xref: /freebsd/sys/cam/ctl/ctl_backend_ramdisk.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2003, 2008 Silicon Graphics International Corp.
5  * Copyright (c) 2012 The FreeBSD Foundation
6  * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Portions of this software were developed by Edward Tomasz Napierala
10  * under sponsorship from the FreeBSD Foundation.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions, and the following disclaimer,
17  *    without modification.
18  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
19  *    substantially similar to the "NO WARRANTY" disclaimer below
20  *    ("Disclaimer") and any redistribution must be conditioned upon
21  *    including a substantially similar Disclaimer requirement for further
22  *    binary redistribution.
23  *
24  * NO WARRANTY
25  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
28  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
29  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
33  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
34  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGES.
36  *
37  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_ramdisk.c#3 $
38  */
39 /*
40  * CAM Target Layer black hole and RAM disk backend.
41  *
42  * Author: Ken Merry <ken@FreeBSD.org>
43  */
44 
45 #include <sys/cdefs.h>
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/condvar.h>
50 #include <sys/types.h>
51 #include <sys/limits.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/malloc.h>
55 #include <sys/sx.h>
56 #include <sys/taskqueue.h>
57 #include <sys/time.h>
58 #include <sys/queue.h>
59 #include <sys/conf.h>
60 #include <sys/ioccom.h>
61 #include <sys/module.h>
62 #include <sys/sysctl.h>
63 #include <sys/nv.h>
64 #include <sys/dnv.h>
65 
66 #include <cam/scsi/scsi_all.h>
67 #include <cam/scsi/scsi_da.h>
68 #include <cam/ctl/ctl_io.h>
69 #include <cam/ctl/ctl.h>
70 #include <cam/ctl/ctl_util.h>
71 #include <cam/ctl/ctl_backend.h>
72 #include <cam/ctl/ctl_debug.h>
73 #include <cam/ctl/ctl_ioctl.h>
74 #include <cam/ctl/ctl_ha.h>
75 #include <cam/ctl/ctl_private.h>
76 #include <cam/ctl/ctl_error.h>
77 
78 #define PRIV(io)	\
79     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
80 #define ARGS(io)	\
81     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
82 
83 #define	PPP	(PAGE_SIZE / sizeof(uint8_t **))
84 #ifdef __LP64__
85 #define	PPPS	(PAGE_SHIFT - 3)
86 #else
87 #define	PPPS	(PAGE_SHIFT - 2)
88 #endif
89 #define	SGPP	(PAGE_SIZE / sizeof(struct ctl_sg_entry))
90 
91 #define	P_UNMAPPED	NULL			/* Page is unmapped. */
92 #define	P_ANCHORED	((void *)(uintptr_t)1)	/* Page is anchored. */
93 
94 typedef enum {
95 	GP_READ,	/* Return data page or zero page. */
96 	GP_WRITE,	/* Return data page, try allocate if none. */
97 	GP_ANCHOR,	/* Return data page, try anchor if none. */
98 	GP_OTHER,	/* Return what present, do not allocate/anchor. */
99 } getpage_op_t;
100 
101 typedef enum {
102 	CTL_BE_RAMDISK_LUN_UNCONFIGURED	= 0x01,
103 	CTL_BE_RAMDISK_LUN_WAITING	= 0x04
104 } ctl_be_ramdisk_lun_flags;
105 
106 struct ctl_be_ramdisk_lun {
107 	struct ctl_be_lun	cbe_lun;	/* Must be first element. */
108 	struct ctl_lun_create_params params;
109 	int			indir;
110 	uint8_t			**pages;
111 	uint8_t			*zero_page;
112 	struct sx		page_lock;
113 	u_int			pblocksize;
114 	u_int			pblockmul;
115 	uint64_t		size_bytes;
116 	uint64_t		size_blocks;
117 	uint64_t		cap_bytes;
118 	uint64_t		cap_used;
119 	struct ctl_be_ramdisk_softc *softc;
120 	ctl_be_ramdisk_lun_flags flags;
121 	SLIST_ENTRY(ctl_be_ramdisk_lun) links;
122 	struct taskqueue	*io_taskqueue;
123 	struct task		io_task;
124 	STAILQ_HEAD(, ctl_io_hdr) cont_queue;
125 	struct mtx_padalign	queue_lock;
126 };
127 
128 struct ctl_be_ramdisk_softc {
129 	struct sx modify_lock;
130 	struct mtx lock;
131 	int num_luns;
132 	SLIST_HEAD(, ctl_be_ramdisk_lun) lun_list;
133 };
134 
135 static struct ctl_be_ramdisk_softc rd_softc;
136 extern struct ctl_softc *control_softc;
137 
138 static int ctl_backend_ramdisk_init(void);
139 static int ctl_backend_ramdisk_shutdown(void);
140 static int ctl_backend_ramdisk_move_done(union ctl_io *io, bool samethr);
141 static void ctl_backend_ramdisk_compare(union ctl_io *io);
142 static void ctl_backend_ramdisk_rw(union ctl_io *io);
143 static int ctl_backend_ramdisk_submit(union ctl_io *io);
144 static void ctl_backend_ramdisk_worker(void *context, int pending);
145 static int ctl_backend_ramdisk_config_read(union ctl_io *io);
146 static int ctl_backend_ramdisk_config_write(union ctl_io *io);
147 static uint64_t ctl_backend_ramdisk_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname);
148 static int ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd,
149 				     caddr_t addr, int flag, struct thread *td);
150 static int ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
151 				  struct ctl_lun_req *req);
152 static int ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
153 				      struct ctl_lun_req *req);
154 static int ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
155 				  struct ctl_lun_req *req);
156 static void ctl_backend_ramdisk_lun_shutdown(struct ctl_be_lun *cbe_lun);
157 
158 static struct ctl_backend_driver ctl_be_ramdisk_driver =
159 {
160 	.name = "ramdisk",
161 	.flags = CTL_BE_FLAG_HAS_CONFIG,
162 	.init = ctl_backend_ramdisk_init,
163 	.shutdown = ctl_backend_ramdisk_shutdown,
164 	.data_submit = ctl_backend_ramdisk_submit,
165 	.config_read = ctl_backend_ramdisk_config_read,
166 	.config_write = ctl_backend_ramdisk_config_write,
167 	.ioctl = ctl_backend_ramdisk_ioctl,
168 	.lun_attr = ctl_backend_ramdisk_lun_attr,
169 };
170 
171 MALLOC_DEFINE(M_RAMDISK, "ctlramdisk", "Memory used for CTL RAMdisk");
172 CTL_BACKEND_DECLARE(cbr, ctl_be_ramdisk_driver);
173 
174 static int
175 ctl_backend_ramdisk_init(void)
176 {
177 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
178 
179 	memset(softc, 0, sizeof(*softc));
180 	sx_init(&softc->modify_lock, "ctlrammod");
181 	mtx_init(&softc->lock, "ctlram", NULL, MTX_DEF);
182 	SLIST_INIT(&softc->lun_list);
183 	return (0);
184 }
185 
186 static int
187 ctl_backend_ramdisk_shutdown(void)
188 {
189 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
190 	struct ctl_be_ramdisk_lun *lun;
191 
192 	mtx_lock(&softc->lock);
193 	while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
194 		SLIST_REMOVE_HEAD(&softc->lun_list, links);
195 		softc->num_luns--;
196 		/*
197 		 * Drop our lock here.  Since ctl_remove_lun() can call
198 		 * back into us, this could potentially lead to a recursive
199 		 * lock of the same mutex, which would cause a hang.
