xref: /freebsd/sys/cam/ctl/ctl_backend_block.c (revision 6e563a1b608438504d963c2d7c70e50d2e75af46)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2003 Silicon Graphics International Corp.
5  * Copyright (c) 2009-2011 Spectra Logic Corporation
6  * Copyright (c) 2012,2021 The FreeBSD Foundation
7  * Copyright (c) 2014-2021 Alexander Motin <mav@FreeBSD.org>
8  * All rights reserved.
9  *
10  * Portions of this software were developed by Edward Tomasz Napierala
11  * under sponsorship from the FreeBSD Foundation.
12  *
13  * Portions of this software were developed by Ka Ho Ng <khng@FreeBSD.org>
14  * under sponsorship from the FreeBSD Foundation.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions, and the following disclaimer,
21  *    without modification.
22  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
23  *    substantially similar to the "NO WARRANTY" disclaimer below
24  *    ("Disclaimer") and any redistribution must be conditioned upon
25  *    including a substantially similar Disclaimer requirement for further
26  *    binary redistribution.
27  *
28  * NO WARRANTY
29  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
32  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
37  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
38  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
39  * POSSIBILITY OF SUCH DAMAGES.
40  *
41  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $
42  */
43 /*
44  * CAM Target Layer driver backend for block devices.
45  *
46  * Author: Ken Merry <ken@FreeBSD.org>
47  */
48 #include <sys/cdefs.h>
49 __FBSDID("$FreeBSD$");
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/kernel.h>
54 #include <sys/types.h>
55 #include <sys/kthread.h>
56 #include <sys/bio.h>
57 #include <sys/fcntl.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/mutex.h>
61 #include <sys/condvar.h>
62 #include <sys/malloc.h>
63 #include <sys/conf.h>
64 #include <sys/ioccom.h>
65 #include <sys/queue.h>
66 #include <sys/sbuf.h>
67 #include <sys/endian.h>
68 #include <sys/uio.h>
69 #include <sys/buf.h>
70 #include <sys/taskqueue.h>
71 #include <sys/vnode.h>
72 #include <sys/namei.h>
73 #include <sys/mount.h>
74 #include <sys/disk.h>
75 #include <sys/fcntl.h>
76 #include <sys/filedesc.h>
77 #include <sys/filio.h>
78 #include <sys/proc.h>
79 #include <sys/pcpu.h>
80 #include <sys/module.h>
81 #include <sys/sdt.h>
82 #include <sys/devicestat.h>
83 #include <sys/sysctl.h>
84 #include <sys/nv.h>
85 #include <sys/dnv.h>
86 #include <sys/sx.h>
87 #include <sys/unistd.h>
88 
89 #include <geom/geom.h>
90 
91 #include <cam/cam.h>
92 #include <cam/scsi/scsi_all.h>
93 #include <cam/scsi/scsi_da.h>
94 #include <cam/ctl/ctl_io.h>
95 #include <cam/ctl/ctl.h>
96 #include <cam/ctl/ctl_backend.h>
97 #include <cam/ctl/ctl_ioctl.h>
98 #include <cam/ctl/ctl_ha.h>
99 #include <cam/ctl/ctl_scsi_all.h>
100 #include <cam/ctl/ctl_private.h>
101 #include <cam/ctl/ctl_error.h>
102 
103 /*
104  * The idea here is to allocate enough S/G space to handle at least 1MB I/Os.
105  * On systems with small maxphys it can be 8 128KB segments.  On large systems
106  * it can be up to 8 1MB segments.  I/Os larger than that we'll split.
107  */
108 #define	CTLBLK_MAX_SEGS		8
109 #define	CTLBLK_HALF_SEGS	(CTLBLK_MAX_SEGS / 2)
110 #define	CTLBLK_MIN_SEG		(128 * 1024)
111 #define	CTLBLK_MAX_SEG		MIN(1024 * 1024, MAX(CTLBLK_MIN_SEG, maxphys))
112 #define	CTLBLK_MAX_IO_SIZE	(CTLBLK_MAX_SEG * CTLBLK_MAX_SEGS)
113 
114 #ifdef CTLBLK_DEBUG
115 #define DPRINTF(fmt, args...) \
116     printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
117 #else
118 #define DPRINTF(fmt, args...) do {} while(0)
119 #endif
120 
121 #define PRIV(io)	\
122     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
123 #define ARGS(io)	\
124     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
125 
126 SDT_PROVIDER_DEFINE(cbb);
127 
128 typedef enum {
129 	CTL_BE_BLOCK_LUN_UNCONFIGURED	= 0x01,
130 	CTL_BE_BLOCK_LUN_WAITING	= 0x04,
131 } ctl_be_block_lun_flags;
132 
133 typedef enum {
134 	CTL_BE_BLOCK_NONE,
135 	CTL_BE_BLOCK_DEV,
136 	CTL_BE_BLOCK_FILE
137 } ctl_be_block_type;
138 
139 struct ctl_be_block_filedata {
140 	struct ucred *cred;
141 };
142 
143 union ctl_be_block_bedata {
144 	struct ctl_be_block_filedata file;
145 };
146 
147 struct ctl_be_block_io;
148 struct ctl_be_block_lun;
149 
150 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
151 			       struct ctl_be_block_io *beio);
152 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
153 				  const char *attrname);
154 
155 /*
156  * Backend LUN structure.  There is a 1:1 mapping between a block device
157  * and a backend block LUN, and between a backend block LUN and a CTL LUN.
158  */
159 struct ctl_be_block_lun {
160 	struct ctl_be_lun cbe_lun;		/* Must be first element. */
161 	struct ctl_lun_create_params params;
162 	char *dev_path;
163 	ctl_be_block_type dev_type;
164 	struct vnode *vn;
165 	union ctl_be_block_bedata backend;
166 	cbb_dispatch_t dispatch;
167 	cbb_dispatch_t lun_flush;
168 	cbb_dispatch_t unmap;
169 	cbb_dispatch_t get_lba_status;
170 	cbb_getattr_t getattr;
171 	uint64_t size_blocks;
172 	uint64_t size_bytes;
173 	struct ctl_be_block_softc *softc;
174 	struct devstat *disk_stats;
175 	ctl_be_block_lun_flags flags;
176 	SLIST_ENTRY(ctl_be_block_lun) links;
177 	struct taskqueue *io_taskqueue;
178 	struct task io_task;
179 	int num_threads;
180 	STAILQ_HEAD(, ctl_io_hdr) input_queue;
181 	STAILQ_HEAD(, ctl_io_hdr) config_read_queue;
182 	STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
183 	STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
184 	struct mtx_padalign io_lock;
185 	struct mtx_padalign queue_lock;
186 };
187 
188 /*
189  * Overall softc structure for the block backend module.
190  */
191 struct ctl_be_block_softc {
192 	struct sx			 modify_lock;
193 	struct mtx			 lock;
194 	int				 num_luns;
195 	SLIST_HEAD(, ctl_be_block_lun)	 lun_list;
196 	uma_zone_t			 beio_zone;
197 	uma_zone_t			 bufmin_zone;
198 	uma_zone_t			 bufmax_zone;
199 };
200 
201 static struct ctl_be_block_softc backend_block_softc;
202 
203 /*
204  * Per-I/O information.
205  */
206 struct ctl_be_block_io {
207 	union ctl_io			*io;
208 	struct ctl_sg_entry		sg_segs[CTLBLK_MAX_SEGS];
209 	struct iovec			xiovecs[CTLBLK_MAX_SEGS];
210 	int				refcnt;
211 	int				bio_cmd;
212 	int				two_sglists;
213 	int				num_segs;
214 	int				num_bios_sent;
215 	int				num_bios_done;
216 	int				send_complete;
217 	int				first_error;
218 	uint64_t			first_error_offset;
219 	struct bintime			ds_t0;
220 	devstat_tag_type		ds_tag_type;
221 	devstat_trans_flags		ds_trans_type;
222 	uint64_t			io_len;
223 	uint64_t			io_offset;
224 	int				io_arg;
225 	struct ctl_be_block_softc	*softc;
226 	struct ctl_be_block_lun		*lun;
227 	void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
228 };
229 
230 extern struct ctl_softc *control_softc;
231 
232 static int cbb_num_threads = 32;
233 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
234 	    "CAM Target Layer Block Backend");
235 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
236            &cbb_num_threads, 0, "Number of threads per backing file");
237 
238 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
239 static void ctl_free_beio(struct ctl_be_block_io *beio);
240 static void ctl_complete_beio(struct ctl_be_block_io *beio);
241 static int ctl_be_block_move_done(union ctl_io *io, bool samethr);
242 static void ctl_be_block_biodone(struct bio *bio);
243 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
244 				    struct ctl_be_block_io *beio);
245 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
246 				       struct ctl_be_block_io *beio);
247 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
248 				  struct ctl_be_block_io *beio);
249 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
250 					 const char *attrname);
251 static void ctl_be_block_unmap_file(struct ctl_be_block_lun *be_lun,
252 				    struct ctl_be_block_io *beio);
253 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
254 				   struct ctl_be_block_io *beio);
255 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
256 				   struct ctl_be_block_io *beio);
257 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
258 				      struct ctl_be_block_io *beio);
259 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
260 					 const char *attrname);
261 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
262 				    union ctl_io *io);
263 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
264 				    union ctl_io *io);
265 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
266 				  union ctl_io *io);
267 static void ctl_be_block_worker(void *context, int pending);
268 static int ctl_be_block_submit(union ctl_io *io);
269 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
270 				   int flag, struct thread *td);
271 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
272 				  struct ctl_lun_req *req);
273 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
274 				 struct ctl_lun_req *req);
275 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
276 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun,
277 			     struct ctl_lun_req *req);
278 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
279 			       struct ctl_lun_req *req);
280 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
281 			   struct ctl_lun_req *req);
282 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
283 			   struct ctl_lun_req *req);
284 static void ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun);
285 static int ctl_be_block_config_write(union ctl_io *io);
286 static int ctl_be_block_config_read(union ctl_io *io);
287 static int ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb);
288 static uint64_t ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname);
289 static int ctl_be_block_init(void);
290 static int ctl_be_block_shutdown(void);
291 
292 static struct ctl_backend_driver ctl_be_block_driver =
293 {
294 	.name = "block",
295 	.flags = CTL_BE_FLAG_HAS_CONFIG,
296 	.init = ctl_be_block_init,
297 	.shutdown = ctl_be_block_shutdown,
298 	.data_submit = ctl_be_block_submit,
299 	.config_read = ctl_be_block_config_read,
300 	.config_write = ctl_be_block_config_write,
301 	.ioctl = ctl_be_block_ioctl,
302 	.lun_info = ctl_be_block_lun_info,
303 	.lun_attr = ctl_be_block_lun_attr
304 };
305 
306 MALLOC_DEFINE(M_CTLBLK, "ctlblock", "Memory used for CTL block backend");
307 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
308 
309 static void
310 ctl_alloc_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg,
311     size_t len)
312 {
313 
314 	if (len <= CTLBLK_MIN_SEG) {
315 		sg->addr = uma_zalloc(softc->bufmin_zone, M_WAITOK);
316 	} else {
317 		KASSERT(len <= CTLBLK_MAX_SEG,
318 		    ("Too large alloc %zu > %lu", len, CTLBLK_MAX_SEG));
319 		sg->addr = uma_zalloc(softc->bufmax_zone, M_WAITOK);
320 	}
321 	sg->len = len;
322 }
323 
324 static void
325 ctl_free_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg)
326 {
327 
328 	if (sg->len <= CTLBLK_MIN_SEG) {
329 		uma_zfree(softc->bufmin_zone, sg->addr);
330 	} else {
331 		KASSERT(sg->len <= CTLBLK_MAX_SEG,
332 		    ("Too large free %zu > %lu", sg->len, CTLBLK_MAX_SEG));
333 		uma_zfree(softc->bufmax_zone, sg->addr);
334 	}
335 }
336 
337 static struct ctl_be_block_io *
338 ctl_alloc_beio(struct ctl_be_block_softc *softc)
339 {
340 	struct ctl_be_block_io *beio;
341 
342 	beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO);
343 	beio->softc = softc;
344 	beio->refcnt = 1;
345 	return (beio);
346 }
347 
348 static void
349 ctl_real_free_beio(struct ctl_be_block_io *beio)
350 {
351 	struct ctl_be_block_softc *softc = beio->softc;
352 	int i;
353 
354 	for (i = 0; i < beio->num_segs; i++) {
355 		ctl_free_seg(softc, &beio->sg_segs[i]);
356 
357 		/* For compare we had two equal S/G lists. */
358 		if (beio->two_sglists) {
359 			ctl_free_seg(softc,
360 			    &beio->sg_segs[i + CTLBLK_HALF_SEGS]);
361 		}
362 	}
363 
364 	uma_zfree(softc->beio_zone, beio);
365 }
366 
367 static void
368 ctl_refcnt_beio(void *arg, int diff)
369 {
370 	struct ctl_be_block_io *beio = arg;
371 
372 	if (atomic_fetchadd_int(&beio->refcnt, diff) + diff == 0)
373 		ctl_real_free_beio(beio);
374 }
375 
376 static void
377 ctl_free_beio(struct ctl_be_block_io *beio)
378 {
379 
380 	ctl_refcnt_beio(beio, -1);
381 }
382 
383 static void
384 ctl_complete_beio(struct ctl_be_block_io *beio)
385 {
386 	union ctl_io *io = beio->io;
387 
388 	if (beio->beio_cont != NULL) {
389 		beio->beio_cont(beio);
390 	} else {
391 		ctl_free_beio(beio);
392 		ctl_data_submit_done(io);
393 	}
394 }
395 
396 static size_t
397 cmp(uint8_t *a, uint8_t *b, size_t size)
398 {
399 	size_t i;
400 
401 	for (i = 0; i < size; i++) {
402 		if (a[i] != b[i])
403 			break;
404 	}
405 	return (i);
406 }
407 
408 static void
409 ctl_be_block_compare(union ctl_io *io)
410 {
411 	struct ctl_be_block_io *beio;
412 	uint64_t off, res;
413 	int i;
414 	uint8_t info[8];
415 
416 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
417 	off = 0;
418 	for (i = 0; i < beio->num_segs; i++) {
419 		res = cmp(beio->sg_segs[i].addr,
420 		    beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
421 		    beio->sg_segs[i].len);
422 		off += res;
423 		if (res < beio->sg_segs[i].len)
424 			break;
425 	}
426 	if (i < beio->num_segs) {
427 		scsi_u64to8b(off, info);
428 		ctl_set_sense(&io->scsiio, /*current_error*/ 1,
429 		    /*sense_key*/ SSD_KEY_MISCOMPARE,
430 		    /*asc*/ 0x1D, /*ascq*/ 0x00,
431 		    /*type*/ SSD_ELEM_INFO,
432 		    /*size*/ sizeof(info), /*data*/ &info,
433 		    /*type*/ SSD_ELEM_NONE);
434 	} else
435 		ctl_set_success(&io->scsiio);
436 }
437 
438 static int
439 ctl_be_block_move_done(union ctl_io *io, bool samethr)
440 {
441 	struct ctl_be_block_io *beio;
442 	struct ctl_be_block_lun *be_lun;
443 	struct ctl_lba_len_flags *lbalen;
444 
445 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
446 
447 	DPRINTF("entered\n");
448 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
449 
450 	/*
451 	 * We set status at this point for read and compare commands.
