xref: /freebsd/sys/cam/ctl/ctl_backend_ramdisk.c (revision 19261079)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 __FBSDID("$FreeBSD$");
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/condvar.h>
52 #include <sys/types.h>
53 #include <sys/limits.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/malloc.h>
57 #include <sys/sx.h>
58 #include <sys/taskqueue.h>
59 #include <sys/time.h>
60 #include <sys/queue.h>
61 #include <sys/conf.h>
62 #include <sys/ioccom.h>
63 #include <sys/module.h>
64 #include <sys/sysctl.h>
65 #include <sys/nv.h>
66 #include <sys/dnv.h>
67 
68 #include <cam/scsi/scsi_all.h>
69 #include <cam/scsi/scsi_da.h>
70 #include <cam/ctl/ctl_io.h>
71 #include <cam/ctl/ctl.h>
72 #include <cam/ctl/ctl_util.h>
73 #include <cam/ctl/ctl_backend.h>
74 #include <cam/ctl/ctl_debug.h>
75 #include <cam/ctl/ctl_ioctl.h>
76 #include <cam/ctl/ctl_ha.h>
77 #include <cam/ctl/ctl_private.h>
78 #include <cam/ctl/ctl_error.h>
79 
80 #define PRIV(io)	\
81     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
82 #define ARGS(io)	\
83     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
84 
85 #define	PPP	(PAGE_SIZE / sizeof(uint8_t **))
86 #ifdef __LP64__
87 #define	PPPS	(PAGE_SHIFT - 3)
88 #else
89 #define	PPPS	(PAGE_SHIFT - 2)
90 #endif
91 #define	SGPP	(PAGE_SIZE / sizeof(struct ctl_sg_entry))
92 
93 #define	P_UNMAPPED	NULL			/* Page is unmapped. */
94 #define	P_ANCHORED	((void *)(uintptr_t)1)	/* Page is anchored. */
95 
96 typedef enum {
97 	GP_READ,	/* Return data page or zero page. */
98 	GP_WRITE,	/* Return data page, try allocate if none. */
99 	GP_ANCHOR,	/* Return data page, try anchor if none. */
100 	GP_OTHER,	/* Return what present, do not allocate/anchor. */
101 } getpage_op_t;
102 
103 typedef enum {
104 	CTL_BE_RAMDISK_LUN_UNCONFIGURED	= 0x01,
105 	CTL_BE_RAMDISK_LUN_WAITING	= 0x04
106 } ctl_be_ramdisk_lun_flags;
107 
108 struct ctl_be_ramdisk_lun {
109 	struct ctl_be_lun	cbe_lun;	/* Must be first element. */
110 	struct ctl_lun_create_params params;
111 	int			indir;
112 	uint8_t			**pages;
113 	uint8_t			*zero_page;
114 	struct sx		page_lock;
115 	u_int			pblocksize;
116 	u_int			pblockmul;
117 	uint64_t		size_bytes;
118 	uint64_t		size_blocks;
119 	uint64_t		cap_bytes;
120 	uint64_t		cap_used;
121 	struct ctl_be_ramdisk_softc *softc;
122 	ctl_be_ramdisk_lun_flags flags;
123 	SLIST_ENTRY(ctl_be_ramdisk_lun) links;
124 	struct taskqueue	*io_taskqueue;
125 	struct task		io_task;
126 	STAILQ_HEAD(, ctl_io_hdr) cont_queue;
127 	struct mtx_padalign	queue_lock;
128 };
129 
130 struct ctl_be_ramdisk_softc {
131 	struct sx modify_lock;
132 	struct mtx lock;
133 	int num_luns;
134 	SLIST_HEAD(, ctl_be_ramdisk_lun) lun_list;
135 };
136 
137 static struct ctl_be_ramdisk_softc rd_softc;
138 extern struct ctl_softc *control_softc;
139 
140 static int ctl_backend_ramdisk_init(void);
141 static int ctl_backend_ramdisk_shutdown(void);
142 static int ctl_backend_ramdisk_move_done(union ctl_io *io, bool samethr);
143 static void ctl_backend_ramdisk_compare(union ctl_io *io);
144 static void ctl_backend_ramdisk_rw(union ctl_io *io);
145 static int ctl_backend_ramdisk_submit(union ctl_io *io);
146 static void ctl_backend_ramdisk_worker(void *context, int pending);
147 static int ctl_backend_ramdisk_config_read(union ctl_io *io);
148 static int ctl_backend_ramdisk_config_write(union ctl_io *io);
149 static uint64_t ctl_backend_ramdisk_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname);
150 static int ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd,
151 				     caddr_t addr, int flag, struct thread *td);
152 static int ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
153 				  struct ctl_lun_req *req);
154 static int ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
155 				      struct ctl_lun_req *req);
156 static int ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
157 				  struct ctl_lun_req *req);
158 static void ctl_backend_ramdisk_lun_shutdown(struct ctl_be_lun *cbe_lun);
