ccd.c revision 1.179.4.1
1 /* $NetBSD: ccd.c,v 1.179.4.1 2020/10/11 12:34:29 martin 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 1.179.4.1 2020/10/11 12:34:29 martin Exp $"); 92 93 #include <sys/param.h> 94 #include <sys/systm.h> 95 #include <sys/kernel.h> 96 #include <sys/proc.h> 97 #include <sys/errno.h> 98 #include <sys/buf.h> 99 #include <sys/kmem.h> 100 #include <sys/pool.h> 101 #include <sys/module.h> 102 #include <sys/namei.h> 103 #include <sys/stat.h> 104 #include <sys/ioctl.h> 105 #include <sys/disklabel.h> 106 #include <sys/device.h> 107 #include <sys/disk.h> 108 #include <sys/syslog.h> 109 #include <sys/fcntl.h> 110 #include <sys/vnode.h> 111 #include <sys/conf.h> 112 #include <sys/mutex.h> 113 #include <sys/queue.h> 114 #include <sys/kauth.h> 115 #include <sys/kthread.h> 116 #include <sys/bufq.h> 117 #include <sys/sysctl.h> 118 #include <sys/compat_stub.h> 119 120 #include <uvm/uvm_extern.h> 121 122 #include <dev/ccdvar.h> 123 #include <dev/dkvar.h> 124 125 #include <miscfs/specfs/specdev.h> /* for v_rdev */ 126 127 #include "ioconf.h" 128 129 #if defined(CCDDEBUG) && !defined(DEBUG) 130 #define DEBUG 131 #endif 132 133 #ifdef DEBUG 134 #define CCDB_FOLLOW 0x01 135 #define CCDB_INIT 0x02 136 #define CCDB_IO 0x04 137 #define CCDB_LABEL 0x08 138 #define CCDB_VNODE 0x10 139 int ccddebug = 0x00; 140 #endif 141 142 #define ccdunit(x) DISKUNIT(x) 143 144 struct ccdbuf { 145 struct buf cb_buf; /* new I/O buf */ 146 struct buf *cb_obp; /* ptr. to original I/O buf */ 147 struct ccd_softc *cb_sc; /* pointer to ccd softc */ 148 int cb_comp; /* target component */ 149 SIMPLEQ_ENTRY(ccdbuf) cb_q; /* fifo of component buffers */ 150 }; 151 152 /* component buffer pool */ 153 static pool_cache_t ccd_cache; 154 155 #define CCD_GETBUF() pool_cache_get(ccd_cache, PR_WAITOK) 156 #define CCD_PUTBUF(cbp) pool_cache_put(ccd_cache, cbp) 157 158 #define CCDLABELDEV(dev) \ 159 (MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART)) 160 161 /* called by main() at boot time */ 162 void ccddetach(void); 163 164 /* called by biodone() at interrupt time */ 165 static void ccdiodone(struct buf *); 166 167 static void ccdinterleave(struct ccd_softc *); 168 static int ccdinit(struct ccd_softc *, char **, struct vnode **, 169 struct lwp *); 170 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *, 171 daddr_t, void *, long); 172 static void ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *); 173 static void ccdgetdisklabel(dev_t); 174 static void ccdmakedisklabel(struct ccd_softc *); 175 static void ccdstart(struct ccd_softc *); 176 static void ccdthread(void *); 177 178 static dev_type_open(ccdopen); 179 static dev_type_close(ccdclose); 180 static dev_type_read(ccdread); 181 static dev_type_write(ccdwrite); 182 static dev_type_ioctl(ccdioctl); 183 static dev_type_strategy(ccdstrategy); 184 static dev_type_size(ccdsize); 185 186 const struct bdevsw ccd_bdevsw = { 187 .d_open = ccdopen, 188 .d_close = ccdclose, 189 .d_strategy = ccdstrategy, 190 .d_ioctl = ccdioctl, 191 .d_dump = nodump, 192 .d_psize = ccdsize, 193 .d_discard = nodiscard, 194 .d_flag = D_DISK | D_MPSAFE 195 }; 196 197 const struct cdevsw ccd_cdevsw = { 198 .d_open = ccdopen, 199 .d_close = ccdclose, 200 .d_read = ccdread, 201 .d_write = ccdwrite, 202 .d_ioctl = ccdioctl, 203 .d_stop = nostop, 204 .d_tty = notty, 205 .d_poll = nopoll, 206 .d_mmap = nommap, 207 .d_kqfilter = nokqfilter, 208 .d_discard = nodiscard, 209 .d_flag = D_DISK | D_MPSAFE 210 }; 211 212 static const struct dkdriver ccddkdriver = { 213 .d_strategy = ccdstrategy, 214 .d_minphys = minphys 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; 223 224 #ifdef _MODULE 225 static struct sysctllog *ccd_clog; 226 #endif 227 228 SYSCTL_SETUP_PROTO(sysctl_kern_ccd_setup); 229 230 static struct ccd_softc * 231 ccdcreate(int unit) { 232 struct ccd_softc *sc = kmem_zalloc(sizeof(*sc), KM_SLEEP); 233 234 /* Initialize per-softc structures. */ 235 snprintf(sc->sc_xname, sizeof(sc->sc_xname), "ccd%d", unit); 236 sc->sc_unit = unit; 237 mutex_init(&sc->sc_dvlock, MUTEX_DEFAULT, IPL_NONE); 238 sc->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 239 cv_init(&sc->sc_stop, "ccdstop"); 240 cv_init(&sc->sc_push, "ccdthr"); 241 disk_init(&sc->sc_dkdev, sc->sc_xname, &ccddkdriver); 242 return sc; 243 } 244 245 static void 246 ccddestroy(struct ccd_softc *sc) { 247 mutex_obj_free(sc->sc_iolock); 248 mutex_exit(&sc->sc_dvlock); 249 mutex_destroy(&sc->sc_dvlock); 250 cv_destroy(&sc->sc_stop); 251 cv_destroy(&sc->sc_push); 252 disk_destroy(&sc->sc_dkdev); 253 kmem_free(sc, sizeof(*sc)); 254 } 255 256 static struct ccd_softc * 257 ccdget(int unit, int make) { 258 struct ccd_softc *sc; 259 if (unit < 0) { 260 #ifdef DIAGNOSTIC 261 panic("%s: unit %d!", __func__, unit); 262 #endif 263 return NULL; 264 } 265 mutex_enter(&ccd_lock); 266 LIST_FOREACH(sc, &ccds, sc_link) { 267 if (sc->sc_unit == unit) { 268 mutex_exit(&ccd_lock); 269 return sc; 270 } 271 } 272 mutex_exit(&ccd_lock); 273 if (!make) 274 return NULL; 275 if ((sc = ccdcreate(unit)) == NULL) 276 return NULL; 277 mutex_enter(&ccd_lock); 278 LIST_INSERT_HEAD(&ccds, sc, sc_link); 279 mutex_exit(&ccd_lock); 280 return sc; 281 } 282 283 static void 284 ccdput(struct ccd_softc *sc) { 285 mutex_enter(&ccd_lock); 286 LIST_REMOVE(sc, sc_link); 287 mutex_exit(&ccd_lock); 288 ccddestroy(sc); 289 } 290 291 /* 292 * Called by main() during pseudo-device attachment. All we need 293 * to do is allocate enough space for devices to be configured later. 