xref: /freebsd/sys/cam/ctl/ctl_backend_block.c (revision 0572ccaa4543b0abef8ef81e384c1d04de9f3da1)
1 /*-
2  * Copyright (c) 2003 Silicon Graphics International Corp.
3  * Copyright (c) 2009-2011 Spectra Logic Corporation
4  * Copyright (c) 2012 The FreeBSD Foundation
5  * All rights reserved.
6  *
7  * Portions of this software were developed by Edward Tomasz Napierala
8  * under sponsorship from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions, and the following disclaimer,
15  *    without modification.
16  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
17  *    substantially similar to the "NO WARRANTY" disclaimer below
18  *    ("Disclaimer") and any redistribution must be conditioned upon
19  *    including a substantially similar Disclaimer requirement for further
20  *    binary redistribution.
21  *
22  * NO WARRANTY
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
31  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
32  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGES.
34  *
35  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $
36  */
37 /*
38  * CAM Target Layer driver backend for block devices.
39  *
40  * Author: Ken Merry <ken@FreeBSD.org>
41  */
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/types.h>
49 #include <sys/kthread.h>
50 #include <sys/bio.h>
51 #include <sys/fcntl.h>
52 #include <sys/limits.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/condvar.h>
56 #include <sys/malloc.h>
57 #include <sys/conf.h>
58 #include <sys/ioccom.h>
59 #include <sys/queue.h>
60 #include <sys/sbuf.h>
61 #include <sys/endian.h>
62 #include <sys/uio.h>
63 #include <sys/buf.h>
64 #include <sys/taskqueue.h>
65 #include <sys/vnode.h>
66 #include <sys/namei.h>
67 #include <sys/mount.h>
68 #include <sys/disk.h>
69 #include <sys/fcntl.h>
70 #include <sys/filedesc.h>
71 #include <sys/proc.h>
72 #include <sys/pcpu.h>
73 #include <sys/module.h>
74 #include <sys/sdt.h>
75 #include <sys/devicestat.h>
76 #include <sys/sysctl.h>
77 
78 #include <geom/geom.h>
79 
80 #include <cam/cam.h>
81 #include <cam/scsi/scsi_all.h>
82 #include <cam/scsi/scsi_da.h>
83 #include <cam/ctl/ctl_io.h>
84 #include <cam/ctl/ctl.h>
85 #include <cam/ctl/ctl_backend.h>
86 #include <cam/ctl/ctl_frontend_internal.h>
87 #include <cam/ctl/ctl_ioctl.h>
88 #include <cam/ctl/ctl_scsi_all.h>
89 #include <cam/ctl/ctl_error.h>
90 
91 /*
92  * The idea here is that we'll allocate enough S/G space to hold a 1MB
93  * I/O.  If we get an I/O larger than that, we'll split it.
94  */
95 #define	CTLBLK_MAX_IO_SIZE	(1024 * 1024)
96 #define	CTLBLK_MAX_SEG		MAXPHYS
97 #define	CTLBLK_MAX_SEGS		MAX(CTLBLK_MAX_IO_SIZE / CTLBLK_MAX_SEG, 1)
98 
99 #ifdef CTLBLK_DEBUG
100 #define DPRINTF(fmt, args...) \
101     printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
102 #else
103 #define DPRINTF(fmt, args...) do {} while(0)
104 #endif
105 
106 SDT_PROVIDER_DEFINE(cbb);
107 
108 typedef enum {
109 	CTL_BE_BLOCK_LUN_UNCONFIGURED	= 0x01,
110 	CTL_BE_BLOCK_LUN_CONFIG_ERR	= 0x02,
111 	CTL_BE_BLOCK_LUN_WAITING	= 0x04,
112 	CTL_BE_BLOCK_LUN_MULTI_THREAD	= 0x08
113 } ctl_be_block_lun_flags;
114 
115 typedef enum {
116 	CTL_BE_BLOCK_NONE,
117 	CTL_BE_BLOCK_DEV,
118 	CTL_BE_BLOCK_FILE
119 } ctl_be_block_type;
120 
121 struct ctl_be_block_devdata {
122 	struct cdev *cdev;
123 	struct cdevsw *csw;
124 	int dev_ref;
125 };
126 
127 struct ctl_be_block_filedata {
128 	struct ucred *cred;
129 };
130 
131 union ctl_be_block_bedata {
132 	struct ctl_be_block_devdata dev;
133 	struct ctl_be_block_filedata file;
134 };
135 
136 struct ctl_be_block_io;
137 struct ctl_be_block_lun;
138 
139 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
140 			       struct ctl_be_block_io *beio);
141 
142 /*
143  * Backend LUN structure.  There is a 1:1 mapping between a block device
144  * and a backend block LUN, and between a backend block LUN and a CTL LUN.
145  */
146 struct ctl_be_block_lun {
147 	struct ctl_block_disk *disk;
148 	char lunname[32];
149 	char *dev_path;
150 	ctl_be_block_type dev_type;
151 	struct vnode *vn;
152 	union ctl_be_block_bedata backend;
153 	cbb_dispatch_t dispatch;
154 	cbb_dispatch_t lun_flush;
155 	cbb_dispatch_t unmap;
156 	struct mtx lock;
157 	uma_zone_t lun_zone;
158 	uint64_t size_blocks;
159 	uint64_t size_bytes;
160 	uint32_t blocksize;
161 	int blocksize_shift;
162 	uint16_t pblockexp;
163 	uint16_t pblockoff;
164 	struct ctl_be_block_softc *softc;
165 	struct devstat *disk_stats;
166 	ctl_be_block_lun_flags flags;
167 	STAILQ_ENTRY(ctl_be_block_lun) links;
168 	struct ctl_be_lun ctl_be_lun;
169 	struct taskqueue *io_taskqueue;
170 	struct task io_task;
171 	int num_threads;
172 	STAILQ_HEAD(, ctl_io_hdr) input_queue;
173 	STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
174 	STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
175 };
176 
177 /*
178  * Overall softc structure for the block backend module.
179  */
180 struct ctl_be_block_softc {
181 	struct mtx			 lock;
182 	int				 num_disks;
183 	STAILQ_HEAD(, ctl_block_disk)	 disk_list;
184 	int				 num_luns;
185 	STAILQ_HEAD(, ctl_be_block_lun)	 lun_list;
186 };
187 
188 static struct ctl_be_block_softc backend_block_softc;
189 
190 /*
191  * Per-I/O information.
192  */
193 struct ctl_be_block_io {
194 	union ctl_io			*io;
195 	struct ctl_sg_entry		sg_segs[CTLBLK_MAX_SEGS];
196 	struct iovec			xiovecs[CTLBLK_MAX_SEGS];
197 	int				bio_cmd;
198 	int				bio_flags;
199 	int				num_segs;
200 	int				num_bios_sent;
201 	int				num_bios_done;
202 	int				send_complete;
203 	int				num_errors;
204 	struct bintime			ds_t0;
205 	devstat_tag_type		ds_tag_type;
206 	devstat_trans_flags		ds_trans_type;
207 	uint64_t			io_len;
208 	uint64_t			io_offset;
209 	struct ctl_be_block_softc	*softc;
210 	struct ctl_be_block_lun		*lun;
211 	void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
212 };
213 
214 static int cbb_num_threads = 14;
215 TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads);
216 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0,
217 	    "CAM Target Layer Block Backend");
218 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW,
219            &cbb_num_threads, 0, "Number of threads per backing file");
220 
221 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
222 static void ctl_free_beio(struct ctl_be_block_io *beio);
223 static void ctl_complete_beio(struct ctl_be_block_io *beio);
224 static int ctl_be_block_move_done(union ctl_io *io);
225 static void ctl_be_block_biodone(struct bio *bio);
226 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
227 				    struct ctl_be_block_io *beio);
228 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
229 				       struct ctl_be_block_io *beio);
230 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
231 				   struct ctl_be_block_io *beio);
232 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
233 				   struct ctl_be_block_io *beio);
234 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
235 				      struct ctl_be_block_io *beio);
236 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
237 				    union ctl_io *io);
238 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
239 				  union ctl_io *io);
240 static void ctl_be_block_worker(void *context, int pending);
241 static int ctl_be_block_submit(union ctl_io *io);
242 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
243 				   int flag, struct thread *td);
244 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
245 				  struct ctl_lun_req *req);
246 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
247 				 struct ctl_lun_req *req);
248 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
249 static int ctl_be_block_open(struct ctl_be_block_softc *softc,
250 			     struct ctl_be_block_lun *be_lun,
251 			     struct ctl_lun_req *req);
252 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
253 			       struct ctl_lun_req *req);
254 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
255 			   struct ctl_lun_req *req);
256 static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
257 				  struct ctl_lun_req *req);
258 static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
259 				 struct ctl_lun_req *req);
260 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
261 			   struct ctl_lun_req *req);
262 static void ctl_be_block_lun_shutdown(void *be_lun);
263 static void ctl_be_block_lun_config_status(void *be_lun,
264 					   ctl_lun_config_status status);
265 static int ctl_be_block_config_write(union ctl_io *io);
266 static int ctl_be_block_config_read(union ctl_io *io);
267 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb);
268 int ctl_be_block_init(void);
269 
270 static struct ctl_backend_driver ctl_be_block_driver =
271 {
272 	.name = "block",
273 	.flags = CTL_BE_FLAG_HAS_CONFIG,
274 	.init = ctl_be_block_init,
275 	.data_submit = ctl_be_block_submit,
276 	.data_move_done = ctl_be_block_move_done,
277 	.config_read = ctl_be_block_config_read,
278 	.config_write = ctl_be_block_config_write,
279 	.ioctl = ctl_be_block_ioctl,
280 	.lun_info = ctl_be_block_lun_info
281 };
282 
283 MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend");
284 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
285 
286 static uma_zone_t beio_zone;
287 
288 static struct ctl_be_block_io *
289 ctl_alloc_beio(struct ctl_be_block_softc *softc)
290 {
291 	struct ctl_be_block_io *beio;
292 
293 	beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO);
294 	beio->softc = softc;
295 	return (beio);
296 }
297 
298 static void
299 ctl_free_beio(struct ctl_be_block_io *beio)
300 {
301 	int duplicate_free;
302 	int i;
303 
304 	duplicate_free = 0;
305 
306 	for (i = 0; i < beio->num_segs; i++) {
307 		if (beio->sg_segs[i].addr == NULL)
308 			duplicate_free++;
309 
310 		uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr);
311 		beio->sg_segs[i].addr = NULL;
312 	}
313 
314 	if (duplicate_free > 0) {
315 		printf("%s: %d duplicate frees out of %d segments\n", __func__,
316 		       duplicate_free, beio->num_segs);
317 	}
318 
319 	uma_zfree(beio_zone, beio);
320 }
321 
322 static void
323 ctl_complete_beio(struct ctl_be_block_io *beio)
324 {
325 	union ctl_io *io;
326 	int io_len;
327 
328 	io = beio->io;
329 
330 	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)
331 		io_len = beio->io_len;
332 	else
333 		io_len = 0;
334 
335 	devstat_end_transaction(beio->lun->disk_stats,
336 				/*bytes*/ io_len,
337 				beio->ds_tag_type,
338 				beio->ds_trans_type,
339 				/*now*/ NULL,
340 				/*then*/&beio->ds_t0);
341 
342 	if (beio->beio_cont != NULL) {
343 		beio->beio_cont(beio);
344 	} else {
345 		ctl_free_beio(beio);
346 		ctl_done(io);
347 	}
348 }
349 
350 static int
351 ctl_be_block_move_done(union ctl_io *io)
352 {
353 	struct ctl_be_block_io *beio;
354 	struct ctl_be_block_lun *be_lun;
355 #ifdef CTL_TIME_IO
356 	struct bintime cur_bt;
357 #endif
358 
359 	beio = (struct ctl_be_block_io *)
360 		io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr;
361 
362 	be_lun = beio->lun;
363 
364 	DPRINTF("entered\n");
365 
366 #ifdef CTL_TIME_IO
367 	getbintime(&cur_bt);
368 	bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
369 	bintime_add(&io->io_hdr.dma_bt, &cur_bt);
370 	io->io_hdr.num_dmas++;
371 #endif
372 
373 	/*
374 	 * We set status at this point for read commands, and write
375 	 * commands with errors.