452 	 */
453 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
454 	    (io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE) {
455 		lbalen = ARGS(io);
456 		if (lbalen->flags & CTL_LLF_READ) {
457 			ctl_set_success(&io->scsiio);
458 		} else if (lbalen->flags & CTL_LLF_COMPARE) {
459 			/* We have two data blocks ready for comparison. */
460 			ctl_be_block_compare(io);
461 		}
462 	}
463 
464 	/*
465 	 * If this is a read, or a write with errors, it is done.
466 	 */
467 	if ((beio->bio_cmd == BIO_READ)
468 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
469 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
470 		ctl_complete_beio(beio);
471 		return (0);
472 	}
473 
474 	/*
475 	 * At this point, we have a write and the DMA completed successfully.
476 	 * If we were called synchronously in the original thread then just
477 	 * dispatch, otherwise we now have to queue it to the task queue to
478 	 * execute the backend I/O.  That is because we do blocking
479 	 * memory allocations, and in the file backing case, blocking I/O.
480 	 * This move done routine is generally called in the SIM's
481 	 * interrupt context, and therefore we cannot block.
482 	 */
483 	be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
484 	if (samethr) {
485 		be_lun->dispatch(be_lun, beio);
486 	} else {
487 		mtx_lock(&be_lun->queue_lock);
488 		STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
489 		mtx_unlock(&be_lun->queue_lock);
490 		taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
491 	}
492 	return (0);
493 }
494 
495 static void
496 ctl_be_block_biodone(struct bio *bio)
497 {
498 	struct ctl_be_block_io *beio = bio->bio_caller1;
499 	struct ctl_be_block_lun *be_lun = beio->lun;
500 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
501 	union ctl_io *io;
502 	int error;
503 
504 	io = beio->io;
505 
506 	DPRINTF("entered\n");
507 
508 	error = bio->bio_error;
509 	mtx_lock(&be_lun->io_lock);
510 	if (error != 0 &&
511 	    (beio->first_error == 0 ||
512 	     bio->bio_offset < beio->first_error_offset)) {
513 		beio->first_error = error;
514 		beio->first_error_offset = bio->bio_offset;
515 	}
516 
517 	beio->num_bios_done++;
518 
519 	/*
520 	 * XXX KDM will this cause WITNESS to complain?  Holding a lock
521 	 * during the free might cause it to complain.
522 	 */
523 	g_destroy_bio(bio);
524 
525 	/*
526 	 * If the send complete bit isn't set, or we aren't the last I/O to
527 	 * complete, then we're done.
528 	 */
529 	if ((beio->send_complete == 0)
530 	 || (beio->num_bios_done < beio->num_bios_sent)) {
531 		mtx_unlock(&be_lun->io_lock);
532 		return;
533 	}
534 
535 	/*
536 	 * At this point, we've verified that we are the last I/O to
537 	 * complete, so it's safe to drop the lock.
538 	 */
539 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
540 	    beio->ds_tag_type, beio->ds_trans_type,
541 	    /*now*/ NULL, /*then*/&beio->ds_t0);
542 	mtx_unlock(&be_lun->io_lock);
543 
544 	/*
545 	 * If there are any errors from the backing device, we fail the
546 	 * entire I/O with a medium error.
547 	 */
548 	error = beio->first_error;
549 	if (error != 0) {
550 		if (error == EOPNOTSUPP) {
551 			ctl_set_invalid_opcode(&io->scsiio);
552 		} else if (error == ENOSPC || error == EDQUOT) {
553 			ctl_set_space_alloc_fail(&io->scsiio);
554 		} else if (error == EROFS || error == EACCES) {
555 			ctl_set_hw_write_protected(&io->scsiio);
556 		} else if (beio->bio_cmd == BIO_FLUSH) {
557 			/* XXX KDM is there is a better error here? */
558 			ctl_set_internal_failure(&io->scsiio,
559 						 /*sks_valid*/ 1,
560 						 /*retry_count*/ 0xbad2);
561 		} else {
562 			ctl_set_medium_error(&io->scsiio,
563 			    beio->bio_cmd == BIO_READ);
564 		}
565 		ctl_complete_beio(beio);
566 		return;
567 	}
568 
569 	/*
570 	 * If this is a write, a flush, a delete or verify, we're all done.
571 	 * If this is a read, we can now send the data to the user.
572 	 */
573 	if ((beio->bio_cmd == BIO_WRITE)
574 	 || (beio->bio_cmd == BIO_FLUSH)
575 	 || (beio->bio_cmd == BIO_DELETE)
576 	 || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
577 		ctl_set_success(&io->scsiio);
578 		ctl_complete_beio(beio);
579 	} else {
580 		if ((ARGS(io)->flags & CTL_LLF_READ) &&
581 		    beio->beio_cont == NULL) {
582 			ctl_set_success(&io->scsiio);
583 			if (cbe_lun->serseq >= CTL_LUN_SERSEQ_SOFT)
584 				ctl_serseq_done(io);
585 		}
586 		ctl_datamove(io);
587 	}
588 }
589 
590 static void
591 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
592 			struct ctl_be_block_io *beio)
593 {
594 	union ctl_io *io = beio->io;
595 	struct mount *mountpoint;
596 	int error;
597 
598 	DPRINTF("entered\n");
599 
600 	binuptime(&beio->ds_t0);
601 	devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
602 
603 	(void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
604 
605 	vn_lock(be_lun->vn, vn_lktype_write(mountpoint, be_lun->vn) |
606 	    LK_RETRY);
607 	error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT,
608 	    curthread);
609 	VOP_UNLOCK(be_lun->vn);
610 
611 	vn_finished_write(mountpoint);
612 
613 	mtx_lock(&be_lun->io_lock);
614 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
615 	    beio->ds_tag_type, beio->ds_trans_type,
616 	    /*now*/ NULL, /*then*/&beio->ds_t0);
617 	mtx_unlock(&be_lun->io_lock);
618 
619 	if (error == 0)
620 		ctl_set_success(&io->scsiio);
621 	else {
622 		/* XXX KDM is there is a better error here? */
623 		ctl_set_internal_failure(&io->scsiio,
624 					 /*sks_valid*/ 1,
625 					 /*retry_count*/ 0xbad1);
626 	}
627 
628 	ctl_complete_beio(beio);
629 }
630 
631 SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t");
632 SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t");
633 SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t");
634 SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t");
635 
636 static void
637 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
638 			   struct ctl_be_block_io *beio)
639 {
640 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
641 	struct ctl_be_block_filedata *file_data;
642 	union ctl_io *io;
643 	struct uio xuio;
644 	struct iovec *xiovec;
645 	size_t s;
646 	int error, flags, i;
647 
648 	DPRINTF("entered\n");
649 
650 	file_data = &be_lun->backend.file;
651 	io = beio->io;
652 	flags = 0;
653 	if (ARGS(io)->flags & CTL_LLF_DPO)
654 		flags |= IO_DIRECT;
655 	if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
656 		flags |= IO_SYNC;
657 
658 	bzero(&xuio, sizeof(xuio));
659 	if (beio->bio_cmd == BIO_READ) {
660 		SDT_PROBE0(cbb, , read, file_start);
661 		xuio.uio_rw = UIO_READ;
662 	} else {
663 		SDT_PROBE0(cbb, , write, file_start);
664 		xuio.uio_rw = UIO_WRITE;
665 	}
666 	xuio.uio_offset = beio->io_offset;
667 	xuio.uio_resid = beio->io_len;
668 	xuio.uio_segflg = UIO_SYSSPACE;
669 	xuio.uio_iov = beio->xiovecs;
670 	xuio.uio_iovcnt = beio->num_segs;
671 	xuio.uio_td = curthread;
672 
673 	for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
674 		xiovec->iov_base = beio->sg_segs[i].addr;
675 		xiovec->iov_len = beio->sg_segs[i].len;
676 	}
677 
678 	binuptime(&beio->ds_t0);
679 	devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
680 
681 	if (beio->bio_cmd == BIO_READ) {
682 		vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
683 
684 		if (beio->beio_cont == NULL &&
685 		    cbe_lun->serseq == CTL_LUN_SERSEQ_SOFT)
686 			ctl_serseq_done(io);
687 		/*
688 		 * UFS pays attention to IO_DIRECT for reads.  If the
689 		 * DIRECTIO option is configured into the kernel, it calls
690 		 * ffs_rawread().  But that only works for single-segment
691 		 * uios with user space addresses.  In our case, with a
692 		 * kernel uio, it still reads into the buffer cache, but it
693 		 * will just try to release the buffer from the cache later
694 		 * on in ffs_read().
695 		 *
696 		 * ZFS does not pay attention to IO_DIRECT for reads.
697 		 *
698 		 * UFS does not pay attention to IO_SYNC for reads.
699 		 *
700 		 * ZFS pays attention to IO_SYNC (which translates into the
701 		 * Solaris define FRSYNC for zfs_read()) for reads.  It
702 		 * attempts to sync the file before reading.
703 		 */
704 		error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
705 
706 		VOP_UNLOCK(be_lun->vn);
707 		SDT_PROBE0(cbb, , read, file_done);
708 		if (error == 0 && xuio.uio_resid > 0) {
709 			/*
710 			 * If we red less then requested (EOF), then
711 			 * we should clean the rest of the buffer.
712 			 */
713 			s = beio->io_len - xuio.uio_resid;
714 			for (i = 0; i < beio->num_segs; i++) {
715 				if (s >= beio->sg_segs[i].len) {
716 					s -= beio->sg_segs[i].len;
717 					continue;
718 				}
719 				bzero((uint8_t *)beio->sg_segs[i].addr + s,
720 				    beio->sg_segs[i].len - s);
721 				s = 0;
722 			}
723 		}
724 	} else {
725 		struct mount *mountpoint;
726 
727 		(void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
728 		vn_lock(be_lun->vn, vn_lktype_write(mountpoint,
729 		    be_lun->vn) | LK_RETRY);
730 
731 		/*
732 		 * UFS pays attention to IO_DIRECT for writes.  The write
733 		 * is done asynchronously.  (Normally the write would just
734 		 * get put into cache.
735 		 *
736 		 * UFS pays attention to IO_SYNC for writes.  It will
737 		 * attempt to write the buffer out synchronously if that
738 		 * flag is set.
