xref: /netbsd-src/sys/dev/md.c (revision 21e37cc72a480a47828990a439cde7ac9ffaf0c6) 
1 /*	$NetBSD: md.c,v 1.37 2004/04/21 18:40:37 itojun 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 1.37 2004/04/21 18:40:37 itojun 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 __P((int));
96 static void md_attach __P((struct device *, struct device *, void *));
97 
98 dev_type_open(mdopen);
99 dev_type_close(mdclose);
100 dev_type_read(mdread);
101 dev_type_write(mdwrite);
102 dev_type_ioctl(mdioctl);
103 dev_type_strategy(mdstrategy);
104 dev_type_size(mdsize);
105 
106 const struct bdevsw md_bdevsw = {
107 	mdopen, mdclose, mdstrategy, mdioctl, nodump, mdsize, D_DISK
108 };
109 
110 const struct cdevsw md_cdevsw = {
111 	mdopen, mdclose, mdread, mdwrite, mdioctl,
112 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
113 };
114 
115 struct dkdriver mddkdriver = { mdstrategy };
116 
117 static int   ramdisk_ndevs;
118 static void *ramdisk_devs[MD_MAX_UNITS];
119 
120 /*
121  * This is called if we are configured as a pseudo-device
122  */
123 void
124 mdattach(n)
125 	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(parent, self, aux)
162 	struct device	*parent, *self;
163 	void		*aux;
164 {
165 	struct md_softc *sc = (struct md_softc *)self;
166 
167 	bufq_alloc(&sc->sc_buflist, BUFQ_FCFS);
168 
169 	/* XXX - Could accept aux info here to set the config. */
170 #ifdef	MEMORY_DISK_HOOKS
171 	/*
172 	 * This external function might setup a pre-loaded disk.
173 	 * All it would need to do is setup the md_conf struct.
174 	 * See sys/dev/md_root.c for an example.
175 	 */
176 	md_attach_hook(sc->sc_dev.dv_unit, &sc->sc_md);
177 #endif
178 
179 	/*
180 	 * Initialize and attach the disk structure.
181 	 */
182 	sc->sc_dkdev.dk_driver = &mddkdriver;
183 	sc->sc_dkdev.dk_name = sc->sc_dev.dv_xname;
184 	disk_attach(&sc->sc_dkdev);
185 }
186 
187 /*
188  * operational routines:
189  * open, close, read, write, strategy,
190  * ioctl, dump, size
191  */
192 
193 #if MEMORY_DISK_SERVER
194 static int md_server_loop __P((struct md_softc *sc));
195 static int md_ioctl_server __P((struct md_softc *sc,
196 		struct md_conf *umd, struct proc *proc));
197 #endif	/* MEMORY_DISK_SERVER */
198 static int md_ioctl_kalloc __P((struct md_softc *sc,
199 		struct md_conf *umd, struct proc *proc));
200 
201 int
202 mdsize(dev_t dev)
203 {
204 	int unit;
205 	struct md_softc *sc;
206 
207 	unit = MD_UNIT(dev);
208 	if (unit >= ramdisk_ndevs)
209 		return 0;
210 	sc = ramdisk_devs[unit];
211 	if (sc == NULL)
212 		return 0;
213 
214 	if (sc->sc_type == MD_UNCONFIGURED)
215 		return 0;
216 
217 	return (sc->sc_size >> DEV_BSHIFT);
218 }
219 
220 int
221 mdopen(dev, flag, fmt, proc)
222 	dev_t dev;
223 	int flag, fmt;
224 	struct proc *proc;
225 {
226 	int unit;
227 	struct md_softc *sc;
228 
229 	unit = MD_UNIT(dev);
230 	if (unit >= ramdisk_ndevs)
231 		return ENXIO;
232 	sc = ramdisk_devs[unit];
233 	if (sc == NULL)
234 		return ENXIO;
235 
236 	/*
237 	 * The raw partition is used for ioctl to configure.
