 1340 } 1341 1342 /* Delete all of our wedges. */ 1343 dkwedge_delall(&cs->sc_dkdev); 1344 1345 /* Stop new I/O, wait for in-flight I/O to complete. */ 1346 mutex_enter(cs->sc_iolock); 1347 cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL); 1348 cs->sc_zap = true; 1349 while (disk_isbusy(&cs->sc_dkdev) || 1350 bufq_peek(cs->sc_bufq) != NULL || 1351 cs->sc_thread != NULL) { 1352 cv_broadcast(&cs->sc_push); 1353 (void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz); 1354 } 1355 mutex_exit(cs->sc_iolock); 1356 1357 /* 1358 * Free ccd_softc information and clear entry. 1359 */ 1360 1361 /* Close the components and free their pathnames. */ 1362 for (i = 0; i < cs->sc_nccdisks; ++i) { 1363 /* 1364 * XXX: this close could potentially fail and 1365 * cause Bad Things. Maybe we need to force 1366 * the close to happen? 1367 */ 1368 #ifdef DEBUG 1369 if (ccddebug & CCDB_VNODE) 1370 vprint("CCDIOCCLR: vnode info", 1371 cs->sc_cinfo[i].ci_vp); 1372 #endif 1373 (void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE, 1374 uc); 1375 kmem_free(cs->sc_cinfo[i].ci_path, 1376 cs->sc_cinfo[i].ci_pathlen); 1377 } 1378 1379 /* Free interleave index. */ 1380 for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) { 1381 kmem_free(cs->sc_itable[i].ii_index, 1382 cs->sc_itable[i].ii_indexsz); 1383 } 1384 1385 /* Free component info and interleave table. */ 1386 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * 1387 sizeof(struct ccdcinfo)); 1388 kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) * 1389 sizeof(struct ccdiinfo)); 1390 1391 aprint_normal("%s: detached\n", cs->sc_xname); 1392 1393 /* Detach the disk. */ 1394 disk_detach(&cs->sc_dkdev); 1395 bufq_free(cs->sc_bufq); 1396 ccdput(cs); 1397 /* Don't break, otherwise cs is read again. */ 1398 return 0; 1399 1400 case DIOCGSTRATEGY: 1401 { 1402 struct disk_strategy *dks = (void *)data; 1403 1404 mutex_enter(cs->sc_iolock); 1405 if (cs->sc_bufq != NULL) 1406 strlcpy(dks->dks_name, 1407 bufq_getstrategyname(cs->sc_bufq), 1408 sizeof(dks->dks_name)); 1409 else 1410 error = EINVAL; 1411 mutex_exit(cs->sc_iolock); 1412 dks->dks_paramlen = 0; 1413 break; 1414 } 1415 1416 case DIOCGCACHE: 1417 { 1418 int dkcache = 0; 1419 1420 /* 1421 * We pass this call down to all components and report 1422 * intersection of the flags returned by the components. 1423 * If any errors out, we return error. CCD components 1424 * can not change unless the device is unconfigured, so 1425 * device feature flags will remain static. RCE/WCE can change 1426 * of course, if set directly on underlying device. 1427 */ 1428 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1429 error = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, &j, 1430 flag, uc); 1431 if (error) 1432 break; 1433 1434 if (i == 0) 1435 dkcache = j; 1436 else 1437 dkcache = DKCACHE_COMBINE(dkcache, j); 1438 } 1439 1440 *((int *)data) = dkcache; 1441 break; 1442 } 1443 1444 case DIOCCACHESYNC: 1445 /* 1446 * We pass this call down to all components and report 1447 * the first error we encounter. 