376 	 */
377 	if ((beio->bio_cmd == BIO_READ)
378 	 && (io->io_hdr.port_status == 0)
379 	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
380 	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
381 		ctl_set_success(&io->scsiio);
382 	else if ((io->io_hdr.port_status != 0)
383 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
384 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
385 		/*
386 		 * For hardware error sense keys, the sense key
387 		 * specific value is defined to be a retry count,
388 		 * but we use it to pass back an internal FETD
389 		 * error code.  XXX KDM  Hopefully the FETD is only
390 		 * using 16 bits for an error code, since that's
391 		 * all the space we have in the sks field.
392 		 */
393 		ctl_set_internal_failure(&io->scsiio,
394 					 /*sks_valid*/ 1,
395 					 /*retry_count*/
396 					 io->io_hdr.port_status);
397 	}
398 
399 	/*
400 	 * If this is a read, or a write with errors, it is done.
401 	 */
402 	if ((beio->bio_cmd == BIO_READ)
403 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
404 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
405 		ctl_complete_beio(beio);
406 		return (0);
407 	}
408 
409 	/*
410 	 * At this point, we have a write and the DMA completed
411 	 * successfully.  We now have to queue it to the task queue to
412 	 * execute the backend I/O.  That is because we do blocking
413 	 * memory allocations, and in the file backing case, blocking I/O.
414 	 * This move done routine is generally called in the SIM's
415 	 * interrupt context, and therefore we cannot block.
416 	 */
417 	mtx_lock(&be_lun->lock);
418 	/*
419 	 * XXX KDM make sure that links is okay to use at this point.
420 	 * Otherwise, we either need to add another field to ctl_io_hdr,
421 	 * or deal with resource allocation here.
422 	 */
423 	STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
424 	mtx_unlock(&be_lun->lock);
425 
426 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
427 
428 	return (0);
429 }
430 
431 static void
432 ctl_be_block_biodone(struct bio *bio)
433 {
434 	struct ctl_be_block_io *beio;
435 	struct ctl_be_block_lun *be_lun;
436 	union ctl_io *io;
437 	int error;
438 
439 	beio = bio->bio_caller1;
440 	be_lun = beio->lun;
441 	io = beio->io;
442 
443 	DPRINTF("entered\n");
444 
445 	error = bio->bio_error;
446 	mtx_lock(&be_lun->lock);
447 	if (error != 0)
448 		beio->num_errors++;
449 
450 	beio->num_bios_done++;
451 
452 	/*
453 	 * XXX KDM will this cause WITNESS to complain?  Holding a lock
454 	 * during the free might cause it to complain.
455 	 */
456 	g_destroy_bio(bio);
457 
458 	/*
459 	 * If the send complete bit isn't set, or we aren't the last I/O to
460 	 * complete, then we're done.
461 	 */
462 	if ((beio->send_complete == 0)
463 	 || (beio->num_bios_done < beio->num_bios_sent)) {
464 		mtx_unlock(&be_lun->lock);
465 		return;
466 	}
467 
468 	/*
469 	 * At this point, we've verified that we are the last I/O to
470 	 * complete, so it's safe to drop the lock.
471 	 */
472 	mtx_unlock(&be_lun->lock);
473 
474 	/*
475 	 * If there are any errors from the backing device, we fail the
476 	 * entire I/O with a medium error.
477 	 */
478 	if (beio->num_errors > 0) {
479 		if (error == EOPNOTSUPP) {
480 			ctl_set_invalid_opcode(&io->scsiio);
481 		} else if (beio->bio_cmd == BIO_FLUSH) {
482 			/* XXX KDM is there is a better error here? */
483 			ctl_set_internal_failure(&io->scsiio,
484 						 /*sks_valid*/ 1,
485 						 /*retry_count*/ 0xbad2);
486 		} else
487 			ctl_set_medium_error(&io->scsiio);
488 		ctl_complete_beio(beio);
489 		return;
490 	}
491 
492 	/*
493 	 * If this is a write, a flush or a delete, we're all done.
494 	 * If this is a read, we can now send the data to the user.
495 	 */
496 	if ((beio->bio_cmd == BIO_WRITE)
497 	 || (beio->bio_cmd == BIO_FLUSH)
498 	 || (beio->bio_cmd == BIO_DELETE)) {
499 		ctl_set_success(&io->scsiio);
500 		ctl_complete_beio(beio);
501 	} else {
502 #ifdef CTL_TIME_IO
503         	getbintime(&io->io_hdr.dma_start_bt);
504 #endif
505 		ctl_datamove(io);
506 	}
507 }
508 
509 static void
510 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
511 			struct ctl_be_block_io *beio)
512 {
513 	union ctl_io *io;
514 	struct mount *mountpoint;
515 	int error, lock_flags;
516 
517 	DPRINTF("entered\n");
518 
519 	io = beio->io;
520 
521        	(void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
522 
523 	if (MNT_SHARED_WRITES(mountpoint)
524 	 || ((mountpoint == NULL)
525 	  && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
526 		lock_flags = LK_SHARED;
527 	else
528 		lock_flags = LK_EXCLUSIVE;
529 
530 	vn_lock(be_lun->vn, lock_flags | LK_RETRY);
531 
532 	binuptime(&beio->ds_t0);
533 	devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
534 
535 	error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread);
536 	VOP_UNLOCK(be_lun->vn, 0);
537 
538 	vn_finished_write(mountpoint);
539 
540 	if (error == 0)
541 		ctl_set_success(&io->scsiio);
542 	else {
543 		/* XXX KDM is there is a better error here? */
544 		ctl_set_internal_failure(&io->scsiio,
545 					 /*sks_valid*/ 1,
546 					 /*retry_count*/ 0xbad1);
547 	}
548 
549 	ctl_complete_beio(beio);
550 }
551 
552 SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t");
553 SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t");
554 SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t");
555 SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t");
556 
557 static void
558 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
559 			   struct ctl_be_block_io *beio)
560 {
561 	struct ctl_be_block_filedata *file_data;
562 	union ctl_io *io;
563 	struct uio xuio;
564 	struct iovec *xiovec;
565 	int flags;
566 	int error, i;
567 
568 	DPRINTF("entered\n");
569 
570 	file_data = &be_lun->backend.file;
571 	io = beio->io;
572 	flags = beio->bio_flags;
573 
574 	if (beio->bio_cmd == BIO_READ) {
575 		SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
576 	} else {
577 		SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
578 	}
579 
580 	bzero(&xuio, sizeof(xuio));
581 	if (beio->bio_cmd == BIO_READ)
582 		xuio.uio_rw = UIO_READ;
583 	else
584 		xuio.uio_rw = UIO_WRITE;
585 
586 	xuio.uio_offset = beio->io_offset;
587 	xuio.uio_resid = beio->io_len;
588 	xuio.uio_segflg = UIO_SYSSPACE;
589 	xuio.uio_iov = beio->xiovecs;
590 	xuio.uio_iovcnt = beio->num_segs;
591 	xuio.uio_td = curthread;
592 
593 	for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
594 		xiovec->iov_base = beio->sg_segs[i].addr;
595 		xiovec->iov_len = beio->sg_segs[i].len;
596 	}
597 
598 	if (beio->bio_cmd == BIO_READ) {
599 		vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
600 
601 		binuptime(&beio->ds_t0);
602 		devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
603 
604 		/*
605 		 * UFS pays attention to IO_DIRECT for reads.  If the
606 		 * DIRECTIO option is configured into the kernel, it calls
607 		 * ffs_rawread().  But that only works for single-segment
608 		 * uios with user space addresses.  In our case, with a
609 		 * kernel uio, it still reads into the buffer cache, but it
610 		 * will just try to release the buffer from the cache later
611 		 * on in ffs_read().
612 		 *
613 		 * ZFS does not pay attention to IO_DIRECT for reads.
614 		 *
615 		 * UFS does not pay attention to IO_SYNC for reads.
616 		 *
617 		 * ZFS pays attention to IO_SYNC (which translates into the
618 		 * Solaris define FRSYNC for zfs_read()) for reads.  It
619 		 * attempts to sync the file before reading.
620 		 *
621 		 * So, to attempt to provide some barrier semantics in the
622 		 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC.
623 		 */
624 		error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ?
625 				 (IO_DIRECT|IO_SYNC) : 0, file_data->cred);
626 
627 		VOP_UNLOCK(be_lun->vn, 0);
628 	} else {
629 		struct mount *mountpoint;
630 		int lock_flags;
631 
632 		(void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
633 
634 		if (MNT_SHARED_WRITES(mountpoint)
635 		 || ((mountpoint == NULL)
636 		  && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
637 			lock_flags = LK_SHARED;
638 		else
639 			lock_flags = LK_EXCLUSIVE;
640 
641 		vn_lock(be_lun->vn, lock_flags | LK_RETRY);
642 
643 		binuptime(&beio->ds_t0);
644 		devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
645 
646 		/*
647 		 * UFS pays attention to IO_DIRECT for writes.  The write
648 		 * is done asynchronously.  (Normally the write would just
649 		 * get put into cache.
650 		 *
651 		 * UFS pays attention to IO_SYNC for writes.  It will
652 		 * attempt to write the buffer out synchronously if that
653 		 * flag is set.
654 		 *
655 		 * ZFS does not pay attention to IO_DIRECT for writes.
656 		 *
657 		 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
658 		 * for writes.  It will flush the transaction from the
659 		 * cache before returning.
660 		 *
661 		 * So if we've got the BIO_ORDERED flag set, we want
662 		 * IO_SYNC in either the UFS or ZFS case.
663 		 */
664 		error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ?
