md.c revision 1.66.8.1
1 /* $NetBSD: md.c,v 1.66.8.1 2012/10/30 17:20:49 yamt Exp $ */ 2 3 /* 4 * Copyright (c) 1995 Gordon W. Ross, Leo Weppelman. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 /* 29 * This implements a general-purpose memory-disk. 30 * See md.h for notes on the config types. 31 * 32 * Note that this driver provides the same functionality 33 * as the MFS filesystem hack, but this is better because 34 * you can use this for any filesystem type you'd like! 35 * 36 * Credit for most of the kmem ramdisk code goes to: 37 * Leo Weppelman (atari) and Phil Nelson (pc532) 38 * Credit for the ideas behind the "user space memory" code goes 39 * to the authors of the MFS implementation. 40 */ 41 42 #include <sys/cdefs.h> 43 __KERNEL_RCSID(0, "$NetBSD: md.c,v 1.66.8.1 2012/10/30 17:20:49 yamt Exp $"); 44 45 #ifdef _KERNEL_OPT 46 #include "opt_md.h" 47 #else 48 #define MEMORY_DISK_SERVER 1 49 #endif 50 51 #include <sys/param.h> 52 #include <sys/kernel.h> 53 #include <sys/malloc.h> 54 #include <sys/systm.h> 55 #include <sys/buf.h> 56 #include <sys/bufq.h> 57 #include <sys/device.h> 58 #include <sys/disk.h> 59 #include <sys/stat.h> 60 #include <sys/proc.h> 61 #include <sys/conf.h> 62 #include <sys/disklabel.h> 63 64 #include <uvm/uvm_extern.h> 65 66 #include <dev/md.h> 67 68 /* 69 * The user-space functionality is included by default. 70 * Use `options MEMORY_DISK_SERVER=0' to turn it off. 71 */ 72 #ifndef MEMORY_DISK_SERVER 73 #error MEMORY_DISK_SERVER should be defined by opt_md.h 74 #endif /* MEMORY_DISK_SERVER */ 75 76 /* 77 * We should use the raw partition for ioctl. 78 */ 79 #define MD_UNIT(unit) DISKUNIT(unit) 80 81 /* autoconfig stuff... */ 82 83 struct md_softc { 84 device_t sc_dev; /* Self. */ 85 struct disk sc_dkdev; /* hook for generic disk handling */ 86 struct md_conf sc_md; 87 kmutex_t sc_lock; /* Protect self. */ 88 kcondvar_t sc_cv; /* Wait here for work. */ 89 struct bufq_state *sc_buflist; 90 }; 91 /* shorthand for fields in sc_md: */ 92 #define sc_addr sc_md.md_addr 93 #define sc_size sc_md.md_size 94 #define sc_type sc_md.md_type 95 96 void mdattach(int); 97 98 static void md_attach(device_t, device_t, void *); 99 static int md_detach(device_t, int); 100 101 static dev_type_open(mdopen); 102 static dev_type_close(mdclose); 103 static dev_type_read(mdread); 104 static dev_type_write(mdwrite); 105 static dev_type_ioctl(mdioctl); 106 static dev_type_strategy(mdstrategy); 107 static dev_type_size(mdsize); 108 109 const struct bdevsw md_bdevsw = { 110 mdopen, mdclose, mdstrategy, mdioctl, nodump, mdsize, D_DISK | D_MPSAFE 111 }; 112 113 const struct cdevsw md_cdevsw = { 114 mdopen, mdclose, mdread, mdwrite, mdioctl, 115 nostop, notty, nopoll, nommap, nokqfilter, D_DISK 116 }; 117 118 static struct dkdriver mddkdriver = { mdstrategy, NULL }; 119 120 extern struct cfdriver md_cd; 121 CFATTACH_DECL3_NEW(md, sizeof(struct md_softc), 122 0, md_attach, md_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN); 123 124 static kmutex_t md_device_lock; /* Protect unit creation / deletion. */ 125 extern size_t md_root_size; 126 127 static void md_set_disklabel(struct md_softc *); 128 129 /* 130 * This is called if we are configured as a pseudo-device 131 */ 132 void 133 mdattach(int n) 134 { 135 136 mutex_init(&md_device_lock, MUTEX_DEFAULT, IPL_NONE); 137 if (config_cfattach_attach(md_cd.cd_name, &md_ca)) { 