1448 */ 1449 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1450 j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data, 1451 flag, uc); 1452 if (j != 0 && error == 0) 1453 error = j; 1454 } 1455 break; 1456 1457 case DIOCWDINFO: 1458 case DIOCSDINFO: 1459 #ifdef __HAVE_OLD_DISKLABEL 1460 case ODIOCWDINFO: 1461 case ODIOCSDINFO: 1462 #endif 1463 { 1464 struct disklabel *lp; 1465 #ifdef __HAVE_OLD_DISKLABEL 1466 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1467 memset(&newlabel, 0, sizeof newlabel); 1468 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1469 lp = &newlabel; 1470 } else 1471 #endif 1472 lp = (struct disklabel *)data; 1473 1474 cs->sc_flags |= CCDF_LABELLING; 1475 1476 error = setdisklabel(cs->sc_dkdev.dk_label, 1477 lp, 0, cs->sc_dkdev.dk_cpulabel); 1478 if (error == 0) { 1479 if (cmd == DIOCWDINFO 1480 #ifdef __HAVE_OLD_DISKLABEL 1481 || cmd == ODIOCWDINFO 1482 #endif 1483 ) 1484 error = writedisklabel(CCDLABELDEV(dev), 1485 ccdstrategy, cs->sc_dkdev.dk_label, 1486 cs->sc_dkdev.dk_cpulabel); 1487 } 1488 1489 cs->sc_flags &= ~CCDF_LABELLING; 1490 break; 1491 } 1492 1493 case DIOCKLABEL: 1494 if (*(int *)data != 0) 1495 cs->sc_flags |= CCDF_KLABEL; 1496 else 1497 cs->sc_flags &= ~CCDF_KLABEL; 1498 break; 1499 1500 case DIOCWLABEL: 1501 if (*(int *)data != 0) 1502 cs->sc_flags |= CCDF_WLABEL; 1503 else 1504 cs->sc_flags &= ~CCDF_WLABEL; 1505 break; 1506 1507 case DIOCGDEFLABEL: 1508 ccdgetdefaultlabel(cs, (struct disklabel *)data); 1509 break; 1510 1511 #ifdef __HAVE_OLD_DISKLABEL 1512 case ODIOCGDEFLABEL: 1513 ccdgetdefaultlabel(cs, &newlabel); 1514 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1515 return ENOTTY; 1516 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1517 break; 1518 #endif 1519 1520 default: 1521 error = ENOTTY; 1522 } 1523 1524 out: 1525 mutex_exit(&cs->sc_dvlock); 1526 return (error); 1527 } 1528 1529 static int 1530 ccdsize(dev_t dev) 1531 { 1532 struct ccd_softc *cs; 1533 struct disklabel *lp; 1534 int part, unit, omask, size; 1535 1536 unit = ccdunit(dev); 1537 if ((cs = ccdget(unit, 0)) == NULL) 1538 return -1; 1539 1540 if ((cs->sc_flags & CCDF_INITED) == 0) 1541 return (-1); 1542 1543 part = DISKPART(dev); 1544 omask = cs->sc_dkdev.dk_openmask & (1 << part); 1545 lp = cs->sc_dkdev.dk_label; 1546 1547 if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp)) 1548 return (-1); 1549 1550 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1551 size = -1; 1552 else 1553 size = lp->d_partitions[part].p_size * 1554 (lp->d_secsize / DEV_BSIZE); 1555 1556 if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp)) 1557 return (-1); 1558 1559 return (size); 1560 } 1561 1562 static void 1563 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp) 1564 { 1565 struct ccdgeom *ccg = &cs->sc_geom; 1566 1567 memset(lp, 0, sizeof(*lp)); 1568 1569 if (cs->sc_size > UINT32_MAX) 1570 lp->d_secperunit = UINT32_MAX; 1571 else 1572 lp->d_secperunit = cs->sc_size; 1573 lp->d_secsize = ccg->ccg_secsize; 1574 lp->d_nsectors = ccg->ccg_nsectors; 1575 lp->d_ntracks = ccg->ccg_ntracks; 1576 lp->d_ncylinders = ccg->ccg_ncylinders; 1577 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1578 1579 strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename)); 1580 lp->d_type = DKTYPE_CCD; 1581 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1582 lp->d_rpm = 3600; 1583 lp->d_interleave = 1; 1584 lp->d_flags = 0; 1585 1586 lp->d_partitions[RAW_PART].p_offset = 0; 1587 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit; 1588 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 1589 lp->d_npartitions = RAW_PART + 1; 1590 1591 lp->d_magic = DISKMAGIC; 1592 lp->d_magic2 = DISKMAGIC; 1593 lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label); 1594 } 1595 1596 /* 1597 * Read the disklabel from the ccd. If one is not present, fake one 1598 * up. 1599 */ 1600 static void 1601 ccdgetdisklabel(dev_t dev) 1602 { 1603 int unit = ccdunit(dev); 1604 struct ccd_softc *cs; 1605 const char *errstring; 1606 struct disklabel *lp; 1607 struct cpu_disklabel *clp; 1608 1609 if ((cs = ccdget(unit, 0)) == NULL) 1610 return; 1611 lp = cs->sc_dkdev.dk_label; 1612 clp = cs->sc_dkdev.dk_cpulabel; 1613 KASSERT(mutex_owned(&cs->sc_dvlock)); 1614 1615 memset(clp, 0, sizeof(*clp)); 1616 1617 ccdgetdefaultlabel(cs, lp); 1618 1619 /* 1620 * Call the generic disklabel extraction routine. 1621 */ 1622 cs->sc_flags |= CCDF_RLABEL; 1623 if ((cs->sc_flags & CCDF_NOLABEL) != 0) 1624 errstring = "CCDF_NOLABEL set; ignoring on-disk label"; 1625 else 1626 errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy, 1627 cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel); 1628 if (errstring) 1629 ccdmakedisklabel(cs); 1630 else { 1631 int i; 1632 struct partition *pp; 1633 1634 /* 1635 * Sanity check whether the found disklabel is valid. 1636 * 1637 * This is necessary since total size of ccd may vary 1638 * when an interleave is changed even though exactly 1639 * same componets are used, and old disklabel may used 1640 * if that is found. 1641 */ 1642 if (lp->d_secperunit < UINT32_MAX ? 1643 lp->d_secperunit != cs->sc_size : 1644 lp->d_secperunit > cs->sc_size) 1645 printf("WARNING: %s: " 1646 "total sector size in disklabel (%ju) != " 1647 "the size of ccd (%ju)\n", cs->sc_xname, 1648 (uintmax_t)lp->d_secperunit, 1649 (uintmax_t)cs->sc_size); 1650 for (i = 0; i < lp->d_npartitions; i++) { 1651 pp = &lp->d_partitions[i]; 1652 if (pp->p_offset + pp->p_size > cs->sc_size) 1653 printf("WARNING: %s: end of partition `%c' " 1654 "exceeds the size of ccd (%ju)\n", 1655 cs->sc_xname, 'a' + i, (uintmax_t)cs->sc_size); 1656 } 1657 } 1658 1659 #ifdef DEBUG 1660 /* It's actually extremely common to have unlabeled ccds. */ 1661 if (ccddebug & CCDB_LABEL) 1662 if (errstring != NULL) 1663 printf("%s: %s\n", cs->sc_xname, errstring); 1664 #endif 1665 1666 /* In-core label now valid. */ 1667 cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL; 1668 } 1669 1670 /* 1671 * Take care of things one might want to take care of in the event 1672 * that a disklabel isn't present. 