294 */ 295 void 296 ccdattach(int num) 297 { 298 mutex_init(&ccd_lock, MUTEX_DEFAULT, IPL_NONE); 299 300 /* Initialize the component buffer pool. */ 301 ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0, 302 0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL); 303 } 304 305 void 306 ccddetach(void) 307 { 308 pool_cache_destroy(ccd_cache); 309 mutex_destroy(&ccd_lock); 310 } 311 312 static int 313 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp, 314 struct lwp *l) 315 { 316 struct ccdcinfo *ci = NULL; 317 int ix; 318 struct ccdgeom *ccg = &cs->sc_geom; 319 char *tmppath; 320 int error, path_alloced; 321 uint64_t psize, minsize; 322 unsigned secsize, maxsecsize; 323 struct disk_geom *dg; 324 325 #ifdef DEBUG 326 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 327 printf("%s: ccdinit\n", cs->sc_xname); 328 #endif 329 330 /* Allocate space for the component info. */ 331 cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo), 332 KM_SLEEP); 333 tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 334 335 cs->sc_size = 0; 336 337 /* 338 * Verify that each component piece exists and record 339 * relevant information about it. 340 */ 341 maxsecsize = 0; 342 minsize = 0; 343 for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) { 344 ci = &cs->sc_cinfo[ix]; 345 ci->ci_vp = vpp[ix]; 346 347 /* 348 * Copy in the pathname of the component. 349 */ 350 memset(tmppath, 0, MAXPATHLEN); /* sanity */ 351 error = copyinstr(cpaths[ix], tmppath, 352 MAXPATHLEN, &ci->ci_pathlen); 353 if (ci->ci_pathlen == 0) 354 error = EINVAL; 355 if (error) { 356 #ifdef DEBUG 357 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 358 printf("%s: can't copy path, error = %d\n", 359 cs->sc_xname, error); 360 #endif 361 goto out; 362 } 363 ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP); 364 memcpy(ci->ci_path, tmppath, ci->ci_pathlen); 365 path_alloced++; 366 367 /* 368 * XXX: Cache the component's dev_t. 369 */ 370 ci->ci_dev = vpp[ix]->v_rdev; 371 372 /* 373 * Get partition information for the component. 374 */ 375 error = getdisksize(vpp[ix], &psize, &secsize); 376 if (error) { 377 #ifdef DEBUG 378 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 379 printf("%s: %s: disksize failed, error = %d\n", 380 cs->sc_xname, ci->ci_path, error); 381 #endif 382 goto out; 383 } 384 385 /* 386 * Calculate the size, truncating to an interleave 387 * boundary if necessary. 388 */ 389 maxsecsize = secsize > maxsecsize ? secsize : maxsecsize; 390 if (cs->sc_ileave > 1) 391 psize -= psize % cs->sc_ileave; 392 393 if (psize == 0) { 394 #ifdef DEBUG 395 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 396 printf("%s: %s: size == 0\n", 397 cs->sc_xname, ci->ci_path); 398 #endif 399 error = ENODEV; 400 goto out; 401 } 402 403 if (minsize == 0 || psize < minsize) 404 minsize = psize; 405 ci->ci_size = psize; 406 cs->sc_size += psize; 407 } 408 409 /* 410 * Don't allow the interleave to be smaller than 411 * the biggest component sector. 412 */ 413 if ((cs->sc_ileave > 0) && 414 (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) { 415 #ifdef DEBUG 416 if (ccddebug & (CCDB_FOLLOW|CCDB_INIT)) 417 printf("%s: interleave must be at least %d\n", 418 cs->sc_xname, (maxsecsize / DEV_BSIZE)); 419 #endif 420 error = EINVAL; 421 goto out; 422 } 423 424 /* 425 * If uniform interleave is desired set all sizes to that of 426 * the smallest component. 427 */ 428 if (cs->sc_flags & CCDF_UNIFORM) { 429 for (ci = cs->sc_cinfo; 430 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 431 ci->ci_size = minsize; 432 433 cs->sc_size = cs->sc_nccdisks * minsize; 434 } 435 436 /* 437 * Construct the interleave table. 438 */ 439 ccdinterleave(cs); 440 441 /* 442 * Create pseudo-geometry based on 1MB cylinders. It's 443 * pretty close. 444 */ 445 ccg->ccg_secsize = DEV_BSIZE; 446 ccg->ccg_ntracks = 1; 447 ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize); 448 ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors; 449 450 dg = &cs->sc_dkdev.dk_geom; 451 memset(dg, 0, sizeof(*dg)); 452 dg->dg_secperunit = cs->sc_size; 453 dg->dg_secsize = ccg->ccg_secsize; 454 dg->dg_nsectors = ccg->ccg_nsectors; 455 dg->dg_ntracks = ccg->ccg_ntracks; 456 dg->dg_ncylinders = ccg->ccg_ncylinders; 457 458 if (cs->sc_ileave > 0) 459 aprint_normal("%s: Interleaving %d component%s " 460 "(%d block interleave)\n", cs->sc_xname, 461 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""), 462 cs->sc_ileave); 463 else 464 aprint_normal("%s: Concatenating %d component%s\n", 465 cs->sc_xname, 466 cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : "")); 467 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 468 ci = &cs->sc_cinfo[ix]; 469 aprint_normal("%s: %s (%ju blocks)\n", cs->sc_xname, 470 ci->ci_path, (uintmax_t)ci->ci_size); 471 } 472 aprint_normal("%s: total %ju blocks\n", cs->sc_xname, cs->sc_size); 473 474 /* 475 * Create thread to handle deferred I/O. 476 */ 477 cs->sc_zap = false; 478 error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread, 479 cs, &cs->sc_thread, "%s", cs->sc_xname); 480 if (error) { 481 printf("ccdinit: can't create thread: %d\n", error); 482 goto out; 483 } 484 485 /* 486 * Only now that everything is set up can we enable the device. 487 */ 488 mutex_enter(cs->sc_iolock); 489 cs->sc_flags |= CCDF_INITED; 490 mutex_exit(cs->sc_iolock); 491 kmem_free(tmppath, MAXPATHLEN); 492 return (0); 493 494 out: 495 for (ix = 0; ix < path_alloced; ix++) { 496 kmem_free(cs->sc_cinfo[ix].ci_path, 497 cs->sc_cinfo[ix].ci_pathlen); 498 } 499 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo)); 500 kmem_free(tmppath, MAXPATHLEN); 501 return (error); 502 } 503 504 static void 505 ccdinterleave(struct ccd_softc *cs) 506 { 507 struct ccdcinfo *ci, *smallci; 508 struct ccdiinfo *ii; 509 daddr_t bn, lbn; 510 int ix; 511 u_long size; 512 513 #ifdef DEBUG 514 if (ccddebug & CCDB_INIT) 515 printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave); 516 #endif 517 /* 518 * Allocate an interleave table. 519 * Chances are this is too big, but we don't care. 520 */ 521 size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo); 522 cs->sc_itable = kmem_zalloc(size, KM_SLEEP); 523 524 /* 525 * Trivial case: no interleave (actually interleave of disk size). 