xref: /netbsd-src/sys/dev/dm/device-mapper.c (revision bbde328be4e75ea9ad02e9715ea13ca54b797ada)
1 /*        $NetBSD: device-mapper.c,v 1.22 2010/03/26 15:46:04 jakllsch Exp $ */
2 
3 /*
4  * Copyright (c) 2010 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Adam Hamsik.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * I want to say thank you to all people who helped me with this project.
34  */
35 
36 #include <sys/types.h>
37 #include <sys/param.h>
38 
39 #include <sys/buf.h>
40 #include <sys/conf.h>
41 #include <sys/device.h>
42 #include <sys/dkio.h>
43 #include <sys/disk.h>
44 #include <sys/disklabel.h>
45 #include <sys/ioctl.h>
46 #include <sys/ioccom.h>
47 #include <sys/kmem.h>
48 
49 #include "netbsd-dm.h"
50 #include "dm.h"
51 
52 static dev_type_open(dmopen);
53 static dev_type_close(dmclose);
54 static dev_type_read(dmread);
55 static dev_type_write(dmwrite);
56 static dev_type_ioctl(dmioctl);
57 static dev_type_strategy(dmstrategy);
58 static dev_type_size(dmsize);
59 
60 /* attach and detach routines */
61 void dmattach(int);
62 #ifdef _MODULE
63 static int dmdestroy(void);
64 #endif
65 
66 static void dm_doinit(void);
67 
68 static int dm_cmd_to_fun(prop_dictionary_t);
69 static int disk_ioctl_switch(dev_t, u_long, void *);
70 static int dm_ioctl_switch(u_long);
71 static void dmminphys(struct buf *);
72 
73 /* CF attach/detach functions used for power management */
74 static int dm_detach(device_t, int);
75 static void dm_attach(device_t, device_t, void *);
76 static int dm_match(device_t, cfdata_t, void *);
77 
78 /* ***Variable-definitions*** */
79 const struct bdevsw dm_bdevsw = {
80 	.d_open = dmopen,
81 	.d_close = dmclose,
82 	.d_strategy = dmstrategy,
83 	.d_ioctl = dmioctl,
84 	.d_dump = nodump,
85 	.d_psize = dmsize,
86 	.d_flag = D_DISK | D_MPSAFE
87 };
88 
89 const struct cdevsw dm_cdevsw = {
90 	.d_open = dmopen,
91 	.d_close = dmclose,
92 	.d_read = dmread,
93 	.d_write = dmwrite,
94 	.d_ioctl = dmioctl,
95 	.d_stop = nostop,
96 	.d_tty = notty,
97 	.d_poll = nopoll,
98 	.d_mmap = nommap,
99 	.d_kqfilter = nokqfilter,
100 	.d_flag = D_DISK | D_MPSAFE
101 };
102 
103 const struct dkdriver dmdkdriver = {
104 	.d_strategy = dmstrategy
105 };
106 
107 CFATTACH_DECL3_NEW(dm, 0,
108      dm_match, dm_attach, dm_detach, NULL, NULL, NULL,
109      DVF_DETACH_SHUTDOWN);
110 
111 extern struct cfdriver dm_cd;
112 
113 extern uint64_t dm_dev_counter;
114 
115 /*
116  * This array is used to translate cmd to function pointer.
117  *
118  * Interface between libdevmapper and lvm2tools uses different
119  * names for one IOCTL call because libdevmapper do another thing
120  * then. When I run "info" or "mknodes" libdevmapper will send same
121  * ioctl to kernel but will do another things in userspace.