665 				  IO_SYNC : 0, file_data->cred);
666 		VOP_UNLOCK(be_lun->vn, 0);
667 
668 		vn_finished_write(mountpoint);
669         }
670 
671 	/*
672 	 * If we got an error, set the sense data to "MEDIUM ERROR" and
673 	 * return the I/O to the user.
674 	 */
675 	if (error != 0) {
676 		char path_str[32];
677 
678 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
679 		/*
680 		 * XXX KDM ZFS returns ENOSPC when the underlying
681 		 * filesystem fills up.  What kind of SCSI error should we
682 		 * return for that?
683 		 */
684 		printf("%s%s command returned errno %d\n", path_str,
685 		       (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error);
686 		ctl_set_medium_error(&io->scsiio);
687 		ctl_complete_beio(beio);
688 		return;
689 	}
690 
691 	/*
692 	 * If this is a write, we're all done.
693 	 * If this is a read, we can now send the data to the user.
694 	 */
695 	if (beio->bio_cmd == BIO_WRITE) {
696 		ctl_set_success(&io->scsiio);
697 		SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
698 		ctl_complete_beio(beio);
699 	} else {
700 		SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
701 #ifdef CTL_TIME_IO
702         	getbintime(&io->io_hdr.dma_start_bt);
703 #endif
704 		ctl_datamove(io);
705 	}
706 }
707 
708 static void
709 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
710 		       struct ctl_be_block_io *beio)
711 {
712 	struct bio *bio;
713 	union ctl_io *io;
714 	struct ctl_be_block_devdata *dev_data;
715 
716 	dev_data = &be_lun->backend.dev;
717 	io = beio->io;
718 
719 	DPRINTF("entered\n");
720 
721 	/* This can't fail, it's a blocking allocation. */
722 	bio = g_alloc_bio();
723 
724 	bio->bio_cmd	    = BIO_FLUSH;
725 	bio->bio_flags	   |= BIO_ORDERED;
726 	bio->bio_dev	    = dev_data->cdev;
727 	bio->bio_offset	    = 0;
728 	bio->bio_data	    = 0;
729 	bio->bio_done	    = ctl_be_block_biodone;
730 	bio->bio_caller1    = beio;
731 	bio->bio_pblkno	    = 0;
732 
733 	/*
734 	 * We don't need to acquire the LUN lock here, because we are only
735 	 * sending one bio, and so there is no other context to synchronize
736 	 * with.
737 	 */
738 	beio->num_bios_sent = 1;
739 	beio->send_complete = 1;
740 
741 	binuptime(&beio->ds_t0);
742 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
743 
744 	(*dev_data->csw->d_strategy)(bio);
745 }
746 
747 static void
748 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
749 		       struct ctl_be_block_io *beio,
750 		       uint64_t off, uint64_t len, int last)
751 {
752 	struct bio *bio;
753 	struct ctl_be_block_devdata *dev_data;
754 	uint64_t maxlen;
755 
756 	dev_data = &be_lun->backend.dev;
757 	maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize);
758 	while (len > 0) {
759 		bio = g_alloc_bio();
760 		bio->bio_cmd	    = BIO_DELETE;
761 		bio->bio_flags	   |= beio->bio_flags;
762 		bio->bio_dev	    = dev_data->cdev;
763 		bio->bio_offset	    = off;
764 		bio->bio_length	    = MIN(len, maxlen);
765 		bio->bio_data	    = 0;
766 		bio->bio_done	    = ctl_be_block_biodone;
767 		bio->bio_caller1    = beio;
768 		bio->bio_pblkno     = off / be_lun->blocksize;
769 
770 		off += bio->bio_length;
771 		len -= bio->bio_length;
772 
773 		mtx_lock(&be_lun->lock);
774 		beio->num_bios_sent++;
775 		if (last && len == 0)
776 			beio->send_complete = 1;
777 		mtx_unlock(&be_lun->lock);
778 
779 		(*dev_data->csw->d_strategy)(bio);
780 	}
781 }
782 
783 static void
784 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
785 		       struct ctl_be_block_io *beio)
786 {
787 	union ctl_io *io;
788 	struct ctl_be_block_devdata *dev_data;
789 	struct ctl_ptr_len_flags ptrlen;
790 	struct scsi_unmap_desc *buf, *end;
791 	uint64_t len;
792 
793 	dev_data = &be_lun->backend.dev;
794 	io = beio->io;
795 
796 	DPRINTF("entered\n");
797 
798 	binuptime(&beio->ds_t0);
799 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
800 
801 	if (beio->io_offset == -1) {
802 		beio->io_len = 0;
803 		memcpy(&ptrlen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
804 		       sizeof(ptrlen));
805 		buf = (struct scsi_unmap_desc *)ptrlen.ptr;
806 		end = buf + ptrlen.len / sizeof(*buf);
807 		for (; buf < end; buf++) {
808 			len = (uint64_t)scsi_4btoul(buf->length) *
809 			    be_lun->blocksize;
810 			beio->io_len += len;
811 			ctl_be_block_unmap_dev_range(be_lun, beio,
812 			    scsi_8btou64(buf->lba) * be_lun->blocksize, len,
813 			    (end - buf < 2) ? TRUE : FALSE);
814 		}
815 	} else
816 		ctl_be_block_unmap_dev_range(be_lun, beio,
817 		    beio->io_offset, beio->io_len, TRUE);
818 }
819 
820 static void
821 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
822 			  struct ctl_be_block_io *beio)
823 {
824 	int i;
825 	struct bio *bio;
826 	struct ctl_be_block_devdata *dev_data;
827 	off_t cur_offset;
828 	int max_iosize;
829 
830 	DPRINTF("entered\n");
831 
832 	dev_data = &be_lun->backend.dev;
833 
834 	/*
835 	 * We have to limit our I/O size to the maximum supported by the
836 	 * backend device.  Hopefully it is MAXPHYS.  If the driver doesn't
837 	 * set it properly, use DFLTPHYS.
838 	 */
839 	max_iosize = dev_data->cdev->si_iosize_max;
840 	if (max_iosize < PAGE_SIZE)
841 		max_iosize = DFLTPHYS;
842 
843 	cur_offset = beio->io_offset;
844 
845 	/*
846 	 * XXX KDM need to accurately reflect the number of I/Os outstanding
847 	 * to a device.
848 	 */
849 	binuptime(&beio->ds_t0);
850 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
851 
852 	for (i = 0; i < beio->num_segs; i++) {
853 		size_t cur_size;
854 		uint8_t *cur_ptr;
855 
856 		cur_size = beio->sg_segs[i].len;
857 		cur_ptr = beio->sg_segs[i].addr;
858 
859 		while (cur_size > 0) {
860 			/* This can't fail, it's a blocking allocation. */
861 			bio = g_alloc_bio();
862 
863 			KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
864 
865 			bio->bio_cmd = beio->bio_cmd;
866 			bio->bio_flags |= beio->bio_flags;
867 			bio->bio_dev = dev_data->cdev;
868 			bio->bio_caller1 = beio;
869 			bio->bio_length = min(cur_size, max_iosize);
870 			bio->bio_offset = cur_offset;
871 			bio->bio_data = cur_ptr;
872 			bio->bio_done = ctl_be_block_biodone;
873 			bio->bio_pblkno = cur_offset / be_lun->blocksize;
874 
875 			cur_offset += bio->bio_length;
876 			cur_ptr += bio->bio_length;
877 			cur_size -= bio->bio_length;
878 
879 			/*
880 			 * Make sure we set the complete bit just before we
881 			 * issue the last bio so we don't wind up with a
882 			 * race.
883 			 *
884 			 * Use the LUN mutex here instead of a combination
885 			 * of atomic variables for simplicity.
886 			 *
887 			 * XXX KDM we could have a per-IO lock, but that
888 			 * would cause additional per-IO setup and teardown
889 			 * overhead.  Hopefully there won't be too much
890 			 * contention on the LUN lock.
891 			 */
892 			mtx_lock(&be_lun->lock);
893 
894 			beio->num_bios_sent++;
895 
896 			if ((i == beio->num_segs - 1)
897 			 && (cur_size == 0))
898 				beio->send_complete = 1;
899 
900 			mtx_unlock(&be_lun->lock);
901 
902 			(*dev_data->csw->d_strategy)(bio);
903 		}
904 	}
905 }
906 
907 static void
908 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
909 {
910 	union ctl_io *io;
911 
912 	io = beio->io;
913 	ctl_free_beio(beio);
914 	if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)
915 	  && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
916 		ctl_config_write_done(io);
917 		return;
918 	}
919 
920 	ctl_be_block_config_write(io);
921 }
922 
923 static void
924 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
925 			    union ctl_io *io)
926 {
927 	struct ctl_be_block_io *beio;
928 	struct ctl_be_block_softc *softc;
929 	struct ctl_lba_len_flags lbalen;
930 	uint64_t len_left, lba;
931 	int i, seglen;
932 	uint8_t *buf, *end;
933 
934 	DPRINTF("entered\n");
935 
936 	beio = io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr;
937 	softc = be_lun->softc;
938 	memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
939 	       sizeof(lbalen));
940 
941 	if (lbalen.flags & ~(SWS_LBDATA | SWS_UNMAP) ||
942 	    (lbalen.flags & SWS_UNMAP && be_lun->unmap == NULL)) {
943 		ctl_free_beio(beio);
944 		ctl_set_invalid_field(&io->scsiio,
945 				      /*sks_valid*/ 1,
946 				      /*command*/ 1,
947 				      /*field*/ 1,
948 				      /*bit_valid*/ 0,
949 				      /*bit*/ 0);
950 		ctl_config_write_done(io);
951 		return;
952 	}
953 
954 	/*
955 	 * If the I/O came down with an ordered or head of queue tag, set
956 	 * the BIO_ORDERED attribute.  For head of queue tags, that's
957 	 * pretty much the best we can do.
958 	 */
959 	if ((io->scsiio.tag_type == CTL_TAG_ORDERED)
960 	 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
961 		beio->bio_flags = BIO_ORDERED;
962 
963 	switch (io->scsiio.tag_type) {
964 	case CTL_TAG_ORDERED:
965 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
966 		break;
967 	case CTL_TAG_HEAD_OF_QUEUE:
968 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
969 		break;
970 	case CTL_TAG_UNTAGGED:
971 	case CTL_TAG_SIMPLE:
972 	case CTL_TAG_ACA:
973 	default:
974 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
975 		break;
976 	}
977 
978 	if (lbalen.flags & SWS_UNMAP) {
979 		beio->io_offset = lbalen.lba * be_lun->blocksize;
980 		beio->io_len = (uint64_t)lbalen.len * be_lun->blocksize;
981 		beio->bio_cmd = BIO_DELETE;
982 		beio->ds_trans_type = DEVSTAT_FREE;
983 
984 		be_lun->unmap(be_lun, beio);
985 		return;
986 	}
987 
988 	beio->bio_cmd = BIO_WRITE;
989 	beio->ds_trans_type = DEVSTAT_WRITE;
990 
991 	DPRINTF("WRITE SAME at LBA %jx len %u\n",
992 	       (uintmax_t)lbalen.lba, lbalen.len);
993 
994 	len_left = (uint64_t)lbalen.len * be_lun->blocksize;
995 	for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
996 
997 		/*
998 		 * Setup the S/G entry for this chunk.
