xref: /netbsd-src/sys/dev/md.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21) 
1 /*	$NetBSD: md.c,v 1.46 2006/10/12 01:30:51 christos 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 1.46 2006/10/12 01:30:51 christos 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  * By default, include the user-space functionality.
71  * Use  `options MEMORY_DISK_SERVER=0' to turn it off.
72  */
73 #ifndef MEMORY_DISK_SERVER
74 #define	MEMORY_DISK_SERVER 1
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 __unused, struct device *self,
163     void *aux __unused)
164 {
165 	struct md_softc *sc = (struct md_softc *)self;
166 
167 	bufq_alloc(&sc->sc_buflist, "fcfs", 0);
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(device_unit(&sc->sc_dev), &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(struct md_softc *sc);
195 static int	md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
196 		    struct lwp *l);
197 #endif	/* MEMORY_DISK_SERVER */
198 static int	md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
199 		    struct lwp *l);
200 
201 static 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 static int
221 mdopen(dev_t dev, int flag __unused, int fmt __unused, struct lwp *l __unused)
222 {
223 	int unit;
224 	struct md_softc *sc;
225 
226 	unit = MD_UNIT(dev);
227 	if (unit >= ramdisk_ndevs)
228 		return ENXIO;
229 	sc = ramdisk_devs[unit];
230 	if (sc == NULL)
231 		return ENXIO;
232 
233 	/*
234 	 * The raw partition is used for ioctl to configure.
235 	 */
236 	if (DISKPART(dev) == RAW_PART)
237 		return 0;
238 
239 #ifdef	MEMORY_DISK_HOOKS
240 	/* Call the open hook to allow loading the device. */
241 	md_open_hook(unit, &sc->sc_md);
242 #endif
243 
244 	/*
245 	 * This is a normal, "slave" device, so
246 	 * enforce initialized.
247 	 */
248 	if (sc->sc_type == MD_UNCONFIGURED)
249 		return ENXIO;
250 
251 	return 0;
252 }
253 
254 static int
255 mdclose(dev_t dev, int flag __unused, int fmt __unused, struct lwp *l __unused)
256 {
257 	int unit;
258 
259 	unit = MD_UNIT(dev);
260 
261 	if (unit >= ramdisk_ndevs)
262 		return ENXIO;
263 
264 	return 0;
265 }
266 
267 static int
268 mdread(dev_t dev, struct uio *uio, int flags __unused)
269 {
270 	int unit;
271 	struct md_softc *sc;
272 
273 	unit = MD_UNIT(dev);
274 
275 	if (unit >= ramdisk_ndevs)
276 		return ENXIO;
277 
278 	sc = ramdisk_devs[unit];
279 
280 	if (sc->sc_type == MD_UNCONFIGURED)
281 		return ENXIO;
282 
283 	return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio));
284 }
285 
286 static int
287 mdwrite(dev_t dev, struct uio *uio, int flags __unused)
288 {
289 	int unit;
290 	struct md_softc *sc;
291 
292 	unit = MD_UNIT(dev);
293 
294 	if (unit >= ramdisk_ndevs)
295 		return ENXIO;
296 
297 	sc = ramdisk_devs[unit];
298 
299 	if (sc->sc_type == MD_UNCONFIGURED)
300 		return ENXIO;
301 
302 	return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio));
303 }
304 
305 /*
306  * Handle I/O requests, either directly, or
307  * by passing them to the server process.
308  */
309 static void
310 mdstrategy(struct buf *bp)
311 {
312 	int unit;
313 	struct md_softc	*sc;
314 	caddr_t	addr;
315 	size_t off, xfer;
316 
317 	unit = MD_UNIT(bp->b_dev);
318 	sc = ramdisk_devs[unit];
319 
320 	if (sc->sc_type == MD_UNCONFIGURED) {
321 		bp->b_error = ENXIO;
322 		bp->b_flags |= B_ERROR;
323 		goto done;
324 	}
325 
326 	switch (sc->sc_type) {
327 #if MEMORY_DISK_SERVER
328 	case MD_UMEM_SERVER:
329 		/* Just add this job to the server's queue. */
330 		BUFQ_PUT(sc->sc_buflist, bp);
331 		wakeup((caddr_t)sc);
332 		/* see md_server_loop() */
333 		/* no biodone in this case */
334 		return;
335 #endif	/* MEMORY_DISK_SERVER */
336 
337 	case MD_KMEM_FIXED:
338 	case MD_KMEM_ALLOCATED:
339 		/* These are in kernel space.  Access directly. */
340 		bp->b_resid = bp->b_bcount;
341 		off = (bp->b_blkno << DEV_BSHIFT);
342 		if (off >= sc->sc_size) {
343 			if (bp->b_flags & B_READ)
344 				break;	/* EOF */
345 			goto set_eio;
346 		}
347 		xfer = bp->b_resid;
348 		if (xfer > (sc->sc_size - off))
349 			xfer = (sc->sc_size - off);
350 		addr = sc->sc_addr + off;
