1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
24 * Copyright 2012 Garrett D'Amore <garrett@damore.org>. All rights reserved.
25 * Copyright 2012 Alexey Zaytsev <alexey.zaytsev@gmail.com> All rights reserved.
26 */
27
28 #include <sys/types.h>
29 #include <sys/ksynch.h>
30 #include <sys/kmem.h>
31 #include <sys/file.h>
32 #include <sys/errno.h>
33 #include <sys/open.h>
34 #include <sys/buf.h>
35 #include <sys/uio.h>
36 #include <sys/aio_req.h>
37 #include <sys/cred.h>
38 #include <sys/modctl.h>
39 #include <sys/cmlb.h>
40 #include <sys/conf.h>
41 #include <sys/devops.h>
42 #include <sys/list.h>
43 #include <sys/sysmacros.h>
44 #include <sys/dkio.h>
45 #include <sys/vtoc.h>
46 #include <sys/scsi/scsi.h> /* for DTYPE_DIRECT */
47 #include <sys/kstat.h>
48 #include <sys/fs/dv_node.h>
49 #include <sys/ddi.h>
50 #include <sys/sunddi.h>
51 #include <sys/note.h>
52 #include <sys/blkdev.h>
53
54 #define BD_MAXPART 64
55 #define BDINST(dev) (getminor(dev) / BD_MAXPART)
56 #define BDPART(dev) (getminor(dev) % BD_MAXPART)
57
58 typedef struct bd bd_t;
59 typedef struct bd_xfer_impl bd_xfer_impl_t;
60
61 struct bd {
62 void *d_private;
63 dev_info_t *d_dip;
64 kmutex_t d_ocmutex;
65 kmutex_t d_iomutex;
66 kmutex_t d_statemutex;
67 kcondvar_t d_statecv;
68 enum dkio_state d_state;
69 cmlb_handle_t d_cmlbh;
70 unsigned d_open_lyr[BD_MAXPART]; /* open count */
71 uint64_t d_open_excl; /* bit mask indexed by partition */
72 uint64_t d_open_reg[OTYPCNT]; /* bit mask */
73
74 uint32_t d_qsize;
75 uint32_t d_qactive;
76 uint32_t d_maxxfer;
77 uint32_t d_blkshift;
78 uint64_t d_numblks;
79 ddi_devid_t d_devid;
80
81 kmem_cache_t *d_cache;
82 list_t d_runq;
83 list_t d_waitq;
84 kstat_t *d_ksp;
85 kstat_io_t *d_kiop;
86
87 boolean_t d_rdonly;
88 boolean_t d_removable;
89 boolean_t d_hotpluggable;
90 boolean_t d_use_dma;
91
92 ddi_dma_attr_t d_dma;
93 bd_ops_t d_ops;
94 bd_handle_t d_handle;
95 };
96
97 struct bd_handle {
98 bd_ops_t h_ops;
99 ddi_dma_attr_t *h_dma;
100 dev_info_t *h_parent;
101 dev_info_t *h_child;
102 void *h_private;
103 bd_t *h_bd;
104 char *h_name;
105 char h_addr[20]; /* enough for %X,%X */
106 };
107
108 struct bd_xfer_impl {
109 bd_xfer_t i_public;
110 list_node_t i_linkage;
111 bd_t *i_bd;
112 buf_t *i_bp;
113 uint_t i_num_win;
114 uint_t i_cur_win;
115 off_t i_offset;
116 int (*i_func)(void *, bd_xfer_t *);
117 uint32_t i_blkshift;
118 size_t i_len;
119 size_t i_resid;
120 };
121
122 #define i_dmah i_public.x_dmah
123 #define i_dmac i_public.x_dmac
124 #define i_ndmac i_public.x_ndmac
125 #define i_kaddr i_public.x_kaddr
126 #define i_nblks i_public.x_nblks
127 #define i_blkno i_public.x_blkno
128 #define i_flags i_public.x_flags
129
130
131 /*
132 * Private prototypes.
133 */
134
135 static int bd_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
136 static int bd_attach(dev_info_t *, ddi_attach_cmd_t);
137 static int bd_detach(dev_info_t *, ddi_detach_cmd_t);
138
139 static int bd_open(dev_t *, int, int, cred_t *);
140 static int bd_close(dev_t, int, int, cred_t *);
141 static int bd_strategy(struct buf *);
142 static int bd_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
143 static int bd_dump(dev_t, caddr_t, daddr_t, int);
144 static int bd_read(dev_t, struct uio *, cred_t *);
145 static int bd_write(dev_t, struct uio *, cred_t *);
146 static int bd_aread(dev_t, struct aio_req *, cred_t *);
147 static int bd_awrite(dev_t, struct aio_req *, cred_t *);
148 static int bd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int, char *,
149 caddr_t, int *);
150
151 static int bd_tg_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t,
152 void *);
153 static int bd_tg_getinfo(dev_info_t *, int, void *, void *);
154 static int bd_xfer_ctor(void *, void *, int);
155 static void bd_xfer_dtor(void *, void *);
156 static void bd_sched(bd_t *);
157 static void bd_submit(bd_t *, bd_xfer_impl_t *);
158 static void bd_runq_exit(bd_xfer_impl_t *, int);
159 static void bd_update_state(bd_t *);
160 static int bd_check_state(bd_t *, enum dkio_state *);
161 static int bd_flush_write_cache(bd_t *, struct dk_callback *);
162
163 struct cmlb_tg_ops bd_tg_ops = {
164 TG_DK_OPS_VERSION_1,
165 bd_tg_rdwr,
166 bd_tg_getinfo,
167 };
168
169 static struct cb_ops bd_cb_ops = {
170 bd_open, /* open */
171 bd_close, /* close */
172 bd_strategy, /* strategy */
173 nodev, /* print */
174 bd_dump, /* dump */
175 bd_read, /* read */
176 bd_write, /* write */
177 bd_ioctl, /* ioctl */
178 nodev, /* devmap */
179 nodev, /* mmap */
180 nodev, /* segmap */
181 nochpoll, /* poll */
182 bd_prop_op, /* cb_prop_op */
183 0, /* streamtab */
184 D_64BIT | D_MP, /* Driver comaptibility flag */
185 CB_REV, /* cb_rev */
186 bd_aread, /* async read */
187 bd_awrite /* async write */
188 };
189
190 struct dev_ops bd_dev_ops = {
191 DEVO_REV, /* devo_rev, */
192 0, /* refcnt */
193 bd_getinfo, /* getinfo */
194 nulldev, /* identify */
195 nulldev, /* probe */
196 bd_attach, /* attach */
197 bd_detach, /* detach */
198 nodev, /* reset */
199 &bd_cb_ops, /* driver operations */
200 NULL, /* bus operations */
201 NULL, /* power */
202 ddi_quiesce_not_needed, /* quiesce */
203 };
204
205 static struct modldrv modldrv = {
206 &mod_driverops,
207 "Generic Block Device",
208 &bd_dev_ops,
209 };
210
211 static struct modlinkage modlinkage = {
212 MODREV_1, { &modldrv, NULL }
213 };
214
215 static void *bd_state;
216 static krwlock_t bd_lock;
217
218 int
219 _init(void)
220 {
221 int rv;
222
223 rv = ddi_soft_state_init(&bd_state, sizeof (struct bd), 2);
224 if (rv != DDI_SUCCESS) {
225 return (rv);
226 }
227 rw_init(&bd_lock, NULL, RW_DRIVER, NULL);
228 rv = mod_install(&modlinkage);
229 if (rv != DDI_SUCCESS) {
230 rw_destroy(&bd_lock);
231 ddi_soft_state_fini(&bd_state);
