ccd.c revision 1.175.2.2
1 /* $NetBSD: ccd.c,v 1.175.2.2 2018/03/20 08:11:25 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 1.175.2.2 2018/03/20 08:11:25 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 #include <sys/compat_stub.h> 123 124 #include <uvm/uvm_extern.h> 125 126 #include <dev/ccdvar.h> 127 #include <dev/dkvar.h> 128 129 #include <miscfs/specfs/specdev.h> /* for v_rdev */ 130 131 #include "ioconf.h" 132 133 #if defined(CCDDEBUG) && !defined(DEBUG) 134 #define DEBUG 135 #endif 136 137 #ifdef DEBUG 138 #define CCDB_FOLLOW 0x01 139 #define CCDB_INIT 0x02 140 #define CCDB_IO 0x04 141 #define CCDB_LABEL 0x08 142 #define CCDB_VNODE 0x10 143 int ccddebug = 0x00; 144 #endif 145 146 #define ccdunit(x) DISKUNIT(x) 147 148 struct ccdbuf { 149 struct buf cb_buf; /* new I/O buf */ 150 struct buf *cb_obp; /* ptr. to original I/O buf */ 151 struct ccd_softc *cb_sc; /* pointer to ccd softc */ 152 int cb_comp; /* target component */ 153 SIMPLEQ_ENTRY(ccdbuf) cb_q; /* fifo of component buffers */ 154 }; 155 156 /* component buffer pool */ 157 static pool_cache_t ccd_cache; 158 159 #define CCD_GETBUF() pool_cache_get(ccd_cache, PR_WAITOK) 160 #define CCD_PUTBUF(cbp) pool_cache_put(ccd_cache, cbp) 161 162 #define CCDLABELDEV(dev) \ 163 (MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART)) 164 165 /* called by main() at boot time */ 166 void ccddetach(void); 167 168 /* called by biodone() at interrupt time */ 169 static void ccdiodone(struct buf *); 170 171 static void ccdinterleave(struct ccd_softc *); 172 static int ccdinit(struct ccd_softc *, char **, struct vnode **, 173 struct lwp *); 174 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *, 175 daddr_t, void *, long); 176 static void ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *); 177 static void ccdgetdisklabel(dev_t); 178 static void ccdmakedisklabel(struct ccd_softc *); 179 static void ccdstart(struct ccd_softc *); 180 static void ccdthread(void *); 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; 222 223 #ifdef _MODULE 224 static struct sysctllog *ccd_clog; 225 #endif 226 227 SYSCTL_SETUP_PROTO(sysctl_kern_ccd_setup); 228 229 static struct ccd_softc * 230 ccdcreate(int unit) { 231 struct ccd_softc *sc = kmem_zalloc(sizeof(*sc), KM_SLEEP); 232 233 /* Initialize per-softc structures. */ 234 snprintf(sc->sc_xname, sizeof(sc->sc_xname), "ccd%d", unit); 235 sc->sc_unit = unit; 236 mutex_init(&sc->sc_dvlock, MUTEX_DEFAULT, IPL_NONE); 237 sc->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 238 cv_init(&sc->sc_stop, "ccdstop"); 239 cv_init(&sc->sc_push, "ccdthr"); 240 disk_init(&sc->sc_dkdev, sc->sc_xname, NULL); /* XXX */ 241 return sc; 242 } 243 244 static void 245 ccddestroy(struct ccd_softc *sc) { 246 mutex_obj_free(sc->sc_iolock); 247 mutex_exit(&sc->sc_dvlock); 248 mutex_destroy(&sc->sc_dvlock); 249 cv_destroy(&sc->sc_stop); 250 cv_destroy(&sc->sc_push); 251 disk_destroy(&sc->sc_dkdev); 252 kmem_free(sc, sizeof(*sc)); 253 } 254 255 static struct ccd_softc * 256 ccdget(int unit, int make) { 257 struct ccd_softc *sc; 258 if (unit < 0) { 259 #ifdef DIAGNOSTIC 260 panic("%s: unit %d!", __func__, unit); 261 #endif 262 return NULL; 263 } 264 mutex_enter(&ccd_lock); 265 LIST_FOREACH(sc, &ccds, sc_link) { 266 if (sc->sc_unit == unit) { 267 mutex_exit(&ccd_lock); 268 return sc; 269 } 270 } 271 mutex_exit(&ccd_lock); 272 if (!make) 273 return NULL; 274 if ((sc = ccdcreate(unit)) == NULL) 275 return NULL; 276 mutex_enter(&ccd_lock); 277 LIST_INSERT_HEAD(&ccds, sc, sc_link); 278 mutex_exit(&ccd_lock); 279 return sc; 280 } 281 282 static void 283 ccdput(struct ccd_softc *sc) { 284 mutex_enter(&ccd_lock); 285 LIST_REMOVE(sc, sc_link); 286 mutex_exit(&ccd_lock); 287 ccddestroy(sc); 288 } 289 290 /* 291 * Called by main() during pseudo-device attachment. All we need 292 * to do is allocate enough space for devices to be configured later. 293 */ 294 void 295 ccdattach(int num) 296 { 297 mutex_init(&ccd_lock, MUTEX_DEFAULT, IPL_NONE); 298 299 /* Initialize the component buffer pool. */ 300 ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0, 301 0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL); 302 } 303 304 void 305 ccddetach(void) 306 { 307 pool_cache_destroy(ccd_cache); 308 mutex_destroy(&ccd_lock); 309 } 310 311 static int 312 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp, 313 struct lwp *l) 314 { 315 struct ccdcinfo *ci = NULL; 316 int ix; 317 struct ccdgeom *ccg = &cs->sc_geom; 318 char *tmppath; 319 int error, path_alloced; 320 uint64_t psize, minsize; 321 unsigned secsize, maxsecsize; 322 struct disk_geom *dg; 323 324 #ifdef DEBUG 325 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 326 printf("%s: ccdinit\n", cs->sc_xname); 327 #endif 328 329 /* Allocate space for the component info. */ 330 cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo), 331 KM_SLEEP); 332 tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 333 334 cs->sc_size = 0; 335 336 /* 337 * Verify that each component piece exists and record 338 * relevant information about it. 339 */ 340 maxsecsize = 0; 341 minsize = 0; 342 for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) { 343 ci = &cs->sc_cinfo[ix]; 344 ci->ci_vp = vpp[ix]; 345 346 /* 347 * Copy in the pathname of the component. 348 */ 349 memset(tmppath, 0, MAXPATHLEN); /* sanity */ 350 error = copyinstr(cpaths[ix], tmppath, 351 MAXPATHLEN, &ci->ci_pathlen); 352 if (ci->ci_pathlen == 0) 353 error = EINVAL; 354 if (error) { 355 #ifdef DEBUG 356 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 357 printf("%s: can't copy path, error = %d\n", 358 cs->sc_xname, error); 359 #endif 360 goto out; 361 } 362 ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP); 363 memcpy(ci->ci_path, tmppath, ci->ci_pathlen); 364 path_alloced++; 365 366 /* 367 * XXX: Cache the component's dev_t. 368 */ 369 ci->ci_dev = vpp[ix]->v_rdev; 370 371 /* 372 * Get partition information for the component. 