122  *
123  */
124 struct cmd_function cmd_fn[] = {
125 		{ .cmd = "version", .fn = dm_get_version_ioctl},
126 		{ .cmd = "targets", .fn = dm_list_versions_ioctl},
127 		{ .cmd = "create",  .fn = dm_dev_create_ioctl},
128 		{ .cmd = "info",    .fn = dm_dev_status_ioctl},
129 		{ .cmd = "mknodes", .fn = dm_dev_status_ioctl},
130 		{ .cmd = "names",   .fn = dm_dev_list_ioctl},
131 		{ .cmd = "suspend", .fn = dm_dev_suspend_ioctl},
132 		{ .cmd = "remove",  .fn = dm_dev_remove_ioctl},
133 		{ .cmd = "rename",  .fn = dm_dev_rename_ioctl},
134 		{ .cmd = "resume",  .fn = dm_dev_resume_ioctl},
135 		{ .cmd = "clear",   .fn = dm_table_clear_ioctl},
136 		{ .cmd = "deps",    .fn = dm_table_deps_ioctl},
137 		{ .cmd = "reload",  .fn = dm_table_load_ioctl},
138 		{ .cmd = "status",  .fn = dm_table_status_ioctl},
139 		{ .cmd = "table",   .fn = dm_table_status_ioctl},
140 		{NULL, NULL}
141 };
142 
143 #ifdef _MODULE
144 #include <sys/module.h>
145 
146 /* Autoconf defines */
147 CFDRIVER_DECL(dm, DV_DISK, NULL);
148 
149 MODULE(MODULE_CLASS_DRIVER, dm, NULL);
150 
151 /* New module handle routine */
152 static int
153 dm_modcmd(modcmd_t cmd, void *arg)
154 {
155 	int error, bmajor, cmajor;
156 
157 	error = 0;
158 	bmajor = -1;
159 	cmajor = -1;
160 
161 	switch (cmd) {
162 	case MODULE_CMD_INIT:
163 		error = config_cfdriver_attach(&dm_cd);
164 		if (error)
165 			break;
166 
167 		error = config_cfattach_attach(dm_cd.cd_name, &dm_ca);
168 		if (error) {
169 			aprint_error("%s: unable to register cfattach\n",
170 			    dm_cd.cd_name);
171 			return error;
172 		}
173 
174 		error = devsw_attach(dm_cd.cd_name, &dm_bdevsw, &bmajor,
175 		    &dm_cdevsw, &cmajor);
176 		if (error) {
177 			config_cfattach_detach(dm_cd.cd_name, &dm_ca);
178 			config_cfdriver_detach(&dm_cd);
179 			break;
180 		}
181 
182 		dm_doinit();
183 
184 		break;
185 
186 	case MODULE_CMD_FINI:
187 		/*
188 		 * Disable unloading of dm module if there are any devices
189 		 * defined in driver. This is probably too strong we need
190 		 * to disable auto-unload only if there is mounted dm device
191 		 * present.
192 		 */
193 		if (dm_dev_counter > 0)
194 			return EBUSY;
195 
196 		error = dmdestroy();
197 		if (error)
198 			break;
199 
200 		config_cfdriver_detach(&dm_cd);
201 
202 		devsw_detach(&dm_bdevsw, &dm_cdevsw);
203 		break;
204 	case MODULE_CMD_STAT:
205 		return ENOTTY;
206 
207 	default:
208 		return ENOTTY;
209 	}
210 
211 	return error;
212 }
213 #endif /* _MODULE */
214 
215 /*
216  * dm_match:
217  *
218  *	Autoconfiguration match function for pseudo-device glue.
219  */
220 static int
221 dm_match(device_t parent, cfdata_t match,
222     void *aux)
223 {
224 
225 	/* Pseudo-device; always present. */
226 	return (1);
227 }
228 
229 /*
230  * dm_attach:
231  *
232  *	Autoconfiguration attach function for pseudo-device glue.
233  */
234 static void
235 dm_attach(device_t parent, device_t self,
236     void *aux)
237 {
238 	return;
239 }
240 
241 
242 /*
243  * dm_detach:
244  *
245  *	Autoconfiguration detach function for pseudo-device glue.
246  * This routine is called by dm_ioctl::dm_dev_remove_ioctl and by autoconf to
247  * remove devices created in device-mapper.