739 		 *
740 		 * ZFS does not pay attention to IO_DIRECT for writes.
741 		 *
742 		 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
743 		 * for writes.  It will flush the transaction from the
744 		 * cache before returning.
745 		 */
746 		error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
747 		VOP_UNLOCK(be_lun->vn);
748 
749 		vn_finished_write(mountpoint);
750 		SDT_PROBE0(cbb, , write, file_done);
751         }
752 
753 	mtx_lock(&be_lun->io_lock);
754 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
755 	    beio->ds_tag_type, beio->ds_trans_type,
756 	    /*now*/ NULL, /*then*/&beio->ds_t0);
757 	mtx_unlock(&be_lun->io_lock);
758 
759 	/*
760 	 * If we got an error, set the sense data to "MEDIUM ERROR" and
761 	 * return the I/O to the user.
762 	 */
763 	if (error != 0) {
764 		if (error == ENOSPC || error == EDQUOT) {
765 			ctl_set_space_alloc_fail(&io->scsiio);
766 		} else if (error == EROFS || error == EACCES) {
767 			ctl_set_hw_write_protected(&io->scsiio);
768 		} else {
769 			ctl_set_medium_error(&io->scsiio,
770 			    beio->bio_cmd == BIO_READ);
771 		}
772 		ctl_complete_beio(beio);
773 		return;
774 	}
775 
776 	/*
777 	 * If this is a write or a verify, we're all done.
778 	 * If this is a read, we can now send the data to the user.
779 	 */
780 	if ((beio->bio_cmd == BIO_WRITE) ||
781 	    (ARGS(io)->flags & CTL_LLF_VERIFY)) {
782 		ctl_set_success(&io->scsiio);
783 		ctl_complete_beio(beio);
784 	} else {
785 		if ((ARGS(io)->flags & CTL_LLF_READ) &&
786 		    beio->beio_cont == NULL) {
787 			ctl_set_success(&io->scsiio);
788 			if (cbe_lun->serseq > CTL_LUN_SERSEQ_SOFT)
789 				ctl_serseq_done(io);
790 		}
791 		ctl_datamove(io);
792 	}
793 }
794 
795 static void
796 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
797 			struct ctl_be_block_io *beio)
798 {
799 	union ctl_io *io = beio->io;
800 	struct ctl_lba_len_flags *lbalen = ARGS(io);
801 	struct scsi_get_lba_status_data *data;
802 	off_t roff, off;
803 	int error, status;
804 
805 	DPRINTF("entered\n");
806 
807 	off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
808 	vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
809 	error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
810 	    0, curthread->td_ucred, curthread);
811 	if (error == 0 && off > roff)
812 		status = 0;	/* mapped up to off */
813 	else {
814 		error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
815 		    0, curthread->td_ucred, curthread);
816 		if (error == 0 && off > roff)
817 			status = 1;	/* deallocated up to off */
818 		else {
819 			status = 0;	/* unknown up to the end */
820 			off = be_lun->size_bytes;
821 		}
822 	}
823 	VOP_UNLOCK(be_lun->vn);
824 
825 	data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
826 	scsi_u64to8b(lbalen->lba, data->descr[0].addr);
827 	scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
828 	    lbalen->lba), data->descr[0].length);
829 	data->descr[0].status = status;
830 
831 	ctl_complete_beio(beio);
832 }
833 
834 static uint64_t
835 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
836 {
837 	struct vattr		vattr;
838 	struct statfs		statfs;
839 	uint64_t		val;
840 	int			error;
841 
842 	val = UINT64_MAX;
843 	if (be_lun->vn == NULL)
844 		return (val);
845 	vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
846 	if (strcmp(attrname, "blocksused") == 0) {
847 		error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
848 		if (error == 0)
849 			val = vattr.va_bytes / be_lun->cbe_lun.blocksize;
850 	}
851 	if (strcmp(attrname, "blocksavail") == 0 &&
852 	    !VN_IS_DOOMED(be_lun->vn)) {
853 		error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
854 		if (error == 0)
855 			val = statfs.f_bavail * statfs.f_bsize /
856 			    be_lun->cbe_lun.blocksize;
857 	}
858 	VOP_UNLOCK(be_lun->vn);
859 	return (val);
860 }
861 
862 static void
863 ctl_be_block_unmap_file(struct ctl_be_block_lun *be_lun,
864 		        struct ctl_be_block_io *beio)
865 {
866 	struct ctl_be_block_filedata *file_data;
867 	union ctl_io *io;
868 	struct ctl_ptr_len_flags *ptrlen;
869 	struct scsi_unmap_desc *buf, *end;
870 	struct mount *mp;
871 	off_t off, len;
872 	int error;
873 
874 	io = beio->io;
875 	file_data = &be_lun->backend.file;
876 	mp = NULL;
877 	error = 0;
878 
879 	binuptime(&beio->ds_t0);
880 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
881 
882 	(void)vn_start_write(be_lun->vn, &mp, V_WAIT);
883 	vn_lock(be_lun->vn, vn_lktype_write(mp, be_lun->vn) | LK_RETRY);
884 	if (beio->io_offset == -1) {
885 		beio->io_len = 0;
886 		ptrlen = (struct ctl_ptr_len_flags *)
887 		    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
888 		buf = (struct scsi_unmap_desc *)ptrlen->ptr;
889 		end = buf + ptrlen->len / sizeof(*buf);
890 		for (; buf < end; buf++) {
891 			off = (off_t)scsi_8btou64(buf->lba) *
892 			    be_lun->cbe_lun.blocksize;
893 			len = (off_t)scsi_4btoul(buf->length) *
894 			    be_lun->cbe_lun.blocksize;
895 			beio->io_len += len;
896 			error = vn_deallocate(be_lun->vn, &off, &len,
897 			    0, IO_NOMACCHECK | IO_NODELOCKED, file_data->cred,
898 			    NOCRED);
899 			if (error != 0)
900 				break;
901 		}
902 	} else {
903 		/* WRITE_SAME */
904 		off = beio->io_offset;
905 		len = beio->io_len;
906 		error = vn_deallocate(be_lun->vn, &off, &len, 0,
907 		    IO_NOMACCHECK | IO_NODELOCKED, file_data->cred, NOCRED);
908 	}
909 	VOP_UNLOCK(be_lun->vn);
910 	vn_finished_write(mp);
911 
912 	mtx_lock(&be_lun->io_lock);
913 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
914 	    beio->ds_tag_type, beio->ds_trans_type,
915 	    /*now*/ NULL, /*then*/&beio->ds_t0);
916 	mtx_unlock(&be_lun->io_lock);
917 
918 	/*
919 	 * If we got an error, set the sense data to "MEDIUM ERROR" and
920 	 * return the I/O to the user.
921 	 */
922 	switch (error) {
923 	case 0:
924 		ctl_set_success(&io->scsiio);
925 		break;
926 	case ENOSPC:
927 	case EDQUOT:
928 		ctl_set_space_alloc_fail(&io->scsiio);
929 		break;
930 	case EROFS:
931 	case EACCES:
932 		ctl_set_hw_write_protected(&io->scsiio);
933 		break;
934 	default:
935 		ctl_set_medium_error(&io->scsiio, false);
936 	}
937 	ctl_complete_beio(beio);
938 }
939 
940 static void
941 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
942 			   struct ctl_be_block_io *beio)
943 {
944 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
945 	union ctl_io *io;
946 	struct cdevsw *csw;
947 	struct cdev *dev;
948 	struct uio xuio;
949 	struct iovec *xiovec;
950 	int error, flags, i, ref;
951 
952 	DPRINTF("entered\n");
953 
954 	io = beio->io;
955 	flags = 0;
956 	if (ARGS(io)->flags & CTL_LLF_DPO)
957 		flags |= IO_DIRECT;
958 	if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
959 		flags |= IO_SYNC;
960 
961 	bzero(&xuio, sizeof(xuio));
962 	if (beio->bio_cmd == BIO_READ) {
963 		SDT_PROBE0(cbb, , read, file_start);
964 		xuio.uio_rw = UIO_READ;
965 	} else {
966 		SDT_PROBE0(cbb, , write, file_start);
967 		xuio.uio_rw = UIO_WRITE;
968 	}
969 	xuio.uio_offset = beio->io_offset;
970 	xuio.uio_resid = beio->io_len;
971 	xuio.uio_segflg = UIO_SYSSPACE;
972 	xuio.uio_iov = beio->xiovecs;
973 	xuio.uio_iovcnt = beio->num_segs;
974 	xuio.uio_td = curthread;
975 
976 	for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
977 		xiovec->iov_base = beio->sg_segs[i].addr;
978 		xiovec->iov_len = beio->sg_segs[i].len;
979 	}
980 
981 	binuptime(&beio->ds_t0);
982 	devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
983 
984 	csw = devvn_refthread(be_lun->vn, &dev, &ref);
985 	if (csw) {
986 		if (beio->bio_cmd == BIO_READ) {
987 			if (beio->beio_cont == NULL &&
988 			    cbe_lun->serseq == CTL_LUN_SERSEQ_SOFT)
989 				ctl_serseq_done(io);
990 			error = csw->d_read(dev, &xuio, flags);
991 		} else
992 			error = csw->d_write(dev, &xuio, flags);
993 		dev_relthread(dev, ref);
994 	} else
995 		error = ENXIO;
996 
997 	if (beio->bio_cmd == BIO_READ)
998 		SDT_PROBE0(cbb, , read, file_done);
999 	else
1000 		SDT_PROBE0(cbb, , write, file_done);
1001 
1002 	mtx_lock(&be_lun->io_lock);
1003 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
1004 	    beio->ds_tag_type, beio->ds_trans_type,
1005 	    /*now*/ NULL, /*then*/&beio->ds_t0);
1006 	mtx_unlock(&be_lun->io_lock);
1007 
1008 	/*
1009 	 * If we got an error, set the sense data to "MEDIUM ERROR" and
1010 	 * return the I/O to the user.
1011 	 */
1012 	if (error != 0) {
1013 		if (error == ENOSPC || error == EDQUOT) {
1014 			ctl_set_space_alloc_fail(&io->scsiio);
1015 		} else if (error == EROFS || error == EACCES) {
1016 			ctl_set_hw_write_protected(&io->scsiio);
1017 		} else {
1018 			ctl_set_medium_error(&io->scsiio,
1019 			    beio->bio_cmd == BIO_READ);
1020 		}
1021 		ctl_complete_beio(beio);
1022 		return;
1023 	}
1024 
1025 	/*
1026 	 * If this is a write or a verify, we're all done.
1027 	 * If this is a read, we can now send the data to the user.
1028 	 */
1029 	if ((beio->bio_cmd == BIO_WRITE) ||
1030 	    (ARGS(io)->flags & CTL_LLF_VERIFY)) {
1031 		ctl_set_success(&io->scsiio);
1032 		ctl_complete_beio(beio);
1033 	} else {
1034 		if ((ARGS(io)->flags & CTL_LLF_READ) &&
1035 		    beio->beio_cont == NULL) {
1036 			ctl_set_success(&io->scsiio);
1037 			if (cbe_lun->serseq > CTL_LUN_SERSEQ_SOFT)
1038 				ctl_serseq_done(io);
1039 		}
1040 		ctl_datamove(io);
1041 	}
1042 }
1043 
1044 static void
1045 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
1046 			struct ctl_be_block_io *beio)
1047 {
1048 	union ctl_io *io = beio->io;
1049 	struct cdevsw *csw;
1050 	struct cdev *dev;
1051 	struct ctl_lba_len_flags *lbalen = ARGS(io);
1052 	struct scsi_get_lba_status_data *data;
1053 	off_t roff, off;
1054 	int error, ref, status;
1055 
1056 	DPRINTF("entered\n");
1057 
1058 	csw = devvn_refthread(be_lun->vn, &dev, &ref);
1059 	if (csw == NULL) {
1060 		status = 0;	/* unknown up to the end */
1061 		off = be_lun->size_bytes;
1062 		goto done;
1063 	}
1064 	off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
1065 	error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD,
1066 	    curthread);
1067 	if (error == 0 && off > roff)
1068 		status = 0;	/* mapped up to off */
1069 	else {
1070 		error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD,
1071 		    curthread);
1072 		if (error == 0 && off > roff)
1073 			status = 1;	/* deallocated up to off */
1074 		else {
1075 			status = 0;	/* unknown up to the end */
1076 			off = be_lun->size_bytes;
1077 		}
1078 	}
1079 	dev_relthread(dev, ref);
1080 
1081 done:
1082 	data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
1083 	scsi_u64to8b(lbalen->lba, data->descr[0].addr);
1084 	scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
1085 	    lbalen->lba), data->descr[0].length);
1086 	data->descr[0].status = status;
1087 
1088 	ctl_complete_beio(beio);
1089 }
1090 
1091 static void
1092 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
1093 		       struct ctl_be_block_io *beio)
1094 {
1095 	struct bio *bio;
1096 	struct cdevsw *csw;
1097 	struct cdev *dev;
1098 	int ref;
1099 
1100 	DPRINTF("entered\n");
1101 
1102 	/* This can't fail, it's a blocking allocation. */
1103 	bio = g_alloc_bio();
1104 
1105 	bio->bio_cmd	    = BIO_FLUSH;
1106 	bio->bio_offset	    = 0;
1107 	bio->bio_data	    = 0;
1108 	bio->bio_done	    = ctl_be_block_biodone;
1109 	bio->bio_caller1    = beio;
1110 	bio->bio_pblkno	    = 0;
1111 
1112 	/*
1113 	 * We don't need to acquire the LUN lock here, because we are only
1114 	 * sending one bio, and so there is no other context to synchronize
1115 	 * with.