138 aprint_error("%s: cfattach_attach failed\n", md_cd.cd_name); 139 return; 140 } 141 } 142 143 static void 144 md_attach(device_t parent, device_t self, void *aux) 145 { 146 struct md_softc *sc = device_private(self); 147 148 sc->sc_dev = self; 149 sc->sc_type = MD_UNCONFIGURED; 150 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 151 cv_init(&sc->sc_cv, "mdidle"); 152 bufq_alloc(&sc->sc_buflist, "fcfs", 0); 153 154 /* XXX - Could accept aux info here to set the config. */ 155 #ifdef MEMORY_DISK_HOOKS 156 /* 157 * This external function might setup a pre-loaded disk. 158 * All it would need to do is setup the md_conf struct. 159 * See sys/dev/md_root.c for an example. 160 */ 161 md_attach_hook(device_unit(self), &sc->sc_md); 162 #endif 163 164 /* 165 * Initialize and attach the disk structure. 166 */ 167 disk_init(&sc->sc_dkdev, device_xname(self), &mddkdriver); 168 disk_attach(&sc->sc_dkdev); 169 170 if (sc->sc_type != MD_UNCONFIGURED) 171 md_set_disklabel(sc); 172 173 if (!pmf_device_register(self, NULL, NULL)) 174 aprint_error_dev(self, "couldn't establish power handler\n"); 175 } 176 177 static int 178 md_detach(device_t self, int flags) 179 { 180 struct md_softc *sc = device_private(self); 181 int rc; 182 183 rc = 0; 184 mutex_enter(&sc->sc_dkdev.dk_openlock); 185 if (sc->sc_dkdev.dk_openmask == 0 && sc->sc_type == MD_UNCONFIGURED) 186 ; /* nothing to do */ 187 else if ((flags & DETACH_FORCE) == 0) 188 rc = EBUSY; 189 mutex_exit(&sc->sc_dkdev.dk_openlock); 190 191 if (rc != 0) 192 return rc; 193 194 pmf_device_deregister(self); 195 disk_detach(&sc->sc_dkdev); 196 disk_destroy(&sc->sc_dkdev); 197 bufq_free(sc->sc_buflist); 198 mutex_destroy(&sc->sc_lock); 199 cv_destroy(&sc->sc_cv); 200 return 0; 201 } 202 203 /* 204 * operational routines: 205 * open, close, read, write, strategy, 206 * ioctl, dump, size 207 */ 208 209 #if MEMORY_DISK_SERVER 210 static int md_server_loop(struct md_softc *sc); 211 static int md_ioctl_server(struct md_softc *sc, struct md_conf *umd, 212 struct lwp *l); 213 #endif /* MEMORY_DISK_SERVER */ 214 static int md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd, 215 struct lwp *l); 216 217 static int 218 mdsize(dev_t dev) 219 { 220 struct md_softc *sc; 221 int res; 222 223 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 224 if (sc == NULL) 225 return 0; 226 227 mutex_enter(&sc->sc_lock); 228 if (sc->sc_type == MD_UNCONFIGURED) 229 res = 0; 230 else 231 res = sc->sc_size >> DEV_BSHIFT; 232 mutex_exit(&sc->sc_lock); 233 234 return res; 235 } 236 237 static int 238 mdopen(dev_t dev, int flag, int fmt, struct lwp *l) 239 { 240 int unit; 241 int part = DISKPART(dev); 242 int pmask = 1 << part; 243 cfdata_t cf; 244 struct md_softc *sc; 245 struct disk *dk; 246 #ifdef MEMORY_DISK_HOOKS 247 bool configured; 248 #endif 249 250 mutex_enter(&md_device_lock); 251 unit = MD_UNIT(dev); 252 sc = device_lookup_private(&md_cd, unit); 253 if (sc == NULL) { 254 if (part != RAW_PART) { 255 mutex_exit(&md_device_lock); 256 return ENXIO; 257 } 258 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 259 cf->cf_name = md_cd.cd_name; 260 cf->cf_atname = md_cd.cd_name; 261 cf->cf_unit = unit; 262 cf->cf_fstate = FSTATE_STAR; 263 sc = device_private(config_attach_pseudo(cf)); 264 if (sc == NULL) { 265 mutex_exit(&md_device_lock); 266 return ENOMEM; 267 } 268 } 269 270 dk = &sc->sc_dkdev; 271 272 /* 273 * The raw partition is used for ioctl to configure. 