999 		 */
1000 		seglen = MIN(CTLBLK_MAX_SEG, len_left);
1001 		seglen -= seglen % be_lun->blocksize;
1002 		beio->sg_segs[i].len = seglen;
1003 		beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1004 
1005 		DPRINTF("segment %d addr %p len %zd\n", i,
1006 			beio->sg_segs[i].addr, beio->sg_segs[i].len);
1007 
1008 		beio->num_segs++;
1009 		len_left -= seglen;
1010 
1011 		buf = beio->sg_segs[i].addr;
1012 		end = buf + seglen;
1013 		for (; buf < end; buf += be_lun->blocksize) {
1014 			memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize);
1015 			if (lbalen.flags & SWS_LBDATA)
1016 				scsi_ulto4b(lbalen.lba + lba, buf);
1017 			lba++;
1018 		}
1019 	}
1020 
1021 	beio->io_offset = lbalen.lba * be_lun->blocksize;
1022 	beio->io_len = lba * be_lun->blocksize;
1023 
1024 	/* We can not do all in one run. Correct and schedule rerun. */
1025 	if (len_left > 0) {
1026 		lbalen.lba += lba;
1027 		lbalen.len -= lba;
1028 		memcpy(io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen,
1029 		       sizeof(lbalen));
1030 		beio->beio_cont = ctl_be_block_cw_done_ws;
1031 	}
1032 
1033 	be_lun->dispatch(be_lun, beio);
1034 }
1035 
1036 static void
1037 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1038 			    union ctl_io *io)
1039 {
1040 	struct ctl_be_block_io *beio;
1041 	struct ctl_be_block_softc *softc;
1042 	struct ctl_ptr_len_flags ptrlen;
1043 
1044 	DPRINTF("entered\n");
1045 
1046 	beio = io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr;
1047 	softc = be_lun->softc;
1048 	memcpy(&ptrlen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
1049 	       sizeof(ptrlen));
1050 
1051 	if (ptrlen.flags != 0 || be_lun->unmap == NULL) {
1052 		ctl_free_beio(beio);
1053 		ctl_set_invalid_field(&io->scsiio,
1054 				      /*sks_valid*/ 0,
1055 				      /*command*/ 1,
1056 				      /*field*/ 0,
1057 				      /*bit_valid*/ 0,
1058 				      /*bit*/ 0);
1059 		ctl_config_write_done(io);
1060 		return;
1061 	}
1062 
1063 	/*
1064 	 * If the I/O came down with an ordered or head of queue tag, set
1065 	 * the BIO_ORDERED attribute.  For head of queue tags, that's
1066 	 * pretty much the best we can do.
1067 	 */
1068 	if ((io->scsiio.tag_type == CTL_TAG_ORDERED)
1069 	 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
1070 		beio->bio_flags = BIO_ORDERED;
1071 
1072 	switch (io->scsiio.tag_type) {
1073 	case CTL_TAG_ORDERED:
1074 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1075 		break;
1076 	case CTL_TAG_HEAD_OF_QUEUE:
1077 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1078 		break;
1079 	case CTL_TAG_UNTAGGED:
1080 	case CTL_TAG_SIMPLE:
1081 	case CTL_TAG_ACA:
1082 	default:
1083 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1084 		break;
1085 	}
1086 
1087 	beio->io_len = 0;
1088 	beio->io_offset = -1;
1089 
1090 	beio->bio_cmd = BIO_DELETE;
1091 	beio->ds_trans_type = DEVSTAT_FREE;
1092 
1093 	DPRINTF("WRITE SAME at LBA %jx len %u\n",
1094 	       (uintmax_t)lbalen.lba, lbalen.len);
1095 
1096 	be_lun->unmap(be_lun, beio);
1097 }
1098 
1099 static void
1100 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1101 {
1102 	union ctl_io *io;
1103 
1104 	io = beio->io;
1105 	ctl_free_beio(beio);
1106 	ctl_config_write_done(io);
1107 }
1108 
1109 static void
1110 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1111 			 union ctl_io *io)
1112 {
1113 	struct ctl_be_block_io *beio;
1114 	struct ctl_be_block_softc *softc;
1115 
1116 	DPRINTF("entered\n");
1117 
1118 	softc = be_lun->softc;
1119 	beio = ctl_alloc_beio(softc);
1120 	beio->io = io;
1121 	beio->lun = be_lun;
1122 	beio->beio_cont = ctl_be_block_cw_done;
1123 	io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio;
1124 
1125 	switch (io->scsiio.cdb[0]) {
1126 	case SYNCHRONIZE_CACHE:
1127 	case SYNCHRONIZE_CACHE_16:
1128 		beio->bio_cmd = BIO_FLUSH;
1129 		beio->ds_trans_type = DEVSTAT_NO_DATA;
1130 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1131 		beio->io_len = 0;
1132 		be_lun->lun_flush(be_lun, beio);
1133 		break;
1134 	case WRITE_SAME_10:
1135 	case WRITE_SAME_16:
1136 		ctl_be_block_cw_dispatch_ws(be_lun, io);
1137 		break;
1138 	case UNMAP:
1139 		ctl_be_block_cw_dispatch_unmap(be_lun, io);
1140 		break;
1141 	default:
1142 		panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1143 		break;
1144 	}
1145 }
1146 
1147 SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t");
1148 SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t");
1149 SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t");
1150 SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t");
1151 
1152 static void
1153 ctl_be_block_next(struct ctl_be_block_io *beio)
1154 {
1155 	struct ctl_be_block_lun *be_lun;
1156 	union ctl_io *io;
1157 
1158 	io = beio->io;
1159 	be_lun = beio->lun;
1160 	ctl_free_beio(beio);
1161 	if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)
1162 	  && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1163 		ctl_done(io);
1164 		return;
1165 	}
1166 
1167 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
1168 	io->io_hdr.status &= ~CTL_STATUS_MASK;
1169 	io->io_hdr.status |= CTL_STATUS_NONE;
1170 
1171 	mtx_lock(&be_lun->lock);
1172 	/*
1173 	 * XXX KDM make sure that links is okay to use at this point.
1174 	 * Otherwise, we either need to add another field to ctl_io_hdr,
1175 	 * or deal with resource allocation here.
1176 	 */
1177 	STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1178 	mtx_unlock(&be_lun->lock);
1179 
1180 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1181 }
1182 
1183 static void
1184 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1185 			   union ctl_io *io)
1186 {
1187 	struct ctl_be_block_io *beio;
1188 	struct ctl_be_block_softc *softc;
1189 	struct ctl_lba_len lbalen;
1190 	uint64_t len_left, lbaoff;
1191 	int i;
1192 
1193 	softc = be_lun->softc;
1194 
1195 	DPRINTF("entered\n");
1196 
1197 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) {
1198 		SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0);
1199 	} else {
1200 		SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0);
1201 	}
1202 
1203 	beio = ctl_alloc_beio(softc);
1204 	beio->io = io;
1205 	beio->lun = be_lun;
1206 	io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio;
1207 
1208 	/*
1209 	 * If the I/O came down with an ordered or head of queue tag, set
1210 	 * the BIO_ORDERED attribute.  For head of queue tags, that's
1211 	 * pretty much the best we can do.
1212 	 *
1213 	 * XXX KDM we don't have a great way to easily know about the FUA
1214 	 * bit right now (it is decoded in ctl_read_write(), but we don't
1215 	 * pass that knowledge to the backend), and in any case we would
1216 	 * need to determine how to handle it.
1217 	 */
1218 	if ((io->scsiio.tag_type == CTL_TAG_ORDERED)
1219 	 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
1220 		beio->bio_flags = BIO_ORDERED;
1221 
1222 	switch (io->scsiio.tag_type) {
1223 	case CTL_TAG_ORDERED:
1224 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1225 		break;
1226 	case CTL_TAG_HEAD_OF_QUEUE:
1227 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1228 		break;
1229 	case CTL_TAG_UNTAGGED:
1230 	case CTL_TAG_SIMPLE:
1231 	case CTL_TAG_ACA:
1232 	default:
1233 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1234 		break;
1235 	}
1236 
1237 	/*
1238 	 * This path handles read and write only.  The config write path
1239 	 * handles flush operations.
1240 	 */
1241 	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) {
1242 		beio->bio_cmd = BIO_READ;
1243 		beio->ds_trans_type = DEVSTAT_READ;
1244 	} else {
1245 		beio->bio_cmd = BIO_WRITE;
1246 		beio->ds_trans_type = DEVSTAT_WRITE;
1247 	}
1248 
1249 	memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
1250 	       sizeof(lbalen));
1251 	DPRINTF("%s at LBA %jx len %u @%ju\n",
1252 	       (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1253 	       (uintmax_t)lbalen.lba, lbalen.len, lbaoff);
1254 	lbaoff = io->scsiio.kern_rel_offset / be_lun->blocksize;
1255 	beio->io_offset = (lbalen.lba + lbaoff) * be_lun->blocksize;
1256 	beio->io_len = MIN((lbalen.len - lbaoff) * be_lun->blocksize,
1257 	    CTLBLK_MAX_IO_SIZE);
1258 	beio->io_len -= beio->io_len % be_lun->blocksize;
1259 
1260 	for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1261 		KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1262 		    i, CTLBLK_MAX_SEGS));
1263 
1264 		/*
1265 		 * Setup the S/G entry for this chunk.
1266 		 */
1267 		beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1268 		beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1269 
1270 		DPRINTF("segment %d addr %p len %zd\n", i,
1271 			beio->sg_segs[i].addr, beio->sg_segs[i].len);
1272 
1273 		beio->num_segs++;
1274 		len_left -= beio->sg_segs[i].len;
1275 	}
1276 	if (io->scsiio.kern_rel_offset + beio->io_len <
1277 	    io->scsiio.kern_total_len)
1278 		beio->beio_cont = ctl_be_block_next;
1279 	io->scsiio.be_move_done = ctl_be_block_move_done;
1280 	io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1281 	io->scsiio.kern_data_len = beio->io_len;
1282 	io->scsiio.kern_data_resid = 0;
1283 	io->scsiio.kern_sg_entries = beio->num_segs;
1284 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST;
1285 
1286 	/*
1287 	 * For the read case, we need to read the data into our buffers and
1288 	 * then we can send it back to the user.  For the write case, we
1289 	 * need to get the data from the user first.