526 * Each table entry represents a single component in its entirety. 527 */ 528 if (cs->sc_ileave == 0) { 529 bn = 0; 530 ii = cs->sc_itable; 531 532 for (ix = 0; ix < cs->sc_nccdisks; ix++) { 533 /* Allocate space for ii_index. */ 534 ii->ii_indexsz = sizeof(int); 535 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 536 ii->ii_ndisk = 1; 537 ii->ii_startblk = bn; 538 ii->ii_startoff = 0; 539 ii->ii_index[0] = ix; 540 bn += cs->sc_cinfo[ix].ci_size; 541 ii++; 542 } 543 ii->ii_ndisk = 0; 544 #ifdef DEBUG 545 if (ccddebug & CCDB_INIT) 546 printiinfo(cs->sc_itable); 547 #endif 548 return; 549 } 550 551 /* 552 * The following isn't fast or pretty; it doesn't have to be. 553 */ 554 size = 0; 555 bn = lbn = 0; 556 for (ii = cs->sc_itable; ; ii++) { 557 /* Allocate space for ii_index. */ 558 ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks; 559 ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP); 560 561 /* 562 * Locate the smallest of the remaining components 563 */ 564 smallci = NULL; 565 for (ci = cs->sc_cinfo; 566 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 567 if (ci->ci_size > size && 568 (smallci == NULL || 569 ci->ci_size < smallci->ci_size)) 570 smallci = ci; 571 572 /* 573 * Nobody left, all done 574 */ 575 if (smallci == NULL) { 576 ii->ii_ndisk = 0; 577 break; 578 } 579 580 /* 581 * Record starting logical block and component offset 582 */ 583 ii->ii_startblk = bn / cs->sc_ileave; 584 ii->ii_startoff = lbn; 585 586 /* 587 * Determine how many disks take part in this interleave 588 * and record their indices. 589 */ 590 ix = 0; 591 for (ci = cs->sc_cinfo; 592 ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) 593 if (ci->ci_size >= smallci->ci_size) 594 ii->ii_index[ix++] = ci - cs->sc_cinfo; 595 ii->ii_ndisk = ix; 596 bn += ix * (smallci->ci_size - size); 597 lbn = smallci->ci_size / cs->sc_ileave; 598 size = smallci->ci_size; 599 } 600 #ifdef DEBUG 601 if (ccddebug & CCDB_INIT) 602 printiinfo(cs->sc_itable); 603 #endif 604 } 605 606 /* ARGSUSED */ 607 static int 608 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l) 609 { 610 int unit = ccdunit(dev); 611 struct ccd_softc *cs; 612 struct disklabel *lp; 613 int error = 0, part, pmask; 614 615 #ifdef DEBUG 616 if (ccddebug & CCDB_FOLLOW) 617 printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags); 618 #endif 619 if ((cs = ccdget(unit, 1)) == NULL) 620 return ENXIO; 621 622 mutex_enter(&cs->sc_dvlock); 623 624 lp = cs->sc_dkdev.dk_label; 625 626 part = DISKPART(dev); 627 pmask = (1 << part); 628 629 /* 630 * If we're initialized, check to see if there are any other 631 * open partitions. If not, then it's safe to update 632 * the in-core disklabel. Only read the disklabel if it is 633 * not already valid. 634 */ 635 if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED && 636 cs->sc_dkdev.dk_openmask == 0) 637 ccdgetdisklabel(dev); 638 639 /* Check that the partition exists. */ 640 if (part != RAW_PART) { 641 if (((cs->sc_flags & CCDF_INITED) == 0) || 642 ((part >= lp->d_npartitions) || 643 (lp->d_partitions[part].p_fstype == FS_UNUSED))) { 644 error = ENXIO; 645 goto done; 646 } 647 } 648 649 /* Prevent our unit from being unconfigured while open. */ 650 switch (fmt) { 651 case S_IFCHR: 652 cs->sc_dkdev.dk_copenmask |= pmask; 653 break; 654 655 case S_IFBLK: 656 cs->sc_dkdev.dk_bopenmask |= pmask; 657 break; 658 } 659 cs->sc_dkdev.dk_openmask = 660 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 661 662 done: 663 mutex_exit(&cs->sc_dvlock); 664 return (error); 665 } 666 667 /* ARGSUSED */ 668 static int 669 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l) 670 { 671 int unit = ccdunit(dev); 672 struct ccd_softc *cs; 673 int part; 674 675 #ifdef DEBUG 676 if (ccddebug & CCDB_FOLLOW) 677 printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags); 678 #endif 679 680 if ((cs = ccdget(unit, 0)) == NULL) 681 return ENXIO; 682 683 mutex_enter(&cs->sc_dvlock); 684 685 part = DISKPART(dev); 686 687 /* ...that much closer to allowing unconfiguration... */ 688 switch (fmt) { 689 case S_IFCHR: 690 cs->sc_dkdev.dk_copenmask &= ~(1 << part); 691 break; 692 693 case S_IFBLK: 694 cs->sc_dkdev.dk_bopenmask &= ~(1 << part); 695 break; 696 } 697 cs->sc_dkdev.dk_openmask = 698 cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask; 699 700 if (cs->sc_dkdev.dk_openmask == 0) { 701 if ((cs->sc_flags & CCDF_KLABEL) == 0) 702 cs->sc_flags &= ~CCDF_VLABEL; 703 } 704 705 mutex_exit(&cs->sc_dvlock); 706 return (0); 707 } 708 709 static bool 710 ccdbackoff(struct ccd_softc *cs) 711 { 712 713 /* XXX Arbitrary, should be a uvm call. */ 714 return uvmexp.free < (uvmexp.freemin >> 1) && 715 disk_isbusy(&cs->sc_dkdev); 716 } 717 718 static void 719 ccdthread(void *cookie) 720 { 721 struct ccd_softc *cs; 722 723 cs = cookie; 724 725 #ifdef DEBUG 726 if (ccddebug & CCDB_FOLLOW) 727 printf("ccdthread: hello\n"); 728 #endif 729 730 mutex_enter(cs->sc_iolock); 731 while (__predict_true(!cs->sc_zap)) { 732 if (bufq_peek(cs->sc_bufq) == NULL) { 733 /* Nothing to do. */ 734 cv_wait(&cs->sc_push, cs->sc_iolock); 735 continue; 736 } 737 if (ccdbackoff(cs)) { 738 /* Wait for memory to become available. */ 739 (void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1); 740 continue; 741 } 742 #ifdef DEBUG 743 if (ccddebug & CCDB_FOLLOW) 744 printf("ccdthread: dispatching I/O\n"); 745 #endif 746 ccdstart(cs); 747 mutex_enter(cs->sc_iolock); 748 } 749 cs->sc_thread = NULL; 750 mutex_exit(cs->sc_iolock); 751 #ifdef DEBUG 752 if (ccddebug & CCDB_FOLLOW) 753 printf("ccdthread: goodbye\n"); 754 #endif 755 kthread_exit(0); 756 } 757 758 static void 759 ccdstrategy(struct buf *bp) 760 { 761 int unit = ccdunit(bp->b_dev); 762 struct ccd_softc *cs; 763 if ((cs = ccdget(unit, 0)) == NULL) 764 return; 765 766 /* Must be open or reading label. */ 767 KASSERT(cs->sc_dkdev.dk_openmask != 0 || 768 (cs->sc_flags & CCDF_RLABEL) != 0); 769 770 mutex_enter(cs->sc_iolock); 771 /* Synchronize with device init/uninit. */ 772 if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) { 773 mutex_exit(cs->sc_iolock); 774 #ifdef DEBUG 775 if (ccddebug & CCDB_FOLLOW) 776 printf("ccdstrategy: unit %d: not inited\n", unit); 777 #endif 778 bp->b_error = ENXIO; 779 bp->b_resid = bp->b_bcount; 780 biodone(bp); 781 return; 782 } 783 784 /* Defer to thread if system is low on memory. */ 785 bufq_put(cs->sc_bufq, bp); 786 if (__predict_false(ccdbackoff(cs))) { 787 mutex_exit(cs->sc_iolock); 788 #ifdef DEBUG 789 if (ccddebug & CCDB_FOLLOW) 790 printf("ccdstrategy: holding off on I/O\n"); 791 #endif 792 return; 793 } 794 ccdstart(cs); 795 } 796 797 static void 798 ccdstart(struct ccd_softc *cs) 799 { 800 daddr_t blkno; 801 int wlabel; 802 struct disklabel *lp; 803 long bcount, rcount; 804 struct ccdbuf *cbp; 805 char *addr; 806 daddr_t bn; 807 vnode_t *vp; 808 buf_t *bp; 809 810 KASSERT(mutex_owned(cs->sc_iolock)); 811 812 bp = bufq_get(cs->sc_bufq); 813 KASSERT(bp != NULL); 814 815 disk_busy(&cs->sc_dkdev); 816 817 #ifdef DEBUG 818 if (ccddebug & CCDB_FOLLOW) 819 printf("ccdstart(%s, %p)\n", cs->sc_xname, bp); 820 #endif 821 822 /* If it's a nil transfer, wake up the top half now. */ 823 if (bp->b_bcount == 0) 824 goto done; 825 826 lp = cs->sc_dkdev.dk_label; 827 828 /* 829 * Do bounds checking and adjust transfer. If there's an 830 * error, the bounds check will flag that for us. Convert 831 * the partition relative block number to an absolute. 832 */ 833 blkno = bp->b_blkno; 834 wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING); 835 if (DISKPART(bp->b_dev) != RAW_PART) { 836 if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0) 837 goto done; 838 blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset; 839 } 840 mutex_exit(cs->sc_iolock); 841 bp->b_rawblkno = blkno; 842 843 /* Allocate the component buffers and start I/O! */ 844 bp->b_resid = bp->b_bcount; 845 bn = bp->b_rawblkno; 846 addr = bp->b_data; 847 for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) { 848 cbp = ccdbuffer(cs, bp, bn, addr, bcount); 849 rcount = cbp->cb_buf.b_bcount; 850 bn += btodb(rcount); 851 addr += rcount; 852 vp = cbp->cb_buf.b_vp; 853 if ((cbp->cb_buf.b_flags & B_READ) == 0) { 854 mutex_enter(vp->v_interlock); 855 vp->v_numoutput++; 856 mutex_exit(vp->v_interlock); 857 } 858 (void)VOP_STRATEGY(vp, &cbp->cb_buf); 859 } 860 return; 861 862 done: 863 disk_unbusy(&cs->sc_dkdev, 0, 0); 864 cv_broadcast(&cs->sc_stop); 865 cv_broadcast(&cs->sc_push); 866 mutex_exit(cs->sc_iolock); 867 bp->b_resid = bp->b_bcount; 868 biodone(bp); 869 } 870 871 /* 872 * Build a component buffer header. 873 */ 874 static struct ccdbuf * 875 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr, 876 long bcount) 877 { 878 struct ccdcinfo *ci; 879 struct ccdbuf *cbp; 880 daddr_t cbn, cboff; 881 u_int64_t cbc; 882 int ccdisk; 883 884 #ifdef DEBUG 885 if (ccddebug & CCDB_IO) 886 printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n", 887 cs, bp, bn, addr, bcount); 888 #endif 889 /* 890 * Determine which component bn falls in. 891 */ 892 cbn = bn; 893 cboff = 0; 894 895 /* 896 * Serially concatenated 897 */ 898 if (cs->sc_ileave == 0) { 899 daddr_t sblk; 900 901 sblk = 0; 902 for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk]; 903 cbn >= sblk + ci->ci_size; 904 ccdisk++, ci = &cs->sc_cinfo[ccdisk]) 905 sblk += ci->ci_size; 906 cbn -= sblk; 907 } 908 /* 909 * Interleaved 910 */ 911 else { 912 struct ccdiinfo *ii; 913 int off; 914 915 cboff = cbn % cs->sc_ileave; 916 cbn /= cs->sc_ileave; 917 for (ii = cs->sc_itable; ii->ii_ndisk; ii++) 918 if (ii->ii_startblk > cbn) 919 break; 920 ii--; 921 off = cbn - ii->ii_startblk; 922 if (ii->ii_ndisk == 1) { 923 ccdisk = ii->ii_index[0]; 924 cbn = ii->ii_startoff + off; 925 } else { 926 ccdisk = ii->ii_index[off % ii->ii_ndisk]; 927 cbn = ii->ii_startoff + off / ii->ii_ndisk; 928 } 929 cbn *= cs->sc_ileave; 930 ci = &cs->sc_cinfo[ccdisk]; 931 } 932 933 /* 934 * Fill in the component buf structure. 935 */ 936 cbp = CCD_GETBUF(); 937 KASSERT(cbp != NULL); 938 buf_init(&cbp->cb_buf); 939 cbp->cb_buf.b_flags = bp->b_flags; 940 cbp->cb_buf.b_oflags = bp->b_oflags; 941 cbp->cb_buf.b_cflags = bp->b_cflags; 942 cbp->cb_buf.b_iodone = ccdiodone; 943 cbp->cb_buf.b_proc = bp->b_proc; 944 cbp->cb_buf.b_dev = ci->ci_dev; 945 cbp->cb_buf.b_blkno = cbn + cboff; 946 cbp->cb_buf.b_data = addr; 947 cbp->cb_buf.b_vp = ci->ci_vp; 948 cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock; 949 if (cs->sc_ileave == 0) 950 cbc = dbtob((u_int64_t)(ci->ci_size - cbn)); 951 else 952 cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff)); 953 cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount; 954 955 /* 956 * context for ccdiodone 957 */ 958 cbp->cb_obp = bp; 959 cbp->cb_sc = cs; 960 cbp->cb_comp = ccdisk; 961 962 BIO_COPYPRIO(&cbp->cb_buf, bp); 963 964 #ifdef DEBUG 965 if (ccddebug & CCDB_IO) 966 printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p" 967 " bcnt %d\n", 968 ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp, 969 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 970 cbp->cb_buf.b_bcount); 971 #endif 972 973 return (cbp); 974 } 975 976 /* 977 * Called at interrupt time. 978 * Mark the component as done and if all components are done, 979 * take a ccd interrupt. 980 */ 981 static void 982 ccdiodone(struct buf *vbp) 983 { 984 struct ccdbuf *cbp = (struct ccdbuf *) vbp; 985 struct buf *bp = cbp->cb_obp; 986 struct ccd_softc *cs = cbp->cb_sc; 987 int count; 988 989 #ifdef DEBUG 990 if (ccddebug & CCDB_FOLLOW) 991 printf("ccdiodone(%p)\n", cbp); 992 if (ccddebug & CCDB_IO) { 993 printf("ccdiodone: bp %p bcount %d resid %d\n", 994 bp, bp->b_bcount, bp->b_resid); 995 printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p" 996 " bcnt %d\n", 997 cbp->cb_buf.b_dev, cbp->cb_comp, cbp, 998 cbp->cb_buf.b_blkno, cbp->cb_buf.b_data, 999 cbp->cb_buf.b_bcount); 1000 } 1001 #endif 1002 1003 if (cbp->cb_buf.b_error != 0) { 1004 bp->b_error = cbp->cb_buf.b_error; 1005 printf("%s: error %d on component %d\n", 1006 cs->sc_xname, bp->b_error, cbp->cb_comp); 1007 } 1008 count = cbp->cb_buf.b_bcount; 1009 buf_destroy(&cbp->cb_buf); 1010 CCD_PUTBUF(cbp); 1011 1012 /* 1013 * If all done, "interrupt". 