1 /* 2 * ---------------------------------------------------------------------------- 3 * "THE BEER-WARE LICENSE" (Revision 42): 4 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you 5 * can do whatever you want with this stuff. If we meet some day, and you think 6 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 7 * ---------------------------------------------------------------------------- 8 * 9 * $FreeBSD: src/sys/dev/md/md.c,v 1.8.2.2 2002/08/19 17:43:34 jdp Exp $ 10 * $DragonFly: src/sys/dev/disk/md/md.c,v 1.20 2008/09/07 08:09:39 swildner Exp $ 11 * 12 */ 13 14 #include "opt_md.h" /* We have adopted some tasks from MFS */ 15 16 #include <sys/param.h> 17 #include <sys/systm.h> 18 #include <sys/buf.h> 19 #include <sys/conf.h> 20 #include <sys/devicestat.h> 21 #include <sys/disk.h> 22 #include <sys/kernel.h> 23 #include <sys/malloc.h> 24 #include <sys/sysctl.h> 25 #include <sys/linker.h> 26 #include <sys/proc.h> 27 #include <sys/buf2.h> 28 #include <sys/thread2.h> 29 #include <sys/queue.h> 30 #include <sys/udev.h> 31 32 #ifndef MD_NSECT 33 #define MD_NSECT (10000 * 2) 34 #endif 35 36 MALLOC_DEFINE(M_MD, "MD disk", "Memory Disk"); 37 MALLOC_DEFINE(M_MDSECT, "MD sectors", "Memory Disk Sectors"); 38 39 static int md_debug; 40 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0, ""); 41 42 #if defined(MD_ROOT) && defined(MD_ROOT_SIZE) 43 /* Image gets put here: */ 44 static u_char mfs_root[MD_ROOT_SIZE*1024] = "MFS Filesystem goes here"; 45 static u_char end_mfs_root[] __unused = "MFS Filesystem had better STOP here"; 46 #endif 47 48 static int mdrootready; 49 50 #define CDEV_MAJOR 95 51 52 static d_strategy_t mdstrategy; 53 static d_strategy_t mdstrategy_preload; 54 static d_strategy_t mdstrategy_malloc; 55 static d_open_t mdopen; 56 static d_close_t mdclose; 57 static d_ioctl_t mdioctl; 58 59 static struct dev_ops md_ops = { 60 { "md", CDEV_MAJOR, D_DISK | D_CANFREE | D_MEMDISK | D_TRACKCLOSE}, 61 .d_open = mdopen, 62 .d_close = mdclose, 63 .d_read = physread, 64 .d_write = physwrite, 65 .d_ioctl = mdioctl, 66 .d_strategy = mdstrategy, 67 }; 68 69 struct md_s { 70 int unit; 71 struct devstat stats; 72 struct bio_queue_head bio_queue; 73 struct disk disk; 74 cdev_t dev; 75 int busy; 76 enum { /* Memory disk type */ 77 MD_MALLOC, 78 MD_PRELOAD 79 } type; 80 unsigned nsect; 81 82 /* MD_MALLOC related fields */ 83 unsigned nsecp; 84 u_char **secp; 85 86 /* MD_PRELOAD related fields */ 87 u_char *pl_ptr; 88 unsigned pl_len; 89 TAILQ_ENTRY(md_s) link; 90 }; 91 TAILQ_HEAD(mdshead, md_s) mdlist = TAILQ_HEAD_INITIALIZER(mdlist); 92 93 static int mdunits; 94 static int refcnt; 95 96 static struct md_s *mdcreate(unsigned); 97 static void mdcreate_malloc(void); 98 static int mdinit(module_t, int, void *); 99 static void md_drvinit(void *); 100 static int md_drvcleanup(void); 101 102 static int 103 mdinit(module_t mod, int cmd, void *arg) 104 { 105 int ret = 0; 106 107 switch(cmd) { 108 case MOD_LOAD: 109 TAILQ_INIT(&mdlist); 110 md_drvinit(NULL); 111 break; 112 case MOD_UNLOAD: 113 ret = md_drvcleanup(); 114 break; 115 default: 116 ret = EINVAL; 117 