dk.c revision 1.2 1 /* $NetBSD: dk.c,v 1.2 2004/10/15 04:42:09 thorpej Exp $ */
2
3 /*-
4 * Copyright (c) 2004 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe.
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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: dk.c,v 1.2 2004/10/15 04:42:09 thorpej Exp $");
41
42 #include "opt_dkwedge.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/proc.h>
47 #include <sys/errno.h>
48 #include <sys/pool.h>
49 #include <sys/ioctl.h>
50 #include <sys/disklabel.h>
51 #include <sys/disk.h>
52 #include <sys/fcntl.h>
53 #include <sys/vnode.h>
54 #include <sys/conf.h>
55 #include <sys/callout.h>
56 #include <sys/kernel.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/device.h>
60
61 #include <miscfs/specfs/specdev.h>
62
63 MALLOC_DEFINE(M_DKWEDGE, "dkwedge", "Disk wedge structures");
64
65 typedef enum {
66 DKW_STATE_LARVAL = 0,
67 DKW_STATE_RUNNING = 1,
68 DKW_STATE_DYING = 2,
69 DKW_STATE_DEAD = 666
70 } dkwedge_state_t;
71
72 struct dkwedge_softc {
73 struct device *sc_dev; /* pointer to our pseudo-device */
74 struct cfdata sc_cfdata; /* our cfdata structure */
75 uint8_t sc_wname[128]; /* wedge name (Unicode, UTF-8) */
76
77 dkwedge_state_t sc_state; /* state this wedge is in */
78
79 struct disk *sc_parent; /* parent disk */
80 daddr_t sc_offset; /* LBA offset of wedge in parent */
81 uint64_t sc_size; /* size of wedge in blocks */
82 char sc_ptype[32]; /* partition type */
83 dev_t sc_pdev; /* cached parent's dev_t */
84 /* link on parent's wedge list */
85 LIST_ENTRY(dkwedge_softc) sc_plink;
86
87 int sc_open; /* locked by parent's rawlock */
88
89 struct disk sc_dk; /* our own disk structure */
90 struct bufq_state sc_bufq; /* buffer queue */
91 struct callout sc_restart_ch; /* callout to restart I/O */
92
93 u_int sc_iopend; /* I/Os pending */
94 int sc_flags; /* flags (splbio) */
95 };
96
97 #define DK_F_WAIT_DRAIN 0x0001 /* waiting for I/O to drain */
98
99 static void dkstart(struct dkwedge_softc *);
100 static void dkiodone(struct buf *);
101 static void dkrestart(void *);
102
103 static dev_type_open(dkopen);
104 static dev_type_close(dkclose);
105 static dev_type_read(dkread);
106 static dev_type_write(dkwrite);
107 static dev_type_ioctl(dkioctl);
108 static dev_type_strategy(dkstrategy);
109 static dev_type_dump(dkdump);
110 static dev_type_size(dksize);
111
112 const struct bdevsw dk_bdevsw = {
113 dkopen, dkclose, dkstrategy, dkioctl, dkdump, dksize, D_DISK
114 };
115
116 const struct cdevsw dk_cdevsw = {
117 dkopen, dkclose, dkread, dkwrite, dkioctl,
118 nostop, notty, nopoll, nommap, nokqfilter, D_DISK
119 };
120
121 static struct dkwedge_softc **dkwedges;
122 static u_int ndkwedges;
123 static struct lock dkwedges_lock = LOCK_INITIALIZER(PRIBIO, "dkwgs", 0, 0);
124
125 static LIST_HEAD(, dkwedge_discovery_method) dkwedge_discovery_methods;
126 static int dkwedge_discovery_methods_initialized;
127 static struct lock dkwedge_discovery_methods_lock =
128 LOCK_INITIALIZER(PRIBIO, "dkddm", 0, 0);
129
130 /*
131 * dkwedge_match:
132 *
133 * Autoconfiguration match function for pseudo-device glue.
134 */
135 static int
136 dkwedge_match(struct device *parent, struct cfdata *match, void *aux)
137 {
138
139 /* Pseudo-device; always present. */
140 return (1);
141 }
142
143 /*
144 * dkwedge_attach:
145 *
146 * Autoconfiguration attach function for pseudo-device glue.
147 */
148 static void
149 dkwedge_attach(struct device *parent, struct device *self, void *aux)
150 {
151
152 /* Nothing to do. */
153 }
154
155 /*
156 * dkwedge_detach:
157 *
158 * Autoconfiguration detach function for pseudo-device glue.