1014 */ 1015 mutex_enter(cs->sc_iolock); 1016 bp->b_resid -= count; 1017 if (bp->b_resid < 0) 1018 panic("ccdiodone: count"); 1019 if (bp->b_resid == 0) { 1020 /* 1021 * Request is done for better or worse, wakeup the top half. 1022 */ 1023 if (bp->b_error != 0) 1024 bp->b_resid = bp->b_bcount; 1025 disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid), 1026 (bp->b_flags & B_READ)); 1027 if (!disk_isbusy(&cs->sc_dkdev)) { 1028 if (bufq_peek(cs->sc_bufq) != NULL) { 1029 cv_broadcast(&cs->sc_push); 1030 } 1031 cv_broadcast(&cs->sc_stop); 1032 } 1033 mutex_exit(cs->sc_iolock); 1034 biodone(bp); 1035 } else 1036 mutex_exit(cs->sc_iolock); 1037 } 1038 1039 /* ARGSUSED */ 1040 static int 1041 ccdread(dev_t dev, struct uio *uio, int flags) 1042 { 1043 int unit = ccdunit(dev); 1044 struct ccd_softc *cs; 1045 1046 #ifdef DEBUG 1047 if (ccddebug & CCDB_FOLLOW) 1048 printf("ccdread(0x%"PRIx64", %p)\n", dev, uio); 1049 #endif 1050 if ((cs = ccdget(unit, 0)) == NULL) 1051 return 0; 1052 1053 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1054 if ((cs->sc_flags & CCDF_INITED) == 0) 1055 return (ENXIO); 1056 1057 return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio)); 1058 } 1059 1060 /* ARGSUSED */ 1061 static int 1062 ccdwrite(dev_t dev, struct uio *uio, int flags) 1063 { 1064 int unit = ccdunit(dev); 1065 struct ccd_softc *cs; 1066 1067 #ifdef DEBUG 1068 if (ccddebug & CCDB_FOLLOW) 1069 printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio); 1070 #endif 1071 if ((cs = ccdget(unit, 0)) == NULL) 1072 return ENOENT; 1073 1074 /* Unlocked advisory check, ccdstrategy check is synchronous. */ 1075 if ((cs->sc_flags & CCDF_INITED) == 0) 1076 return (ENXIO); 1077 1078 return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio)); 1079 } 1080 1081 int (*compat_ccd_ioctl_60)(dev_t, u_long, void *, int, struct lwp *, 1082 int (*)(dev_t, u_long, void *, int, struct lwp *)) = (void *)enosys; 1083 1084 static int 1085 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 1086 { 1087 int unit = ccdunit(dev); 1088 int i, j, lookedup = 0, error = 0; 1089 int part, pmask, make, hook; 1090 struct ccd_softc *cs; 1091 struct ccd_ioctl *ccio = (struct ccd_ioctl *)data; 1092 kauth_cred_t uc; 1093 char **cpp; 1094 struct pathbuf *pb; 1095 struct vnode **vpp; 1096 #ifdef __HAVE_OLD_DISKLABEL 1097 struct disklabel newlabel; 1098 #endif 1099 1100 switch (cmd) { 1101 case CCDIOCSET: 1102 make = 1; 1103 break; 1104 default: 1105 MODULE_HOOK_CALL(ccd_ioctl_60_hook, 1106 (0, cmd, NULL, 0, NULL, NULL), 1107 enosys(), hook); 1108 if (hook == 0) 1109 make = 1; 1110 else 1111 make = 0; 1112 break; 1113 } 1114 1115 if ((cs = ccdget(unit, make)) == NULL) 1116 return ENOENT; 1117 uc = kauth_cred_get(); 1118 1119 MODULE_HOOK_CALL(ccd_ioctl_60_hook, 1120 (dev, cmd, data, flag, l, ccdioctl), 1121 enosys(), error); 1122 if (error != ENOSYS) 1123 return error; 1124 1125 /* Must be open for writes for these commands... */ 1126 switch (cmd) { 1127 case CCDIOCSET: 1128 case CCDIOCCLR: 1129 case DIOCSDINFO: 1130 case DIOCWDINFO: 1131 case DIOCCACHESYNC: 1132 case DIOCAWEDGE: 1133 case DIOCDWEDGE: 1134 case DIOCRMWEDGES: 1135 case DIOCMWEDGES: 1136 #ifdef __HAVE_OLD_DISKLABEL 1137 case ODIOCSDINFO: 1138 case ODIOCWDINFO: 1139 #endif 1140 case DIOCKLABEL: 1141 case DIOCWLABEL: 1142 if ((flag & FWRITE) == 0) 1143 return (EBADF); 1144 } 1145 1146 /* Must be initialized for these... */ 1147 switch (cmd) { 1148 case CCDIOCCLR: 1149 case DIOCGDINFO: 1150 case DIOCGSTRATEGY: 1151 case DIOCGCACHE: 1152 case DIOCCACHESYNC: 1153 case DIOCAWEDGE: 1154 case DIOCDWEDGE: 1155 case DIOCLWEDGES: 1156 case DIOCMWEDGES: 1157 case DIOCSDINFO: 1158 case DIOCWDINFO: 1159 case DIOCGPARTINFO: 1160 case DIOCWLABEL: 1161 case DIOCKLABEL: 1162 case DIOCGDEFLABEL: 1163 #ifdef __HAVE_OLD_DISKLABEL 1164 case ODIOCGDINFO: 1165 case ODIOCSDINFO: 1166 case ODIOCWDINFO: 1167 case ODIOCGDEFLABEL: 1168 #endif 1169 if ((cs->sc_flags & CCDF_INITED) == 0) 1170 return ENXIO; 1171 } 1172 1173 error = disk_ioctl(&cs->sc_dkdev, dev, cmd, data, flag, l); 1174 if (error != EPASSTHROUGH) 1175 return error; 1176 1177 switch (cmd) { 1178 case DIOCGSTRATEGY: 1179 { 1180 struct disk_strategy *dks = (void *)data; 1181 1182 mutex_enter(cs->sc_iolock); 1183 if (cs->sc_bufq != NULL) 1184 strlcpy(dks->dks_name, 1185 bufq_getstrategyname(cs->sc_bufq), 1186 sizeof(dks->dks_name)); 1187 else 1188 error = EINVAL; 1189 mutex_exit(cs->sc_iolock); 1190 dks->dks_paramlen = 0; 1191 break; 1192 } 1193 1194 case DIOCWDINFO: 1195 case DIOCSDINFO: 1196 #ifdef __HAVE_OLD_DISKLABEL 1197 case ODIOCWDINFO: 1198 case ODIOCSDINFO: 1199 #endif 1200 { 1201 struct disklabel *lp; 1202 #ifdef __HAVE_OLD_DISKLABEL 1203 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) { 1204 memset(&newlabel, 0, sizeof newlabel); 1205 memcpy(&newlabel, data, sizeof (struct olddisklabel)); 1206 lp = &newlabel; 1207 } else 1208 #endif 1209 lp = (struct disklabel *)data; 1210 1211 cs->sc_flags |= CCDF_LABELLING; 1212 1213 error = setdisklabel(cs->sc_dkdev.dk_label, 1214 lp, 0, cs->sc_dkdev.dk_cpulabel); 1215 if (error == 0) { 1216 if (cmd == DIOCWDINFO 1217 #ifdef __HAVE_OLD_DISKLABEL 1218 || cmd == ODIOCWDINFO 1219 #endif 1220 ) 1221 error = writedisklabel(CCDLABELDEV(dev), 1222 ccdstrategy, cs->sc_dkdev.dk_label, 1223 cs->sc_dkdev.dk_cpulabel); 1224 } 1225 1226 cs->sc_flags &= ~CCDF_LABELLING; 1227 break; 1228 } 1229 1230 case DIOCKLABEL: 1231 