xref: /netbsd-src/sys/dev/md.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: md.c,v 1.51 2007/10/08 16:41:10 ad 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 2007/10/08 16:41:10 ad 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 
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 	disk_init(&sc->sc_dkdev, sc->sc_dev.dv_xname, &mddkdriver);
183 	disk_attach(&sc->sc_dkdev);
184 }
185 
186 /*
187  * operational routines:
188  * open, close, read, write, strategy,
189  * ioctl, dump, size
190  */
191 
192 #if MEMORY_DISK_SERVER
193 static int	md_server_loop(struct md_softc *sc);
194 static int	md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
195 		    struct lwp *l);
196 #endif	/* MEMORY_DISK_SERVER */
197 static int	md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
198 		    struct lwp *l);
199 
200 static int
201 mdsize(dev_t dev)
202 {
203 	int unit;
204 	struct md_softc *sc;
205 
206 	unit = MD_UNIT(dev);
207 	if (unit >= ramdisk_ndevs)
208 		return 0;
209 	sc = ramdisk_devs[unit];
210 	if (sc == NULL)
211 		return 0;
212 
213 	if (sc->sc_type == MD_UNCONFIGURED)
214 		return 0;
215 
216 	return (sc->sc_size >> DEV_BSHIFT);
217 }
218 
219 static int
220 mdopen(dev_t dev, int flag, int fmt, struct lwp *l)
221 {
222 	int unit;
223 	struct md_softc *sc;
224 
225 	unit = MD_UNIT(dev);
226 	if (unit >= ramdisk_ndevs)
227 		return ENXIO;
228 	sc = ramdisk_devs[unit];
229 	if (sc == NULL)
230 		return ENXIO;
231 
232 	/*
233 	 * The raw partition is used for ioctl to configure.
234 	 */
235 	if (DISKPART(dev) == RAW_PART)
236 		return 0;
237 
238 #ifdef	MEMORY_DISK_HOOKS
239 	/* Call the open hook to allow loading the device. */
240 	md_open_hook(unit, &sc->sc_md);
241 #endif
242 
243 	/*
244 	 * This is a normal, "slave" device, so
245 	 * enforce initialized.
246 	 */
247 	if (sc->sc_type == MD_UNCONFIGURED)
248 		return ENXIO;
249 
250 	return 0;
251 }
252 
253 static int
254 mdclose(dev_t dev, int flag, int fmt, struct lwp *l)
255 {
256 	int unit;
257 
258 	unit = MD_UNIT(dev);
259 
260 	if (unit >= ramdisk_ndevs)
261 		return ENXIO;
262 
263 	return 0;
264 }
265 
266 static int
267 mdread(dev_t dev, struct uio *uio, int flags)
268 {
269 	int unit;
270 	struct md_softc *sc;
271 
272 	unit = MD_UNIT(dev);
273 
274 	if (unit >= ramdisk_ndevs)
275 		return ENXIO;
276 
277 	sc = ramdisk_devs[unit];
278 
279 	if (sc->sc_type == MD_UNCONFIGURED)
280 		return ENXIO;
281 
282 	return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio));
283 }
284 
285 static int
286 mdwrite(dev_t dev, struct uio *uio, int flags)
287 {
288 	int unit;
289 	struct md_softc *sc;
290 
291 	unit = MD_UNIT(dev);
292 
293 	if (unit >= ramdisk_ndevs)
294 		return ENXIO;
295 
296 	sc = ramdisk_devs[unit];
297 
298 	if (sc->sc_type == MD_UNCONFIGURED)
299 		return ENXIO;
300 
301 	return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio));
302 }
303 
304 /*
305  * Handle I/O requests, either directly, or
306  * by passing them to the server process.