159 
160 static struct ctl_backend_driver ctl_be_ramdisk_driver =
161 {
162 	.name = "ramdisk",
163 	.flags = CTL_BE_FLAG_HAS_CONFIG,
164 	.init = ctl_backend_ramdisk_init,
165 	.shutdown = ctl_backend_ramdisk_shutdown,
166 	.data_submit = ctl_backend_ramdisk_submit,
167 	.config_read = ctl_backend_ramdisk_config_read,
168 	.config_write = ctl_backend_ramdisk_config_write,
169 	.ioctl = ctl_backend_ramdisk_ioctl,
170 	.lun_attr = ctl_backend_ramdisk_lun_attr,
171 };
172 
173 MALLOC_DEFINE(M_RAMDISK, "ctlramdisk", "Memory used for CTL RAMdisk");
174 CTL_BACKEND_DECLARE(cbr, ctl_be_ramdisk_driver);
175 
176 static int
177 ctl_backend_ramdisk_init(void)
178 {
179 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
180 
181 	memset(softc, 0, sizeof(*softc));
182 	sx_init(&softc->modify_lock, "ctlrammod");
183 	mtx_init(&softc->lock, "ctlram", NULL, MTX_DEF);
184 	SLIST_INIT(&softc->lun_list);
185 	return (0);
186 }
187 
188 static int
189 ctl_backend_ramdisk_shutdown(void)
190 {
191 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
192 	struct ctl_be_ramdisk_lun *lun;
193 
194 	mtx_lock(&softc->lock);
195 	while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
196 		SLIST_REMOVE_HEAD(&softc->lun_list, links);
197 		softc->num_luns--;
198 		/*
199 		 * Drop our lock here.  Since ctl_remove_lun() can call
200 		 * back into us, this could potentially lead to a recursive
201 		 * lock of the same mutex, which would cause a hang.
202 		 */
203 		mtx_unlock(&softc->lock);
204 		ctl_remove_lun(&lun->cbe_lun);
205 		mtx_lock(&softc->lock);
206 	}
207 	mtx_unlock(&softc->lock);
208 	mtx_destroy(&softc->lock);
209 	sx_destroy(&softc->modify_lock);
210 	return (0);
211 }
212 
213 static uint8_t *
214 ctl_backend_ramdisk_getpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn,
215     getpage_op_t op)
216 {
217 	uint8_t **p, ***pp;
218 	off_t i;
219 	int s;
220 
221 	if (be_lun->cap_bytes == 0) {
222 		switch (op) {
223 		case GP_READ:
224 			return (be_lun->zero_page);
225 		case GP_WRITE:
226 			return ((uint8_t *)be_lun->pages);
227 		case GP_ANCHOR:
228 			return (P_ANCHORED);
229 		default:
230 			return (P_UNMAPPED);
231 		}
232 	}
233 	if (op == GP_WRITE || op == GP_ANCHOR) {
234 		sx_xlock(&be_lun->page_lock);
235 		pp = &be_lun->pages;
236 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
237 			if (*pp == NULL) {
238 				*pp = malloc(PAGE_SIZE, M_RAMDISK,
239 				    M_WAITOK|M_ZERO);
240 			}
241 			i = pn >> s;
242 			pp = (uint8_t ***)&(*pp)[i];
243 			pn -= i << s;
244 		}
245 		if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
246 			if (op == GP_WRITE) {
247 				*pp = malloc(be_lun->pblocksize, M_RAMDISK,
248 				    M_WAITOK|M_ZERO);
249 			} else
250 				*pp = P_ANCHORED;
251 			be_lun->cap_used += be_lun->pblocksize;
252 		} else if (*pp == P_ANCHORED && op == GP_WRITE) {
253 			*pp = malloc(be_lun->pblocksize, M_RAMDISK,
254 			    M_WAITOK|M_ZERO);
255 		}
256 		sx_xunlock(&be_lun->page_lock);
257 		return ((uint8_t *)*pp);
258 	} else {
259 		sx_slock(&be_lun->page_lock);
260 		p = be_lun->pages;
261 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
262 			if (p == NULL)
263 				break;
264 			i = pn >> s;
265 			p = (uint8_t **)p[i];
266 			pn -= i << s;
267 		}
268 		sx_sunlock(&be_lun->page_lock);
269 		if ((p == P_UNMAPPED || p == P_ANCHORED) && op == GP_READ)
270 			return (be_lun->zero_page);
271 		return ((uint8_t *)p);
272 	}
273 };
274 
275 static void
276 ctl_backend_ramdisk_unmappage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
277 {
278 	uint8_t ***pp;
279 	off_t i;
280 	int s;
281 
282 	if (be_lun->cap_bytes == 0)
283 		return;
284 	sx_xlock(&be_lun->page_lock);
285 	pp = &be_lun->pages;
286 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
287 		if (*pp == NULL)
288 			goto noindir;
289 		i = pn >> s;
290 		pp = (uint8_t ***)&(*pp)[i];
291 		pn -= i << s;
292 	}
293 	if (*pp == P_ANCHORED) {
294 		be_lun->cap_used -= be_lun->pblocksize;
