xref: /netbsd-src/sys/dev/md.c (revision de1dfb1250df962f1ff3a011772cf58e605aed11) 
1 /*	$NetBSD: md.c,v 1.38 2004/08/30 00:34:42 thorpej 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 1.38 2004/08/30 00:34:42 thorpej 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/device.h>
59 #include <sys/disk.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  * By default, include the user-space functionality.
70  * Use  `options MEMORY_DISK_SERVER=0' to turn it off.
71  */
72 #ifndef MEMORY_DISK_SERVER
73 #define	MEMORY_DISK_SERVER 1
74 #endif	/* MEMORY_DISK_SERVER */
75 
76 /*
77  * We should use the raw partition for ioctl.
78  */
79 #define MD_MAX_UNITS	0x10
80 #define MD_UNIT(unit)	DISKUNIT(unit)
81 
82 /* autoconfig stuff... */
83 
84 struct md_softc {
85 	struct device sc_dev;	/* REQUIRED first entry */
86 	struct disk sc_dkdev;	/* hook for generic disk handling */
87 	struct md_conf sc_md;
88 	struct bufq_state sc_buflist;
89 };
90 /* shorthand for fields in sc_md: */
91 #define sc_addr sc_md.md_addr
92 #define sc_size sc_md.md_size
93 #define sc_type sc_md.md_type
94 
95 void	mdattach(int);
96 
97 static void	md_attach(struct device *, struct device *, void *);
98 
99 static dev_type_open(mdopen);
100 static dev_type_close(mdclose);
101 static dev_type_read(mdread);
102 static dev_type_write(mdwrite);
103 static dev_type_ioctl(mdioctl);
104 static dev_type_strategy(mdstrategy);
105 static dev_type_size(mdsize);
106 
107 const struct bdevsw md_bdevsw = {
108 	mdopen, mdclose, mdstrategy, mdioctl, nodump, mdsize, D_DISK
109 };
110 
111 const struct cdevsw md_cdevsw = {
112 	mdopen, mdclose, mdread, mdwrite, mdioctl,
113 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
114 };
115 
116 static struct dkdriver mddkdriver = { mdstrategy };
117 
118 static int   ramdisk_ndevs;
119 static void *ramdisk_devs[MD_MAX_UNITS];
120 
121 /*
122  * This is called if we are configured as a pseudo-device
123  */
124 void
125 mdattach(int n)
126 {
127 	struct md_softc *sc;
128 	int i;
129 
130 #ifdef	DIAGNOSTIC
131 	if (ramdisk_ndevs) {
132 		aprint_error("ramdisk: multiple attach calls?\n");
133 		return;
134 	}
135 #endif
136 
137 	/* XXX:  Are we supposed to provide a default? */
138 	if (n <= 1)
139 		n = 1;
140 	if (n > MD_MAX_UNITS)
141 		n = MD_MAX_UNITS;
142 	ramdisk_ndevs = n;
143 
144 	/* Attach as if by autoconfig. */
145 	for (i = 0; i < n; i++) {
146 
147 		sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT|M_ZERO);
148 		if (!sc) {
149 			aprint_error("ramdisk: malloc for attach failed!\n");
150 			return;
151 		}
152 		ramdisk_devs[i] = sc;
153 		sc->sc_dev.dv_unit = i;
154 		snprintf(sc->sc_dev.dv_xname, sizeof(sc->sc_dev.dv_xname),
155 		    "md%d", i);
156 		md_attach(NULL, &sc->sc_dev, NULL);
157 	}
158 }
159 
160 static void
161 md_attach(struct device *parent, struct device *self, void *aux)
162 {
163 	struct md_softc *sc = (struct md_softc *)self;
164 
165 	bufq_alloc(&sc->sc_buflist, BUFQ_FCFS);
166 
167 	/* XXX - Could accept aux info here to set the config. */
168 #ifdef	MEMORY_DISK_HOOKS
169 	/*
170 	 * This external function might setup a pre-loaded disk.
171 	 * All it would need to do is setup the md_conf struct.
172 	 * See sys/dev/md_root.c for an example.
173 	 */
174 	md_attach_hook(sc->sc_dev.dv_unit, &sc->sc_md);
175 #endif
176 
177 	/*
178 	 * Initialize and attach the disk structure.
