xref: /freebsd/sys/cam/ctl/ctl_backend_block.c (revision 3fe8969a749c0e4a62ffdbf4f6883898027a9e19)
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_HALF_IO_SIZE	(512 * 1024)
96 #define	CTLBLK_MAX_IO_SIZE	(CTLBLK_HALF_IO_SIZE * 2)
97 #define	CTLBLK_MAX_SEG		MAXPHYS
98 #define	CTLBLK_HALF_SEGS	MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1)
99 #define	CTLBLK_MAX_SEGS		(CTLBLK_HALF_SEGS * 2)
100 
101 #ifdef CTLBLK_DEBUG
102 #define DPRINTF(fmt, args...) \
103     printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
104 #else
105 #define DPRINTF(fmt, args...) do {} while(0)
106 #endif
107 
108 #define PRIV(io)	\
109     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
110 #define ARGS(io)	\
111     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
112 
113 SDT_PROVIDER_DEFINE(cbb);
114 
115 typedef enum {
116 	CTL_BE_BLOCK_LUN_UNCONFIGURED	= 0x01,
117 	CTL_BE_BLOCK_LUN_CONFIG_ERR	= 0x02,
118 	CTL_BE_BLOCK_LUN_WAITING	= 0x04,
119 	CTL_BE_BLOCK_LUN_MULTI_THREAD	= 0x08
120 } ctl_be_block_lun_flags;
121 
122 typedef enum {
123 	CTL_BE_BLOCK_NONE,
124 	CTL_BE_BLOCK_DEV,
125 	CTL_BE_BLOCK_FILE
126 } ctl_be_block_type;
127 
128 struct ctl_be_block_devdata {
129 	struct cdev *cdev;
130 	struct cdevsw *csw;
131 	int dev_ref;
132 };
133 
134 struct ctl_be_block_filedata {
135 	struct ucred *cred;
136 };
137 
138 union ctl_be_block_bedata {
139 	struct ctl_be_block_devdata dev;
140 	struct ctl_be_block_filedata file;
141 };
142 
143 struct ctl_be_block_io;
144 struct ctl_be_block_lun;
145 
146 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
147 			       struct ctl_be_block_io *beio);
148 
149 /*
150  * Backend LUN structure.  There is a 1:1 mapping between a block device
151  * and a backend block LUN, and between a backend block LUN and a CTL LUN.
152  */
153 struct ctl_be_block_lun {
154 	struct ctl_lun_create_params params;
155 	struct ctl_block_disk *disk;
156 	char lunname[32];
157 	char *dev_path;
158 	ctl_be_block_type dev_type;
159 	struct vnode *vn;
160 	union ctl_be_block_bedata backend;
161 	cbb_dispatch_t dispatch;
162 	cbb_dispatch_t lun_flush;
163 	cbb_dispatch_t unmap;
164 	uma_zone_t lun_zone;
165 	uint64_t size_blocks;
166 	uint64_t size_bytes;
167 	uint32_t blocksize;
168 	int blocksize_shift;
169 	uint16_t pblockexp;
170 	uint16_t pblockoff;
171 	struct ctl_be_block_softc *softc;
172 	struct devstat *disk_stats;
173 	ctl_be_block_lun_flags flags;
174 	STAILQ_ENTRY(ctl_be_block_lun) links;
175 	struct ctl_be_lun ctl_be_lun;
176 	struct taskqueue *io_taskqueue;
177 	struct task io_task;
178 	int num_threads;
179 	STAILQ_HEAD(, ctl_io_hdr) input_queue;
180 	STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
181 	STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
182 	struct mtx_padalign io_lock;
183 	struct mtx_padalign queue_lock;
184 };
185 
186 /*
187  * Overall softc structure for the block backend module.
188  */
189 struct ctl_be_block_softc {
190 	struct mtx			 lock;
191 	int				 num_disks;
192 	STAILQ_HEAD(, ctl_block_disk)	 disk_list;
193 	int				 num_luns;
194 	STAILQ_HEAD(, ctl_be_block_lun)	 lun_list;
195 };
196 
197 static struct ctl_be_block_softc backend_block_softc;
198 
199 /*
200  * Per-I/O information.
201  */
202 struct ctl_be_block_io {
203 	union ctl_io			*io;
204 	struct ctl_sg_entry		sg_segs[CTLBLK_MAX_SEGS];
205 	struct iovec			xiovecs[CTLBLK_MAX_SEGS];
206 	int				bio_cmd;
207 	int				num_segs;
208 	int				num_bios_sent;
209 	int				num_bios_done;
210 	int				send_complete;
211 	int				num_errors;
212 	struct bintime			ds_t0;
213 	devstat_tag_type		ds_tag_type;
214 	devstat_trans_flags		ds_trans_type;
215 	uint64_t			io_len;
216 	uint64_t			io_offset;
217 	struct ctl_be_block_softc	*softc;
218 	struct ctl_be_block_lun		*lun;
219 	void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
220 };
221 
222 static int cbb_num_threads = 14;
223 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0,
224 	    "CAM Target Layer Block Backend");
225 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
226            &cbb_num_threads, 0, "Number of threads per backing file");
227 
228 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
229 static void ctl_free_beio(struct ctl_be_block_io *beio);
230 static void ctl_complete_beio(struct ctl_be_block_io *beio);
231 static int ctl_be_block_move_done(union ctl_io *io);
232 static void ctl_be_block_biodone(struct bio *bio);
233 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
234 				    struct ctl_be_block_io *beio);
235 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
236 				       struct ctl_be_block_io *beio);
237 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
238 				   struct ctl_be_block_io *beio);
239 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
240 				   struct ctl_be_block_io *beio);
241 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
242 				      struct ctl_be_block_io *beio);
243 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
244 				    union ctl_io *io);
245 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
246 				  union ctl_io *io);
247 static void ctl_be_block_worker(void *context, int pending);
248 static int ctl_be_block_submit(union ctl_io *io);
249 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
250 				   int flag, struct thread *td);
251 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
252 				  struct ctl_lun_req *req);
253 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
254 				 struct ctl_lun_req *req);
255 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
256 static int ctl_be_block_open(struct ctl_be_block_softc *softc,
257 			     struct ctl_be_block_lun *be_lun,
258 			     struct ctl_lun_req *req);
259 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
260 			       struct ctl_lun_req *req);
261 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
262 			   struct ctl_lun_req *req);
263 static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
264 				  struct ctl_lun_req *req);
265 static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
266 				 struct ctl_lun_req *req);
267 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
268 			   struct ctl_lun_req *req);
269 static void ctl_be_block_lun_shutdown(void *be_lun);
270 static void ctl_be_block_lun_config_status(void *be_lun,
271 					   ctl_lun_config_status status);
272 static int ctl_be_block_config_write(union ctl_io *io);
273 static int ctl_be_block_config_read(union ctl_io *io);
274 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb);
275 int ctl_be_block_init(void);
276 
277 static struct ctl_backend_driver ctl_be_block_driver =
278 {
279 	.name = "block",
280 	.flags = CTL_BE_FLAG_HAS_CONFIG,
281 	.init = ctl_be_block_init,
282 	.data_submit = ctl_be_block_submit,
283 	.data_move_done = ctl_be_block_move_done,
284 	.config_read = ctl_be_block_config_read,
285 	.config_write = ctl_be_block_config_write,
286 	.ioctl = ctl_be_block_ioctl,
287 	.lun_info = ctl_be_block_lun_info
288 };
289 
290 MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend");
291 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
292 
293 static uma_zone_t beio_zone;
294 
295 static struct ctl_be_block_io *
296 ctl_alloc_beio(struct ctl_be_block_softc *softc)
297 {
298 	struct ctl_be_block_io *beio;
299 
300 	beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO);
301 	beio->softc = softc;
302 	return (beio);
303 }
304 
305 static void
306 ctl_free_beio(struct ctl_be_block_io *beio)
307 {
308 	int duplicate_free;
309 	int i;
310 
311 	duplicate_free = 0;
312 
313 	for (i = 0; i < beio->num_segs; i++) {
314 		if (beio->sg_segs[i].addr == NULL)
315 			duplicate_free++;
316 
317 		uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr);
318 		beio->sg_segs[i].addr = NULL;
319 
320 		/* For compare we had two equal S/G lists. */
321 		if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) {
322 			uma_zfree(beio->lun->lun_zone,
323 			    beio->sg_segs[i + CTLBLK_HALF_SEGS].addr);
324 			beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL;
325 		}
326 	}
327 
328 	if (duplicate_free > 0) {
329 		printf("%s: %d duplicate frees out of %d segments\n", __func__,
330 		       duplicate_free, beio->num_segs);
331 	}
332 
333 	uma_zfree(beio_zone, beio);
334 }
335 
336 static void
337 ctl_complete_beio(struct ctl_be_block_io *beio)
338 {
339 	union ctl_io *io = beio->io;
340 
341 	if (beio->beio_cont != NULL) {
342 		beio->beio_cont(beio);
343 	} else {
344 		ctl_free_beio(beio);
345 		ctl_data_submit_done(io);
346 	}
347 }
348 
349 static int
350 ctl_be_block_move_done(union ctl_io *io)
351 {
352 	struct ctl_be_block_io *beio;
353 	struct ctl_be_block_lun *be_lun;
354 	struct ctl_lba_len_flags *lbalen;
355 #ifdef CTL_TIME_IO
356 	struct bintime cur_bt;
357 #endif
358 	int i;
359 
360 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
361 	be_lun = beio->lun;
362 
363 	DPRINTF("entered\n");
364 
365 #ifdef CTL_TIME_IO
366 	getbintime(&cur_bt);
367 	bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
368 	bintime_add(&io->io_hdr.dma_bt, &cur_bt);
369 	io->io_hdr.num_dmas++;
370 #endif
371 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
372 
373 	/*
374 	 * We set status at this point for read commands, and write
375 	 * commands with errors.
376 	 */
377 	if ((io->io_hdr.port_status == 0) &&
378 	    ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) &&
379 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
380 		lbalen = ARGS(beio->io);
381 		if (lbalen->flags & CTL_LLF_READ) {
382 			ctl_set_success(&io->scsiio);
383 		} else if (lbalen->flags & CTL_LLF_COMPARE) {
384 			/* We have two data blocks ready for comparison. */
385 			for (i = 0; i < beio->num_segs; i++) {
386 				if (memcmp(beio->sg_segs[i].addr,
387 				    beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
388 				    beio->sg_segs[i].len) != 0)
389 					break;
390 			}
391 			if (i < beio->num_segs)
392 				ctl_set_sense(&io->scsiio,
393 				    /*current_error*/ 1,
394 				    /*sense_key*/ SSD_KEY_MISCOMPARE,
395 				    /*asc*/ 0x1D,
396 				    /*ascq*/ 0x00,
397 				    SSD_ELEM_NONE);
398 			else
399 				ctl_set_success(&io->scsiio);
400 		}
401 	}
402 	else if ((io->io_hdr.port_status != 0)
403 	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
404 	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
405 		/*
406 		 * For hardware error sense keys, the sense key
407 		 * specific value is defined to be a retry count,
408 		 * but we use it to pass back an internal FETD
409 		 * error code.  XXX KDM  Hopefully the FETD is only
410 		 * using 16 bits for an error code, since that's
411 		 * all the space we have in the sks field.