200 		 */
201 		mtx_unlock(&softc->lock);
202 		ctl_remove_lun(&lun->cbe_lun);
203 		mtx_lock(&softc->lock);
204 	}
205 	mtx_unlock(&softc->lock);
206 	mtx_destroy(&softc->lock);
207 	sx_destroy(&softc->modify_lock);
208 	return (0);
209 }
210 
211 static uint8_t *
212 ctl_backend_ramdisk_getpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn,
213     getpage_op_t op)
214 {
215 	uint8_t **p, ***pp;
216 	off_t i;
217 	int s;
218 
219 	if (be_lun->cap_bytes == 0) {
220 		switch (op) {
221 		case GP_READ:
222 			return (be_lun->zero_page);
223 		case GP_WRITE:
224 			return ((uint8_t *)be_lun->pages);
225 		case GP_ANCHOR:
226 			return (P_ANCHORED);
227 		default:
228 			return (P_UNMAPPED);
229 		}
230 	}
231 	if (op == GP_WRITE || op == GP_ANCHOR) {
232 		sx_xlock(&be_lun->page_lock);
233 		pp = &be_lun->pages;
234 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
235 			if (*pp == NULL) {
236 				*pp = malloc(PAGE_SIZE, M_RAMDISK,
237 				    M_WAITOK|M_ZERO);
238 			}
239 			i = pn >> s;
240 			pp = (uint8_t ***)&(*pp)[i];
241 			pn -= i << s;
242 		}
243 		if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
244 			if (op == GP_WRITE) {
245 				*pp = malloc(be_lun->pblocksize, M_RAMDISK,
246 				    M_WAITOK|M_ZERO);
247 			} else
248 				*pp = P_ANCHORED;
249 			be_lun->cap_used += be_lun->pblocksize;
250 		} else if (*pp == P_ANCHORED && op == GP_WRITE) {
251 			*pp = malloc(be_lun->pblocksize, M_RAMDISK,
252 			    M_WAITOK|M_ZERO);
253 		}
254 		sx_xunlock(&be_lun->page_lock);
255 		return ((uint8_t *)*pp);
256 	} else {
257 		sx_slock(&be_lun->page_lock);
258 		p = be_lun->pages;
259 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
260 			if (p == NULL)
261 				break;
262 			i = pn >> s;
263 			p = (uint8_t **)p[i];
264 			pn -= i << s;
265 		}
266 		sx_sunlock(&be_lun->page_lock);
267 		if ((p == P_UNMAPPED || p == P_ANCHORED) && op == GP_READ)
268 			return (be_lun->zero_page);
269 		return ((uint8_t *)p);
270 	}
271 };
272 
273 static void
274 ctl_backend_ramdisk_unmappage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
275 {
276 	uint8_t ***pp;
277 	off_t i;
278 	int s;
279 
280 	if (be_lun->cap_bytes == 0)
281 		return;
282 	sx_xlock(&be_lun->page_lock);
283 	pp = &be_lun->pages;
284 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
285 		if (*pp == NULL)
286 			goto noindir;
287 		i = pn >> s;
288 		pp = (uint8_t ***)&(*pp)[i];
289 		pn -= i << s;
290 	}
291 	if (*pp == P_ANCHORED) {
292 		be_lun->cap_used -= be_lun->pblocksize;
293 		*pp = P_UNMAPPED;
294 	} else if (*pp != P_UNMAPPED) {
295 		free(*pp, M_RAMDISK);
296 		be_lun->cap_used -= be_lun->pblocksize;
297 		*pp = P_UNMAPPED;
298 	}
299 noindir:
300 	sx_xunlock(&be_lun->page_lock);
301 };
302 
303 static void
304 ctl_backend_ramdisk_anchorpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
305 {
306 	uint8_t ***pp;
307 	off_t i;
308 	int s;
309 
310 	if (be_lun->cap_bytes == 0)
311 		return;
312 	sx_xlock(&be_lun->page_lock);
313 	pp = &be_lun->pages;
314 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
315 		if (*pp == NULL)
316 			goto noindir;
317 		i = pn >> s;
318 		pp = (uint8_t ***)&(*pp)[i];
319 		pn -= i << s;
320 	}
321 	if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
322 		be_lun->cap_used += be_lun->pblocksize;
323 		*pp = P_ANCHORED;
324 	} else if (*pp != P_ANCHORED) {
325 		free(*pp, M_RAMDISK);
326 		*pp = P_ANCHORED;
327 	}
328 noindir:
329 	sx_xunlock(&be_lun->page_lock);
330 };
331 
332 static void
333 ctl_backend_ramdisk_freeallpages(uint8_t **p, int indir)
334 {
335 	int i;
336 
337 	if (p == NULL)
338 		return;
339 	if (indir == 0) {
340 		free(p, M_RAMDISK);
341 		return;
342 	}
343 	for (i = 0; i < PPP; i++) {
344 		if (p[i] == NULL)
345 			continue;
346 		ctl_backend_ramdisk_freeallpages((uint8_t **)p[i], indir - 1);
347 	}
348 	free(p, M_RAMDISK);
349 };
350 
351 static size_t
352 cmp(uint8_t *a, uint8_t *b, size_t size)
353 {
354 	size_t i;
355 
356 	for (i = 0; i < size; i++) {
357 		if (a[i] != b[i])
358 			break;
359 	}
360 	return (i);
361 }
362 
363 static int
364 ctl_backend_ramdisk_cmp(union ctl_io *io)
365 {
366 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
367 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
368 	uint8_t *page;
369 	uint8_t info[8];
370 	uint64_t lba;
371 	u_int lbaoff, lbas, res, off;
372 
373 	lbas = io->scsiio.kern_data_len / cbe_lun->blocksize;
374 	lba = ARGS(io)->lba + PRIV(io)->len - lbas;
375 	off = 0;
376 	for (; lbas > 0; lbas--, lba++) {
377 		page = ctl_backend_ramdisk_getpage(be_lun,
378 		    lba >> cbe_lun->pblockexp, GP_READ);
379 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
380 		page += lbaoff * cbe_lun->blocksize;
381 		res = cmp(io->scsiio.kern_data_ptr + off, page,