179 	 */
180 	sc->sc_dkdev.dk_driver = &mddkdriver;
181 	sc->sc_dkdev.dk_name = sc->sc_dev.dv_xname;
182 	disk_attach(&sc->sc_dkdev);
183 }
184 
185 /*
186  * operational routines:
187  * open, close, read, write, strategy,
188  * ioctl, dump, size
189  */
190 
191 #if MEMORY_DISK_SERVER
192 static int	md_server_loop(struct md_softc *sc);
193 static int	md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
194 		    struct proc *proc);
195 #endif	/* MEMORY_DISK_SERVER */
196 static int	md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
197 		    struct proc *proc);
198 
199 static int
200 mdsize(dev_t dev)
201 {
202 	int unit;
203 	struct md_softc *sc;
204 
205 	unit = MD_UNIT(dev);
206 	if (unit >= ramdisk_ndevs)
207 		return 0;
208 	sc = ramdisk_devs[unit];
209 	if (sc == NULL)
210 		return 0;
211 
212 	if (sc->sc_type == MD_UNCONFIGURED)
213 		return 0;
214 
215 	return (sc->sc_size >> DEV_BSHIFT);
216 }
217 
218 static int
219 mdopen(dev_t dev, int flag, int fmt, struct proc *proc)
220 {
221 	int unit;
222 	struct md_softc *sc;
223 
224 	unit = MD_UNIT(dev);
225 	if (unit >= ramdisk_ndevs)
226 		return ENXIO;
227 	sc = ramdisk_devs[unit];
228 	if (sc == NULL)
229 		return ENXIO;
230 
231 	/*
232 	 * The raw partition is used for ioctl to configure.
233 	 */
234 	if (DISKPART(dev) == RAW_PART)
235 		return 0;
236 
237 #ifdef	MEMORY_DISK_HOOKS
238 	/* Call the open hook to allow loading the device. */
239 	md_open_hook(unit, &sc->sc_md);
240 #endif
241 
242 	/*
243 	 * This is a normal, "slave" device, so
244 	 * enforce initialized.
245 	 */
246 	if (sc->sc_type == MD_UNCONFIGURED)
247 		return ENXIO;
248 
249 	return 0;
250 }
251 
252 static int
253 mdclose(dev_t dev, int flag, int fmt, struct proc *proc)
254 {
255 	int unit;
256 
257 	unit = MD_UNIT(dev);
258 
259 	if (unit >= ramdisk_ndevs)
260 		return ENXIO;
261 
262 	return 0;
263 }
264 
265 static int
266 mdread(dev_t dev, struct uio *uio, int flags)
267 {
268 	int unit;
269 	struct md_softc *sc;
270 
271 	unit = MD_UNIT(dev);
272 
273 	if (unit >= ramdisk_ndevs)
274 		return ENXIO;
275 
276 	sc = ramdisk_devs[unit];
277 
278 	if (sc->sc_type == MD_UNCONFIGURED)
279 		return ENXIO;
280 
281 	return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio));
282 }
283 
284 static int
285 mdwrite(dev_t dev, struct uio *uio, int flags)
286 {
287 	int unit;
288 	struct md_softc *sc;
289 
290 	unit = MD_UNIT(dev);
291 
292 	if (unit >= ramdisk_ndevs)
293 		return ENXIO;
294 
295 	sc = ramdisk_devs[unit];
296 
297 	if (sc->sc_type == MD_UNCONFIGURED)
298 		return ENXIO;
299 
300 	return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio));
301 }
302 
303 /*
304  * Handle I/O requests, either directly, or
305  * by passing them to the server process.
306  */
307 static void
308 mdstrategy(struct buf *bp)
309 {
310 	int unit;
311 	struct md_softc	*sc;
312 	caddr_t	addr;
313 	size_t off, xfer;
314 
315 	unit = MD_UNIT(bp->b_dev);
316 	sc = ramdisk_devs[unit];
317 
318 	if (sc->sc_type == MD_UNCONFIGURED) {
319 		bp->b_error = ENXIO;
320 		bp->b_flags |= B_ERROR;
321 		goto done;
322 	}
323 
324 	switch (sc->sc_type) {
325 #if MEMORY_DISK_SERVER
326 	case MD_UMEM_SERVER:
327 		/* Just add this job to the server's queue. */
328 		BUFQ_PUT(&sc->sc_buflist, bp);
329 		wakeup((caddr_t)sc);
330 		/* see md_server_loop() */
331 		/* no biodone in this case */
332 		return;
333 #endif	/* MEMORY_DISK_SERVER */
334 
335 	case MD_KMEM_FIXED:
336 	case MD_KMEM_ALLOCATED:
337 		/* These are in kernel space.  Access directly. */
338 		bp->b_resid = bp->b_bcount;
339 		off = (bp->b_blkno << DEV_BSHIFT);
340 		if (off >= sc->sc_size) {
341 			if (bp->b_flags & B_READ)
342 				break;	/* EOF */
343 			goto set_eio;
344 		}
345 		xfer = bp->b_resid;
346 		if (xfer > (sc->sc_size - off))
347 			xfer = (sc->sc_size - off);
348 		addr = sc->sc_addr + off;
349 		if (bp->b_flags & B_READ)
350 			memcpy(bp->b_data, addr, xfer);
351 		else
352 			memcpy(addr, bp->b_data, xfer);
353 		bp->b_resid -= xfer;
354 		break;
355 
356 	default:
357 		bp->b_resid = bp->b_bcount;
358 	set_eio:
359 		bp->b_error = EIO;
360 		bp->b_flags |= B_ERROR;
361 		break;
362 	}
363  done:
364 	biodone(bp);
365 }
366 
367 static int
368 mdioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *proc)
369 {
370 	int unit;
371 	struct md_softc *sc;
372 	struct md_conf *umd;
373 
374 	unit = MD_UNIT(dev);
375 	sc = ramdisk_devs[unit];
376 
377 	/* If this is not the raw partition, punt! */
378 	if (DISKPART(dev) != RAW_PART)
379 		return ENOTTY;
380 
381 	umd = (struct md_conf *)data;
382 	switch (cmd) {
383 	case MD_GETCONF:
384 		*umd = sc->sc_md;
385 		return 0;
386 
387 	case MD_SETCONF:
388 		/* Can only set it once. */
389 		if (sc->sc_type != MD_UNCONFIGURED)
390 			break;
391 		switch (umd->md_type) {
392 		case MD_KMEM_ALLOCATED:
393 			return md_ioctl_kalloc(sc, umd, proc);
394 #if MEMORY_DISK_SERVER
395 		case MD_UMEM_SERVER:
396 			return md_ioctl_server(sc, umd, proc);
397 #endif	/* MEMORY_DISK_SERVER */
398 		default:
399 			break;
400 		}
401 		break;
402 	}
403 	return EINVAL;
404 }
405 
406 /*
407  * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED)
408  * Just allocate some kernel memory and return.