295 		*pp = P_UNMAPPED;
296 	} else if (*pp != P_UNMAPPED) {
297 		free(*pp, M_RAMDISK);
298 		be_lun->cap_used -= be_lun->pblocksize;
299 		*pp = P_UNMAPPED;
300 	}
301 noindir:
302 	sx_xunlock(&be_lun->page_lock);
303 };
304 
305 static void
306 ctl_backend_ramdisk_anchorpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
307 {
308 	uint8_t ***pp;
309 	off_t i;
310 	int s;
311 
312 	if (be_lun->cap_bytes == 0)
313 		return;
314 	sx_xlock(&be_lun->page_lock);
315 	pp = &be_lun->pages;
316 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
317 		if (*pp == NULL)
318 			goto noindir;
319 		i = pn >> s;
320 		pp = (uint8_t ***)&(*pp)[i];
321 		pn -= i << s;
322 	}
323 	if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
324 		be_lun->cap_used += be_lun->pblocksize;
325 		*pp = P_ANCHORED;
326 	} else if (*pp != P_ANCHORED) {
327 		free(*pp, M_RAMDISK);
328 		*pp = P_ANCHORED;
329 	}
330 noindir:
331 	sx_xunlock(&be_lun->page_lock);
332 };
333 
334 static void
335 ctl_backend_ramdisk_freeallpages(uint8_t **p, int indir)
336 {
337 	int i;
338 
339 	if (p == NULL)
340 		return;
341 	if (indir == 0) {
342 		free(p, M_RAMDISK);
343 		return;
344 	}
345 	for (i = 0; i < PPP; i++) {
346 		if (p[i] == NULL)
347 			continue;
348 		ctl_backend_ramdisk_freeallpages((uint8_t **)p[i], indir - 1);
349 	}
350 	free(p, M_RAMDISK);
351 };
352 
353 static size_t
354 cmp(uint8_t *a, uint8_t *b, size_t size)
355 {
356 	size_t i;
357 
358 	for (i = 0; i < size; i++) {
359 		if (a[i] != b[i])
360 			break;
361 	}
362 	return (i);
363 }
364 
365 static int
366 ctl_backend_ramdisk_cmp(union ctl_io *io)
367 {
368 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
369 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
370 	uint8_t *page;
371 	uint8_t info[8];
372 	uint64_t lba;
373 	u_int lbaoff, lbas, res, off;
374 
375 	lbas = io->scsiio.kern_data_len / cbe_lun->blocksize;
376 	lba = ARGS(io)->lba + PRIV(io)->len - lbas;
377 	off = 0;
378 	for (; lbas > 0; lbas--, lba++) {
379 		page = ctl_backend_ramdisk_getpage(be_lun,
380 		    lba >> cbe_lun->pblockexp, GP_READ);
381 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
382 		page += lbaoff * cbe_lun->blocksize;
383 		res = cmp(io->scsiio.kern_data_ptr + off, page,
384 		    cbe_lun->blocksize);
385 		off += res;
386 		if (res < cbe_lun->blocksize)
387 			break;
388 	}
389 	if (lbas > 0) {
390 		off += io->scsiio.kern_rel_offset - io->scsiio.kern_data_len;
391 		scsi_u64to8b(off, info);
392 		ctl_set_sense(&io->scsiio, /*current_error*/ 1,
393 		    /*sense_key*/ SSD_KEY_MISCOMPARE,
394 		    /*asc*/ 0x1D, /*ascq*/ 0x00,
395 		    /*type*/ SSD_ELEM_INFO,
396 		    /*size*/ sizeof(info), /*data*/ &info,
397 		    /*type*/ SSD_ELEM_NONE);
398 		return (1);
399 	}
400 	return (0);
401 }
402 
403 static int
404 ctl_backend_ramdisk_move_done(union ctl_io *io, bool samethr)
405 {
406 	struct ctl_be_ramdisk_lun *be_lun =
407 	    (struct ctl_be_ramdisk_lun *)CTL_BACKEND_LUN(io);
408 
409 	CTL_DEBUG_PRINT(("ctl_backend_ramdisk_move_done\n"));
410 	if (io->scsiio.kern_sg_entries > 0)
411 		free(io->scsiio.kern_data_ptr, M_RAMDISK);
412 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
413 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
414 	    (io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE) {
415 		if (ARGS(io)->flags & CTL_LLF_COMPARE) {
416 			/* We have data block ready for comparison. */
417 			if (ctl_backend_ramdisk_cmp(io))
418 				goto done;
419 		}
420 		if (ARGS(io)->len > PRIV(io)->len) {
421 			mtx_lock(&be_lun->queue_lock);
422 			STAILQ_INSERT_TAIL(&be_lun->cont_queue,
423 			    &io->io_hdr, links);
424 			mtx_unlock(&be_lun->queue_lock);
425 			taskqueue_enqueue(be_lun->io_taskqueue,
426 			    &be_lun->io_task);
427 			return (0);
428 		}
429 		ctl_set_success(&io->scsiio);
430 	}
431 done:
432 	ctl_data_submit_done(io);
433 	return(0);
434 }
435 
436 static void
437 ctl_backend_ramdisk_compare(union ctl_io *io)
438 {
