xref: /netbsd-src/sys/dev/md.c (revision 0d34d14bb949d7e0fe657c29532de778397cd58d) 
1 /*	$NetBSD: md.c,v 1.60 2009/07/28 17:55:27 dyoung 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 1.60 2009/07/28 17:55:27 dyoung 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/stat.h>
63 #include <sys/proc.h>
64 #include <sys/conf.h>
65 #include <sys/disklabel.h>
66 
67 #include <uvm/uvm_extern.h>
68 
69 #include <dev/md.h>
70 
71 /*
72  * The user-space functionality is included by default.
73  * Use  `options MEMORY_DISK_SERVER=0' to turn it off.
74  */
75 #ifndef MEMORY_DISK_SERVER
76 #error MEMORY_DISK_SERVER should be defined by opt_md.h
77 #endif	/* MEMORY_DISK_SERVER */
78 
79 /*
80  * We should use the raw partition for ioctl.
81  */
82 #define MD_UNIT(unit)	DISKUNIT(unit)
83 
84 /* autoconfig stuff... */
85 
86 struct md_softc {
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(device_t, device_t, void *);
99 static int	md_detach(device_t, int);
100 
101 static dev_type_open(mdopen);
102 static dev_type_close(mdclose);
103 static dev_type_read(mdread);
104 static dev_type_write(mdwrite);
105 static dev_type_ioctl(mdioctl);
106 static dev_type_strategy(mdstrategy);
107 static dev_type_size(mdsize);
108 
109 const struct bdevsw md_bdevsw = {
110 	mdopen, mdclose, mdstrategy, mdioctl, nodump, mdsize, D_DISK
111 };
112 
113 const struct cdevsw md_cdevsw = {
114 	mdopen, mdclose, mdread, mdwrite, mdioctl,
115 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
116 };
117 
118 static struct dkdriver mddkdriver = { mdstrategy, NULL };
119 
120 extern struct cfdriver md_cd;
121 CFATTACH_DECL3_NEW(md, sizeof(struct md_softc),
122 	0, md_attach, md_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
123 
124 extern size_t md_root_size;
125 
126 /*
127  * This is called if we are configured as a pseudo-device
128  */
129 void
130 mdattach(int n)
131 {
132 	int i;
133 	cfdata_t cf;
134 
135 #ifdef TFTPROOT
136 	/*
137 	 * Attachement of md0 must be done after md_root_setconf(),
138 	 * because the RAMdisk is not loaded yet.
139 	 */
140 	if (md_root_size == 0)
141 		return;
142 #endif
143 	if (config_cfattach_attach("md", &md_ca)) {
144 		printf("md: cfattach_attach failed\n");
145 		return;
146 	}
147 
148 	/* XXX:  Are we supposed to provide a default? */
149 	if (n <= 1)
150 		n = 1;
151 
152 	/* Attach as if by autoconfig. */
153 	for (i = 0; i < n; i++) {
154 		cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
155 		cf->cf_name = "md";
156 		cf->cf_atname = "md";
157 		cf->cf_unit = i;
158 		cf->cf_fstate = FSTATE_NOTFOUND;
159 		(void)config_attach_pseudo(cf);
160 	}
161 }
162 
163 static void
164 md_attach(device_t parent, device_t self, void *aux)
165 {
166 	struct md_softc *sc = device_private(self);
167 
168 	bufq_alloc(&sc->sc_buflist, "fcfs", 0);
169 
170 	/* XXX - Could accept aux info here to set the config. */
171 #ifdef	MEMORY_DISK_HOOKS
172 	/*
173 	 * This external function might setup a pre-loaded disk.
174 	 * All it would need to do is setup the md_conf struct.
175 	 * See sys/dev/md_root.c for an example.
176 	 */
177 	md_attach_hook(device_unit(self), &sc->sc_md);
178 #endif
179 
180 	/*
181 	 * Initialize and attach the disk structure.
