md.c revision 1.73
1 /* $NetBSD: md.c,v 1.73 2015/01/02 19:42:06 christos 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 1.73 2015/01/02 19:42:06 christos 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 .d_open = mdopen, 111 .d_close = mdclose, 112 .d_strategy = mdstrategy, 113 .d_ioctl = mdioctl, 114 .d_dump = nodump, 115 .d_psize = mdsize, 116 .d_discard = nodiscard, 117 .d_flag = D_DISK | D_MPSAFE 118 }; 119 120 const struct cdevsw md_cdevsw = { 121 .d_open = mdopen, 122 .d_close = mdclose, 123 .d_read = mdread, 124 .d_write = mdwrite, 125 .d_ioctl = mdioctl, 126 .d_stop = nostop, 127 .d_tty = notty, 128 .d_poll = nopoll, 129 .d_mmap = nommap, 130 .d_kqfilter = nokqfilter, 131 .d_discard = nodiscard, 132 .d_flag = D_DISK 133 }; 134 135 static struct dkdriver mddkdriver = { mdstrategy, NULL }; 136 137 extern struct cfdriver md_cd; 138 CFATTACH_DECL3_NEW(md, sizeof(struct md_softc), 139 0, md_attach, md_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN); 140 141 static kmutex_t md_device_lock; /* Protect unit creation / deletion. */ 142 extern size_t md_root_size; 143 144 static void md_set_disklabel(struct md_softc *); 145 146 /* 147 * This is called if we are configured as a pseudo-device 148 */ 149 void 150 mdattach(int n) 151 { 152 153 mutex_init(&md_device_lock, MUTEX_DEFAULT, IPL_NONE); 154 if (config_cfattach_attach(md_cd.cd_name, &md_ca)) { 155 aprint_error("%s: cfattach_attach failed\n", md_cd.cd_name); 156 return; 157 } 158 } 159 160 static void 161 md_attach(device_t parent, device_t self, void *aux) 162 { 163 struct md_softc *sc = device_private(self); 164 165 sc->sc_dev = self; 166 sc->sc_type = MD_UNCONFIGURED; 167 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 168 cv_init(&sc->sc_cv, "mdidle"); 169 bufq_alloc(&sc->sc_buflist, "fcfs", 0); 170 171 /* XXX - Could accept aux info here to set the config. */ 172 #ifdef MEMORY_DISK_HOOKS 173 /* 174 * This external function might setup a pre-loaded disk. 175 * All it would need to do is setup the md_conf struct. 176 * See sys/dev/md_root.c for an example. 177 */ 178 md_attach_hook(device_unit(self), &sc->sc_md); 179 #endif 180 181 /* 182 * Initialize and attach the disk structure. 183 */ 184 disk_init(&sc->sc_dkdev, device_xname(self), &mddkdriver); 185 disk_attach(&sc->sc_dkdev); 186 187 if (sc->sc_type != MD_UNCONFIGURED) 188 md_set_disklabel(sc); 189 190 if (!pmf_device_register(self, NULL, NULL)) 191 aprint_error_dev(self, "couldn't establish power handler\n"); 192 } 193 194 static int 195 md_detach(device_t self, int flags) 196 { 197 struct md_softc *sc = device_private(self); 198 int rc; 199 200 rc = 0; 201 mutex_enter(&sc->sc_dkdev.dk_openlock); 202 if (sc->sc_dkdev.dk_openmask == 0 && sc->sc_type == MD_UNCONFIGURED) 203 ; /* nothing to do */ 204 else if ((flags & DETACH_FORCE) == 0) 205 rc = EBUSY; 206 mutex_exit(&sc->sc_dkdev.dk_openlock); 207 208 if (rc != 0) 209 return rc; 210 211 pmf_device_deregister(self); 212 disk_detach(&sc->sc_dkdev); 213 disk_destroy(&sc->sc_dkdev); 214 bufq_free(sc->sc_buflist); 215 mutex_destroy(&sc->sc_lock); 216 cv_destroy(&sc->sc_cv); 217 return 0; 218 } 219 220 /* 221 * operational routines: 222 * open, close, read, write, strategy, 223 * ioctl, dump, size 224 */ 225 226 #if MEMORY_DISK_SERVER 227 static int md_server_loop(struct md_softc *sc); 228 static int md_ioctl_server(struct md_softc *sc, struct md_conf *umd, 229 struct lwp *l); 230 #endif /* MEMORY_DISK_SERVER */ 231 static int md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd, 232 struct lwp *l); 233 234 static int 235 mdsize(dev_t dev) 236 { 237 struct md_softc *sc; 238 int res; 239 240 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 