238 	 */
239 	if (DISKPART(dev) == RAW_PART)
240 		return 0;
241 
242 #ifdef	MEMORY_DISK_HOOKS
243 	/* Call the open hook to allow loading the device. */
244 	md_open_hook(unit, &sc->sc_md);
245 #endif
246 
247 	/*
248 	 * This is a normal, "slave" device, so
249 	 * enforce initialized.
250 	 */
251 	if (sc->sc_type == MD_UNCONFIGURED)
252 		return ENXIO;
253 
254 	return 0;
255 }
256 
257 int
258 mdclose(dev, flag, fmt, proc)
259 	dev_t dev;
260 	int flag, fmt;
261 	struct proc *proc;
262 {
263 	int unit;
264 
265 	unit = MD_UNIT(dev);
266 
267 	if (unit >= ramdisk_ndevs)
268 		return ENXIO;
269 
270 	return 0;
271 }
272 
273 int
274 mdread(dev, uio, flags)
275 	dev_t dev;
276 	struct uio *uio;
277 	int flags;
278 {
279 	int unit;
280 	struct md_softc *sc;
281 
282 	unit = MD_UNIT(dev);
283 
284 	if (unit >= ramdisk_ndevs)
285 		return ENXIO;
286 
287 	sc = ramdisk_devs[unit];
288 
289 	if (sc->sc_type == MD_UNCONFIGURED)
290 		return ENXIO;
291 
292 	return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio));
293 }
294 
295 int
296 mdwrite(dev, uio, flags)
297 	dev_t dev;
298 	struct uio *uio;
299 	int flags;
300 {
301 	int unit;
302 	struct md_softc *sc;
303 
304 	unit = MD_UNIT(dev);
305 
306 	if (unit >= ramdisk_ndevs)
307 		return ENXIO;
308 
309 	sc = ramdisk_devs[unit];
310 
311 	if (sc->sc_type == MD_UNCONFIGURED)
312 		return ENXIO;
313 
314 	return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio));
315 }
316 
317 /*
318  * Handle I/O requests, either directly, or
319  * by passing them to the server process.
320  */
321 void
322 mdstrategy(bp)
323 	struct buf *bp;
324 {
325 	int unit;
326 	struct md_softc	*sc;
327 	caddr_t	addr;
328 	size_t off, xfer;
329 
330 	unit = MD_UNIT(bp->b_dev);
331 	sc = ramdisk_devs[unit];
332 
333 	if (sc->sc_type == MD_UNCONFIGURED) {
334 		bp->b_error = ENXIO;
335 		bp->b_flags |= B_ERROR;
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((caddr_t)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 = 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 		bp->b_flags |= B_ERROR;
376 		break;
377 	}
378  done:
379 	biodone(bp);
380 }
381 
382 int
383 mdioctl(dev, cmd, data, flag, proc)
384 	dev_t dev;
385 	u_long cmd;
386 	int flag;
387 	caddr_t data;
388 	struct proc *proc;
389 {
390 	int unit;
391 	struct md_softc *sc;
392 	struct md_conf *umd;
393 
394 	unit = MD_UNIT(dev);
395 	sc = ramdisk_devs[unit];
396 
397 	/* If this is not the raw partition, punt! */
398 	if (DISKPART(dev) != RAW_PART)
399 		return ENOTTY;
400 
401 	umd = (struct md_conf *)data;
402 	switch (cmd) {
403 	case MD_GETCONF:
404 		*umd = sc->sc_md;
405 		return 0;
406 
407 	case MD_SETCONF:
408 		/* Can only set it once. */
409 		if (sc->sc_type != MD_UNCONFIGURED)
410 			break;
411 		switch (umd->md_type) {
412 		case MD_KMEM_ALLOCATED:
413 			return md_ioctl_kalloc(sc, umd, proc);
414 #if MEMORY_DISK_SERVER
415 		case MD_UMEM_SERVER:
416 			return md_ioctl_server(sc, umd, proc);
417 #endif	/* MEMORY_DISK_SERVER */
418 		default:
419 			break;
420 		}
421 		break;
422 	}
423 	return EINVAL;
424 }
425 
426 /*
427  * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED)
428  * Just allocate some kernel memory and return.