182 	 */
183 	disk_init(&sc->sc_dkdev, device_xname(self), &mddkdriver);
184 	disk_attach(&sc->sc_dkdev);
185 
186 	if (!pmf_device_register(self, NULL, NULL))
187 		aprint_error_dev(self, "couldn't establish power handler\n");
188 }
189 
190 static int
191 md_detach(device_t self, int flags)
192 {
193 	struct md_softc *sc = device_private(self);
194 	int rc;
195 
196 	rc = 0;
197 	mutex_enter(&sc->sc_dkdev.dk_openlock);
198 	if (sc->sc_dkdev.dk_openmask == 0)
199 		;	/* nothing to do */
200 	else if ((flags & DETACH_FORCE) == 0)
201 		rc = EBUSY;
202 	mutex_exit(&sc->sc_dkdev.dk_openlock);
203 
204 	if (rc != 0)
205 		return rc;
206 
207 	pmf_device_deregister(self);
208 	disk_detach(&sc->sc_dkdev);
209 	disk_destroy(&sc->sc_dkdev);
210 	bufq_free(sc->sc_buflist);
211 	return 0;
212 }
213 
214 /*
215  * operational routines:
216  * open, close, read, write, strategy,
217  * ioctl, dump, size
218  */
219 
220 #if MEMORY_DISK_SERVER
221 static int	md_server_loop(struct md_softc *sc);
222 static int	md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
223 		    struct lwp *l);
224 #endif	/* MEMORY_DISK_SERVER */
225 static int	md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
226 		    struct lwp *l);
227 
228 static int
229 mdsize(dev_t dev)
230 {
231 	struct md_softc *sc;
232 
233 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
234 	if (sc == NULL)
235 		return 0;
236 
237 	if (sc->sc_type == MD_UNCONFIGURED)
238 		return 0;
239 
240 	return (sc->sc_size >> DEV_BSHIFT);
241 }
242 
243 static int
244 mdopen(dev_t dev, int flag, int fmt, struct lwp *l)
245 {
246 	int unit;
247 	int part = DISKPART(dev);
248 	int pmask = 1 << part;
249 	struct md_softc *sc;
250 	struct disk *dk;
251 
252 	unit = MD_UNIT(dev);
253 	sc = device_lookup_private(&md_cd, unit);
254 	if (sc == NULL)
255 		return ENXIO;
256 
257 	dk = &sc->sc_dkdev;
258 
259 	/*
260 	 * The raw partition is used for ioctl to configure.
261 	 */
262 	if (part == RAW_PART)
263 		goto ok;
264 
265 #ifdef	MEMORY_DISK_HOOKS
266 	/* Call the open hook to allow loading the device. */
267 	md_open_hook(unit, &sc->sc_md);
268 #endif
269 
270 	/*
271 	 * This is a normal, "slave" device, so
272 	 * enforce initialized.
273 	 */
274 	if (sc->sc_type == MD_UNCONFIGURED)
275 		return ENXIO;
276 
277 ok:
278 	/* XXX duplicates code in dk_open().  Call dk_open(), instead? */
279 	mutex_enter(&dk->dk_openlock);
280 	/* Mark our unit as open. */
281 	switch (fmt) {
282 	case S_IFCHR:
283 		dk->dk_copenmask |= pmask;
284 		break;
285 	case S_IFBLK:
286 		dk->dk_bopenmask |= pmask;
287 		break;
288 	}
289 
290 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
291 
292 	mutex_exit(&dk->dk_openlock);
293 	return 0;
294 }
295 
296 static int
297 mdclose(dev_t dev, int flag, int fmt, struct lwp *l)
298 {
299 	int part = DISKPART(dev);
300 	int pmask = 1 << part;
301 	struct md_softc *sc;
302 	struct disk *dk;
303 
304 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
305 	if (sc == NULL)
306 		return ENXIO;
307 
308 	dk = &sc->sc_dkdev;
309 
310 	mutex_enter(&dk->dk_openlock);
311 
312 	switch (fmt) {
313 	case S_IFCHR:
314 		dk->dk_copenmask &= ~pmask;
315 		break;
316 	case S_IFBLK:
317 		dk->dk_bopenmask &= ~pmask;
318 		break;
319 	}
320 	dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
321 
322 	mutex_exit(&dk->dk_openlock);
323 	return 0;
324 }
325 
326 static int
327 mdread(dev_t dev, struct uio *uio, int flags)
328 {
329 	struct md_softc *sc;
330 
331 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
332 
333 	if (sc->sc_type == MD_UNCONFIGURED)
334 		return ENXIO;
335 
336 	return (physio(mdstrategy, NULL, dev, B_READ, minphys, uio));
337 }
338 
339 static int
340 mdwrite(dev_t dev, struct uio *uio, int flags)
341 {
342 	struct md_softc *sc;
343 
344 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
345 
346 	if (sc->sc_type == MD_UNCONFIGURED)
347 		return ENXIO;
348 
349 	return (physio(mdstrategy, NULL, dev, B_WRITE, minphys, uio));
350 }
351 
352 /*
353  * Handle I/O requests, either directly, or
354  * by passing them to the server process.