412 		 */
413 		ctl_set_internal_failure(&io->scsiio,
414 					 /*sks_valid*/ 1,
415 					 /*retry_count*/
416 					 io->io_hdr.port_status);
417 	}
418 
419 	/*
420 	 * If this is a read, or a write with errors, it is done.
421 	 */
422 	if ((beio->bio_cmd == BIO_READ)
423 	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
424 	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
425 		ctl_complete_beio(beio);
426 		return (0);
427 	}
428 
429 	/*
430 	 * At this point, we have a write and the DMA completed
431 	 * successfully.  We now have to queue it to the task queue to
432 	 * execute the backend I/O.  That is because we do blocking
433 	 * memory allocations, and in the file backing case, blocking I/O.
434 	 * This move done routine is generally called in the SIM's
435 	 * interrupt context, and therefore we cannot block.
436 	 */
437 	mtx_lock(&be_lun->queue_lock);
438 	/*
439 	 * XXX KDM make sure that links is okay to use at this point.
440 	 * Otherwise, we either need to add another field to ctl_io_hdr,
441 	 * or deal with resource allocation here.
442 	 */
443 	STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
444 	mtx_unlock(&be_lun->queue_lock);
445 
446 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
447 
448 	return (0);
449 }
450 
451 static void
452 ctl_be_block_biodone(struct bio *bio)
453 {
454 	struct ctl_be_block_io *beio;
455 	struct ctl_be_block_lun *be_lun;
456 	union ctl_io *io;
457 	int error;
458 
459 	beio = bio->bio_caller1;
460 	be_lun = beio->lun;
461 	io = beio->io;
462 
463 	DPRINTF("entered\n");
464 
465 	error = bio->bio_error;
466 	mtx_lock(&be_lun->io_lock);
467 	if (error != 0)
468 		beio->num_errors++;
469 
470 	beio->num_bios_done++;
471 
472 	/*
473 	 * XXX KDM will this cause WITNESS to complain?  Holding a lock
474 	 * during the free might cause it to complain.
475 	 */
476 	g_destroy_bio(bio);
477 
478 	/*
479 	 * If the send complete bit isn't set, or we aren't the last I/O to
480 	 * complete, then we're done.
481 	 */
482 	if ((beio->send_complete == 0)
483 	 || (beio->num_bios_done < beio->num_bios_sent)) {
484 		mtx_unlock(&be_lun->io_lock);
485 		return;
486 	}
487 
488 	/*
489 	 * At this point, we've verified that we are the last I/O to
490 	 * complete, so it's safe to drop the lock.
491 	 */
492 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
493 	    beio->ds_tag_type, beio->ds_trans_type,
494 	    /*now*/ NULL, /*then*/&beio->ds_t0);
495 	mtx_unlock(&be_lun->io_lock);
496 
497 	/*
498 	 * If there are any errors from the backing device, we fail the
499 	 * entire I/O with a medium error.
500 	 */
501 	if (beio->num_errors > 0) {
502 		if (error == EOPNOTSUPP) {
503 			ctl_set_invalid_opcode(&io->scsiio);
504 		} else if (error == ENOSPC) {
505 			ctl_set_space_alloc_fail(&io->scsiio);
506 		} else if (beio->bio_cmd == BIO_FLUSH) {
507 			/* XXX KDM is there is a better error here? */
508 			ctl_set_internal_failure(&io->scsiio,
509 						 /*sks_valid*/ 1,
510 						 /*retry_count*/ 0xbad2);
511 		} else
512 			ctl_set_medium_error(&io->scsiio);
513 		ctl_complete_beio(beio);
514 		return;
515 	}
516 
517 	/*
518 	 * If this is a write, a flush, a delete or verify, we're all done.
519 	 * If this is a read, we can now send the data to the user.
520 	 */
521 	if ((beio->bio_cmd == BIO_WRITE)
522 	 || (beio->bio_cmd == BIO_FLUSH)
523 	 || (beio->bio_cmd == BIO_DELETE)
524 	 || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
525 		ctl_set_success(&io->scsiio);
526 		ctl_complete_beio(beio);
527 	} else {
528 #ifdef CTL_TIME_IO
529         	getbintime(&io->io_hdr.dma_start_bt);
530 #endif
531 		ctl_datamove(io);
532 	}
533 }
534 
535 static void
536 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
537 			struct ctl_be_block_io *beio)
538 {
539 	union ctl_io *io = beio->io;
540 	struct mount *mountpoint;
541 	int error, lock_flags;
542 
543 	DPRINTF("entered\n");
544 
545 	binuptime(&beio->ds_t0);
546 	mtx_lock(&be_lun->io_lock);
547 	devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
548 	mtx_unlock(&be_lun->io_lock);
549 
550 	(void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
551 
552 	if (MNT_SHARED_WRITES(mountpoint)
553 	 || ((mountpoint == NULL)
554 	  && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
555 		lock_flags = LK_SHARED;
556 	else
557 		lock_flags = LK_EXCLUSIVE;
558 
559 	vn_lock(be_lun->vn, lock_flags | LK_RETRY);
560 
561 	error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread);
562 	VOP_UNLOCK(be_lun->vn, 0);
563 
564 	vn_finished_write(mountpoint);
565 
566 	mtx_lock(&be_lun->io_lock);
567 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
568 	    beio->ds_tag_type, beio->ds_trans_type,
569 	    /*now*/ NULL, /*then*/&beio->ds_t0);
570 	mtx_unlock(&be_lun->io_lock);
571 
572 	if (error == 0)
573 		ctl_set_success(&io->scsiio);
574 	else {
575 		/* XXX KDM is there is a better error here? */
576 		ctl_set_internal_failure(&io->scsiio,
577 					 /*sks_valid*/ 1,
578 					 /*retry_count*/ 0xbad1);
579 	}
580 
581 	ctl_complete_beio(beio);
582 }
583 
584 SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t");
585 SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t");
586 SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t");
587 SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t");
588 
589 static void
590 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
591 			   struct ctl_be_block_io *beio)
592 {
593 	struct ctl_be_block_filedata *file_data;
594 	union ctl_io *io;
595 	struct uio xuio;
596 	struct iovec *xiovec;
597 	int flags;
598 	int error, i;
599 
600 	DPRINTF("entered\n");
601 
602 	file_data = &be_lun->backend.file;
603 	io = beio->io;
604 	flags = 0;
605 	if (ARGS(io)->flags & CTL_LLF_DPO)
606 		flags |= IO_DIRECT;
607 	if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
608 		flags |= IO_SYNC;
609 
610 	bzero(&xuio, sizeof(xuio));
611 	if (beio->bio_cmd == BIO_READ) {
612 		SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
613 		xuio.uio_rw = UIO_READ;
614 	} else {
615 		SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
616 		xuio.uio_rw = UIO_WRITE;
617 	}
618 	xuio.uio_offset = beio->io_offset;
619 	xuio.uio_resid = beio->io_len;
620 	xuio.uio_segflg = UIO_SYSSPACE;
621 	xuio.uio_iov = beio->xiovecs;
622 	xuio.uio_iovcnt = beio->num_segs;
623 	xuio.uio_td = curthread;
624 
625 	for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
626 		xiovec->iov_base = beio->sg_segs[i].addr;
627 		xiovec->iov_len = beio->sg_segs[i].len;
628 	}
629 
630 	binuptime(&beio->ds_t0);
631 	mtx_lock(&be_lun->io_lock);
632 	devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
633 	mtx_unlock(&be_lun->io_lock);
634 
635 	if (beio->bio_cmd == BIO_READ) {
636 		vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
637 
638 		/*
639 		 * UFS pays attention to IO_DIRECT for reads.  If the
640 		 * DIRECTIO option is configured into the kernel, it calls
641 		 * ffs_rawread().  But that only works for single-segment
642 		 * uios with user space addresses.  In our case, with a
643 		 * kernel uio, it still reads into the buffer cache, but it
644 		 * will just try to release the buffer from the cache later
645 		 * on in ffs_read().
646 		 *
647 		 * ZFS does not pay attention to IO_DIRECT for reads.
648 		 *
649 		 * UFS does not pay attention to IO_SYNC for reads.
650 		 *
651 		 * ZFS pays attention to IO_SYNC (which translates into the
652 		 * Solaris define FRSYNC for zfs_read()) for reads.  It
653 		 * attempts to sync the file before reading.
654 		 *
655 		 * So, to attempt to provide some barrier semantics in the
656 		 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC.
657 		 */
658 		error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
659 
660 		VOP_UNLOCK(be_lun->vn, 0);
661 		SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
662 	} else {
663 		struct mount *mountpoint;
664 		int lock_flags;
665 
666 		(void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
667 
668 		if (MNT_SHARED_WRITES(mountpoint)
669 		 || ((mountpoint == NULL)
670 		  && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
671 			lock_flags = LK_SHARED;
672 		else
673 			lock_flags = LK_EXCLUSIVE;
674 
675 		vn_lock(be_lun->vn, lock_flags | LK_RETRY);
676 
677 		/*
678 		 * UFS pays attention to IO_DIRECT for writes.  The write
679 		 * is done asynchronously.  (Normally the write would just
680 		 * get put into cache.
681 		 *
682 		 * UFS pays attention to IO_SYNC for writes.  It will
683 		 * attempt to write the buffer out synchronously if that
684 		 * flag is set.
685 		 *
686 		 * ZFS does not pay attention to IO_DIRECT for writes.
687 		 *
688 		 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
689 		 * for writes.  It will flush the transaction from the
690 		 * cache before returning.
691 		 *
692 		 * So if we've got the BIO_ORDERED flag set, we want
693 		 * IO_SYNC in either the UFS or ZFS case.
694 		 */
695 		error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
696 		VOP_UNLOCK(be_lun->vn, 0);
697 
698 		vn_finished_write(mountpoint);
699 		SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
700         }
701 
702 	mtx_lock(&be_lun->io_lock);
703 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
704 	    beio->ds_tag_type, beio->ds_trans_type,
705 	    /*now*/ NULL, /*then*/&beio->ds_t0);
706 	mtx_unlock(&be_lun->io_lock);
707 
708 	/*
709 	 * If we got an error, set the sense data to "MEDIUM ERROR" and
710 	 * return the I/O to the user.
