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