232 }
233 return (rv);
234 }
235
236 int
237 _fini(void)
238 {
239 int rv;
240
241 rv = mod_remove(&modlinkage);
242 if (rv == DDI_SUCCESS) {
243 rw_destroy(&bd_lock);
244 ddi_soft_state_fini(&bd_state);
245 }
246 return (rv);
247 }
248
249 int
250 _info(struct modinfo *modinfop)
251 {
252 return (mod_info(&modlinkage, modinfop));
253 }
254
255 static int
256 bd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resultp)
257 {
258 bd_t *bd;
259 minor_t inst;
260
261 _NOTE(ARGUNUSED(dip));
262
263 inst = BDINST((dev_t)arg);
264
265 switch (cmd) {
266 case DDI_INFO_DEVT2DEVINFO:
267 bd = ddi_get_soft_state(bd_state, inst);
268 if (bd == NULL) {
269 return (DDI_FAILURE);
270 }
271 *resultp = (void *)bd->d_dip;
272 break;
273
274 case DDI_INFO_DEVT2INSTANCE:
275 *resultp = (void *)(intptr_t)inst;
276 break;
277
278 default:
279 return (DDI_FAILURE);
280 }
281 return (DDI_SUCCESS);
282 }
283
284 static int
285 bd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
286 {
287 int inst;
288 bd_handle_t hdl;
289 bd_t *bd;
290 bd_drive_t drive;
291 int rv;
292 char name[16];
293 char kcache[32];
294
295 switch (cmd) {
296 case DDI_ATTACH:
297 break;
298 case DDI_RESUME:
299 /* We don't do anything native for suspend/resume */
300 return (DDI_SUCCESS);
301 default:
302 return (DDI_FAILURE);
303 }
304
305 inst = ddi_get_instance(dip);
306 hdl = ddi_get_parent_data(dip);
307
308 (void) snprintf(name, sizeof (name), "%s%d",
309 ddi_driver_name(dip), ddi_get_instance(dip));
310 (void) snprintf(kcache, sizeof (kcache), "%s_xfer", name);
311
312 if (hdl == NULL) {
313 cmn_err(CE_WARN, "%s: missing parent data!", name);
314 return (DDI_FAILURE);
315 }
316
317 if (ddi_soft_state_zalloc(bd_state, inst) != DDI_SUCCESS) {
318 cmn_err(CE_WARN, "%s: unable to zalloc soft state!", name);
319 return (DDI_FAILURE);
320 }
321 bd = ddi_get_soft_state(bd_state, inst);
322
323 if (hdl->h_dma) {
324 bd->d_dma = *(hdl->h_dma);
325 bd->d_dma.dma_attr_granular =
326 max(DEV_BSIZE, bd->d_dma.dma_attr_granular);
327 bd->d_use_dma = B_TRUE;
328
329 if (bd->d_maxxfer &&
330 (bd->d_maxxfer != bd->d_dma.dma_attr_maxxfer)) {
331 cmn_err(CE_WARN,
332 "%s: inconsistent maximum transfer size!",
333 name);
334 /* We force it */
335 bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
336 } else {
337 bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
338 }
339 } else {
340 bd->d_use_dma = B_FALSE;
341 if (bd->d_maxxfer == 0) {
342 bd->d_maxxfer = 1024 * 1024;
343 }
344 }
345 bd->d_ops = hdl->h_ops;
346 bd->d_private = hdl->h_private;
347 bd->d_blkshift = 9; /* 512 bytes, to start */
348
349 if (bd->d_maxxfer % DEV_BSIZE) {
350 cmn_err(CE_WARN, "%s: maximum transfer misaligned!", name);
351 bd->d_maxxfer &= ~(DEV_BSIZE - 1);
352 }
353 if (bd->d_maxxfer < DEV_BSIZE) {
354 cmn_err(CE_WARN, "%s: maximum transfer size too small!", name);
355 ddi_soft_state_free(bd_state, inst);
356 return (DDI_FAILURE);
357 }
358
359 bd->d_dip = dip;
360 bd->d_handle = hdl;
361 hdl->h_bd = bd;
362 ddi_set_driver_private(dip, bd);
363
364 mutex_init(&bd->d_iomutex, NULL, MUTEX_DRIVER, NULL);
365 mutex_init(&bd->d_ocmutex, NULL, MUTEX_DRIVER, NULL);
366 mutex_init(&bd->d_statemutex, NULL, MUTEX_DRIVER, NULL);
367 cv_init(&bd->d_statecv, NULL, CV_DRIVER, NULL);
368
369 list_create(&bd->d_waitq, sizeof (bd_xfer_impl_t),
370 offsetof(struct bd_xfer_impl, i_linkage));
371 list_create(&bd->d_runq, sizeof (bd_xfer_impl_t),
372 offsetof(struct bd_xfer_impl, i_linkage));
373
374 bd->d_cache = kmem_cache_create(kcache, sizeof (bd_xfer_impl_t), 8,
375 bd_xfer_ctor, bd_xfer_dtor, NULL, bd, NULL, 0);
376
377 bd->d_ksp = kstat_create(ddi_driver_name(dip), inst, NULL, "disk",
378 KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
379 if (bd->d_ksp != NULL) {
380 bd->d_ksp->ks_lock = &bd->d_iomutex;
381 kstat_install(bd->d_ksp);
382 bd->d_kiop = bd->d_ksp->ks_data;
383 } else {
384 /*
385 * Even if we cannot create the kstat, we create a
386 * scratch kstat. The reason for this is to ensure
387 * that we can update the kstat all of the time,
388 * without adding an extra branch instruction.
389 */
390 bd->d_kiop = kmem_zalloc(sizeof (kstat_io_t), KM_SLEEP);
391 }
392
393 cmlb_alloc_handle(&bd->d_cmlbh);
394
395 bd->d_state = DKIO_NONE;
396
397 bzero(&drive, sizeof (drive));
398 bd->d_ops.o_drive_info(bd->d_private, &drive);
399 bd->d_qsize = drive.d_qsize;
400 bd->d_removable = drive.d_removable;
401 bd->d_hotpluggable = drive.d_hotpluggable;
402
403 if (drive.d_maxxfer && drive.d_maxxfer < bd->d_maxxfer)
404 bd->d_maxxfer = drive.d_maxxfer;
405
406
407 rv = cmlb_attach(dip, &bd_tg_ops, DTYPE_DIRECT,
408 bd->d_removable, bd->d_hotpluggable,
409 drive.d_lun >= 0 ? DDI_NT_BLOCK_CHAN : DDI_NT_BLOCK,
410 CMLB_FAKE_LABEL_ONE_PARTITION, bd->d_cmlbh, 0);
411 if (rv != 0) {
412 cmlb_free_handle(&bd->d_cmlbh);
413 kmem_cache_destroy(bd->d_cache);
414 mutex_destroy(&bd->d_iomutex);
415 mutex_destroy(&bd->d_ocmutex);
416 mutex_destroy(&bd->d_statemutex);
417 cv_destroy(&bd->d_statecv);
418 list_destroy(&bd->d_waitq);
419 list_destroy(&bd->d_runq);
420 if (bd->d_ksp != NULL) {
421 kstat_delete(bd->d_ksp);
422 bd->d_ksp = NULL;
423 } else {
424 kmem_free(bd->d_kiop, sizeof (kstat_io_t));
425 }
426 ddi_soft_state_free(bd_state, inst);
427 return (DDI_FAILURE);
428 }
429
430 if (bd->d_ops.o_devid_init != NULL) {
431 rv = bd->d_ops.o_devid_init(bd->d_private, dip, &bd->d_devid);
432 if (rv == DDI_SUCCESS) {
433 if (ddi_devid_register(dip, bd->d_devid) !=
434 DDI_SUCCESS) {
435 cmn_err(CE_WARN,
436 "%s: unable to register devid", name);
437 }
438 }
439 }
440
441 /*
442 * Add a zero-length attribute to tell the world we support
443 * kernel ioctls (for layered drivers). Also set up properties
444 * used by HAL to identify removable media.