439 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
440 	u_int lbas, len;
441 
442 	lbas = ARGS(io)->len - PRIV(io)->len;
443 	lbas = MIN(lbas, 131072 / cbe_lun->blocksize);
444 	len = lbas * cbe_lun->blocksize;
445 
446 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
447 	io->scsiio.kern_data_ptr = malloc(len, M_RAMDISK, M_WAITOK);
448 	io->scsiio.kern_data_len = len;
449 	io->scsiio.kern_sg_entries = 0;
450 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
451 	PRIV(io)->len += lbas;
452 	ctl_datamove(io);
453 }
454 
455 static void
456 ctl_backend_ramdisk_rw(union ctl_io *io)
457 {
458 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
459 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
460 	struct ctl_sg_entry *sg_entries;
461 	uint8_t *page;
462 	uint64_t lba;
463 	u_int i, len, lbaoff, lbas, sgs, off;
464 	getpage_op_t op;
465 
466 	lba = ARGS(io)->lba + PRIV(io)->len;
467 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
468 	lbas = ARGS(io)->len - PRIV(io)->len;
469 	lbas = MIN(lbas, (SGPP << cbe_lun->pblockexp) - lbaoff);
470 	sgs = (lbas + lbaoff + be_lun->pblockmul - 1) >> cbe_lun->pblockexp;
471 	off = lbaoff * cbe_lun->blocksize;
472 	op = (ARGS(io)->flags & CTL_LLF_WRITE) ? GP_WRITE : GP_READ;
473 	if (sgs > 1) {
474 		io->scsiio.kern_data_ptr = malloc(sizeof(struct ctl_sg_entry) *
475 		    sgs, M_RAMDISK, M_WAITOK);
476 		sg_entries = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
477 		len = lbas * cbe_lun->blocksize;
478 		for (i = 0; i < sgs; i++) {
479 			page = ctl_backend_ramdisk_getpage(be_lun,
480 			    (lba >> cbe_lun->pblockexp) + i, op);
481 			if (page == P_UNMAPPED || page == P_ANCHORED) {
482 				free(io->scsiio.kern_data_ptr, M_RAMDISK);
483 nospc:
484 				ctl_set_space_alloc_fail(&io->scsiio);
485 				ctl_data_submit_done(io);
486 				return;
487 			}
488 			sg_entries[i].addr = page + off;
489 			sg_entries[i].len = MIN(len, be_lun->pblocksize - off);
490 			len -= sg_entries[i].len;
491 			off = 0;
492 		}
493 	} else {
494 		page = ctl_backend_ramdisk_getpage(be_lun,
495 		    lba >> cbe_lun->pblockexp, op);
496 		if (page == P_UNMAPPED || page == P_ANCHORED)
497 			goto nospc;
498 		sgs = 0;
499 		io->scsiio.kern_data_ptr = page + off;
500 	}
501 
502 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
503 	io->scsiio.kern_data_len = lbas * cbe_lun->blocksize;
504 	io->scsiio.kern_sg_entries = sgs;
505 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
506 	PRIV(io)->len += lbas;
507 	if ((ARGS(io)->flags & CTL_LLF_READ) &&
508 	    ARGS(io)->len <= PRIV(io)->len) {
509 		ctl_set_success(&io->scsiio);
510 		if (cbe_lun->serseq >= CTL_LUN_SERSEQ_SOFT)
511 			ctl_serseq_done(io);
512 	}
513 	ctl_datamove(io);
514 }
515 
516 static int
517 ctl_backend_ramdisk_submit(union ctl_io *io)
518 {
519 	struct ctl_lba_len_flags *lbalen = ARGS(io);
520 
521 	if (lbalen->flags & CTL_LLF_VERIFY) {
522 		ctl_set_success(&io->scsiio);
523 		ctl_data_submit_done(io);
524 		return (CTL_RETVAL_COMPLETE);
525 	}
526 	PRIV(io)->len = 0;
527 	if (lbalen->flags & CTL_LLF_COMPARE)
528 		ctl_backend_ramdisk_compare(io);
529 	else
530 		ctl_backend_ramdisk_rw(io);
531 	return (CTL_RETVAL_COMPLETE);
532 }
533 
534 static void
535 ctl_backend_ramdisk_worker(void *context, int pending)
536 {
537 	struct ctl_be_ramdisk_lun *be_lun;
538 	union ctl_io *io;
539 
540 	be_lun = (struct ctl_be_ramdisk_lun *)context;
541 	mtx_lock(&be_lun->queue_lock);
542 	for (;;) {
543 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->cont_queue);
544 		if (io != NULL) {
545 			STAILQ_REMOVE_HEAD(&be_lun->cont_queue, links);
546 			mtx_unlock(&be_lun->queue_lock);
547 			if (ARGS(io)->flags & CTL_LLF_COMPARE)
548 				ctl_backend_ramdisk_compare(io);
549 			else
550 				ctl_backend_ramdisk_rw(io);
551 			mtx_lock(&be_lun->queue_lock);
552 			continue;
553 		}
554 
555 		/*
556 		 * If we get here, there is no work left in the queues, so
557 		 * just break out and let the task queue go to sleep.