711 	 */
712 	if (error != 0) {
713 		char path_str[32];
714 
715 		ctl_scsi_path_string(io, path_str, sizeof(path_str));
716 		printf("%s%s command returned errno %d\n", path_str,
717 		       (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error);
718 		if (error == ENOSPC) {
719 			ctl_set_space_alloc_fail(&io->scsiio);
720 		} else
721 			ctl_set_medium_error(&io->scsiio);
722 		ctl_complete_beio(beio);
723 		return;
724 	}
725 
726 	/*
727 	 * If this is a write or a verify, we're all done.
728 	 * If this is a read, we can now send the data to the user.
729 	 */
730 	if ((beio->bio_cmd == BIO_WRITE) ||
731 	    (ARGS(io)->flags & CTL_LLF_VERIFY)) {
732 		ctl_set_success(&io->scsiio);
733 		ctl_complete_beio(beio);
734 	} else {
735 #ifdef CTL_TIME_IO
736         	getbintime(&io->io_hdr.dma_start_bt);
737 #endif
738 		ctl_datamove(io);
739 	}
740 }
741 
742 static void
743 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
744 			   struct ctl_be_block_io *beio)
745 {
746 	struct ctl_be_block_devdata *dev_data;
747 	union ctl_io *io;
748 	struct uio xuio;
749 	struct iovec *xiovec;
750 	int flags;
751 	int error, i;
752 
753 	DPRINTF("entered\n");
754 
755 	dev_data = &be_lun->backend.dev;
756 	io = beio->io;
757 	flags = 0;
758 	if (ARGS(io)->flags & CTL_LLF_DPO)
759 		flags |= IO_DIRECT;
760 	if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
761 		flags |= IO_SYNC;
762 
763 	bzero(&xuio, sizeof(xuio));
764 	if (beio->bio_cmd == BIO_READ) {
765 		SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
766 		xuio.uio_rw = UIO_READ;
767 	} else {
768 		SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
769 		xuio.uio_rw = UIO_WRITE;
770 	}
771 	xuio.uio_offset = beio->io_offset;
772 	xuio.uio_resid = beio->io_len;
773 	xuio.uio_segflg = UIO_SYSSPACE;
774 	xuio.uio_iov = beio->xiovecs;
775 	xuio.uio_iovcnt = beio->num_segs;
776 	xuio.uio_td = curthread;
777 
778 	for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
779 		xiovec->iov_base = beio->sg_segs[i].addr;
780 		xiovec->iov_len = beio->sg_segs[i].len;
781 	}
782 
783 	binuptime(&beio->ds_t0);
784 	mtx_lock(&be_lun->io_lock);
785 	devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
786 	mtx_unlock(&be_lun->io_lock);
787 
788 	if (beio->bio_cmd == BIO_READ) {
789 		error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags);
790 		SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
791 	} else {
792 		error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags);
793 		SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
794 	}
795 
796 	mtx_lock(&be_lun->io_lock);
797 	devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
798 	    beio->ds_tag_type, beio->ds_trans_type,
799 	    /*now*/ NULL, /*then*/&beio->ds_t0);
800 	mtx_unlock(&be_lun->io_lock);
801 
802 	/*
803 	 * If we got an error, set the sense data to "MEDIUM ERROR" and
804 	 * return the I/O to the user.
805 	 */
806 	if (error != 0) {
807 		if (error == ENOSPC) {
808 			ctl_set_space_alloc_fail(&io->scsiio);
809 		} else
810 			ctl_set_medium_error(&io->scsiio);
811 		ctl_complete_beio(beio);
812 		return;
813 	}
814 
815 	/*
816 	 * If this is a write or a verify, we're all done.
817 	 * If this is a read, we can now send the data to the user.
818 	 */
819 	if ((beio->bio_cmd == BIO_WRITE) ||
820 	    (ARGS(io)->flags & CTL_LLF_VERIFY)) {
821 		ctl_set_success(&io->scsiio);
822 		ctl_complete_beio(beio);
823 	} else {
824 #ifdef CTL_TIME_IO
825         	getbintime(&io->io_hdr.dma_start_bt);
826 #endif
827 		ctl_datamove(io);
828 	}
829 }
830 
831 static void
832 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
833 		       struct ctl_be_block_io *beio)
834 {
835 	struct bio *bio;
836 	union ctl_io *io;
837 	struct ctl_be_block_devdata *dev_data;
838 
839 	dev_data = &be_lun->backend.dev;
840 	io = beio->io;
841 
842 	DPRINTF("entered\n");
843 
844 	/* This can't fail, it's a blocking allocation. */
845 	bio = g_alloc_bio();
846 
847 	bio->bio_cmd	    = BIO_FLUSH;
848 	bio->bio_flags	   |= BIO_ORDERED;
849 	bio->bio_dev	    = dev_data->cdev;
850 	bio->bio_offset	    = 0;
851 	bio->bio_data	    = 0;
852 	bio->bio_done	    = ctl_be_block_biodone;
853 	bio->bio_caller1    = beio;
854 	bio->bio_pblkno	    = 0;
855 
856 	/*
857 	 * We don't need to acquire the LUN lock here, because we are only
858 	 * sending one bio, and so there is no other context to synchronize
859 	 * with.
860 	 */
861 	beio->num_bios_sent = 1;
862 	beio->send_complete = 1;
863 
864 	binuptime(&beio->ds_t0);
865 	mtx_lock(&be_lun->io_lock);
866 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
867 	mtx_unlock(&be_lun->io_lock);
868 
869 	(*dev_data->csw->d_strategy)(bio);
870 }
871 
872 static void
873 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
874 		       struct ctl_be_block_io *beio,
875 		       uint64_t off, uint64_t len, int last)
876 {
877 	struct bio *bio;
878 	struct ctl_be_block_devdata *dev_data;
879 	uint64_t maxlen;
880 
881 	dev_data = &be_lun->backend.dev;
882 	maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize);
883 	while (len > 0) {
884 		bio = g_alloc_bio();
885 		bio->bio_cmd	    = BIO_DELETE;
886 		bio->bio_dev	    = dev_data->cdev;
887 		bio->bio_offset	    = off;
888 		bio->bio_length	    = MIN(len, maxlen);
889 		bio->bio_data	    = 0;
890 		bio->bio_done	    = ctl_be_block_biodone;
891 		bio->bio_caller1    = beio;
892 		bio->bio_pblkno     = off / be_lun->blocksize;
893 
894 		off += bio->bio_length;
895 		len -= bio->bio_length;
896 
897 		mtx_lock(&be_lun->io_lock);
898 		beio->num_bios_sent++;
899 		if (last && len == 0)
900 			beio->send_complete = 1;
901 		mtx_unlock(&be_lun->io_lock);
902 
903 		(*dev_data->csw->d_strategy)(bio);
904 	}
905 }
906 
907 static void
908 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
909 		       struct ctl_be_block_io *beio)
910 {
911 	union ctl_io *io;
912 	struct ctl_be_block_devdata *dev_data;
913 	struct ctl_ptr_len_flags *ptrlen;
914 	struct scsi_unmap_desc *buf, *end;
915 	uint64_t len;
916 
917 	dev_data = &be_lun->backend.dev;
918 	io = beio->io;
919 
920 	DPRINTF("entered\n");
921 
922 	binuptime(&beio->ds_t0);
923 	mtx_lock(&be_lun->io_lock);
924 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
925 	mtx_unlock(&be_lun->io_lock);
926 
927 	if (beio->io_offset == -1) {
928 		beio->io_len = 0;
929 		ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
930 		buf = (struct scsi_unmap_desc *)ptrlen->ptr;
931 		end = buf + ptrlen->len / sizeof(*buf);
932 		for (; buf < end; buf++) {
933 			len = (uint64_t)scsi_4btoul(buf->length) *
934 			    be_lun->blocksize;
935 			beio->io_len += len;
936 			ctl_be_block_unmap_dev_range(be_lun, beio,
937 			    scsi_8btou64(buf->lba) * be_lun->blocksize, len,
938 			    (end - buf < 2) ? TRUE : FALSE);
939 		}
940 	} else
941 		ctl_be_block_unmap_dev_range(be_lun, beio,
942 		    beio->io_offset, beio->io_len, TRUE);
943 }
944 
945 static void
946 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
947 			  struct ctl_be_block_io *beio)
948 {
949 	TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
950 	int i;
951 	struct bio *bio;
952 	struct ctl_be_block_devdata *dev_data;
953 	off_t cur_offset;
954 	int max_iosize;
955 
956 	DPRINTF("entered\n");
957 
958 	dev_data = &be_lun->backend.dev;
959 
960 	/*
961 	 * We have to limit our I/O size to the maximum supported by the
962 	 * backend device.  Hopefully it is MAXPHYS.  If the driver doesn't
963 	 * set it properly, use DFLTPHYS.
964 	 */
965 	max_iosize = dev_data->cdev->si_iosize_max;
966 	if (max_iosize < PAGE_SIZE)
967 		max_iosize = DFLTPHYS;
968 
969 	cur_offset = beio->io_offset;
970 	for (i = 0; i < beio->num_segs; i++) {
971 		size_t cur_size;
972 		uint8_t *cur_ptr;
973 
974 		cur_size = beio->sg_segs[i].len;
975 		cur_ptr = beio->sg_segs[i].addr;
976 
977 		while (cur_size > 0) {
978 			/* This can't fail, it's a blocking allocation. */
979 			bio = g_alloc_bio();
980 
981 			KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
982 
983 			bio->bio_cmd = beio->bio_cmd;
984 			bio->bio_dev = dev_data->cdev;
985 			bio->bio_caller1 = beio;
986 			bio->bio_length = min(cur_size, max_iosize);
987 			bio->bio_offset = cur_offset;
988 			bio->bio_data = cur_ptr;
989 			bio->bio_done = ctl_be_block_biodone;
990 			bio->bio_pblkno = cur_offset / be_lun->blocksize;
991 
992 			cur_offset += bio->bio_length;
993 			cur_ptr += bio->bio_length;
994 			cur_size -= bio->bio_length;
995 
996 			TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
997 			beio->num_bios_sent++;
998 		}
999 	}
1000 	binuptime(&beio->ds_t0);
1001 	mtx_lock(&be_lun->io_lock);
1002 	devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1003 	beio->send_complete = 1;
1004 	mtx_unlock(&be_lun->io_lock);
1005 
1006 	/*
1007 	 * Fire off all allocated requests!