248  */
249 static int
250 dm_detach(device_t self, int flags)
251 {
252 	dm_dev_t *dmv;
253 
254 	/* Detach device from global device list */
255 	if ((dmv = dm_dev_detach(self)) == NULL)
256 		return ENOENT;
257 
258 	/* Destroy active table first.  */
259 	dm_table_destroy(&dmv->table_head, DM_TABLE_ACTIVE);
260 
261 	/* Destroy inactive table if exits, too. */
262 	dm_table_destroy(&dmv->table_head, DM_TABLE_INACTIVE);
263 
264 	dm_table_head_destroy(&dmv->table_head);
265 
266 	/* Destroy disk device structure */
267 	disk_detach(dmv->diskp);
268 	disk_destroy(dmv->diskp);
269 
270 	/* Destroy device */
271 	(void)dm_dev_free(dmv);
272 
273 	/* Decrement device counter After removing device */
274 	atomic_dec_64(&dm_dev_counter);
275 
276 	return 0;
277 }
278 
279 static void
280 dm_doinit(void)
281 {
282 	dm_target_init();
283 	dm_dev_init();
284 	dm_pdev_init();
285 }
286 
287 /* attach routine */
288 void
289 dmattach(int n)
290 {
291 	int error;
292 
293 	error = config_cfattach_attach(dm_cd.cd_name, &dm_ca);
294 	if (error) {
295 		aprint_error("%s: unable to register cfattach\n",
296 		    dm_cd.cd_name);
297 	} else {
298 		dm_doinit();
299 	}
300 }
301 
302 #ifdef _MODULE
303 /* Destroy routine */
304 static int
305 dmdestroy(void)
306 {
307 	int error;
308 
309 	error = config_cfattach_detach(dm_cd.cd_name, &dm_ca);
310 	if (error)
311 		return error;
312 
313 	dm_dev_destroy();
314 	dm_pdev_destroy();
315 	dm_target_destroy();
316 
317 	return 0;
318 }
319 #endif /* _MODULE */
320 
321 static int
322 dmopen(dev_t dev, int flags, int mode, struct lwp *l)
323 {
324 
325 	aprint_debug("dm open routine called %" PRIu32 "\n", minor(dev));
326 	return 0;
327 }
328 
329 static int
330 dmclose(dev_t dev, int flags, int mode, struct lwp *l)
331 {
332 
333 	aprint_debug("dm close routine called %" PRIu32 "\n", minor(dev));
334 	return 0;
335 }
336 
337 
338 static int
339 dmioctl(dev_t dev, const u_long cmd, void *data, int flag, struct lwp *l)
340 {
341 	int r;
342 	prop_dictionary_t dm_dict_in;
343 
344 	r = 0;
345 
346 	aprint_debug("dmioctl called\n");
347 
348 	KASSERT(data != NULL);
349 
350 	if (( r = disk_ioctl_switch(dev, cmd, data)) == ENOTTY) {
351 		struct plistref *pref = (struct plistref *) data;
352 
353 		/* Check if we were called with NETBSD_DM_IOCTL ioctl
354 		   otherwise quit. */
355 		if ((r = dm_ioctl_switch(cmd)) != 0)
356 			return r;
357 
358 		if((r = prop_dictionary_copyin_ioctl(pref, cmd, &dm_dict_in)) != 0)
359 			return r;
360 
361 		if ((r = dm_check_version(dm_dict_in)) != 0)
362 			goto cleanup_exit;
363 
364 		/* run ioctl routine */
365 		if ((r = dm_cmd_to_fun(dm_dict_in)) != 0)
366 			goto cleanup_exit;
367 
368 cleanup_exit:
369 		r = prop_dictionary_copyout_ioctl(pref, cmd, dm_dict_in);
370 		prop_object_release(dm_dict_in);
371 	}
372 
373 	return r;
374 }
375 
376 /*
377  * Translate command sent from libdevmapper to func.
378  */
379 static int
380 dm_cmd_to_fun(prop_dictionary_t dm_dict){
381 	int i, r;
382 	prop_string_t command;
383 
384 	r = 0;
385 
386 	if ((command = prop_dictionary_get(dm_dict, DM_IOCTL_COMMAND)) == NULL)
387 		return EINVAL;
388 
389 	for(i = 0; cmd_fn[i].cmd != NULL; i++)
390 		if (prop_string_equals_cstring(command, cmd_fn[i].cmd))
391 			break;
392 
393 	if (cmd_fn[i].cmd == NULL)
394 		return EINVAL;
395 
396 	aprint_debug("ioctl %s called\n", cmd_fn[i].cmd);
397 	r = cmd_fn[i].fn(dm_dict);
398 
399 	return r;
400 }
401 
402 /* Call apropriate ioctl handler function. */
403 static int
404 dm_ioctl_switch(u_long cmd)
405 {
406 
407 	switch(cmd) {
408 
409 	case NETBSD_DM_IOCTL:
410 		aprint_debug("dm NetBSD_DM_IOCTL called\n");
411 		break;
412 	default:
413 		 aprint_debug("dm unknown ioctl called\n");
414 		 return ENOTTY;
415 		 break; /* NOT REACHED */
416 	}
417 
418 	 return 0;
419 }
420 
421  /*
422   * Check for disk specific ioctls.