382 		    cbe_lun->blocksize);
383 		off += res;
384 		if (res < cbe_lun->blocksize)
385 			break;
386 	}
387 	free(io->scsiio.kern_data_ptr, M_RAMDISK);
388 	if (lbas > 0) {
389 		off += io->scsiio.kern_rel_offset - io->scsiio.kern_data_len;
390 		scsi_u64to8b(off, info);
391 		ctl_set_sense(&io->scsiio, /*current_error*/ 1,
392 		    /*sense_key*/ SSD_KEY_MISCOMPARE,
393 		    /*asc*/ 0x1D, /*ascq*/ 0x00,
394 		    /*type*/ SSD_ELEM_INFO,
395 		    /*size*/ sizeof(info), /*data*/ &info,
396 		    /*type*/ SSD_ELEM_NONE);
397 		return (1);
398 	}
399 	return (0);
400 }
401 
402 static int
403 ctl_backend_ramdisk_move_done(union ctl_io *io, bool samethr)
404 {
405 	struct ctl_be_ramdisk_lun *be_lun =
406 	    (struct ctl_be_ramdisk_lun *)CTL_BACKEND_LUN(io);
407 
408 	CTL_DEBUG_PRINT(("ctl_backend_ramdisk_move_done\n"));
409 	if (io->scsiio.kern_sg_entries > 0)
410 		free(io->scsiio.kern_data_ptr, M_RAMDISK);
411 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
412 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
413 	    (io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE) {
414 		if (ARGS(io)->flags & CTL_LLF_COMPARE) {
415 			/* We have data block ready for comparison. */
416 			if (ctl_backend_ramdisk_cmp(io))
417 				goto done;
418 		}
419 		if (ARGS(io)->len > PRIV(io)->len) {
420 			mtx_lock(&be_lun->queue_lock);
421 			STAILQ_INSERT_TAIL(&be_lun->cont_queue,
422 			    &io->io_hdr, links);
423 			mtx_unlock(&be_lun->queue_lock);
424 			taskqueue_enqueue(be_lun->io_taskqueue,
425 			    &be_lun->io_task);
426 			return (0);
427 		}
428 		ctl_set_success(&io->scsiio);
429 	}
430 done:
431 	ctl_data_submit_done(io);
432 	return(0);
433 }
434 
435 static void
436 ctl_backend_ramdisk_compare(union ctl_io *io)
437 {
438 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
439 	u_int lbas, len;
440 
441 	lbas = ARGS(io)->len - PRIV(io)->len;
442 	lbas = MIN(lbas, 131072 / cbe_lun->blocksize);
443 	len = lbas * cbe_lun->blocksize;
444 
445 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
446 	io->scsiio.kern_data_ptr = malloc(len, M_RAMDISK, M_WAITOK);
447 	io->scsiio.kern_data_len = len;
448 	io->scsiio.kern_sg_entries = 0;
449 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
450 	PRIV(io)->len += lbas;
451 	ctl_datamove(io);
452 }
453 
454 static void
455 ctl_backend_ramdisk_rw(union ctl_io *io)
456 {
457 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
458 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
459 	struct ctl_sg_entry *sg_entries;
460 	uint8_t *page;
461 	uint64_t lba;
462 	u_int i, len, lbaoff, lbas, sgs, off;
463 	getpage_op_t op;
464 
465 	lba = ARGS(io)->lba + PRIV(io)->len;
466 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
467 	lbas = ARGS(io)->len - PRIV(io)->len;
468 	lbas = MIN(lbas, (SGPP << cbe_lun->pblockexp) - lbaoff);
469 	sgs = (lbas + lbaoff + be_lun->pblockmul - 1) >> cbe_lun->pblockexp;
470 	off = lbaoff * cbe_lun->blocksize;
471 	op = (ARGS(io)->flags & CTL_LLF_WRITE) ? GP_WRITE : GP_READ;
472 	if (sgs > 1) {
473 		io->scsiio.kern_data_ptr = malloc(sizeof(struct ctl_sg_entry) *
474 		    sgs, M_RAMDISK, M_WAITOK);
475 		sg_entries = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
476 		len = lbas * cbe_lun->blocksize;
477 		for (i = 0; i < sgs; i++) {
478 			page = ctl_backend_ramdisk_getpage(be_lun,
479 			    (lba >> cbe_lun->pblockexp) + i, op);
480 			if (page == P_UNMAPPED || page == P_ANCHORED) {
481 				free(io->scsiio.kern_data_ptr, M_RAMDISK);
482 nospc:
483 				ctl_set_space_alloc_fail(&io->scsiio);
484 				ctl_data_submit_done(io);
485 				return;
486 			}
487 			sg_entries[i].addr = page + off;
488 			sg_entries[i].len = MIN(len, be_lun->pblocksize - off);
489 			len -= sg_entries[i].len;
490 			off = 0;
491 		}
492 	} else {
493 		page = ctl_backend_ramdisk_getpage(be_lun,
494 		    lba >> cbe_lun->pblockexp, op);
495 		if (page == P_UNMAPPED || page == P_ANCHORED)
496 			goto nospc;
497 		sgs = 0;
498 		io->scsiio.kern_data_ptr = page + off;
499 	}
500 
501 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
502 	io->scsiio.kern_data_len = lbas * cbe_lun->blocksize;
503 	io->scsiio.kern_sg_entries = sgs;
504 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
505 	PRIV(io)->len += lbas;
506 	if ((ARGS(io)->flags & CTL_LLF_READ) &&
507 	    ARGS(io)->len <= PRIV(io)->len) {
508 		ctl_set_success(&io->scsiio);
509 		if (cbe_lun->serseq >= CTL_LUN_SERSEQ_SOFT)
510 			ctl_serseq_done(io);
511 	}
512 	ctl_datamove(io);
513 }
514 
515 static int
516 ctl_backend_ramdisk_submit(union ctl_io *io)
517 {
518 	struct ctl_lba_len_flags *lbalen = ARGS(io);
519 
520 	if (lbalen->flags & CTL_LLF_VERIFY) {
521 		ctl_set_success(&io->scsiio);