1008 	 */
1009 	while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1010 		TAILQ_REMOVE(&queue, bio, bio_queue);
1011 		(*dev_data->csw->d_strategy)(bio);
1012 	}
1013 }
1014 
1015 static void
1016 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1017 {
1018 	union ctl_io *io;
1019 
1020 	io = beio->io;
1021 	ctl_free_beio(beio);
1022 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1023 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1024 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1025 		ctl_config_write_done(io);
1026 		return;
1027 	}
1028 
1029 	ctl_be_block_config_write(io);
1030 }
1031 
1032 static void
1033 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1034 			    union ctl_io *io)
1035 {
1036 	struct ctl_be_block_io *beio;
1037 	struct ctl_be_block_softc *softc;
1038 	struct ctl_lba_len_flags *lbalen;
1039 	uint64_t len_left, lba;
1040 	int i, seglen;
1041 	uint8_t *buf, *end;
1042 
1043 	DPRINTF("entered\n");
1044 
1045 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1046 	softc = be_lun->softc;
1047 	lbalen = ARGS(beio->io);
1048 
1049 	if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1050 	    (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1051 		ctl_free_beio(beio);
1052 		ctl_set_invalid_field(&io->scsiio,
1053 				      /*sks_valid*/ 1,
1054 				      /*command*/ 1,
1055 				      /*field*/ 1,
1056 				      /*bit_valid*/ 0,
1057 				      /*bit*/ 0);
1058 		ctl_config_write_done(io);
1059 		return;
1060 	}
1061 
1062 	switch (io->scsiio.tag_type) {
1063 	case CTL_TAG_ORDERED:
1064 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1065 		break;
1066 	case CTL_TAG_HEAD_OF_QUEUE:
1067 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1068 		break;
1069 	case CTL_TAG_UNTAGGED:
1070 	case CTL_TAG_SIMPLE:
1071 	case CTL_TAG_ACA:
1072 	default:
1073 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1074 		break;
1075 	}
1076 
1077 	if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1078 		beio->io_offset = lbalen->lba * be_lun->blocksize;
1079 		beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize;
1080 		beio->bio_cmd = BIO_DELETE;
1081 		beio->ds_trans_type = DEVSTAT_FREE;
1082 
1083 		be_lun->unmap(be_lun, beio);
1084 		return;
1085 	}
1086 
1087 	beio->bio_cmd = BIO_WRITE;
1088 	beio->ds_trans_type = DEVSTAT_WRITE;
1089 
1090 	DPRINTF("WRITE SAME at LBA %jx len %u\n",
1091 	       (uintmax_t)lbalen->lba, lbalen->len);
1092 
1093 	len_left = (uint64_t)lbalen->len * be_lun->blocksize;
1094 	for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1095 
1096 		/*
1097 		 * Setup the S/G entry for this chunk.
1098 		 */
1099 		seglen = MIN(CTLBLK_MAX_SEG, len_left);
1100 		seglen -= seglen % be_lun->blocksize;
1101 		beio->sg_segs[i].len = seglen;
1102 		beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1103 
1104 		DPRINTF("segment %d addr %p len %zd\n", i,
1105 			beio->sg_segs[i].addr, beio->sg_segs[i].len);
1106 
1107 		beio->num_segs++;
1108 		len_left -= seglen;
1109 
1110 		buf = beio->sg_segs[i].addr;
1111 		end = buf + seglen;
1112 		for (; buf < end; buf += be_lun->blocksize) {
1113 			memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize);
1114 			if (lbalen->flags & SWS_LBDATA)
1115 				scsi_ulto4b(lbalen->lba + lba, buf);
1116 			lba++;
1117 		}
1118 	}
1119 
1120 	beio->io_offset = lbalen->lba * be_lun->blocksize;
1121 	beio->io_len = lba * be_lun->blocksize;
1122 
1123 	/* We can not do all in one run. Correct and schedule rerun. */
1124 	if (len_left > 0) {
1125 		lbalen->lba += lba;
1126 		lbalen->len -= lba;
1127 		beio->beio_cont = ctl_be_block_cw_done_ws;
1128 	}
1129 
1130 	be_lun->dispatch(be_lun, beio);
1131 }
1132 
1133 static void
1134 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1135 			    union ctl_io *io)
1136 {
1137 	struct ctl_be_block_io *beio;
1138 	struct ctl_be_block_softc *softc;
1139 	struct ctl_ptr_len_flags *ptrlen;
1140 
1141 	DPRINTF("entered\n");
1142 
1143 	beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1144 	softc = be_lun->softc;
1145 	ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1146 
1147 	if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1148 		ctl_free_beio(beio);
1149 		ctl_set_invalid_field(&io->scsiio,
1150 				      /*sks_valid*/ 0,
1151 				      /*command*/ 1,
1152 				      /*field*/ 0,
1153 				      /*bit_valid*/ 0,
1154 				      /*bit*/ 0);
1155 		ctl_config_write_done(io);
1156 		return;
1157 	}
1158 
1159 	switch (io->scsiio.tag_type) {
1160 	case CTL_TAG_ORDERED:
1161 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1162 		break;
1163 	case CTL_TAG_HEAD_OF_QUEUE:
1164 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1165 		break;
1166 	case CTL_TAG_UNTAGGED:
1167 	case CTL_TAG_SIMPLE:
1168 	case CTL_TAG_ACA:
1169 	default:
1170 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1171 		break;
1172 	}
1173 
1174 	beio->io_len = 0;
1175 	beio->io_offset = -1;
1176 
1177 	beio->bio_cmd = BIO_DELETE;
1178 	beio->ds_trans_type = DEVSTAT_FREE;
1179 
1180 	DPRINTF("UNMAP\n");
1181 
1182 	be_lun->unmap(be_lun, beio);
1183 }
1184 
1185 static void
1186 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1187 {
1188 	union ctl_io *io;
1189 
1190 	io = beio->io;
1191 	ctl_free_beio(beio);
1192 	ctl_config_write_done(io);
1193 }
1194 
1195 static void
1196 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1197 			 union ctl_io *io)
1198 {
1199 	struct ctl_be_block_io *beio;
1200 	struct ctl_be_block_softc *softc;
1201 
1202 	DPRINTF("entered\n");
1203 
1204 	softc = be_lun->softc;
1205 	beio = ctl_alloc_beio(softc);
1206 	beio->io = io;
1207 	beio->lun = be_lun;
1208 	beio->beio_cont = ctl_be_block_cw_done;
1209 	PRIV(io)->ptr = (void *)beio;
1210 
1211 	switch (io->scsiio.cdb[0]) {
1212 	case SYNCHRONIZE_CACHE:
1213 	case SYNCHRONIZE_CACHE_16:
1214 		beio->bio_cmd = BIO_FLUSH;
1215 		beio->ds_trans_type = DEVSTAT_NO_DATA;
1216 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1217 		beio->io_len = 0;
1218 		be_lun->lun_flush(be_lun, beio);
1219 		break;
1220 	case WRITE_SAME_10:
1221 	case WRITE_SAME_16:
1222 		ctl_be_block_cw_dispatch_ws(be_lun, io);
1223 		break;
1224 	case UNMAP:
1225 		ctl_be_block_cw_dispatch_unmap(be_lun, io);
1226 		break;
1227 	default:
1228 		panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1229 		break;
1230 	}
1231 }
1232 
1233 SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t");
1234 SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t");
1235 SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t");
1236 SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t");
1237 
1238 static void
1239 ctl_be_block_next(struct ctl_be_block_io *beio)
1240 {
1241 	struct ctl_be_block_lun *be_lun;
1242 	union ctl_io *io;
1243 
1244 	io = beio->io;
1245 	be_lun = beio->lun;
1246 	ctl_free_beio(beio);
1247 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1248 	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1249 	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1250 		ctl_data_submit_done(io);
1251 		return;
1252 	}
1253 
1254 	io->io_hdr.status &= ~CTL_STATUS_MASK;
1255 	io->io_hdr.status |= CTL_STATUS_NONE;
1256 
1257 	mtx_lock(&be_lun->queue_lock);
1258 	/*
1259 	 * XXX KDM make sure that links is okay to use at this point.
1260 	 * Otherwise, we either need to add another field to ctl_io_hdr,
1261 	 * or deal with resource allocation here.
1262 	 */
1263 	STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1264 	mtx_unlock(&be_lun->queue_lock);
1265 
1266 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1267 }
1268 
1269 static void
1270 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1271 			   union ctl_io *io)
1272 {
1273 	struct ctl_be_block_io *beio;
1274 	struct ctl_be_block_softc *softc;
1275 	struct ctl_lba_len_flags *lbalen;
1276 	struct ctl_ptr_len_flags *bptrlen;
1277 	uint64_t len_left, lbas;
1278 	int i;
1279 
1280 	softc = be_lun->softc;
1281 
1282 	DPRINTF("entered\n");
1283 
1284 	lbalen = ARGS(io);
1285 	if (lbalen->flags & CTL_LLF_WRITE) {
1286 		SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0);
1287 	} else {
1288 		SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0);
1289 	}
1290 
1291 	beio = ctl_alloc_beio(softc);
1292 	beio->io = io;
1293 	beio->lun = be_lun;
1294 	bptrlen = PRIV(io);
1295 	bptrlen->ptr = (void *)beio;
1296 
1297 	switch (io->scsiio.tag_type) {
1298 	case CTL_TAG_ORDERED:
1299 		beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1300 		break;
1301 	case CTL_TAG_HEAD_OF_QUEUE:
1302 		beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1303 		break;
1304 	case CTL_TAG_UNTAGGED:
1305 	case CTL_TAG_SIMPLE:
1306 	case CTL_TAG_ACA:
1307 	default:
1308 		beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1309 		break;
1310 	}
1311 
1312 	if (lbalen->flags & CTL_LLF_WRITE) {
1313 		beio->bio_cmd = BIO_WRITE;
1314 		beio->ds_trans_type = DEVSTAT_WRITE;
1315 	} else {
1316 		beio->bio_cmd = BIO_READ;
1317 		beio->ds_trans_type = DEVSTAT_READ;
1318 	}
1319 
1320 	DPRINTF("%s at LBA %jx len %u @%ju\n",
1321 	       (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1322 	       (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1323 	if (lbalen->flags & CTL_LLF_COMPARE)
1324 		lbas = CTLBLK_HALF_IO_SIZE;
1325 	else
1326 		lbas = CTLBLK_MAX_IO_SIZE;
1327 	lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize);
1328 	beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize;
1329 	beio->io_len = lbas * be_lun->blocksize;
1330 	bptrlen->len += lbas;
1331 
1332 	for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1333 		KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1334 		    i, CTLBLK_MAX_SEGS));
1335 
1336 		/*
1337 		 * Setup the S/G entry for this chunk.
1338 		 */
1339 		beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1340 		beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1341 
1342 		DPRINTF("segment %d addr %p len %zd\n", i,
1343 			beio->sg_segs[i].addr, beio->sg_segs[i].len);
1344 
1345 		/* Set up second segment for compare operation. */
1346 		if (lbalen->flags & CTL_LLF_COMPARE) {
1347 			beio->sg_segs[i + CTLBLK_HALF_SEGS].len =
1348 			    beio->sg_segs[i].len;
1349 			beio->sg_segs[i + CTLBLK_HALF_SEGS].addr =
1350 			    uma_zalloc(be_lun->lun_zone, M_WAITOK);
1351 		}
1352 
1353 		beio->num_segs++;
1354 		len_left -= beio->sg_segs[i].len;
1355 	}
1356 	if (bptrlen->len < lbalen->len)
1357 		beio->beio_cont = ctl_be_block_next;
1358 	io->scsiio.be_move_done = ctl_be_block_move_done;
1359 	/* For compare we have separate S/G lists for read and datamove. */
1360 	if (lbalen->flags & CTL_LLF_COMPARE)
1361 		io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1362 	else
1363 		io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1364 	io->scsiio.kern_data_len = beio->io_len;
1365 	io->scsiio.kern_data_resid = 0;
1366 	io->scsiio.kern_sg_entries = beio->num_segs;
1367 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST;
1368 
1369 	/*
1370 	 * For the read case, we need to read the data into our buffers and
1371 	 * then we can send it back to the user.  For the write case, we
1372 	 * need to get the data from the user first.