274 */ 275 if (part == RAW_PART) 276 goto ok; 277 278 #ifdef MEMORY_DISK_HOOKS 279 /* Call the open hook to allow loading the device. */ 280 configured = (sc->sc_type != MD_UNCONFIGURED); 281 md_open_hook(unit, &sc->sc_md); 282 /* initialize disklabel if the device is configured in open hook */ 283 if (!configured && sc->sc_type != MD_UNCONFIGURED) 284 md_set_disklabel(sc); 285 #endif 286 287 /* 288 * This is a normal, "slave" device, so 289 * enforce initialized. 290 */ 291 if (sc->sc_type == MD_UNCONFIGURED) { 292 mutex_exit(&md_device_lock); 293 return ENXIO; 294 } 295 296 ok: 297 /* XXX duplicates code in dk_open(). Call dk_open(), instead? */ 298 mutex_enter(&dk->dk_openlock); 299 /* Mark our unit as open. */ 300 switch (fmt) { 301 case S_IFCHR: 302 dk->dk_copenmask |= pmask; 303 break; 304 case S_IFBLK: 305 dk->dk_bopenmask |= pmask; 306 break; 307 } 308 309 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 310 311 mutex_exit(&dk->dk_openlock); 312 mutex_exit(&md_device_lock); 313 return 0; 314 } 315 316 static int 317 mdclose(dev_t dev, int flag, int fmt, struct lwp *l) 318 { 319 int part = DISKPART(dev); 320 int pmask = 1 << part; 321 int error; 322 cfdata_t cf; 323 struct md_softc *sc; 324 struct disk *dk; 325 326 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 327 if (sc == NULL) 328 return ENXIO; 329 330 dk = &sc->sc_dkdev; 331 332 mutex_enter(&dk->dk_openlock); 333 334 switch (fmt) { 335 case S_IFCHR: 336 dk->dk_copenmask &= ~pmask; 337 break; 338 case S_IFBLK: 339 dk->dk_bopenmask &= ~pmask; 340 break; 341 } 342 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 343 344 mutex_exit(&dk->dk_openlock); 345 346 mutex_enter(&md_device_lock); 347 cf = device_cfdata(sc->sc_dev); 348 error = config_detach(sc->sc_dev, DETACH_QUIET); 349 if (! error) 350 free(cf, M_DEVBUF); 351 mutex_exit(&md_device_lock); 352 return error; 353 } 354 355 static int 356 mdread(dev_t dev, struct uio *uio, int flags) 357 { 358 struct md_softc *sc; 359 360 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 361 362 if (sc == NULL || sc->sc_type == MD_UNCONFIGURED) 363 return ENXIO; 364 365 return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio)); 366 } 367 368 static int 369 mdwrite(dev_t dev, struct uio *uio, int flags) 370 { 371 struct md_softc *sc; 372 373 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 374 375 if (sc == NULL || sc->sc_type == MD_UNCONFIGURED) 376 return ENXIO; 377 378 return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio)); 379 } 380 381 /* 382 * Handle I/O requests, either directly, or 383 * by passing them to the server process. 384 */ 385 static void 386 mdstrategy(struct buf *bp) 387 { 388 struct md_softc *sc; 389 void * addr; 390 size_t off, xfer; 391 bool is_read; 392 393 sc = device_lookup_private(&md_cd, MD_UNIT(bp->b_dev)); 394 395 mutex_enter(&sc->sc_lock); 396 397 if (sc == NULL || sc->sc_type == MD_UNCONFIGURED) { 398 bp->b_error = ENXIO; 399 goto done; 400 } 401 402 switch (sc->sc_type) { 403 #if MEMORY_DISK_SERVER 404 case MD_UMEM_SERVER: 405 /* Just add this job to the server's queue. */ 406 bufq_put(sc->sc_buflist, bp); 407 cv_signal(&sc->sc_cv); 408 mutex_exit(&sc->sc_lock); 409 /* see md_server_loop() */ 410 /* no biodone in this case */ 411 return; 412 #endif /* MEMORY_DISK_SERVER */ 413 414 case MD_KMEM_FIXED: 415 