522 		ctl_data_submit_done(io);
523 		return (CTL_RETVAL_COMPLETE);
524 	}
525 	PRIV(io)->len = 0;
526 	if (lbalen->flags & CTL_LLF_COMPARE)
527 		ctl_backend_ramdisk_compare(io);
528 	else
529 		ctl_backend_ramdisk_rw(io);
530 	return (CTL_RETVAL_COMPLETE);
531 }
532 
533 static void
534 ctl_backend_ramdisk_worker(void *context, int pending)
535 {
536 	struct ctl_be_ramdisk_lun *be_lun;
537 	union ctl_io *io;
538 
539 	be_lun = (struct ctl_be_ramdisk_lun *)context;
540 	mtx_lock(&be_lun->queue_lock);
541 	for (;;) {
542 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->cont_queue);
543 		if (io != NULL) {
544 			STAILQ_REMOVE_HEAD(&be_lun->cont_queue, links);
545 			mtx_unlock(&be_lun->queue_lock);
546 			if (ARGS(io)->flags & CTL_LLF_COMPARE)
547 				ctl_backend_ramdisk_compare(io);
548 			else
549 				ctl_backend_ramdisk_rw(io);
550 			mtx_lock(&be_lun->queue_lock);
551 			continue;
552 		}
553 
554 		/*
555 		 * If we get here, there is no work left in the queues, so
556 		 * just break out and let the task queue go to sleep.
557 		 */
558 		break;
559 	}
560 	mtx_unlock(&be_lun->queue_lock);
561 }
562 
563 static int
564 ctl_backend_ramdisk_gls(union ctl_io *io)
565 {
566 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
567 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
568 	struct scsi_get_lba_status_data *data;
569 	uint8_t *page;
570 	u_int lbaoff;
571 
572 	data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
573 	scsi_u64to8b(ARGS(io)->lba, data->descr[0].addr);
574 	lbaoff = ARGS(io)->lba & ~(UINT_MAX << cbe_lun->pblockexp);
575 	scsi_ulto4b(be_lun->pblockmul - lbaoff, data->descr[0].length);
576 	page = ctl_backend_ramdisk_getpage(be_lun,
577 	    ARGS(io)->lba >> cbe_lun->pblockexp, GP_OTHER);
578 	if (page == P_UNMAPPED)
579 		data->descr[0].status = 1;
580 	else if (page == P_ANCHORED)
581 		data->descr[0].status = 2;
582 	else
583 		data->descr[0].status = 0;
584 	ctl_config_read_done(io);
585 	return (CTL_RETVAL_COMPLETE);
586 }
587 
588 static int
589 ctl_backend_ramdisk_config_read(union ctl_io *io)
590 {
591 	int retval = 0;
592 
593 	switch (io->scsiio.cdb[0]) {
594 	case SERVICE_ACTION_IN:
595 		if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
596 			retval = ctl_backend_ramdisk_gls(io);
597 			break;
598 		}
599 		ctl_set_invalid_field(&io->scsiio,
600 				      /*sks_valid*/ 1,
601 				      /*command*/ 1,
602 				      /*field*/ 1,
603 				      /*bit_valid*/ 1,
604 				      /*bit*/ 4);
605 		ctl_config_read_done(io);
606 		retval = CTL_RETVAL_COMPLETE;
607 		break;
608 	default:
609 		ctl_set_invalid_opcode(&io->scsiio);
610 		ctl_config_read_done(io);
611 		retval = CTL_RETVAL_COMPLETE;
612 		break;
613 	}
614 	return (retval);
615 }
616 
617 static void
618 ctl_backend_ramdisk_delete(struct ctl_be_lun *cbe_lun, off_t lba, off_t len,
619     int anchor)
620 {
621 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
622 	uint8_t *page;
623 	uint64_t p, lp;
624 	u_int lbaoff;
625 	getpage_op_t op = anchor ? GP_ANCHOR : GP_OTHER;
626 
627 	/* Partially zero first partial page. */
628 	p = lba >> cbe_lun->pblockexp;
629 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
630 	if (lbaoff != 0) {
631 		page = ctl_backend_ramdisk_getpage(be_lun, p, op);
632 		if (page != P_UNMAPPED && page != P_ANCHORED) {
633 			memset(page + lbaoff * cbe_lun->blocksize, 0,
634 			    min(len, be_lun->pblockmul - lbaoff) *
635 			    cbe_lun->blocksize);
636 		}
637 		p++;
638 	}
639 
640 	/* Partially zero last partial page. */
641 	lp = (lba + len) >> cbe_lun->pblockexp;
642 	lbaoff = (lba + len) & ~(UINT_MAX << cbe_lun->pblockexp);
643 	if (p <= lp && lbaoff != 0) {
644 		page = ctl_backend_ramdisk_getpage(be_lun, lp, op);
645 		if (page != P_UNMAPPED && page != P_ANCHORED)
646 			memset(page, 0, lbaoff * cbe_lun->blocksize);
647 	}
648 
649 	/* Delete remaining full pages. */
650 	if (anchor) {
651 		for (; p < lp; p++)
652 			ctl_backend_ramdisk_anchorpage(be_lun, p);
653 	} else {
654 		for (; p < lp; p++)
655 			ctl_backend_ramdisk_unmappage(be_lun, p);
656 	}
657 }
658 
659 static void
660 ctl_backend_ramdisk_ws(union ctl_io *io)
661 {
662 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
663 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
664 	struct ctl_lba_len_flags *lbalen = ARGS(io);
665 	uint8_t *page;
666 	uint64_t lba;
667 	u_int lbaoff, lbas;
668 
669 	if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB)) {
670 		ctl_set_invalid_field(&io->scsiio,
671 				      /*sks_valid*/ 1,
672 				      /*command*/ 1,
673 				      /*field*/ 1,
674 				      /*bit_valid*/ 0,
675 				      /*bit*/ 0);
676 		ctl_config_write_done(io);
677 		return;
678 	}
679 	if (lbalen->flags & SWS_UNMAP) {