1373 	 */
1374 	if (beio->bio_cmd == BIO_READ) {
1375 		SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0);
1376 		be_lun->dispatch(be_lun, beio);
1377 	} else {
1378 		SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0);
1379 #ifdef CTL_TIME_IO
1380         	getbintime(&io->io_hdr.dma_start_bt);
1381 #endif
1382 		ctl_datamove(io);
1383 	}
1384 }
1385 
1386 static void
1387 ctl_be_block_worker(void *context, int pending)
1388 {
1389 	struct ctl_be_block_lun *be_lun;
1390 	struct ctl_be_block_softc *softc;
1391 	union ctl_io *io;
1392 
1393 	be_lun = (struct ctl_be_block_lun *)context;
1394 	softc = be_lun->softc;
1395 
1396 	DPRINTF("entered\n");
1397 
1398 	mtx_lock(&be_lun->queue_lock);
1399 	for (;;) {
1400 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1401 		if (io != NULL) {
1402 			struct ctl_be_block_io *beio;
1403 
1404 			DPRINTF("datamove queue\n");
1405 
1406 			STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1407 				      ctl_io_hdr, links);
1408 
1409 			mtx_unlock(&be_lun->queue_lock);
1410 
1411 			beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1412 
1413 			be_lun->dispatch(be_lun, beio);
1414 
1415 			mtx_lock(&be_lun->queue_lock);
1416 			continue;
1417 		}
1418 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1419 		if (io != NULL) {
1420 
1421 			DPRINTF("config write queue\n");
1422 
1423 			STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1424 				      ctl_io_hdr, links);
1425 
1426 			mtx_unlock(&be_lun->queue_lock);
1427 
1428 			ctl_be_block_cw_dispatch(be_lun, io);
1429 
1430 			mtx_lock(&be_lun->queue_lock);
1431 			continue;
1432 		}
1433 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1434 		if (io != NULL) {
1435 			DPRINTF("input queue\n");
1436 
1437 			STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1438 				      ctl_io_hdr, links);
1439 			mtx_unlock(&be_lun->queue_lock);
1440 
1441 			/*
1442 			 * We must drop the lock, since this routine and
1443 			 * its children may sleep.
1444 			 */
1445 			ctl_be_block_dispatch(be_lun, io);
1446 
1447 			mtx_lock(&be_lun->queue_lock);
1448 			continue;
1449 		}
1450 
1451 		/*
1452 		 * If we get here, there is no work left in the queues, so
1453 		 * just break out and let the task queue go to sleep.
1454 		 */
1455 		break;
1456 	}
1457 	mtx_unlock(&be_lun->queue_lock);
1458 }
1459 
1460 /*
1461  * Entry point from CTL to the backend for I/O.  We queue everything to a
1462  * work thread, so this just puts the I/O on a queue and wakes up the
1463  * thread.
1464  */
1465 static int
1466 ctl_be_block_submit(union ctl_io *io)
1467 {
1468 	struct ctl_be_block_lun *be_lun;
1469 	struct ctl_be_lun *ctl_be_lun;
1470 
1471 	DPRINTF("entered\n");
1472 
1473 	ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
1474 		CTL_PRIV_BACKEND_LUN].ptr;
1475 	be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
1476 
1477 	/*
1478 	 * Make sure we only get SCSI I/O.
1479 	 */
1480 	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1481 		"%#x) encountered", io->io_hdr.io_type));
1482 
1483 	PRIV(io)->len = 0;
1484 
1485 	mtx_lock(&be_lun->queue_lock);
1486 	/*
1487 	 * XXX KDM make sure that links is okay to use at this point.
1488 	 * Otherwise, we either need to add another field to ctl_io_hdr,
1489 	 * or deal with resource allocation here.
1490 	 */
1491 	STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1492 	mtx_unlock(&be_lun->queue_lock);
1493 	taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1494 
1495 	return (CTL_RETVAL_COMPLETE);
1496 }
1497 
1498 static int
1499 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1500 			int flag, struct thread *td)
1501 {
1502 	struct ctl_be_block_softc *softc;
1503 	int error;
1504 
1505 	softc = &backend_block_softc;
1506 
1507 	error = 0;
1508 
1509 	switch (cmd) {
1510 	case CTL_LUN_REQ: {
1511 		struct ctl_lun_req *lun_req;
1512 
1513 		lun_req = (struct ctl_lun_req *)addr;
1514 
1515 		switch (lun_req->reqtype) {
1516 		case CTL_LUNREQ_CREATE:
1517 			error = ctl_be_block_create(softc, lun_req);
1518 			break;
1519 		case CTL_LUNREQ_RM:
1520 			error = ctl_be_block_rm(softc, lun_req);
1521 			break;
1522 		case CTL_LUNREQ_MODIFY:
1523 			error = ctl_be_block_modify(softc, lun_req);
1524 			break;
1525 		default:
1526 			lun_req->status = CTL_LUN_ERROR;
1527 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1528 				 "invalid LUN request type %d",
1529 				 lun_req->reqtype);
1530 			break;
1531 		}
1532 		break;
1533 	}
1534 	default:
1535 		error = ENOTTY;
1536 		break;
1537 	}
1538 
1539 	return (error);
1540 }
1541 
1542 static int
1543 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1544 {
1545 	struct ctl_be_block_filedata *file_data;
1546 	struct ctl_lun_create_params *params;
1547 	struct vattr		      vattr;
1548 	off_t			      pss;
1549 	int			      error;
1550 
1551 	error = 0;
1552 	file_data = &be_lun->backend.file;
1553 	params = &be_lun->params;
1554 
1555 	be_lun->dev_type = CTL_BE_BLOCK_FILE;
1556 	be_lun->dispatch = ctl_be_block_dispatch_file;
1557 	be_lun->lun_flush = ctl_be_block_flush_file;
1558 
1559 	error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1560 	if (error != 0) {
1561 		snprintf(req->error_str, sizeof(req->error_str),
1562 			 "error calling VOP_GETATTR() for file %s",
1563 			 be_lun->dev_path);
1564 		return (error);
1565 	}
1566 
1567 	/*
1568 	 * Verify that we have the ability to upgrade to exclusive
1569 	 * access on this file so we can trap errors at open instead
1570 	 * of reporting them during first access.
1571 	 */
1572 	if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) {
1573 		vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY);
1574 		if (be_lun->vn->v_iflag & VI_DOOMED) {
1575 			error = EBADF;
1576 			snprintf(req->error_str, sizeof(req->error_str),
1577 				 "error locking file %s", be_lun->dev_path);
1578 			return (error);
1579 		}
1580 	}
1581 
1582 
1583 	file_data->cred = crhold(curthread->td_ucred);
1584 	if (params->lun_size_bytes != 0)
1585 		be_lun->size_bytes = params->lun_size_bytes;
1586 	else
1587 		be_lun->size_bytes = vattr.va_size;
1588 	/*
1589 	 * We set the multi thread flag for file operations because all
1590 	 * filesystems (in theory) are capable of allowing multiple readers
1591 	 * of a file at once.  So we want to get the maximum possible
1592 	 * concurrency.
1593 	 */
1594 	be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD;
1595 
1596 	/*
1597 	 * For files we can use any logical block size.  Prefer 512 bytes
1598 	 * for compatibility reasons.  If file's vattr.va_blocksize
1599 	 * (preferred I/O block size) is bigger and multiple to chosen
1600 	 * logical block size -- report it as physical block size.
1601 	 */
1602 	if (params->blocksize_bytes != 0)
1603 		be_lun->blocksize = params->blocksize_bytes;
1604 	else
1605 		be_lun->blocksize = 512;
1606 	pss = vattr.va_blocksize / be_lun->blocksize;
1607 	if ((pss > 0) && (pss * be_lun->blocksize == vattr.va_blocksize) &&
1608 	    ((pss & (pss - 1)) == 0)) {
1609 		be_lun->pblockexp = fls(pss) - 1;
1610 		be_lun->pblockoff = 0;
1611 	}
1612 
1613 	/*
1614 	 * Sanity check.  The media size has to be at least one
1615 	 * sector long.
1616 	 */
1617 	if (be_lun->size_bytes < be_lun->blocksize) {
1618 		error = EINVAL;
1619 		snprintf(req->error_str, sizeof(req->error_str),
1620 			 "file %s size %ju < block size %u", be_lun->dev_path,
1621 			 (uintmax_t)be_lun->size_bytes, be_lun->blocksize);
1622 	}
1623 	return (error);
1624 }
1625 
1626 static int
1627 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1628 {
1629 	struct ctl_lun_create_params *params;
1630 	struct vattr		      vattr;
1631 	struct cdev		     *dev;
1632 	struct cdevsw		     *devsw;
1633 	int			      error;
1634 	off_t			      ps, pss, po, pos;
1635 
1636 	params = &be_lun->params;
1637 
1638 	be_lun->dev_type = CTL_BE_BLOCK_DEV;
1639 	be_lun->backend.dev.cdev = be_lun->vn->v_rdev;
1640 	be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev,
1641 					     &be_lun->backend.dev.dev_ref);
1642 	if (be_lun->backend.dev.csw == NULL)
1643 		panic("Unable to retrieve device switch");
1644 	if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0)
1645 		be_lun->dispatch = ctl_be_block_dispatch_zvol;
1646 	else
1647 		be_lun->dispatch = ctl_be_block_dispatch_dev;
1648 	be_lun->lun_flush = ctl_be_block_flush_dev;
1649 	be_lun->unmap = ctl_be_block_unmap_dev;
1650 
1651 	error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED);
1652 	if (error) {
1653 		snprintf(req->error_str, sizeof(req->error_str),
1654 			 "error getting vnode attributes for device %s",
1655 			 be_lun->dev_path);
1656 		return (error);
1657 	}
1658 
1659 	dev = be_lun->vn->v_rdev;
1660 	devsw = dev->si_devsw;
1661 	if (!devsw->d_ioctl) {
1662 		snprintf(req->error_str, sizeof(req->error_str),
1663 			 "no d_ioctl for device %s!",
1664 			 be_lun->dev_path);
1665 		return (ENODEV);
1666 	}
1667 
1668 	error = devsw->d_ioctl(dev, DIOCGSECTORSIZE,
1669 			       (caddr_t)&be_lun->blocksize, FREAD,
1670 			       curthread);
1671 	if (error) {
1672 		snprintf(req->error_str, sizeof(req->error_str),
1673 			 "error %d returned for DIOCGSECTORSIZE ioctl "
1674 			 "on %s!", error, be_lun->dev_path);
1675 		return (error);
1676 	}
1677 
1678 	/*
1679 	 * If the user has asked for a blocksize that is greater than the
1680 	 * backing device's blocksize, we can do it only if the blocksize
1681 	 * the user is asking for is an even multiple of the underlying
1682 	 * device's blocksize.