241 if (sc == NULL) 242 return 0; 243 244 mutex_enter(&sc->sc_lock); 245 if (sc->sc_type == MD_UNCONFIGURED) 246 res = 0; 247 else 248 res = sc->sc_size >> DEV_BSHIFT; 249 mutex_exit(&sc->sc_lock); 250 251 return res; 252 } 253 254 static int 255 mdopen(dev_t dev, int flag, int fmt, struct lwp *l) 256 { 257 int unit; 258 int part = DISKPART(dev); 259 int pmask = 1 << part; 260 cfdata_t cf; 261 struct md_softc *sc; 262 struct disk *dk; 263 #ifdef MEMORY_DISK_HOOKS 264 bool configured; 265 #endif 266 267 mutex_enter(&md_device_lock); 268 unit = MD_UNIT(dev); 269 sc = device_lookup_private(&md_cd, unit); 270 if (sc == NULL) { 271 if (part != RAW_PART) { 272 mutex_exit(&md_device_lock); 273 return ENXIO; 274 } 275 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 276 cf->cf_name = md_cd.cd_name; 277 cf->cf_atname = md_cd.cd_name; 278 cf->cf_unit = unit; 279 cf->cf_fstate = FSTATE_STAR; 280 sc = device_private(config_attach_pseudo(cf)); 281 if (sc == NULL) { 282 mutex_exit(&md_device_lock); 283 return ENOMEM; 284 } 285 } 286 287 dk = &sc->sc_dkdev; 288 289 /* 290 * The raw partition is used for ioctl to configure. 291 */ 292 if (part == RAW_PART) 293 goto ok; 294 295 #ifdef MEMORY_DISK_HOOKS 296 /* Call the open hook to allow loading the device. */ 297 configured = (sc->sc_type != MD_UNCONFIGURED); 298 md_open_hook(unit, &sc->sc_md); 299 /* initialize disklabel if the device is configured in open hook */ 300 if (!configured && sc->sc_type != MD_UNCONFIGURED) 301 md_set_disklabel(sc); 302 #endif 303 304 /* 305 * This is a normal, "slave" device, so 306 * enforce initialized. 307 */ 308 if (sc->sc_type == MD_UNCONFIGURED) { 309 mutex_exit(&md_device_lock); 310 return ENXIO; 311 } 312 313 ok: 314 /* XXX duplicates code in dk_open(). Call dk_open(), instead? */ 315 mutex_enter(&dk->dk_openlock); 316 /* Mark our unit as open. */ 317 switch (fmt) { 318 case S_IFCHR: 319 dk->dk_copenmask |= pmask; 320 break; 321 case S_IFBLK: 322 dk->dk_bopenmask |= pmask; 323 break; 324 } 325 326 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 327 328 mutex_exit(&dk->dk_openlock); 329 mutex_exit(&md_device_lock); 330 return 0; 331 } 332 333 static int 334 mdclose(dev_t dev, int flag, int fmt, struct lwp *l) 335 { 336 int part = DISKPART(dev); 337 int pmask = 1 << part; 338 int error; 339 cfdata_t cf; 340 struct md_softc *sc; 341 struct disk *dk; 342 343 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 344 if (sc == NULL) 345 return ENXIO; 346 347 dk = &sc->sc_dkdev; 348 349 mutex_enter(&dk->dk_openlock); 350 351 switch (fmt) { 352 case S_IFCHR: 353 dk->dk_copenmask &= ~pmask; 354 break; 355 case S_IFBLK: 356 dk->dk_bopenmask &= ~pmask; 357 break; 358 } 359 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask; 360 if (dk->dk_openmask != 0) { 361 mutex_exit(&dk->dk_openlock); 362 return 0; 363 } 364 365 mutex_exit(&dk->dk_openlock); 366 367 mutex_enter(&md_device_lock); 368 cf = device_cfdata(sc->sc_dev); 369 error = config_detach(sc->sc_dev, DETACH_QUIET); 370 if (! error) 371 free(cf, M_DEVBUF); 372 mutex_exit(&md_device_lock); 373 return error; 374 } 375 376 static int 377 mdread(dev_t dev, struct uio *uio, int flags) 378 { 379 struct md_softc *sc; 380 381 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 382 383 if (sc == NULL || sc->sc_type == MD_UNCONFIGURED) 384 return ENXIO; 385 386 return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio)); 387 } 388 389 static int 390 mdwrite(dev_t dev, struct uio *uio, int flags) 391 { 392 struct md_softc *sc; 393 394 sc = device_lookup_private(&md_cd, MD_UNIT(dev)); 395 396 if (sc == NULL || sc->sc_type == MD_UNCONFIGURED) 397 return ENXIO; 398 399 return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio)); 400 } 401 402 /* 403 * Handle I/O requests, either directly, or 404 * by passing them to the server process. 