373 */ 374 error = getdisksize(vpp[ix], &psize, &secsize); 375 if (error) { 376 #ifdef DEBUG 377 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 378 printf("%s: %s: disksize failed, error = %d\n", 379 cs->sc_xname, ci->ci_path, error); 380 #endif 381 goto out; 382 } 383 384 /* 385 * Calculate the size, truncating to an interleave 386 * boundary if necessary. 387 */ 388 maxsecsize = secsize > maxsecsize ? secsize : maxsecsize; 389 if (cs->sc_ileave > 1) 390 psize -= psize % cs->sc_ileave; 391 392 if (psize == 0) { 393 #ifdef DEBUG 394 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 395 printf("%s: %s: size == 0\n", 396 cs->sc_xname, ci->ci_path); 397 #endif 398 error = ENODEV; 399 goto out; 400 } 401 402 if (minsize == 0 || psize < minsize) 403 minsize = psize; 404 ci->ci_size = psize; 405 cs->sc_size += psize; 406 } 407 408 /* 409 * Don't allow the interleave to be smaller than 410 * the biggest component sector. 411 */ 412 if ((cs->sc_ileave > 0) && 413 (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) { 414 #ifdef DEBUG 415 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 416 printf("%s: interleave must be at least %d\n", 417 cs->sc_xname, (maxsecsize / DEV_BSIZE)); 418 #endif 419 error = EINVAL; 420 goto out; 421 } 422 423 /* 424 * If uniform interleave is desired set all sizes to that of 425 * the smallest component. 426 */ 427 if (cs->sc_flags & CCDF_UNIFORM) { 428 for (ci = cs->sc_cinfo; 429 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 430 ci->ci_size = minsize; 431 432 cs->sc_size = cs->sc_nccdisks * minsize; 433 } 434 435 /* 436 * Construct the interleave table. 437 */ 438 ccdinterleave(cs); 439 440 /* 441 * Create pseudo-geometry based on 1MB cylinders. It's 442 * pretty close. 443 */ 444 ccg->ccg_secsize = DEV_BSIZE; 445 ccg->ccg_ntracks = 1; 446 ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize); 447 ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors; 448 449 dg = &cs->sc_dkdev.dk_geom; 450 memset(dg, 0, sizeof(*dg)); 451 dg->dg_secperunit = cs->sc_size; 452 dg->dg_secsize = ccg->ccg_secsize; 453 dg->dg_nsectors = ccg->ccg_nsectors; 454 dg->dg_ntracks = ccg->ccg_ntracks; 455 dg->dg_ncylinders = ccg->ccg_ncylinders; 456 457 if (cs->sc_ileave > 0) 458 aprint_normal("%s: Interleaving %d component%s " 459 "(%d block interleave)\n", cs->sc_xname, 460 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""), 461 cs->sc_ileave); 462 else 463 aprint_normal("%s: Concatenating %d component%s\n", 464 cs->sc_xname, 465 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : "")); 466 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 467 ci = &cs->sc_cinfo[ix]; 468 aprint_normal("%s: %s (%ju blocks)\n", cs->sc_xname, 469 ci->ci_path, (uintmax_t)ci->ci_size); 470 } 471 aprint_normal("%s: total %ju blocks\n", cs->sc_xname, cs->sc_size); 472 473 /* 474 * Create thread to handle deferred I/O. 475 */ 476 cs->sc_zap = false; 477 error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread, 478 cs, &cs->sc_thread, "%s", cs->sc_xname); 479 if (error) { 480 printf("ccdinit: can't create thread: %d\n", error); 481 goto out; 482 } 483 484 /* 485 * Only now that everything is set up can we enable the device. 486 */ 487 mutex_enter(cs->sc_iolock); 488 cs->sc_flags |= CCDF_INITED; 489 mutex_exit(cs->sc_iolock); 490 kmem_free(tmppath, MAXPATHLEN); 491 return (0); 492 493 out: 494 for (ix = 0; ix < path_alloced; ix++) { 495 kmem_free(cs->sc_cinfo[ix].ci_path, 496 cs->sc_cinfo[ix].ci_pathlen); 497 } 498 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo)); 499 kmem_free(tmppath, MAXPATHLEN); 500 return (error); 501 } 502 503 static void 504 ccdinterleave(struct ccd_softc *cs) 505 { 506 struct ccdcinfo *ci, *smallci; 507 struct ccdiinfo *ii; 508 daddr_t bn, lbn; 509 int ix; 510 u_long size; 511 512 #ifdef DEBUG 513 if (ccddebug & CCDB_INIT) 514 printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave); 515 #endif 516 /* 517 * Allocate an interleave table. 518 * Chances are this is too big, but we don't care. 519 */ 520 size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo); 521 cs->sc_itable = kmem_zalloc(size, KM_SLEEP); 522 523 /* 524 * Trivial case: no interleave (actually interleave of disk size). 525 * Each table entry represents a single component in its entirety. 526 */ 527 if (cs->sc_ileave == 0) { 528 bn = 0; 529 ii = cs->sc_itable; 530 531 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 532 /* Allocate space for ii_index. */ 533 ii->ii_indexsz = sizeof(int); 534 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 535 ii->ii_ndisk = 1; 536 ii->ii_startblk = bn; 537 ii->ii_startoff = 0; 538 ii->ii_index[0] = ix; 539 bn += cs->sc_cinfo[ix].ci_size; 540 ii++; 541 } 542 ii->ii_ndisk = 0; 543 #ifdef DEBUG 544 if (ccddebug & CCDB_INIT) 545 printiinfo(cs->sc_itable); 546 #endif 547 return; 548 } 549 550 /* 551 * The following isn't fast or pretty; it doesn't have to be. 552 */ 553 size = 0; 554 bn = lbn = 0; 555 for (ii = cs->sc_itable; ; ii++) { 556 /* Allocate space for ii_index. */ 557 ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks; 558 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 559 560 /* 561 * Locate the smallest of the remaining components 562 */ 563 smallci = NULL; 564 for (ci = cs->sc_cinfo; 565 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 566 if (ci->ci_size > size && 567 (smallci == NULL || 568 ci->ci_size < smallci->ci_size)) 569 smallci = ci; 570 571 /* 572 * Nobody left, all done 573 */ 574 if (smallci == NULL) { 575 ii->ii_ndisk = 0; 576 break; 577 } 578 579 /* 580 * Record starting logical block and component offset 581 */ 582 ii->ii_startblk = bn / cs->sc_ileave; 583 ii->ii_startoff = lbn; 584 585 /* 586 * Determine how many disks take part in this interleave 587 * and record their indices. 