680 		ctl_backend_ramdisk_delete(cbe_lun, lbalen->lba, lbalen->len,
681 		    (lbalen->flags & SWS_ANCHOR) != 0);
682 		ctl_set_success(&io->scsiio);
683 		ctl_config_write_done(io);
684 		return;
685 	}
686 
687 	for (lba = lbalen->lba, lbas = lbalen->len; lbas > 0; lba++, lbas--) {
688 		page = ctl_backend_ramdisk_getpage(be_lun,
689 		    lba >> cbe_lun->pblockexp, GP_WRITE);
690 		if (page == P_UNMAPPED || page == P_ANCHORED) {
691 			ctl_set_space_alloc_fail(&io->scsiio);
692 			ctl_data_submit_done(io);
693 			return;
694 		}
695 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
696 		page += lbaoff * cbe_lun->blocksize;
697 		if (lbalen->flags & SWS_NDOB) {
698 			memset(page, 0, cbe_lun->blocksize);
699 		} else {
700 			memcpy(page, io->scsiio.kern_data_ptr,
701 			    cbe_lun->blocksize);
702 		}
703 		if (lbalen->flags & SWS_LBDATA)
704 			scsi_ulto4b(lba, page);
705 	}
706 	ctl_set_success(&io->scsiio);
707 	ctl_config_write_done(io);
708 }
709 
710 static void
711 ctl_backend_ramdisk_unmap(union ctl_io *io)
712 {
713 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
714 	struct ctl_ptr_len_flags *ptrlen = (struct ctl_ptr_len_flags *)ARGS(io);
715 	struct scsi_unmap_desc *buf, *end;
716 
717 	if ((ptrlen->flags & ~SU_ANCHOR) != 0) {
718 		ctl_set_invalid_field(&io->scsiio,
719 				      /*sks_valid*/ 0,
720 				      /*command*/ 0,
721 				      /*field*/ 0,
722 				      /*bit_valid*/ 0,
723 				      /*bit*/ 0);
724 		ctl_config_write_done(io);
725 		return;
726 	}
727 
728 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
729 	end = buf + ptrlen->len / sizeof(*buf);
730 	for (; buf < end; buf++) {
731 		ctl_backend_ramdisk_delete(cbe_lun,
732 		    scsi_8btou64(buf->lba), scsi_4btoul(buf->length),
733 		    (ptrlen->flags & SU_ANCHOR) != 0);
734 	}
735 
736 	ctl_set_success(&io->scsiio);
737 	ctl_config_write_done(io);
738 }
739 
740 static int
741 ctl_backend_ramdisk_config_write(union ctl_io *io)
742 {
743 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
744 	int retval = 0;
745 
746 	switch (io->scsiio.cdb[0]) {
747 	case SYNCHRONIZE_CACHE:
748 	case SYNCHRONIZE_CACHE_16:
749 		/* We have no cache to flush. */
750 		ctl_set_success(&io->scsiio);
751 		ctl_config_write_done(io);
752 		break;
753 	case START_STOP_UNIT: {
754 		struct scsi_start_stop_unit *cdb;
755 
756 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
757 		if ((cdb->how & SSS_PC_MASK) != 0) {
758 			ctl_set_success(&io->scsiio);
759 			ctl_config_write_done(io);
760 			break;
761 		}
762 		if (cdb->how & SSS_START) {
763 			if (cdb->how & SSS_LOEJ)
764 				ctl_lun_has_media(cbe_lun);
765 			ctl_start_lun(cbe_lun);
766 		} else {
767 			ctl_stop_lun(cbe_lun);
768 			if (cdb->how & SSS_LOEJ)
769 				ctl_lun_ejected(cbe_lun);
770 		}
771 		ctl_set_success(&io->scsiio);
772 		ctl_config_write_done(io);
773 		break;
774 	}
775 	case PREVENT_ALLOW:
776 		ctl_set_success(&io->scsiio);
777 		ctl_config_write_done(io);
778 		break;
779 	case WRITE_SAME_10:
780 	case WRITE_SAME_16:
781 		ctl_backend_ramdisk_ws(io);
782 		break;
783 	case UNMAP:
784 		ctl_backend_ramdisk_unmap(io);
785 		break;
786 	default:
787 		ctl_set_invalid_opcode(&io->scsiio);
788 		ctl_config_write_done(io);
789 		retval = CTL_RETVAL_COMPLETE;
790 		break;
791 	}
792 
793 	return (retval);
794 }
795 
796 static uint64_t
797 ctl_backend_ramdisk_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
798 {
799 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
800 	uint64_t		val;
801 
802 	val = UINT64_MAX;
803 	if (be_lun->cap_bytes == 0)
804 		return (val);
805 	sx_slock(&be_lun->page_lock);
806 	if (strcmp(attrname, "blocksused") == 0) {
807 		val = be_lun->cap_used / be_lun->cbe_lun.blocksize;
808 	} else if (strcmp(attrname, "blocksavail") == 0) {
809 		val = (be_lun->cap_bytes - be_lun->cap_used) /
810 		    be_lun->cbe_lun.blocksize;
811 	}
812 	sx_sunlock(&be_lun->page_lock);
813 	return (val);
814 }
815 
816 static int
817 ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
818 			  int flag, struct thread *td)
819 {
820 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
821 	struct ctl_lun_req *lun_req;
822 	int retval;
823 
824 	retval = 0;
825 	switch (cmd) {
826 	case CTL_LUN_REQ:
827 		lun_req = (struct ctl_lun_req *)addr;
828 		switch (lun_req->reqtype) {
829 		case CTL_LUNREQ_CREATE:
830 			retval = ctl_backend_ramdisk_create(softc, lun_req);
831 			break;
832 		case CTL_LUNREQ_RM:
833 			retval = ctl_backend_ramdisk_rm(softc, lun_req);
834 			break;
835 		case CTL_LUNREQ_MODIFY:
836 			retval = ctl_backend_ramdisk_modify(softc, lun_req);
837 			break;
838 		default:
839 			lun_req->status = CTL_LUN_ERROR;
840 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
841 				 "%s: invalid LUN request type %d", __func__,