405 */ 406 static void 407 mdstrategy(struct buf *bp) 408 { 409 struct md_softc *sc; 410 void * addr; 411 size_t off, xfer; 412 bool is_read; 413 414 sc = device_lookup_private(&md_cd, MD_UNIT(bp->b_dev)); 415 416 mutex_enter(&sc->sc_lock); 417 418 if (sc == NULL || sc->sc_type == MD_UNCONFIGURED) { 419 bp->b_error = ENXIO; 420 goto done; 421 } 422 423 switch (sc->sc_type) { 424 #if MEMORY_DISK_SERVER 425 case MD_UMEM_SERVER: 426 /* Just add this job to the server's queue. */ 427 bufq_put(sc->sc_buflist, bp); 428 cv_signal(&sc->sc_cv); 429 mutex_exit(&sc->sc_lock); 430 /* see md_server_loop() */ 431 /* no biodone in this case */ 432 return; 433 #endif /* MEMORY_DISK_SERVER */ 434 435 case MD_KMEM_FIXED: 436 case MD_KMEM_ALLOCATED: 437 /* These are in kernel space. Access directly. */ 438 is_read = ((bp->b_flags & B_READ) == B_READ); 439 bp->b_resid = bp->b_bcount; 440 off = (bp->b_blkno << DEV_BSHIFT); 441 if (off >= sc->sc_size) { 442 if (is_read) 443 break; /* EOF */ 444 goto set_eio; 445 } 446 xfer = bp->b_resid; 447 if (xfer > (sc->sc_size - off)) 448 xfer = (sc->sc_size - off); 449 addr = (char *)sc->sc_addr + off; 450 disk_busy(&sc->sc_dkdev); 451 if (is_read) 452 memcpy(bp->b_data, addr, xfer); 453 else 454 memcpy(addr, bp->b_data, xfer); 455 disk_unbusy(&sc->sc_dkdev, xfer, is_read); 456 bp->b_resid -= xfer; 457 break; 458 459 default: 460 bp->b_resid = bp->b_bcount; 461 set_eio: 462 bp->b_error = EIO; 463 break; 464 } 465 466 done: 467 mutex_exit(&sc->sc_lock); 468 469 biodone(bp); 470 } 471 472 static int 473 mdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 474 { 475 struct md_softc *sc; 476 struct md_conf *umd; 477 int error; 478 479 if ((sc = device_lookup_private(&md_cd, MD_UNIT(dev))) == NULL) 480 return ENXIO; 481 482 mutex_enter(&sc->sc_lock); 483 if (sc->sc_type != MD_UNCONFIGURED) { 484 error = disk_ioctl(&sc->sc_dkdev, dev, cmd, data, flag, l); 485 if (error != EPASSTHROUGH) { 486 mutex_exit(&sc->sc_lock); 487 return 0; 488 } 489 } 490 491 /* If this is not the raw partition, punt! */ 492 if (DISKPART(dev) != RAW_PART) { 493 mutex_exit(&sc->sc_lock); 494 return ENOTTY; 495 } 496 497 umd = (struct md_conf *)data; 498 error = EINVAL; 499 switch (cmd) { 500 case MD_GETCONF: 501 *umd = sc->sc_md; 502 error = 0; 503 break; 504 505 case MD_SETCONF: 506 /* Can only set it once. */ 507 if (sc->sc_type != MD_UNCONFIGURED) 508 break; 509 switch (umd->md_type) { 510 case MD_KMEM_ALLOCATED: 511 error = md_ioctl_kalloc(sc, umd, l); 512 break; 513 #if MEMORY_DISK_SERVER 514 case MD_UMEM_SERVER: 515 error = md_ioctl_server(sc, umd, l); 516 break; 517 #endif /* MEMORY_DISK_SERVER */ 518 default: 519 break; 520 } 521 break; 522 } 523 mutex_exit(&sc->sc_lock); 524 return error; 525 } 526 527 static void 528 md_set_disklabel(struct md_softc *sc) 529 { 530 struct disklabel *lp = sc->sc_dkdev.dk_label; 531 struct partition *pp; 532 533 memset(lp, 0, sizeof(*lp)); 534 535 lp->d_secsize = DEV_BSIZE; 536 lp->d_secperunit = sc->sc_size / DEV_BSIZE; 537 if (lp->d_secperunit >= (32*64)) { 538 lp->d_nsectors = 32; 539 lp->d_ntracks = 64; 540 lp->d_ncylinders = lp->d_secperunit / (32*64); 541 } else { 542 lp->d_nsectors = 1; 543 lp->d_ntracks = 1; 544 lp->d_ncylinders = lp->d_secperunit; 545 } 546 lp->d_secpercyl = lp->d_ntracks*lp->d_nsectors; 547 548 strncpy(lp->d_typename, md_cd.cd_name, sizeof(lp->d_typename)); 549 lp->d_type = DKTYPE_MD; 550 strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname)); 551 lp->d_rpm = 3600; 552 lp->d_interleave = 1; 553 lp->d_flags = 0; 554 555 pp = &lp->d_partitions[0]; 556 pp->p_offset = 0; 557 pp->p_size = lp->d_secperunit; 558 pp->p_fstype = FS_BSDFFS; 559 560 pp = &lp->d_partitions[RAW_PART]; 561 pp->p_offset = 0; 562 pp->p_size = lp->d_secperunit; 563 pp->p_fstype = FS_UNUSED; 564 565 lp->d_npartitions = RAW_PART+1; 566 lp->d_magic = DISKMAGIC; 567 lp->d_magic2 = DISKMAGIC; 568 lp->d_checksum = dkcksum(lp); 569 } 570 571 /* 572 * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED) 573 * Just allocate some kernel memory and return. 