1290 	 */
1291 	if (beio->bio_cmd == BIO_READ) {
1292 		SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0);
1293 		be_lun->dispatch(be_lun, beio);
1294 	} else {
1295 		SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0);
1296 #ifdef CTL_TIME_IO
1297         	getbintime(&io->io_hdr.dma_start_bt);
1298 #endif
1299 		ctl_datamove(io);
1300 	}
1301 }
1302 
1303 static void
1304 ctl_be_block_worker(void *context, int pending)
1305 {
1306 	struct ctl_be_block_lun *be_lun;
1307 	struct ctl_be_block_softc *softc;
1308 	union ctl_io *io;
1309 
1310 	be_lun = (struct ctl_be_block_lun *)context;
1311 	softc = be_lun->softc;
1312 
1313 	DPRINTF("entered\n");
1314 
1315 	mtx_lock(&be_lun->lock);
1316 	for (;;) {
1317 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1318 		if (io != NULL) {
1319 			struct ctl_be_block_io *beio;
1320 
1321 			DPRINTF("datamove queue\n");
1322 
1323 			STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1324 				      ctl_io_hdr, links);
1325 
1326 			mtx_unlock(&be_lun->lock);
1327 
1328 			beio = (struct ctl_be_block_io *)
1329 			    io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr;
1330 
1331 			be_lun->dispatch(be_lun, beio);
1332 
1333 			mtx_lock(&be_lun->lock);
1334 			continue;
1335 		}
1336 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1337 		if (io != NULL) {
1338 
1339 			DPRINTF("config write queue\n");
1340 
1341 			STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1342 				      ctl_io_hdr, links);
1343 
1344 			mtx_unlock(&be_lun->lock);
1345 
1346 			ctl_be_block_cw_dispatch(be_lun, io);
1347 
1348 			mtx_lock(&be_lun->lock);
1349 			continue;
1350 		}
1351 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1352 		if (io != NULL) {
1353 			DPRINTF("input queue\n");
1354 
1355 			STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1356 				      ctl_io_hdr, links);
1357 			mtx_unlock(&be_lun->lock);
1358 
1359 			/*
1360 			 * We must drop the lock, since this routine and
1361 			 * its children may sleep.
1362 			 */
1363 			ctl_be_block_dispatch(be_lun, io);
1364 
1365 			mtx_lock(&be_lun->lock);
1366 			continue;
1367 		}
1368 
1369 		/*
1370 		 * If we get here, there is no work left in the queues, so
1371 		 * just break out and let the task queue go to sleep.
1372 		 */
1373 		break;
1374 	}
1375 	mtx_unlock(&be_lun->lock);
1376 }
1377 
1378 /*
1379  * Entry point from CTL to the backend for I/O.  We queue everything to a
1380  * work thread, so this just puts the I/O on a queue and wakes up the
1381  * thread.
1382  */
1383 static int
1384 ctl_be_block_submit(union ctl_io *io)
1385 {
1386 	struct ctl_lba_len lbalen;
1387 	struct ctl_be_block_lun *be_lun;
1388 	struct ctl_be_lun *ctl_be_lun;
1389 	int retval;
1390 
1391 	DPRINTF("entered\n");
1392 
1393 	retval = CTL_RETVAL_COMPLETE;
1394 
1395 	ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
1396 		CTL_PRIV_BACKEND_LUN].ptr;
1397 	be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
1398 
1399 	/*
1400 	 * Make sure we only get SCSI I/O.
1401 	 */
1402 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1403 		"%#x) encountered", io->io_hdr.io_type));
1404 
1405 	memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
1406 	       sizeof(lbalen));
1407 	io->scsiio.kern_total_len = lbalen.len * be_lun->blocksize;
1408 	io->scsiio.kern_rel_offset = 0;
1409 
1410 	mtx_lock(&be_lun->lock);
1411 	/*
1412 	 * XXX KDM make sure that links is okay to use at this point.
1413 	 * Otherwise, we either need to add another field to ctl_io_hdr,
1414 	 * or deal with resource allocation here.
1415 	 */
1416 	STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1417 	mtx_unlock(&be_lun->lock);
1418 
1419 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1420 
1421 	return (retval);
1422 }
1423 
1424 static int
1425 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1426 			int flag, struct thread *td)
1427 {
1428 	struct ctl_be_block_softc *softc;
1429 	int error;
1430 
1431 	softc = &backend_block_softc;
1432 
1433 	error = 0;
1434 
1435 	switch (cmd) {
1436 	case CTL_LUN_REQ: {
1437 		struct ctl_lun_req *lun_req;
1438 
1439 		lun_req = (struct ctl_lun_req *)addr;
1440 
1441 		switch (lun_req->reqtype) {
1442 		case CTL_LUNREQ_CREATE:
1443 			error = ctl_be_block_create(softc, lun_req);
1444 			break;
1445 		case CTL_LUNREQ_RM:
1446 			error = ctl_be_block_rm(softc, lun_req);
1447 			break;
1448 		case CTL_LUNREQ_MODIFY:
1449 			error = ctl_be_block_modify(softc, lun_req);
1450 			break;
1451 		default:
1452 			lun_req->status = CTL_LUN_ERROR;
1453 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1454 				 "%s: invalid LUN request type %d", __func__,
1455 				 lun_req->reqtype);
1456 			break;
1457 		}
1458 		break;
1459 	}
1460 	default:
1461 		error = ENOTTY;
1462 		break;
1463 	}
1464 
1465 	return (error);
1466 }
1467 
1468 static int
1469 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1470 {
1471 	struct ctl_be_block_filedata *file_data;
1472 	struct ctl_lun_create_params *params;
1473 	struct vattr		      vattr;
1474 	int			      error;
1475 
1476 	error = 0;
1477 	file_data = &be_lun->backend.file;
1478 	params = &req->reqdata.create;
1479 
1480 	be_lun->dev_type = CTL_BE_BLOCK_FILE;
1481 	be_lun->dispatch = ctl_be_block_dispatch_file;
1482 	be_lun->lun_flush = ctl_be_block_flush_file;
1483 
1484 	error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1485 	if (error != 0) {
1486 		snprintf(req->error_str, sizeof(req->error_str),
1487 			 "error calling VOP_GETATTR() for file %s",
1488 			 be_lun->dev_path);
1489 		return (error);
1490 	}
1491 
1492 	/*
1493 	 * Verify that we have the ability to upgrade to exclusive
1494 	 * access on this file so we can trap errors at open instead
1495 	 * of reporting them during first access.
1496 	 */
1497 	if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) {
1498 		vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY);
1499 		if (be_lun->vn->v_iflag & VI_DOOMED) {
1500 			error = EBADF;
1501 			snprintf(req->error_str, sizeof(req->error_str),
1502 				 "error locking file %s", be_lun->dev_path);
1503 			return (error);
1504 		}
1505 	}
1506 
1507 
1508 	file_data->cred = crhold(curthread->td_ucred);
1509 	if (params->lun_size_bytes != 0)
1510 		be_lun->size_bytes = params->lun_size_bytes;
1511 	else
1512 		be_lun->size_bytes = vattr.va_size;
1513 	/*
1514 	 * We set the multi thread flag for file operations because all
1515 	 * filesystems (in theory) are capable of allowing multiple readers
1516 	 * of a file at once.  So we want to get the maximum possible
1517 	 * concurrency.
1518 	 */
1519 	be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD;
1520 
1521 	/*
1522 	 * XXX KDM vattr.va_blocksize may be larger than 512 bytes here.
1523 	 * With ZFS, it is 131072 bytes.  Block sizes that large don't work
1524 	 * with disklabel and UFS on FreeBSD at least.  Large block sizes
1525 	 * may not work with other OSes as well.  So just export a sector
1526 	 * size of 512 bytes, which should work with any OS or
1527 	 * application.  Since our backing is a file, any block size will
1528 	 * work fine for the backing store.
1529 	 */
1530 #if 0
1531 	be_lun->blocksize= vattr.va_blocksize;
1532 #endif
1533 	if (params->blocksize_bytes != 0)
1534 		be_lun->blocksize = params->blocksize_bytes;
1535 	else
1536 		be_lun->blocksize = 512;
1537 
1538 	/*
1539 	 * Sanity check.  The media size has to be at least one
1540 	 * sector long.
1541 	 */
1542 	if (be_lun->size_bytes < be_lun->blocksize) {
1543 		error = EINVAL;
1544 		snprintf(req->error_str, sizeof(req->error_str),
1545 			 "file %s size %ju < block size %u", be_lun->dev_path,
1546 			 (uintmax_t)be_lun->size_bytes, be_lun->blocksize);
1547 	}
1548 	return (error);
1549 }
1550 
1551 static int
1552 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1553 {
1554 	struct ctl_lun_create_params *params;
1555 	struct vattr		      vattr;
1556 	struct cdev		     *dev;
1557 	struct cdevsw		     *devsw;
1558 	int			      error;
1559 	off_t			      ps, pss, po, pos;
1560 
1561 	params = &req->reqdata.create;
1562 
1563 	be_lun->dev_type = CTL_BE_BLOCK_DEV;
1564 	be_lun->dispatch = ctl_be_block_dispatch_dev;
1565 	be_lun->lun_flush = ctl_be_block_flush_dev;
1566 	be_lun->unmap = ctl_be_block_unmap_dev;
1567 	be_lun->backend.dev.cdev = be_lun->vn->v_rdev;
1568 	be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev,
1569 					     &be_lun->backend.dev.dev_ref);
1570 	if (be_lun->backend.dev.csw == NULL)
1571 		panic("Unable to retrieve device switch");
1572 
1573 	error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED);
1574 	if (error) {
1575 		snprintf(req->error_str, sizeof(req->error_str),
1576 			 "%s: error getting vnode attributes for device %s",
1577 			 __func__, be_lun->dev_path);
1578 		return (error);
1579 	}
1580 
1581 	dev = be_lun->vn->v_rdev;
1582 	devsw = dev->si_devsw;
1583 	if (!devsw->d_ioctl) {
1584 		snprintf(req->error_str, sizeof(req->error_str),
1585 			 "%s: no d_ioctl for device %s!", __func__,
1586 			 be_lun->dev_path);
1587 		return (ENODEV);
1588 	}
1589 
1590 	error = devsw->d_ioctl(dev, DIOCGSECTORSIZE,
1591 			       (caddr_t)&be_lun->blocksize, FREAD,
1592 			       curthread);
1593 	if (error) {
1594 		snprintf(req->error_str, sizeof(req->error_str),
1595 			 "%s: error %d returned for DIOCGSECTORSIZE ioctl "
1596 			 "on %s!", __func__, error, be_lun->dev_path);
1597 		return (error);
1598 	}
1599 
1600 	/*
1601 	 * If the user has asked for a blocksize that is greater than the
1602 	 * backing device's blocksize, we can do it only if the blocksize
1603 	 * the user is asking for is an even multiple of the underlying
1604 	 * device's blocksize.