1116 	 */
1117 	beio->num_bios_sent = 1;
1118 	beio->send_complete = 1;
1119 
1120 	binuptime(&beio->ds_t0);
1121 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1122 
1123 	csw = devvn_refthread(be_lun->vn, &dev, &ref);
1124 	if (csw) {
1125 		bio->bio_dev = dev;
1126 		csw->d_strategy(bio);
1127 		dev_relthread(dev, ref);
1128 	} else {
1129 		bio->bio_error = ENXIO;
1130 		ctl_be_block_biodone(bio);
1131 	}
1132 }
1133 
1134 static void
1135 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1136 		       struct ctl_be_block_io *beio,
1137 		       uint64_t off, uint64_t len, int last)
1138 {
1139 	struct bio *bio;
1140 	uint64_t maxlen;
1141 	struct cdevsw *csw;
1142 	struct cdev *dev;
1143 	int ref;
1144 
1145 	csw = devvn_refthread(be_lun->vn, &dev, &ref);
1146 	maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize);
1147 	while (len > 0) {
1148 		bio = g_alloc_bio();
1149 		bio->bio_cmd	    = BIO_DELETE;
1150 		bio->bio_dev	    = dev;
1151 		bio->bio_offset	    = off;
1152 		bio->bio_length	    = MIN(len, maxlen);
1153 		bio->bio_data	    = 0;
1154 		bio->bio_done	    = ctl_be_block_biodone;
1155 		bio->bio_caller1    = beio;
1156 		bio->bio_pblkno     = off / be_lun->cbe_lun.blocksize;
1157 
1158 		off += bio->bio_length;
1159 		len -= bio->bio_length;
1160 
1161 		mtx_lock(&be_lun->io_lock);
1162 		beio->num_bios_sent++;
1163 		if (last && len == 0)
1164 			beio->send_complete = 1;
1165 		mtx_unlock(&be_lun->io_lock);
1166 
1167 		if (csw) {
1168 			csw->d_strategy(bio);
1169 		} else {
1170 			bio->bio_error = ENXIO;
1171 			ctl_be_block_biodone(bio);
1172 		}
1173 	}
1174 	if (csw)
1175 		dev_relthread(dev, ref);
1176 }
1177 
1178 static void
1179 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1180 		       struct ctl_be_block_io *beio)
1181 {
1182 	union ctl_io *io;
1183 	struct ctl_ptr_len_flags *ptrlen;
1184 	struct scsi_unmap_desc *buf, *end;
1185 	uint64_t len;
1186 
1187 	io = beio->io;
1188 
1189 	DPRINTF("entered\n");
1190 
1191 	binuptime(&beio->ds_t0);
1192 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1193 
1194 	if (beio->io_offset == -1) {
1195 		beio->io_len = 0;
1196 		ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1197 		buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1198 		end = buf + ptrlen->len / sizeof(*buf);
1199 		for (; buf < end; buf++) {
1200 			len = (uint64_t)scsi_4btoul(buf->length) *
1201 			    be_lun->cbe_lun.blocksize;
1202 			beio->io_len += len;
1203 			ctl_be_block_unmap_dev_range(be_lun, beio,
1204 			    scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize,
1205 			    len, (end - buf < 2) ? TRUE : FALSE);
1206 		}
1207 	} else
1208 		ctl_be_block_unmap_dev_range(be_lun, beio,
1209 		    beio->io_offset, beio->io_len, TRUE);
1210 }
1211 
1212 static void
1213 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1214 			  struct ctl_be_block_io *beio)
1215 {
1216 	TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1217 	struct bio *bio;
1218 	struct cdevsw *csw;
1219 	struct cdev *dev;
1220 	off_t cur_offset;
1221 	int i, max_iosize, ref;
1222 
1223 	DPRINTF("entered\n");
1224 	csw = devvn_refthread(be_lun->vn, &dev, &ref);
1225 
1226 	/*
1227 	 * We have to limit our I/O size to the maximum supported by the
1228 	 * backend device.
1229 	 */
1230 	if (csw) {
1231 		max_iosize = dev->si_iosize_max;
1232 		if (max_iosize <= 0)
1233 			max_iosize = DFLTPHYS;
1234 	} else
1235 		max_iosize = maxphys;
1236 
1237 	cur_offset = beio->io_offset;
1238 	for (i = 0; i < beio->num_segs; i++) {
1239 		size_t cur_size;
1240 		uint8_t *cur_ptr;
1241 
1242 		cur_size = beio->sg_segs[i].len;
1243 		cur_ptr = beio->sg_segs[i].addr;
1244 
1245 		while (cur_size > 0) {
1246 			/* This can't fail, it's a blocking allocation. */
1247 			bio = g_alloc_bio();
1248 
1249 			KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1250 
1251 			bio->bio_cmd = beio->bio_cmd;
1252 			bio->bio_dev = dev;
1253 			bio->bio_caller1 = beio;
1254 			bio->bio_length = min(cur_size, max_iosize);
1255 			bio->bio_offset = cur_offset;
1256 			bio->bio_data = cur_ptr;
1257 			bio->bio_done = ctl_be_block_biodone;
1258 			bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize;
1259 
1260 			cur_offset += bio->bio_length;
1261 			cur_ptr += bio->bio_length;
1262 			cur_size -= bio->bio_length;
1263 
1264 			TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1265 			beio->num_bios_sent++;
1266 		}
1267 	}
1268 	beio->send_complete = 1;
1269 	binuptime(&beio->ds_t0);
1270 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1271 
1272 	/*
1273 	 * Fire off all allocated requests!
1274 	 */
1275 	while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1276 		TAILQ_REMOVE(&queue, bio, bio_queue);
1277 		if (csw)
1278 			csw->d_strategy(bio);
1279 		else {
1280 			bio->bio_error = ENXIO;
1281 			ctl_be_block_biodone(bio);
1282 		}
1283 	}
1284 	if (csw)
1285 		dev_relthread(dev, ref);
1286 }
1287 
1288 static uint64_t
1289 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1290 {
1291 	struct diocgattr_arg	arg;
1292 	struct cdevsw *csw;
1293 	struct cdev *dev;
1294 	int error, ref;
1295 
1296 	csw = devvn_refthread(be_lun->vn, &dev, &ref);
1297 	if (csw == NULL)
1298 		return (UINT64_MAX);
1299 	strlcpy(arg.name, attrname, sizeof(arg.name));
1300 	arg.len = sizeof(arg.value.off);
1301 	if (csw->d_ioctl) {
1302 		error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
1303 		    curthread);
1304 	} else
1305 		error = ENODEV;
1306 	dev_relthread(dev, ref);
1307 	if (error != 0)
1308 		return (UINT64_MAX);
1309 	return (arg.value.off);
1310 }
1311 
1312 static void
1313 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun,
1314 			    union ctl_io *io)
1315 {
1316 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1317 	struct ctl_be_block_io *beio;
1318 	struct ctl_lba_len_flags *lbalen;
1319 
1320 	DPRINTF("entered\n");
1321 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1322 	lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1323 
1324 	beio->io_len = lbalen->len * cbe_lun->blocksize;
1325 	beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1326 	beio->io_arg = (lbalen->flags & SSC_IMMED) != 0;
1327 	beio->bio_cmd = BIO_FLUSH;
1328 	beio->ds_trans_type = DEVSTAT_NO_DATA;
1329 	DPRINTF("SYNC\n");
1330 	be_lun->lun_flush(be_lun, beio);
1331 }
1332 
1333 static void
1334 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1335 {
1336 	union ctl_io *io;
1337 
1338 	io = beio->io;
1339 	ctl_free_beio(beio);
1340 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1341 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1342 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1343 		ctl_config_write_done(io);
1344 		return;
1345 	}
1346 
1347 	ctl_be_block_config_write(io);
1348 }
1349 
1350 static void
1351 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1352 			    union ctl_io *io)
1353 {
1354 	struct ctl_be_block_softc *softc = be_lun->softc;
1355 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1356 	struct ctl_be_block_io *beio;
1357 	struct ctl_lba_len_flags *lbalen;
1358 	uint64_t len_left, lba;
1359 	uint32_t pb, pbo, adj;
1360 	int i, seglen;
1361 	uint8_t *buf, *end;
1362 
1363 	DPRINTF("entered\n");
1364 
1365 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1366 	lbalen = ARGS(io);
1367 
1368 	if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1369 	    (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1370 		ctl_free_beio(beio);
1371 		ctl_set_invalid_field(&io->scsiio,
1372 				      /*sks_valid*/ 1,
1373 				      /*command*/ 1,
1374 				      /*field*/ 1,
1375 				      /*bit_valid*/ 0,
1376 				      /*bit*/ 0);
1377 		ctl_config_write_done(io);
1378 		return;
1379 	}
1380 
1381 	if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1382 		beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1383 		beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1384 		beio->bio_cmd = BIO_DELETE;
1385 		beio->ds_trans_type = DEVSTAT_FREE;
1386 
1387 		be_lun->unmap(be_lun, beio);
1388 		return;
1389 	}
1390 
1391 	beio->bio_cmd = BIO_WRITE;
1392 	beio->ds_trans_type = DEVSTAT_WRITE;
1393 
1394 	DPRINTF("WRITE SAME at LBA %jx len %u\n",
1395 	       (uintmax_t)lbalen->lba, lbalen->len);
1396 
1397 	pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1398 	if (be_lun->cbe_lun.pblockoff > 0)
1399 		pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1400 	else
1401 		pbo = 0;
1402 	len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1403 	for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1404 		/*
1405 		 * Setup the S/G entry for this chunk.