445 */
446 (void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
447 DDI_KERNEL_IOCTL, NULL, 0);
448 if (bd->d_removable) {
449 (void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
450 "removable-media", NULL, 0);
451 }
452 if (bd->d_hotpluggable) {
453 (void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
454 "hotpluggable", NULL, 0);
455 }
456
457 ddi_report_dev(dip);
458
459 return (DDI_SUCCESS);
460 }
461
462 static int
463 bd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
464 {
465 bd_t *bd;
466
467 bd = ddi_get_driver_private(dip);
468
469 switch (cmd) {
470 case DDI_DETACH:
471 break;
472 case DDI_SUSPEND:
473 /* We don't suspend, but our parent does */
474 return (DDI_SUCCESS);
475 default:
476 return (DDI_FAILURE);
477 }
478 if (bd->d_ksp != NULL) {
479 kstat_delete(bd->d_ksp);
480 bd->d_ksp = NULL;
481 } else {
482 kmem_free(bd->d_kiop, sizeof (kstat_io_t));
483 }
484 cmlb_detach(bd->d_cmlbh, 0);
485 cmlb_free_handle(&bd->d_cmlbh);
486 if (bd->d_devid)
487 ddi_devid_free(bd->d_devid);
488 kmem_cache_destroy(bd->d_cache);
489 mutex_destroy(&bd->d_iomutex);
490 mutex_destroy(&bd->d_ocmutex);
491 mutex_destroy(&bd->d_statemutex);
492 cv_destroy(&bd->d_statecv);
493 list_destroy(&bd->d_waitq);
494 list_destroy(&bd->d_runq);
495 ddi_soft_state_free(bd_state, ddi_get_instance(dip));
496 return (DDI_SUCCESS);
497 }
498
499 static int
500 bd_xfer_ctor(void *buf, void *arg, int kmflag)
501 {
502 bd_xfer_impl_t *xi;
503 bd_t *bd = arg;
504 int (*dcb)(caddr_t);
505
506 if (kmflag == KM_PUSHPAGE || kmflag == KM_SLEEP) {
507 dcb = DDI_DMA_SLEEP;
508 } else {
509 dcb = DDI_DMA_DONTWAIT;
510 }
511
512 xi = buf;
513 bzero(xi, sizeof (*xi));
514 xi->i_bd = bd;
515
516 if (bd->d_use_dma) {
517 if (ddi_dma_alloc_handle(bd->d_dip, &bd->d_dma, dcb, NULL,
518 &xi->i_dmah) != DDI_SUCCESS) {
519 return (-1);
520 }
521 }
522
523 return (0);
524 }
525
526 static void
527 bd_xfer_dtor(void *buf, void *arg)
528 {
529 bd_xfer_impl_t *xi = buf;
530
531 _NOTE(ARGUNUSED(arg));
532
533 if (xi->i_dmah)
534 ddi_dma_free_handle(&xi->i_dmah);
535 xi->i_dmah = NULL;
536 }
537
538 static bd_xfer_impl_t *
539 bd_xfer_alloc(bd_t *bd, struct buf *bp, int (*func)(void *, bd_xfer_t *),
540 int kmflag)
541 {
542 bd_xfer_impl_t *xi;
543 int rv;
544 int status;
545 unsigned dir;
546 int (*cb)(caddr_t);
547 size_t len;
548 uint32_t shift;
549
550 if (kmflag == KM_SLEEP) {
551 cb = DDI_DMA_SLEEP;
552 } else {
553 cb = DDI_DMA_DONTWAIT;
554 }
555
556 xi = kmem_cache_alloc(bd->d_cache, kmflag);
557 if (xi == NULL) {
558 bioerror(bp, ENOMEM);
559 return (NULL);
560 }
561
562 ASSERT(bp);
563
564 xi->i_bp = bp;
565 xi->i_func = func;
566 xi->i_blkno = bp->b_lblkno;
567
568 if (bp->b_bcount == 0) {
569 xi->i_len = 0;
570 xi->i_nblks = 0;
571 xi->i_kaddr = NULL;
572 xi->i_resid = 0;
573 xi->i_num_win = 0;
574 goto done;
575 }
576
577 if (bp->b_flags & B_READ) {
578 dir = DDI_DMA_READ;
579 xi->i_func = bd->d_ops.o_read;
580 } else {
581 dir = DDI_DMA_WRITE;
582 xi->i_func = bd->d_ops.o_write;
583 }
584
585 shift = bd->d_blkshift;
586 xi->i_blkshift = shift;
587
588 if (!bd->d_use_dma) {
589 bp_mapin(bp);
590 rv = 0;
591 xi->i_offset = 0;
592 xi->i_num_win =
593 (bp->b_bcount + (bd->d_maxxfer - 1)) / bd->d_maxxfer;
594 xi->i_cur_win = 0;
595 xi->i_len = min(bp->b_bcount, bd->d_maxxfer);
596 xi->i_nblks = xi->i_len >> shift;
597 xi->i_kaddr = bp->b_un.b_addr;
598 xi->i_resid = bp->b_bcount;
599 } else {
600
601 /*
602 * We have to use consistent DMA if the address is misaligned.
603 */
604 if (((bp->b_flags & (B_PAGEIO | B_REMAPPED)) != B_PAGEIO) &&
605 ((uintptr_t)bp->b_un.b_addr & 0x7)) {
606 dir |= DDI_DMA_CONSISTENT | DDI_DMA_PARTIAL;
607 } else {
608 dir |= DDI_DMA_STREAMING | DDI_DMA_PARTIAL;
609 }
610
611 status = ddi_dma_buf_bind_handle(xi->i_dmah, bp, dir, cb,
612 NULL, &xi->i_dmac, &xi->i_ndmac);
613 switch (status) {
614 case DDI_DMA_MAPPED:
615 xi->i_num_win = 1;
616 xi->i_cur_win = 0;
617 xi->i_offset = 0;
618 xi->i_len = bp->b_bcount;
619 xi->i_nblks = xi->i_len >> shift;
620 xi->i_resid = bp->b_bcount;
621 rv = 0;
622 break;
623 case DDI_DMA_PARTIAL_MAP:
624 xi->i_cur_win = 0;
625
626 if ((ddi_dma_numwin(xi->i_dmah, &xi->i_num_win) !=
627 DDI_SUCCESS) ||
628 (ddi_dma_getwin(xi->i_dmah, 0, &xi->i_offset,
629 &len, &xi->i_dmac, &xi->i_ndmac) !=
630 DDI_SUCCESS) ||
631 (P2PHASE(len, shift) != 0)) {
632 (void) ddi_dma_unbind_handle(xi->i_dmah);
633 rv = EFAULT;
634 goto done;
635 }
636 xi->i_len = len;
637 xi->i_nblks = xi->i_len >> shift;
638 xi->i_resid = bp->b_bcount;
639 rv = 0;
640 break;
641 case DDI_DMA_NORESOURCES:
642 rv = EAGAIN;
643 goto done;
644 case DDI_DMA_TOOBIG:
645 rv = EINVAL;
646 goto done;
647 case DDI_DMA_NOMAPPING:
648 case DDI_DMA_INUSE:
649 default:
650 rv = EFAULT;
651 goto done;
652 }
653 }
654
655 done:
656 if (rv != 0) {
657 kmem_cache_free(bd->d_cache, xi);
658 bioerror(bp, rv);
659 return (NULL);
660 }
661
662 return (xi);
663 }
664
665 static void
666 bd_xfer_free(bd_xfer_impl_t *xi)
667 {
668 if (xi->i_dmah) {
669 (void) ddi_dma_unbind_handle(xi->i_dmah);
670 }
671 kmem_cache_free(xi->i_bd->d_cache, xi);
672 }
673
674 static int
675 bd_open(dev_t *devp, int flag, int otyp, cred_t *credp)
676 {
677 dev_t dev = *devp;
678 bd_t *bd;
679 minor_t part;
680 minor_t inst;
681 uint64_t mask;
682 boolean_t ndelay;
683 int rv;
684 diskaddr_t nblks;
685 diskaddr_t lba;
686
687 _NOTE(ARGUNUSED(credp));
688
689 part = BDPART(dev);
690 inst = BDINST(dev);
691
692 if (otyp >= OTYPCNT)
693 return (EINVAL);
694
695 ndelay = (flag & (FNDELAY | FNONBLOCK)) ? B_TRUE : B_FALSE;
696
697 /*
698 * Block any DR events from changing the set of registered
699 * devices while we function.