588 */ 589 ix = 0; 590 for (ci = cs->sc_cinfo; 591 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 592 if (ci->ci_size >= smallci->ci_size) 593 ii->ii_index[ix++] = ci - cs->sc_cinfo; 594 ii->ii_ndisk = ix; 595 bn += ix * (smallci->ci_size - size); 596 lbn = smallci->ci_size / cs->sc_ileave; 597 size = smallci->ci_size; 598 } 599 #ifdef DEBUG 600 if (ccddebug & CCDB_INIT) 601 printiinfo(cs->sc_itable); 602 #endif 603 } 604 605 /* ARGSUSED */ 606 static int 607 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l) 608 { 609 int unit = ccdunit(dev); 610 struct ccd_softc *cs; 611 struct disklabel *lp; 612 int error = 0, part, pmask; 613 614 #ifdef DEBUG 615 if (ccddebug & CCDB_FOLLOW) 616 printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags); 617 #endif 618 if ((cs = ccdget(unit, 1)) == NULL) 619 return ENXIO; 620 621 mutex_enter(&cs->sc_dvlock); 622 623 lp = cs->sc_dkdev.dk_label; 624 625 part = DISKPART(dev); 626 pmask = (1 << part); 627 628 /* 629 * If we're initialized, check to see if there are any other 630 * open partitions. If not, then it's safe to update 631 * the in-core disklabel. Only read the disklabel if it is 632 * not already valid. 633 */ 634 if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED && 635 cs->sc_dkdev.dk_openmask == 0) 636 ccdgetdisklabel(dev); 637 638 /* Check that the partition exists. */ 639 if (part != RAW_PART) { 640 if (((cs->sc_flags & CCDF_INITED) == 0) || 641 ((part >= lp->d_npartitions) || 642 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 643 error = ENXIO; 644 goto done; 645 } 646 } 647 648 /* Prevent our unit from being unconfigured while open. */ 649 switch (fmt) { 650 case S_IFCHR: 651 cs->sc_dkdev.dk_copenmask |= pmask; 652 break; 653 654 case S_IFBLK: 655 cs->sc_dkdev.dk_bopenmask |= pmask; 656 break; 657 } 658 cs->sc_dkdev.dk_openmask = 659 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 660 661 done: 662 mutex_exit(&cs->sc_dvlock); 663 return (error); 664 } 665 666 /* ARGSUSED */ 667 static int 668 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l) 669 { 670 int unit = ccdunit(dev); 671 struct ccd_softc *cs; 672 int part; 673 674 #ifdef DEBUG 675 if (ccddebug & CCDB_FOLLOW) 676 printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags); 677 #endif 678 679 if ((cs = ccdget(unit, 0)) == NULL) 680 return ENXIO; 681 682 mutex_enter(&cs->sc_dvlock); 683 684 part = DISKPART(dev); 685 686 /* ...that much closer to allowing unconfiguration... */ 687 switch (fmt) { 688 case S_IFCHR: 689 cs->sc_dkdev.dk_copenmask &= ~(1 << part); 690 break; 691 692 case S_IFBLK: 693 cs->sc_dkdev.dk_bopenmask &= ~(1 << part); 694 break; 695 } 696 cs->sc_dkdev.dk_openmask = 697 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 698 699 if (cs->sc_dkdev.dk_openmask == 0) { 700 if ((cs->sc_flags & CCDF_KLABEL) == 0) 701 cs->sc_flags &= ~CCDF_VLABEL; 702 } 703 704 mutex_exit(&cs->sc_dvlock); 705 return (0); 706 } 707 708 static bool 709 ccdbackoff(struct ccd_softc *cs) 710 { 711 712 /* XXX Arbitrary, should be a uvm call. */ 713 return uvmexp.free < (uvmexp.freemin >> 1) && 714 disk_isbusy(&cs->sc_dkdev); 715 } 716 717 static void 718 ccdthread(void *cookie) 719 { 720 struct ccd_softc *cs; 721 722 cs = cookie; 723 724 #ifdef DEBUG 725 if (ccddebug & CCDB_FOLLOW) 726 printf("ccdthread: hello\n"); 727 #endif 728 729 mutex_enter(cs->sc_iolock); 730 while (__predict_true(!cs->sc_zap)) { 731 if (bufq_peek(cs->sc_bufq) == NULL) { 732 /* Nothing to do. */ 733 cv_wait(&cs->sc_push, cs->sc_iolock); 734 continue; 735 } 736 if (ccdbackoff(cs)) { 737 /* Wait for memory to become available. */ 738 (void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1); 739 continue; 740 } 741 #ifdef DEBUG 742 if (ccddebug & CCDB_FOLLOW) 743 printf("ccdthread: dispatching I/O\n"); 744 #endif 745 ccdstart(cs); 746 mutex_enter(cs->sc_iolock); 747 } 748 cs->sc_thread = NULL; 749 mutex_exit(cs->sc_iolock); 750 #ifdef DEBUG 751 if (ccddebug & CCDB_FOLLOW) 752 printf("ccdthread: goodbye\n"); 753 #endif 754 kthread_exit(0); 755 } 756 757 static void 758 ccdstrategy(struct buf *bp) 759 { 760 int unit = ccdunit(bp->b_dev); 761 struct ccd_softc *cs; 762 if ((cs = ccdget(unit, 0)) == NULL) 763 return; 764 765 /* Must be open or reading label. */ 766 KASSERT(cs->sc_dkdev.dk_openmask != 0 || 767 (cs->sc_flags & CCDF_RLABEL) != 0); 768 769 mutex_enter(cs->sc_iolock); 770 /* Synchronize with device init/uninit. */ 771 if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) { 772 mutex_exit(cs->sc_iolock); 773 #ifdef DEBUG 774 if (ccddebug & CCDB_FOLLOW) 775 printf("ccdstrategy: unit %d: not inited\n", unit); 776 #endif 777 bp->b_error = ENXIO; 778 bp->b_resid = bp->b_bcount; 779 biodone(bp); 780 return; 781 } 782 783 /* Defer to thread if system is low on memory. */ 784 bufq_put(cs->sc_bufq, bp); 785 if (__predict_false(ccdbackoff(cs))) { 786 mutex_exit(cs->sc_iolock); 787 #ifdef DEBUG 788 if (ccddebug & CCDB_FOLLOW) 789 printf("ccdstrategy: holding off on I/O\n"); 790 #endif 791 return; 792 } 793 ccdstart(cs); 794 } 795 796 static void 797 ccdstart(struct ccd_softc *cs) 798 { 799 daddr_t blkno; 800 int wlabel; 801 struct disklabel *lp; 802 long bcount, rcount; 803 struct ccdbuf *cbp; 804 char *addr; 805 daddr_t bn; 806 vnode_t *vp; 807 buf_t *bp; 808 809 KASSERT(mutex_owned(cs->sc_iolock)); 810 811 bp = bufq_get(cs->sc_bufq); 812 KASSERT(bp != NULL); 813 814 disk_busy(&cs->sc_dkdev); 815 816 #ifdef DEBUG 817 if (ccddebug & CCDB_FOLLOW) 818 printf("ccdstart(%s, %p)\n", cs->sc_xname, bp); 819 #endif 820 821 /* If it's a nil transfer, wake up the top half now. */ 822 if (bp->b_bcount == 0) 823 goto done; 824 825 lp = cs->sc_dkdev.dk_label; 826 827 /* 828 * Do bounds checking and adjust transfer. If there's an 829 * error, the bounds check will flag that for us. Convert 830 * the partition relative block number to an absolute. 