159 */
160 static int
161 dkwedge_detach(struct device *self, int flags)
162 {
163
164 /* Always succeeds. */
165 return (0);
166 }
167
168 CFDRIVER_DECL(dk, DV_DISK, NULL);
169 CFATTACH_DECL(dk, sizeof(struct device),
170 dkwedge_match, dkwedge_attach, dkwedge_detach, NULL);
171
172 static int dkwedge_cfglue_initialized;
173 static struct simplelock dkwedge_cfglue_initialized_slock =
174 SIMPLELOCK_INITIALIZER;
175
176 static void
177 dkwedge_cfglue_init(void)
178 {
179
180 simple_lock(&dkwedge_cfglue_initialized_slock);
181 if (dkwedge_cfglue_initialized == 0) {
182 if (config_cfdriver_attach(&dk_cd) != 0)
183 panic("dkwedge: unable to attach cfdriver");
184 if (config_cfattach_attach(dk_cd.cd_name, &dk_ca) != 0)
185 panic("dkwedge: unable to attach cfattach");
186
187 dkwedge_cfglue_initialized = 1;
188 }
189 simple_unlock(&dkwedge_cfglue_initialized_slock);
190 }
191
192 /*
193 * dkwedge_wait_drain:
194 *
195 * Wait for I/O on the wedge to drain.
196 * NOTE: Must be called at splbio()!
197 */
198 static void
199 dkwedge_wait_drain(struct dkwedge_softc *sc)
200 {
201
202 while (sc->sc_iopend != 0) {
203 sc->sc_flags |= DK_F_WAIT_DRAIN;
204 (void) tsleep(&sc->sc_iopend, PRIBIO, "dkdrn", 0);
205 }
206 }
207
208 /*
209 * dkwedge_compute_pdev:
210 *
211 * Compute the parent disk's dev_t.
212 */
213 static int
214 dkwedge_compute_pdev(const char *pname, dev_t *pdevp)
215 {
216 const char *name, *cp;
217 int punit, pmaj;
218 char devname[16];
219
220 name = pname;
221 if ((pmaj = devsw_name2blk(name, devname, sizeof(devname))) == -1)
222 return (ENODEV);
223
224 name += strlen(devname);
225 for (cp = name, punit = 0; *cp >= '0' && *cp <= '9'; cp++)
226 punit = (punit * 10) + (*cp - '0');
227 if (cp == name) {
228 /* Invalid parent disk name. */
229 return (ENODEV);
230 }
231
232 *pdevp = MAKEDISKDEV(pmaj, punit, RAW_PART);
233
234 return (0);
235 }
236
237 /*
238 * dkwedge_array_expand:
239 *
240 * Expand the dkwedges array.
241 */
242 static void
243 dkwedge_array_expand(void)
244 {
245 int newcnt = ndkwedges + 16;
246 struct dkwedge_softc **newarray, **oldarray;
247
248 newarray = malloc(newcnt * sizeof(*newarray), M_DKWEDGE,
249 M_WAITOK|M_ZERO);
250 if ((oldarray = dkwedges) != NULL)
251 memcpy(newarray, dkwedges, ndkwedges * sizeof(*newarray));
252 dkwedges = newarray;
253 ndkwedges = newcnt;
254 if (oldarray != NULL)
255 free(oldarray, M_DKWEDGE);
256 }
257
258 /*
259 * dkwedge_add: [exported function]
260 *
261 * Add a disk wedge based on the provided information.
262 *
263 * The incoming dkw_devname[] is ignored, instead being
264 * filled in and returned to the caller.
265 */
266 int
267 dkwedge_add(struct dkwedge_info *dkw)
268 {
269 struct dkwedge_softc *sc, *lsc;
270 struct disk *pdk;
271 u_int unit;
272 int error;
273 dev_t pdev;
274
275 if (dkwedge_cfglue_initialized == 0)
276 dkwedge_cfglue_init();
277
278 dkw->dkw_parent[sizeof(dkw->dkw_parent) - 1] = '\0';
279 pdk = disk_find(dkw->dkw_parent);
280 if (pdk == NULL)
281 return (ENODEV);
282
283 error = dkwedge_compute_pdev(pdk->dk_name, &pdev);
284 if (error)
285 return (error);
286
287 if (dkw->dkw_offset < 0)
288 return (EINVAL);
289
290 sc = malloc(sizeof(*sc), M_DKWEDGE, M_WAITOK|M_ZERO);
291 sc->sc_state = DKW_STATE_LARVAL;
292 sc->sc_parent = pdk;
293 sc->sc_pdev = pdev;
294 sc->sc_offset = dkw->dkw_offset;
295 sc->sc_size = dkw->dkw_size;
296
297 memcpy(sc->sc_wname, dkw->dkw_wname, sizeof(sc->sc_wname));
298 sc->sc_wname[sizeof(sc->sc_wname) - 1] = '\0';
299
300 memcpy(sc->sc_ptype, dkw->dkw_ptype, sizeof(sc->sc_ptype));
301 sc->sc_ptype[sizeof(sc->sc_ptype) - 1] = '\0';
302
303 bufq_alloc(&sc->sc_bufq, BUFQ_FCFS);
304
305 callout_init(&sc->sc_restart_ch);
306 callout_setfunc(&sc->sc_restart_ch, dkrestart, sc);
307
308 /*
309 * Wedge will be added; increment the wedge count for the parent.
310 * Only allow this to happend if RAW_PART is the only thing open.
