xref: /netbsd-src/sys/dev/md.c (revision c8da0e5fefd3800856b306200a18b2315c7fbb9f) 
1 /*	$NetBSD: md.c,v 1.58 2009/05/04 16:20:41 manu 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 1.58 2009/05/04 16:20:41 manu Exp $");
50 
51 #include "opt_md.h"
52 #include "opt_tftproot.h"
53 
54 #include <sys/param.h>
55 #include <sys/kernel.h>
56 #include <sys/malloc.h>
57 #include <sys/systm.h>
58 #include <sys/buf.h>
59 #include <sys/bufq.h>
60 #include <sys/device.h>
61 #include <sys/disk.h>
62 #include <sys/proc.h>
63 #include <sys/conf.h>
64 #include <sys/disklabel.h>
65 
66 #include <uvm/uvm_extern.h>
67 
68 #include <dev/md.h>
69 
70 /*
71  * The user-space functionality is included by default.
72  * Use  `options MEMORY_DISK_SERVER=0' to turn it off.
73  */
74 #ifndef MEMORY_DISK_SERVER
75 #error MEMORY_DISK_SERVER should be defined by opt_md.h
76 #endif	/* MEMORY_DISK_SERVER */
77 
78 /*
79  * We should use the raw partition for ioctl.
80  */
81 #define MD_UNIT(unit)	DISKUNIT(unit)
82 
83 /* autoconfig stuff... */
84 
85 struct md_softc {
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(device_t, device_t, 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, NULL };
117 
118 extern struct cfdriver md_cd;
119 CFATTACH_DECL_NEW(md, sizeof(struct md_softc),
120 	0, md_attach, 0, NULL);
121 
122 extern size_t md_root_size;
123 
124 /*
125  * This is called if we are configured as a pseudo-device
126  */
127 void
128 mdattach(int n)
129 {
130 	int i;
131 	cfdata_t cf;
132 
133 #ifdef TFTPROOT
134 	/*
135 	 * Attachement of md0 must be done after md_root_setconf(),
136 	 * because the RAMdisk is not loaded yet.
137 	 */
138 	if (md_root_size == 0)
139 		return;
140 #endif
141 	if (config_cfattach_attach("md", &md_ca)) {
142 		printf("md: cfattach_attach failed\n");
143 		return;
144 	}
145 
146 	/* XXX:  Are we supposed to provide a default? */
147 	if (n <= 1)
148 		n = 1;
149 
150 	/* Attach as if by autoconfig. */
151 	for (i = 0; i < n; i++) {
152 		cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
153 		cf->cf_name = "md";
154 		cf->cf_atname = "md";
155 		cf->cf_unit = i;
156 		cf->cf_fstate = FSTATE_NOTFOUND;
157 		(void)config_attach_pseudo(cf);
158 	}
159 }
160 
161 static void
162 md_attach(device_t parent, device_t self,
163     void *aux)
164 {
165 	struct md_softc *sc = device_private(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(self), &sc->sc_md);
177 #endif
178 
179 	/*
180 	 * Initialize and attach the disk structure.
181 	 */
182 	disk_init(&sc->sc_dkdev, device_xname(self), &mddkdriver);
183 	disk_attach(&sc->sc_dkdev);
184 
185 	if (!pmf_device_register(self, NULL, NULL))
186 		aprint_error_dev(self, "couldn't establish power handler\n");
187 }
188 
189 /*
190  * operational routines:
191  * open, close, read, write, strategy,
192  * ioctl, dump, size
193  */
194 
195 #if MEMORY_DISK_SERVER
196 static int	md_server_loop(struct md_softc *sc);
197 static int	md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
198 		    struct lwp *l);
199 #endif	/* MEMORY_DISK_SERVER */
200 static int	md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
201 		    struct lwp *l);
202 
203 static int
204 mdsize(dev_t dev)
205 {
206 	struct md_softc *sc;
207 
208 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
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 lwp *l)
220 {
221 	int unit;
222 	struct md_softc *sc;
223 
224 	unit = MD_UNIT(dev);
225 	sc = device_lookup_private(&md_cd, unit);
226 	if (sc == NULL)
227 		return ENXIO;
228 
229 	/*
230 	 * The raw partition is used for ioctl to configure.
