Home | History | Annotate | Line # | Download | only in dev
md.c revision 1.51.18.2
      1 /*	$NetBSD: md.c,v 1.51.18.2 2008/06/02 13:23:11 mjf 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  * 3. The name of the author may not be used to endorse or promote products
     16  *    derived from this software without specific prior written permission.
     17  * 4. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *      This product includes software developed by
     20  *			Gordon W. Ross and Leo Weppelman.
     21  *
     22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     32  */
     33 
     34 /*
     35  * This implements a general-purpose memory-disk.
     36  * See md.h for notes on the config types.
     37  *
     38  * Note that this driver provides the same functionality
     39  * as the MFS filesystem hack, but this is better because
     40  * you can use this for any filesystem type you'd like!
     41  *
     42  * Credit for most of the kmem ramdisk code goes to:
     43  *   Leo Weppelman (atari) and Phil Nelson (pc532)
     44  * Credit for the ideas behind the "user space memory" code goes
     45  * to the authors of the MFS implementation.
     46  */
     47 
     48 #include <sys/cdefs.h>
     49 __KERNEL_RCSID(0, "$NetBSD: md.c,v 1.51.18.2 2008/06/02 13:23:11 mjf Exp $");
     50 
     51 #include "opt_md.h"
     52 
     53 #include <sys/param.h>
     54 #include <sys/kernel.h>
     55 #include <sys/malloc.h>
     56 #include <sys/systm.h>
     57 #include <sys/buf.h>
     58 #include <sys/bufq.h>
     59 #include <sys/device.h>
     60 #include <sys/disk.h>
     61 #include <sys/proc.h>
     62 #include <sys/conf.h>
     63 #include <sys/disklabel.h>
     64 
     65 #include <uvm/uvm_extern.h>
     66 
     67 #include <dev/md.h>
     68 
     69 /*
     70  * The user-space functionality is included by default.
     71  * Use  `options MEMORY_DISK_SERVER=0' to turn it off.
     72  */
     73 #ifndef MEMORY_DISK_SERVER
     74 #error MEMORY_DISK_SERVER should be defined by opt_md.h
     75 #endif	/* MEMORY_DISK_SERVER */
     76 
     77 /*
     78  * We should use the raw partition for ioctl.
     79  */
     80 #define MD_MAX_UNITS	0x10
     81 #define MD_UNIT(unit)	DISKUNIT(unit)
     82 
     83 /* autoconfig stuff... */
     84 
     85 struct md_softc {
     86 	struct device sc_dev;	/* REQUIRED first entry */
     87 	struct disk sc_dkdev;	/* hook for generic disk handling */
     88 	struct md_conf sc_md;
     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(struct device *, struct device *, void *);
     99 
    100 static dev_type_open(mdopen);
    101 static dev_type_close(mdclose);
    102 static dev_type_read(mdread);
    103 static dev_type_write(mdwrite);
    104 static dev_type_ioctl(mdioctl);
    105 static dev_type_strategy(mdstrategy);
    106 static dev_type_size(mdsize);
    107 
    108 const struct bdevsw md_bdevsw = {
    109 	mdopen, mdclose, mdstrategy, mdioctl, nodump, mdsize, D_DISK
    110 };
    111 
    112 const struct cdevsw md_cdevsw = {
    113 	mdopen, mdclose, mdread, mdwrite, mdioctl,
    114 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    115 };
    116 
    117 static struct dkdriver mddkdriver = { mdstrategy, NULL };
    118 
    119 static int   ramdisk_ndevs;
    120 static void *ramdisk_devs[MD_MAX_UNITS];
    121 
    122 /*
    123  * This is called if we are configured as a pseudo-device
    124  */
    125 void
    126 mdattach(int n)
    127 {
    128 	struct md_softc *sc;
    129 	int i;
    130 
    131 #ifdef	DIAGNOSTIC
    132 	if (ramdisk_ndevs) {
    133 		aprint_error("ramdisk: multiple attach calls?\n");
    134 		return;
    135 	}
    136 #endif
    137 
    138 	/* XXX:  Are we supposed to provide a default? */
    139 	if (n <= 1)
    140 		n = 1;
    141 	if (n > MD_MAX_UNITS)
    142 		n = MD_MAX_UNITS;
    143 	ramdisk_ndevs = n;
    144 
    145 	/* Attach as if by autoconfig. */
    146 	for (i = 0; i < n; i++) {
    147 
    148 		sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT|M_ZERO);
    149 		if (!sc) {
    150 			aprint_error("ramdisk: malloc for attach failed!\n");
    151 			return;
    152 		}
    153 		ramdisk_devs[i] = sc;
    154 		sc->sc_dev.dv_unit = i;
    155 		snprintf(sc->sc_dev.dv_xname, sizeof(sc->sc_dev.dv_xname),
    156 		    "md%d", i);
    157 		md_attach(NULL, &sc->sc_dev, NULL);
    158 	}
    159 }
    160 
    161 static void
    162 md_attach(struct device *parent, struct device *self,
    163     void *aux)
    164 {
    165 	struct md_softc *sc = (struct md_softc *)self;
    166 	int bmaj, cmaj, unit;
    167 
    168 	bufq_alloc(&sc->sc_buflist, "fcfs", 0);
    169 
    170 	/* XXX - Could accept aux info here to set the config. */
    171 #ifdef	MEMORY_DISK_HOOKS
    172 	/*
    173 	 * This external function might setup a pre-loaded disk.