429  */
430 static int
431 md_ioctl_kalloc(sc, umd, proc)
432 	struct md_softc *sc;
433 	struct md_conf *umd;
434 	struct proc *proc;
435 {
436 	vaddr_t addr;
437 	vsize_t size;
438 
439 	/* Sanity check the size. */
440 	size = umd->md_size;
441 	addr = uvm_km_zalloc(kernel_map, size);
442 	if (!addr)
443 		return ENOMEM;
444 
445 	/* This unit is now configured. */
446 	sc->sc_addr = (caddr_t)addr; 	/* kernel space */
447 	sc->sc_size = (size_t)size;
448 	sc->sc_type = MD_KMEM_ALLOCATED;
449 	return 0;
450 }
451 
452 #if MEMORY_DISK_SERVER
453 
454 /*
455  * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER)
456  * Set config, then become the I/O server for this unit.
457  */
458 static int
459 md_ioctl_server(sc, umd, proc)
460 	struct md_softc *sc;
461 	struct md_conf *umd;
462 	struct proc *proc;
463 {
464 	vaddr_t end;
465 	int error;
466 
467 	/* Sanity check addr, size. */
468 	end = (vaddr_t) (umd->md_addr + umd->md_size);
469 
470 	if ((end >= VM_MAXUSER_ADDRESS) ||
471 		(end < ((vaddr_t) umd->md_addr)) )
472 		return EINVAL;
473 
474 	/* This unit is now configured. */
475 	sc->sc_addr = umd->md_addr; 	/* user space */
476 	sc->sc_size = umd->md_size;
477 	sc->sc_type = MD_UMEM_SERVER;
478 
479 	/* Become the server daemon */
480 	error = md_server_loop(sc);
481 
482 	/* This server is now going away! */
483 	sc->sc_type = MD_UNCONFIGURED;
484 	sc->sc_addr = 0;
485 	sc->sc_size = 0;
486 
487 	return (error);
488 }
489 
490 int md_sleep_pri = PWAIT | PCATCH;
491 
492 static int
493 md_server_loop(sc)
494 	struct md_softc *sc;
495 {
496 	struct buf *bp;
497 	caddr_t addr;	/* user space address */
498 	size_t off;	/* offset into "device" */
499 	size_t xfer;	/* amount to transfer */
500 	int error;
501 
502 	for (;;) {
503 		/* Wait for some work to arrive. */
504 		while ((bp = BUFQ_GET(&sc->sc_buflist)) == NULL) {
505 			error = tsleep((caddr_t)sc, md_sleep_pri, "md_idle", 0);
506 			if (error)
507 				return error;
508 		}
509 
510 		/* Do the transfer to/from user space. */
511 		error = 0;
512 		bp->b_resid = bp->b_bcount;
513 		off = (bp->b_blkno << DEV_BSHIFT);
514 		if (off >= sc->sc_size) {
515 			if (bp->b_flags & B_READ)
516 				goto done;	/* EOF (not an error) */
517 			error = EIO;
518 			goto done;
519 		}
520 		xfer = bp->b_resid;
521 		if (xfer > (sc->sc_size - off))
522 			xfer = (sc->sc_size - off);
523 		addr = sc->sc_addr + off;
524 		if (bp->b_flags & B_READ)
525 			error = copyin(addr, bp->b_data, xfer);
526 		else
527 			error = copyout(bp->b_data, addr, xfer);
528 		if (!error)
529 			bp->b_resid -= xfer;
530 
531 	done:
532 		if (error) {
533 			bp->b_error = error;
534 			bp->b_flags |= B_ERROR;
535 		}
536 		biodone(bp);
537 	}
538 }
539 #endif	/* MEMORY_DISK_SERVER */
540