1605 	 */
1606 	if ((params->blocksize_bytes != 0)
1607 	 && (params->blocksize_bytes > be_lun->blocksize)) {
1608 		uint32_t bs_multiple, tmp_blocksize;
1609 
1610 		bs_multiple = params->blocksize_bytes / be_lun->blocksize;
1611 
1612 		tmp_blocksize = bs_multiple * be_lun->blocksize;
1613 
1614 		if (tmp_blocksize == params->blocksize_bytes) {
1615 			be_lun->blocksize = params->blocksize_bytes;
1616 		} else {
1617 			snprintf(req->error_str, sizeof(req->error_str),
1618 				 "%s: requested blocksize %u is not an even "
1619 				 "multiple of backing device blocksize %u",
1620 				 __func__, params->blocksize_bytes,
1621 				 be_lun->blocksize);
1622 			return (EINVAL);
1623 
1624 		}
1625 	} else if ((params->blocksize_bytes != 0)
1626 		&& (params->blocksize_bytes != be_lun->blocksize)) {
1627 		snprintf(req->error_str, sizeof(req->error_str),
1628 			 "%s: requested blocksize %u < backing device "
1629 			 "blocksize %u", __func__, params->blocksize_bytes,
1630 			 be_lun->blocksize);
1631 		return (EINVAL);
1632 	}
1633 
1634 	error = devsw->d_ioctl(dev, DIOCGMEDIASIZE,
1635 			       (caddr_t)&be_lun->size_bytes, FREAD,
1636 			       curthread);
1637 	if (error) {
1638 		snprintf(req->error_str, sizeof(req->error_str),
1639 			 "%s: error %d returned for DIOCGMEDIASIZE "
1640 			 " ioctl on %s!", __func__, error,
1641 			 be_lun->dev_path);
1642 		return (error);
1643 	}
1644 
1645 	if (params->lun_size_bytes != 0) {
1646 		if (params->lun_size_bytes > be_lun->size_bytes) {
1647 			snprintf(req->error_str, sizeof(req->error_str),
1648 				 "%s: requested LUN size %ju > backing device "
1649 				 "size %ju", __func__,
1650 				 (uintmax_t)params->lun_size_bytes,
1651 				 (uintmax_t)be_lun->size_bytes);
1652 			return (EINVAL);
1653 		}
1654 
1655 		be_lun->size_bytes = params->lun_size_bytes;
1656 	}
1657 
1658 	error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE,
1659 			       (caddr_t)&ps, FREAD, curthread);
1660 	if (error)
1661 		ps = po = 0;
1662 	else {
1663 		error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET,
1664 				       (caddr_t)&po, FREAD, curthread);
1665 		if (error)
1666 			po = 0;
1667 	}
1668 	pss = ps / be_lun->blocksize;
1669 	pos = po / be_lun->blocksize;
1670 	if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) &&
1671 	    ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) {
1672 		be_lun->pblockexp = fls(pss) - 1;
1673 		be_lun->pblockoff = (pss - pos) % pss;
1674 	}
1675 
1676 	return (0);
1677 }
1678 
1679 static int
1680 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
1681 {
1682 	DROP_GIANT();
1683 	if (be_lun->vn) {
1684 		int flags = FREAD | FWRITE;
1685 
1686 		switch (be_lun->dev_type) {
1687 		case CTL_BE_BLOCK_DEV:
1688 			if (be_lun->backend.dev.csw) {
1689 				dev_relthread(be_lun->backend.dev.cdev,
1690 					      be_lun->backend.dev.dev_ref);
1691 				be_lun->backend.dev.csw  = NULL;
1692 				be_lun->backend.dev.cdev = NULL;
1693 			}
1694 			break;
1695 		case CTL_BE_BLOCK_FILE:
1696 			break;
1697 		case CTL_BE_BLOCK_NONE:
1698 			break;
1699 		default:
1700 			panic("Unexpected backend type.");
1701 			break;
1702 		}
1703 
1704 		(void)vn_close(be_lun->vn, flags, NOCRED, curthread);
1705 		be_lun->vn = NULL;
1706 
1707 		switch (be_lun->dev_type) {
1708 		case CTL_BE_BLOCK_DEV:
1709 			break;
1710 		case CTL_BE_BLOCK_FILE:
1711 			if (be_lun->backend.file.cred != NULL) {
1712 				crfree(be_lun->backend.file.cred);
1713 				be_lun->backend.file.cred = NULL;
1714 			}
1715 			break;
1716 		case CTL_BE_BLOCK_NONE:
1717 			break;
1718 		default:
1719 			panic("Unexpected backend type.");
1720 			break;
1721 		}
1722 	}
1723 	PICKUP_GIANT();
1724 
1725 	return (0);
1726 }
1727 
1728 static int
1729 ctl_be_block_open(struct ctl_be_block_softc *softc,
1730 		       struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1731 {
1732 	struct nameidata nd;
1733 	int		 flags;
1734 	int		 error;
1735 
1736 	/*
1737 	 * XXX KDM allow a read-only option?
1738 	 */
1739 	flags = FREAD | FWRITE;
1740 	error = 0;
1741 
1742 	if (rootvnode == NULL) {
1743 		snprintf(req->error_str, sizeof(req->error_str),
1744 			 "%s: Root filesystem is not mounted", __func__);
1745 		return (1);
1746 	}
1747 
1748 	if (!curthread->td_proc->p_fd->fd_cdir) {
1749 		curthread->td_proc->p_fd->fd_cdir = rootvnode;
1750 		VREF(rootvnode);
1751 	}
1752 	if (!curthread->td_proc->p_fd->fd_rdir) {
1753 		curthread->td_proc->p_fd->fd_rdir = rootvnode;
1754 		VREF(rootvnode);
1755 	}
1756 	if (!curthread->td_proc->p_fd->fd_jdir) {
1757 		curthread->td_proc->p_fd->fd_jdir = rootvnode;
1758 		VREF(rootvnode);
1759 	}
1760 
1761  again:
1762 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread);
1763 	error = vn_open(&nd, &flags, 0, NULL);
1764 	if (error) {
1765 		/*
1766 		 * This is the only reasonable guess we can make as far as
1767 		 * path if the user doesn't give us a fully qualified path.
1768 		 * If they want to specify a file, they need to specify the
1769 		 * full path.
1770 		 */
1771 		if (be_lun->dev_path[0] != '/') {
1772 			char *dev_path = "/dev/";
1773 			char *dev_name;
1774 
1775 			/* Try adding device path at beginning of name */
1776 			dev_name = malloc(strlen(be_lun->dev_path)
1777 					+ strlen(dev_path) + 1,
1778 					  M_CTLBLK, M_WAITOK);
1779 			if (dev_name) {
1780 				sprintf(dev_name, "%s%s", dev_path,
1781 					be_lun->dev_path);
1782 				free(be_lun->dev_path, M_CTLBLK);
1783 				be_lun->dev_path = dev_name;
1784 				goto again;
1785 			}
1786 		}
1787 		snprintf(req->error_str, sizeof(req->error_str),
1788 			 "%s: error opening %s", __func__, be_lun->dev_path);
1789 		return (error);
1790 	}
1791 
1792 	NDFREE(&nd, NDF_ONLY_PNBUF);
1793 
1794 	be_lun->vn = nd.ni_vp;
1795 
1796 	/* We only support disks and files. */
1797 	if (vn_isdisk(be_lun->vn, &error)) {
1798 		error = ctl_be_block_open_dev(be_lun, req);
1799 	} else if (be_lun->vn->v_type == VREG) {
1800 		error = ctl_be_block_open_file(be_lun, req);
1801 	} else {
1802 		error = EINVAL;
1803 		snprintf(req->error_str, sizeof(req->error_str),
1804 			 "%s is not a disk or plain file", be_lun->dev_path);
1805 	}
1806 	VOP_UNLOCK(be_lun->vn, 0);
1807 
1808 	if (error != 0) {
1809 		ctl_be_block_close(be_lun);
1810 		return (error);
1811 	}
1812 
1813 	be_lun->blocksize_shift = fls(be_lun->blocksize) - 1;
1814 	be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift;
1815 
1816 	return (0);
1817 }
1818 
1819 static int
1820 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
1821 {
1822 	struct ctl_be_block_lun *be_lun;
1823 	struct ctl_lun_create_params *params;
1824 	struct ctl_be_arg *file_arg;
1825 	char tmpstr[32];
1826 	int retval, num_threads, unmap;
1827 	int i;
1828 
1829 	params = &req->reqdata.create;
1830 	retval = 0;
1831 
1832 	num_threads = cbb_num_threads;
1833 
1834 	file_arg = NULL;
1835 
1836 	be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
1837 
1838 	be_lun->softc = softc;
1839 	STAILQ_INIT(&be_lun->input_queue);
1840 	STAILQ_INIT(&be_lun->config_write_queue);
1841 	STAILQ_INIT(&be_lun->datamove_queue);
1842 	STAILQ_INIT(&be_lun->ctl_be_lun.options);
1843 	sprintf(be_lun->lunname, "cblk%d", softc->num_luns);
1844 	mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF);
1845 
1846 	be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG,
1847 	    NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
1848 
1849 	if (be_lun->lun_zone == NULL) {
1850 		snprintf(req->error_str, sizeof(req->error_str),
1851 			 "%s: error allocating UMA zone", __func__);
1852 		goto bailout_error;
1853 	}
1854 
1855 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
1856 		be_lun->ctl_be_lun.lun_type = params->device_type;
1857 	else
1858 		be_lun->ctl_be_lun.lun_type = T_DIRECT;
1859 
1860 	if (be_lun->ctl_be_lun.lun_type == T_DIRECT) {
1861 		for (i = 0; i < req->num_be_args; i++) {
1862 			if (strcmp(req->kern_be_args[i].kname, "file") == 0) {
1863 				file_arg = &req->kern_be_args[i];
1864 				break;
1865 			}
1866 		}
1867 
1868 		if (file_arg == NULL) {
1869 			snprintf(req->error_str, sizeof(req->error_str),
1870 				 "%s: no file argument specified", __func__);
1871 			goto bailout_error;
1872 		}
1873 
1874 		be_lun->dev_path = malloc(file_arg->vallen, M_CTLBLK,
1875 					  M_WAITOK | M_ZERO);
1876 
1877 		strlcpy(be_lun->dev_path, (char *)file_arg->kvalue,
1878 			file_arg->vallen);
1879 
1880 		retval = ctl_be_block_open(softc, be_lun, req);
1881 		if (retval != 0) {
1882 			retval = 0;
1883 			goto bailout_error;
1884 		}
1885 
1886 		/*
1887 		 * Tell the user the size of the file/device.
1888 		 */
1889 		params->lun_size_bytes = be_lun->size_bytes;
1890 
1891 		/*
1892 		 * The maximum LBA is the size - 1.
1893 		 */
1894 		be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1;
1895 	} else {
1896 		/*
1897 		 * For processor devices, we don't have any size.
1898 		 */
1899 		be_lun->blocksize = 0;
1900 		be_lun->pblockexp = 0;
1901 		be_lun->pblockoff = 0;
1902 		be_lun->size_blocks = 0;
1903 		be_lun->size_bytes = 0;
1904 		be_lun->ctl_be_lun.maxlba = 0;
1905 		params->lun_size_bytes = 0;
1906 
1907 		/*
1908 		 * Default to just 1 thread for processor devices.