831 */ 832 blkno = bp->b_blkno; 833 wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING); 834 if (DISKPART(bp->b_dev) != RAW_PART) { 835 if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0) 836 goto done; 837 blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset; 838 } 839 mutex_exit(cs->sc_iolock); 840 bp->b_rawblkno = blkno; 841 842 /* Allocate the component buffers and start I/O! */ 843 bp->b_resid = bp->b_bcount; 844 bn = bp->b_rawblkno; 845 addr = bp->b_data; 846 for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) { 847 cbp = ccdbuffer(cs, bp, bn, addr, bcount); 848 rcount = cbp->cb_buf.b_bcount; 849 bn += btodb(rcount); 850 addr += rcount; 851 vp = cbp->cb_buf.b_vp; 852 if ((cbp->cb_buf.b_flags & B_READ) == 0) { 853 mutex_enter(vp->v_interlock); 854 vp->v_numoutput++; 855 mutex_exit(vp->v_interlock); 856 } 857 (void)VOP_STRATEGY(vp, &cbp->cb_buf); 858 } 859 return; 860 861 done: 862 disk_unbusy(&cs->sc_dkdev, 0, 0); 863 cv_broadcast(&cs->sc_stop); 864 cv_broadcast(&cs->sc_push); 865 mutex_exit(cs->sc_iolock); 866 bp->b_resid = bp->b_bcount; 867 biodone(bp); 868 } 869 870 /* 871 * Build a component buffer header. 872 */ 873 static struct ccdbuf * 874 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr, 875 long bcount) 876 { 877 struct ccdcinfo *ci; 878 struct ccdbuf *cbp; 879 daddr_t cbn, cboff; 880 u_int64_t cbc; 881 int ccdisk; 882 883 #ifdef DEBUG 884 if (ccddebug & CCDB_IO) 885 printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n", 886 cs, bp, bn, addr, bcount); 887 #endif 888 /* 889 * Determine which component bn falls in. 890 */ 891 cbn = bn; 892 cboff = 0; 893 894 /* 895 * Serially concatenated 896 */ 897 if (cs->sc_ileave == 0) { 898 daddr_t sblk; 899 900 sblk = 0; 901 for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk]; 902 cbn >= sblk + ci->ci_size; 903 ccdisk++, ci = &cs->sc_cinfo[ccdisk]) 904 sblk += ci->ci_size; 905 cbn -= sblk; 906 } 907 /* 908 * Interleaved 909 */ 910 else { 911 struct ccdiinfo *ii; 912 int off; 913 914 cboff = cbn % cs->sc_ileave; 915 cbn /= cs->sc_ileave; 916 for (ii = cs->sc_itable; ii->ii_ndisk; ii++) 917 if (ii->ii_startblk > cbn) 918 break; 919 ii--; 920 off = cbn - ii->ii_startblk; 921 if (ii->ii_ndisk == 1) { 922 ccdisk = ii->ii_index[0]; 923 cbn = ii->ii_startoff + off; 924 } else { 925 ccdisk = ii->ii_index[off % ii->ii_ndisk]; 926 cbn = ii->ii_startoff + off / ii->ii_ndisk; 927 } 928 cbn *= cs->sc_ileave; 929 ci = &cs->sc_cinfo[ccdisk]; 930 } 931 932 /* 933 * Fill in the component buf structure. 934 */ 935 cbp = CCD_GETBUF(); 936 KASSERT(cbp != NULL); 937 buf_init(&cbp->cb_buf); 938 cbp->cb_buf.b_flags = bp->b_flags; 939 cbp->cb_buf.b_oflags = bp->b_oflags; 940 cbp->cb_buf.b_cflags = bp->b_cflags; 941 cbp->cb_buf.b_iodone = ccdiodone; 942 cbp->cb_buf.b_proc = bp->b_proc; 943 cbp->cb_buf.b_dev = ci->ci_dev; 944 cbp->cb_buf.b_blkno = cbn + cboff; 945 cbp->cb_buf.b_data = addr; 946 cbp->cb_buf.b_vp = ci->ci_vp; 947 cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock; 948 if (cs->sc_ileave == 0) 949 cbc = dbtob((u_int64_t)(ci->ci_size - cbn)); 950 else 951 cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff)); 952 cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount; 953 954 /* 955 * context for ccdiodone 956 */ 957 cbp->cb_obp = bp; 958 cbp->cb_sc = cs; 959 cbp->cb_comp = ccdisk; 960 961 BIO_COPYPRIO(&cbp->cb_buf, bp); 962 963 #ifdef DEBUG 964 if (ccddebug & CCDB_IO) 965 printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p" 966 " bcnt %d\n", 967 ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp, 968 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 969 cbp->cb_buf.b_bcount); 970 #endif 971 972 return (cbp); 973 } 974 975 /* 976 * Called at interrupt time. 977 * Mark the component as done and if all components are done, 978 * take a ccd interrupt. 979 */ 980 static void 981 ccdiodone(struct buf *vbp) 982 { 983 struct ccdbuf *cbp = (struct ccdbuf *) vbp; 984 struct buf *bp = cbp->cb_obp; 985 struct ccd_softc *cs = cbp->cb_sc; 986 int count; 987 988 #ifdef DEBUG 989 if (ccddebug & CCDB_FOLLOW) 990 printf("ccdiodone(%p)\n", cbp); 991 if (ccddebug & CCDB_IO) { 992 printf("ccdiodone: bp %p bcount %d resid %d\n", 993 bp, bp->b_bcount, bp->b_resid); 994 printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p" 995 " bcnt %d\n", 996 cbp->cb_buf.b_dev, cbp->cb_comp, cbp, 997 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 998 cbp->cb_buf.b_bcount); 999 } 1000 #endif 1001 1002 if (cbp->cb_buf.b_error != 0) { 1003 bp->b_error = cbp->cb_buf.b_error; 1004 printf("%s: error %d on component %d\n", 1005 cs->sc_xname, bp->b_error, cbp->cb_comp); 1006 } 1007 count = cbp->cb_buf.b_bcount; 1008 buf_destroy(&cbp->cb_buf); 1009 CCD_PUTBUF(cbp); 1010 1011 /* 1012 * If all done, "interrupt". 