700 */
701 rw_enter(&bd_lock, RW_READER);
702 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
703 rw_exit(&bd_lock);
704 return (ENXIO);
705 }
706
707 mutex_enter(&bd->d_ocmutex);
708
709 ASSERT(part < 64);
710 mask = (1U << part);
711
712 bd_update_state(bd);
713
714 if (cmlb_validate(bd->d_cmlbh, 0, 0) != 0) {
715
716 /* non-blocking opens are allowed to succeed */
717 if (!ndelay) {
718 rv = ENXIO;
719 goto done;
720 }
721 } else if (cmlb_partinfo(bd->d_cmlbh, part, &nblks, &lba,
722 NULL, NULL, 0) == 0) {
723
724 /*
725 * We read the partinfo, verify valid ranges. If the
726 * partition is invalid, and we aren't blocking or
727 * doing a raw access, then fail. (Non-blocking and
728 * raw accesses can still succeed to allow a disk with
729 * bad partition data to opened by format and fdisk.)
730 */
731 if ((!nblks) && ((!ndelay) || (otyp != OTYP_CHR))) {
732 rv = ENXIO;
733 goto done;
734 }
735 } else if (!ndelay) {
736 /*
737 * cmlb_partinfo failed -- invalid partition or no
738 * disk label.
739 */
740 rv = ENXIO;
741 goto done;
742 }
743
744 if ((flag & FWRITE) && bd->d_rdonly) {
745 rv = EROFS;
746 goto done;
747 }
748
749 if ((bd->d_open_excl) & (mask)) {
750 rv = EBUSY;
751 goto done;
752 }
753 if (flag & FEXCL) {
754 if (bd->d_open_lyr[part]) {
755 rv = EBUSY;
756 goto done;
757 }
758 for (int i = 0; i < OTYP_LYR; i++) {
759 if (bd->d_open_reg[i] & mask) {
760 rv = EBUSY;
761 goto done;
762 }
763 }
764 }
765
766 if (otyp == OTYP_LYR) {
767 bd->d_open_lyr[part]++;
768 } else {
769 bd->d_open_reg[otyp] |= mask;
770 }
771 if (flag & FEXCL) {
772 bd->d_open_excl |= mask;
773 }
774
775 rv = 0;
776 done:
777 mutex_exit(&bd->d_ocmutex);
778 rw_exit(&bd_lock);
779
780 return (rv);
781 }
782
783 static int
784 bd_close(dev_t dev, int flag, int otyp, cred_t *credp)
785 {
786 bd_t *bd;
787 minor_t inst;
788 minor_t part;
789 uint64_t mask;
790 boolean_t last = B_TRUE;
791
792 _NOTE(ARGUNUSED(flag));
793 _NOTE(ARGUNUSED(credp));
794
795 part = BDPART(dev);
796 inst = BDINST(dev);
797
798 ASSERT(part < 64);
799 mask = (1U << part);
800
801 rw_enter(&bd_lock, RW_READER);
802
803 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
804 rw_exit(&bd_lock);
805 return (ENXIO);
806 }
807
808 mutex_enter(&bd->d_ocmutex);
809 if (bd->d_open_excl & mask) {
810 bd->d_open_excl &= ~mask;
811 }
812 if (otyp == OTYP_LYR) {
813 bd->d_open_lyr[part]--;
814 } else {
815 bd->d_open_reg[otyp] &= ~mask;
816 }
817 for (int i = 0; i < 64; i++) {
818 if (bd->d_open_lyr[part]) {
819 last = B_FALSE;
820 }
821 }
822 for (int i = 0; last && (i < OTYP_LYR); i++) {
823 if (bd->d_open_reg[i]) {
824 last = B_FALSE;
825 }
826 }
827 mutex_exit(&bd->d_ocmutex);
828
829 if (last) {
830 cmlb_invalidate(bd->d_cmlbh, 0);
831 }
832 rw_exit(&bd_lock);
833
834 return (0);
835 }
836
837 static int
838 bd_dump(dev_t dev, caddr_t caddr, daddr_t blkno, int nblk)
839 {
840 minor_t inst;
841 minor_t part;
842 diskaddr_t pstart;
843 diskaddr_t psize;
844 bd_t *bd;
845 bd_xfer_impl_t *xi;
846 buf_t *bp;
847 int rv;
848
849 rw_enter(&bd_lock, RW_READER);
850
851 part = BDPART(dev);
852 inst = BDINST(dev);
853
854 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
855 rw_exit(&bd_lock);
856 return (ENXIO);
857 }
858 /*
859 * do cmlb, but do it synchronously unless we already have the
860 * partition (which we probably should.)
861 */
862 if (cmlb_partinfo(bd->d_cmlbh, part, &psize, &pstart, NULL, NULL,
863 (void *)1)) {
864 rw_exit(&bd_lock);
865 return (ENXIO);
866 }
867
868 if ((blkno + nblk) > psize) {
869 rw_exit(&bd_lock);
870 return (EINVAL);
871 }
872 bp = getrbuf(KM_NOSLEEP);
873 if (bp == NULL) {
874 rw_exit(&bd_lock);
875 return (ENOMEM);
876 }
877
878 bp->b_bcount = nblk << bd->d_blkshift;
879 bp->b_resid = bp->b_bcount;
880 bp->b_lblkno = blkno;
881 bp->b_un.b_addr = caddr;
882
883 xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_write, KM_NOSLEEP);
884 if (xi == NULL) {
885 rw_exit(&bd_lock);
886 freerbuf(bp);
887 return (ENOMEM);
888 }
889 xi->i_blkno = blkno + pstart;
890 xi->i_flags = BD_XFER_POLL;
891 bd_submit(bd, xi);
892 rw_exit(&bd_lock);
893
894 /*
895 * Generally, we should have run this entirely synchronously
896 * at this point and the biowait call should be a no-op. If
897 * it didn't happen this way, it's a bug in the underlying
898 * driver not honoring BD_XFER_POLL.
899 */
900 (void) biowait(bp);
901 rv = geterror(bp);
902 freerbuf(bp);
903 return (rv);
904 }
905
906 void
907 bd_minphys(struct buf *bp)
908 {
909 minor_t inst;
910 bd_t *bd;
911 inst = BDINST(bp->b_edev);
912
913 bd = ddi_get_soft_state(bd_state, inst);
914
915 /*
916 * In a non-debug kernel, bd_strategy will catch !bd as
917 * well, and will fail nicely.