if (*(int *)data != 0) 1232 cs->sc_flags |= CCDF_KLABEL; 1233 else 1234 cs->sc_flags &= ~CCDF_KLABEL; 1235 break; 1236 1237 case DIOCWLABEL: 1238 if (*(int *)data != 0) 1239 cs->sc_flags |= CCDF_WLABEL; 1240 else 1241 cs->sc_flags &= ~CCDF_WLABEL; 1242 break; 1243 1244 case DIOCGDEFLABEL: 1245 ccdgetdefaultlabel(cs, (struct disklabel *)data); 1246 break; 1247 1248 #ifdef __HAVE_OLD_DISKLABEL 1249 case ODIOCGDEFLABEL: 1250 ccdgetdefaultlabel(cs, &newlabel); 1251 if (newlabel.d_npartitions > OLDMAXPARTITIONS) 1252 return ENOTTY; 1253 memcpy(data, &newlabel, sizeof (struct olddisklabel)); 1254 break; 1255 #endif 1256 default: 1257 error = ENOTTY; 1258 break; 1259 } 1260 1261 if (error != ENOTTY) 1262 return error; 1263 1264 mutex_enter(&cs->sc_dvlock); 1265 1266 error = 0; 1267 switch (cmd) { 1268 case CCDIOCSET: 1269 if (cs->sc_flags & CCDF_INITED) { 1270 error = EBUSY; 1271 goto out; 1272 } 1273 1274 /* Validate the flags. */ 1275 if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) { 1276 error = EINVAL; 1277 goto out; 1278 } 1279 1280 if (ccio->ccio_ndisks > CCD_MAXNDISKS || 1281 ccio->ccio_ndisks == 0) { 1282 error = EINVAL; 1283 goto out; 1284 } 1285 1286 /* Fill in some important bits. */ 1287 cs->sc_ileave = ccio->ccio_ileave; 1288 cs->sc_nccdisks = ccio->ccio_ndisks; 1289 cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK; 1290 1291 /* 1292 * Allocate space for and copy in the array of 1293 * component pathnames and device numbers. 1294 */ 1295 cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP); 1296 vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP); 1297 error = copyin(ccio->ccio_disks, cpp, 1298 ccio->ccio_ndisks * sizeof(*cpp)); 1299 if (error) { 1300 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1301 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1302 goto out; 1303 } 1304 1305 #ifdef DEBUG 1306 if (ccddebug & CCDB_INIT) 1307 for (i = 0; i < ccio->ccio_ndisks; ++i) 1308 printf("ccdioctl: component %d: %p\n", 1309 i, cpp[i]); 1310 #endif 1311 1312 for (i = 0; i < ccio->ccio_ndisks; ++i) { 1313 #ifdef DEBUG 1314 if (ccddebug & CCDB_INIT) 1315 printf("ccdioctl: lookedup = %d\n", lookedup); 1316 #endif 1317 error = pathbuf_copyin(cpp[i], &pb); 1318 if (error == 0) { 1319 error = dk_lookup(pb, l, &vpp[i]); 1320 } 1321 pathbuf_destroy(pb); 1322 if (error != 0) { 1323 for (j = 0; j < lookedup; ++j) 1324 (void)vn_close(vpp[j], FREAD|FWRITE, 1325 uc); 1326 kmem_free(vpp, ccio->ccio_ndisks * 1327 sizeof(*vpp)); 1328 kmem_free(cpp, ccio->ccio_ndisks * 1329 sizeof(*cpp)); 1330 1331 /* 1332 * No component data is allocated, 1333 * nothing is to be freed. 1334 */ 1335 cs->sc_nccdisks = 0; 1336 goto out; 1337 } 1338 ++lookedup; 1339 } 1340 1341 /* Attach the disk. */ 1342 disk_attach(&cs->sc_dkdev); 1343 bufq_alloc(&cs->sc_bufq, "fcfs", 0); 1344 1345 /* 1346 * Initialize the ccd. Fills in the softc for us. 1347 */ 1348 if ((error = ccdinit(cs, cpp, vpp, l)) != 0) { 1349 for (j = 0; j < lookedup; ++j) 1350 (void)vn_close(vpp[j], FREAD|FWRITE, 1351 uc); 1352 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1353 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1354 disk_detach(&cs->sc_dkdev); 1355 mutex_exit(&cs->sc_dvlock); 1356 bufq_free(cs->sc_bufq); 1357 return error; 1358 } 1359 1360 /* We can free the temporary variables now. */ 1361 kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp)); 1362 kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp)); 1363 1364 /* 1365 * The ccd has been successfully initialized, so 1366 * we can place it into the array. Don't try to 1367 * read the disklabel until the disk has been attached, 1368 * because space for the disklabel is allocated 1369 * in disk_attach(); 1370 */ 1371 ccio->ccio_unit = unit; 1372 ccio->ccio_size = cs->sc_size; 1373 1374 /* Try and read the disklabel. */ 1375 ccdgetdisklabel(dev); 1376 disk_set_info(NULL, &cs->sc_dkdev, NULL); 1377 1378 /* discover wedges */ 1379 mutex_exit(&cs->sc_dvlock); 1380 dkwedge_discover(&cs->sc_dkdev); 1381 return 0; 1382 1383 case CCDIOCCLR: 1384 /* 1385 * Don't unconfigure if any other partitions are open 1386 * or if both the character and block flavors of this 1387 * partition are open. 1388 */ 1389 part = DISKPART(dev); 1390 pmask = (1 << part); 1391 if ((cs->sc_dkdev.dk_openmask & ~pmask) || 1392 ((cs->sc_dkdev.dk_bopenmask & pmask) && 1393 (cs->sc_dkdev.dk_copenmask & pmask))) { 1394 error = EBUSY; 1395 goto out; 1396 } 1397 1398 /* Delete all of our wedges. */ 1399 dkwedge_delall(&cs->sc_dkdev); 1400 1401 /* Stop new I/O, wait for in-flight I/O to complete. */ 1402 mutex_enter(cs->sc_iolock); 1403 cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL); 1404 cs->sc_zap = true; 1405 while (disk_isbusy(&cs->sc_dkdev) || 1406 bufq_peek(cs->sc_bufq) != NULL || 1407 cs->sc_thread != NULL) { 1408 cv_broadcast(&cs->sc_push); 1409 (void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz); 1410 } 1411 mutex_exit(cs->sc_iolock); 1412 1413 /* 1414 * Free ccd_softc information and clear entry. 1415 */ 1416 1417 /* Close the components and free their pathnames. */ 1418 for (i = 0; i < cs->sc_nccdisks; ++i) { 1419 /* 1420 * XXX: this close could potentially fail and 1421 * cause Bad Things. Maybe we need to force 1422 * the close to happen? 1423 */ 1424 #ifdef DEBUG 1425 if (ccddebug & CCDB_VNODE) 1426 vprint("CCDIOCCLR: vnode info", 1427 cs->sc_cinfo[i].ci_vp); 1428 #endif 1429 (void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE, 1430 uc); 1431 kmem_free(cs->sc_cinfo[i].ci_path, 1432 cs->sc_cinfo[i].ci_pathlen); 1433 } 1434 1435 if (cs->sc_nccdisks != 0) { 1436 /* Free interleave index. */ 1437 for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) { 1438 kmem_free(cs->sc_itable[i].ii_index, 1439 cs->sc_itable[i].ii_indexsz); 1440 } 1441 /* Free component info and interleave table. */ 1442 kmem_free(cs->sc_cinfo, cs->sc_nccdisks * 1443 sizeof(struct ccdcinfo)); 1444 kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) * 1445 sizeof(struct ccdiinfo)); 1446 } 1447 1448 aprint_normal("%s: detached\n", cs->sc_xname); 1449 1450 /* Detach the disk. */ 1451 disk_detach(&cs->sc_dkdev); 1452 bufq_free(cs->sc_bufq); 1453 1454 /* also releases dv_lock */ 1455 ccdput(cs); 1456 1457 /* Don't break, otherwise cs is read again. */ 1458 return 0; 1459 1460 case DIOCGCACHE: 1461 { 1462 int dkcache = 0; 1463 1464 /* 1465 * We pass this call down to all components and report 1466 * intersection of the flags returned by the components. 