423   */
424 
425 static int
426 disk_ioctl_switch(dev_t dev, u_long cmd, void *data)
427 {
428 	dm_dev_t *dmv;
429 
430 	/* disk ioctls make sense only on block devices */
431 	if (minor(dev) == 0)
432 		return ENOTTY;
433 
434 	switch(cmd) {
435 	case DIOCGWEDGEINFO:
436 	{
437 		struct dkwedge_info *dkw = (void *) data;
438 
439 		if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
440 			return ENODEV;
441 
442 		aprint_debug("DIOCGWEDGEINFO ioctl called\n");
443 
444 		strlcpy(dkw->dkw_devname, dmv->name, 16);
445 		strlcpy(dkw->dkw_wname, dmv->name, DM_NAME_LEN);
446 		strlcpy(dkw->dkw_parent, dmv->name, 16);
447 
448 		dkw->dkw_offset = 0;
449 		dkw->dkw_size = dm_table_size(&dmv->table_head);
450 		strcpy(dkw->dkw_ptype, DKW_PTYPE_FFS);
451 
452 		dm_dev_unbusy(dmv);
453 		break;
454 	}
455 
456 	case DIOCGDISKINFO:
457 	{
458 		struct plistref *pref = (struct plistref *) data;
459 
460 		if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
461 			return ENODEV;
462 
463 		if (dmv->diskp->dk_info == NULL) {
464 			dm_dev_unbusy(dmv);
465 			return ENOTSUP;
466 		} else
467 			prop_dictionary_copyout_ioctl(pref, cmd,
468 			    dmv->diskp->dk_info);
469 
470 		dm_dev_unbusy(dmv);
471 		break;
472 	}
473 
474 	default:
475 		aprint_debug("unknown disk_ioctl called\n");
476 		return ENOTTY;
477 		break; /* NOT REACHED */
478 	}
479 
480 	return 0;
481 }
482 
483 /*
484  * Do all IO operations on dm logical devices.
485  */
486 static void
487 dmstrategy(struct buf *bp)
488 {
489 	dm_dev_t *dmv;
490 	dm_table_t  *tbl;
491 	dm_table_entry_t *table_en;
492 	struct buf *nestbuf;
493 
494 	uint32_t dev_type;
495 
496 	uint64_t buf_start, buf_len, issued_len;
497 	uint64_t table_start, table_end;
498 	uint64_t start, end;
499 
500 	buf_start = bp->b_blkno * DEV_BSIZE;
501 	buf_len = bp->b_bcount;
502 
503 	tbl = NULL;
504 
505 	table_end = 0;
506 	dev_type = 0;
507 	issued_len = 0;
508 
509 	if ((dmv = dm_dev_lookup(NULL, NULL, minor(bp->b_dev))) == NULL) {
510 		bp->b_error = EIO;
511 		bp->b_resid = bp->b_bcount;
512 		biodone(bp);
513 		return;
514 	}
515 
516 	if (bounds_check_with_mediasize(bp, DEV_BSIZE,
517 	    dm_table_size(&dmv->table_head)) <= 0) {
518 		dm_dev_unbusy(dmv);
519 		bp->b_resid = bp->b_bcount;
520 		biodone(bp);
521 		return;
522 	}
523 
524 	/*
525 	 * disk(9) is part of device structure and it can't be used without
526 	 * mutual exclusion, use diskp_mtx until it will be fixed.
527 	 */
528 	mutex_enter(&dmv->diskp_mtx);
529 	disk_busy(dmv->diskp);
530 	mutex_exit(&dmv->diskp_mtx);
531 
532 	/* Select active table */
533 	tbl = dm_table_get_entry(&dmv->table_head, DM_TABLE_ACTIVE);
534 
535 	 /* Nested buffers count down to zero therefore I have
536 	    to set bp->b_resid to maximal value. */
537 	bp->b_resid = bp->b_bcount;
538 
539 	/*
540 	 * Find out what tables I want to select.