558 		 */
559 		break;
560 	}
561 	mtx_unlock(&be_lun->queue_lock);
562 }
563 
564 static int
565 ctl_backend_ramdisk_gls(union ctl_io *io)
566 {
567 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
568 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
569 	struct scsi_get_lba_status_data *data;
570 	uint8_t *page;
571 	u_int lbaoff;
572 
573 	data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
574 	scsi_u64to8b(ARGS(io)->lba, data->descr[0].addr);
575 	lbaoff = ARGS(io)->lba & ~(UINT_MAX << cbe_lun->pblockexp);
576 	scsi_ulto4b(be_lun->pblockmul - lbaoff, data->descr[0].length);
577 	page = ctl_backend_ramdisk_getpage(be_lun,
578 	    ARGS(io)->lba >> cbe_lun->pblockexp, GP_OTHER);
579 	if (page == P_UNMAPPED)
580 		data->descr[0].status = 1;
581 	else if (page == P_ANCHORED)
582 		data->descr[0].status = 2;
583 	else
584 		data->descr[0].status = 0;
585 	ctl_config_read_done(io);
586 	return (CTL_RETVAL_COMPLETE);
587 }
588 
589 static int
590 ctl_backend_ramdisk_config_read(union ctl_io *io)
591 {
592 	int retval = 0;
593 
594 	switch (io->scsiio.cdb[0]) {
595 	case SERVICE_ACTION_IN:
596 		if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
597 			retval = ctl_backend_ramdisk_gls(io);
598 			break;
599 		}
600 		ctl_set_invalid_field(&io->scsiio,
601 				      /*sks_valid*/ 1,
602 				      /*command*/ 1,
603 				      /*field*/ 1,
604 				      /*bit_valid*/ 1,
605 				      /*bit*/ 4);
606 		ctl_config_read_done(io);
607 		retval = CTL_RETVAL_COMPLETE;
608 		break;
609 	default:
610 		ctl_set_invalid_opcode(&io->scsiio);
611 		ctl_config_read_done(io);
612 		retval = CTL_RETVAL_COMPLETE;
613 		break;
614 	}
615 	return (retval);
616 }
617 
618 static void
619 ctl_backend_ramdisk_delete(struct ctl_be_lun *cbe_lun, off_t lba, off_t len,
620     int anchor)
621 {
622 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
623 	uint8_t *page;
624 	uint64_t p, lp;
625 	u_int lbaoff;
626 	getpage_op_t op = anchor ? GP_ANCHOR : GP_OTHER;
627 
628 	/* Partially zero first partial page. */
629 	p = lba >> cbe_lun->pblockexp;
630 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
631 	if (lbaoff != 0) {
632 		page = ctl_backend_ramdisk_getpage(be_lun, p, op);
633 		if (page != P_UNMAPPED && page != P_ANCHORED) {
634 			memset(page + lbaoff * cbe_lun->blocksize, 0,
635 			    min(len, be_lun->pblockmul - lbaoff) *
636 			    cbe_lun->blocksize);
637 		}
638 		p++;
639 	}
640 
641 	/* Partially zero last partial page. */
642 	lp = (lba + len) >> cbe_lun->pblockexp;
643 	lbaoff = (lba + len) & ~(UINT_MAX << cbe_lun->pblockexp);
644 	if (p <= lp && lbaoff != 0) {
645 		page = ctl_backend_ramdisk_getpage(be_lun, lp, op);
646 		if (page != P_UNMAPPED && page != P_ANCHORED)
647 			memset(page, 0, lbaoff * cbe_lun->blocksize);
648 	}
649 
650 	/* Delete remaining full pages. */
651 	if (anchor) {
652 		for (; p < lp; p++)
653 			ctl_backend_ramdisk_anchorpage(be_lun, p);
654 	} else {
655 		for (; p < lp; p++)
656 			ctl_backend_ramdisk_unmappage(be_lun, p);
657 	}
658 }
659 
660 static void
661 ctl_backend_ramdisk_ws(union ctl_io *io)
662 {
663 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
664 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
665 	struct ctl_lba_len_flags *lbalen = ARGS(io);
666 	uint8_t *page;
667 	uint64_t lba;
668 	u_int lbaoff, lbas;
669 
670 	if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB)) {
671 		ctl_set_invalid_field(&io->scsiio,
672 				      /*sks_valid*/ 1,
673 				      /*command*/ 1,
674 				      /*field*/ 1,
675 				      /*bit_valid*/ 0,
676 				      /*bit*/ 0);
677 		ctl_config_write_done(io);
678 		return;
679 	}
680 	if (lbalen->flags & SWS_UNMAP) {
681 		ctl_backend_ramdisk_delete(cbe_lun, lbalen->lba, lbalen->len,
682 		    (lbalen->flags & SWS_ANCHOR) != 0);
683 		ctl_set_success(&io->scsiio);
684 		ctl_config_write_done(io);
685 		return;
686 	}
687 
688 	for (lba = lbalen->lba, lbas = lbalen->len; lbas > 0; lba++, lbas--) {
689 		page = ctl_backend_ramdisk_getpage(be_lun,
690 		    lba >> cbe_lun->pblockexp, GP_WRITE);
691 		if (page == P_UNMAPPED || page == P_ANCHORED) {
692 			ctl_set_space_alloc_fail(&io->scsiio);
693 			ctl_data_submit_done(io);
694 			return;
695 		}
696 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