311 */
312 (void) lockmgr(&pdk->dk_openlock, LK_EXCLUSIVE, NULL);
313 if (pdk->dk_openmask & ~(1 << RAW_PART))
314 error = EBUSY;
315 else {
316 /* Check for wedge overlap. */
317 LIST_FOREACH(lsc, &pdk->dk_wedges, sc_plink) {
318 daddr_t lastblk = sc->sc_offset + sc->sc_size - 1;
319 daddr_t llastblk = lsc->sc_offset + lsc->sc_size - 1;
320
321 if (sc->sc_offset >= lsc->sc_offset &&
322 sc->sc_offset <= llastblk) {
323 /* Overlaps the tail of the exsiting wedge. */
324 break;
325 }
326 if (lastblk >= lsc->sc_offset &&
327 lastblk <= llastblk) {
328 /* Overlaps the head of the existing wedge. */
329 break;
330 }
331 }
332 if (lsc != NULL)
333 error = EINVAL;
334 else {
335 pdk->dk_nwedges++;
336 LIST_INSERT_HEAD(&pdk->dk_wedges, sc, sc_plink);
337 }
338 }
339 (void) lockmgr(&pdk->dk_openlock, LK_RELEASE, NULL);
340 if (error) {
341 bufq_free(&sc->sc_bufq);
342 free(sc, M_DKWEDGE);
343 return (error);
344 }
345
346 /* Fill in our cfdata for the pseudo-device glue. */
347 sc->sc_cfdata.cf_name = dk_cd.cd_name;
348 sc->sc_cfdata.cf_atname = dk_ca.ca_name;
349 /* sc->sc_cfdata.cf_unit set below */
350 sc->sc_cfdata.cf_fstate = FSTATE_NOTFOUND;
351
352 /* Insert the larval wedge into the array. */
353 (void) lockmgr(&dkwedges_lock, LK_EXCLUSIVE, NULL);
354 for (error = 0;;) {
355 struct dkwedge_softc **scpp;
356
357 /*
358 * Check for a duplicate wname while searching for
359 * a slot.
360 */
361 for (scpp = NULL, unit = 0; unit < ndkwedges; unit++) {
362 if (dkwedges[unit] == NULL) {
363 if (scpp == NULL) {
364 scpp = &dkwedges[unit];
365 sc->sc_cfdata.cf_unit = unit;
366 }
367 } else {
368 /* XXX Unicode. */
369 if (strcmp(dkwedges[unit]->sc_wname,
370 sc->sc_wname) == 0) {
371 error = EEXIST;
372 break;
373 }
374 }
375 }
376 if (error)
377 break;
378 KASSERT(unit == ndkwedges);
379 if (scpp == NULL)
380 dkwedge_array_expand();
381 else {
382 KASSERT(scpp == &dkwedges[sc->sc_cfdata.cf_unit]);
383 *scpp = sc;
384 break;
385 }
386 }
387 (void) lockmgr(&dkwedges_lock, LK_RELEASE, NULL);
388 if (error) {
389 (void) lockmgr(&pdk->dk_openlock, LK_EXCLUSIVE, NULL);
390 pdk->dk_nwedges--;
391 LIST_REMOVE(sc, sc_plink);
392 (void) lockmgr(&pdk->dk_openlock, LK_RELEASE, NULL);
393
394 bufq_free(&sc->sc_bufq);
395 free(sc, M_DKWEDGE);
396 return (error);
397 }
398
399 /*
400 * Now that we know the unit #, attach a pseudo-device for
401 * this wedge instance. This will provide us with the
402 * "struct device" necessary for glue to other parts of the
403 * system.
404 *
405 * This should never fail, unless we're almost totally out of
406 * memory.
407 */
408 if ((sc->sc_dev = config_attach_pseudo(&sc->sc_cfdata)) == NULL) {
409 aprint_error("%s%u: unable to attach pseudo-device\n",
410 sc->sc_cfdata.cf_name, sc->sc_cfdata.cf_unit);
411
412 (void) lockmgr(&dkwedges_lock, LK_EXCLUSIVE, NULL);
413 dkwedges[sc->sc_cfdata.cf_unit] = NULL;
414 (void) lockmgr(&dkwedges_lock, LK_RELEASE, NULL);
415
416 (void) lockmgr(&pdk->dk_openlock, LK_EXCLUSIVE, NULL);
417 pdk->dk_nwedges--;
418 LIST_REMOVE(sc, sc_plink);
419 (void) lockmgr(&pdk->dk_openlock, LK_RELEASE, NULL);
420
421 bufq_free(&sc->sc_bufq);
422 free(sc, M_DKWEDGE);
423 return (ENOMEM);
424 }
425 sc->sc_dk.dk_name = sc->sc_dev->dv_xname;
426
427 /* Return the devname to the caller. */
428 strcpy(dkw->dkw_devname, sc->sc_dev->dv_xname);
429
430 /*
431 * XXX Really ought to make the disk_attach() and the changing
432 * of state to RUNNING atomic.
433 */
434
435 disk_attach(&sc->sc_dk);
436
437 /* Disk wedge is ready for use! */
438 sc->sc_state = DKW_STATE_RUNNING;
439
440 /* Announce our arrival. */
441 aprint_normal("%s at %s: %s\n", sc->sc_dev->dv_xname, pdk->dk_name,
442 sc->sc_wname); /* XXX Unicode */
443 aprint_normal("%s: %"PRIu64" blocks at %"PRId64", type: %s\n",
444 sc->sc_dev->dv_xname, sc->sc_size, sc->sc_offset, sc->sc_ptype);
445
446 return (0);
447 }
448
449 /*
450 * dkwedge_del: [exported function]
451 *
452 * Delete a disk wedge based on the provided information.