1013 */ 1014 mutex_enter(cs->sc_iolock); 1015 bp->b_resid -= count; 1016 if (bp->b_resid < 0) 1017 panic("ccdiodone: count"); 1018 if (bp->b_resid == 0) { 1019 /* 1020 * Request is done for better or worse, wakeup the top half. 1021 */ 1022 if (bp->b_error != 0) 1023 bp->b_resid = bp->b_bcount; 1024 disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid), 1025 (bp->b_flags & B_READ)); 1026 if (!disk_isbusy(&cs->sc_dkdev)) { 1027 if (bufq_peek(cs->sc_bufq) != NULL) { 1028 cv_broadcast(&cs->sc_push); 1029 } 1030 cv_broadcast(&cs->sc_stop); 1031 } 1032 mutex_exit(cs->sc_iolock); 1033 biodone(bp); 1034 } else 1035 mutex_exit(cs->sc_iolock); 1036 } 1037 1038 /* ARGSUSED */ 1039 static int 1040 ccdread(dev_t dev, struct uio *uio, int flags) 1041 { 1042 int unit = ccdunit(dev); 1043 struct ccd_softc *cs; 1044 1045 #ifdef DEBUG 1046 if (ccddebug & CCDB_FOLLOW) 1047 printf("ccdread(0x%"PRIx64", %p)\n", dev, uio); 1048 #endif 1049 if ((cs = ccdget(unit, 0)) == NULL) 1050 return 0; 1051 1052 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1053 if ((cs->sc_flags & CCDF_INITED) == 0) 1054 return (ENXIO); 1055 1056 return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio)); 1057 } 1058 1059 /* ARGSUSED */ 1060 static int 1061 ccdwrite(dev_t dev, struct uio *uio, int flags) 1062 { 1063 int unit = ccdunit(dev); 1064 struct ccd_softc *cs; 1065 1066 #ifdef DEBUG 1067 if (ccddebug & CCDB_FOLLOW) 1068 printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio); 1069 #endif 1070 if ((cs = ccdget(unit, 0)) == NULL) 1071 return ENOENT; 1072 1073 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1074 if ((cs->sc_flags & CCDF_INITED) == 0) 1075 return (ENXIO); 1076 1077 return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio)); 1078 } 1079 1080 static int 1081 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1082 { 1083 int unit = ccdunit(dev); 1084 int i, j, lookedup = 0, error = 0; 1085 int part, pmask, make; 1086 struct ccd_softc *cs; 1087 struct ccd_ioctl *ccio = (struct ccd_ioctl *)data; 1088 kauth_cred_t uc; 1089 char **cpp; 1090 struct pathbuf *pb; 1091 struct vnode **vpp; 1092 #ifdef __HAVE_OLD_DISKLABEL 1093 struct disklabel newlabel; 1094 #endif 1095 1096 switch (cmd) { 1097 case CCDIOCSET: 1098 make = 1; 1099 break; 1100 default: 1101 if ((*compat_ccd_ioctl_60)(0, cmd, NULL, 0, NULL, NULL) == 0) 1102 make = 1; 1103 else 1104 make = 0; 1105 break; 1106 } 1107 1108 if ((cs = ccdget(unit, make)) == NULL) 1109 return ENOENT; 1110 uc = kauth_cred_get(); 1111 1112 error = (*compat_ccd_ioctl_60)(dev, cmd, data, flag, l, ccdioctl); 1113 if (error != ENOSYS) 1114 return error; 1115 1116 /* Must be open for writes for these commands... */ 1117 switch (cmd) { 1118 case CCDIOCSET: 1119 case CCDIOCCLR: 1120 case DIOCSDINFO: 1121 case DIOCWDINFO: 1122 case DIOCCACHESYNC: 1123 case DIOCAWEDGE: 1124 case DIOCDWEDGE: 1125 case DIOCMWEDGES: 1126 #ifdef __HAVE_OLD_DISKLABEL 1127 case ODIOCSDINFO: 1128 case ODIOCWDINFO: 1129 #endif 1130 case DIOCKLABEL: 1131 case DIOCWLABEL: 1132 if ((flag & FWRITE) == 0) 1133 return (EBADF); 1134 } 1135 1136 mutex_enter(&cs->sc_dvlock); 1137 1138 /* Must be initialized for these... */ 1139 switch (cmd) { 1140 case CCDIOCCLR: 1141 case DIOCGDINFO: 1142 case DIOCGSTRATEGY: 1143 case DIOCGCACHE: 1144 case DIOCCACHESYNC: 1145 case DIOCAWEDGE: 1146 case DIOCDWEDGE: 1147 case DIOCLWEDGES: 1148 case DIOCMWEDGES: 1149 case DIOCSDINFO: 1150 case DIOCWDINFO: 1151 case DIOCGPARTINFO: 1152 case DIOCWLABEL: 1153 case DIOCKLABEL: 1154 case DIOCGDEFLABEL: 1155 #ifdef __HAVE_OLD_DISKLABEL 1156 case ODIOCGDINFO: 1157 case ODIOCSDINFO: 1158 case ODIOCWDINFO: 1159 case ODIOCGDEFLABEL: 1160 #endif 1161 if ((cs->sc_flags & CCDF_INITED) == 0) { 1162 error = ENXIO; 1163 goto out; 1164 } 1165 } 1166 1167 error = disk_ioctl(&cs->sc_dkdev, dev, cmd, data, flag, l); 1168 if (error != EPASSTHROUGH) 1169 goto out; 1170 1171 error = 0; 1172 switch (cmd) { 1173 case CCDIOCSET: 1174 if (cs->sc_flags & CCDF_INITED) { 1175 error = EBUSY; 1176 goto out; 1177 } 1178 1179 /* Validate the flags. */ 1180 if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) { 1181 error = EINVAL; 1182 goto out; 1183 } 1184 1185 if (ccio->ccio_ndisks > CCD_MAXNDISKS || 1186 ccio->ccio_ndisks == 0) { 1187 error = EINVAL; 1188 goto out; 1189 } 1190 1191 /* Fill in some important bits. */ 1192 cs->sc_ileave = ccio->ccio_ileave; 1193 cs->sc_nccdisks = ccio->ccio_ndisks; 1194 cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK; 1195 1196 /* 1197 * Allocate space for and copy in the array of 1198 * component pathnames and device numbers. 1199 */ 1200 cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP); 1201 vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP); 1202 error = copyin(ccio->ccio_disks, cpp, 1203 ccio->ccio_ndisks * sizeof(*cpp)); 1204 if (error) { 1205 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1206 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1207 goto out; 1208 } 1209 1210 #ifdef DEBUG 1211 if (ccddebug & CCDB_INIT) 1212 for (i = 0; i < ccio->ccio_ndisks; ++i) 1213 printf("ccdioctl: component %d: %p\n", 1214 i, cpp[i]); 1215 #endif 1216 1217 for (i = 0; i < ccio->ccio_ndisks; ++i) { 1218 #ifdef DEBUG 1219 if (ccddebug & CCDB_INIT) 1220 printf("ccdioctl: lookedup = %d\n", lookedup); 1221 #endif 1222 error = pathbuf_copyin(cpp[i], &pb); 1223 if (error == 0) { 1224 error = dk_lookup(pb, l, &vpp[i]); 1225 } 1226 pathbuf_destroy(pb); 1227 if (error != 0) { 1228 for (j = 0; j < lookedup; ++j) 1229 (void)vn_close(vpp[j], FREAD|FWRITE, 1230 uc); 1231 kmem_free(vpp, ccio->ccio_ndisks * 1232 sizeof(*vpp)); 1233 kmem_free(cpp, ccio->ccio_ndisks * 1234 sizeof(*cpp)); 1235 goto out; 1236 } 1237 ++lookedup; 1238 } 1239 1240 /* Attach the disk. */ 1241 disk_attach(&cs->sc_dkdev); 1242 bufq_alloc(&cs->sc_bufq, "fcfs", 0); 1243 1244 /* 1245 * Initialize the ccd. Fills in the softc for us. 1246 */ 1247 if ((error = ccdinit(cs, cpp, vpp, l)) != 0) { 1248 for (j = 0; j < lookedup; ++j) 1249 (void)vn_close(vpp[j], FREAD|FWRITE, 1250 uc); 1251 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1252 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1253 disk_detach(&cs->sc_dkdev); 1254 mutex_exit(&cs->sc_dvlock); 1255 bufq_free(cs->sc_bufq); 1256 return error; 1257 } 1258 1259 /* We can free the temporary variables now. */ 1260 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1261 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1262 1263 /* 1264 * The ccd has been successfully initialized, so 1265 * we can place it into the array. Don't try to 1266 * read the disklabel until the disk has been attached, 1267 * because space for the disklabel is allocated 1268 * in disk_attach(); 1269 */ 1270 ccio->ccio_unit = unit; 1271 ccio->ccio_size = cs->sc_size; 1272 1273 /* Try and read the disklabel. */ 1274 ccdgetdisklabel(dev); 1275 disk_set_info(NULL, &cs->sc_dkdev, NULL); 1276 1277 /* discover wedges */ 1278 mutex_exit(&cs->sc_dvlock); 1279 dkwedge_discover(&cs->sc_dkdev); 1280 return 0; 1281 1282 case CCDIOCCLR: 1283 /* 1284 * Don't unconfigure if any other partitions are open 1285 * or if both the character and block flavors of this 1286 * partition are open. 1287 */ 1288 part = DISKPART(dev); 1289 pmask = (1 << part); 1290 if ((cs->sc_dkdev.dk_openmask & ~pmask) || 1291 ((cs->sc_dkdev.dk_bopenmask & pmask) && 1292 (cs->sc_dkdev.dk_copenmask & pmask))) { 1293 error = EBUSY; 1294 goto out; 1295 } 1296 1297 /* Delete all of our wedges. */ 1298 dkwedge_delall(&cs->sc_dkdev); 1299 1300 /* Stop new I/O, wait for in-flight I/O to complete. */ 1301 mutex_enter(cs->sc_iolock); 1302 cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL); 1303 cs->sc_zap = true; 1304 while (disk_isbusy(&cs->sc_dkdev) || 1305 bufq_peek(cs->sc_bufq) != NULL || 1306 cs->sc_thread != NULL) { 1307 cv_broadcast(&cs->sc_push); 1308 (void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz); 1309 } 1310 mutex_exit(cs->sc_iolock); 1311 1312 /* 1313 * Free ccd_softc information and clear entry. 1314 */ 1315 1316 /* Close the components and free their pathnames. */ 1317 for (i = 0; i < cs->sc_nccdisks; ++i) { 1318 /* 1319 * XXX: this close could potentially fail and 1320 * cause Bad Things. Maybe we need to force 1321 * the close to happen? 1322 */ 1323 #ifdef DEBUG 1324 if (ccddebug & CCDB_VNODE) 1325 vprint("CCDIOCCLR: vnode info", 1326 cs->sc_cinfo[i].ci_vp); 1327 #endif 1328 (void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE, 1329 uc); 1330 kmem_free(cs->sc_cinfo[i].ci_path, 1331 cs->sc_cinfo[i].ci_pathlen); 1332 } 1333 1334 /* Free interleave index. */ 1335 for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) { 1336 kmem_free(cs->sc_itable[i].ii_index, 1337 cs->sc_itable[i].ii_indexsz); 1338 } 1339 1340 /* Free component info and interleave table. */ 1341 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * 1342 sizeof(struct ccdcinfo)); 1343 kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) * 1344 sizeof(struct ccdiinfo)); 1345 1346 aprint_normal("%s: detached\n", cs->sc_xname); 1347 1348 /* Detach the disk. */ 1349 disk_detach(&cs->sc_dkdev); 1350 bufq_free(cs->sc_bufq); 1351 ccdput(cs); 1352 /* Don't break, otherwise cs is read again. */ 1353 return 0; 1354 1355 case DIOCGSTRATEGY: 1356 { 1357 struct disk_strategy *dks = (void *)data; 1358 1359 mutex_enter(cs->sc_iolock); 1360 if (cs->sc_bufq != NULL) 1361 strlcpy(dks->dks_name, 1362 bufq_getstrategyname(cs->sc_bufq), 1363 sizeof(dks->dks_name)); 1364 else 1365 error = EINVAL; 1366 mutex_exit(cs->sc_iolock); 1367 dks->dks_paramlen = 0; 1368 break; 1369 } 1370 1371 case DIOCGCACHE: 1372 { 1373 int dkcache = 0; 1374 1375 /* 1376 * We pass this call down to all components and report 1377 * intersection of the flags returned by the components. 1378 * If any errors out, we return error. CCD components 1379 * can not change unless the device is unconfigured, so 1380 * device feature flags will remain static. RCE/WCE can change 1381 * of course, if set directly on underlying device. 1382 */ 1383 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1384 error = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, &j, 1385 flag, uc); 1386 if (error) 1387 break; 1388 1389 if (i == 0) 1390 dkcache = j; 1391 else 1392 dkcache = DKCACHE_COMBINE(dkcache, j); 1393 } 1394 1395 *((int *)data) = dkcache; 1396 break; 1397 } 1398 1399 case DIOCCACHESYNC: 1400 /* 1401 * We pass this call down to all components and report 1402 * the first error we encounter. 