1406 		 */
1407 		seglen = MIN(CTLBLK_MAX_SEG, len_left);
1408 		if (pb > cbe_lun->blocksize) {
1409 			adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1410 			    seglen - pbo) % pb;
1411 			if (seglen > adj)
1412 				seglen -= adj;
1413 			else
1414 				seglen -= seglen % cbe_lun->blocksize;
1415 		} else
1416 			seglen -= seglen % cbe_lun->blocksize;
1417 		ctl_alloc_seg(softc, &beio->sg_segs[i], seglen);
1418 
1419 		DPRINTF("segment %d addr %p len %zd\n", i,
1420 			beio->sg_segs[i].addr, beio->sg_segs[i].len);
1421 
1422 		beio->num_segs++;
1423 		len_left -= seglen;
1424 
1425 		buf = beio->sg_segs[i].addr;
1426 		end = buf + seglen;
1427 		for (; buf < end; buf += cbe_lun->blocksize) {
1428 			if (lbalen->flags & SWS_NDOB) {
1429 				memset(buf, 0, cbe_lun->blocksize);
1430 			} else {
1431 				memcpy(buf, io->scsiio.kern_data_ptr,
1432 				    cbe_lun->blocksize);
1433 			}
1434 			if (lbalen->flags & SWS_LBDATA)
1435 				scsi_ulto4b(lbalen->lba + lba, buf);
1436 			lba++;
1437 		}
1438 	}
1439 
1440 	beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1441 	beio->io_len = lba * cbe_lun->blocksize;
1442 
1443 	/* We can not do all in one run. Correct and schedule rerun. */
1444 	if (len_left > 0) {
1445 		lbalen->lba += lba;
1446 		lbalen->len -= lba;
1447 		beio->beio_cont = ctl_be_block_cw_done_ws;
1448 	}
1449 
1450 	be_lun->dispatch(be_lun, beio);
1451 }
1452 
1453 static void
1454 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1455 			    union ctl_io *io)
1456 {
1457 	struct ctl_be_block_io *beio;
1458 	struct ctl_ptr_len_flags *ptrlen;
1459 
1460 	DPRINTF("entered\n");
1461 
1462 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1463 	ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1464 
1465 	if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1466 		ctl_free_beio(beio);
1467 		ctl_set_invalid_field(&io->scsiio,
1468 				      /*sks_valid*/ 0,
1469 				      /*command*/ 1,
1470 				      /*field*/ 0,
1471 				      /*bit_valid*/ 0,
1472 				      /*bit*/ 0);
1473 		ctl_config_write_done(io);
1474 		return;
1475 	}
1476 
1477 	beio->io_len = 0;
1478 	beio->io_offset = -1;
1479 	beio->bio_cmd = BIO_DELETE;
1480 	beio->ds_trans_type = DEVSTAT_FREE;
1481 	DPRINTF("UNMAP\n");
1482 	be_lun->unmap(be_lun, beio);
1483 }
1484 
1485 static void
1486 ctl_be_block_cr_done(struct ctl_be_block_io *beio)
1487 {
1488 	union ctl_io *io;
1489 
1490 	io = beio->io;
1491 	ctl_free_beio(beio);
1492 	ctl_config_read_done(io);
1493 }
1494 
1495 static void
1496 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1497 			 union ctl_io *io)
1498 {
1499 	struct ctl_be_block_io *beio;
1500 	struct ctl_be_block_softc *softc;
1501 
1502 	DPRINTF("entered\n");
1503 
1504 	softc = be_lun->softc;
1505 	beio = ctl_alloc_beio(softc);
1506 	beio->io = io;
1507 	beio->lun = be_lun;
1508 	beio->beio_cont = ctl_be_block_cr_done;
1509 	PRIV(io)->ptr = (void *)beio;
1510 
1511 	switch (io->scsiio.cdb[0]) {
1512 	case SERVICE_ACTION_IN:		/* GET LBA STATUS */
1513 		beio->bio_cmd = -1;
1514 		beio->ds_trans_type = DEVSTAT_NO_DATA;
1515 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1516 		beio->io_len = 0;
1517 		if (be_lun->get_lba_status)
1518 			be_lun->get_lba_status(be_lun, beio);
1519 		else
1520 			ctl_be_block_cr_done(beio);
1521 		break;
1522 	default:
1523 		panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1524 		break;
1525 	}
1526 }
1527 
1528 static void
1529 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1530 {
1531 	union ctl_io *io;
1532 
1533 	io = beio->io;
1534 	ctl_free_beio(beio);
1535 	ctl_config_write_done(io);
1536 }
1537 
1538 static void
1539 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1540 			 union ctl_io *io)
1541 {
1542 	struct ctl_be_block_io *beio;
1543 	struct ctl_be_block_softc *softc;
1544 
1545 	DPRINTF("entered\n");
1546 
1547 	softc = be_lun->softc;
1548 	beio = ctl_alloc_beio(softc);
1549 	beio->io = io;
1550 	beio->lun = be_lun;
1551 	beio->beio_cont = ctl_be_block_cw_done;
1552 	switch (io->scsiio.tag_type) {
1553 	case CTL_TAG_ORDERED:
1554 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1555 		break;
1556 	case CTL_TAG_HEAD_OF_QUEUE:
1557 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1558 		break;
1559 	case CTL_TAG_UNTAGGED:
1560 	case CTL_TAG_SIMPLE:
1561 	case CTL_TAG_ACA:
1562 	default:
1563 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1564 		break;
1565 	}
1566 	PRIV(io)->ptr = (void *)beio;
1567 
1568 	switch (io->scsiio.cdb[0]) {
1569 	case SYNCHRONIZE_CACHE:
1570 	case SYNCHRONIZE_CACHE_16:
1571 		ctl_be_block_cw_dispatch_sync(be_lun, io);
1572 		break;
1573 	case WRITE_SAME_10:
1574 	case WRITE_SAME_16:
1575 		ctl_be_block_cw_dispatch_ws(be_lun, io);
1576 		break;
1577 	case UNMAP:
1578 		ctl_be_block_cw_dispatch_unmap(be_lun, io);
1579 		break;
1580 	default:
1581 		panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1582 		break;
1583 	}
1584 }
1585 
1586 SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t");
1587 SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t");
1588 SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t");
1589 SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t");
1590 
1591 static void
1592 ctl_be_block_next(struct ctl_be_block_io *beio)
1593 {
1594 	struct ctl_be_block_lun *be_lun;
1595 	union ctl_io *io;
1596 
1597 	io = beio->io;
1598 	be_lun = beio->lun;
1599 	ctl_free_beio(beio);
1600 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1601 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1602 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1603 		ctl_data_submit_done(io);
1604 		return;
1605 	}
1606 
1607 	io->io_hdr.status &= ~CTL_STATUS_MASK;
1608 	io->io_hdr.status |= CTL_STATUS_NONE;
1609 
1610 	mtx_lock(&be_lun->queue_lock);
1611 	STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1612 	mtx_unlock(&be_lun->queue_lock);
1613 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1614 }
1615 
1616 static void
1617 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1618 			   union ctl_io *io)
1619 {
1620 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1621 	struct ctl_be_block_io *beio;
1622 	struct ctl_be_block_softc *softc;
1623 	struct ctl_lba_len_flags *lbalen;
1624 	struct ctl_ptr_len_flags *bptrlen;
1625 	uint64_t len_left, lbas;
1626 	int i;
1627 
1628 	softc = be_lun->softc;
1629 
1630 	DPRINTF("entered\n");
1631 
1632 	lbalen = ARGS(io);
1633 	if (lbalen->flags & CTL_LLF_WRITE) {
1634 		SDT_PROBE0(cbb, , write, start);
1635 	} else {
1636 		SDT_PROBE0(cbb, , read, start);
1637 	}
1638 
1639 	beio = ctl_alloc_beio(softc);
1640 	beio->io = io;
1641 	beio->lun = be_lun;
1642 	bptrlen = PRIV(io);
1643 	bptrlen->ptr = (void *)beio;
1644 
1645 	switch (io->scsiio.tag_type) {
1646 	case CTL_TAG_ORDERED:
1647 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1648 		break;
1649 	case CTL_TAG_HEAD_OF_QUEUE:
1650 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1651 		break;
1652 	case CTL_TAG_UNTAGGED:
1653 	case CTL_TAG_SIMPLE:
1654 	case CTL_TAG_ACA:
1655 	default:
1656 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1657 		break;
1658 	}
1659 
1660 	if (lbalen->flags & CTL_LLF_WRITE) {
1661 		beio->bio_cmd = BIO_WRITE;
1662 		beio->ds_trans_type = DEVSTAT_WRITE;
1663 	} else {
1664 		beio->bio_cmd = BIO_READ;
1665 		beio->ds_trans_type = DEVSTAT_READ;
1666 	}
1667 
1668 	DPRINTF("%s at LBA %jx len %u @%ju\n",
1669 	       (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1670 	       (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1671 	lbas = CTLBLK_MAX_IO_SIZE;
1672 	if (lbalen->flags & CTL_LLF_COMPARE) {
1673 		beio->two_sglists = 1;
1674 		lbas /= 2;
1675 	}
1676 	lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize);
1677 	beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize;
1678 	beio->io_len = lbas * cbe_lun->blocksize;
1679 	bptrlen->len += lbas;
1680 
1681 	for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1682 		KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1683 		    i, CTLBLK_MAX_SEGS));
1684 
1685 		/*
1686 		 * Setup the S/G entry for this chunk.
1687 		 */
1688 		ctl_alloc_seg(softc, &beio->sg_segs[i],
1689 		    MIN(CTLBLK_MAX_SEG, len_left));
1690 
1691 		DPRINTF("segment %d addr %p len %zd\n", i,
1692 			beio->sg_segs[i].addr, beio->sg_segs[i].len);
1693 
1694 		/* Set up second segment for compare operation. */
1695 		if (beio->two_sglists) {
1696 			ctl_alloc_seg(softc,
1697 			    &beio->sg_segs[i + CTLBLK_HALF_SEGS],
1698 			    beio->sg_segs[i].len);
1699 		}
1700 
1701 		beio->num_segs++;
1702 		len_left -= beio->sg_segs[i].len;
1703 	}
1704 	if (bptrlen->len < lbalen->len)
1705 		beio->beio_cont = ctl_be_block_next;
1706 	io->scsiio.be_move_done = ctl_be_block_move_done;
1707 	/* For compare we have separate S/G lists for read and datamove. */
1708 	if (beio->two_sglists)
1709 		io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1710 	else
1711 		io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1712 	io->scsiio.kern_data_len = beio->io_len;
1713 	io->scsiio.kern_sg_entries = beio->num_segs;
1714 	io->scsiio.kern_data_ref = ctl_refcnt_beio;
1715 	io->scsiio.kern_data_arg = beio;
1716 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
1717 
1718 	/*
1719 	 * For the read case, we need to read the data into our buffers and
1720 	 * then we can send it back to the user.  For the write case, we
1721 	 * need to get the data from the user first.
1722 	 */
1723 	if (beio->bio_cmd == BIO_READ) {
1724 		SDT_PROBE0(cbb, , read, alloc_done);
1725 		be_lun->dispatch(be_lun, beio);
1726 	} else {
1727 		SDT_PROBE0(cbb, , write, alloc_done);
1728 		ctl_datamove(io);
1729 	}
1730 }
1731 
1732 static void
1733 ctl_be_block_worker(void *context, int pending)
1734 {
1735 	struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context;
1736 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1737 	union ctl_io *io;
1738 	struct ctl_be_block_io *beio;
1739 
1740 	DPRINTF("entered\n");
1741 	/*
1742 	 * Fetch and process I/Os from all queues.  If we detect LUN
1743 	 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race,
1744 	 * so make response maximally opaque to not confuse initiator.
1745 	 */
1746 	for (;;) {
1747 		mtx_lock(&be_lun->queue_lock);
1748 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1749 		if (io != NULL) {
1750 			DPRINTF("datamove queue\n");
1751 			STAILQ_REMOVE_HEAD(&be_lun->datamove_queue, links);
1752 			mtx_unlock(&be_lun->queue_lock);
1753 			beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1754 			if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1755 				ctl_set_busy(&io->scsiio);
1756 				ctl_complete_beio(beio);
1757 				continue;
1758 			}
1759 			be_lun->dispatch(be_lun, beio);
1760 			continue;
1761 		}
1762 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1763 		if (io != NULL) {
1764 			DPRINTF("config write queue\n");
1765 			STAILQ_REMOVE_HEAD(&be_lun->config_write_queue, links);
1766 			mtx_unlock(&be_lun->queue_lock);
1767 			if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1768 				ctl_set_busy(&io->scsiio);
1769 				ctl_config_write_done(io);
1770 				continue;
1771 			}
1772 			ctl_be_block_cw_dispatch(be_lun, io);
1773 			continue;
1774 		}
1775 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
1776 		if (io != NULL) {
1777 			DPRINTF("config read queue\n");
1778 			STAILQ_REMOVE_HEAD(&be_lun->config_read_queue, links);
1779 			mtx_unlock(&be_lun->queue_lock);
1780 			if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1781 				ctl_set_busy(&io->scsiio);
1782 				ctl_config_read_done(io);
1783 				continue;
1784 			}
1785 			ctl_be_block_cr_dispatch(be_lun, io);
1786 			continue;
1787 		}
1788 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1789 		if (io != NULL) {
1790 			DPRINTF("input queue\n");
1791 			STAILQ_REMOVE_HEAD(&be_lun->input_queue, links);
1792 			mtx_unlock(&be_lun->queue_lock);
1793 			if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1794 				ctl_set_busy(&io->scsiio);
1795 				ctl_data_submit_done(io);
1796 				continue;
1797 			}
1798 			ctl_be_block_dispatch(be_lun, io);
1799 			continue;
1800 		}
1801 
1802 		/*
1803 		 * If we get here, there is no work left in the queues, so
1804 		 * just break out and let the task queue go to sleep.
1805 		 */
1806 		mtx_unlock(&be_lun->queue_lock);
1807 		break;
1808 	}
1809 }
1810 
1811 /*
1812  * Entry point from CTL to the backend for I/O.  We queue everything to a
1813  * work thread, so this just puts the I/O on a queue and wakes up the
1814  * thread.