574 */ 575 static int 576 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd, 577 struct lwp *l) 578 { 579 vaddr_t addr; 580 vsize_t size; 581 582 mutex_exit(&sc->sc_lock); 583 584 /* Sanity check the size. */ 585 size = umd->md_size; 586 addr = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 587 588 mutex_enter(&sc->sc_lock); 589 590 if (!addr) 591 return ENOMEM; 592 593 /* If another thread beat us to configure this unit: fail. */ 594 if (sc->sc_type != MD_UNCONFIGURED) { 595 uvm_km_free(kernel_map, addr, size, UVM_KMF_WIRED); 596 return EINVAL; 597 } 598 599 /* This unit is now configured. */ 600 sc->sc_addr = (void *)addr; /* kernel space */ 601 sc->sc_size = (size_t)size; 602 sc->sc_type = MD_KMEM_ALLOCATED; 603 md_set_disklabel(sc); 604 return 0; 605 } 606 607 #if MEMORY_DISK_SERVER 608 609 /* 610 * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER) 611 * Set config, then become the I/O server for this unit. 612 */ 613 static int 614 md_ioctl_server(struct md_softc *sc, struct md_conf *umd, 615 struct lwp *l) 616 { 617 vaddr_t end; 618 int error; 619 620 KASSERT(mutex_owned(&sc->sc_lock)); 621 622 /* Sanity check addr, size. */ 623 end = (vaddr_t) ((char *)umd->md_addr + umd->md_size); 624 625 if ((end >= VM_MAXUSER_ADDRESS) || 626 (end < ((vaddr_t) umd->md_addr)) ) 627 return EINVAL; 628 629 /* This unit is now configured. */ 630 sc->sc_addr = umd->md_addr; /* user space */ 631 sc->sc_size = umd->md_size; 632 sc->sc_type = MD_UMEM_SERVER; 633 md_set_disklabel(sc); 634 635 /* Become the server daemon */ 636 error = md_server_loop(sc); 637 638 /* This server is now going away! */ 639 sc->sc_type = MD_UNCONFIGURED; 640 sc->sc_addr = 0; 641 sc->sc_size = 0; 642 643 return (error); 644 } 645 646 static int 647 md_server_loop(struct md_softc *sc) 648 { 649 struct buf *bp; 650 void *addr; /* user space address */ 651 size_t off; /* offset into "device" */ 652 size_t xfer; /* amount to transfer */ 653 int error; 654 bool is_read; 655 656 KASSERT(mutex_owned(&sc->sc_lock)); 657 658 for (;;) { 659 /* Wait for some work to arrive. */ 660 while ((bp = bufq_get(sc->sc_buflist)) == NULL) { 661 error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock); 662 if (error) 663 return error; 664 } 665 666 /* Do the transfer to/from user space. */ 667 mutex_exit(&sc->sc_lock); 668 error = 0; 669 is_read = ((bp->b_flags & B_READ) == B_READ); 670 bp->b_resid = bp->b_bcount; 671 off = (bp->b_blkno << DEV_BSHIFT); 672 if (off >= sc->sc_size) { 673 if (is_read) 674 goto done; /* EOF (not an error) */ 675 error = EIO; 676 goto done; 677 } 678 xfer = bp->b_resid; 679 if (xfer > (sc->sc_size - off)) 680 xfer = (sc->sc_size - off); 681 addr = (char *)sc->sc_addr + off; 682 disk_busy(&sc->sc_dkdev); 683 if (is_read) 684 error = copyin(addr, bp->b_data, xfer); 685 else 686 error = copyout(bp->b_data, addr, xfer); 687 disk_unbusy(&sc->sc_dkdev, (error ? 0 : xfer), is_read); 688 if (!error) 689 bp->b_resid -= xfer; 690 691 done: 692 if (error) { 693 bp->b_error = error; 694 } 695 biodone(bp); 696 mutex_enter(&sc->sc_lock); 697 } 698 } 699 #endif /* MEMORY_DISK_SERVER */ 700
Indexes created Thu Sep 25 16:09:42 GMT 2025