break; 118 } 119 120 return (ret); 121 } 122 123 static int 124 mdopen(struct dev_open_args *ap) 125 { 126 cdev_t dev = ap->a_head.a_dev; 127 struct md_s *sc; 128 129 if (md_debug) 130 kprintf("mdopen(%s %x %x)\n", 131 devtoname(dev), ap->a_oflags, ap->a_devtype); 132 133 sc = dev->si_drv1; 134 if (sc->unit + 1 == mdunits) 135 mdcreate_malloc(); 136 137 atomic_add_int(&refcnt, 1); 138 return (0); 139 } 140 141 static int 142 mdclose(struct dev_close_args *ap) 143 { 144 cdev_t dev = ap->a_head.a_dev; 145 struct md_s *sc; 146 147 if (md_debug) 148 kprintf("mdclose(%s %x %x)\n", 149 devtoname(dev), ap->a_fflag, ap->a_devtype); 150 sc = dev->si_drv1; 151 atomic_add_int(&refcnt, -1); 152 153 return (0); 154 } 155 156 static int 157 mdioctl(struct dev_ioctl_args *ap) 158 { 159 cdev_t dev = ap->a_head.a_dev; 160 161 if (md_debug) 162 kprintf("mdioctl(%s %lx %p %x)\n", 163 devtoname(dev), ap->a_cmd, ap->a_data, ap->a_fflag); 164 165 return (ENOIOCTL); 166 } 167 168 static int 169 mdstrategy(struct dev_strategy_args *ap) 170 { 171 cdev_t dev = ap->a_head.a_dev; 172 struct bio *bio = ap->a_bio; 173 struct buf *bp = bio->bio_buf; 174 struct md_s *sc; 175 176 if (md_debug > 1) { 177 kprintf("mdstrategy(%p) %s %08x, %lld, %d, %p)\n", 178 bp, devtoname(dev), bp->b_flags, 179 (long long)bio->bio_offset, 180 bp->b_bcount, bp->b_data); 181 } 182 bio->bio_driver_info = dev; 183 sc = dev->si_drv1; 184 if (sc->type == MD_MALLOC) { 185 mdstrategy_malloc(ap); 186 } else { 187 mdstrategy_preload(ap); 188 } 189 return(0); 190 } 191 192 193 static int 194 mdstrategy_malloc(struct dev_strategy_args *ap) 195 { 196 cdev_t dev = ap->a_head.a_dev; 197 struct bio *bio = ap->a_bio; 198 struct buf *bp = bio->bio_buf; 199 unsigned secno, nsec, secval, uc; 200 u_char *secp, **secpp, *dst; 201 struct md_s *sc; 202 int i; 203 204 if (md_debug > 1) 205 kprintf("mdstrategy_malloc(%p) %s %08xx, %lld, %d, %p)\n", 206 bp, devtoname(dev), bp->b_flags, 207 (long long)bio->bio_offset, 208 bp->b_bcount, bp->b_data); 209 210 sc = dev->si_drv1; 211 212 crit_enter(); 213 214 bioqdisksort(&sc->bio_queue, bio); 215 216 if (sc->busy) { 217 crit_exit(); 218 return(0); 219 } 220 221 sc->busy++; 222 223 while (1) { 224 bio = bioq_first(&sc->bio_queue); 225 if (bio == NULL) { 226 crit_exit(); 227 break; 228 } 229 crit_exit(); 230 bioq_remove(&sc->bio_queue, bio); 231 bp = bio->bio_buf; 232 233 devstat_start_transaction(&sc->stats); 234 235 switch (bp->b_cmd) { 236 case BUF_CMD_FREEBLKS: 237 case BUF_CMD_READ: 238 case BUF_CMD_WRITE: 239 break; 240 default: 241 panic("md: bad b_cmd %d", bp->b_cmd); 242 } 243 244 nsec = bp->b_bcount >> DEV_BSHIFT; 245 secno = (unsigned)(bio->bio_offset >> DEV_BSHIFT); 246 dst = bp->b_data; 247 while (nsec--) { 248 if (secno < sc->nsecp) { 249 secpp = &sc->secp[secno]; 250 if ((u_int)(uintptr_t)*secpp > 255) { 251 secp = *secpp; 252 secval = 0; 253 } else { 254 secp = 0; 255 secval = (u_int)(uintptr_t)*secpp; 256 } 257 } else { 258 secpp = 0; 259 secp = 0; 260 secval = 0; 261 } 262 if (md_debug > 2) 263 kprintf("%08x %p %p %d\n", bp->b_flags, secpp, secp, secval); 264 265 switch (bp->b_cmd) { 266 case BUF_CMD_FREEBLKS: 267 if (secpp) { 268 if (secp) 269 FREE(secp, M_MDSECT); 270 *secpp = 0; 271 } 272 break; 273 case BUF_CMD_READ: 274 if (secp) { 275 bcopy(secp, dst, DEV_BSIZE); 276 } else if (secval) { 277 for (i = 0; i < DEV_BSIZE; i++) 278 dst[i] = secval; 279 } else { 280 bzero(dst, DEV_BSIZE); 281 } 282 break; 283 case BUF_CMD_WRITE: 284 uc = dst[0]; 285 for (i = 1; i < DEV_BSIZE; i++) 286 if (dst[i] != uc) 287 break; 288 if (i == DEV_BSIZE && !uc) { 289 if (secp) 290 FREE(secp, M_MDSECT); 291 if (secpp) 292 *secpp = (u_char *)(uintptr_t)uc; 293 } else { 294 if (!secpp) { 295 MALLOC(secpp, u_char **, (secno + nsec + 1) * sizeof(u_char *), M_MD, M_WAITOK | M_ZERO); 296 bcopy(sc->secp, secpp, sc->nsecp * sizeof(u_char *)); 297 FREE(sc->secp, M_MD); 298 sc->secp = secpp; 299 sc->nsecp = secno + nsec + 1; 300 secpp = &sc->secp[secno]; 301 } 302 if (i == DEV_BSIZE) { 303 if (secp) 304 FREE(secp, M_MDSECT); 305 *secpp = (u_char *)(uintptr_t)uc; 306 } else { 307 if (!secp) 308 MALLOC(secp, u_char *, DEV_BSIZE, M_MDSECT, M_WAITOK); 309 bcopy(dst, secp, DEV_BSIZE); 310 311 *secpp = secp; 312 } 313 } 314 break; 315 default: 316 panic("md: bad b_cmd %d", bp->b_cmd); 317 318 } 319 secno++; 320 dst += DEV_BSIZE; 321 } 322 bp->b_resid = 0; 323 devstat_end_transaction_buf(&sc->stats, bp); 324 biodone(bio); 325 crit_enter(); 326 } 327 sc->busy = 0; 328 return(0); 329 } 330 331 332 static int 333 mdstrategy_preload(struct dev_strategy_args *ap) 334 { 335 cdev_t dev = ap->a_head.a_dev; 336 struct bio *bio = ap->a_bio; 337 struct buf *bp = bio->bio_buf; 338 struct md_s *sc; 339 340 if (md_debug > 1) 341 kprintf("mdstrategy_preload(%p) %s %08x, %lld, %d, %p)\n", 342 bp, devtoname(dev), bp->b_flags, 343 (long long)bio->bio_offset, 344 bp->b_bcount, bp->b_data); 345 346 sc = dev->si_drv1; 347 348 crit_enter(); 349 350 bioqdisksort(&sc->bio_queue, bio); 351 352 if (sc->busy) { 353 crit_exit(); 354 return(0); 355 } 356 357 sc->busy++; 358 359 while (1) { 360 bio = bioq_first(&sc->bio_queue); 361 if (bio) 362 bioq_remove(&sc->bio_queue, bio); 363 crit_exit(); 364 if (bio == NULL) 365 break; 366 367 devstat_start_transaction(&sc->stats); 368 369 switch (bp->b_cmd) { 370 case BUF_CMD_FREEBLKS: 371 break; 372 case BUF_CMD_READ: 373 bcopy(sc->pl_ptr + bio->bio_offset, 374 bp->b_data, bp->b_bcount); 375 break; 376 case BUF_CMD_WRITE: 377 bcopy(bp->b_data, sc->pl_ptr + bio->bio_offset, 378 bp->b_bcount); 379 break; 380 default: 381 panic("md: bad cmd %d\n", bp->b_cmd); 382 } 383 bp->b_resid = 0; 384 devstat_end_transaction_buf(&sc->stats, bp); 385 biodone(bio); 386 crit_enter(); 387 } 388 sc->busy = 0; 389 return(0); 390 } 391 392 static struct md_s * 393 mdcreate(unsigned length) 394 { 395 struct md_s *sc; 396 struct disk_info info; 397 398 MALLOC(sc, struct md_s *,sizeof(*sc), M_MD, M_WAITOK | M_ZERO); 399 sc->unit = mdunits++; 400 bioq_init(&sc->bio_queue); 401 devstat_add_entry(&sc->stats, "md", sc->unit, DEV_BSIZE, 402 DEVSTAT_NO_ORDERED_TAGS, 403 DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER, 404 DEVSTAT_PRIORITY_OTHER); 405 sc->dev = disk_create(sc->unit, &sc->disk, &md_ops); 406 sc->dev->si_drv1 = sc; 407 sc->dev->si_iosize_max = DFLTPHYS; 408 disk_setdisktype(&sc->disk, "memory"); 409 410 bzero(&info, sizeof(info)); 411 info.d_media_blksize = DEV_BSIZE; /* mandatory */ 412 info.d_media_blocks = length / DEV_BSIZE; 413 414 info.d_secpertrack = 1024; /* optional */ 415 info.d_nheads = 1; 416 info.d_secpercyl = info.d_secpertrack * info.d_nheads; 417 info.d_ncylinders = (u_int)(info.d_media_blocks / info.d_secpercyl); 418 disk_setdiskinfo(&sc->disk, &info); 419 TAILQ_INSERT_HEAD(&mdlist, sc, link); 420 421 return (sc); 422 } 423 424 425 static void 426 mdcreate_preload(u_char *image, unsigned length) 427 { 428 struct md_s *sc; 429 430 sc = mdcreate(length); 431 sc->type = MD_PRELOAD; 432 sc->nsect = length / DEV_BSIZE; 433 sc->pl_ptr = image; 434 sc->pl_len = length; 435 436 if (sc->unit == 0) 437 mdrootready = 1; 438 } 439 440 static void 441 mdcreate_malloc(void) 442 { 443 struct md_s *sc; 444 445 sc = mdcreate(MD_NSECT*DEV_BSIZE); 446 sc->type = MD_MALLOC; 447 448 sc->nsect = MD_NSECT; /* for now */ 449 MALLOC(sc->secp, u_char **, sizeof(u_char *), M_MD, M_WAITOK | M_ZERO); 450 sc->nsecp = 1; 451 kprintf("md%d: Malloc disk\n", sc->unit); 452 } 453 454 static int 455 md_drvcleanup(void) 456 { 457 458 int secno; 459 struct md_s *sc, *sc_temp; 460 461 if (atomic_fetchadd_int(&refcnt, 0) != 0) 462 return EBUSY; 463 464 /* 465 * Go through all the md devices, freeing up all the 466 * memory allocated for sectors, and the md_s struct 467 * itself. 468 */ 469 TAILQ_FOREACH_MUTABLE(sc, &mdlist, link, sc_temp) { 470 for (secno = 0; secno < sc->nsecp; secno++) { 471 if ((u_int)(uintptr_t)sc->secp[secno] > 255) 472 FREE(sc->secp[secno], M_MDSECT); 473 } 474 475 if (sc->dev != NULL) 476 disk_destroy(&sc->disk); 477 478 devstat_remove_entry(&sc->stats); 479 TAILQ_REMOVE(&mdlist, sc, link); 480 481 FREE(sc->secp, M_MD); 482 FREE(sc, M_MD); 483 } 484 485 return 0; 486 487 } 488 489 static void 490 md_drvinit(void *unused) 491 { 492 493 caddr_t mod; 494 caddr_t c; 495 u_char *ptr, *name, *type; 496 unsigned len; 497 498 #ifdef MD_ROOT_SIZE 499 mdcreate_preload(mfs_root, MD_ROOT_SIZE*1024); 500 #endif 501 mod = NULL; 502 while ((mod = preload_search_next_name(mod)) != NULL) { 503 name = (char *)preload_search_info(mod, MODINFO_NAME); 504 type = (char *)preload_search_info(mod, MODINFO_TYPE); 505 if (name == NULL) 506 continue; 507 if (type == NULL) 508 continue; 509 if (strcmp(type, "md_image") && strcmp(type, "mfs_root")) 510 continue; 511 c = preload_search_info(mod, MODINFO_ADDR); 512 ptr = *(u_char **)c; 513 c = preload_search_info(mod, MODINFO_SIZE); 514 len = *(unsigned *)c; 515 kprintf("md%d: Preloaded image <%s> %d bytes at %p\n", 516 mdunits, name, len, ptr); 517 mdcreate_preload(ptr, len); 518 } 519 mdcreate_malloc(); 520 } 521 522 DEV_MODULE(md, mdinit, NULL); 523 524 #ifdef MD_ROOT 525 static void 526 md_takeroot(void *junk) 527 { 528 if (mdrootready) 529 rootdevnames[0] = "ufs:/dev/md0s0"; 530 } 531 532 SYSINIT(md_root, SI_SUB_MOUNT_ROOT, SI_ORDER_FIRST, md_takeroot, NULL); 533 #endif 534