918 */
919 ASSERT(bd);
920
921 if (bp->b_bcount > bd->d_maxxfer)
922 bp->b_bcount = bd->d_maxxfer;
923 }
924
925 static int
926 bd_read(dev_t dev, struct uio *uio, cred_t *credp)
927 {
928 _NOTE(ARGUNUSED(credp));
929 return (physio(bd_strategy, NULL, dev, B_READ, bd_minphys, uio));
930 }
931
932 static int
933 bd_write(dev_t dev, struct uio *uio, cred_t *credp)
934 {
935 _NOTE(ARGUNUSED(credp));
936 return (physio(bd_strategy, NULL, dev, B_WRITE, bd_minphys, uio));
937 }
938
939 static int
940 bd_aread(dev_t dev, struct aio_req *aio, cred_t *credp)
941 {
942 _NOTE(ARGUNUSED(credp));
943 return (aphysio(bd_strategy, anocancel, dev, B_READ, bd_minphys, aio));
944 }
945
946 static int
947 bd_awrite(dev_t dev, struct aio_req *aio, cred_t *credp)
948 {
949 _NOTE(ARGUNUSED(credp));
950 return (aphysio(bd_strategy, anocancel, dev, B_WRITE, bd_minphys, aio));
951 }
952
953 static int
954 bd_strategy(struct buf *bp)
955 {
956 minor_t inst;
957 minor_t part;
958 bd_t *bd;
959 diskaddr_t p_lba;
960 diskaddr_t p_nblks;
961 diskaddr_t b_nblks;
962 bd_xfer_impl_t *xi;
963 uint32_t shift;
964 int (*func)(void *, bd_xfer_t *);
965
966 part = BDPART(bp->b_edev);
967 inst = BDINST(bp->b_edev);
968
969 ASSERT(bp);
970
971 bp->b_resid = bp->b_bcount;
972
973 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
974 bioerror(bp, ENXIO);
975 biodone(bp);
976 return (0);
977 }
978
979 if (cmlb_partinfo(bd->d_cmlbh, part, &p_nblks, &p_lba,
980 NULL, NULL, 0)) {
981 bioerror(bp, ENXIO);
982 biodone(bp);
983 return (0);
984 }
985
986 shift = bd->d_blkshift;
987
988 if ((P2PHASE(bp->b_bcount, (1U << shift)) != 0) ||
989 (bp->b_lblkno > p_nblks)) {
990 bioerror(bp, ENXIO);
991 biodone(bp);
992 return (0);
993 }
994 b_nblks = bp->b_bcount >> shift;
995 if ((bp->b_lblkno == p_nblks) || (bp->b_bcount == 0)) {
996 biodone(bp);
997 return (0);
998 }
999
1000 if ((b_nblks + bp->b_lblkno) > p_nblks) {
1001 bp->b_resid = ((bp->b_lblkno + b_nblks - p_nblks) << shift);
1002 bp->b_bcount -= bp->b_resid;
1003 } else {
1004 bp->b_resid = 0;
1005 }
1006 func = (bp->b_flags & B_READ) ? bd->d_ops.o_read : bd->d_ops.o_write;
1007
1008 xi = bd_xfer_alloc(bd, bp, func, KM_NOSLEEP);
1009 if (xi == NULL) {
1010 xi = bd_xfer_alloc(bd, bp, func, KM_PUSHPAGE);
1011 }
1012 if (xi == NULL) {
1013 /* bd_request_alloc will have done bioerror */
1014 biodone(bp);
1015 return (0);
1016 }
1017 xi->i_blkno = bp->b_lblkno + p_lba;
1018
1019 bd_submit(bd, xi);
1020
1021 return (0);
1022 }
1023
1024 static int
1025 bd_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *credp, int *rvalp)
1026 {
1027 minor_t inst;
1028 uint16_t part;
1029 bd_t *bd;
1030 void *ptr = (void *)arg;
1031 int rv;
1032
1033 part = BDPART(dev);
1034 inst = BDINST(dev);
1035
1036 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1037 return (ENXIO);
1038 }
1039
1040 rv = cmlb_ioctl(bd->d_cmlbh, dev, cmd, arg, flag, credp, rvalp, 0);
1041 if (rv != ENOTTY)
1042 return (rv);
1043
1044 switch (cmd) {
1045 case DKIOCGMEDIAINFO: {
1046 struct dk_minfo minfo;
1047
1048 /* make sure our state information is current */
1049 bd_update_state(bd);
1050 bzero(&minfo, sizeof (minfo));
1051 minfo.dki_media_type = DK_FIXED_DISK;
1052 minfo.dki_lbsize = (1U << bd->d_blkshift);
1053 minfo.dki_capacity = bd->d_numblks;
1054 if (ddi_copyout(&minfo, ptr, sizeof (minfo), flag)) {
1055 return (EFAULT);
1056 }
1057 return (0);
1058 }
1059 case DKIOCINFO: {
1060 struct dk_cinfo cinfo;
1061 bzero(&cinfo, sizeof (cinfo));
1062 cinfo.dki_ctype = DKC_BLKDEV;
1063 cinfo.dki_cnum = ddi_get_instance(ddi_get_parent(bd->d_dip));
1064 (void) snprintf(cinfo.dki_cname, sizeof (cinfo.dki_cname),
1065 "%s", ddi_driver_name(ddi_get_parent(bd->d_dip)));
1066 (void) snprintf(cinfo.dki_dname, sizeof (cinfo.dki_dname),
1067 "%s", ddi_driver_name(bd->d_dip));
1068 cinfo.dki_unit = inst;
1069 cinfo.dki_flags = DKI_FMTVOL;
1070 cinfo.dki_partition = part;
1071 cinfo.dki_maxtransfer = bd->d_maxxfer / DEV_BSIZE;
1072 cinfo.dki_addr = 0;
1073 cinfo.dki_slave = 0;
1074 cinfo.dki_space = 0;
1075 cinfo.dki_prio = 0;
1076 cinfo.dki_vec = 0;
1077 if (ddi_copyout(&cinfo, ptr, sizeof (cinfo), flag)) {
1078 return (EFAULT);
1079 }
1080 return (0);
1081 }
1082 case DKIOCREMOVABLE: {
1083 int i;
1084 i = bd->d_removable ? 1 : 0;
1085 if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1086 return (EFAULT);
1087 }
1088 return (0);
1089 }
1090 case DKIOCHOTPLUGGABLE: {
1091 int i;
1092 i = bd->d_hotpluggable ? 1 : 0;
1093 if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1094 return (EFAULT);
1095 }
1096 return (0);
1097 }
1098 case DKIOCREADONLY: {
1099 int i;
1100 i = bd->d_rdonly ? 1 : 0;
1101 if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1102 return (EFAULT);
1103 }
1104 return (0);
1105 }
1106 case DKIOCSTATE: {
1107 enum dkio_state state;
1108 if (ddi_copyin(ptr, &state, sizeof (state), flag)) {
1109 return (EFAULT);
1110 }
1111 if ((rv = bd_check_state(bd, &state)) != 0) {
1112 return (rv);
1113 }
1114 if (ddi_copyout(&state, ptr, sizeof (state), flag)) {
1115 return (EFAULT);
1116 }
1117 return (0);
1118 }
1119 case DKIOCFLUSHWRITECACHE: {
1120 struct dk_callback *dkc = NULL;
1121
1122 if (flag & FKIOCTL)
1123 dkc = (void *)arg;
1124
1125 rv = bd_flush_write_cache(bd, dkc);
1126 return (rv);
1127 }
1128
1129 default:
1130 if (bd->d_ops.o_ioctl != NULL) {
1131 rv = bd->d_ops.o_ioctl(dev, cmd, arg, flag, credp,
1132 rvalp);
1133 } else {
1134 /* Unsupported ioctl ==> return ENOTTY. */
1135 rv = ENOTTY;
1136 }
1137 /* FALLTHRU */
1138 }
1139 return (rv);
1140 }
1141
1142 static int
1143 bd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1144 char *name, caddr_t valuep, int *lengthp)
1145 {
1146 bd_t *bd;
1147
1148 bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1149 if (bd == NULL)
1150 return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1151 name, valuep, lengthp));
1152
1153 return (cmlb_prop_op(bd->d_cmlbh, dev, dip, prop_op, mod_flags, name,
1154 valuep, lengthp, BDPART(dev), 0));
1155 }
1156
1157
1158 static int
1159 bd_tg_rdwr(dev_info_t *dip, uchar_t cmd, void *bufaddr, diskaddr_t start,
1160 size_t length, void *tg_cookie)
1161 {
1162 bd_t *bd;
1163 buf_t *bp;
1164 bd_xfer_impl_t *xi;
1165 int rv;
1166 int (*func)(void *, bd_xfer_t *);
1167 int kmflag;
1168
1169 /*
1170 * If we are running in polled mode (such as during dump(9e)
1171 * execution), then we cannot sleep for kernel allocations.