541 	 */
542 	SLIST_FOREACH(table_en, tbl, next)
543 	{
544 		/* I need need number of bytes not blocks. */
545 		table_start = table_en->start * DEV_BSIZE;
546 		/*
547 		 * I have to sub 1 from table_en->length to prevent
548 		 * off by one error
549 		 */
550 		table_end = table_start + (table_en->length)* DEV_BSIZE;
551 
552 		start = MAX(table_start, buf_start);
553 
554 		end = MIN(table_end, buf_start + buf_len);
555 
556 		aprint_debug("----------------------------------------\n");
557 		aprint_debug("table_start %010" PRIu64", table_end %010"
558 		    PRIu64 "\n", table_start, table_end);
559 		aprint_debug("buf_start %010" PRIu64", buf_len %010"
560 		    PRIu64"\n", buf_start, buf_len);
561 		aprint_debug("start-buf_start %010"PRIu64", end %010"
562 		    PRIu64"\n", start - buf_start, end);
563 		aprint_debug("start %010" PRIu64" , end %010"
564                     PRIu64"\n", start, end);
565 		aprint_debug("\n----------------------------------------\n");
566 
567 		if (start < end) {
568 			/* create nested buffer  */
569 			nestbuf = getiobuf(NULL, true);
570 
571 			nestiobuf_setup(bp, nestbuf, start - buf_start,
572 			    (end - start));
573 
574 			issued_len += end - start;
575 
576 			/* I need number of blocks. */
577 			nestbuf->b_blkno = (start - table_start) / DEV_BSIZE;
578 
579 			table_en->target->strategy(table_en, nestbuf);
580 		}
581 	}
582 
583 	if (issued_len < buf_len)
584 		nestiobuf_done(bp, buf_len - issued_len, EINVAL);
585 
586 	mutex_enter(&dmv->diskp_mtx);
587 	disk_unbusy(dmv->diskp, buf_len, bp != NULL ? bp->b_flags & B_READ : 0);
588 	mutex_exit(&dmv->diskp_mtx);
589 
590 	dm_table_release(&dmv->table_head, DM_TABLE_ACTIVE);
591 	dm_dev_unbusy(dmv);
592 
593 	return;
594 }
595 
596 
597 static int
598 dmread(dev_t dev, struct uio *uio, int flag)
599 {
600 
601 	return (physio(dmstrategy, NULL, dev, B_READ, dmminphys, uio));
602 }
603 
604 static int
605 dmwrite(dev_t dev, struct uio *uio, int flag)
606 {
607 
608 	return (physio(dmstrategy, NULL, dev, B_WRITE, dmminphys, uio));
609 }
610 
611 static int
612 dmsize(dev_t dev)
613 {
614 	dm_dev_t *dmv;
615 	uint64_t size;
616 
617 	size = 0;
618 
619 	if ((dmv = dm_dev_lookup(NULL, NULL, minor(dev))) == NULL)
620 			return -ENOENT;
621 
622 	size = dm_table_size(&dmv->table_head);
623 	dm_dev_unbusy(dmv);
624 
625   	return size;
626 }
627 
628 static void
629 dmminphys(struct buf *bp)
630 {
631 
632 	bp->b_bcount = MIN(bp->b_bcount, MAXPHYS);
633 }
634 
635 void
636 dmgetproperties(struct disk *disk, dm_table_head_t *head)
637 {
638 	prop_dictionary_t disk_info, odisk_info, geom;
639 	int dmp_size;
640 
641 	dmp_size = dm_table_size(head);
642 	disk_info = prop_dictionary_create();
643 	geom = prop_dictionary_create();
644 
645 	prop_dictionary_set_cstring_nocopy(disk_info, "type", "ESDI");
646 	prop_dictionary_set_uint64(geom, "sectors-per-unit", dmp_size);
647 	prop_dictionary_set_uint32(geom, "sector-size",
648 	    DEV_BSIZE /* XXX 512? */);
649 	prop_dictionary_set_uint32(geom, "sectors-per-track", 32);
650 	prop_dictionary_set_uint32(geom, "tracks-per-cylinder", 64);
651 	prop_dictionary_set_uint32(geom, "cylinders-per-unit", dmp_size / 2048);
652 	prop_dictionary_set(disk_info, "geometry", geom);
653 	prop_object_release(geom);
654 
655 	odisk_info = disk->dk_info;
656 	disk->dk_info = disk_info;
657 
658 	if (odisk_info != NULL)
659 		prop_object_release(odisk_info);
660 }
661