231 	 */
232 	if (DISKPART(dev) == RAW_PART)
233 		return 0;
234 
235 #ifdef	MEMORY_DISK_HOOKS
236 	/* Call the open hook to allow loading the device. */
237 	md_open_hook(unit, &sc->sc_md);
238 #endif
239 
240 	/*
241 	 * This is a normal, "slave" device, so
242 	 * enforce initialized.
243 	 */
244 	if (sc->sc_type == MD_UNCONFIGURED)
245 		return ENXIO;
246 
247 	return 0;
248 }
249 
250 static int
251 mdclose(dev_t dev, int flag, int fmt, struct lwp *l)
252 {
253 
254 	return 0;
255 }
256 
257 static int
258 mdread(dev_t dev, struct uio *uio, int flags)
259 {
260 	struct md_softc *sc;
261 
262 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
263 
264 	if (sc->sc_type == MD_UNCONFIGURED)
265 		return ENXIO;
266 
267 	return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio));
268 }
269 
270 static int
271 mdwrite(dev_t dev, struct uio *uio, int flags)
272 {
273 	struct md_softc *sc;
274 
275 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
276 
277 	if (sc->sc_type == MD_UNCONFIGURED)
278 		return ENXIO;
279 
280 	return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio));
281 }
282 
283 /*
284  * Handle I/O requests, either directly, or
285  * by passing them to the server process.
286  */
287 static void
288 mdstrategy(struct buf *bp)
289 {
290 	struct md_softc	*sc;
291 	void *	addr;
292 	size_t off, xfer;
293 
294 	sc = device_lookup_private(&md_cd, MD_UNIT(bp->b_dev));
295 
296 	if (sc->sc_type == MD_UNCONFIGURED) {
297 		bp->b_error = ENXIO;
298 		goto done;
299 	}
300 
301 	switch (sc->sc_type) {
302 #if MEMORY_DISK_SERVER
303 	case MD_UMEM_SERVER:
304 		/* Just add this job to the server's queue. */
305 		bufq_put(sc->sc_buflist, bp);
306 		wakeup((void *)sc);
307 		/* see md_server_loop() */
308 		/* no biodone in this case */
309 		return;
310 #endif	/* MEMORY_DISK_SERVER */
311 
312 	case MD_KMEM_FIXED:
313 	case MD_KMEM_ALLOCATED:
314 		/* These are in kernel space.  Access directly. */
315 		bp->b_resid = bp->b_bcount;
316 		off = (bp->b_blkno << DEV_BSHIFT);
317 		if (off >= sc->sc_size) {
318 			if (bp->b_flags & B_READ)
319 				break;	/* EOF */
320 			goto set_eio;
321 		}
322 		xfer = bp->b_resid;
323 		if (xfer > (sc->sc_size - off))
324 			xfer = (sc->sc_size - off);
325 		addr = (char *)sc->sc_addr + off;
326 		if (bp->b_flags & B_READ)
327 			memcpy(bp->b_data, addr, xfer);
328 		else
329 			memcpy(addr, bp->b_data, xfer);
330 		bp->b_resid -= xfer;
331 		break;
332 
333 	default:
334 		bp->b_resid = bp->b_bcount;
335 	set_eio:
336 		bp->b_error = EIO;
337 		break;
338 	}
339  done:
340 	biodone(bp);
341 }
342 
343 static int
344 mdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
345 {
346 	struct md_softc *sc;
347 	struct md_conf *umd;
348 
349 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
350 
351 	/* If this is not the raw partition, punt! */
352 	if (DISKPART(dev) != RAW_PART)
353 		return ENOTTY;
354 
355 	umd = (struct md_conf *)data;
356 	switch (cmd) {
357 	case MD_GETCONF:
358 		*umd = sc->sc_md;
359 		return 0;
360 
361 	case MD_SETCONF:
362 		/* Can only set it once. */
363 		if (sc->sc_type != MD_UNCONFIGURED)
364 			break;
365 		switch (umd->md_type) {
366 		case MD_KMEM_ALLOCATED:
367 			return md_ioctl_kalloc(sc, umd, l);
368 #if MEMORY_DISK_SERVER
369 		case MD_UMEM_SERVER:
370 			return md_ioctl_server(sc, umd, l);
371 #endif	/* MEMORY_DISK_SERVER */
372 		default:
373 			break;
374 		}
375 		break;
376 	}
377 	return EINVAL;
378 }
379 
380 /*
381  * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED)
382  * Just allocate some kernel memory and return.