1172 */
1173 kmflag = tg_cookie ? KM_NOSLEEP : KM_SLEEP;
1174
1175 bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1176
1177 if (P2PHASE(length, (1U << bd->d_blkshift)) != 0) {
1178 /* We can only transfer whole blocks at a time! */
1179 return (EINVAL);
1180 }
1181
1182 if ((bp = getrbuf(kmflag)) == NULL) {
1183 return (ENOMEM);
1184 }
1185
1186 switch (cmd) {
1187 case TG_READ:
1188 bp->b_flags = B_READ;
1189 func = bd->d_ops.o_read;
1190 break;
1191 case TG_WRITE:
1192 bp->b_flags = B_WRITE;
1193 func = bd->d_ops.o_write;
1194 break;
1195 default:
1196 freerbuf(bp);
1197 return (EINVAL);
1198 }
1199
1200 bp->b_un.b_addr = bufaddr;
1201 bp->b_bcount = length;
1202 xi = bd_xfer_alloc(bd, bp, func, kmflag);
1203 if (xi == NULL) {
1204 rv = geterror(bp);
1205 freerbuf(bp);
1206 return (rv);
1207 }
1208 xi->i_flags = tg_cookie ? BD_XFER_POLL : 0;
1209 xi->i_blkno = start;
1210 bd_submit(bd, xi);
1211 (void) biowait(bp);
1212 rv = geterror(bp);
1213 freerbuf(bp);
1214
1215 return (rv);
1216 }
1217
1218 static int
1219 bd_tg_getinfo(dev_info_t *dip, int cmd, void *arg, void *tg_cookie)
1220 {
1221 bd_t *bd;
1222
1223 _NOTE(ARGUNUSED(tg_cookie));
1224 bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1225
1226 switch (cmd) {
1227 case TG_GETPHYGEOM:
1228 case TG_GETVIRTGEOM:
1229 /*
1230 * We don't have any "geometry" as such, let cmlb
1231 * fabricate something.
1232 */
1233 return (ENOTTY);
1234
1235 case TG_GETCAPACITY:
1236 bd_update_state(bd);
1237 *(diskaddr_t *)arg = bd->d_numblks;
1238 return (0);
1239
1240 case TG_GETBLOCKSIZE:
1241 *(uint32_t *)arg = (1U << bd->d_blkshift);
1242 return (0);
1243
1244 case TG_GETATTR:
1245 /*
1246 * It turns out that cmlb really doesn't do much for
1247 * non-writable media, but lets make the information
1248 * available for it in case it does more in the
1249 * future. (The value is currently used for
1250 * triggering special behavior for CD-ROMs.)
1251 */
1252 bd_update_state(bd);
1253 ((tg_attribute_t *)arg)->media_is_writable =
1254 bd->d_rdonly ? B_FALSE : B_TRUE;
1255 return (0);
1256
1257 default:
1258 return (EINVAL);
1259 }
1260 }
1261
1262
1263 static void
1264 bd_sched(bd_t *bd)
1265 {
1266 bd_xfer_impl_t *xi;
1267 struct buf *bp;
1268 int rv;
1269
1270 mutex_enter(&bd->d_iomutex);
1271
1272 while ((bd->d_qactive < bd->d_qsize) &&
1273 ((xi = list_remove_head(&bd->d_waitq)) != NULL)) {
1274 bd->d_qactive++;
1275 kstat_waitq_to_runq(bd->d_kiop);
1276 list_insert_tail(&bd->d_runq, xi);
1277
1278 /*
1279 * Submit the job to the driver. We drop the I/O mutex
1280 * so that we can deal with the case where the driver
1281 * completion routine calls back into us synchronously.
1282 */
1283
1284 mutex_exit(&bd->d_iomutex);
1285
1286 rv = xi->i_func(bd->d_private, &xi->i_public);
1287 if (rv != 0) {
1288 bp = xi->i_bp;
1289 bd_xfer_free(xi);
1290 bioerror(bp, rv);
1291 biodone(bp);
1292
1293 mutex_enter(&bd->d_iomutex);
1294 bd->d_qactive--;
1295 kstat_runq_exit(bd->d_kiop);
1296 list_remove(&bd->d_runq, xi);
1297 } else {
1298 mutex_enter(&bd->d_iomutex);
1299 }
1300 }
1301
1302 mutex_exit(&bd->d_iomutex);
1303 }
1304
1305 static void
1306 bd_submit(bd_t *bd, bd_xfer_impl_t *xi)
1307 {
1308 mutex_enter(&bd->d_iomutex);
1309 list_insert_tail(&bd->d_waitq, xi);
1310 kstat_waitq_enter(bd->d_kiop);
1311 mutex_exit(&bd->d_iomutex);
1312
1313 bd_sched(bd);
1314 }
1315
1316 static void
1317 bd_runq_exit(bd_xfer_impl_t *xi, int err)
1318 {
1319 bd_t *bd = xi->i_bd;
1320 buf_t *bp = xi->i_bp;
1321
1322 mutex_enter(&bd->d_iomutex);
1323 bd->d_qactive--;
1324 kstat_runq_exit(bd->d_kiop);
1325 list_remove(&bd->d_runq, xi);
1326 mutex_exit(&bd->d_iomutex);
1327
1328 if (err == 0) {
1329 if (bp->b_flags & B_READ) {
1330 bd->d_kiop->reads++;
1331 bd->d_kiop->nread += (bp->b_bcount - xi->i_resid);
1332 } else {
1333 bd->d_kiop->writes++;
1334 bd->d_kiop->nwritten += (bp->b_bcount - xi->i_resid);
1335 }
1336 }
1337 bd_sched(bd);
1338 }
1339
1340 static void
1341 bd_update_state(bd_t *bd)
1342 {
1343 enum dkio_state state;
1344 bd_media_t media;
1345 boolean_t docmlb = B_FALSE;
1346
1347 bzero(&media, sizeof (media));
1348
1349 mutex_enter(&bd->d_statemutex);
1350 if (bd->d_ops.o_media_info(bd->d_private, &media) == 0) {
1351 if ((1U << bd->d_blkshift) != media.m_blksize) {
1352 if ((media.m_blksize < 512) ||
1353 (!ISP2(media.m_blksize)) ||
1354 (P2PHASE(bd->d_maxxfer, media.m_blksize))) {
1355 cmn_err(CE_WARN,
1356 "%s%d: Invalid media block size (%d)",
1357 ddi_driver_name(bd->d_dip),
1358 ddi_get_instance(bd->d_dip),
1359 media.m_blksize);
1360 /*
1361 * We can't use the media, treat it as
1362 * not present.