697 		page += lbaoff * cbe_lun->blocksize;
698 		if (lbalen->flags & SWS_NDOB) {
699 			memset(page, 0, cbe_lun->blocksize);
700 		} else {
701 			memcpy(page, io->scsiio.kern_data_ptr,
702 			    cbe_lun->blocksize);
703 		}
704 		if (lbalen->flags & SWS_LBDATA)
705 			scsi_ulto4b(lba, page);
706 	}
707 	ctl_set_success(&io->scsiio);
708 	ctl_config_write_done(io);
709 }
710 
711 static void
712 ctl_backend_ramdisk_unmap(union ctl_io *io)
713 {
714 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
715 	struct ctl_ptr_len_flags *ptrlen = (struct ctl_ptr_len_flags *)ARGS(io);
716 	struct scsi_unmap_desc *buf, *end;
717 
718 	if ((ptrlen->flags & ~SU_ANCHOR) != 0) {
719 		ctl_set_invalid_field(&io->scsiio,
720 				      /*sks_valid*/ 0,
721 				      /*command*/ 0,
722 				      /*field*/ 0,
723 				      /*bit_valid*/ 0,
724 				      /*bit*/ 0);
725 		ctl_config_write_done(io);
726 		return;
727 	}
728 
729 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
730 	end = buf + ptrlen->len / sizeof(*buf);
731 	for (; buf < end; buf++) {
732 		ctl_backend_ramdisk_delete(cbe_lun,
733 		    scsi_8btou64(buf->lba), scsi_4btoul(buf->length),
734 		    (ptrlen->flags & SU_ANCHOR) != 0);
735 	}
736 
737 	ctl_set_success(&io->scsiio);
738 	ctl_config_write_done(io);
739 }
740 
741 static int
742 ctl_backend_ramdisk_config_write(union ctl_io *io)
743 {
744 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
745 	int retval = 0;
746 
747 	switch (io->scsiio.cdb[0]) {
748 	case SYNCHRONIZE_CACHE:
749 	case SYNCHRONIZE_CACHE_16:
750 		/* We have no cache to flush. */
751 		ctl_set_success(&io->scsiio);
752 		ctl_config_write_done(io);
753 		break;
754 	case START_STOP_UNIT: {
755 		struct scsi_start_stop_unit *cdb;
756 
757 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
758 		if ((cdb->how & SSS_PC_MASK) != 0) {
759 			ctl_set_success(&io->scsiio);
760 			ctl_config_write_done(io);
761 			break;
762 		}
763 		if (cdb->how & SSS_START) {
764 			if (cdb->how & SSS_LOEJ)
765 				ctl_lun_has_media(cbe_lun);
766 			ctl_start_lun(cbe_lun);
767 		} else {
768 			ctl_stop_lun(cbe_lun);
769 			if (cdb->how & SSS_LOEJ)
770 				ctl_lun_ejected(cbe_lun);
771 		}
772 		ctl_set_success(&io->scsiio);
773 		ctl_config_write_done(io);
774 		break;
775 	}
776 	case PREVENT_ALLOW:
777 		ctl_set_success(&io->scsiio);
778 		ctl_config_write_done(io);
779 		break;
780 	case WRITE_SAME_10:
781 	case WRITE_SAME_16:
782 		ctl_backend_ramdisk_ws(io);
783 		break;
784 	case UNMAP:
785 		ctl_backend_ramdisk_unmap(io);
786 		break;
787 	default:
788 		ctl_set_invalid_opcode(&io->scsiio);
789 		ctl_config_write_done(io);
790 		retval = CTL_RETVAL_COMPLETE;
791 		break;
792 	}
793 
794 	return (retval);
795 }
796 
797 static uint64_t
798 ctl_backend_ramdisk_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
799 {
800 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
801 	uint64_t		val;
802 
803 	val = UINT64_MAX;
804 	if (be_lun->cap_bytes == 0)
805 		return (val);
806 	sx_slock(&be_lun->page_lock);
807 	if (strcmp(attrname, "blocksused") == 0) {
808 		val = be_lun->cap_used / be_lun->cbe_lun.blocksize;
809 	} else if (strcmp(attrname, "blocksavail") == 0) {
810 		val = (be_lun->cap_bytes - be_lun->cap_used) /
811 		    be_lun->cbe_lun.blocksize;
812 	}
813 	sx_sunlock(&be_lun->page_lock);
814 	return (val);
815 }
816 
817 static int
818 ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
819 			  int flag, struct thread *td)
820 {
821 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
822 	struct ctl_lun_req *lun_req;
823 	int retval;
824 
825 	retval = 0;
826 	switch (cmd) {
827 	case CTL_LUN_REQ:
828 		lun_req = (struct ctl_lun_req *)addr;
829 		switch (lun_req->reqtype) {
830 		case CTL_LUNREQ_CREATE:
831 			retval = ctl_backend_ramdisk_create(softc, lun_req);
832 			break;
833 		case CTL_LUNREQ_RM:
834 			retval = ctl_backend_ramdisk_rm(softc, lun_req);
835 			break;
836 		case CTL_LUNREQ_MODIFY:
837 			retval = ctl_backend_ramdisk_modify(softc, lun_req);
838 			break;
839 		default:
840 			lun_req->status = CTL_LUN_ERROR;
841 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
842 				 "%s: invalid LUN request type %d", __func__,
843 				 lun_req->reqtype);
844 			break;
845 		}
846 		break;
847 	default:
848 		retval = ENOTTY;
849 		break;