351 		if (bp->b_flags & B_READ)
352 			memcpy(bp->b_data, addr, xfer);
353 		else
354 			memcpy(addr, bp->b_data, xfer);
355 		bp->b_resid -= xfer;
356 		break;
357 
358 	default:
359 		bp->b_resid = bp->b_bcount;
360 	set_eio:
361 		bp->b_error = EIO;
362 		bp->b_flags |= B_ERROR;
363 		break;
364 	}
365  done:
366 	biodone(bp);
367 }
368 
369 static int
370 mdioctl(dev_t dev, u_long cmd, caddr_t data, int flag __unused, struct lwp *l)
371 {
372 	int unit;
373 	struct md_softc *sc;
374 	struct md_conf *umd;
375 
376 	unit = MD_UNIT(dev);
377 	sc = ramdisk_devs[unit];
378 
379 	/* If this is not the raw partition, punt! */
380 	if (DISKPART(dev) != RAW_PART)
381 		return ENOTTY;
382 
383 	umd = (struct md_conf *)data;
384 	switch (cmd) {
385 	case MD_GETCONF:
386 		*umd = sc->sc_md;
387 		return 0;
388 
389 	case MD_SETCONF:
390 		/* Can only set it once. */
391 		if (sc->sc_type != MD_UNCONFIGURED)
392 			break;
393 		switch (umd->md_type) {
394 		case MD_KMEM_ALLOCATED:
395 			return md_ioctl_kalloc(sc, umd, l);
396 #if MEMORY_DISK_SERVER
397 		case MD_UMEM_SERVER:
398 			return md_ioctl_server(sc, umd, l);
399 #endif	/* MEMORY_DISK_SERVER */
400 		default:
401 			break;
402 		}
403 		break;
404 	}
405 	return EINVAL;
406 }
407 
408 /*
409  * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED)
410  * Just allocate some kernel memory and return.
411  */
412 static int
413 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
414     struct lwp *l __unused)
415 {
416 	vaddr_t addr;
417 	vsize_t size;
418 
419 	/* Sanity check the size. */
420 	size = umd->md_size;
421 	addr = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
422 	if (!addr)
423 		return ENOMEM;
424 
425 	/* This unit is now configured. */
426 	sc->sc_addr = (caddr_t)addr; 	/* kernel space */
427 	sc->sc_size = (size_t)size;
428 	sc->sc_type = MD_KMEM_ALLOCATED;
429 	return 0;
430 }
431 
432 #if MEMORY_DISK_SERVER
433 
434 /*
435  * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER)
436  * Set config, then become the I/O server for this unit.
437  */
438 static int
439 md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
440     struct lwp *l __unused)
441 {
442 	vaddr_t end;
443 	int error;
444 
445 	/* Sanity check addr, size. */
446 	end = (vaddr_t) (umd->md_addr + umd->md_size);
447 
448 	if ((end >= VM_MAXUSER_ADDRESS) ||
449 		(end < ((vaddr_t) umd->md_addr)) )
450 		return EINVAL;
451 
452 	/* This unit is now configured. */
453 	sc->sc_addr = umd->md_addr; 	/* user space */
454 	sc->sc_size = umd->md_size;
455 	sc->sc_type = MD_UMEM_SERVER;
456 
457 	/* Become the server daemon */
458 	error = md_server_loop(sc);
459 
460 	/* This server is now going away! */
461 	sc->sc_type = MD_UNCONFIGURED;
462 	sc->sc_addr = 0;
463 	sc->sc_size = 0;
464 
465 	return (error);
466 }
467 
468 static int md_sleep_pri = PWAIT | PCATCH;
469 
470 static int
471 md_server_loop(struct md_softc *sc)
472 {
473 	struct buf *bp;
474 	caddr_t addr;	/* user space address */
475 	size_t off;	/* offset into "device" */
476 	size_t xfer;	/* amount to transfer */
477 	int error;
478 
479 	for (;;) {
480 		/* Wait for some work to arrive. */
481 		while ((bp = BUFQ_GET(sc->sc_buflist)) == NULL) {
482 			error = tsleep((caddr_t)sc, md_sleep_pri, "md_idle", 0);
483 			if (error)
484 				return error;
485 		}
486 
487 		/* Do the transfer to/from user space. */
488 		error = 0;
489 		bp->b_resid = bp->b_bcount;
490 		off = (bp->b_blkno << DEV_BSHIFT);
491 		if (off >= sc->sc_size) {
492 			if (bp->b_flags & B_READ)
493 				goto done;	/* EOF (not an error) */
494 			error = EIO;
495 			goto done;
496 		}
497 		xfer = bp->b_resid;
498 		if (xfer > (sc->sc_size - off))
499 			xfer = (sc->sc_size - off);
500 		addr = sc->sc_addr + off;
501 		if (bp->b_flags & B_READ)
502 			error = copyin(addr, bp->b_data, xfer);
503 		else
504 			error = copyout(bp->b_data, addr, xfer);
505 		if (!error)
506 			bp->b_resid -= xfer;
507 
508 	done:
509 		if (error) {
510 			bp->b_error = error;
511 			bp->b_flags |= B_ERROR;
512 		}
513 		biodone(bp);
514 	}
515 }
516 #endif	/* MEMORY_DISK_SERVER */
517