355  */
356 static void
357 mdstrategy(struct buf *bp)
358 {
359 	struct md_softc	*sc;
360 	void *	addr;
361 	size_t off, xfer;
362 
363 	sc = device_lookup_private(&md_cd, MD_UNIT(bp->b_dev));
364 
365 	if (sc->sc_type == MD_UNCONFIGURED) {
366 		bp->b_error = ENXIO;
367 		goto done;
368 	}
369 
370 	switch (sc->sc_type) {
371 #if MEMORY_DISK_SERVER
372 	case MD_UMEM_SERVER:
373 		/* Just add this job to the server's queue. */
374 		bufq_put(sc->sc_buflist, bp);
375 		wakeup((void *)sc);
376 		/* see md_server_loop() */
377 		/* no biodone in this case */
378 		return;
379 #endif	/* MEMORY_DISK_SERVER */
380 
381 	case MD_KMEM_FIXED:
382 	case MD_KMEM_ALLOCATED:
383 		/* These are in kernel space.  Access directly. */
384 		bp->b_resid = bp->b_bcount;
385 		off = (bp->b_blkno << DEV_BSHIFT);
386 		if (off >= sc->sc_size) {
387 			if (bp->b_flags & B_READ)
388 				break;	/* EOF */
389 			goto set_eio;
390 		}
391 		xfer = bp->b_resid;
392 		if (xfer > (sc->sc_size - off))
393 			xfer = (sc->sc_size - off);
394 		addr = (char *)sc->sc_addr + off;
395 		if (bp->b_flags & B_READ)
396 			memcpy(bp->b_data, addr, xfer);
397 		else
398 			memcpy(addr, bp->b_data, xfer);
399 		bp->b_resid -= xfer;
400 		break;
401 
402 	default:
403 		bp->b_resid = bp->b_bcount;
404 	set_eio:
405 		bp->b_error = EIO;
406 		break;
407 	}
408  done:
409 	biodone(bp);
410 }
411 
412 static int
413 mdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
414 {
415 	struct md_softc *sc;
416 	struct md_conf *umd;
417 
418 	sc = device_lookup_private(&md_cd, MD_UNIT(dev));
419 
420 	/* If this is not the raw partition, punt! */
421 	if (DISKPART(dev) != RAW_PART)
422 		return ENOTTY;
423 
424 	umd = (struct md_conf *)data;
425 	switch (cmd) {
426 	case MD_GETCONF:
427 		*umd = sc->sc_md;
428 		return 0;
429 
430 	case MD_SETCONF:
431 		/* Can only set it once. */
432 		if (sc->sc_type != MD_UNCONFIGURED)
433 			break;
434 		switch (umd->md_type) {
435 		case MD_KMEM_ALLOCATED:
436 			return md_ioctl_kalloc(sc, umd, l);
437 #if MEMORY_DISK_SERVER
438 		case MD_UMEM_SERVER:
439 			return md_ioctl_server(sc, umd, l);
440 #endif	/* MEMORY_DISK_SERVER */
441 		default:
442 			break;
443 		}
444 		break;
445 	}
446 	return EINVAL;
447 }
448 
449 /*
450  * Handle ioctl MD_SETCONF for (sc_type == MD_KMEM_ALLOCATED)
451  * Just allocate some kernel memory and return.