1815  */
1816 static int
1817 ctl_be_block_submit(union ctl_io *io)
1818 {
1819 	struct ctl_be_block_lun *be_lun;
1820 
1821 	DPRINTF("entered\n");
1822 
1823 	be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
1824 
1825 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
1826 	    ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
1827 
1828 	PRIV(io)->len = 0;
1829 
1830 	mtx_lock(&be_lun->queue_lock);
1831 	STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1832 	mtx_unlock(&be_lun->queue_lock);
1833 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1834 
1835 	return (CTL_RETVAL_COMPLETE);
1836 }
1837 
1838 static int
1839 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1840 			int flag, struct thread *td)
1841 {
1842 	struct ctl_be_block_softc *softc = &backend_block_softc;
1843 	int error;
1844 
1845 	error = 0;
1846 	switch (cmd) {
1847 	case CTL_LUN_REQ: {
1848 		struct ctl_lun_req *lun_req;
1849 
1850 		lun_req = (struct ctl_lun_req *)addr;
1851 
1852 		switch (lun_req->reqtype) {
1853 		case CTL_LUNREQ_CREATE:
1854 			error = ctl_be_block_create(softc, lun_req);
1855 			break;
1856 		case CTL_LUNREQ_RM:
1857 			error = ctl_be_block_rm(softc, lun_req);
1858 			break;
1859 		case CTL_LUNREQ_MODIFY:
1860 			error = ctl_be_block_modify(softc, lun_req);
1861 			break;
1862 		default:
1863 			lun_req->status = CTL_LUN_ERROR;
1864 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1865 				 "invalid LUN request type %d",
1866 				 lun_req->reqtype);
1867 			break;
1868 		}
1869 		break;
1870 	}
1871 	default:
1872 		error = ENOTTY;
1873 		break;
1874 	}
1875 
1876 	return (error);
1877 }
1878 
1879 static int
1880 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1881 {
1882 	struct ctl_be_lun *cbe_lun;
1883 	struct ctl_be_block_filedata *file_data;
1884 	struct ctl_lun_create_params *params;
1885 	const char		     *value;
1886 	struct vattr		      vattr;
1887 	off_t			      ps, pss, po, pos, us, uss, uo, uos;
1888 	int			      error;
1889 	long			      pconf;
1890 
1891 	cbe_lun = &be_lun->cbe_lun;
1892 	file_data = &be_lun->backend.file;
1893 	params = &be_lun->params;
1894 
1895 	be_lun->dev_type = CTL_BE_BLOCK_FILE;
1896 	be_lun->dispatch = ctl_be_block_dispatch_file;
1897 	be_lun->lun_flush = ctl_be_block_flush_file;
1898 	be_lun->get_lba_status = ctl_be_block_gls_file;
1899 	be_lun->getattr = ctl_be_block_getattr_file;
1900 	be_lun->unmap = ctl_be_block_unmap_file;
1901 	cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
1902 
1903 	error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1904 	if (error != 0) {
1905 		snprintf(req->error_str, sizeof(req->error_str),
1906 			 "error calling VOP_GETATTR() for file %s",
1907 			 be_lun->dev_path);
1908 		return (error);
1909 	}
1910 
1911 	error = VOP_PATHCONF(be_lun->vn, _PC_DEALLOC_PRESENT, &pconf);
1912 	if (error != 0) {
1913 		snprintf(req->error_str, sizeof(req->error_str),
1914 		    "error calling VOP_PATHCONF() for file %s",
1915 		    be_lun->dev_path);
1916 		return (error);
1917 	}
1918 	if (pconf == 1)
1919 		cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
1920 
1921 	file_data->cred = crhold(curthread->td_ucred);
1922 	if (params->lun_size_bytes != 0)
1923 		be_lun->size_bytes = params->lun_size_bytes;
1924 	else
1925 		be_lun->size_bytes = vattr.va_size;
1926 
1927 	/*
1928 	 * For files we can use any logical block size.  Prefer 512 bytes
1929 	 * for compatibility reasons.  If file's vattr.va_blocksize
1930 	 * (preferred I/O block size) is bigger and multiple to chosen
1931 	 * logical block size -- report it as physical block size.
1932 	 */
1933 	if (params->blocksize_bytes != 0)
1934 		cbe_lun->blocksize = params->blocksize_bytes;
1935 	else if (cbe_lun->lun_type == T_CDROM)
1936 		cbe_lun->blocksize = 2048;
1937 	else
1938 		cbe_lun->blocksize = 512;
1939 	be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
1940 	cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
1941 	    0 : (be_lun->size_blocks - 1);
1942 
1943 	us = ps = vattr.va_blocksize;
1944 	uo = po = 0;
1945 
1946 	value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
1947 	if (value != NULL)
1948 		ctl_expand_number(value, &ps);
1949 	value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
1950 	if (value != NULL)
1951 		ctl_expand_number(value, &po);
1952 	pss = ps / cbe_lun->blocksize;
1953 	pos = po / cbe_lun->blocksize;
1954 	if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
1955 	    ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
1956 		cbe_lun->pblockexp = fls(pss) - 1;
1957 		cbe_lun->pblockoff = (pss - pos) % pss;
1958 	}
1959 
1960 	value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
1961 	if (value != NULL)
1962 		ctl_expand_number(value, &us);
1963 	value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
1964 	if (value != NULL)
1965 		ctl_expand_number(value, &uo);
1966 	uss = us / cbe_lun->blocksize;
1967 	uos = uo / cbe_lun->blocksize;
1968 	if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
1969 	    ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
1970 		cbe_lun->ublockexp = fls(uss) - 1;
1971 		cbe_lun->ublockoff = (uss - uos) % uss;
1972 	}
1973 
1974 	/*
1975 	 * Sanity check.  The media size has to be at least one
1976 	 * sector long.
1977 	 */
1978 	if (be_lun->size_bytes < cbe_lun->blocksize) {
1979 		error = EINVAL;
1980 		snprintf(req->error_str, sizeof(req->error_str),
1981 			 "file %s size %ju < block size %u", be_lun->dev_path,
1982 			 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize);
1983 	}
1984 
1985 	cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize;
1986 	return (error);
1987 }
1988 
1989 static int
1990 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1991 {
1992 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1993 	struct ctl_lun_create_params *params;
1994 	struct cdevsw		     *csw;
1995 	struct cdev		     *dev;
1996 	const char		     *value;
1997 	int			      error, atomic, maxio, ref, unmap, tmp;
1998 	off_t			      ps, pss, po, pos, us, uss, uo, uos, otmp;
1999 
2000 	params = &be_lun->params;
2001 
2002 	be_lun->dev_type = CTL_BE_BLOCK_DEV;
2003 	csw = devvn_refthread(be_lun->vn, &dev, &ref);
2004 	if (csw == NULL)
2005 		return (ENXIO);
2006 	if (strcmp(csw->d_name, "zvol") == 0) {
2007 		be_lun->dispatch = ctl_be_block_dispatch_zvol;
2008 		be_lun->get_lba_status = ctl_be_block_gls_zvol;
2009 		atomic = maxio = CTLBLK_MAX_IO_SIZE;
2010 	} else {
2011 		be_lun->dispatch = ctl_be_block_dispatch_dev;
2012 		be_lun->get_lba_status = NULL;
2013 		atomic = 0;
2014 		maxio = dev->si_iosize_max;
2015 		if (maxio <= 0)
2016 			maxio = DFLTPHYS;
2017 		if (maxio > CTLBLK_MAX_SEG)
2018 			maxio = CTLBLK_MAX_SEG;
2019 	}
2020 	be_lun->lun_flush = ctl_be_block_flush_dev;
2021 	be_lun->getattr = ctl_be_block_getattr_dev;
2022 	be_lun->unmap = ctl_be_block_unmap_dev;
2023 
2024 	if (!csw->d_ioctl) {
2025 		dev_relthread(dev, ref);
2026 		snprintf(req->error_str, sizeof(req->error_str),
2027 			 "no d_ioctl for device %s!", be_lun->dev_path);
2028 		return (ENODEV);
2029 	}
2030 
2031 	error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD,
2032 			       curthread);
2033 	if (error) {
2034 		dev_relthread(dev, ref);
2035 		snprintf(req->error_str, sizeof(req->error_str),
2036 			 "error %d returned for DIOCGSECTORSIZE ioctl "
2037 			 "on %s!", error, be_lun->dev_path);
2038 		return (error);
2039 	}
2040 
2041 	/*
2042 	 * If the user has asked for a blocksize that is greater than the
2043 	 * backing device's blocksize, we can do it only if the blocksize
2044 	 * the user is asking for is an even multiple of the underlying
2045 	 * device's blocksize.
2046 	 */
2047 	if ((params->blocksize_bytes != 0) &&
2048 	    (params->blocksize_bytes >= tmp)) {
2049 		if (params->blocksize_bytes % tmp == 0) {
2050 			cbe_lun->blocksize = params->blocksize_bytes;
2051 		} else {
2052 			dev_relthread(dev, ref);
2053 			snprintf(req->error_str, sizeof(req->error_str),
2054 				 "requested blocksize %u is not an even "
2055 				 "multiple of backing device blocksize %u",
2056 				 params->blocksize_bytes, tmp);
2057 			return (EINVAL);
2058 		}
2059 	} else if (params->blocksize_bytes != 0) {
2060 		dev_relthread(dev, ref);
2061 		snprintf(req->error_str, sizeof(req->error_str),
2062 			 "requested blocksize %u < backing device "
2063 			 "blocksize %u", params->blocksize_bytes, tmp);
2064 		return (EINVAL);
2065 	} else if (cbe_lun->lun_type == T_CDROM)
2066 		cbe_lun->blocksize = MAX(tmp, 2048);
2067 	else
2068 		cbe_lun->blocksize = tmp;
2069 
2070 	error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD,
2071 			     curthread);
2072 	if (error) {
2073 		dev_relthread(dev, ref);
2074 		snprintf(req->error_str, sizeof(req->error_str),
2075 			 "error %d returned for DIOCGMEDIASIZE "
2076 			 " ioctl on %s!", error,
2077 			 be_lun->dev_path);
2078 		return (error);
2079 	}
2080 
2081 	if (params->lun_size_bytes != 0) {
2082 		if (params->lun_size_bytes > otmp) {
2083 			dev_relthread(dev, ref);
2084 			snprintf(req->error_str, sizeof(req->error_str),
2085 				 "requested LUN size %ju > backing device "
2086 				 "size %ju",
2087 				 (uintmax_t)params->lun_size_bytes,
2088 				 (uintmax_t)otmp);
2089 			return (EINVAL);
2090 		}
2091 
2092 		be_lun->size_bytes = params->lun_size_bytes;
2093 	} else
2094 		be_lun->size_bytes = otmp;
2095 	be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2096 	cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2097 	    0 : (be_lun->size_blocks - 1);
2098 
2099 	error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD,
2100 	    curthread);
2101 	if (error)
2102 		ps = po = 0;
2103 	else {
2104 		error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po,
2105 		    FREAD, curthread);
2106 		if (error)
2107 			po = 0;
2108 	}
2109 	us = ps;
2110 	uo = po;
2111 
2112 	value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
2113 	if (value != NULL)
2114 		ctl_expand_number(value, &ps);
2115 	value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
2116 	if (value != NULL)
2117 		ctl_expand_number(value, &po);
2118 	pss = ps / cbe_lun->blocksize;
2119 	pos = po / cbe_lun->blocksize;
2120 	if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
2121 	    ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
2122 		cbe_lun->pblockexp = fls(pss) - 1;
2123 		cbe_lun->pblockoff = (pss - pos) % pss;
2124 	}
2125 
2126 	value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
2127 	if (value != NULL)
2128 		ctl_expand_number(value, &us);
2129 	value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
2130 	if (value != NULL)
2131 		ctl_expand_number(value, &uo);
2132 	uss = us / cbe_lun->blocksize;
2133 	uos = uo / cbe_lun->blocksize;
2134 	if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
2135 	    ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
2136 		cbe_lun->ublockexp = fls(uss) - 1;
2137 		cbe_lun->ublockoff = (uss - uos) % uss;
2138 	}
2139 
2140 	cbe_lun->atomicblock = atomic / cbe_lun->blocksize;
2141 	cbe_lun->opttxferlen = maxio / cbe_lun->blocksize;
2142 
2143 	if (be_lun->dispatch == ctl_be_block_dispatch_zvol) {
2144 		unmap = 1;
2145 	} else {
2146 		struct diocgattr_arg	arg;
2147 
2148 		strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
2149 		arg.len = sizeof(arg.value.i);
2150 		error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
2151 		    curthread);
2152 		unmap = (error == 0) ? arg.value.i : 0;
2153 	}
2154 	value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
2155 	if (value != NULL)
2156 		unmap = (strcmp(value, "on") == 0);
2157 	if (unmap)
2158 		cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
2159 	else
2160 		cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
2161 
2162 	dev_relthread(dev, ref);
2163 	return (0);
2164 }
2165 
2166 static int
2167 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
2168 {
2169 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2170 	int flags;
2171 
2172 	if (be_lun->vn) {
2173 		flags = FREAD;
2174 		if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0)
2175 			flags |= FWRITE;
2176 		(void)vn_close(be_lun->vn, flags, NOCRED, curthread);
2177 		be_lun->vn = NULL;
2178 
2179 		switch (be_lun->dev_type) {
2180 		case CTL_BE_BLOCK_DEV:
2181 			break;
2182 		case CTL_BE_BLOCK_FILE:
2183 			if (be_lun->backend.file.cred != NULL) {
2184 				crfree(be_lun->backend.file.cred);
2185 				be_lun->backend.file.cred = NULL;
2186 			}
2187 			break;
2188 		case CTL_BE_BLOCK_NONE:
2189 			break;
2190 		default:
2191 			panic("Unexpected backend type %d", be_lun->dev_type);
2192 			break;
2193 		}
2194 		be_lun->dev_type = CTL_BE_BLOCK_NONE;
2195 	}
2196 	return (0);
2197 }
2198 
2199 static int
2200 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2201 {
2202 	struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2203 	struct nameidata nd;
2204 	const char	*value;
2205 	int		 error, flags;
2206 
2207 	error = 0;
2208 	if (rootvnode == NULL) {
2209 		snprintf(req->error_str, sizeof(req->error_str),
2210 			 "Root filesystem is not mounted");
2211 		return (1);
2212 	}
2213 	pwd_ensure_dirs();
2214 
2215 	value = dnvlist_get_string(cbe_lun->options, "file", NULL);
2216 	if (value == NULL) {
2217 		snprintf(req->error_str, sizeof(req->error_str),
2218 			 "no file argument specified");
2219 		return (1);
2220 	}
2221 	free(be_lun->dev_path, M_CTLBLK);
2222 	be_lun->dev_path = strdup(value, M_CTLBLK);
2223 
2224 	flags = FREAD;
2225 	value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
2226 	if (value != NULL) {
2227 		if (strcmp(value, "on") != 0)
2228 			flags |= FWRITE;
2229 	} else if (cbe_lun->lun_type == T_DIRECT)
2230 		flags |= FWRITE;
2231 
2232 again:
2233 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path);
2234 	error = vn_open(&nd, &flags, 0, NULL);
2235 	if ((error == EROFS || error == EACCES) && (flags & FWRITE)) {
2236 		flags &= ~FWRITE;
2237 		goto again;
2238 	}
2239 	if (error) {
2240 		/*
2241 		 * This is the only reasonable guess we can make as far as
2242 		 * path if the user doesn't give us a fully qualified path.