453 * NOTE: We look up the wedge based on the wedge devname,
454 * not wname.
455 */
456 int
457 dkwedge_del(struct dkwedge_info *dkw)
458 {
459 struct dkwedge_softc *sc = NULL;
460 u_int unit;
461 int bmaj, cmaj, i, mn, s;
462
463 /* Find our softc. */
464 dkw->dkw_devname[sizeof(dkw->dkw_devname) - 1] = '\0';
465 (void) lockmgr(&dkwedges_lock, LK_EXCLUSIVE, NULL);
466 for (unit = 0; unit < ndkwedges; unit++) {
467 if ((sc = dkwedges[unit]) != NULL &&
468 strcmp(sc->sc_dev->dv_xname, dkw->dkw_devname) == 0 &&
469 strcmp(sc->sc_parent->dk_name, dkw->dkw_parent) == 0) {
470 /* Mark the wedge as dying. */
471 sc->sc_state = DKW_STATE_DYING;
472 break;
473 }
474 }
475 (void) lockmgr(&dkwedges_lock, LK_RELEASE, NULL);
476 if (unit == ndkwedges)
477 return (ESRCH);
478
479 KASSERT(sc != NULL);
480
481 /* Locate the wedge major numbers. */
482 bmaj = bdevsw_lookup_major(&dk_bdevsw);
483 cmaj = cdevsw_lookup_major(&dk_cdevsw);
484
485 /* Kill any pending restart. */
486 callout_stop(&sc->sc_restart_ch);
487
488 /*
489 * dkstart() will kill any queued buffers now that the
490 * state of the wedge is not RUNNING. Once we've done
491 * that, wait for any other pending I/O to complete.
492 */
493 s = splbio();
494 dkstart(sc);
495 dkwedge_wait_drain(sc);
496 splx(s);
497
498 /* Nuke the vnodes for any open instances. */
499 for (i = 0; i < MAXPARTITIONS; i++) {
500 mn = DISKMINOR(unit, i);
501 vdevgone(bmaj, mn, mn, VBLK);
502 vdevgone(cmaj, mn, mn, VCHR);
503 }
504
505 /* Clean up the parent. */
506 (void) lockmgr(&sc->sc_parent->dk_rawlock, LK_EXCLUSIVE, NULL);
507 if (sc->sc_open) {
508 if (sc->sc_parent->dk_rawopens-- == 1) {
509 KASSERT(sc->sc_parent->dk_rawvp != NULL);
510 (void) vn_close(sc->sc_parent->dk_rawvp, FREAD | FWRITE,
511 NOCRED, curproc);
512 sc->sc_parent->dk_rawvp = NULL;
513 }
514 sc->sc_open = 0;
515 }
516 (void) lockmgr(&sc->sc_parent->dk_rawlock, LK_RELEASE, NULL);
517
518 /* Announce our departure. */
519 aprint_normal("%s at %s (%s) deleted\n", sc->sc_dev->dv_xname,
520 sc->sc_parent->dk_name,
521 sc->sc_wname); /* XXX Unicode */
522
523 /* Delete our pseudo-device. */
524 (void) config_detach(sc->sc_dev, DETACH_FORCE | DETACH_QUIET);
525
526 (void) lockmgr(&sc->sc_parent->dk_openlock, LK_EXCLUSIVE, NULL);
527 sc->sc_parent->dk_nwedges--;
528 LIST_REMOVE(sc, sc_plink);
529 (void) lockmgr(&sc->sc_parent->dk_openlock, LK_RELEASE, NULL);
530
531 /* Delete our buffer queue. */
532 bufq_free(&sc->sc_bufq);
533
534 /* Detach from the disk list. */
535 disk_detach(&sc->sc_dk);
536
537 /* Poof. */
538 (void) lockmgr(&dkwedges_lock, LK_EXCLUSIVE, NULL);
539 dkwedges[unit] = NULL;
540 sc->sc_state = DKW_STATE_DEAD;
541 (void) lockmgr(&dkwedges_lock, LK_RELEASE, NULL);
542
543 free(sc, M_DKWEDGE);
544
545 return (0);
546 }
547
548 /*
549 * dkwedge_delall: [exported function]
550 *
551 * Delete all of the wedges on the specified disk. Used when
552 * a disk is being detached.
553 */
554 void
555 dkwedge_delall(struct disk *pdk)
556 {
557 struct dkwedge_info dkw;
558 struct dkwedge_softc *sc;
559
560 for (;;) {
561 (void) lockmgr(&pdk->dk_openlock, LK_EXCLUSIVE, NULL);
562 if ((sc = LIST_FIRST(&pdk->dk_wedges)) == NULL) {
563 KASSERT(pdk->dk_nwedges == 0);
564 (void) lockmgr(&pdk->dk_openlock, LK_RELEASE, NULL);
565 return;
566 }
567 strcpy(dkw.dkw_parent, pdk->dk_name);
568 strcpy(dkw.dkw_devname, sc->sc_dev->dv_xname);
569 (void) lockmgr(&pdk->dk_openlock, LK_RELEASE, NULL);
570 (void) dkwedge_del(&dkw);
571 }
572 }
573
574 /*
575 * dkwedge_list: [exported function]
576 *
577 * List all of the wedges on a particular disk.