452  */
453 static int
454 md_ioctl_kalloc(struct md_softc *sc, struct md_conf *umd,
455     struct lwp *l)
456 {
457 	vaddr_t addr;
458 	vsize_t size;
459 
460 	/* Sanity check the size. */
461 	size = umd->md_size;
462 	addr = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
463 	if (!addr)
464 		return ENOMEM;
465 
466 	/* This unit is now configured. */
467 	sc->sc_addr = (void *)addr; 	/* kernel space */
468 	sc->sc_size = (size_t)size;
469 	sc->sc_type = MD_KMEM_ALLOCATED;
470 	return 0;
471 }
472 
473 #if MEMORY_DISK_SERVER
474 
475 /*
476  * Handle ioctl MD_SETCONF for (sc_type == MD_UMEM_SERVER)
477  * Set config, then become the I/O server for this unit.
478  */
479 static int
480 md_ioctl_server(struct md_softc *sc, struct md_conf *umd,
481     struct lwp *l)
482 {
483 	vaddr_t end;
484 	int error;
485 
486 	/* Sanity check addr, size. */
487 	end = (vaddr_t) ((char *)umd->md_addr + umd->md_size);
488 
489 	if ((end >= VM_MAXUSER_ADDRESS) ||
490 		(end < ((vaddr_t) umd->md_addr)) )
491 		return EINVAL;
492 
493 	/* This unit is now configured. */
494 	sc->sc_addr = umd->md_addr; 	/* user space */
495 	sc->sc_size = umd->md_size;
496 	sc->sc_type = MD_UMEM_SERVER;
497 
498 	/* Become the server daemon */
499 	error = md_server_loop(sc);
500 
501 	/* This server is now going away! */
502 	sc->sc_type = MD_UNCONFIGURED;
503 	sc->sc_addr = 0;
504 	sc->sc_size = 0;
505 
506 	return (error);
507 }
508 
509 static int md_sleep_pri = PWAIT | PCATCH;
510 
511 static int
512 md_server_loop(struct md_softc *sc)
513 {
514 	struct buf *bp;
515 	void *addr;	/* user space address */
516 	size_t off;	/* offset into "device" */
517 	size_t xfer;	/* amount to transfer */
518 	int error;
519 
520 	for (;;) {
521 		/* Wait for some work to arrive. */
522 		while ((bp = bufq_get(sc->sc_buflist)) == NULL) {
523 			error = tsleep((void *)sc, md_sleep_pri, "md_idle", 0);
524 			if (error)
525 				return error;
526 		}
527 
528 		/* Do the transfer to/from user space. */
529 		error = 0;
530 		bp->b_resid = bp->b_bcount;
531 		off = (bp->b_blkno << DEV_BSHIFT);
532 		if (off >= sc->sc_size) {
533 			if (bp->b_flags & B_READ)
534 				goto done;	/* EOF (not an error) */
535 			error = EIO;
536 			goto done;
537 		}
538 		xfer = bp->b_resid;
539 		if (xfer > (sc->sc_size - off))
540 			xfer = (sc->sc_size - off);
541 		addr = (char *)sc->sc_addr + off;
542 		if (bp->b_flags & B_READ)
543 			error = copyin(addr, bp->b_data, xfer);
544 		else
545 			error = copyout(bp->b_data, addr, xfer);
546 		if (!error)
547 			bp->b_resid -= xfer;
548 
549 	done:
550 		if (error) {
551 			bp->b_error = error;
552 		}
553 		biodone(bp);
554 	}
555 }
556 #endif	/* MEMORY_DISK_SERVER */
557