1683 	 */
1684 	if ((params->blocksize_bytes != 0)
1685 	 && (params->blocksize_bytes > be_lun->blocksize)) {
1686 		uint32_t bs_multiple, tmp_blocksize;
1687 
1688 		bs_multiple = params->blocksize_bytes / be_lun->blocksize;
1689 
1690 		tmp_blocksize = bs_multiple * be_lun->blocksize;
1691 
1692 		if (tmp_blocksize == params->blocksize_bytes) {
1693 			be_lun->blocksize = params->blocksize_bytes;
1694 		} else {
1695 			snprintf(req->error_str, sizeof(req->error_str),
1696 				 "requested blocksize %u is not an even "
1697 				 "multiple of backing device blocksize %u",
1698 				 params->blocksize_bytes,
1699 				 be_lun->blocksize);
1700 			return (EINVAL);
1701 
1702 		}
1703 	} else if ((params->blocksize_bytes != 0)
1704 		&& (params->blocksize_bytes != be_lun->blocksize)) {
1705 		snprintf(req->error_str, sizeof(req->error_str),
1706 			 "requested blocksize %u < backing device "
1707 			 "blocksize %u", params->blocksize_bytes,
1708 			 be_lun->blocksize);
1709 		return (EINVAL);
1710 	}
1711 
1712 	error = devsw->d_ioctl(dev, DIOCGMEDIASIZE,
1713 			       (caddr_t)&be_lun->size_bytes, FREAD,
1714 			       curthread);
1715 	if (error) {
1716 		snprintf(req->error_str, sizeof(req->error_str),
1717 			 "error %d returned for DIOCGMEDIASIZE "
1718 			 " ioctl on %s!", error,
1719 			 be_lun->dev_path);
1720 		return (error);
1721 	}
1722 
1723 	if (params->lun_size_bytes != 0) {
1724 		if (params->lun_size_bytes > be_lun->size_bytes) {
1725 			snprintf(req->error_str, sizeof(req->error_str),
1726 				 "requested LUN size %ju > backing device "
1727 				 "size %ju",
1728 				 (uintmax_t)params->lun_size_bytes,
1729 				 (uintmax_t)be_lun->size_bytes);
1730 			return (EINVAL);
1731 		}
1732 
1733 		be_lun->size_bytes = params->lun_size_bytes;
1734 	}
1735 
1736 	error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE,
1737 			       (caddr_t)&ps, FREAD, curthread);
1738 	if (error)
1739 		ps = po = 0;
1740 	else {
1741 		error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET,
1742 				       (caddr_t)&po, FREAD, curthread);
1743 		if (error)
1744 			po = 0;
1745 	}
1746 	pss = ps / be_lun->blocksize;
1747 	pos = po / be_lun->blocksize;
1748 	if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) &&
1749 	    ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) {
1750 		be_lun->pblockexp = fls(pss) - 1;
1751 		be_lun->pblockoff = (pss - pos) % pss;
1752 	}
1753 
1754 	return (0);
1755 }
1756 
1757 static int
1758 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
1759 {
1760 	DROP_GIANT();
1761 	if (be_lun->vn) {
1762 		int flags = FREAD | FWRITE;
1763 
1764 		switch (be_lun->dev_type) {
1765 		case CTL_BE_BLOCK_DEV:
1766 			if (be_lun->backend.dev.csw) {
1767 				dev_relthread(be_lun->backend.dev.cdev,
1768 					      be_lun->backend.dev.dev_ref);
1769 				be_lun->backend.dev.csw  = NULL;
1770 				be_lun->backend.dev.cdev = NULL;
1771 			}
1772 			break;
1773 		case CTL_BE_BLOCK_FILE:
1774 			break;
1775 		case CTL_BE_BLOCK_NONE:
1776 			break;
1777 		default:
1778 			panic("Unexpected backend type.");
1779 			break;
1780 		}
1781 
1782 		(void)vn_close(be_lun->vn, flags, NOCRED, curthread);
1783 		be_lun->vn = NULL;
1784 
1785 		switch (be_lun->dev_type) {
1786 		case CTL_BE_BLOCK_DEV:
1787 			break;
1788 		case CTL_BE_BLOCK_FILE:
1789 			if (be_lun->backend.file.cred != NULL) {
1790 				crfree(be_lun->backend.file.cred);
1791 				be_lun->backend.file.cred = NULL;
1792 			}
1793 			break;
1794 		case CTL_BE_BLOCK_NONE:
1795 			break;
1796 		default:
1797 			panic("Unexpected backend type.");
1798 			break;
1799 		}
1800 		be_lun->dev_type = CTL_BE_BLOCK_NONE;
1801 	}
1802 	PICKUP_GIANT();
1803 
1804 	return (0);
1805 }
1806 
1807 static int
1808 ctl_be_block_open(struct ctl_be_block_softc *softc,
1809 		       struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1810 {
1811 	struct nameidata nd;
1812 	int		 flags;
1813 	int		 error;
1814 
1815 	/*
1816 	 * XXX KDM allow a read-only option?
1817 	 */
1818 	flags = FREAD | FWRITE;
1819 	error = 0;
1820 
1821 	if (rootvnode == NULL) {
1822 		snprintf(req->error_str, sizeof(req->error_str),
1823 			 "Root filesystem is not mounted");
1824 		return (1);
1825 	}
1826 
1827 	if (!curthread->td_proc->p_fd->fd_cdir) {
1828 		curthread->td_proc->p_fd->fd_cdir = rootvnode;
1829 		VREF(rootvnode);
1830 	}
1831 	if (!curthread->td_proc->p_fd->fd_rdir) {
1832 		curthread->td_proc->p_fd->fd_rdir = rootvnode;
1833 		VREF(rootvnode);
1834 	}
1835 	if (!curthread->td_proc->p_fd->fd_jdir) {
1836 		curthread->td_proc->p_fd->fd_jdir = rootvnode;
1837 		VREF(rootvnode);
1838 	}
1839 
1840  again:
1841 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread);
1842 	error = vn_open(&nd, &flags, 0, NULL);
1843 	if (error) {
1844 		/*
1845 		 * This is the only reasonable guess we can make as far as
1846 		 * path if the user doesn't give us a fully qualified path.
1847 		 * If they want to specify a file, they need to specify the
1848 		 * full path.
1849 		 */
1850 		if (be_lun->dev_path[0] != '/') {
1851 			char *dev_path = "/dev/";
1852 			char *dev_name;
1853 
1854 			/* Try adding device path at beginning of name */
1855 			dev_name = malloc(strlen(be_lun->dev_path)
1856 					+ strlen(dev_path) + 1,
1857 					  M_CTLBLK, M_WAITOK);
1858 			if (dev_name) {
1859 				sprintf(dev_name, "%s%s", dev_path,
1860 					be_lun->dev_path);
1861 				free(be_lun->dev_path, M_CTLBLK);
1862 				be_lun->dev_path = dev_name;
1863 				goto again;
1864 			}
1865 		}
1866 		snprintf(req->error_str, sizeof(req->error_str),
1867 		    "error opening %s: %d", be_lun->dev_path, error);
1868 		return (error);
1869 	}
1870 
1871 	NDFREE(&nd, NDF_ONLY_PNBUF);
1872 
1873 	be_lun->vn = nd.ni_vp;
1874 
1875 	/* We only support disks and files. */
1876 	if (vn_isdisk(be_lun->vn, &error)) {
1877 		error = ctl_be_block_open_dev(be_lun, req);
1878 	} else if (be_lun->vn->v_type == VREG) {
1879 		error = ctl_be_block_open_file(be_lun, req);
1880 	} else {
1881 		error = EINVAL;
1882 		snprintf(req->error_str, sizeof(req->error_str),
1883 			 "%s is not a disk or plain file", be_lun->dev_path);
1884 	}
1885 	VOP_UNLOCK(be_lun->vn, 0);
1886 
1887 	if (error != 0) {
1888 		ctl_be_block_close(be_lun);
1889 		return (error);
1890 	}
1891 
1892 	be_lun->blocksize_shift = fls(be_lun->blocksize) - 1;
1893 	be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift;
1894 
1895 	return (0);
1896 }
1897 
1898 static int
1899 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
1900 {
1901 	struct ctl_be_block_lun *be_lun;
1902 	struct ctl_lun_create_params *params;
1903 	char num_thread_str[16];
1904 	char tmpstr[32];
1905 	char *value;
1906 	int retval, num_threads, unmap;
1907 	int tmp_num_threads;
1908 
1909 	params = &req->reqdata.create;
1910 	retval = 0;
1911 	req->status = CTL_LUN_OK;
1912 
1913 	num_threads = cbb_num_threads;
1914 
1915 	be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
1916 
1917 	be_lun->params = req->reqdata.create;
1918 	be_lun->softc = softc;
1919 	STAILQ_INIT(&be_lun->input_queue);
1920 	STAILQ_INIT(&be_lun->config_write_queue);
1921 	STAILQ_INIT(&be_lun->datamove_queue);
1922 	sprintf(be_lun->lunname, "cblk%d", softc->num_luns);
1923 	mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF);
1924 	mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF);
1925 	ctl_init_opts(&be_lun->ctl_be_lun.options,
1926 	    req->num_be_args, req->kern_be_args);
1927 
1928 	be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG,
1929 	    NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
1930 
1931 	if (be_lun->lun_zone == NULL) {
1932 		snprintf(req->error_str, sizeof(req->error_str),
1933 			 "error allocating UMA zone");
1934 		goto bailout_error;
1935 	}
1936 
1937 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
1938 		be_lun->ctl_be_lun.lun_type = params->device_type;
1939 	else
1940 		be_lun->ctl_be_lun.lun_type = T_DIRECT;
1941 
1942 	if (be_lun->ctl_be_lun.lun_type == T_DIRECT) {
1943 		value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file");
1944 		if (value == NULL) {
1945 			snprintf(req->error_str, sizeof(req->error_str),
1946 				 "no file argument specified");
1947 			goto bailout_error;
1948 		}
1949 		be_lun->dev_path = strdup(value, M_CTLBLK);
1950 		be_lun->blocksize = 512;
1951 		be_lun->blocksize_shift = fls(be_lun->blocksize) - 1;
1952 
1953 		retval = ctl_be_block_open(softc, be_lun, req);
1954 		if (retval != 0) {
1955 			retval = 0;
1956 			req->status = CTL_LUN_WARNING;
1957 		}
1958 	} else {
1959 		/*
1960 		 * For processor devices, we don't have any size.