1363 */
1364 state = DKIO_EJECTED;
1365 bd->d_numblks = 0;
1366 } else {
1367 bd->d_blkshift = ddi_ffs(media.m_blksize) - 1;
1368 bd->d_numblks = media.m_nblks;
1369 bd->d_rdonly = media.m_readonly;
1370 state = DKIO_INSERTED;
1371 }
1372
1373 /* Device size changed */
1374 docmlb = B_TRUE;
1375
1376 } else {
1377 if (bd->d_numblks != media.m_nblks) {
1378 /* Device size changed */
1379 docmlb = B_TRUE;
1380 }
1381 bd->d_numblks = media.m_nblks;
1382 bd->d_rdonly = media.m_readonly;
1383 state = DKIO_INSERTED;
1384 }
1385
1386 } else {
1387 bd->d_numblks = 0;
1388 state = DKIO_EJECTED;
1389 }
1390 if (state != bd->d_state) {
1391 bd->d_state = state;
1392 cv_broadcast(&bd->d_statecv);
1393 docmlb = B_TRUE;
1394 }
1395 mutex_exit(&bd->d_statemutex);
1396
1397 if (docmlb) {
1398 if (state == DKIO_INSERTED) {
1399 (void) cmlb_validate(bd->d_cmlbh, 0, 0);
1400 } else {
1401 cmlb_invalidate(bd->d_cmlbh, 0);
1402 }
1403 }
1404 }
1405
1406 static int
1407 bd_check_state(bd_t *bd, enum dkio_state *state)
1408 {
1409 clock_t when;
1410
1411 for (;;) {
1412
1413 bd_update_state(bd);
1414
1415 mutex_enter(&bd->d_statemutex);
1416
1417 if (bd->d_state != *state) {
1418 *state = bd->d_state;
1419 mutex_exit(&bd->d_statemutex);
1420 break;
1421 }
1422
1423 when = drv_usectohz(1000000);
1424 if (cv_reltimedwait_sig(&bd->d_statecv, &bd->d_statemutex,
1425 when, TR_CLOCK_TICK) == 0) {
1426 mutex_exit(&bd->d_statemutex);
1427 return (EINTR);
1428 }
1429
1430 mutex_exit(&bd->d_statemutex);
1431 }
1432
1433 return (0);
1434 }
1435
1436 static int
1437 bd_flush_write_cache_done(struct buf *bp)
1438 {
1439 struct dk_callback *dc = (void *)bp->b_private;
1440
1441 (*dc->dkc_callback)(dc->dkc_cookie, geterror(bp));
1442 kmem_free(dc, sizeof (*dc));
1443 freerbuf(bp);
1444 return (0);
1445 }
1446
1447 static int
1448 bd_flush_write_cache(bd_t *bd, struct dk_callback *dkc)
1449 {
1450 buf_t *bp;
1451 struct dk_callback *dc;
1452 bd_xfer_impl_t *xi;
1453 int rv;
1454
1455 if (bd->d_ops.o_sync_cache == NULL) {
1456 return (ENOTSUP);
1457 }
1458 if ((bp = getrbuf(KM_SLEEP)) == NULL) {
1459 return (ENOMEM);
1460 }
1461 bp->b_resid = 0;
1462 bp->b_bcount = 0;
1463
1464 xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_sync_cache, KM_SLEEP);
1465 if (xi == NULL) {
1466 rv = geterror(bp);
1467 freerbuf(bp);
1468 return (rv);
1469 }
1470
1471 /* Make an asynchronous flush, but only if there is a callback */
1472 if (dkc != NULL && dkc->dkc_callback != NULL) {
1473 /* Make a private copy of the callback structure */
1474 dc = kmem_alloc(sizeof (*dc), KM_SLEEP);
1475 *dc = *dkc;
1476 bp->b_private = dc;
1477 bp->b_iodone = bd_flush_write_cache_done;
1478
1479 bd_submit(bd, xi);
1480 return (0);
1481 }
1482
1483 /* In case there is no callback, perform a synchronous flush */
1484 bd_submit(bd, xi);
1485 (void) biowait(bp);
1486 rv = geterror(bp);
1487 freerbuf(bp);
1488
1489 return (rv);
1490 }
1491
1492 /*
1493 * Nexus support.
1494 */
1495 int
1496 bd_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t ctlop,
1497 void *arg, void *result)
1498 {
1499 bd_handle_t hdl;
1500
1501 switch (ctlop) {
1502 case DDI_CTLOPS_REPORTDEV:
1503 cmn_err(CE_CONT, "?Block device: %s@%s, %s%d\n",
1504 ddi_node_name(rdip), ddi_get_name_addr(rdip),
1505 ddi_driver_name(rdip), ddi_get_instance(rdip));
1506 return (DDI_SUCCESS);
1507
1508 case DDI_CTLOPS_INITCHILD:
1509 hdl = ddi_get_parent_data((dev_info_t *)arg);
1510 if (hdl == NULL) {
1511 return (DDI_NOT_WELL_FORMED);
1512 }
1513 ddi_set_name_addr((dev_info_t *)arg, hdl->h_addr);
1514 return (DDI_SUCCESS);
1515
1516 case DDI_CTLOPS_UNINITCHILD:
1517 ddi_set_name_addr((dev_info_t *)arg, NULL);
1518 ndi_prop_remove_all((dev_info_t *)arg);
1519 return (DDI_SUCCESS);
1520
1521 default:
1522 return (ddi_ctlops(dip, rdip, ctlop, arg, result));
1523 }
1524 }
1525
1526 /*
1527 * Functions for device drivers.
1528 */
1529 bd_handle_t
1530 bd_alloc_handle(void *private, bd_ops_t *ops, ddi_dma_attr_t *dma, int kmflag)
1531 {
1532 bd_handle_t hdl;
1533
1534 hdl = kmem_zalloc(sizeof (*hdl), kmflag);
1535 if (hdl == NULL)
1536 return (NULL);
1537
1538 /*
1539 * Cheesy versioning handling. We've only appended members into
1540 * bd_ops as we grew from v0 to v1. Since we zalloc hdl, the
1541 * ioctl ops will be NULL anyway. So for the old version, we
1542 * copy over only the v0 elements.