1403 */ 1404 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1405 j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data, 1406 flag, uc); 1407 if (j != 0 && error == 0) 1408 error = j; 1409 } 1410 break; 1411 1412 case DIOCWDINFO: 1413 case DIOCSDINFO: 1414 #ifdef __HAVE_OLD_DISKLABEL 1415 case ODIOCWDINFO: 1416 case ODIOCSDINFO: 1417 #endif 1418 { 1419 struct disklabel *lp; 1420 #ifdef __HAVE_OLD_DISKLABEL 1421 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1422 memset(&newlabel, 0, sizeof newlabel); 1423 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1424 lp = &newlabel; 1425 } else 1426 #endif 1427 lp = (struct disklabel *)data; 1428 1429 cs->sc_flags |= CCDF_LABELLING; 1430 1431 error = setdisklabel(cs->sc_dkdev.dk_label, 1432 lp, 0, cs->sc_dkdev.dk_cpulabel); 1433 if (error == 0) { 1434 if (cmd == DIOCWDINFO 1435 #ifdef __HAVE_OLD_DISKLABEL 1436 || cmd == ODIOCWDINFO 1437 #endif 1438 ) 1439 error = writedisklabel(CCDLABELDEV(dev), 1440 ccdstrategy, cs->sc_dkdev.dk_label, 1441 cs->sc_dkdev.dk_cpulabel); 1442 } 1443 1444 cs->sc_flags &= ~CCDF_LABELLING; 1445 break; 1446 } 1447 1448 case DIOCKLABEL: 1449 if (*(int *)data != 0) 1450 cs->sc_flags |= CCDF_KLABEL; 1451 else 1452 cs->sc_flags &= ~CCDF_KLABEL; 1453 break; 1454 1455 case DIOCWLABEL: 1456 if (*(int *)data != 0) 1457 cs->sc_flags |= CCDF_WLABEL; 1458 else 1459 cs->sc_flags &= ~CCDF_WLABEL; 1460 break; 1461 1462 case DIOCGDEFLABEL: 1463 ccdgetdefaultlabel(cs, (struct disklabel *)data); 1464 break; 1465 1466 #ifdef __HAVE_OLD_DISKLABEL 1467 case ODIOCGDEFLABEL: 1468 ccdgetdefaultlabel(cs, &newlabel); 1469 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1470 return ENOTTY; 1471 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1472 break; 1473 #endif 1474 1475 default: 1476 error = ENOTTY; 1477 } 1478 1479 out: 1480 mutex_exit(&cs->sc_dvlock); 1481 return (error); 1482 } 1483 1484 static int 1485 ccdsize(dev_t dev) 1486 { 1487 struct ccd_softc *cs; 1488 struct disklabel *lp; 1489 int part, unit, omask, size; 1490 1491 unit = ccdunit(dev); 1492 if ((cs = ccdget(unit, 0)) == NULL) 1493 return -1; 1494 1495 if ((cs->sc_flags & CCDF_INITED) == 0) 1496 return (-1); 1497 1498 part = DISKPART(dev); 1499 omask = cs->sc_dkdev.dk_openmask & (1 << part); 1500 lp = cs->sc_dkdev.dk_label; 1501 1502 if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp)) 1503 return (-1); 1504 1505 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1506 size = -1; 1507 else 1508 size = lp->d_partitions[part].p_size * 1509 (lp->d_secsize / DEV_BSIZE); 1510 1511 if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp)) 1512 return (-1); 1513 1514 return (size); 1515 } 1516 1517 static void 1518 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp) 1519 { 1520 struct ccdgeom *ccg = &cs->sc_geom; 1521 1522 memset(lp, 0, sizeof(*lp)); 1523 1524 if (cs->sc_size > UINT32_MAX) 1525 lp->d_secperunit = UINT32_MAX; 1526 else 1527 lp->d_secperunit = cs->sc_size; 1528 lp->d_secsize = ccg->ccg_secsize; 1529 lp->d_nsectors = ccg->ccg_nsectors; 1530 lp->d_ntracks = ccg->ccg_ntracks; 1531 lp->d_ncylinders = ccg->ccg_ncylinders; 1532 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1533 1534 strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename)); 1535 lp->d_type = DKTYPE_CCD; 1536 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1537 lp->d_rpm = 3600; 1538 lp->d_interleave = 1; 1539 lp->d_flags = 0; 1540 1541 lp->d_partitions[RAW_PART].p_offset = 0; 1542 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit; 1543 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 1544 lp->d_npartitions = RAW_PART + 1; 1545 1546 lp->d_magic = DISKMAGIC; 1547 lp->d_magic2 = DISKMAGIC; 1548 lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label); 1549 } 1550 1551 /* 1552 * Read the disklabel from the ccd. If one is not present, fake one 1553 * up. 1554 */ 1555 static void 1556 ccdgetdisklabel(dev_t dev) 1557 { 1558 int unit = ccdunit(dev); 1559 struct ccd_softc *cs; 1560 const char *errstring; 1561 struct disklabel *lp; 1562 struct cpu_disklabel *clp; 1563 1564 if ((cs = ccdget(unit, 0)) == NULL) 1565 return; 1566 lp = cs->sc_dkdev.dk_label; 1567 clp = cs->sc_dkdev.dk_cpulabel; 1568 KASSERT(mutex_owned(&cs->sc_dvlock)); 1569 1570 memset(clp, 0, sizeof(*clp)); 1571 1572 ccdgetdefaultlabel(cs, lp); 1573 1574 /* 1575 * Call the generic disklabel extraction routine. 1576 */ 1577 cs->sc_flags |= CCDF_RLABEL; 1578 if ((cs->sc_flags & CCDF_NOLABEL) != 0) 1579 errstring = "CCDF_NOLABEL set; ignoring on-disk label"; 1580 else 1581 errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy, 1582 cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel); 1583 if (errstring) 1584 ccdmakedisklabel(cs); 1585 else { 1586 int i; 1587 struct partition *pp; 1588 1589 /* 1590 * Sanity check whether the found disklabel is valid. 1591 * 1592 * This is necessary since total size of ccd may vary 1593 * when an interleave is changed even though exactly 1594 * same componets are used, and old disklabel may used 1595 * if that is found. 1596 */ 1597 if (lp->d_secperunit < UINT32_MAX ? 1598 lp->d_secperunit != cs->sc_size : 1599 lp->d_secperunit > cs->sc_size) 1600 printf("WARNING: %s: " 1601 "total sector size in disklabel (%ju) != " 1602 "the size of ccd (%ju)\n", cs->sc_xname, 1603 (uintmax_t)lp->d_secperunit, 1604 (uintmax_t)cs->sc_size); 1605 for (i = 0; i < lp->d_npartitions; i++) { 1606 pp = &lp->d_partitions[i]; 1607 if (pp->p_offset + pp->p_size > cs->sc_size) 1608 printf("WARNING: %s: end of partition `%c' " 1609 "exceeds the size of ccd (%ju)\n", 1610 cs->sc_xname, 'a' + i, (uintmax_t)cs->sc_size); 1611 } 1612 } 1613 1614 #ifdef DEBUG 1615 /* It's actually extremely common to have unlabeled ccds. */ 1616 if (ccddebug & CCDB_LABEL) 1617 if (errstring != NULL) 1618 printf("%s: %s\n", cs->sc_xname, errstring); 1619 #endif 1620 1621 /* In-core label now valid. */ 1622 cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL; 1623 } 1624 1625 /* 1626 * Take care of things one might want to take care of in the event 1627 * that a disklabel isn't present. 1628 */ 1629 static void 1630 ccdmakedisklabel(struct ccd_softc *cs) 1631 { 1632 struct disklabel *lp = cs->sc_dkdev.dk_label; 1633 1634 /* 1635 * For historical reasons, if there's no disklabel present 1636 * the raw partition must be marked FS_BSDFFS. 