1961 		 */
1962 		be_lun->blocksize = 0;
1963 		be_lun->pblockexp = 0;
1964 		be_lun->pblockoff = 0;
1965 		be_lun->size_blocks = 0;
1966 		be_lun->size_bytes = 0;
1967 		be_lun->ctl_be_lun.maxlba = 0;
1968 
1969 		/*
1970 		 * Default to just 1 thread for processor devices.
1971 		 */
1972 		num_threads = 1;
1973 	}
1974 
1975 	/*
1976 	 * XXX This searching loop might be refactored to be combined with
1977 	 * the loop above,
1978 	 */
1979 	value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads");
1980 	if (value != NULL) {
1981 		tmp_num_threads = strtol(value, NULL, 0);
1982 
1983 		/*
1984 		 * We don't let the user specify less than one
1985 		 * thread, but hope he's clueful enough not to
1986 		 * specify 1000 threads.
1987 		 */
1988 		if (tmp_num_threads < 1) {
1989 			snprintf(req->error_str, sizeof(req->error_str),
1990 				 "invalid number of threads %s",
1991 				 num_thread_str);
1992 			goto bailout_error;
1993 		}
1994 		num_threads = tmp_num_threads;
1995 	}
1996 	unmap = 0;
1997 	value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap");
1998 	if (value != NULL && strcmp(value, "on") == 0)
1999 		unmap = 1;
2000 
2001 	be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED;
2002 	be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY;
2003 	if (be_lun->vn == NULL)
2004 		be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE;
2005 	if (unmap)
2006 		be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP;
2007 	be_lun->ctl_be_lun.be_lun = be_lun;
2008 	be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ?
2009 	    0 : (be_lun->size_blocks - 1);
2010 	be_lun->ctl_be_lun.blocksize = be_lun->blocksize;
2011 	be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp;
2012 	be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff;
2013 	if (be_lun->dispatch == ctl_be_block_dispatch_zvol &&
2014 	    be_lun->blocksize != 0)
2015 		be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE /
2016 		    be_lun->blocksize;
2017 	/* Tell the user the blocksize we ended up using */
2018 	params->lun_size_bytes = be_lun->size_bytes;
2019 	params->blocksize_bytes = be_lun->blocksize;
2020 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2021 		be_lun->ctl_be_lun.req_lun_id = params->req_lun_id;
2022 		be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ;
2023 	} else
2024 		be_lun->ctl_be_lun.req_lun_id = 0;
2025 
2026 	be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown;
2027 	be_lun->ctl_be_lun.lun_config_status =
2028 		ctl_be_block_lun_config_status;
2029 	be_lun->ctl_be_lun.be = &ctl_be_block_driver;
2030 
2031 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2032 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d",
2033 			 softc->num_luns);
2034 		strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr,
2035 			ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
2036 			sizeof(tmpstr)));
2037 
2038 		/* Tell the user what we used for a serial number */
2039 		strncpy((char *)params->serial_num, tmpstr,
2040 			ctl_min(sizeof(params->serial_num), sizeof(tmpstr)));
2041 	} else {
2042 		strncpy((char *)be_lun->ctl_be_lun.serial_num,
2043 			params->serial_num,
2044 			ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
2045 			sizeof(params->serial_num)));
2046 	}
2047 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2048 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns);
2049 		strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr,
2050 			ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2051 			sizeof(tmpstr)));
2052 
2053 		/* Tell the user what we used for a device ID */
2054 		strncpy((char *)params->device_id, tmpstr,
2055 			ctl_min(sizeof(params->device_id), sizeof(tmpstr)));
2056 	} else {
2057 		strncpy((char *)be_lun->ctl_be_lun.device_id,
2058 			params->device_id,
2059 			ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2060 				sizeof(params->device_id)));
2061 	}
2062 
2063 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2064 
2065 	be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK,
2066 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2067 
2068 	if (be_lun->io_taskqueue == NULL) {
2069 		snprintf(req->error_str, sizeof(req->error_str),
2070 			 "unable to create taskqueue");
2071 		goto bailout_error;
2072 	}
2073 
2074 	/*
2075 	 * Note that we start the same number of threads by default for
2076 	 * both the file case and the block device case.  For the file
2077 	 * case, we need multiple threads to allow concurrency, because the
2078 	 * vnode interface is designed to be a blocking interface.  For the
2079 	 * block device case, ZFS zvols at least will block the caller's
2080 	 * context in many instances, and so we need multiple threads to
2081 	 * overcome that problem.  Other block devices don't need as many
2082 	 * threads, but they shouldn't cause too many problems.
2083 	 *
2084 	 * If the user wants to just have a single thread for a block
2085 	 * device, he can specify that when the LUN is created, or change
2086 	 * the tunable/sysctl to alter the default number of threads.
2087 	 */
2088 	retval = taskqueue_start_threads(&be_lun->io_taskqueue,
2089 					 /*num threads*/num_threads,
2090 					 /*priority*/PWAIT,
2091 					 /*thread name*/
2092 					 "%s taskq", be_lun->lunname);
2093 
2094 	if (retval != 0)
2095 		goto bailout_error;
2096 
2097 	be_lun->num_threads = num_threads;
2098 
2099 	mtx_lock(&softc->lock);
2100 	softc->num_luns++;
2101 	STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links);
2102 
2103 	mtx_unlock(&softc->lock);
2104 
2105 	retval = ctl_add_lun(&be_lun->ctl_be_lun);
2106 	if (retval != 0) {
2107 		mtx_lock(&softc->lock);
2108 		STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2109 			      links);
2110 		softc->num_luns--;
2111 		mtx_unlock(&softc->lock);
2112 		snprintf(req->error_str, sizeof(req->error_str),
2113 			 "ctl_add_lun() returned error %d, see dmesg for "
2114 			 "details", retval);
2115 		retval = 0;
2116 		goto bailout_error;
2117 	}
2118 
2119 	mtx_lock(&softc->lock);
2120 
2121 	/*
2122 	 * Tell the config_status routine that we're waiting so it won't
2123 	 * clean up the LUN in the event of an error.
2124 	 */
2125 	be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2126 
2127 	while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2128 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2129 		if (retval == EINTR)
2130 			break;
2131 	}
2132 	be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2133 
2134 	if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) {
2135 		snprintf(req->error_str, sizeof(req->error_str),
2136 			 "LUN configuration error, see dmesg for details");
2137 		STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2138 			      links);
2139 		softc->num_luns--;
2140 		mtx_unlock(&softc->lock);
2141 		goto bailout_error;
2142 	} else {
2143 		params->req_lun_id = be_lun->ctl_be_lun.lun_id;
2144 	}
2145 
2146 	mtx_unlock(&softc->lock);
2147 
2148 	be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id,
2149 					       be_lun->blocksize,
2150 					       DEVSTAT_ALL_SUPPORTED,
2151 					       be_lun->ctl_be_lun.lun_type
2152 					       | DEVSTAT_TYPE_IF_OTHER,
2153 					       DEVSTAT_PRIORITY_OTHER);
2154 
2155 	return (retval);
2156 
2157 bailout_error:
2158 	req->status = CTL_LUN_ERROR;
2159 
2160 	if (be_lun->io_taskqueue != NULL)
2161 		taskqueue_free(be_lun->io_taskqueue);
2162 	ctl_be_block_close(be_lun);
2163 	if (be_lun->dev_path != NULL)
2164 		free(be_lun->dev_path, M_CTLBLK);
2165 	if (be_lun->lun_zone != NULL)
2166 		uma_zdestroy(be_lun->lun_zone);
2167 	ctl_free_opts(&be_lun->ctl_be_lun.options);
2168 	mtx_destroy(&be_lun->queue_lock);
2169 	mtx_destroy(&be_lun->io_lock);
2170 	free(be_lun, M_CTLBLK);
2171 
2172 	return (retval);
2173 }
2174 
2175 static int
2176 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2177 {
2178 	struct ctl_lun_rm_params *params;
2179 	struct ctl_be_block_lun *be_lun;
2180 	int retval;
2181 
2182 	params = &req->reqdata.rm;
2183 
2184 	mtx_lock(&softc->lock);
2185 
2186 	be_lun = NULL;
2187 
2188 	STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2189 		if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2190 			break;
2191 	}
2192 	mtx_unlock(&softc->lock);
2193 
2194 	if (be_lun == NULL) {
2195 		snprintf(req->error_str, sizeof(req->error_str),
2196 			 "LUN %u is not managed by the block backend",
2197 			 params->lun_id);
2198 		goto bailout_error;
2199 	}
2200 
2201 	retval = ctl_disable_lun(&be_lun->ctl_be_lun);
2202 
2203 	if (retval != 0) {
2204 		snprintf(req->error_str, sizeof(req->error_str),
2205 			 "error %d returned from ctl_disable_lun() for "
2206 			 "LUN %d", retval, params->lun_id);
2207 		goto bailout_error;
2208 
2209 	}
2210 
2211 	retval = ctl_invalidate_lun(&be_lun->ctl_be_lun);
2212 	if (retval != 0) {
2213 		snprintf(req->error_str, sizeof(req->error_str),
2214 			 "error %d returned from ctl_invalidate_lun() for "
2215 			 "LUN %d", retval, params->lun_id);
2216 		goto bailout_error;
2217 	}
2218 
2219 	mtx_lock(&softc->lock);
2220 
2221 	be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2222 
2223 	while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2224                 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2225                 if (retval == EINTR)
2226                         break;
2227         }
2228 
2229 	be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2230 
2231 	if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2232 		snprintf(req->error_str, sizeof(req->error_str),
2233 			 "interrupted waiting for LUN to be freed");
2234 		mtx_unlock(&softc->lock);
2235 		goto bailout_error;
2236 	}
2237 
2238 	STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links);
2239 
2240 	softc->num_luns--;
2241 	mtx_unlock(&softc->lock);
2242 
2243 	taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task);
2244 
2245 	taskqueue_free(be_lun->io_taskqueue);
2246 
2247 	ctl_be_block_close(be_lun);
2248 
2249 	if (be_lun->disk_stats != NULL)