842 				 lun_req->reqtype);
843 			break;
844 		}
845 		break;
846 	default:
847 		retval = ENOTTY;
848 		break;
849 	}
850 
851 	return (retval);
852 }
853 
854 static int
855 ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
856 		       struct ctl_lun_req *req)
857 {
858 	struct ctl_be_ramdisk_lun *be_lun;
859 	struct ctl_lun_rm_params *params;
860 	int retval;
861 
862 	params = &req->reqdata.rm;
863 	sx_xlock(&softc->modify_lock);
864 	mtx_lock(&softc->lock);
865 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
866 		if (be_lun->cbe_lun.lun_id == params->lun_id) {
867 			SLIST_REMOVE(&softc->lun_list, be_lun,
868 			    ctl_be_ramdisk_lun, links);
869 			softc->num_luns--;
870 			break;
871 		}
872 	}
873 	mtx_unlock(&softc->lock);
874 	sx_xunlock(&softc->modify_lock);
875 	if (be_lun == NULL) {
876 		snprintf(req->error_str, sizeof(req->error_str),
877 			 "%s: LUN %u is not managed by the ramdisk backend",
878 			 __func__, params->lun_id);
879 		goto bailout_error;
880 	}
881 
882 	/*
883 	 * Set the waiting flag before we invalidate the LUN.  Our shutdown
884 	 * routine can be called any time after we invalidate the LUN,
885 	 * and can be called from our context.
886 	 *
887 	 * This tells the shutdown routine that we're waiting, or we're
888 	 * going to wait for the shutdown to happen.
889 	 */
890 	mtx_lock(&softc->lock);
891 	be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING;
892 	mtx_unlock(&softc->lock);
893 
894 	retval = ctl_remove_lun(&be_lun->cbe_lun);
895 	if (retval != 0) {
896 		snprintf(req->error_str, sizeof(req->error_str),
897 			 "%s: error %d returned from ctl_remove_lun() for "
898 			 "LUN %d", __func__, retval, params->lun_id);
899 		mtx_lock(&softc->lock);
900 		be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
901 		mtx_unlock(&softc->lock);
902 		goto bailout_error;
903 	}
904 
905 	mtx_lock(&softc->lock);
906 	while ((be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) == 0) {
907 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlramrm", 0);
908 		if (retval == EINTR)
909 			break;
910 	}
911 	be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
912 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) {
913 		mtx_unlock(&softc->lock);
914 		free(be_lun, M_RAMDISK);
915 	} else {
916 		mtx_unlock(&softc->lock);
917 		return (EINTR);
918 	}
919 
920 	req->status = CTL_LUN_OK;
921 	return (retval);
922 
923 bailout_error:
924 	req->status = CTL_LUN_ERROR;
925 	return (0);
926 }
927 
928 static int
929 ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
930 			   struct ctl_lun_req *req)
931 {
932 	struct ctl_be_ramdisk_lun *be_lun;
933 	struct ctl_be_lun *cbe_lun;
934 	struct ctl_lun_create_params *params;
935 	const char *value;
936 	char tmpstr[32];
937 	uint64_t t;
938 	int retval;
939 
940 	retval = 0;
941 	params = &req->reqdata.create;
942 
943 	be_lun = malloc(sizeof(*be_lun), M_RAMDISK, M_ZERO | M_WAITOK);
944 	cbe_lun = &be_lun->cbe_lun;
945 	cbe_lun->options = nvlist_clone(req->args_nvl);
946 	be_lun->params = req->reqdata.create;
947 	be_lun->softc = softc;
948 
949 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
950 		cbe_lun->lun_type = params->device_type;
951 	else
952 		cbe_lun->lun_type = T_DIRECT;
953 	be_lun->flags = 0;
954 	cbe_lun->flags = 0;
955 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
956 	if (value != NULL) {
957 		if (strcmp(value, "primary") == 0)
958 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
959 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
960 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
961 
962 	be_lun->pblocksize = PAGE_SIZE;
963 	value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
964 	if (value != NULL) {
965 		ctl_expand_number(value, &t);
966 		be_lun->pblocksize = t;
967 	}
968 	if (be_lun->pblocksize < 512 || be_lun->pblocksize > 131072) {
969 		snprintf(req->error_str, sizeof(req->error_str),
970 			 "%s: unsupported pblocksize %u", __func__,
971 			 be_lun->pblocksize);
972 		goto bailout_error;
973 	}
974 
975 	if (cbe_lun->lun_type == T_DIRECT ||
976 	    cbe_lun->lun_type == T_CDROM) {
977 		if (params->blocksize_bytes != 0)
978 			cbe_lun->blocksize = params->blocksize_bytes;
979 		else if (cbe_lun->lun_type == T_CDROM)
980 			cbe_lun->blocksize = 2048;
981 		else
982 			cbe_lun->blocksize = 512;
983 		be_lun->pblockmul = be_lun->pblocksize / cbe_lun->blocksize;
984 		if (be_lun->pblockmul < 1 || !powerof2(be_lun->pblockmul)) {
985 			snprintf(req->error_str, sizeof(req->error_str),
986 				 "%s: pblocksize %u not exp2 of blocksize %u",
987 				 __func__,
988 				 be_lun->pblocksize, cbe_lun->blocksize);
989 			goto bailout_error;
990 		}