850 	}
851 
852 	return (retval);
853 }
854 
855 static int
856 ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
857 		       struct ctl_lun_req *req)
858 {
859 	struct ctl_be_ramdisk_lun *be_lun;
860 	struct ctl_lun_rm_params *params;
861 	int retval;
862 
863 	params = &req->reqdata.rm;
864 	sx_xlock(&softc->modify_lock);
865 	mtx_lock(&softc->lock);
866 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
867 		if (be_lun->cbe_lun.lun_id == params->lun_id) {
868 			SLIST_REMOVE(&softc->lun_list, be_lun,
869 			    ctl_be_ramdisk_lun, links);
870 			softc->num_luns--;
871 			break;
872 		}
873 	}
874 	mtx_unlock(&softc->lock);
875 	sx_xunlock(&softc->modify_lock);
876 	if (be_lun == NULL) {
877 		snprintf(req->error_str, sizeof(req->error_str),
878 			 "%s: LUN %u is not managed by the ramdisk backend",
879 			 __func__, params->lun_id);
880 		goto bailout_error;
881 	}
882 
883 	/*
884 	 * Set the waiting flag before we invalidate the LUN.  Our shutdown
885 	 * routine can be called any time after we invalidate the LUN,
886 	 * and can be called from our context.
887 	 *
888 	 * This tells the shutdown routine that we're waiting, or we're
889 	 * going to wait for the shutdown to happen.
890 	 */
891 	mtx_lock(&softc->lock);
892 	be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING;
893 	mtx_unlock(&softc->lock);
894 
895 	retval = ctl_remove_lun(&be_lun->cbe_lun);
896 	if (retval != 0) {
897 		snprintf(req->error_str, sizeof(req->error_str),
898 			 "%s: error %d returned from ctl_remove_lun() for "
899 			 "LUN %d", __func__, retval, params->lun_id);
900 		mtx_lock(&softc->lock);
901 		be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
902 		mtx_unlock(&softc->lock);
903 		goto bailout_error;
904 	}
905 
906 	mtx_lock(&softc->lock);
907 	while ((be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) == 0) {
908 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlramrm", 0);
909 		if (retval == EINTR)
910 			break;
911 	}
912 	be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
913 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) {
914 		mtx_unlock(&softc->lock);
915 		free(be_lun, M_RAMDISK);
916 	} else {
917 		mtx_unlock(&softc->lock);
918 		return (EINTR);
919 	}
920 
921 	req->status = CTL_LUN_OK;
922 	return (retval);
923 
924 bailout_error:
925 	req->status = CTL_LUN_ERROR;
926 	return (0);
927 }
928 
929 static int
930 ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
931 			   struct ctl_lun_req *req)
932 {
933 	struct ctl_be_ramdisk_lun *be_lun;
934 	struct ctl_be_lun *cbe_lun;
935 	struct ctl_lun_create_params *params;
936 	const char *value;
937 	char tmpstr[32];
938 	uint64_t t;
939 	int retval;
940 
941 	retval = 0;
942 	params = &req->reqdata.create;
943 
944 	be_lun = malloc(sizeof(*be_lun), M_RAMDISK, M_ZERO | M_WAITOK);
945 	cbe_lun = &be_lun->cbe_lun;
946 	cbe_lun->options = nvlist_clone(req->args_nvl);
947 	be_lun->params = req->reqdata.create;
948 	be_lun->softc = softc;
949 
950 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
951 		cbe_lun->lun_type = params->device_type;
952 	else
953 		cbe_lun->lun_type = T_DIRECT;
954 	be_lun->flags = 0;
955 	cbe_lun->flags = 0;
956 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
957 	if (value != NULL) {
958 		if (strcmp(value, "primary") == 0)
959 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
960 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
961 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
962 
963 	be_lun->pblocksize = PAGE_SIZE;
964 	value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
965 	if (value != NULL) {
966 		ctl_expand_number(value, &t);
967 		be_lun->pblocksize = t;
968 	}
969 	if (be_lun->pblocksize < 512 || be_lun->pblocksize > 131072) {
970 		snprintf(req->error_str, sizeof(req->error_str),
971 			 "%s: unsupported pblocksize %u", __func__,
972 			 be_lun->pblocksize);
973 		goto bailout_error;
974 	}
975 
976 	if (cbe_lun->lun_type == T_DIRECT ||
977 	    cbe_lun->lun_type == T_CDROM) {
978 		if (params->blocksize_bytes != 0)
979 			cbe_lun->blocksize = params->blocksize_bytes;
980 		else if (cbe_lun->lun_type == T_CDROM)
981 			cbe_lun->blocksize = 2048;
982 		else
983 			cbe_lun->blocksize = 512;
984 		be_lun->pblockmul = be_lun->pblocksize / cbe_lun->blocksize;