409  */
410 static int
411 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd, struct proc *proc)
412 {
413 	vaddr_t addr;
414 	vsize_t size;
415 
416 	/* Sanity check the size. */
417 	size = umd->md_size;
418 	addr = uvm_km_zalloc(kernel_map, size);
419 	if (!addr)
420 		return ENOMEM;
421 
422 	/* This unit is now configured. */
423 	sc->sc_addr = (caddr_t)addr; 	/* kernel space */
424 	sc->sc_size = (size_t)size;
425 	sc->sc_type = MD_KMEM_ALLOCATED;
426 	return 0;
427 }
428 
429 #if MEMORY_DISK_SERVER
430 
431 /*
432  * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER)
433  * Set config, then become the I/O server for this unit.
434  */
435 static int
436 md_ioctl_server(struct md_softc *sc, struct md_conf *umd, struct proc *proc)
437 {
438 	vaddr_t end;
439 	int error;
440 
441 	/* Sanity check addr, size. */
442 	end = (vaddr_t) (umd->md_addr + umd->md_size);
443 
444 	if ((end >= VM_MAXUSER_ADDRESS) ||
445 		(end < ((vaddr_t) umd->md_addr)) )
446 		return EINVAL;
447 
448 	/* This unit is now configured. */
449 	sc->sc_addr = umd->md_addr; 	/* user space */
450 	sc->sc_size = umd->md_size;
451 	sc->sc_type = MD_UMEM_SERVER;
452 
453 	/* Become the server daemon */
454 	error = md_server_loop(sc);
455 
456 	/* This server is now going away! */
457 	sc->sc_type = MD_UNCONFIGURED;
458 	sc->sc_addr = 0;
459 	sc->sc_size = 0;
460 
461 	return (error);
462 }
463 
464 static int md_sleep_pri = PWAIT | PCATCH;
465 
466 static int
467 md_server_loop(struct md_softc *sc)
468 {
469 	struct buf *bp;
470 	caddr_t addr;	/* user space address */
471 	size_t off;	/* offset into "device" */
472 	size_t xfer;	/* amount to transfer */
473 	int error;
474 
475 	for (;;) {
476 		/* Wait for some work to arrive. */
477 		while ((bp = BUFQ_GET(&sc->sc_buflist)) == NULL) {
478 			error = tsleep((caddr_t)sc, md_sleep_pri, "md_idle", 0);
479 			if (error)
480 				return error;
481 		}
482 
483 		/* Do the transfer to/from user space. */
484 		error = 0;
485 		bp->b_resid = bp->b_bcount;
486 		off = (bp->b_blkno << DEV_BSHIFT);
487 		if (off >= sc->sc_size) {
488 			if (bp->b_flags & B_READ)
489 				goto done;	/* EOF (not an error) */
490 			error = EIO;
491 			goto done;
492 		}
493 		xfer = bp->b_resid;
494 		if (xfer > (sc->sc_size - off))
495 			xfer = (sc->sc_size - off);
496 		addr = sc->sc_addr + off;
497 		if (bp->b_flags & B_READ)
498 			error = copyin(addr, bp->b_data, xfer);
499 		else
500 			error = copyout(bp->b_data, addr, xfer);
501 		if (!error)
502 			bp->b_resid -= xfer;
503 
504 	done:
505 		if (error) {
506 			bp->b_error = error;
507 			bp->b_flags |= B_ERROR;
508 		}
509 		biodone(bp);
510 	}
511 }
512 #endif	/* MEMORY_DISK_SERVER */
513