1467 * If any errors out, we return error. CCD components 1468 * can not change unless the device is unconfigured, so 1469 * device feature flags will remain static. RCE/WCE can change 1470 * of course, if set directly on underlying device. 1471 */ 1472 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1473 error = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, &j, 1474 flag, uc); 1475 if (error) 1476 break; 1477 1478 if (i == 0) 1479 dkcache = j; 1480 else 1481 dkcache = DKCACHE_COMBINE(dkcache, j); 1482 } 1483 1484 *((int *)data) = dkcache; 1485 break; 1486 } 1487 1488 case DIOCCACHESYNC: 1489 /* 1490 * We pass this call down to all components and report 1491 * the first error we encounter. 1492 */ 1493 for (error = 0, i = 0; i < cs->sc_nccdisks; i++) { 1494 j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data, 1495 flag, uc); 1496 if (j != 0 && error == 0) 1497 error = j; 1498 } 1499 break; 1500 1501 default: 1502 error = ENOTTY; 1503 break; 1504 } 1505 1506 out: 1507 mutex_exit(&cs->sc_dvlock); 1508 return (error); 1509 } 1510 1511 static int 1512 ccdsize(dev_t dev) 1513 { 1514 struct ccd_softc *cs; 1515 struct disklabel *lp; 1516 int part, unit, omask, size; 1517 1518 unit = ccdunit(dev); 1519 if ((cs = ccdget(unit, 0)) == NULL) 1520 return -1; 1521 1522 if ((cs->sc_flags & CCDF_INITED) == 0) 1523 return (-1); 1524 1525 part = DISKPART(dev); 1526 omask = cs->sc_dkdev.dk_openmask & (1 << part); 1527 lp = cs->sc_dkdev.dk_label; 1528 1529 if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp)) 1530 return (-1); 1531 1532 if (lp->d_partitions[part].p_fstype != FS_SWAP) 1533 size = -1; 1534 else 1535 size = lp->d_partitions[part].p_size * 1536 (lp->d_secsize / DEV_BSIZE); 1537 1538 if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp)) 1539 return (-1); 1540 1541 return (size); 1542 } 1543 1544 static void 1545 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp) 1546 { 1547 struct ccdgeom *ccg = &cs->sc_geom; 1548 1549 memset(lp, 0, sizeof(*lp)); 1550 1551 if (cs->sc_size > UINT32_MAX) 1552 lp->d_secperunit = UINT32_MAX; 1553 else 1554 lp->d_secperunit = cs->sc_size; 1555 lp->d_secsize = ccg->ccg_secsize; 1556 lp->d_nsectors = ccg->ccg_nsectors; 1557 lp->d_ntracks = ccg->ccg_ntracks; 1558 lp->d_ncylinders = ccg->ccg_ncylinders; 1559 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors; 1560 1561 strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename)); 1562 lp->d_type = DKTYPE_CCD; 1563 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 1564 lp->d_rpm = 3600; 1565 lp->d_interleave = 1; 1566 lp->d_flags = 0; 1567 1568 lp->d_partitions[RAW_PART].p_offset = 0; 1569 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit; 1570 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED; 1571 lp->d_npartitions = RAW_PART + 1; 1572 1573 lp->d_magic = DISKMAGIC; 1574 lp->d_magic2 = DISKMAGIC; 1575 lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label); 1576 } 1577 1578 /* 1579 * Read the disklabel from the ccd. If one is not present, fake one 1580 * up. 1581 */ 1582 static void 1583 ccdgetdisklabel(dev_t dev) 1584 { 1585 int unit = ccdunit(dev); 1586 struct ccd_softc *cs; 1587 const char *errstring; 1588 struct disklabel *lp; 1589 struct cpu_disklabel *clp; 1590 1591 if ((cs = ccdget(unit, 0)) == NULL) 1592 return; 1593 lp = cs->sc_dkdev.dk_label; 1594 clp = cs->sc_dkdev.dk_cpulabel; 1595 KASSERT(mutex_owned(&cs->sc_dvlock)); 1596 1597 memset(clp, 0, sizeof(*clp)); 1598 1599 ccdgetdefaultlabel(cs, lp); 1600 1601 /* 1602 * Call the generic disklabel extraction routine. 1603 */ 1604 cs->sc_flags |= CCDF_RLABEL; 1605 if ((cs->sc_flags & CCDF_NOLABEL) != 0) 1606 errstring = "CCDF_NOLABEL set; ignoring on-disk label"; 1607 else 1608 errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy, 1609 cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel); 1610 if (errstring) 1611 ccdmakedisklabel(cs); 1612 else { 1613 int i; 1614 struct partition *pp; 1615 1616 /* 1617 * Sanity check whether the found disklabel is valid. 1618 * 1619 * This is necessary since total size of ccd may vary 1620 * when an interleave is changed even though exactly 1621 * same componets are used, and old disklabel may used 1622 * if that is found. 1623 */ 1624 if (lp->d_secperunit < UINT32_MAX ? 1625 lp->d_secperunit != cs->sc_size : 1626 lp->d_secperunit > cs->sc_size) 1627 printf("WARNING: %s: " 1628 "total sector size in disklabel (%ju) != " 1629 "the size of ccd (%ju)\n", cs->sc_xname, 1630 (uintmax_t)lp->d_secperunit, 1631 (uintmax_t)cs->sc_size); 1632 for (i = 0; i < lp->d_npartitions; i++) { 1633 pp = &lp->d_partitions[i]; 1634 if (pp->p_offset + pp->p_size > cs->sc_size) 1635 printf("WARNING: %s: end of partition `%c' " 1636 "exceeds the size of ccd (%ju)\n", 1637 cs->sc_xname, 'a' + i, (uintmax_t)cs->sc_size); 1638 } 1639 } 1640 1641 #ifdef DEBUG 1642 /* It's actually extremely common to have unlabeled ccds. */ 1643 if (ccddebug & CCDB_LABEL) 1644 if (errstring != NULL) 1645 printf("%s: %s\n", cs->sc_xname, errstring); 1646 #endif 1647 1648 /* In-core label now valid. */ 1649 cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL; 1650 } 1651 1652 /* 1653 * Take care of things one might want to take care of in the event 1654 * that a disklabel isn't present. 