991 		if (params->lun_size_bytes < cbe_lun->blocksize) {
992 			snprintf(req->error_str, sizeof(req->error_str),
993 				 "%s: LUN size %ju < blocksize %u", __func__,
994 				 params->lun_size_bytes, cbe_lun->blocksize);
995 			goto bailout_error;
996 		}
997 		be_lun->size_blocks = params->lun_size_bytes / cbe_lun->blocksize;
998 		be_lun->size_bytes = be_lun->size_blocks * cbe_lun->blocksize;
999 		be_lun->indir = 0;
1000 		t = be_lun->size_bytes / be_lun->pblocksize;
1001 		while (t > 1) {
1002 			t /= PPP;
1003 			be_lun->indir++;
1004 		}
1005 		cbe_lun->maxlba = be_lun->size_blocks - 1;
1006 		cbe_lun->pblockexp = fls(be_lun->pblockmul) - 1;
1007 		cbe_lun->pblockoff = 0;
1008 		cbe_lun->ublockexp = cbe_lun->pblockexp;
1009 		cbe_lun->ublockoff = 0;
1010 		cbe_lun->atomicblock = be_lun->pblocksize;
1011 		cbe_lun->opttxferlen = SGPP * be_lun->pblocksize;
1012 		value = dnvlist_get_string(cbe_lun->options, "capacity", NULL);
1013 		if (value != NULL)
1014 			ctl_expand_number(value, &be_lun->cap_bytes);
1015 	} else {
1016 		be_lun->pblockmul = 1;
1017 		cbe_lun->pblockexp = 0;
1018 	}
1019 
1020 	/* Tell the user the blocksize we ended up using */
1021 	params->blocksize_bytes = cbe_lun->blocksize;
1022 	params->lun_size_bytes = be_lun->size_bytes;
1023 
1024 	value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
1025 	if (value == NULL || strcmp(value, "off") != 0)
1026 		cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
1027 	value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
1028 	if (value != NULL) {
1029 		if (strcmp(value, "on") == 0)
1030 			cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1031 	} else if (cbe_lun->lun_type != T_DIRECT)
1032 		cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1033 	cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1034 	value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
1035 	if (value != NULL && strcmp(value, "on") == 0)
1036 		cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
1037 	else if (value != NULL && strcmp(value, "read") == 0)
1038 		cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
1039 	else if (value != NULL && strcmp(value, "soft") == 0)
1040 		cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
1041 	else if (value != NULL && strcmp(value, "off") == 0)
1042 		cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1043 
1044 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
1045 		cbe_lun->req_lun_id = params->req_lun_id;
1046 		cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
1047 	} else
1048 		cbe_lun->req_lun_id = 0;
1049 
1050 	cbe_lun->lun_shutdown = ctl_backend_ramdisk_lun_shutdown;
1051 	cbe_lun->be = &ctl_be_ramdisk_driver;
1052 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
1053 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
1054 			 softc->num_luns);
1055 		strncpy((char *)cbe_lun->serial_num, tmpstr,
1056 			MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
1057 
1058 		/* Tell the user what we used for a serial number */
1059 		strncpy((char *)params->serial_num, tmpstr,
1060 			MIN(sizeof(params->serial_num), sizeof(tmpstr)));
1061 	} else {
1062 		strncpy((char *)cbe_lun->serial_num, params->serial_num,
1063 			MIN(sizeof(cbe_lun->serial_num),
1064 			    sizeof(params->serial_num)));
1065 	}
1066 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
1067 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
1068 		strncpy((char *)cbe_lun->device_id, tmpstr,
1069 			MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
1070 
1071 		/* Tell the user what we used for a device ID */
1072 		strncpy((char *)params->device_id, tmpstr,
1073 			MIN(sizeof(params->device_id), sizeof(tmpstr)));
1074 	} else {
1075 		strncpy((char *)cbe_lun->device_id, params->device_id,
1076 			MIN(sizeof(cbe_lun->device_id),
1077 			    sizeof(params->device_id)));
1078 	}
1079 
1080 	STAILQ_INIT(&be_lun->cont_queue);
1081 	sx_init(&be_lun->page_lock, "ctlram page");
1082 	if (be_lun->cap_bytes == 0) {
1083 		be_lun->indir = 0;
1084 		be_lun->pages = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK);
1085 	}
1086 	be_lun->zero_page = malloc(be_lun->pblocksize, M_RAMDISK,
1087 	    M_WAITOK|M_ZERO);
1088 	mtx_init(&be_lun->queue_lock, "ctlram queue", NULL, MTX_DEF);
1089 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_backend_ramdisk_worker,
1090 	    be_lun);
1091 
1092 	be_lun->io_taskqueue = taskqueue_create("ctlramtq", M_WAITOK,
1093 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
1094 	if (be_lun->io_taskqueue == NULL) {
1095 		snprintf(req->error_str, sizeof(req->error_str),
1096 			 "%s: Unable to create taskqueue", __func__);
1097 		goto bailout_error;
1098 	}
1099 
1100 	retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