1673 */ 1674 static void 1675 ccdmakedisklabel(struct ccd_softc *cs) 1676 { 1677 struct disklabel *lp = cs->sc_dkdev.dk_label; 1678 1679 /* 1680 * For historical reasons, if there's no disklabel present 1681 * the raw partition must be marked FS_BSDFFS. 1682 */ 1683 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 1684 1685 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 1686 1687 lp->d_checksum = dkcksum(lp); 1688 } 1689 1690 #ifdef DEBUG 1691 static void 1692 printiinfo(struct ccdiinfo *ii) 1693 { 1694 int ix, i; 1695 1696 for (ix = 0; ii->ii_ndisk; ix++, ii++) { 1697 printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64, 1698 ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff); 1699 for (i = 0; i < ii->ii_ndisk; i++) 1700 printf(" %d", ii->ii_index[i]); 1701 printf("\n"); 1702 } 1703 } 1704 #endif 1705 1706 MODULE(MODULE_CLASS_DRIVER, ccd, "dk_subr"); 1707 1708 static int 1709 ccd_modcmd(modcmd_t cmd, void *arg) 1710 { 1711 int error = 0; 1712 #ifdef _MODULE 1713 int bmajor = -1, cmajor = -1; 1714 #endif 1715 1716 1717 switch (cmd) { 1718 case MODULE_CMD_INIT: 1719 #ifdef _MODULE 1720 ccdattach(0); 1721 1722 error = devsw_attach("ccd", &ccd_bdevsw, &bmajor, 1723 &ccd_cdevsw, &cmajor); 1724 sysctl_kern_ccd_setup(&ccd_clog); 1725 #endif 1726 break; 1727 1728 case MODULE_CMD_FINI: 1729 #ifdef _MODULE 1730 mutex_enter(&ccd_lock); 1731 if (ccd_nactive) { 1732 mutex_exit(&ccd_lock); 1733 error = EBUSY; 1734 } else { 1735 mutex_exit(&ccd_lock); 1736 error = devsw_detach(&ccd_bdevsw, &ccd_cdevsw); 1737 ccddetach(); 1738 } 1739 sysctl_teardown(&ccd_clog); 1740 #endif 1741 break; 1742 1743 case MODULE_CMD_STAT: 1744 return ENOTTY; 1745 1746 default: 1747 return ENOTTY; 1748 } 1749 1750 return error; 1751 } 1752 1753 static int 1754 ccd_units_sysctl(SYSCTLFN_ARGS) 1755 { 1756 struct sysctlnode node; 1757 struct ccd_softc *sc; 1758 int error, i, nccd, *units; 1759 size_t size; 1760 1761 nccd = 0; 1762 mutex_enter(&ccd_lock); 1763 LIST_FOREACH(sc, &ccds, sc_link) 1764 nccd++; 1765 mutex_exit(&ccd_lock); 1766 1767 if (nccd != 0) { 1768 size = nccd * sizeof(*units); 1769 units = kmem_zalloc(size, KM_SLEEP); 1770 if (units == NULL) 1771 return ENOMEM; 1772 1773 i = 0; 1774 mutex_enter(&ccd_lock); 1775 LIST_FOREACH(sc, &ccds, sc_link) { 1776 if (i >= nccd) 1777 break; 1778 units[i] = sc->sc_unit; 1779 } 1780 mutex_exit(&ccd_lock); 1781 } else { 1782 units = NULL; 1783 size = 0; 1784 } 1785 1786 node = *rnode; 1787 node.sysctl_data = units; 1788 node.sysctl_size = size; 1789 1790 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1791 if (units) 1792 kmem_free(units, size); 1793 return error; 1794 } 1795 1796 static int 1797 ccd_info_sysctl(SYSCTLFN_ARGS) 1798 { 1799 struct sysctlnode node; 1800 struct ccddiskinfo ccd; 1801 struct ccd_softc *sc; 1802 int unit; 1803 1804 if (newp == NULL || newlen != sizeof(int)) 1805 return EINVAL; 1806 1807 unit = *(const int *)newp; 1808 newp = NULL; 1809 newlen = 0; 1810 ccd.ccd_ndisks = ~0; 1811 mutex_enter(&ccd_lock); 1812 LIST_FOREACH(sc, &ccds, sc_link) { 1813 if (sc->sc_unit == unit) { 1814 ccd.ccd_ileave = sc->sc_ileave; 1815 ccd.ccd_size = sc->sc_size; 1816 ccd.ccd_ndisks = sc->sc_nccdisks; 1817 ccd.ccd_flags = sc->sc_flags; 1818 break; 1819 } 1820 } 1821 mutex_exit(&ccd_lock); 1822 1823 if (ccd.ccd_ndisks == ~0) 1824 return ENOENT; 1825 1826 node = *rnode; 1827 node.sysctl_data = &ccd; 1828 node.sysctl_size = sizeof(ccd); 1829 1830 return sysctl_lookup(SYSCTLFN_CALL(&node)); 1831 } 1832 1833 static int 1834 ccd_components_sysctl(SYSCTLFN_ARGS) 1835 { 1836 struct sysctlnode node; 1837 int error, unit; 1838 size_t size; 1839 char *names, *p, *ep; 1840 struct ccd_softc *sc; 1841 1842 if (newp == NULL || newlen != sizeof(int)) 1843 return EINVAL; 1844 1845 size = 0; 1846 unit = *(const int *)newp; 1847 newp = NULL; 1848 newlen = 0; 1849 mutex_enter(&ccd_lock); 1850 LIST_FOREACH(sc, &ccds, sc_link) 1851 if (sc->sc_unit == unit) { 1852 for (size_t i = 0; i < sc->sc_nccdisks; i++) 1853 size += strlen(sc->sc_cinfo[i].ci_path) + 1; 1854 break; 1855 } 1856 mutex_exit(&ccd_lock); 1857 1858 if (size == 0) 1859 return ENOENT; 1860 names = kmem_zalloc(size, KM_SLEEP); 1861 if (names == NULL) 1862 return ENOMEM; 1863 1864 p = names; 1865 ep = names + size; 1866 mutex_enter(&ccd_lock); 1867 LIST_FOREACH(sc, &ccds, sc_link) 1868 if (sc->sc_unit == unit) { 1869 for (size_t i = 0; i < sc->sc_nccdisks; i++) { 1870 char *d = sc->sc_cinfo[i].ci_path; 1871 while (p < ep && (*p++ = *d++) != '\0') 1872 continue; 1873 } 1874 break; 1875 } 1876 mutex_exit(&ccd_lock); 1877 1878 node = *rnode; 1879 node.sysctl_data = names; 1880 node.sysctl_size = ep - names; 1881 1882 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1883 kmem_free(names, size); 1884 return error; 1885 } 1886 1887 SYSCTL_SETUP(sysctl_kern_ccd_setup, "sysctl kern.ccd subtree setup") 1888 { 1889 const struct sysctlnode *node = NULL; 1890 1891 sysctl_createv(clog, 0, NULL, &node, 1892 CTLFLAG_PERMANENT, 1893 CTLTYPE_NODE, "ccd", 1894 SYSCTL_DESCR("ConCatenated Disk state"), 1895 NULL, 0, NULL, 0, 1896 CTL_KERN, CTL_CREATE, CTL_EOL); 1897 1898 if (node == NULL) 1899 return; 1900 1901 sysctl_createv(clog, 0, &node, NULL, 1902 CTLFLAG_PERMANENT | CTLFLAG_READONLY, 1903 CTLTYPE_STRUCT, "units", 1904 SYSCTL_DESCR("List of ccd unit numbers"), 1905 ccd_units_sysctl, 0, NULL, 0, 1906 CTL_CREATE, CTL_EOL); 1907 sysctl_createv(clog, 0, &node, NULL, 1908 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1909 CTLTYPE_STRUCT, "info", 1910 SYSCTL_DESCR("Information about a CCD unit"), 1911 ccd_info_sysctl, 0, NULL, 0, 1912 CTL_CREATE, CTL_EOL); 1913 sysctl_createv(clog, 0, &node, NULL, 1914 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1915 CTLTYPE_STRUCT, "components", 1916 SYSCTL_DESCR("Information about CCD components"), 1917 ccd_components_sysctl, 0, NULL, 0, 1918 CTL_CREATE, CTL_EOL); 1919 } 1920
Indexes created Thu Oct 02 14:10:14 GMT 2025