    174 	 * All it would need to do is setup the md_conf struct.
    175 	 * See sys/dev/md_root.c for an example.
    176 	 */
    177 	md_attach_hook(device_unit(&sc->sc_dev), &sc->sc_md);
    178 #endif
    179 
    180 	/*
    181 	 * Initialize and attach the disk structure.
    182 	 */
    183 	disk_init(&sc->sc_dkdev, device_xname(&sc->sc_dev), &mddkdriver);
    184 	disk_attach(&sc->sc_dkdev);
    185 
    186 	cmaj = cdevsw_lookup_major(&md_cdevsw);
    187 	bmaj = bdevsw_lookup_major(&md_bdevsw);
    188 	unit = device_unit(&sc->sc_dev);
    189 
    190 	device_register_name(MAKEDISKDEV(cmaj, unit, 0), &sc->sc_dev, true,
    191 	    DEV_DISK, "r%sa", device_xname(&sc->sc_dev));
    192 	device_register_name(MAKEDISKDEV(cmaj, unit, 3), &sc->sc_dev, true,
    193 	    DEV_DISK, "r%sd", device_xname(&sc->sc_dev));
    194 
    195 	device_register_name(MAKEDISKDEV(bmaj, unit, 0), &sc->sc_dev, false,
    196 	    DEV_DISK, "%sa", device_xname(&sc->sc_dev));
    197 	device_register_name(MAKEDISKDEV(bmaj, unit, 3), &sc->sc_dev, false,
    198 	    DEV_DISK, "%sd", device_xname(&sc->sc_dev));
    199 }
    200 
    201 /*
    202  * operational routines:
    203  * open, close, read, write, strategy,
    204  * ioctl, dump, size
    205  */
    206 
    207 #if MEMORY_DISK_SERVER
    208 static int	md_server_loop(struct md_softc *sc);
    209 static int	md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
    210 		    struct lwp *l);
    211 #endif	/* MEMORY_DISK_SERVER */
    212 static int	md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
    213 		    struct lwp *l);
    214 
    215 static int
    216 mdsize(dev_t dev)
    217 {
    218 	int unit;
    219 	struct md_softc *sc;
    220 
    221 	unit = MD_UNIT(dev);
    222 	if (unit >= ramdisk_ndevs)
    223 		return 0;
    224 	sc = ramdisk_devs[unit];
    225 	if (sc == NULL)
    226 		return 0;
    227 
    228 	if (sc->sc_type == MD_UNCONFIGURED)
    229 		return 0;
    230 
    231 	return (sc->sc_size >> DEV_BSHIFT);
    232 }
    233 
    234 static int
    235 mdopen(dev_t dev, int flag, int fmt, struct lwp *l)
    236 {
    237 	int unit;
    238 	struct md_softc *sc;
    239 
    240 	unit = MD_UNIT(dev);
    241 	if (unit >= ramdisk_ndevs)
    242 		return ENXIO;
    243 	sc = ramdisk_devs[unit];
    244 	if (sc == NULL)
    245 		return ENXIO;
    246 
    247 	/*
    248 	 * The raw partition is used for ioctl to configure.
    249 	 */
    250 	if (DISKPART(dev) == RAW_PART)
    251 		return 0;
    252 
    253 #ifdef	MEMORY_DISK_HOOKS
    254 	/* Call the open hook to allow loading the device. */
    255 	md_open_hook(unit, &sc->sc_md);
    256 #endif
    257 
    258 	/*
    259 	 * This is a normal, "slave" device, so
    260 	 * enforce initialized.