578 * If p == NULL, the buffer is in kernel space. Otherwise, it is
579 * in user space of the specified process.
580 */
581 int
582 dkwedge_list(struct disk *pdk, struct dkwedge_list *dkwl, struct proc *p)
583 {
584 struct uio uio;
585 struct iovec iov;
586 struct dkwedge_softc *sc;
587 struct dkwedge_info dkw;
588 int error = 0;
589
590 iov.iov_base = dkwl->dkwl_buf;
591 iov.iov_len = dkwl->dkwl_bufsize;
592
593 uio.uio_iov = &iov;
594 uio.uio_iovcnt = 1;
595 uio.uio_offset = 0;
596 uio.uio_resid = dkwl->dkwl_bufsize;
597 uio.uio_segflg = p != NULL ? UIO_USERSPACE : UIO_SYSSPACE;
598 uio.uio_rw = UIO_READ;
599 uio.uio_procp = p;
600
601 dkwl->dkwl_ncopied = 0;
602
603 (void) lockmgr(&pdk->dk_openlock, LK_EXCLUSIVE, NULL);
604 LIST_FOREACH(sc, &pdk->dk_wedges, sc_plink) {
605 if (uio.uio_resid < sizeof(dkw))
606 break;
607
608 if (sc->sc_state != DKW_STATE_RUNNING)
609 continue;
610
611 strcpy(dkw.dkw_devname, sc->sc_dev->dv_xname);
612 memcpy(dkw.dkw_wname, sc->sc_wname, sizeof(dkw.dkw_wname));
613 dkw.dkw_wname[sizeof(dkw.dkw_wname) - 1] = '\0';
614 strcpy(dkw.dkw_parent, sc->sc_parent->dk_name);
615 dkw.dkw_offset = sc->sc_offset;
616 dkw.dkw_size = sc->sc_size;
617 strcpy(dkw.dkw_ptype, sc->sc_ptype);
618
619 error = uiomove(&dkw, sizeof(dkw), &uio);
620 if (error)
621 break;
622 dkwl->dkwl_ncopied++;
623 }
624 dkwl->dkwl_nwedges = pdk->dk_nwedges;
625 (void) lockmgr(&pdk->dk_openlock, LK_RELEASE, NULL);
626
627 return (error);
628 }
629
630 /*
631 * We need a dummy objet to stuff into the dkwedge discovery method link
632 * set to ensure that there is always at least one object in the set.
633 */
634 static struct dkwedge_discovery_method dummy_discovery_method;
635 __link_set_add_bss(dkwedge_methods, dummy_discovery_method);
636
637 /*
638 * dkwedge_discover_init:
639 *
640 * Initialize the disk wedge discovery method list.
641 */
642 static void
643 dkwedge_discover_init(void)
644 {
645 __link_set_decl(dkwedge_methods, struct dkwedge_discovery_method);
646 struct dkwedge_discovery_method * const *ddmp;
647 struct dkwedge_discovery_method *lddm, *ddm;
648
649 (void) lockmgr(&dkwedge_discovery_methods_lock, LK_EXCLUSIVE, NULL);
650
651 if (dkwedge_discovery_methods_initialized) {
652 (void) lockmgr(&dkwedge_discovery_methods_lock, LK_RELEASE,
653 NULL);
654 return;
655 }
656
657 LIST_INIT(&dkwedge_discovery_methods);
658
659 __link_set_foreach(ddmp, dkwedge_methods) {
660 ddm = *ddmp;
661 if (ddm == &dummy_discovery_method)
662 continue;
663 if (LIST_EMPTY(&dkwedge_discovery_methods)) {
664 LIST_INSERT_HEAD(&dkwedge_discovery_methods,
665 ddm, ddm_list);
666 continue;
667 }
668 LIST_FOREACH(lddm, &dkwedge_discovery_methods, ddm_list) {
669 if (ddm->ddm_priority == lddm->ddm_priority) {
670 aprint_error("dk-method-%s: method \"%s\" "
671 "already exists at priority %d\n",
672 ddm->ddm_name, lddm->ddm_name,
673 lddm->ddm_priority);
674 /* Not inserted. */
675 break;
676 }
677 if (ddm->ddm_priority < lddm->ddm_priority) {
678 /* Higher priority; insert before. */
679 LIST_INSERT_BEFORE(lddm, ddm, ddm_list);
680 break;
681 }
682 if (LIST_NEXT(lddm, ddm_list) == NULL) {
683 /* Last one; insert after. */
684 KASSERT(lddm->ddm_priority < ddm->ddm_priority);
685 LIST_INSERT_AFTER(lddm, ddm, ddm_list);
686 break;
687 }
688 }
689 }
690
691 dkwedge_discovery_methods_initialized = 1;
692
693 (void) lockmgr(&dkwedge_discovery_methods_lock, LK_RELEASE, NULL);
694 }
695
696 #ifdef DKWEDGE_AUTODISCOVER
697 int dkwedge_autodiscover = 1;
698 #else
699 int dkwedge_autodiscover = 0;
700 #endif
701
702 /*
703 * dkwedge_discover: [exported function]
704 *
705 * Discover the wedges on a newly attached disk.