1655 */ 1656 static void 1657 ccdmakedisklabel(struct ccd_softc *cs) 1658 { 1659 struct disklabel *lp = cs->sc_dkdev.dk_label; 1660 1661 /* 1662 * For historical reasons, if there's no disklabel present 1663 * the raw partition must be marked FS_BSDFFS. 1664 */ 1665 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS; 1666 1667 strncpy(lp->d_packname, "default label", sizeof(lp->d_packname)); 1668 1669 lp->d_checksum = dkcksum(lp); 1670 } 1671 1672 #ifdef DEBUG 1673 static void 1674 printiinfo(struct ccdiinfo *ii) 1675 { 1676 int ix, i; 1677 1678 for (ix = 0; ii->ii_ndisk; ix++, ii++) { 1679 printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64, 1680 ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff); 1681 for (i = 0; i < ii->ii_ndisk; i++) 1682 printf(" %d", ii->ii_index[i]); 1683 printf("\n"); 1684 } 1685 } 1686 #endif 1687 1688 MODULE(MODULE_CLASS_DRIVER, ccd, "dk_subr,bufq_fcfs"); 1689 1690 static int 1691 ccd_modcmd(modcmd_t cmd, void *arg) 1692 { 1693 int error = 0; 1694 #ifdef _MODULE 1695 int bmajor = -1, cmajor = -1; 1696 #endif 1697 1698 1699 switch (cmd) { 1700 case MODULE_CMD_INIT: 1701 #ifdef _MODULE 1702 ccdattach(0); 1703 1704 error = devsw_attach("ccd", &ccd_bdevsw, &bmajor, 1705 &ccd_cdevsw, &cmajor); 1706 sysctl_kern_ccd_setup(&ccd_clog); 1707 #endif 1708 break; 1709 1710 case MODULE_CMD_FINI: 1711 #ifdef _MODULE 1712 mutex_enter(&ccd_lock); 1713 if (!LIST_EMPTY(&ccds)) { 1714 mutex_exit(&ccd_lock); 1715 error = EBUSY; 1716 } else { 1717 mutex_exit(&ccd_lock); 1718 error = devsw_detach(&ccd_bdevsw, &ccd_cdevsw); 1719 ccddetach(); 1720 } 1721 sysctl_teardown(&ccd_clog); 1722 #endif 1723 break; 1724 1725 case MODULE_CMD_STAT: 1726 return ENOTTY; 1727 1728 default: 1729 return ENOTTY; 1730 } 1731 1732 return error; 1733 } 1734 1735 static int 1736 ccd_units_sysctl(SYSCTLFN_ARGS) 1737 { 1738 struct sysctlnode node; 1739 struct ccd_softc *sc; 1740 int error, i, nccd, *units; 1741 size_t size; 1742 1743 nccd = 0; 1744 mutex_enter(&ccd_lock); 1745 LIST_FOREACH(sc, &ccds, sc_link) 1746 nccd++; 1747 mutex_exit(&ccd_lock); 1748 1749 if (nccd != 0) { 1750 size = nccd * sizeof(*units); 1751 units = kmem_zalloc(size, KM_SLEEP); 1752 i = 0; 1753 mutex_enter(&ccd_lock); 1754 LIST_FOREACH(sc, &ccds, sc_link) { 1755 if (i >= nccd) 1756 break; 1757 units[i] = sc->sc_unit; 1758 } 1759 mutex_exit(&ccd_lock); 1760 } else { 1761 units = NULL; 1762 size = 0; 1763 } 1764 1765 node = *rnode; 1766 node.sysctl_data = units; 1767 node.sysctl_size = size; 1768 1769 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1770 if (units) 1771 kmem_free(units, size); 1772 return error; 1773 } 1774 1775 static int 1776 ccd_info_sysctl(SYSCTLFN_ARGS) 1777 { 1778 struct sysctlnode node; 1779 struct ccddiskinfo ccd; 1780 struct ccd_softc *sc; 1781 int unit; 1782 1783 if (newp == NULL || newlen != sizeof(int)) 1784 return EINVAL; 1785 1786 unit = *(const int *)newp; 1787 newp = NULL; 1788 newlen = 0; 1789 ccd.ccd_ndisks = ~0; 1790 mutex_enter(&ccd_lock); 1791 LIST_FOREACH(sc, &ccds, sc_link) { 1792 if (sc->sc_unit == unit) { 1793 ccd.ccd_ileave = sc->sc_ileave; 1794 ccd.ccd_size = sc->sc_size; 1795 ccd.ccd_ndisks = sc->sc_nccdisks; 1796 ccd.ccd_flags = sc->sc_flags; 1797 break; 1798 } 1799 } 1800 mutex_exit(&ccd_lock); 1801 1802 if (ccd.ccd_ndisks == ~0) 1803 return ENOENT; 1804 1805 node = *rnode; 1806 node.sysctl_data = &ccd; 1807 node.sysctl_size = sizeof(ccd); 1808 1809 return sysctl_lookup(SYSCTLFN_CALL(&node)); 1810 } 1811 1812 static int 1813 ccd_components_sysctl(SYSCTLFN_ARGS) 1814 { 1815 struct sysctlnode node; 1816 int error, unit; 1817 size_t size; 1818 char *names, *p, *ep; 1819 struct ccd_softc *sc; 1820 1821 if (newp == NULL || newlen != sizeof(int)) 1822 return EINVAL; 1823 1824 size = 0; 1825 unit = *(const int *)newp; 1826 newp = NULL; 1827 newlen = 0; 1828 mutex_enter(&ccd_lock); 1829 LIST_FOREACH(sc, &ccds, sc_link) 1830 if (sc->sc_unit == unit) { 1831 for (size_t i = 0; i < sc->sc_nccdisks; i++) 1832 size += strlen(sc->sc_cinfo[i].ci_path) + 1; 1833 break; 1834 } 1835 mutex_exit(&ccd_lock); 1836 1837 if (size == 0) 1838 return ENOENT; 1839 names = kmem_zalloc(size, KM_SLEEP); 1840 p = names; 1841 ep = names + size; 1842 mutex_enter(&ccd_lock); 1843 LIST_FOREACH(sc, &ccds, sc_link) 1844 if (sc->sc_unit == unit) { 1845 for (size_t i = 0; i < sc->sc_nccdisks; i++) { 1846 char *d = sc->sc_cinfo[i].ci_path; 1847 while (p < ep && (*p++ = *d++) != '\0') 1848 continue; 1849 } 1850 break; 1851 } 1852 mutex_exit(&ccd_lock); 1853 1854 node = *rnode; 1855 node.sysctl_data = names; 1856 node.sysctl_size = ep - names; 1857 1858 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1859 kmem_free(names, size); 1860 return error; 1861 } 1862 1863 SYSCTL_SETUP(sysctl_kern_ccd_setup, "sysctl kern.ccd subtree setup") 1864 { 1865 const struct sysctlnode *node = NULL; 1866 1867 sysctl_createv(clog, 0, NULL, &node, 1868 CTLFLAG_PERMANENT, 1869 CTLTYPE_NODE, "ccd", 1870 SYSCTL_DESCR("ConCatenated Disk state"), 1871 NULL, 0, NULL, 0, 1872 CTL_KERN, CTL_CREATE, CTL_EOL); 1873 1874 if (node == NULL) 1875 return; 1876 1877 sysctl_createv(clog, 0, &node, NULL, 1878 CTLFLAG_PERMANENT | CTLFLAG_READONLY, 1879 CTLTYPE_STRUCT, "units", 1880 SYSCTL_DESCR("List of ccd unit numbers"), 1881 ccd_units_sysctl, 0, NULL, 0, 1882 CTL_CREATE, CTL_EOL); 1883 sysctl_createv(clog, 0, &node, NULL, 1884 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1885 CTLTYPE_STRUCT, "info", 1886 SYSCTL_DESCR("Information about a CCD unit"), 1887 ccd_info_sysctl, 0, NULL, 0, 1888 CTL_CREATE, CTL_EOL); 1889 sysctl_createv(clog, 0, &node, NULL, 1890 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1891 CTLTYPE_STRUCT, "components", 1892 SYSCTL_DESCR("Information about CCD components"), 1893 ccd_components_sysctl, 0, NULL, 0, 1894 CTL_CREATE, CTL_EOL); 1895 } 1896
Indexes created Thu Oct 02 14:10:14 GMT 2025