1909 		 */
1910 		num_threads = 1;
1911 	}
1912 
1913 	/*
1914 	 * XXX This searching loop might be refactored to be combined with
1915 	 * the loop above,
1916 	 */
1917 	unmap = 0;
1918 	for (i = 0; i < req->num_be_args; i++) {
1919 		if (strcmp(req->kern_be_args[i].kname, "num_threads") == 0) {
1920 			struct ctl_be_arg *thread_arg;
1921 			char num_thread_str[16];
1922 			int tmp_num_threads;
1923 
1924 
1925 			thread_arg = &req->kern_be_args[i];
1926 
1927 			strlcpy(num_thread_str, (char *)thread_arg->kvalue,
1928 				min(thread_arg->vallen,
1929 				sizeof(num_thread_str)));
1930 
1931 			tmp_num_threads = strtol(num_thread_str, NULL, 0);
1932 
1933 			/*
1934 			 * We don't let the user specify less than one
1935 			 * thread, but hope he's clueful enough not to
1936 			 * specify 1000 threads.
1937 			 */
1938 			if (tmp_num_threads < 1) {
1939 				snprintf(req->error_str, sizeof(req->error_str),
1940 					 "%s: invalid number of threads %s",
1941 				         __func__, num_thread_str);
1942 				goto bailout_error;
1943 			}
1944 
1945 			num_threads = tmp_num_threads;
1946 		} else if (strcmp(req->kern_be_args[i].kname, "unmap") == 0 &&
1947 		    strcmp(req->kern_be_args[i].kvalue, "on") == 0) {
1948 			unmap = 1;
1949 		} else if (strcmp(req->kern_be_args[i].kname, "file") != 0 &&
1950 		    strcmp(req->kern_be_args[i].kname, "dev") != 0) {
1951 			struct ctl_be_lun_option *opt;
1952 
1953 			opt = malloc(sizeof(*opt), M_CTLBLK, M_WAITOK);
1954 			opt->name = malloc(strlen(req->kern_be_args[i].kname) + 1, M_CTLBLK, M_WAITOK);
1955 			strcpy(opt->name, req->kern_be_args[i].kname);
1956 			opt->value = malloc(strlen(req->kern_be_args[i].kvalue) + 1, M_CTLBLK, M_WAITOK);
1957 			strcpy(opt->value, req->kern_be_args[i].kvalue);
1958 			STAILQ_INSERT_TAIL(&be_lun->ctl_be_lun.options, opt, links);
1959 		}
1960 	}
1961 
1962 	be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED;
1963 	be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY;
1964 	if (unmap)
1965 		be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP;
1966 	be_lun->ctl_be_lun.be_lun = be_lun;
1967 	be_lun->ctl_be_lun.blocksize = be_lun->blocksize;
1968 	be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp;
1969 	be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff;
1970 	/* Tell the user the blocksize we ended up using */
1971 	params->blocksize_bytes = be_lun->blocksize;
1972 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
1973 		be_lun->ctl_be_lun.req_lun_id = params->req_lun_id;
1974 		be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ;
1975 	} else
1976 		be_lun->ctl_be_lun.req_lun_id = 0;
1977 
1978 	be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown;
1979 	be_lun->ctl_be_lun.lun_config_status =
1980 		ctl_be_block_lun_config_status;
1981 	be_lun->ctl_be_lun.be = &ctl_be_block_driver;
1982 
1983 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
1984 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d",
1985 			 softc->num_luns);
1986 		strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr,
1987 			ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
1988 			sizeof(tmpstr)));
1989 
1990 		/* Tell the user what we used for a serial number */
1991 		strncpy((char *)params->serial_num, tmpstr,
1992 			ctl_min(sizeof(params->serial_num), sizeof(tmpstr)));
1993 	} else {
1994 		strncpy((char *)be_lun->ctl_be_lun.serial_num,
1995 			params->serial_num,
1996 			ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
1997 			sizeof(params->serial_num)));
1998 	}
1999 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2000 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns);
2001 		strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr,
2002 			ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2003 			sizeof(tmpstr)));
2004 
2005 		/* Tell the user what we used for a device ID */
2006 		strncpy((char *)params->device_id, tmpstr,
2007 			ctl_min(sizeof(params->device_id), sizeof(tmpstr)));
2008 	} else {
2009 		strncpy((char *)be_lun->ctl_be_lun.device_id,
2010 			params->device_id,
2011 			ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2012 				sizeof(params->device_id)));
2013 	}
2014 
2015 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2016 
2017 	be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK,
2018 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2019 
2020 	if (be_lun->io_taskqueue == NULL) {
2021 		snprintf(req->error_str, sizeof(req->error_str),
2022 			 "%s: Unable to create taskqueue", __func__);
2023 		goto bailout_error;
2024 	}
2025 
2026 	/*
2027 	 * Note that we start the same number of threads by default for
2028 	 * both the file case and the block device case.  For the file
2029 	 * case, we need multiple threads to allow concurrency, because the
2030 	 * vnode interface is designed to be a blocking interface.  For the
2031 	 * block device case, ZFS zvols at least will block the caller's
2032 	 * context in many instances, and so we need multiple threads to
2033 	 * overcome that problem.  Other block devices don't need as many
2034 	 * threads, but they shouldn't cause too many problems.
2035 	 *
2036 	 * If the user wants to just have a single thread for a block
2037 	 * device, he can specify that when the LUN is created, or change
2038 	 * the tunable/sysctl to alter the default number of threads.
2039 	 */
2040 	retval = taskqueue_start_threads(&be_lun->io_taskqueue,
2041 					 /*num threads*/num_threads,
2042 					 /*priority*/PWAIT,
2043 					 /*thread name*/
2044 					 "%s taskq", be_lun->lunname);
2045 
2046 	if (retval != 0)
2047 		goto bailout_error;
2048 
2049 	be_lun->num_threads = num_threads;
2050 
2051 	mtx_lock(&softc->lock);
2052 	softc->num_luns++;
2053 	STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links);
2054 
2055 	mtx_unlock(&softc->lock);
2056 
2057 	retval = ctl_add_lun(&be_lun->ctl_be_lun);
2058 	if (retval != 0) {
2059 		mtx_lock(&softc->lock);
2060 		STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2061 			      links);
2062 		softc->num_luns--;
2063 		mtx_unlock(&softc->lock);
2064 		snprintf(req->error_str, sizeof(req->error_str),
2065 			 "%s: ctl_add_lun() returned error %d, see dmesg for "
2066 			"details", __func__, retval);
2067 		retval = 0;
2068 		goto bailout_error;
2069 	}
2070 
2071 	mtx_lock(&softc->lock);
2072 
2073 	/*
2074 	 * Tell the config_status routine that we're waiting so it won't
2075 	 * clean up the LUN in the event of an error.
2076 	 */
2077 	be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2078 
2079 	while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2080 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2081 		if (retval == EINTR)
2082 			break;
2083 	}
2084 	be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2085 
2086 	if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) {
2087 		snprintf(req->error_str, sizeof(req->error_str),
2088 			 "%s: LUN configuration error, see dmesg for details",
2089 			 __func__);
2090 		STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2091 			      links);
2092 		softc->num_luns--;
2093 		mtx_unlock(&softc->lock);
2094 		goto bailout_error;
2095 	} else {
2096 		params->req_lun_id = be_lun->ctl_be_lun.lun_id;
2097 	}
2098 
2099 	mtx_unlock(&softc->lock);
2100 
2101 	be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id,
2102 					       be_lun->blocksize,
2103 					       DEVSTAT_ALL_SUPPORTED,
2104 					       be_lun->ctl_be_lun.lun_type
2105 					       | DEVSTAT_TYPE_IF_OTHER,
2106 					       DEVSTAT_PRIORITY_OTHER);
2107 
2108 
2109 	req->status = CTL_LUN_OK;
2110 
2111 	return (retval);
2112 
2113 bailout_error:
2114 	req->status = CTL_LUN_ERROR;
2115 
2116 	ctl_be_block_close(be_lun);
2117 
2118 	free(be_lun->dev_path, M_CTLBLK);
2119 	free(be_lun, M_CTLBLK);
2120 
2121 	return (retval);
2122 }
2123 
2124 static int
2125 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2126 {
2127 	struct ctl_lun_rm_params *params;
2128 	struct ctl_be_block_lun *be_lun;
2129 	int retval;
2130 
2131 	params = &req->reqdata.rm;
2132 
2133 	mtx_lock(&softc->lock);
2134 
2135 	be_lun = NULL;
2136 
2137 	STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2138 		if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2139 			break;
2140 	}
2141 	mtx_unlock(&softc->lock);
2142 
2143 	if (be_lun == NULL) {
2144 		snprintf(req->error_str, sizeof(req->error_str),
2145 			 "%s: LUN %u is not managed by the block backend",
2146 			 __func__, params->lun_id);
2147 		goto bailout_error;
2148 	}
2149 
2150 	retval = ctl_disable_lun(&be_lun->ctl_be_lun);
2151 
2152 	if (retval != 0) {
2153 		snprintf(req->error_str, sizeof(req->error_str),
2154 			 "%s: error %d returned from ctl_disable_lun() for "
2155 			 "LUN %d", __func__, retval, params->lun_id);
2156 		goto bailout_error;
2157 
2158 	}
2159 
2160 	retval = ctl_invalidate_lun(&be_lun->ctl_be_lun);
2161 	if (retval != 0) {
2162 		snprintf(req->error_str, sizeof(req->error_str),
2163 			 "%s: error %d returned from ctl_invalidate_lun() for "
2164 			 "LUN %d", __func__, retval, params->lun_id);
2165 		goto bailout_error;
2166 	}
2167 
2168 	mtx_lock(&softc->lock);
2169 
2170 	be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2171 
2172 	while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2173                 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2174                 if (retval == EINTR)
2175                         break;
2176         }
2177 
2178 	be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2179 
2180 	if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2181 		snprintf(req->error_str, sizeof(req->error_str),
2182 			 "%s: interrupted waiting for LUN to be freed",
2183 			 __func__);
2184 		mtx_unlock(&softc->lock);
2185 		goto bailout_error;
2186 	}
2187 
2188 	STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links);
2189 
2190 	softc->num_luns--;
2191 	mtx_unlock(&softc->lock);
2192 
2193 	taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task);
2194 
2195 	taskqueue_free(be_lun->io_taskqueue);
2196 