706 */
707 void
708 dkwedge_discover(struct disk *pdk)
709 {
710 struct dkwedge_discovery_method *ddm;
711 struct vnode *vp;
712 int error;
713 dev_t pdev;
714
715 /*
716 * Require people playing with wedges to enable this explicitly.
717 */
718 if (dkwedge_autodiscover == 0)
719 return;
720
721 if (dkwedge_discovery_methods_initialized == 0)
722 dkwedge_discover_init();
723
724 (void) lockmgr(&dkwedge_discovery_methods_lock, LK_SHARED, NULL);
725
726 error = dkwedge_compute_pdev(pdk->dk_name, &pdev);
727 if (error) {
728 aprint_error("%s: unable to compute pdev, error = %d\n",
729 pdk->dk_name, error);
730 goto out;
731 }
732
733 error = bdevvp(pdev, &vp);
734 if (error) {
735 aprint_error("%s: unable to find vnode for pdev, error = %d\n",
736 pdk->dk_name, error);
737 goto out;
738 }
739
740 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
741 if (error) {
742 aprint_error("%s: unable to lock vnode for pdev, error = %d\n",
743 pdk->dk_name, error);
744 vrele(vp);
745 goto out;
746 }
747
748 error = VOP_OPEN(vp, FREAD | FWRITE, NOCRED, 0);
749 if (error) {
750 aprint_error("%s: unable to open device, error = %d\n",
751 pdk->dk_name, error);
752 vput(vp);
753 goto out;
754 }
755 /* VOP_OPEN() doesn't do this for us. */
756 vp->v_writecount++;
757 VOP_UNLOCK(vp, 0);
758
759 /*
760 * For each supported partition map type, look to see if
761 * this map type exists. If so, parse it and add the
762 * corresponding wedges.
763 */
764 LIST_FOREACH(ddm, &dkwedge_discovery_methods, ddm_list) {
765 error = (*ddm->ddm_discover)(pdk, vp);
766 if (error == 0) {
767 /* Successfully created wedges; we're done. */
768 break;
769 }
770 }
771
772 error = vn_close(vp, FREAD | FWRITE, NOCRED, curproc);
773 if (error) {
774 aprint_error("%s: unable to close device, error = %d\n",
775 pdk->dk_name, error);
776 /* We'll just assume the vnode has been cleaned up. */
777 }
778 out:
779 (void) lockmgr(&dkwedge_discovery_methods_lock, LK_RELEASE, NULL);
780 }
781
782 /*
783 * dkwedge_read:
784 *
785 * Read the some data from the specified disk, used for
786 * partition discovery.
787 */
788 int
789 dkwedge_read(struct disk *pdk, struct vnode *vp, daddr_t blkno, void *buf,
790 size_t len)
791 {
792 struct buf b;
793
794 BUF_INIT(&b);
795
796 b.b_vp = vp;
797 b.b_dev = vp->v_rdev;
798 b.b_blkno = blkno;
799 b.b_bcount = b.b_resid = len;
800 b.b_flags = B_READ;
801 b.b_proc = curproc;
802 b.b_data = buf;
803
804 VOP_STRATEGY(vp, &b);
805 return (biowait(&b));
806 }
807
808 /*
809 * dkwedge_lookup:
810 *
811 * Look up a dkwedge_softc based on the provided dev_t.
812 */
813 static struct dkwedge_softc *
814 dkwedge_lookup(dev_t dev)
815 {
816 int unit = DISKUNIT(dev);
817
818 if (unit >= ndkwedges)
819 return (NULL);
820
821 KASSERT(dkwedges != NULL);
822
823 return (dkwedges[unit]);
824 }
825
826 /*
827 * dkopen: [devsw entry point]
828 *
829 * Open a wedge.
830 */
831 static int
832 dkopen(dev_t dev, int flags, int fmt, struct proc *p)
833 {
834 struct dkwedge_softc *sc = dkwedge_lookup(dev);
835 struct vnode *vp;
836 int error;
837
838 if (sc == NULL)
839 return (ENODEV);
840
841 if (sc->sc_state != DKW_STATE_RUNNING)
842 return (ENXIO);
843
844 /*
845 * We go through a complicated little dance to only open the parent
846 * vnode once per wedge, no matter how many times the wedge is
847 * opened. The reason? We see one dkopen() per open call, but
848 * only dkclose() on the last close.