2243 		 * If they want to specify a file, they need to specify the
2244 		 * full path.
2245 		 */
2246 		if (be_lun->dev_path[0] != '/') {
2247 			char *dev_name;
2248 
2249 			asprintf(&dev_name, M_CTLBLK, "/dev/%s",
2250 				be_lun->dev_path);
2251 			free(be_lun->dev_path, M_CTLBLK);
2252 			be_lun->dev_path = dev_name;
2253 			goto again;
2254 		}
2255 		snprintf(req->error_str, sizeof(req->error_str),
2256 		    "error opening %s: %d", be_lun->dev_path, error);
2257 		return (error);
2258 	}
2259 	if (flags & FWRITE)
2260 		cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY;
2261 	else
2262 		cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
2263 
2264 	NDFREE_PNBUF(&nd);
2265 	be_lun->vn = nd.ni_vp;
2266 
2267 	/* We only support disks and files. */
2268 	if (vn_isdisk_error(be_lun->vn, &error)) {
2269 		error = ctl_be_block_open_dev(be_lun, req);
2270 	} else if (be_lun->vn->v_type == VREG) {
2271 		error = ctl_be_block_open_file(be_lun, req);
2272 	} else {
2273 		error = EINVAL;
2274 		snprintf(req->error_str, sizeof(req->error_str),
2275 			 "%s is not a disk or plain file", be_lun->dev_path);
2276 	}
2277 	VOP_UNLOCK(be_lun->vn);
2278 
2279 	if (error != 0)
2280 		ctl_be_block_close(be_lun);
2281 	cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2282 	if (be_lun->dispatch != ctl_be_block_dispatch_dev)
2283 		cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
2284 	value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
2285 	if (value != NULL && strcmp(value, "on") == 0)
2286 		cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
2287 	else if (value != NULL && strcmp(value, "read") == 0)
2288 		cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2289 	else if (value != NULL && strcmp(value, "soft") == 0)
2290 		cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
2291 	else if (value != NULL && strcmp(value, "off") == 0)
2292 		cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2293 	return (0);
2294 }
2295 
2296 static int
2297 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2298 {
2299 	struct ctl_be_lun *cbe_lun;
2300 	struct ctl_be_block_lun *be_lun;
2301 	struct ctl_lun_create_params *params;
2302 	char num_thread_str[16];
2303 	char tmpstr[32];
2304 	const char *value;
2305 	int retval, num_threads;
2306 	int tmp_num_threads;
2307 
2308 	params = &req->reqdata.create;
2309 	retval = 0;
2310 	req->status = CTL_LUN_OK;
2311 
2312 	be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
2313 	cbe_lun = &be_lun->cbe_lun;
2314 	be_lun->params = req->reqdata.create;
2315 	be_lun->softc = softc;
2316 	STAILQ_INIT(&be_lun->input_queue);
2317 	STAILQ_INIT(&be_lun->config_read_queue);
2318 	STAILQ_INIT(&be_lun->config_write_queue);
2319 	STAILQ_INIT(&be_lun->datamove_queue);
2320 	mtx_init(&be_lun->io_lock, "ctlblock io", NULL, MTX_DEF);
2321 	mtx_init(&be_lun->queue_lock, "ctlblock queue", NULL, MTX_DEF);
2322 	cbe_lun->options = nvlist_clone(req->args_nvl);
2323 
2324 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
2325 		cbe_lun->lun_type = params->device_type;
2326 	else
2327 		cbe_lun->lun_type = T_DIRECT;
2328 	be_lun->flags = 0;
2329 	cbe_lun->flags = 0;
2330 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2331 	if (value != NULL) {
2332 		if (strcmp(value, "primary") == 0)
2333 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2334 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2335 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2336 
2337 	if (cbe_lun->lun_type == T_DIRECT ||
2338 	    cbe_lun->lun_type == T_CDROM) {
2339 		be_lun->size_bytes = params->lun_size_bytes;
2340 		if (params->blocksize_bytes != 0)
2341 			cbe_lun->blocksize = params->blocksize_bytes;
2342 		else if (cbe_lun->lun_type == T_CDROM)
2343 			cbe_lun->blocksize = 2048;
2344 		else
2345 			cbe_lun->blocksize = 512;
2346 		be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2347 		cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2348 		    0 : (be_lun->size_blocks - 1);
2349 
2350 		if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2351 		    control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2352 			retval = ctl_be_block_open(be_lun, req);
2353 			if (retval != 0) {
2354 				retval = 0;
2355 				req->status = CTL_LUN_WARNING;
2356 			}
2357 		}
2358 		num_threads = cbb_num_threads;
2359 	} else {
2360 		num_threads = 1;
2361 	}
2362 
2363 	value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL);
2364 	if (value != NULL) {
2365 		tmp_num_threads = strtol(value, NULL, 0);
2366 
2367 		/*
2368 		 * We don't let the user specify less than one
2369 		 * thread, but hope he's clueful enough not to
2370 		 * specify 1000 threads.
2371 		 */
2372 		if (tmp_num_threads < 1) {
2373 			snprintf(req->error_str, sizeof(req->error_str),
2374 				 "invalid number of threads %s",
2375 				 num_thread_str);
2376 			goto bailout_error;
2377 		}
2378 		num_threads = tmp_num_threads;
2379 	}
2380 
2381 	if (be_lun->vn == NULL)
2382 		cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2383 	/* Tell the user the blocksize we ended up using */
2384 	params->lun_size_bytes = be_lun->size_bytes;
2385 	params->blocksize_bytes = cbe_lun->blocksize;
2386 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2387 		cbe_lun->req_lun_id = params->req_lun_id;
2388 		cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
2389 	} else
2390 		cbe_lun->req_lun_id = 0;
2391 
2392 	cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown;
2393 	cbe_lun->be = &ctl_be_block_driver;
2394 
2395 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2396 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
2397 			 softc->num_luns);
2398 		strncpy((char *)cbe_lun->serial_num, tmpstr,
2399 			MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
2400 
2401 		/* Tell the user what we used for a serial number */
2402 		strncpy((char *)params->serial_num, tmpstr,
2403 			MIN(sizeof(params->serial_num), sizeof(tmpstr)));
2404 	} else {
2405 		strncpy((char *)cbe_lun->serial_num, params->serial_num,
2406 			MIN(sizeof(cbe_lun->serial_num),
2407 			sizeof(params->serial_num)));
2408 	}
2409 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2410 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
2411 		strncpy((char *)cbe_lun->device_id, tmpstr,
2412 			MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
2413 
2414 		/* Tell the user what we used for a device ID */
2415 		strncpy((char *)params->device_id, tmpstr,
2416 			MIN(sizeof(params->device_id), sizeof(tmpstr)));
2417 	} else {
2418 		strncpy((char *)cbe_lun->device_id, params->device_id,
2419 			MIN(sizeof(cbe_lun->device_id),
2420 			    sizeof(params->device_id)));
2421 	}
2422 
2423 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2424 
2425 	be_lun->io_taskqueue = taskqueue_create("ctlblocktq", M_WAITOK,
2426 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2427 
2428 	if (be_lun->io_taskqueue == NULL) {
2429 		snprintf(req->error_str, sizeof(req->error_str),
2430 			 "unable to create taskqueue");
2431 		goto bailout_error;
2432 	}
2433 
2434 	/*
2435 	 * Note that we start the same number of threads by default for
2436 	 * both the file case and the block device case.  For the file
2437 	 * case, we need multiple threads to allow concurrency, because the
2438 	 * vnode interface is designed to be a blocking interface.  For the
2439 	 * block device case, ZFS zvols at least will block the caller's
2440 	 * context in many instances, and so we need multiple threads to
2441 	 * overcome that problem.  Other block devices don't need as many
2442 	 * threads, but they shouldn't cause too many problems.
2443 	 *
2444 	 * If the user wants to just have a single thread for a block
2445 	 * device, he can specify that when the LUN is created, or change
2446 	 * the tunable/sysctl to alter the default number of threads.