case MD_KMEM_ALLOCATED: 416 /* These are in kernel space. Access directly. */ 417 is_read = ((bp->b_flags & B_READ) == B_READ); 418 bp->b_resid = bp->b_bcount; 419 off = (bp->b_blkno << DEV_BSHIFT); 420 if (off >= sc->sc_size) { 421 if (is_read) 422 break; /* EOF */ 423 goto set_eio; 424 } 425 xfer = bp->b_resid; 426 if (xfer > (sc->sc_size - off)) 427 xfer = (sc->sc_size - off); 428 addr = (char *)sc->sc_addr + off; 429 disk_busy(&sc->sc_dkdev); 430 if (is_read) 431 memcpy(bp->b_data, addr, xfer); 432 else 433 memcpy(addr, bp->b_data, xfer); 434 disk_unbusy(&sc->sc_dkdev, xfer, is_read); 435 bp->b_resid -= xfer; 436 break; 437 438 default: 439 bp->b_resid = bp->b_bcount; 440 set_eio: 441 bp->b_error = EIO; 442 break; 443 } 444 445 done: 446 mutex_exit(&sc->sc_lock); 447 448 biodone(bp); 449 } 450 451 static int 452 mdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 453 { 454 struct md_softc *sc; 455 struct md_conf *umd; 456 struct disklabel *lp; 457 struct partinfo *pp; 458 int error; 459 460 if ((sc = device_lookup_private(&md_cd, MD_UNIT(dev))) == NULL) 461 return ENXIO; 462 463 mutex_enter(&sc->sc_lock); 464 if (sc->sc_type != MD_UNCONFIGURED) { 465 switch (cmd) { 466 case DIOCGDINFO: 467 lp = (struct disklabel *)data; 468 *lp = *sc->sc_dkdev.dk_label; 469 mutex_exit(&sc->sc_lock); 470 return 0; 471 472 case DIOCGPART: 473 pp = (struct partinfo *)data; 474 pp->disklab = sc->sc_dkdev.dk_label; 475 pp->part = 476 &sc->sc_dkdev.dk_label->d_partitions[DISKPART(dev)]; 477 mutex_exit(&sc->sc_lock); 478 return 0; 479 } 480 } 481 482 /* If this is not the raw partition, punt! */ 483 if (DISKPART(dev) != RAW_PART) { 484 mutex_exit(&sc->sc_lock); 485 return ENOTTY; 486 } 487 488 umd = (struct md_conf *)data; 489 error = EINVAL; 490 switch (cmd) { 491 case MD_GETCONF: 492 *umd = sc->sc_md; 493 error = 0; 494 break; 495 496 case MD_SETCONF: 497 /* Can only set it once. */ 498 if (sc->sc_type != MD_UNCONFIGURED) 499 break; 500 switch (umd->md_type) { 501 case MD_KMEM_ALLOCATED: 502 error = md_ioctl_kalloc(sc, umd, l); 503 break; 504 #if MEMORY_DISK_SERVER 505 case MD_UMEM_SERVER: 506 error = md_ioctl_server(sc, umd, l); 507 break; 508 #endif /* MEMORY_DISK_SERVER */ 509 default: 510 break; 511 } 512 break; 513 } 514 mutex_exit(&sc->sc_lock); 515 return error; 516 } 517 518 static void 519 md_set_disklabel(struct md_softc *sc) 520 { 521 struct disklabel *lp = sc->sc_dkdev.dk_label; 522 struct partition *pp; 523 524 memset(lp, 0, sizeof(*lp)); 525 526 lp->d_secsize = DEV_BSIZE; 527 lp->d_secperunit = sc->sc_size / DEV_BSIZE; 528 if (lp->d_secperunit >= (32*64)) { 529 lp->d_nsectors = 32; 530 lp->d_ntracks = 64; 531 lp->d_ncylinders = lp->d_secperunit / (32*64); 532 } else { 533 lp->d_nsectors = 1; 534 lp->d_ntracks = 1; 535 lp->d_ncylinders = lp->d_secperunit; 536 } 537 lp->d_secpercyl = lp->d_ntracks*lp->d_nsectors; 538 539 strncpy(lp->d_typename, md_cd.cd_name, sizeof(lp->d_typename)); 540 lp->d_type = DTYPE_UNKNOWN; 541 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 542 lp->d_rpm = 3600; 543 lp->d_interleave = 1; 544 lp->d_flags = 0; 545 546 pp = &lp->d_partitions[0]; 547 pp->p_offset = 0; 548 pp->p_size = lp->d_secperunit; 549 pp->p_fstype = FS_BSDFFS; 550 551 pp = &lp->d_partitions[RAW_PART]; 552 pp->p_offset = 0; 553 pp->p_size = lp->d_secperunit; 554 pp->p_fstype = FS_UNUSED; 555 556 lp->d_npartitions = RAW_PART+1; 557 lp->d_magic = DISKMAGIC; 558 lp->d_magic2 = DISKMAGIC; 559 lp->d_checksum = dkcksum(lp); 560 } 561 562 /* 563 * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED) 564 * Just allocate some kernel memory and return. 