    261 	 */
    262 	if (sc->sc_type == MD_UNCONFIGURED)
    263 		return ENXIO;
    264 
    265 	return 0;
    266 }
    267 
    268 static int
    269 mdclose(dev_t dev, int flag, int fmt, struct lwp *l)
    270 {
    271 	int unit;
    272 
    273 	unit = MD_UNIT(dev);
    274 
    275 	if (unit >= ramdisk_ndevs)
    276 		return ENXIO;
    277 
    278 	return 0;
    279 }
    280 
    281 static int
    282 mdread(dev_t dev, struct uio *uio, int flags)
    283 {
    284 	int unit;
    285 	struct md_softc *sc;
    286 
    287 	unit = MD_UNIT(dev);
    288 
    289 	if (unit >= ramdisk_ndevs)
    290 		return ENXIO;
    291 
    292 	sc = ramdisk_devs[unit];
    293 
    294 	if (sc->sc_type == MD_UNCONFIGURED)
    295 		return ENXIO;
    296 
    297 	return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio));
    298 }
    299 
    300 static int
    301 mdwrite(dev_t dev, struct uio *uio, int flags)
    302 {
    303 	int unit;
    304 	struct md_softc *sc;
    305 
    306 	unit = MD_UNIT(dev);
    307 
    308 	if (unit >= ramdisk_ndevs)
    309 		return ENXIO;
    310 
    311 	sc = ramdisk_devs[unit];
    312 
    313 	if (sc->sc_type == MD_UNCONFIGURED)
    314 		return ENXIO;
    315 
    316 	return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio));
    317 }
    318 
    319 /*
    320  * Handle I/O requests, either directly, or
    321  * by passing them to the server process.
    322  */
    323 static void
    324 mdstrategy(struct buf *bp)
    325 {
    326 	int unit;
    327 	struct md_softc	*sc;
    328 	void *	addr;
    329 	size_t off, xfer;
    330 
    331 	unit = MD_UNIT(bp->b_dev);
    332 	sc = ramdisk_devs[unit];
    333 
    334 	if (sc->sc_type == MD_UNCONFIGURED) {
    335 		bp->b_error = ENXIO;
    336 		goto done;
    337 	}
    338 
    339 	switch (sc->sc_type) {
    340 #if MEMORY_DISK_SERVER
    341 	case MD_UMEM_SERVER:
    342 		/* Just add this job to the server's queue. */
    343 		BUFQ_PUT(sc->sc_buflist, bp);
    344 		wakeup((void *)sc);
    345 		/* see md_server_loop() */
    346 		/* no biodone in this case */
    347 		return;
    348 #endif	/* MEMORY_DISK_SERVER */
    349 
    350 	case MD_KMEM_FIXED:
    351 	case MD_KMEM_ALLOCATED:
    352 		/* These are in kernel space.  Access directly. */
    353 		bp->b_resid = bp->b_bcount;
    354 		off = (bp->b_blkno << DEV_BSHIFT);
    355 		if (off >= sc->sc_size) {
    356 			if (bp->b_flags & B_READ)
    357 				break;	/* EOF */
    358 			goto set_eio;
    359 		}
    360 		xfer = bp->b_resid;
    361 		if (xfer > (sc->sc_size - off))
    362 			xfer = (sc->sc_size - off);
    363 		addr = (char *)sc->sc_addr + off;
    364 		if (bp->b_flags & B_READ)
    365 			memcpy(bp->b_data, addr, xfer);
    366 		else
    367 			memcpy(addr, bp->b_data, xfer);
    368 		bp->b_resid -= xfer;
    369 		break;
    370 
    371 	default:
    372 		bp->b_resid = bp->b_bcount;
    373 	set_eio:
    374 		bp->b_error = EIO;
    375 		break;
    376 	}
    377  done:
    378 	biodone(bp);
    379 }
    380 
    381 static int
    382 mdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    383 {
    384 	int unit;
    385 	struct md_softc *sc;
    386 	struct md_conf *umd;
    387 
    388 	unit = MD_UNIT(dev);
    389 	sc = ramdisk_devs[unit];
    390 
    391 	/* If this is not the raw partition, punt! */
    392 	if (DISKPART(dev) != RAW_PART)
    393 		return ENOTTY;
    394 
    395 	umd = (struct md_conf *)data;
    396 	switch (cmd) {
    397 	case MD_GETCONF:
    398 		*umd = sc->sc_md;
    399 		return 0;
    400 
    401 	case MD_SETCONF:
    402 		/* Can only set it once. */
    403 		if (sc->sc_type != MD_UNCONFIGURED)
    404 			break;
    405 		switch (umd->md_type) {
    406 		case MD_KMEM_ALLOCATED:
    407 			return md_ioctl_kalloc(sc, umd, l);
    408 #if MEMORY_DISK_SERVER
    409 		case MD_UMEM_SERVER:
    410 			return md_ioctl_server(sc, umd, l);
    411 #endif	/* MEMORY_DISK_SERVER */
    412 		default:
    413 			break;
    414 		}
    415 		break;
    416 	}
    417 	return EINVAL;
    418 }
    419 
    420 /*
    421  * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED)
    422  * Just allocate some kernel memory and return.