849 */
850 (void) lockmgr(&sc->sc_parent->dk_rawlock, LK_EXCLUSIVE, NULL);
851 if (sc->sc_open == 0) {
852 if (sc->sc_parent->dk_rawopens++ == 0) {
853 KASSERT(sc->sc_parent->dk_rawvp == NULL);
854 error = bdevvp(sc->sc_pdev, &vp);
855 if (error)
856 goto popen_fail;
857 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
858 if (error) {
859 vrele(vp);
860 goto popen_fail;
861 }
862 error = VOP_OPEN(vp, FREAD | FWRITE, NOCRED, 0);
863 if (error) {
864 vput(vp);
865 goto popen_fail;
866 }
867 /* VOP_OPEN() doesn't do this for us. */
868 vp->v_writecount++;
869 VOP_UNLOCK(vp, 0);
870 sc->sc_parent->dk_rawvp = vp;
871 }
872 sc->sc_open = 1;
873 }
874 (void) lockmgr(&sc->sc_parent->dk_rawlock, LK_RELEASE, NULL);
875
876 return (0);
877
878 popen_fail:
879 (void) lockmgr(&sc->sc_parent->dk_rawlock, LK_RELEASE, NULL);
880 return (error);
881 }
882
883 /*
884 * dkclose: [devsw entry point]
885 *
886 * Close a wedge.
887 */
888 static int
889 dkclose(dev_t dev, int flags, int fmt, struct proc *p)
890 {
891 struct dkwedge_softc *sc = dkwedge_lookup(dev);
892 int error = 0;
893
894 KASSERT(sc->sc_open);
895
896 (void) lockmgr(&sc->sc_parent->dk_rawlock, LK_EXCLUSIVE, NULL);
897
898 if (sc->sc_parent->dk_rawopens-- == 1) {
899 KASSERT(sc->sc_parent->dk_rawvp != NULL);
900 error = vn_close(sc->sc_parent->dk_rawvp, FREAD | FWRITE,
901 NOCRED, p);
902 sc->sc_parent->dk_rawvp = NULL;
903 }
904 sc->sc_open = 0;
905
906 (void) lockmgr(&sc->sc_parent->dk_rawlock, LK_RELEASE, NULL);
907
908 return (error);
909 }
910
911 /*
912 * dkstragegy: [devsw entry point]
913 *
914 * Perform I/O based on the wedge I/O strategy.
915 */
916 static void
917 dkstrategy(struct buf *bp)
918 {
919 struct dkwedge_softc *sc = dkwedge_lookup(bp->b_dev);
920 int s;
921
922 if (sc->sc_state != DKW_STATE_RUNNING) {
923 bp->b_error = ENXIO;
924 bp->b_flags |= B_ERROR;
925 goto done;
926 }
927
928 /* If it's an empty transfer, wake up the top half now. */
929 if (bp->b_bcount == 0)
930 goto done;
931
932 /* Make sure it's in-range. */
933 if (bounds_check_with_mediasize(bp, DEV_BSIZE, sc->sc_size) <= 0)
934 goto done;
935
936 /* Translate it to the parent's raw LBA. */
937 bp->b_rawblkno = bp->b_blkno + sc->sc_offset;
938
939 /* Place it in the queue and start I/O on the unit. */
940 s = splbio();
941 sc->sc_iopend++;
942 BUFQ_PUT(&sc->sc_bufq, bp);
943 dkstart(sc);
944 splx(s);
945 return;
946
947 done:
948 bp->b_resid = bp->b_bcount;
949 biodone(bp);
950 }
951
952 /*
953 * dkstart:
954 *
955 * Start I/O that has been enqueued on the wedge.
956 * NOTE: Must be called at splbio()!
957 */
958 static void
959 dkstart(struct dkwedge_softc *sc)
960 {
961 struct buf *bp, *nbp;
962
963 /* Do as much work as has been enqueued. */
964 while ((bp = BUFQ_PEEK(&sc->sc_bufq)) != NULL) {
965 if (sc->sc_state != DKW_STATE_RUNNING) {
966 (void) BUFQ_GET(&sc->sc_bufq);
967 if (sc->sc_iopend-- == 1 &&
968 (sc->sc_flags & DK_F_WAIT_DRAIN) != 0) {
969 sc->sc_flags &= ~DK_F_WAIT_DRAIN;
970 wakeup(&sc->sc_iopend);
971 }
972 bp->b_error = ENXIO;
973 bp->b_flags |= B_ERROR;
974 bp->b_resid = bp->b_bcount;
975 biodone(bp);
976 }
977
978 /* Instrumentation. */
979 disk_busy(&sc->sc_dk);
980
981 nbp = pool_get(&bufpool, PR_NOWAIT);
982 if (nbp == NULL) {
983 /*
984 * No resources to run this request; leave the
985 * buffer queued up, and schedule a timer to
986 * restart the queue in 1/2 a second.
987 */
988 disk_unbusy(&sc->sc_dk, 0, bp->b_flags & B_READ);
989 callout_schedule(&sc->sc_restart_ch, hz / 2);
990 return;
991 }
992
993 (void) BUFQ_GET(&sc->sc_bufq);
994
995 BUF_INIT(nbp);
996 nbp->b_data = bp->b_data;
997 nbp->b_flags = bp->b_flags | B_CALL;
998 nbp->b_iodone = dkiodone;
999 nbp->b_proc = bp->b_proc;
1000 nbp->b_blkno = bp->b_rawblkno;
1001 nbp->b_dev = sc->sc_parent->dk_rawvp->v_rdev;
1002 nbp->b_vp = sc->sc_parent->dk_rawvp;
1003 nbp->b_bcount = bp->b_bcount;
1004 nbp->b_private = bp;
1005 BIO_COPYPRIO(nbp, bp);
1006
1007 if ((nbp->b_flags & B_READ) == 0)
1008 V_INCR_NUMOUTPUT(nbp->b_vp);
1009 VOP_STRATEGY(nbp->b_vp, nbp);
1010 }
1011 }
1012
1013 /*
1014 * dkiodone:
1015 *
1016 * I/O to a wedge has completed; alert the top half.