2447 	 */
2448 	retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
2449 					 /*num threads*/num_threads,
2450 					 /*priority*/PUSER,
2451 					 /*proc*/control_softc->ctl_proc,
2452 					 /*thread name*/"block");
2453 
2454 	if (retval != 0)
2455 		goto bailout_error;
2456 
2457 	be_lun->num_threads = num_threads;
2458 
2459 	retval = ctl_add_lun(&be_lun->cbe_lun);
2460 	if (retval != 0) {
2461 		snprintf(req->error_str, sizeof(req->error_str),
2462 			 "ctl_add_lun() returned error %d, see dmesg for "
2463 			 "details", retval);
2464 		retval = 0;
2465 		goto bailout_error;
2466 	}
2467 
2468 	be_lun->disk_stats = devstat_new_entry("cbb", cbe_lun->lun_id,
2469 					       cbe_lun->blocksize,
2470 					       DEVSTAT_ALL_SUPPORTED,
2471 					       cbe_lun->lun_type
2472 					       | DEVSTAT_TYPE_IF_OTHER,
2473 					       DEVSTAT_PRIORITY_OTHER);
2474 
2475 	mtx_lock(&softc->lock);
2476 	softc->num_luns++;
2477 	SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
2478 	mtx_unlock(&softc->lock);
2479 
2480 	params->req_lun_id = cbe_lun->lun_id;
2481 
2482 	return (retval);
2483 
2484 bailout_error:
2485 	req->status = CTL_LUN_ERROR;
2486 
2487 	if (be_lun->io_taskqueue != NULL)
2488 		taskqueue_free(be_lun->io_taskqueue);
2489 	ctl_be_block_close(be_lun);
2490 	if (be_lun->dev_path != NULL)
2491 		free(be_lun->dev_path, M_CTLBLK);
2492 	nvlist_destroy(cbe_lun->options);
2493 	mtx_destroy(&be_lun->queue_lock);
2494 	mtx_destroy(&be_lun->io_lock);
2495 	free(be_lun, M_CTLBLK);
2496 
2497 	return (retval);
2498 }
2499 
2500 static int
2501 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2502 {
2503 	struct ctl_lun_rm_params *params;
2504 	struct ctl_be_block_lun *be_lun;
2505 	struct ctl_be_lun *cbe_lun;
2506 	int retval;
2507 
2508 	params = &req->reqdata.rm;
2509 
2510 	sx_xlock(&softc->modify_lock);
2511 	mtx_lock(&softc->lock);
2512 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2513 		if (be_lun->cbe_lun.lun_id == params->lun_id) {
2514 			SLIST_REMOVE(&softc->lun_list, be_lun,
2515 			    ctl_be_block_lun, links);
2516 			softc->num_luns--;
2517 			break;
2518 		}
2519 	}
2520 	mtx_unlock(&softc->lock);
2521 	sx_xunlock(&softc->modify_lock);
2522 	if (be_lun == NULL) {
2523 		snprintf(req->error_str, sizeof(req->error_str),
2524 			 "LUN %u is not managed by the block backend",
2525 			 params->lun_id);
2526 		goto bailout_error;
2527 	}
2528 	cbe_lun = &be_lun->cbe_lun;
2529 
2530 	if (be_lun->vn != NULL) {
2531 		cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2532 		ctl_lun_no_media(cbe_lun);
2533 		taskqueue_drain_all(be_lun->io_taskqueue);
2534 		ctl_be_block_close(be_lun);
2535 	}
2536 
2537 	mtx_lock(&softc->lock);
2538 	be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2539 	mtx_unlock(&softc->lock);
2540 
2541 	retval = ctl_remove_lun(cbe_lun);
2542 	if (retval != 0) {
2543 		snprintf(req->error_str, sizeof(req->error_str),
2544 			 "error %d returned from ctl_remove_lun() for "
2545 			 "LUN %d", retval, params->lun_id);
2546 		mtx_lock(&softc->lock);
2547 		be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2548 		mtx_unlock(&softc->lock);
2549 		goto bailout_error;
2550 	}
2551 
2552 	mtx_lock(&softc->lock);
2553 	while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2554 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblockrm", 0);
2555 		if (retval == EINTR)
2556 			break;
2557 	}
2558 	be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2559 	if (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2560 		mtx_unlock(&softc->lock);
2561 		free(be_lun, M_CTLBLK);
2562 	} else {
2563 		mtx_unlock(&softc->lock);
2564 		return (EINTR);
2565 	}
2566 
2567 	req->status = CTL_LUN_OK;
2568 	return (0);
2569 
2570 bailout_error:
2571 	req->status = CTL_LUN_ERROR;
2572 	return (0);
2573 }
2574 
2575 static int
2576 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2577 {
2578 	struct ctl_lun_modify_params *params;
2579 	struct ctl_be_block_lun *be_lun;
2580 	struct ctl_be_lun *cbe_lun;
2581 	const char *value;
2582 	uint64_t oldsize;
2583 	int error, wasprim;
2584 
2585 	params = &req->reqdata.modify;
2586 
2587 	sx_xlock(&softc->modify_lock);
2588 	mtx_lock(&softc->lock);
2589 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2590 		if (be_lun->cbe_lun.lun_id == params->lun_id)
2591 			break;
2592 	}
2593 	mtx_unlock(&softc->lock);
2594 	if (be_lun == NULL) {
2595 		snprintf(req->error_str, sizeof(req->error_str),
2596 			 "LUN %u is not managed by the block backend",
2597 			 params->lun_id);
2598 		goto bailout_error;
2599 	}
2600 	cbe_lun = &be_lun->cbe_lun;
2601 
2602 	if (params->lun_size_bytes != 0)
2603 		be_lun->params.lun_size_bytes = params->lun_size_bytes;
2604 
2605 	if (req->args_nvl != NULL) {
2606 		nvlist_destroy(cbe_lun->options);
2607 		cbe_lun->options = nvlist_clone(req->args_nvl);
2608 	}
2609 
2610 	wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
2611 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2612 	if (value != NULL) {
2613 		if (strcmp(value, "primary") == 0)
2614 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2615 		else
2616 			cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2617 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2618 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2619 	else
2620 		cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2621 	if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
2622 		if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
2623 			ctl_lun_primary(cbe_lun);
2624 		else
2625 			ctl_lun_secondary(cbe_lun);
2626 	}
2627 
2628 	oldsize = be_lun->size_blocks;
2629 	if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2630 	    control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2631 		if (be_lun->vn == NULL)
2632 			error = ctl_be_block_open(be_lun, req);
2633 		else if (vn_isdisk_error(be_lun->vn, &error))
2634 			error = ctl_be_block_open_dev(be_lun, req);
2635 		else if (be_lun->vn->v_type == VREG) {
2636 			vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2637 			error = ctl_be_block_open_file(be_lun, req);
2638 			VOP_UNLOCK(be_lun->vn);
2639 		} else
2640 			error = EINVAL;
2641 		if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) &&
2642 		    be_lun->vn != NULL) {
2643 			cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2644 			ctl_lun_has_media(cbe_lun);
2645 		} else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 &&
2646 		    be_lun->vn == NULL) {
2647 			cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2648 			ctl_lun_no_media(cbe_lun);
2649 		}
2650 		cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2651 	} else {
2652 		if (be_lun->vn != NULL) {
2653 			cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2654 			ctl_lun_no_media(cbe_lun);
2655 			taskqueue_drain_all(be_lun->io_taskqueue);
2656 			error = ctl_be_block_close(be_lun);
2657 		} else
2658 			error = 0;
2659 	}
2660 	if (be_lun->size_blocks != oldsize)
2661 		ctl_lun_capacity_changed(cbe_lun);
2662 
2663 	/* Tell the user the exact size we ended up using */
2664 	params->lun_size_bytes = be_lun->size_bytes;
2665 
2666 	sx_xunlock(&softc->modify_lock);
2667 	req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2668 	return (0);
2669 
2670 bailout_error:
2671 	sx_xunlock(&softc->modify_lock);
2672 	req->status = CTL_LUN_ERROR;
2673 	return (0);
2674 }
2675 
2676 static void
2677 ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun)
2678 {
2679 	struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)cbe_lun;
2680 	struct ctl_be_block_softc *softc = be_lun->softc;
2681 
2682 	taskqueue_drain_all(be_lun->io_taskqueue);
2683 	taskqueue_free(be_lun->io_taskqueue);
2684 	if (be_lun->disk_stats != NULL)
2685 		devstat_remove_entry(be_lun->disk_stats);
2686 	nvlist_destroy(be_lun->cbe_lun.options);
2687 	free(be_lun->dev_path, M_CTLBLK);
2688 	mtx_destroy(&be_lun->queue_lock);
2689 	mtx_destroy(&be_lun->io_lock);
2690 
2691 	mtx_lock(&softc->lock);
2692 	be_lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2693 	if (be_lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2694 		wakeup(be_lun);
2695 	else
2696 		free(be_lun, M_CTLBLK);
2697 	mtx_unlock(&softc->lock);
2698 }
2699 
2700 static int
2701 ctl_be_block_config_write(union ctl_io *io)
2702 {
2703 	struct ctl_be_block_lun *be_lun;
2704 	struct ctl_be_lun *cbe_lun;
2705 	int retval;
2706 
2707 	DPRINTF("entered\n");
2708 
2709 	cbe_lun = CTL_BACKEND_LUN(io);
2710 	be_lun = (struct ctl_be_block_lun *)cbe_lun;
2711 
2712 	retval = 0;
2713 	switch (io->scsiio.cdb[0]) {
2714 	case SYNCHRONIZE_CACHE:
2715 	case SYNCHRONIZE_CACHE_16:
2716 	case WRITE_SAME_10:
2717 	case WRITE_SAME_16:
2718 	case UNMAP:
2719 		/*
2720 		 * The upper level CTL code will filter out any CDBs with
2721 		 * the immediate bit set and return the proper error.
2722 		 *
2723 		 * We don't really need to worry about what LBA range the
2724 		 * user asked to be synced out.  When they issue a sync
2725 		 * cache command, we'll sync out the whole thing.
2726 		 */
2727 		mtx_lock(&be_lun->queue_lock);
2728 		STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2729 				   links);
2730 		mtx_unlock(&be_lun->queue_lock);
2731 		taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2732 		break;
2733 	case START_STOP_UNIT: {
2734 		struct scsi_start_stop_unit *cdb;
2735 		struct ctl_lun_req req;
2736 
2737 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2738 		if ((cdb->how & SSS_PC_MASK) != 0) {
2739 			ctl_set_success(&io->scsiio);
2740 			ctl_config_write_done(io);
2741 			break;
2742 		}
2743 		if (cdb->how & SSS_START) {
2744 			if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) {
2745 				retval = ctl_be_block_open(be_lun, &req);
2746 				cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2747 				if (retval == 0) {
2748 					cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2749 					ctl_lun_has_media(cbe_lun);
2750 				} else {
2751 					cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2752 					ctl_lun_no_media(cbe_lun);
2753 				}
2754 			}
2755 			ctl_start_lun(cbe_lun);
2756 		} else {
2757 			ctl_stop_lun(cbe_lun);
2758 			if (cdb->how & SSS_LOEJ) {
2759 				cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2760 				cbe_lun->flags |= CTL_LUN_FLAG_EJECTED;
2761 				ctl_lun_ejected(cbe_lun);
2762 				if (be_lun->vn != NULL)
2763 					ctl_be_block_close(be_lun);
2764 			}
2765 		}
2766 
2767 		ctl_set_success(&io->scsiio);
2768 		ctl_config_write_done(io);
2769 		break;
2770 	}
2771 	case PREVENT_ALLOW:
2772 		ctl_set_success(&io->scsiio);
2773 		ctl_config_write_done(io);
2774 		break;
2775 	default:
2776 		ctl_set_invalid_opcode(&io->scsiio);
2777 		ctl_config_write_done(io);
2778 		retval = CTL_RETVAL_COMPLETE;
2779 		break;
2780 	}
2781 
2782 	return (retval);
2783 }
2784 
2785 static int
2786 ctl_be_block_config_read(union ctl_io *io)
2787 {
2788 	struct ctl_be_block_lun *be_lun;
2789 	int retval = 0;
2790 
2791 	DPRINTF("entered\n");
2792 
2793 	be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
2794 
2795 	switch (io->scsiio.cdb[0]) {
2796 	case SERVICE_ACTION_IN:
2797 		if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
2798 			mtx_lock(&be_lun->queue_lock);
2799 			STAILQ_INSERT_TAIL(&be_lun->config_read_queue,
2800 			    &io->io_hdr, links);
2801 			mtx_unlock(&be_lun->queue_lock);
2802 			taskqueue_enqueue(be_lun->io_taskqueue,
2803 			    &be_lun->io_task);
2804 			retval = CTL_RETVAL_QUEUED;
2805 			break;
2806 		}
2807 		ctl_set_invalid_field(&io->scsiio,
2808 				      /*sks_valid*/ 1,
2809 				      /*command*/ 1,
2810 				      /*field*/ 1,
2811 				      /*bit_valid*/ 1,
2812 				      /*bit*/ 4);
2813 		ctl_config_read_done(io);
2814 		retval = CTL_RETVAL_COMPLETE;
2815 		break;
2816 	default:
2817 		ctl_set_invalid_opcode(&io->scsiio);
2818 		ctl_config_read_done(io);
2819 		retval = CTL_RETVAL_COMPLETE;
2820 		break;
2821 	}
2822 
2823 	return (retval);
2824 }
2825 
2826 static int
2827 ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb)
2828 {
2829 	struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
2830 	int retval;
2831 
2832 	retval = sbuf_printf(sb, "\t<num_threads>");
2833 	if (retval != 0)
2834 		goto bailout;
2835 	retval = sbuf_printf(sb, "%d", lun->num_threads);
2836 	if (retval != 0)
2837 		goto bailout;
2838 	retval = sbuf_printf(sb, "</num_threads>\n");
2839 
2840 bailout:
2841 	return (retval);
2842 }
2843 
2844 static uint64_t
2845 ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
2846 {
2847 	struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
2848 
2849 	if (lun->getattr == NULL)
2850 		return (UINT64_MAX);
2851 	return (lun->getattr(lun, attrname));
2852 }
2853 
2854 static int
2855 ctl_be_block_init(void)
2856 {
2857 	struct ctl_be_block_softc *softc = &backend_block_softc;
2858 
2859 	sx_init(&softc->modify_lock, "ctlblock modify");
2860 	mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
2861 	softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2862 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2863 	softc->bufmin_zone = uma_zcreate("ctlblockmin", CTLBLK_MIN_SEG,
2864 	    NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
2865 	if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG)
2866 		softc->bufmax_zone = uma_zcreate("ctlblockmax", CTLBLK_MAX_SEG,
2867 		    NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
2868 	SLIST_INIT(&softc->lun_list);
2869 	return (0);
2870 }
2871 
2872 static int
2873 ctl_be_block_shutdown(void)
2874 {
2875 	struct ctl_be_block_softc *softc = &backend_block_softc;
2876 	struct ctl_be_block_lun *lun;
2877 
2878 	mtx_lock(&softc->lock);
2879 	while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
2880 		SLIST_REMOVE_HEAD(&softc->lun_list, links);
2881 		softc->num_luns--;
2882 		/*
2883 		 * Drop our lock here.  Since ctl_remove_lun() can call
2884 		 * back into us, this could potentially lead to a recursive
2885 		 * lock of the same mutex, which would cause a hang.
2886 		 */
2887 		mtx_unlock(&softc->lock);
2888 		ctl_remove_lun(&lun->cbe_lun);
2889 		mtx_lock(&softc->lock);
2890 	}
2891 	mtx_unlock(&softc->lock);
2892 	uma_zdestroy(softc->bufmin_zone);
2893 	if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG)
2894 		uma_zdestroy(softc->bufmax_zone);
2895 	uma_zdestroy(softc->beio_zone);
2896 	mtx_destroy(&softc->lock);
2897 	sx_destroy(&softc->modify_lock);
2898 	return (0);
2899 }
2900