307  */
308 static void
309 mdstrategy(struct buf *bp)
310 {
311 	int unit;
312 	struct md_softc	*sc;
313 	void *	addr;
314 	size_t off, xfer;
315 
316 	unit = MD_UNIT(bp->b_dev);
317 	sc = ramdisk_devs[unit];
318 
319 	if (sc->sc_type == MD_UNCONFIGURED) {
320 		bp->b_error = ENXIO;
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((void *)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 = (char *)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 		break;
361 	}
362  done:
363 	biodone(bp);
364 }
365 
366 static int
367 mdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
368 {
369 	int unit;
370 	struct md_softc *sc;
371 	struct md_conf *umd;
372 
373 	unit = MD_UNIT(dev);
374 	sc = ramdisk_devs[unit];
375 
376 	/* If this is not the raw partition, punt! */
377 	if (DISKPART(dev) != RAW_PART)
378 		return ENOTTY;
379 
380 	umd = (struct md_conf *)data;
381 	switch (cmd) {
382 	case MD_GETCONF:
383 		*umd = sc->sc_md;
384 		return 0;
385 
386 	case MD_SETCONF:
387 		/* Can only set it once. */
388 		if (sc->sc_type != MD_UNCONFIGURED)
389 			break;
390 		switch (umd->md_type) {
391 		case MD_KMEM_ALLOCATED:
392 			return md_ioctl_kalloc(sc, umd, l);
393 #if MEMORY_DISK_SERVER
394 		case MD_UMEM_SERVER:
395 			return md_ioctl_server(sc, umd, l);
396 #endif	/* MEMORY_DISK_SERVER */
397 		default:
398 			break;
399 		}
400 		break;
401 	}
402 	return EINVAL;
403 }
404 
405 /*
406  * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED)
407  * Just allocate some kernel memory and return.
408  */
409 static int
410 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
411     struct lwp *l)
412 {
413 	vaddr_t addr;
414 	vsize_t size;
415 
416 	/* Sanity check the size. */
417 	size = umd->md_size;
418 	addr = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
419 	if (!addr)
420 		return ENOMEM;
421 
422 	/* This unit is now configured. */
423 	sc->sc_addr = (void *)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,
437     struct lwp *l)
438 {
439 	vaddr_t end;
440 	int error;
441 
442 	/* Sanity check addr, size. */
443 	end = (vaddr_t) ((char *)umd->md_addr + umd->md_size);
444 
445 	if ((end >= VM_MAXUSER_ADDRESS) ||
446 		(end < ((vaddr_t) umd->md_addr)) )
447 		return EINVAL;
448 
449 	/* This unit is now configured. */
450 	sc->sc_addr = umd->md_addr; 	/* user space */
451 	sc->sc_size = umd->md_size;
452 	sc->sc_type = MD_UMEM_SERVER;
453 
454 	/* Become the server daemon */
455 	error = md_server_loop(sc);
456 
457 	/* This server is now going away! */
458 	sc->sc_type = MD_UNCONFIGURED;
459 	sc->sc_addr = 0;
460 	sc->sc_size = 0;
461 
462 	return (error);
463 }
464 
465 static int md_sleep_pri = PWAIT | PCATCH;
466 
467 static int
468 md_server_loop(struct md_softc *sc)
469 {
470 	struct buf *bp;
471 	void *addr;	/* user space address */
472 	size_t off;	/* offset into "device" */
473 	size_t xfer;	/* amount to transfer */
474 	int error;
475 
476 	for (;;) {
477 		/* Wait for some work to arrive. */
478 		while ((bp = BUFQ_GET(sc->sc_buflist)) == NULL) {
479 			error = tsleep((void *)sc, md_sleep_pri, "md_idle", 0);
480 			if (error)
481 				return error;
482 		}
483 
484 		/* Do the transfer to/from user space. */
485 		error = 0;
486 		bp->b_resid = bp->b_bcount;
487 		off = (bp->b_blkno << DEV_BSHIFT);
488 		if (off >= sc->sc_size) {
489 			if (bp->b_flags & B_READ)
490 				goto done;	/* EOF (not an error) */
491 			error = EIO;
492 			goto done;
493 		}
494 		xfer = bp->b_resid;
495 		if (xfer > (sc->sc_size - off))
496 			xfer = (sc->sc_size - off);
497 		addr = (char *)sc->sc_addr + off;
498 		if (bp->b_flags & B_READ)
499 			error = copyin(addr, bp->b_data, xfer);
500 		else
501 			error = copyout(bp->b_data, addr, xfer);
502 		if (!error)
503 			bp->b_resid -= xfer;
504 
505 	done:
506 		if (error) {
507 			bp->b_error = error;
508 		}
509 		biodone(bp);
510 	}
511 }
512 #endif	/* MEMORY_DISK_SERVER */
513