565 */ 566 static int 567 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd, 568 struct lwp *l) 569 { 570 vaddr_t addr; 571 vsize_t size; 572 573 mutex_exit(&sc->sc_lock); 574 575 /* Sanity check the size. */ 576 size = umd->md_size; 577 addr = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 578 579 mutex_enter(&sc->sc_lock); 580 581 if (!addr) 582 return ENOMEM; 583 584 /* If another thread beat us to configure this unit: fail. */ 585 if (sc->sc_type != MD_UNCONFIGURED) { 586 uvm_km_free(kernel_map, addr, size, UVM_KMF_WIRED); 587 return EINVAL; 588 } 589 590 /* This unit is now configured. */ 591 sc->sc_addr = (void *)addr; /* kernel space */ 592 sc->sc_size = (size_t)size; 593 sc->sc_type = MD_KMEM_ALLOCATED; 594 md_set_disklabel(sc); 595 return 0; 596 } 597 598 #if MEMORY_DISK_SERVER 599 600 /* 601 * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER) 602 * Set config, then become the I/O server for this unit. 603 */ 604 static int 605 md_ioctl_server(struct md_softc *sc, struct md_conf *umd, 606 struct lwp *l) 607 { 608 vaddr_t end; 609 int error; 610 611 KASSERT(mutex_owned(&sc->sc_lock)); 612 613 /* Sanity check addr, size. */ 614 end = (vaddr_t) ((char *)umd->md_addr + umd->md_size); 615 616 if ((end >= VM_MAXUSER_ADDRESS) || 617 (end < ((vaddr_t) umd->md_addr)) ) 618 return EINVAL; 619 620 /* This unit is now configured. */ 621 sc->sc_addr = umd->md_addr; /* user space */ 622 sc->sc_size = umd->md_size; 623 sc->sc_type = MD_UMEM_SERVER; 624 md_set_disklabel(sc); 625 626 /* Become the server daemon */ 627 error = md_server_loop(sc); 628 629 /* This server is now going away! */ 630 sc->sc_type = MD_UNCONFIGURED; 631 sc->sc_addr = 0; 632 sc->sc_size = 0; 633 634 return (error); 635 } 636 637 static int 638 md_server_loop(struct md_softc *sc) 639 { 640 struct buf *bp; 641 void *addr; /* user space address */ 642 size_t off; /* offset into "device" */ 643 size_t xfer; /* amount to transfer */ 644 int error; 645 bool is_read; 646 647 KASSERT(mutex_owned(&sc->sc_lock)); 648 649 for (;;) { 650 /* Wait for some work to arrive. */ 651 while ((bp = bufq_get(sc->sc_buflist)) == NULL) { 652 error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock); 653 if (error) 654 return error; 655 } 656 657 /* Do the transfer to/from user space. */ 658 mutex_exit(&sc->sc_lock); 659 error = 0; 660 is_read = ((bp->b_flags & B_READ) == B_READ); 661 bp->b_resid = bp->b_bcount; 662 off = (bp->b_blkno << DEV_BSHIFT); 663 if (off >= sc->sc_size) { 664 if (is_read) 665 goto done; /* EOF (not an error) */ 666 error = EIO; 667 goto done; 668 } 669 xfer = bp->b_resid; 670 if (xfer > (sc->sc_size - off)) 671 xfer = (sc->sc_size - off); 672 addr = (char *)sc->sc_addr + off; 673 disk_busy(&sc->sc_dkdev); 674 if (is_read) 675 error = copyin(addr, bp->b_data, xfer); 676 else 677 error = copyout(bp->b_data, addr, xfer); 678 disk_unbusy(&sc->sc_dkdev, (error ? 0 : xfer), is_read); 679 if (!error) 680 bp->b_resid -= xfer; 681 682 done: 683 if (error) { 684 bp->b_error = error; 685 } 686 biodone(bp); 687 mutex_enter(&sc->sc_lock); 688 } 689 } 690 #endif /* MEMORY_DISK_SERVER */ 691
Indexes created Tue Sep 30 20:09:53 GMT 2025