    423  */
    424 static int
    425 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
    426     struct lwp *l)
    427 {
    428 	vaddr_t addr;
    429 	vsize_t size;
    430 
    431 	/* Sanity check the size. */
    432 	size = umd->md_size;
    433 	addr = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
    434 	if (!addr)
    435 		return ENOMEM;
    436 
    437 	/* This unit is now configured. */
    438 	sc->sc_addr = (void *)addr; 	/* kernel space */
    439 	sc->sc_size = (size_t)size;
    440 	sc->sc_type = MD_KMEM_ALLOCATED;
    441 	return 0;
    442 }
    443 
    444 #if MEMORY_DISK_SERVER
    445 
    446 /*
    447  * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER)
    448  * Set config, then become the I/O server for this unit.
    449  */
    450 static int
    451 md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
    452     struct lwp *l)
    453 {
    454 	vaddr_t end;
    455 	int error;
    456 
    457 	/* Sanity check addr, size. */
    458 	end = (vaddr_t) ((char *)umd->md_addr + umd->md_size);
    459 
    460 	if ((end >= VM_MAXUSER_ADDRESS) ||
    461 		(end < ((vaddr_t) umd->md_addr)) )
    462 		return EINVAL;
    463 
    464 	/* This unit is now configured. */
    465 	sc->sc_addr = umd->md_addr; 	/* user space */
    466 	sc->sc_size = umd->md_size;
    467 	sc->sc_type = MD_UMEM_SERVER;
    468 
    469 	/* Become the server daemon */
    470 	error = md_server_loop(sc);
    471 
    472 	/* This server is now going away! */
    473 	sc->sc_type = MD_UNCONFIGURED;
    474 	sc->sc_addr = 0;
    475 	sc->sc_size = 0;
    476 
    477 	return (error);
    478 }
    479 
    480 static int md_sleep_pri = PWAIT | PCATCH;
    481 
    482 static int
    483 md_server_loop(struct md_softc *sc)
    484 {
    485 	struct buf *bp;
    486 	void *addr;	/* user space address */
    487 	size_t off;	/* offset into "device" */
    488 	size_t xfer;	/* amount to transfer */
    489 	int error;
    490 
    491 	for (;;) {
    492 		/* Wait for some work to arrive. */
    493 		while ((bp = BUFQ_GET(sc->sc_buflist)) == NULL) {
    494 			error = tsleep((void *)sc, md_sleep_pri, "md_idle", 0);
    495 			if (error)
    496 				return error;
    497 		}
    498 
    499 		/* Do the transfer to/from user space. */
    500 		error = 0;
    501 		bp->b_resid = bp->b_bcount;
    502 		off = (bp->b_blkno << DEV_BSHIFT);
    503 		if (off >= sc->sc_size) {
    504 			if (bp->b_flags & B_READ)
    505 				goto done;	/* EOF (not an error) */
    506 			error = EIO;
    507 			goto done;
    508 		}
    509 		xfer = bp->b_resid;
    510 		if (xfer > (sc->sc_size - off))
    511 			xfer = (sc->sc_size - off);
    512 		addr = (char *)sc->sc_addr + off;
    513 		if (bp->b_flags & B_READ)
    514 			error = copyin(addr, bp->b_data, xfer);
    515 		else
    516 			error = copyout(bp->b_data, addr, xfer);
    517 		if (!error)
    518 			bp->b_resid -= xfer;
    519 
    520 	done:
    521 		if (error) {
    522 			bp->b_error = error;
    523 		}
    524 		biodone(bp);
    525 	}
    526 }
    527 #endif	/* MEMORY_DISK_SERVER */
    528