1543 */
1544 switch (ops->o_version) {
1545 case BD_OPS_VERSION_0:
1546 /* Don't copy the last pointer in the structure. */
1547 bcopy(ops, &hdl->h_ops, sizeof (*ops) - sizeof (void *));
1548 break;
1549 case BD_OPS_VERSION_1:
1550 hdl->h_ops = *ops;
1551 break;
1552 default:
1553 kmem_free(hdl, sizeof (*hdl));
1554 cmn_err(CE_WARN, "Unsupported blkdev ops version %d.\n",
1555 ops->o_version);
1556 return (NULL);
1557 /* NOTREACHED */
1558 }
1559 hdl->h_dma = dma;
1560 hdl->h_private = private;
1561
1562 return (hdl);
1563 }
1564
1565 void
1566 bd_free_handle(bd_handle_t hdl)
1567 {
1568 kmem_free(hdl, sizeof (*hdl));
1569 }
1570
1571 int
1572 bd_attach_handle(dev_info_t *dip, bd_handle_t hdl)
1573 {
1574 dev_info_t *child;
1575 bd_drive_t drive;
1576
1577 /* if drivers don't override this, make it assume none */
1578 drive.d_lun = -1;
1579 hdl->h_ops.o_drive_info(hdl->h_private, &drive);
1580
1581 hdl->h_parent = dip;
1582 hdl->h_name = "blkdev";
1583
1584 if (drive.d_lun >= 0) {
1585 (void) snprintf(hdl->h_addr, sizeof (hdl->h_addr), "%X,%X",
1586 drive.d_target, drive.d_lun);
1587 } else {
1588 (void) snprintf(hdl->h_addr, sizeof (hdl->h_addr), "%X",
1589 drive.d_target);
1590 }
1591 if (ndi_devi_alloc(dip, hdl->h_name, (pnode_t)DEVI_SID_NODEID,
1592 &child) != NDI_SUCCESS) {
1593 cmn_err(CE_WARN, "%s%d: unable to allocate node %s@%s",
1594 ddi_driver_name(dip), ddi_get_instance(dip),
1595 "blkdev", hdl->h_addr);
1596 return (DDI_FAILURE);
1597 }
1598
1599 ddi_set_parent_data(child, hdl);
1600 hdl->h_child = child;
1601
1602 if (ndi_devi_online(child, 0) == NDI_FAILURE) {
1603 cmn_err(CE_WARN, "%s%d: failed bringing node %s@%s online",
1604 ddi_driver_name(dip), ddi_get_instance(dip),
1605 hdl->h_name, hdl->h_addr);
1606 (void) ndi_devi_free(child);
1607 return (DDI_FAILURE);
1608 }
1609
1610 return (DDI_SUCCESS);
1611 }
1612
1613 int
1614 bd_detach_handle(bd_handle_t hdl)
1615 {
1616 int circ;
1617 int rv;
1618 char *devnm;
1619
1620 if (hdl->h_child == NULL) {
1621 return (DDI_SUCCESS);
1622 }
1623 ndi_devi_enter(hdl->h_parent, &circ);
1624 if (i_ddi_node_state(hdl->h_child) < DS_INITIALIZED) {
1625 rv = ddi_remove_child(hdl->h_child, 0);
1626 } else {
1627 devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
1628 (void) ddi_deviname(hdl->h_child, devnm);
1629 (void) devfs_clean(hdl->h_parent, devnm + 1, DV_CLEAN_FORCE);
1630 rv = ndi_devi_unconfig_one(hdl->h_parent, devnm + 1, NULL,
1631 NDI_DEVI_REMOVE | NDI_UNCONFIG);
1632 kmem_free(devnm, MAXNAMELEN + 1);
1633 }
1634 if (rv == 0) {
1635 hdl->h_child = NULL;
1636 }
1637
1638 ndi_devi_exit(hdl->h_parent, circ);
1639 return (rv = NDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
1640 }
1641
1642 void
1643 bd_xfer_done(bd_xfer_t *xfer, int err)
1644 {
1645 bd_xfer_impl_t *xi = (void *)xfer;
1646 buf_t *bp = xi->i_bp;
1647 int rv = DDI_SUCCESS;
1648 bd_t *bd = xi->i_bd;
1649 size_t len;
1650
1651 if (err != 0) {
1652 bd_runq_exit(xi, err);
1653
1654 bp->b_resid += xi->i_resid;
1655 bd_xfer_free(xi);
1656 bioerror(bp, err);
1657 biodone(bp);
1658 return;
1659 }
1660
1661 xi->i_cur_win++;
1662 xi->i_resid -= xi->i_len;
1663
1664 if (xi->i_resid == 0) {
1665 /* Job completed succcessfully! */
1666 bd_runq_exit(xi, 0);
1667
1668 bd_xfer_free(xi);
1669 biodone(bp);
1670 return;
1671 }
1672
1673 xi->i_blkno += xi->i_nblks;
1674
1675 if (bd->d_use_dma) {
1676 /* More transfer still pending... advance to next DMA window. */
1677 rv = ddi_dma_getwin(xi->i_dmah, xi->i_cur_win,
1678 &xi->i_offset, &len, &xi->i_dmac, &xi->i_ndmac);
1679 } else {
1680 /* Advance memory window. */
1681 xi->i_kaddr += xi->i_len;
1682 xi->i_offset += xi->i_len;
1683 len = min(bp->b_bcount - xi->i_offset, bd->d_maxxfer);
1684 }
1685
1686
1687 if ((rv != DDI_SUCCESS) ||
1688 (P2PHASE(len, (1U << xi->i_blkshift) != 0))) {
1689 bd_runq_exit(xi, EFAULT);
1690
1691 bp->b_resid += xi->i_resid;
1692 bd_xfer_free(xi);
1693 bioerror(bp, EFAULT);
1694 biodone(bp);
1695 return;
1696 }
1697 xi->i_len = len;
1698 xi->i_nblks = len >> xi->i_blkshift;
1699
1700 /* Submit next window to hardware. */
1701 rv = xi->i_func(bd->d_private, &xi->i_public);
1702 if (rv != 0) {
1703 bd_runq_exit(xi, rv);
1704
1705 bp->b_resid += xi->i_resid;
1706 bd_xfer_free(xi);
1707 bioerror(bp, rv);
1708 biodone(bp);
1709 }
1710 }
1711
1712 void
1713 bd_state_change(bd_handle_t hdl)
1714 {
1715 bd_t *bd;
1716
1717 if ((bd = hdl->h_bd) != NULL) {
1718 bd_update_state(bd);
1719 }
1720 }
1721
1722 void
1723 bd_mod_init(struct dev_ops *devops)
1724 {
1725 static struct bus_ops bd_bus_ops = {
1726 BUSO_REV, /* busops_rev */
1727 nullbusmap, /* bus_map */
1728 NULL, /* bus_get_intrspec (OBSOLETE) */
1729 NULL, /* bus_add_intrspec (OBSOLETE) */
1730 NULL, /* bus_remove_intrspec (OBSOLETE) */
1731 i_ddi_map_fault, /* bus_map_fault */
1732 NULL, /* bus_dma_map (OBSOLETE) */
1733 ddi_dma_allochdl, /* bus_dma_allochdl */
1734 ddi_dma_freehdl, /* bus_dma_freehdl */
1735 ddi_dma_bindhdl, /* bus_dma_bindhdl */
1736 ddi_dma_unbindhdl, /* bus_dma_unbindhdl */
1737 ddi_dma_flush, /* bus_dma_flush */
1738 ddi_dma_win, /* bus_dma_win */
1739 ddi_dma_mctl, /* bus_dma_ctl */
1740 bd_bus_ctl, /* bus_ctl */
1741 ddi_bus_prop_op, /* bus_prop_op */
1742 NULL, /* bus_get_eventcookie */
1743 NULL, /* bus_add_eventcall */
1744 NULL, /* bus_remove_eventcall */
1745 NULL, /* bus_post_event */
1746 NULL, /* bus_intr_ctl (OBSOLETE) */
1747 NULL, /* bus_config */
1748 NULL, /* bus_unconfig */
1749 NULL, /* bus_fm_init */
1750 NULL, /* bus_fm_fini */
1751 NULL, /* bus_fm_access_enter */
1752 NULL, /* bus_fm_access_exit */
1753 NULL, /* bus_power */
1754 NULL, /* bus_intr_op */
1755 };
1756
1757 devops->devo_bus_ops = &bd_bus_ops;
1758
1759 /*
1760 * NB: The device driver is free to supply its own
1761 * character entry device support.
1762 */
1763 }
1764
1765 void
1766 bd_mod_fini(struct dev_ops *devops)
1767 {
1768 devops->devo_bus_ops = NULL;
1769 }