1637 */ 1638 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 1639 1640 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 1641 1642 lp->d_checksum = dkcksum(lp); 1643 } 1644 1645 #ifdef DEBUG 1646 static void 1647 printiinfo(struct ccdiinfo *ii) 1648 { 1649 int ix, i; 1650 1651 for (ix = 0; ii->ii_ndisk; ix++, ii++) { 1652 printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64, 1653 ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff); 1654 for (i = 0; i < ii->ii_ndisk; i++) 1655 printf(" %d", ii->ii_index[i]); 1656 printf("\n"); 1657 } 1658 } 1659 #endif 1660 1661 MODULE(MODULE_CLASS_DRIVER, ccd, "dk_subr,bufq_fcfs"); 1662 1663 static int 1664 ccd_modcmd(modcmd_t cmd, void *arg) 1665 { 1666 int error = 0; 1667 #ifdef _MODULE 1668 int bmajor = -1, cmajor = -1; 1669 #endif 1670 1671 1672 switch (cmd) { 1673 case MODULE_CMD_INIT: 1674 #ifdef _MODULE 1675 ccdattach(0); 1676 1677 error = devsw_attach("ccd", &ccd_bdevsw, &bmajor, 1678 &ccd_cdevsw, &cmajor); 1679 sysctl_kern_ccd_setup(&ccd_clog); 1680 #endif 1681 break; 1682 1683 case MODULE_CMD_FINI: 1684 #ifdef _MODULE 1685 mutex_enter(&ccd_lock); 1686 if (!LIST_EMPTY(&ccds)) { 1687 mutex_exit(&ccd_lock); 1688 error = EBUSY; 1689 } else { 1690 mutex_exit(&ccd_lock); 1691 error = devsw_detach(&ccd_bdevsw, &ccd_cdevsw); 1692 ccddetach(); 1693 } 1694 sysctl_teardown(&ccd_clog); 1695 #endif 1696 break; 1697 1698 case MODULE_CMD_STAT: 1699 return ENOTTY; 1700 1701 default: 1702 return ENOTTY; 1703 } 1704 1705 return error; 1706 } 1707 1708 static int 1709 ccd_units_sysctl(SYSCTLFN_ARGS) 1710 { 1711 struct sysctlnode node; 1712 struct ccd_softc *sc; 1713 int error, i, nccd, *units; 1714 size_t size; 1715 1716 nccd = 0; 1717 mutex_enter(&ccd_lock); 1718 LIST_FOREACH(sc, &ccds, sc_link) 1719 nccd++; 1720 mutex_exit(&ccd_lock); 1721 1722 if (nccd != 0) { 1723 size = nccd * sizeof(*units); 1724 units = kmem_zalloc(size, KM_SLEEP); 1725 i = 0; 1726 mutex_enter(&ccd_lock); 1727 LIST_FOREACH(sc, &ccds, sc_link) { 1728 if (i >= nccd) 1729 break; 1730 units[i] = sc->sc_unit; 1731 } 1732 mutex_exit(&ccd_lock); 1733 } else { 1734 units = NULL; 1735 size = 0; 1736 } 1737 1738 node = *rnode; 1739 node.sysctl_data = units; 1740 node.sysctl_size = size; 1741 1742 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1743 if (units) 1744 kmem_free(units, size); 1745 return error; 1746 } 1747 1748 static int 1749 ccd_info_sysctl(SYSCTLFN_ARGS) 1750 { 1751 struct sysctlnode node; 1752 struct ccddiskinfo ccd; 1753 struct ccd_softc *sc; 1754 int unit; 1755 1756 if (newp == NULL || newlen != sizeof(int)) 1757 return EINVAL; 1758 1759 unit = *(const int *)newp; 1760 newp = NULL; 1761 newlen = 0; 1762 ccd.ccd_ndisks = ~0; 1763 mutex_enter(&ccd_lock); 1764 LIST_FOREACH(sc, &ccds, sc_link) { 1765 if (sc->sc_unit == unit) { 1766 ccd.ccd_ileave = sc->sc_ileave; 1767 ccd.ccd_size = sc->sc_size; 1768 ccd.ccd_ndisks = sc->sc_nccdisks; 1769 ccd.ccd_flags = sc->sc_flags; 1770 break; 1771 } 1772 } 1773 mutex_exit(&ccd_lock); 1774 1775 if (ccd.ccd_ndisks == ~0) 1776 return ENOENT; 1777 1778 node = *rnode; 1779 node.sysctl_data = &ccd; 1780 node.sysctl_size = sizeof(ccd); 1781 1782 return sysctl_lookup(SYSCTLFN_CALL(&node)); 1783 } 1784 1785 static int 1786 ccd_components_sysctl(SYSCTLFN_ARGS) 1787 { 1788 struct sysctlnode node; 1789 int error, unit; 1790 size_t size; 1791 char *names, *p, *ep; 1792 struct ccd_softc *sc; 1793 1794 if (newp == NULL || newlen != sizeof(int)) 1795 return EINVAL; 1796 1797 size = 0; 1798 unit = *(const int *)newp; 1799 newp = NULL; 1800 newlen = 0; 1801 mutex_enter(&ccd_lock); 1802 LIST_FOREACH(sc, &ccds, sc_link) 1803 if (sc->sc_unit == unit) { 1804 for (size_t i = 0; i < sc->sc_nccdisks; i++) 1805 size += strlen(sc->sc_cinfo[i].ci_path) + 1; 1806 break; 1807 } 1808 mutex_exit(&ccd_lock); 1809 1810 if (size == 0) 1811 return ENOENT; 1812 names = kmem_zalloc(size, KM_SLEEP); 1813 p = names; 1814 ep = names + size; 1815 mutex_enter(&ccd_lock); 1816 LIST_FOREACH(sc, &ccds, sc_link) 1817 if (sc->sc_unit == unit) { 1818 for (size_t i = 0; i < sc->sc_nccdisks; i++) { 1819 char *d = sc->sc_cinfo[i].ci_path; 1820 while (p < ep && (*p++ = *d++) != '\0') 1821 continue; 1822 } 1823 break; 1824 } 1825 mutex_exit(&ccd_lock); 1826 1827 node = *rnode; 1828 node.sysctl_data = names; 1829 node.sysctl_size = ep - names; 1830 1831 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1832 kmem_free(names, size); 1833 return error; 1834 } 1835 1836 SYSCTL_SETUP(sysctl_kern_ccd_setup, "sysctl kern.ccd subtree setup") 1837 { 1838 const struct sysctlnode *node = NULL; 1839 1840 sysctl_createv(clog, 0, NULL, &node, 1841 CTLFLAG_PERMANENT, 1842 CTLTYPE_NODE, "ccd", 1843 SYSCTL_DESCR("ConCatenated Disk state"), 1844 NULL, 0, NULL, 0, 1845 CTL_KERN, CTL_CREATE, CTL_EOL); 1846 1847 if (node == NULL) 1848 return; 1849 1850 sysctl_createv(clog, 0, &node, NULL, 1851 CTLFLAG_PERMANENT | CTLFLAG_READONLY, 1852 CTLTYPE_STRUCT, "units", 1853 SYSCTL_DESCR("List of ccd unit numbers"), 1854 ccd_units_sysctl, 0, NULL, 0, 1855 CTL_CREATE, CTL_EOL); 1856 sysctl_createv(clog, 0, &node, NULL, 1857 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1858 CTLTYPE_STRUCT, "info", 1859 SYSCTL_DESCR("Information about a CCD unit"), 1860 ccd_info_sysctl, 0, NULL, 0, 1861 CTL_CREATE, CTL_EOL); 1862 sysctl_createv(clog, 0, &node, NULL, 1863 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1864 CTLTYPE_STRUCT, "components", 1865 SYSCTL_DESCR("Information about CCD components"), 1866 ccd_components_sysctl, 0, NULL, 0, 1867 CTL_CREATE, CTL_EOL); 1868 } 1869
Indexes created Thu Oct 02 14:10:14 GMT 2025