2197 	ctl_be_block_close(be_lun);
2198 
2199 	if (be_lun->disk_stats != NULL)
2200 		devstat_remove_entry(be_lun->disk_stats);
2201 
2202 	uma_zdestroy(be_lun->lun_zone);
2203 
2204 	free(be_lun->dev_path, M_CTLBLK);
2205 
2206 	free(be_lun, M_CTLBLK);
2207 
2208 	req->status = CTL_LUN_OK;
2209 
2210 	return (0);
2211 
2212 bailout_error:
2213 
2214 	req->status = CTL_LUN_ERROR;
2215 
2216 	return (0);
2217 }
2218 
2219 static int
2220 ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
2221 			 struct ctl_lun_req *req)
2222 {
2223 	struct vattr vattr;
2224 	int error;
2225 	struct ctl_lun_modify_params *params;
2226 
2227 	params = &req->reqdata.modify;
2228 
2229 	if (params->lun_size_bytes != 0) {
2230 		be_lun->size_bytes = params->lun_size_bytes;
2231 	} else  {
2232 		error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
2233 		if (error != 0) {
2234 			snprintf(req->error_str, sizeof(req->error_str),
2235 				 "error calling VOP_GETATTR() for file %s",
2236 				 be_lun->dev_path);
2237 			return (error);
2238 		}
2239 
2240 		be_lun->size_bytes = vattr.va_size;
2241 	}
2242 
2243 	return (0);
2244 }
2245 
2246 static int
2247 ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
2248 			struct ctl_lun_req *req)
2249 {
2250 	struct cdev *dev;
2251 	struct cdevsw *devsw;
2252 	int error;
2253 	struct ctl_lun_modify_params *params;
2254 	uint64_t size_bytes;
2255 
2256 	params = &req->reqdata.modify;
2257 
2258 	dev = be_lun->vn->v_rdev;
2259 	devsw = dev->si_devsw;
2260 	if (!devsw->d_ioctl) {
2261 		snprintf(req->error_str, sizeof(req->error_str),
2262 			 "%s: no d_ioctl for device %s!", __func__,
2263 			 be_lun->dev_path);
2264 		return (ENODEV);
2265 	}
2266 
2267 	error = devsw->d_ioctl(dev, DIOCGMEDIASIZE,
2268 			       (caddr_t)&size_bytes, FREAD,
2269 			       curthread);
2270 	if (error) {
2271 		snprintf(req->error_str, sizeof(req->error_str),
2272 			 "%s: error %d returned for DIOCGMEDIASIZE ioctl "
2273 			 "on %s!", __func__, error, be_lun->dev_path);
2274 		return (error);
2275 	}
2276 
2277 	if (params->lun_size_bytes != 0) {
2278 		if (params->lun_size_bytes > size_bytes) {
2279 			snprintf(req->error_str, sizeof(req->error_str),
2280 				 "%s: requested LUN size %ju > backing device "
2281 				 "size %ju", __func__,
2282 				 (uintmax_t)params->lun_size_bytes,
2283 				 (uintmax_t)size_bytes);
2284 			return (EINVAL);
2285 		}
2286 
2287 		be_lun->size_bytes = params->lun_size_bytes;
2288 	} else {
2289 		be_lun->size_bytes = size_bytes;
2290 	}
2291 
2292 	return (0);
2293 }
2294 
2295 static int
2296 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2297 {
2298 	struct ctl_lun_modify_params *params;
2299 	struct ctl_be_block_lun *be_lun;
2300 	int error;
2301 
2302 	params = &req->reqdata.modify;
2303 
2304 	mtx_lock(&softc->lock);
2305 
2306 	be_lun = NULL;
2307 
2308 	STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2309 		if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2310 			break;
2311 	}
2312 	mtx_unlock(&softc->lock);
2313 
2314 	if (be_lun == NULL) {
2315 		snprintf(req->error_str, sizeof(req->error_str),
2316 			 "%s: LUN %u is not managed by the block backend",
2317 			 __func__, params->lun_id);
2318 		goto bailout_error;
2319 	}
2320 
2321 	if (params->lun_size_bytes != 0) {
2322 		if (params->lun_size_bytes < be_lun->blocksize) {
2323 			snprintf(req->error_str, sizeof(req->error_str),
2324 				"%s: LUN size %ju < blocksize %u", __func__,
2325 				params->lun_size_bytes, be_lun->blocksize);
2326 			goto bailout_error;
2327 		}
2328 	}
2329 
2330 	vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2331 
2332 	if (be_lun->vn->v_type == VREG)
2333 		error = ctl_be_block_modify_file(be_lun, req);
2334 	else
2335 		error = ctl_be_block_modify_dev(be_lun, req);
2336 
2337 	VOP_UNLOCK(be_lun->vn, 0);
2338 
2339 	if (error != 0)
2340 		goto bailout_error;
2341 
2342 	be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift;
2343 
2344 	/*
2345 	 * The maximum LBA is the size - 1.
2346 	 *
2347 	 * XXX: Note that this field is being updated without locking,
2348 	 * 	which might cause problems on 32-bit architectures.
2349 	 */
2350 	be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1;
2351 	ctl_lun_capacity_changed(&be_lun->ctl_be_lun);
2352 
2353 	/* Tell the user the exact size we ended up using */
2354 	params->lun_size_bytes = be_lun->size_bytes;
2355 
2356 	req->status = CTL_LUN_OK;
2357 
2358 	return (0);
2359 
2360 bailout_error:
2361 	req->status = CTL_LUN_ERROR;
2362 
2363 	return (0);
2364 }
2365 
2366 static void
2367 ctl_be_block_lun_shutdown(void *be_lun)
2368 {
2369 	struct ctl_be_block_lun *lun;
2370 	struct ctl_be_block_softc *softc;
2371 
2372 	lun = (struct ctl_be_block_lun *)be_lun;
2373 
2374 	softc = lun->softc;
2375 
2376 	mtx_lock(&softc->lock);
2377 	lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2378 	if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2379 		wakeup(lun);
2380 	mtx_unlock(&softc->lock);
2381 
2382 }
2383 
2384 static void
2385 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status)
2386 {
2387 	struct ctl_be_block_lun *lun;
2388 	struct ctl_be_block_softc *softc;
2389 
2390 	lun = (struct ctl_be_block_lun *)be_lun;
2391 	softc = lun->softc;
2392 
2393 	if (status == CTL_LUN_CONFIG_OK) {
2394 		mtx_lock(&softc->lock);
2395 		lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2396 		if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2397 			wakeup(lun);
2398 		mtx_unlock(&softc->lock);
2399 
2400 		/*
2401 		 * We successfully added the LUN, attempt to enable it.
2402 		 */
2403 		if (ctl_enable_lun(&lun->ctl_be_lun) != 0) {
2404 			printf("%s: ctl_enable_lun() failed!\n", __func__);
2405 			if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) {
2406 				printf("%s: ctl_invalidate_lun() failed!\n",
2407 				       __func__);
2408 			}
2409 		}
2410 
2411 		return;
2412 	}
2413 
2414 
2415 	mtx_lock(&softc->lock);
2416 	lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2417 	lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR;
2418 	wakeup(lun);
2419 	mtx_unlock(&softc->lock);
2420 }
2421 
2422 
2423 static int
2424 ctl_be_block_config_write(union ctl_io *io)
2425 {
2426 	struct ctl_be_block_lun *be_lun;
2427 	struct ctl_be_lun *ctl_be_lun;
2428 	int retval;
2429 
2430 	retval = 0;
2431 
2432 	DPRINTF("entered\n");
2433 
2434 	ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
2435 		CTL_PRIV_BACKEND_LUN].ptr;
2436 	be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
2437 
2438 	switch (io->scsiio.cdb[0]) {
2439 	case SYNCHRONIZE_CACHE:
2440 	case SYNCHRONIZE_CACHE_16:
2441 	case WRITE_SAME_10:
2442 	case WRITE_SAME_16:
2443 	case UNMAP:
2444 		/*
2445 		 * The upper level CTL code will filter out any CDBs with
2446 		 * the immediate bit set and return the proper error.
2447 		 *
2448 		 * We don't really need to worry about what LBA range the
2449 		 * user asked to be synced out.  When they issue a sync
2450 		 * cache command, we'll sync out the whole thing.
2451 		 */
2452 		mtx_lock(&be_lun->lock);
2453 		STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2454 				   links);
2455 		mtx_unlock(&be_lun->lock);
2456 		taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2457 		break;
2458 	case START_STOP_UNIT: {
2459 		struct scsi_start_stop_unit *cdb;
2460 
2461 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2462 
2463 		if (cdb->how & SSS_START)
2464 			retval = ctl_start_lun(ctl_be_lun);
2465 		else {
2466 			retval = ctl_stop_lun(ctl_be_lun);
2467 			/*
2468 			 * XXX KDM Copan-specific offline behavior.
2469 			 * Figure out a reasonable way to port this?
2470 			 */
2471 #ifdef NEEDTOPORT
2472 			if ((retval == 0)
2473 			 && (cdb->byte2 & SSS_ONOFFLINE))
2474 				retval = ctl_lun_offline(ctl_be_lun);
2475 #endif
2476 		}
2477 
2478 		/*
2479 		 * In general, the above routines should not fail.  They
2480 		 * just set state for the LUN.  So we've got something
2481 		 * pretty wrong here if we can't start or stop the LUN.
2482 		 */
2483 		if (retval != 0) {
2484 			ctl_set_internal_failure(&io->scsiio,
2485 						 /*sks_valid*/ 1,
2486 						 /*retry_count*/ 0xf051);
2487 			retval = CTL_RETVAL_COMPLETE;
2488 		} else {
2489 			ctl_set_success(&io->scsiio);
2490 		}
2491 		ctl_config_write_done(io);
2492 		break;
2493 	}
2494 	default:
2495 		ctl_set_invalid_opcode(&io->scsiio);
2496 		ctl_config_write_done(io);
2497 		retval = CTL_RETVAL_COMPLETE;
2498 		break;
2499 	}
2500 
2501 	return (retval);
2502 
2503 }
2504 
2505 static int
2506 ctl_be_block_config_read(union ctl_io *io)
2507 {
2508 	return (0);
2509 }
2510 
2511 static int
2512 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb)
2513 {
2514 	struct ctl_be_block_lun *lun;
2515 	int retval;
2516 
2517 	lun = (struct ctl_be_block_lun *)be_lun;
2518 	retval = 0;
2519 
2520 	retval = sbuf_printf(sb, "<num_threads>");
2521 
2522 	if (retval != 0)
2523 		goto bailout;
2524 
2525 	retval = sbuf_printf(sb, "%d", lun->num_threads);
2526 
2527 	if (retval != 0)
2528 		goto bailout;
2529 
2530 	retval = sbuf_printf(sb, "</num_threads>");
2531 
2532 	/*
2533 	 * For processor devices, we don't have a path variable.
2534 	 */
2535 	if ((retval != 0)
2536 	 || (lun->dev_path == NULL))
2537 		goto bailout;
2538 
2539 	retval = sbuf_printf(sb, "<file>");
2540 
2541 	if (retval != 0)
2542 		goto bailout;
2543 
2544 	retval = ctl_sbuf_printf_esc(sb, lun->dev_path);
2545 
2546 	if (retval != 0)
2547 		goto bailout;
2548 
2549 	retval = sbuf_printf(sb, "</file>\n");
2550 
2551 bailout:
2552 
2553 	return (retval);
2554 }
2555 
2556 int
2557 ctl_be_block_init(void)
2558 {
2559 	struct ctl_be_block_softc *softc;
2560 	int retval;
2561 
2562 	softc = &backend_block_softc;
2563 	retval = 0;
2564 
2565 	mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF);
2566 	beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2567 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2568 	STAILQ_INIT(&softc->disk_list);
2569 	STAILQ_INIT(&softc->lun_list);
2570 
2571 	return (retval);
2572 }
2573