383  */
384 static int
385 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
386     struct lwp *l)
387 {
388 	vaddr_t addr;
389 	vsize_t size;
390 
391 	/* Sanity check the size. */
392 	size = umd->md_size;
393 	addr = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
394 	if (!addr)
395 		return ENOMEM;
396 
397 	/* This unit is now configured. */
398 	sc->sc_addr = (void *)addr; 	/* kernel space */
399 	sc->sc_size = (size_t)size;
400 	sc->sc_type = MD_KMEM_ALLOCATED;
401 	return 0;
402 }
403 
404 #if MEMORY_DISK_SERVER
405 
406 /*
407  * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER)
408  * Set config, then become the I/O server for this unit.
409  */
410 static int
411 md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
412     struct lwp *l)
413 {
414 	vaddr_t end;
415 	int error;
416 
417 	/* Sanity check addr, size. */
418 	end = (vaddr_t) ((char *)umd->md_addr + umd->md_size);
419 
420 	if ((end >= VM_MAXUSER_ADDRESS) ||
421 		(end < ((vaddr_t) umd->md_addr)) )
422 		return EINVAL;
423 
424 	/* This unit is now configured. */
425 	sc->sc_addr = umd->md_addr; 	/* user space */
426 	sc->sc_size = umd->md_size;
427 	sc->sc_type = MD_UMEM_SERVER;
428 
429 	/* Become the server daemon */
430 	error = md_server_loop(sc);
431 
432 	/* This server is now going away! */
433 	sc->sc_type = MD_UNCONFIGURED;
434 	sc->sc_addr = 0;
435 	sc->sc_size = 0;
436 
437 	return (error);
438 }
439 
440 static int md_sleep_pri = PWAIT | PCATCH;
441 
442 static int
443 md_server_loop(struct md_softc *sc)
444 {
445 	struct buf *bp;
446 	void *addr;	/* user space address */
447 	size_t off;	/* offset into "device" */
448 	size_t xfer;	/* amount to transfer */
449 	int error;
450 
451 	for (;;) {
452 		/* Wait for some work to arrive. */
453 		while ((bp = bufq_get(sc->sc_buflist)) == NULL) {
454 			error = tsleep((void *)sc, md_sleep_pri, "md_idle", 0);
455 			if (error)
456 				return error;
457 		}
458 
459 		/* Do the transfer to/from user space. */
460 		error = 0;
461 		bp->b_resid = bp->b_bcount;
462 		off = (bp->b_blkno << DEV_BSHIFT);
463 		if (off >= sc->sc_size) {
464 			if (bp->b_flags & B_READ)
465 				goto done;	/* EOF (not an error) */
466 			error = EIO;
467 			goto done;
468 		}
469 		xfer = bp->b_resid;
470 		if (xfer > (sc->sc_size - off))
471 			xfer = (sc->sc_size - off);
472 		addr = (char *)sc->sc_addr + off;
473 		if (bp->b_flags & B_READ)
474 			error = copyin(addr, bp->b_data, xfer);
475 		else
476 			error = copyout(bp->b_data, addr, xfer);
477 		if (!error)
478 			bp->b_resid -= xfer;
479 
480 	done:
481 		if (error) {
482 			bp->b_error = error;
483 		}
484 		biodone(bp);
485 	}
486 }
487 #endif	/* MEMORY_DISK_SERVER */
488