2250 		devstat_remove_entry(be_lun->disk_stats);
2251 
2252 	uma_zdestroy(be_lun->lun_zone);
2253 
2254 	ctl_free_opts(&be_lun->ctl_be_lun.options);
2255 	free(be_lun->dev_path, M_CTLBLK);
2256 	mtx_destroy(&be_lun->queue_lock);
2257 	mtx_destroy(&be_lun->io_lock);
2258 	free(be_lun, M_CTLBLK);
2259 
2260 	req->status = CTL_LUN_OK;
2261 
2262 	return (0);
2263 
2264 bailout_error:
2265 
2266 	req->status = CTL_LUN_ERROR;
2267 
2268 	return (0);
2269 }
2270 
2271 static int
2272 ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
2273 			 struct ctl_lun_req *req)
2274 {
2275 	struct vattr vattr;
2276 	int error;
2277 	struct ctl_lun_create_params *params = &be_lun->params;
2278 
2279 	if (params->lun_size_bytes != 0) {
2280 		be_lun->size_bytes = params->lun_size_bytes;
2281 	} else  {
2282 		vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2283 		error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
2284 		VOP_UNLOCK(be_lun->vn, 0);
2285 		if (error != 0) {
2286 			snprintf(req->error_str, sizeof(req->error_str),
2287 				 "error calling VOP_GETATTR() for file %s",
2288 				 be_lun->dev_path);
2289 			return (error);
2290 		}
2291 
2292 		be_lun->size_bytes = vattr.va_size;
2293 	}
2294 
2295 	return (0);
2296 }
2297 
2298 static int
2299 ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
2300 			struct ctl_lun_req *req)
2301 {
2302 	struct ctl_be_block_devdata *dev_data;
2303 	int error;
2304 	struct ctl_lun_create_params *params = &be_lun->params;
2305 	uint64_t size_bytes;
2306 
2307 	dev_data = &be_lun->backend.dev;
2308 	if (!dev_data->csw->d_ioctl) {
2309 		snprintf(req->error_str, sizeof(req->error_str),
2310 			 "no d_ioctl for device %s!", be_lun->dev_path);
2311 		return (ENODEV);
2312 	}
2313 
2314 	error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE,
2315 			       (caddr_t)&size_bytes, FREAD,
2316 			       curthread);
2317 	if (error) {
2318 		snprintf(req->error_str, sizeof(req->error_str),
2319 			 "error %d returned for DIOCGMEDIASIZE ioctl "
2320 			 "on %s!", error, be_lun->dev_path);
2321 		return (error);
2322 	}
2323 
2324 	if (params->lun_size_bytes != 0) {
2325 		if (params->lun_size_bytes > size_bytes) {
2326 			snprintf(req->error_str, sizeof(req->error_str),
2327 				 "requested LUN size %ju > backing device "
2328 				 "size %ju",
2329 				 (uintmax_t)params->lun_size_bytes,
2330 				 (uintmax_t)size_bytes);
2331 			return (EINVAL);
2332 		}
2333 
2334 		be_lun->size_bytes = params->lun_size_bytes;
2335 	} else {
2336 		be_lun->size_bytes = size_bytes;
2337 	}
2338 
2339 	return (0);
2340 }
2341 
2342 static int
2343 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2344 {
2345 	struct ctl_lun_modify_params *params;
2346 	struct ctl_be_block_lun *be_lun;
2347 	uint64_t oldsize;
2348 	int error;
2349 
2350 	params = &req->reqdata.modify;
2351 
2352 	mtx_lock(&softc->lock);
2353 	be_lun = NULL;
2354 	STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2355 		if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2356 			break;
2357 	}
2358 	mtx_unlock(&softc->lock);
2359 
2360 	if (be_lun == NULL) {
2361 		snprintf(req->error_str, sizeof(req->error_str),
2362 			 "LUN %u is not managed by the block backend",
2363 			 params->lun_id);
2364 		goto bailout_error;
2365 	}
2366 
2367 	be_lun->params.lun_size_bytes = params->lun_size_bytes;
2368 
2369 	oldsize = be_lun->size_blocks;
2370 	if (be_lun->vn == NULL)
2371 		error = ctl_be_block_open(softc, be_lun, req);
2372 	else if (be_lun->vn->v_type == VREG)
2373 		error = ctl_be_block_modify_file(be_lun, req);
2374 	else
2375 		error = ctl_be_block_modify_dev(be_lun, req);
2376 
2377 	if (error == 0 && be_lun->size_blocks != oldsize) {
2378 		be_lun->size_blocks = be_lun->size_bytes >>
2379 		    be_lun->blocksize_shift;
2380 
2381 		/*
2382 		 * The maximum LBA is the size - 1.
2383 		 *
2384 		 * XXX: Note that this field is being updated without locking,
2385 		 * 	which might cause problems on 32-bit architectures.
2386 		 */
2387 		be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ?
2388 		    0 : (be_lun->size_blocks - 1);
2389 		be_lun->ctl_be_lun.blocksize = be_lun->blocksize;
2390 		be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp;
2391 		be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff;
2392 		if (be_lun->dispatch == ctl_be_block_dispatch_zvol &&
2393 		    be_lun->blocksize != 0)
2394 			be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE /
2395 			    be_lun->blocksize;
2396 		ctl_lun_capacity_changed(&be_lun->ctl_be_lun);
2397 		if (oldsize == 0 && be_lun->size_blocks != 0)
2398 			ctl_lun_online(&be_lun->ctl_be_lun);
2399 	}
2400 
2401 	/* Tell the user the exact size we ended up using */
2402 	params->lun_size_bytes = be_lun->size_bytes;
2403 
2404 	req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2405 
2406 	return (0);
2407 
2408 bailout_error:
2409 	req->status = CTL_LUN_ERROR;
2410 
2411 	return (0);
2412 }
2413 
2414 static void
2415 ctl_be_block_lun_shutdown(void *be_lun)
2416 {
2417 	struct ctl_be_block_lun *lun;
2418 	struct ctl_be_block_softc *softc;
2419 
2420 	lun = (struct ctl_be_block_lun *)be_lun;
2421 
2422 	softc = lun->softc;
2423 
2424 	mtx_lock(&softc->lock);
2425 	lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2426 	if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2427 		wakeup(lun);
2428 	mtx_unlock(&softc->lock);
2429 
2430 }
2431 
2432 static void
2433 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status)
2434 {
2435 	struct ctl_be_block_lun *lun;
2436 	struct ctl_be_block_softc *softc;
2437 
2438 	lun = (struct ctl_be_block_lun *)be_lun;
2439 	softc = lun->softc;
2440 
2441 	if (status == CTL_LUN_CONFIG_OK) {
2442 		mtx_lock(&softc->lock);
2443 		lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2444 		if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2445 			wakeup(lun);
2446 		mtx_unlock(&softc->lock);
2447 
2448 		/*
2449 		 * We successfully added the LUN, attempt to enable it.
2450 		 */
2451 		if (ctl_enable_lun(&lun->ctl_be_lun) != 0) {
2452 			printf("%s: ctl_enable_lun() failed!\n", __func__);
2453 			if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) {
2454 				printf("%s: ctl_invalidate_lun() failed!\n",
2455 				       __func__);
2456 			}
2457 		}
2458 
2459 		return;
2460 	}
2461 
2462 
2463 	mtx_lock(&softc->lock);
2464 	lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2465 	lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR;
2466 	wakeup(lun);
2467 	mtx_unlock(&softc->lock);
2468 }
2469 
2470 
2471 static int
2472 ctl_be_block_config_write(union ctl_io *io)
2473 {
2474 	struct ctl_be_block_lun *be_lun;
2475 	struct ctl_be_lun *ctl_be_lun;
2476 	int retval;
2477 
2478 	retval = 0;
2479 
2480 	DPRINTF("entered\n");
2481 
2482 	ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
2483 		CTL_PRIV_BACKEND_LUN].ptr;
2484 	be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
2485 
2486 	switch (io->scsiio.cdb[0]) {
2487 	case SYNCHRONIZE_CACHE:
2488 	case SYNCHRONIZE_CACHE_16:
2489 	case WRITE_SAME_10:
2490 	case WRITE_SAME_16:
2491 	case UNMAP:
2492 		/*
2493 		 * The upper level CTL code will filter out any CDBs with
2494 		 * the immediate bit set and return the proper error.
2495 		 *
2496 		 * We don't really need to worry about what LBA range the
2497 		 * user asked to be synced out.  When they issue a sync
2498 		 * cache command, we'll sync out the whole thing.
2499 		 */
2500 		mtx_lock(&be_lun->queue_lock);
2501 		STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2502 				   links);
2503 		mtx_unlock(&be_lun->queue_lock);
2504 		taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2505 		break;
2506 	case START_STOP_UNIT: {
2507 		struct scsi_start_stop_unit *cdb;
2508 
2509 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2510 
2511 		if (cdb->how & SSS_START)
2512 			retval = ctl_start_lun(ctl_be_lun);
2513 		else {
2514 			retval = ctl_stop_lun(ctl_be_lun);
2515 			/*
2516 			 * XXX KDM Copan-specific offline behavior.
2517 			 * Figure out a reasonable way to port this?
2518 			 */
2519 #ifdef NEEDTOPORT
2520 			if ((retval == 0)
2521 			 && (cdb->byte2 & SSS_ONOFFLINE))
2522 				retval = ctl_lun_offline(ctl_be_lun);
2523 #endif
2524 		}
2525 
2526 		/*
2527 		 * In general, the above routines should not fail.  They
2528 		 * just set state for the LUN.  So we've got something
2529 		 * pretty wrong here if we can't start or stop the LUN.
2530 		 */
2531 		if (retval != 0) {
2532 			ctl_set_internal_failure(&io->scsiio,
2533 						 /*sks_valid*/ 1,
2534 						 /*retry_count*/ 0xf051);
2535 			retval = CTL_RETVAL_COMPLETE;
2536 		} else {
2537 			ctl_set_success(&io->scsiio);
2538 		}
2539 		ctl_config_write_done(io);
2540 		break;
2541 	}
2542 	default:
2543 		ctl_set_invalid_opcode(&io->scsiio);
2544 		ctl_config_write_done(io);
2545 		retval = CTL_RETVAL_COMPLETE;
2546 		break;
2547 	}
2548 
2549 	return (retval);
2550 
2551 }
2552 
2553 static int
2554 ctl_be_block_config_read(union ctl_io *io)
2555 {
2556 	return (0);
2557 }
2558 
2559 static int
2560 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb)
2561 {
2562 	struct ctl_be_block_lun *lun;
2563 	int retval;
2564 
2565 	lun = (struct ctl_be_block_lun *)be_lun;
2566 	retval = 0;
2567 
2568 	retval = sbuf_printf(sb, "\t<num_threads>");
2569 
2570 	if (retval != 0)
2571 		goto bailout;
2572 
2573 	retval = sbuf_printf(sb, "%d", lun->num_threads);
2574 
2575 	if (retval != 0)
2576 		goto bailout;
2577 
2578 	retval = sbuf_printf(sb, "</num_threads>\n");
2579 
2580 bailout:
2581 
2582 	return (retval);
2583 }
2584 
2585 int
2586 ctl_be_block_init(void)
2587 {
2588 	struct ctl_be_block_softc *softc;
2589 	int retval;
2590 
2591 	softc = &backend_block_softc;
2592 	retval = 0;
2593 
2594 	mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
2595 	beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2596 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2597 	STAILQ_INIT(&softc->disk_list);
2598 	STAILQ_INIT(&softc->lun_list);
2599 
2600 	return (retval);
2601 }
2602