1101 					 /*num threads*/1,
1102 					 /*priority*/PUSER,
1103 					 /*proc*/control_softc->ctl_proc,
1104 					 /*thread name*/"ramdisk");
1105 	if (retval != 0)
1106 		goto bailout_error;
1107 
1108 	retval = ctl_add_lun(&be_lun->cbe_lun);
1109 	if (retval != 0) {
1110 		snprintf(req->error_str, sizeof(req->error_str),
1111 			 "%s: ctl_add_lun() returned error %d, see dmesg for "
1112 			"details", __func__, retval);
1113 		retval = 0;
1114 		goto bailout_error;
1115 	}
1116 
1117 	mtx_lock(&softc->lock);
1118 	softc->num_luns++;
1119 	SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
1120 	mtx_unlock(&softc->lock);
1121 
1122 	params->req_lun_id = cbe_lun->lun_id;
1123 
1124 	req->status = CTL_LUN_OK;
1125 	return (retval);
1126 
1127 bailout_error:
1128 	req->status = CTL_LUN_ERROR;
1129 	if (be_lun != NULL) {
1130 		if (be_lun->io_taskqueue != NULL)
1131 			taskqueue_free(be_lun->io_taskqueue);
1132 		nvlist_destroy(cbe_lun->options);
1133 		free(be_lun->zero_page, M_RAMDISK);
1134 		ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1135 		sx_destroy(&be_lun->page_lock);
1136 		mtx_destroy(&be_lun->queue_lock);
1137 		free(be_lun, M_RAMDISK);
1138 	}
1139 	return (retval);
1140 }
1141 
1142 static int
1143 ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
1144 		       struct ctl_lun_req *req)
1145 {
1146 	struct ctl_be_ramdisk_lun *be_lun;
1147 	struct ctl_be_lun *cbe_lun;
1148 	struct ctl_lun_modify_params *params;
1149 	const char *value;
1150 	uint32_t blocksize;
1151 	int wasprim;
1152 
1153 	params = &req->reqdata.modify;
1154 	sx_xlock(&softc->modify_lock);
1155 	mtx_lock(&softc->lock);
1156 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
1157 		if (be_lun->cbe_lun.lun_id == params->lun_id)
1158 			break;
1159 	}
1160 	mtx_unlock(&softc->lock);
1161 	if (be_lun == NULL) {
1162 		snprintf(req->error_str, sizeof(req->error_str),
1163 			 "%s: LUN %u is not managed by the ramdisk backend",
1164 			 __func__, params->lun_id);
1165 		goto bailout_error;
1166 	}
1167 	cbe_lun = &be_lun->cbe_lun;
1168 
1169 	if (params->lun_size_bytes != 0)
1170 		be_lun->params.lun_size_bytes = params->lun_size_bytes;
1171 
1172 	if (req->args_nvl != NULL) {
1173 		nvlist_destroy(cbe_lun->options);
1174 		cbe_lun->options = nvlist_clone(req->args_nvl);
1175 	}
1176 
1177 	wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
1178 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
1179 	if (value != NULL) {
1180 		if (strcmp(value, "primary") == 0)
1181 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1182 		else
1183 			cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1184 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
1185 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1186 	else
1187 		cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1188 	if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
1189 		if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
1190 			ctl_lun_primary(cbe_lun);
1191 		else
1192 			ctl_lun_secondary(cbe_lun);
1193 	}
1194 
1195 	blocksize = be_lun->cbe_lun.blocksize;
1196 	if (be_lun->params.lun_size_bytes < blocksize) {
1197 		snprintf(req->error_str, sizeof(req->error_str),
1198 			"%s: LUN size %ju < blocksize %u", __func__,
1199 			be_lun->params.lun_size_bytes, blocksize);
1200 		goto bailout_error;
1201 	}
1202 	be_lun->size_blocks = be_lun->params.lun_size_bytes / blocksize;
1203 	be_lun->size_bytes = be_lun->size_blocks * blocksize;
1204 	be_lun->cbe_lun.maxlba = be_lun->size_blocks - 1;
1205 	ctl_lun_capacity_changed(&be_lun->cbe_lun);
1206 
1207 	/* Tell the user the exact size we ended up using */
1208 	params->lun_size_bytes = be_lun->size_bytes;
1209 
1210 	sx_xunlock(&softc->modify_lock);
1211 	req->status = CTL_LUN_OK;
1212 	return (0);
1213 
1214 bailout_error:
1215 	sx_xunlock(&softc->modify_lock);
1216 	req->status = CTL_LUN_ERROR;
1217 	return (0);
1218 }
1219 
1220 static void
1221 ctl_backend_ramdisk_lun_shutdown(struct ctl_be_lun *cbe_lun)
1222 {
1223 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
1224 	struct ctl_be_ramdisk_softc *softc = be_lun->softc;
1225 
1226 	taskqueue_drain_all(be_lun->io_taskqueue);
1227 	taskqueue_free(be_lun->io_taskqueue);
1228 	nvlist_destroy(be_lun->cbe_lun.options);
1229 	free(be_lun->zero_page, M_RAMDISK);
1230 	ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1231 	sx_destroy(&be_lun->page_lock);
1232 	mtx_destroy(&be_lun->queue_lock);
1233 
1234 	mtx_lock(&softc->lock);
1235 	be_lun->flags |= CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1236 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_WAITING)
1237 		wakeup(be_lun);
1238 	else
1239 		free(be_lun, M_RAMDISK);
1240 	mtx_unlock(&softc->lock);
1241 }
1242