1017 * NOTE: Must be called at splbio()!
1018 */
1019 static void
1020 dkiodone(struct buf *bp)
1021 {
1022 struct buf *obp = bp->b_private;
1023 struct dkwedge_softc *sc = dkwedge_lookup(obp->b_dev);
1024
1025 if (bp->b_flags & B_ERROR) {
1026 obp->b_flags |= B_ERROR;
1027 obp->b_error = bp->b_error;
1028 }
1029 obp->b_resid = bp->b_resid;
1030 pool_put(&bufpool, bp);
1031
1032 if (sc->sc_iopend-- == 1 && (sc->sc_flags & DK_F_WAIT_DRAIN) != 0) {
1033 sc->sc_flags &= ~DK_F_WAIT_DRAIN;
1034 wakeup(&sc->sc_iopend);
1035 }
1036
1037 disk_unbusy(&sc->sc_dk, obp->b_bcount - obp->b_resid,
1038 obp->b_flags & B_READ);
1039
1040 biodone(obp);
1041
1042 /* Kick the queue in case there is more work we can do. */
1043 dkstart(sc);
1044 }
1045
1046 /*
1047 * dkrestart:
1048 *
1049 * Restart the work queue after it was stalled due to
1050 * a resource shortage. Invoked via a callout.
1051 */
1052 static void
1053 dkrestart(void *v)
1054 {
1055 struct dkwedge_softc *sc = v;
1056 int s;
1057
1058 s = splbio();
1059 dkstart(sc);
1060 splx(s);
1061 }
1062
1063 /*
1064 * dkread: [devsw entry point]
1065 *
1066 * Read from a wedge.
1067 */
1068 static int
1069 dkread(dev_t dev, struct uio *uio, int flags)
1070 {
1071 struct dkwedge_softc *sc = dkwedge_lookup(dev);
1072
1073 if (sc->sc_state != DKW_STATE_RUNNING)
1074 return (ENXIO);
1075
1076 return (physio(dkstrategy, NULL, dev, B_READ,
1077 sc->sc_parent->dk_driver->d_minphys, uio));
1078 }
1079
1080 /*
1081 * dkwrite: [devsw entry point]
1082 *
1083 * Write to a wedge.
1084 */
1085 static int
1086 dkwrite(dev_t dev, struct uio *uio, int flags)
1087 {
1088 struct dkwedge_softc *sc = dkwedge_lookup(dev);
1089
1090 if (sc->sc_state != DKW_STATE_RUNNING)
1091 return (ENXIO);
1092
1093 return (physio(dkstrategy, NULL, dev, B_WRITE,
1094 sc->sc_parent->dk_driver->d_minphys, uio));
1095 }
1096
1097 /*
1098 * dkioctl: [devsw entry point]
1099 *
1100 * Perform an ioctl request on a wedge.
1101 */
1102 static int
1103 dkioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
1104 {
1105 struct dkwedge_softc *sc = dkwedge_lookup(dev);
1106 int error = 0;
1107
1108 if (sc->sc_state != DKW_STATE_RUNNING)
1109 return (ENXIO);
1110
1111 switch (cmd) {
1112 case DIOCGWEDGEINFO:
1113 {
1114 struct dkwedge_info *dkw = (void *) data;
1115
1116 strcpy(dkw->dkw_devname, sc->sc_dev->dv_xname);
1117 memcpy(dkw->dkw_wname, sc->sc_wname, sizeof(dkw->dkw_wname));
1118 dkw->dkw_wname[sizeof(dkw->dkw_wname) - 1] = '\0';
1119 strcpy(dkw->dkw_parent, sc->sc_parent->dk_name);
1120 dkw->dkw_offset = sc->sc_offset;
1121 dkw->dkw_size = sc->sc_size;
1122 strcpy(dkw->dkw_ptype, sc->sc_ptype);
1123
1124 break;
1125 }
1126
1127 default:
1128 error = ENOTTY;
1129 }
1130
1131 return (error);
1132 }
1133
1134 /*
1135 * dksize: [devsw entry point]
1136 *
1137 * Query the size of a wedge for the purpose of performing a dump
1138 * or for swapping to.
1139 */
1140 static int
1141 dksize(dev_t dev)
1142 {
1143
1144 /* XXX */
1145 return (-1);
1146 }
1147
1148 /*
1149 * dkdump: [devsw entry point]
1150 *
1151 * Perform a crash dump to a wedge.
1152 */
1153 static int
1154 dkdump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1155 {
1156
1157 /* XXX */
1158 return (ENXIO);
1159 }
1160