985 		if (be_lun->pblockmul < 1 || !powerof2(be_lun->pblockmul)) {
986 			snprintf(req->error_str, sizeof(req->error_str),
987 				 "%s: pblocksize %u not exp2 of blocksize %u",
988 				 __func__,
989 				 be_lun->pblocksize, cbe_lun->blocksize);
990 			goto bailout_error;
991 		}
992 		if (params->lun_size_bytes < cbe_lun->blocksize) {
993 			snprintf(req->error_str, sizeof(req->error_str),
994 				 "%s: LUN size %ju < blocksize %u", __func__,
995 				 params->lun_size_bytes, cbe_lun->blocksize);
996 			goto bailout_error;
997 		}
998 		be_lun->size_blocks = params->lun_size_bytes / cbe_lun->blocksize;
999 		be_lun->size_bytes = be_lun->size_blocks * cbe_lun->blocksize;
1000 		be_lun->indir = 0;
1001 		t = be_lun->size_bytes / be_lun->pblocksize;
1002 		while (t > 1) {
1003 			t /= PPP;
1004 			be_lun->indir++;
1005 		}
1006 		cbe_lun->maxlba = be_lun->size_blocks - 1;
1007 		cbe_lun->pblockexp = fls(be_lun->pblockmul) - 1;
1008 		cbe_lun->pblockoff = 0;
1009 		cbe_lun->ublockexp = cbe_lun->pblockexp;
1010 		cbe_lun->ublockoff = 0;
1011 		cbe_lun->atomicblock = be_lun->pblocksize;
1012 		cbe_lun->opttxferlen = SGPP * be_lun->pblocksize;
1013 		value = dnvlist_get_string(cbe_lun->options, "capacity", NULL);
1014 		if (value != NULL)
1015 			ctl_expand_number(value, &be_lun->cap_bytes);
1016 	} else {
1017 		be_lun->pblockmul = 1;
1018 		cbe_lun->pblockexp = 0;
1019 	}
1020 
1021 	/* Tell the user the blocksize we ended up using */
1022 	params->blocksize_bytes = cbe_lun->blocksize;
1023 	params->lun_size_bytes = be_lun->size_bytes;
1024 
1025 	value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
1026 	if (value == NULL || strcmp(value, "off") != 0)
1027 		cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
1028 	value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
1029 	if (value != NULL) {
1030 		if (strcmp(value, "on") == 0)
1031 			cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1032 	} else if (cbe_lun->lun_type != T_DIRECT)
1033 		cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1034 	cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1035 	value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
1036 	if (value != NULL && strcmp(value, "on") == 0)
1037 		cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
1038 	else if (value != NULL && strcmp(value, "read") == 0)
1039 		cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
1040 	else if (value != NULL && strcmp(value, "soft") == 0)
1041 		cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
1042 	else if (value != NULL && strcmp(value, "off") == 0)
1043 		cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1044 
1045 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
1046 		cbe_lun->req_lun_id = params->req_lun_id;
1047 		cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
1048 	} else
1049 		cbe_lun->req_lun_id = 0;
1050 
1051 	cbe_lun->lun_shutdown = ctl_backend_ramdisk_lun_shutdown;
1052 	cbe_lun->be = &ctl_be_ramdisk_driver;
1053 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
1054 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
1055 			 softc->num_luns);
1056 		strncpy((char *)cbe_lun->serial_num, tmpstr,
1057 			MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
1058 
1059 		/* Tell the user what we used for a serial number */
1060 		strncpy((char *)params->serial_num, tmpstr,
1061 			MIN(sizeof(params->serial_num), sizeof(tmpstr)));
1062 	} else {
1063 		strncpy((char *)cbe_lun->serial_num, params->serial_num,
1064 			MIN(sizeof(cbe_lun->serial_num),
1065 			    sizeof(params->serial_num)));
1066 	}
1067 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
1068 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
1069 		strncpy((char *)cbe_lun->device_id, tmpstr,
1070 			MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
1071 
1072 		/* Tell the user what we used for a device ID */
1073 		strncpy((char *)params->device_id, tmpstr,
1074 			MIN(sizeof(params->device_id), sizeof(tmpstr)));
1075 	} else {
1076 		strncpy((char *)cbe_lun->device_id, params->device_id,
1077 			MIN(sizeof(cbe_lun->device_id),
1078 			    sizeof(params->device_id)));
1079 	}
1080 
1081 	STAILQ_INIT(&be_lun->cont_queue);
1082 	sx_init(&be_lun->page_lock, "ctlram page");
1083 	if (be_lun->cap_bytes == 0) {
1084 		be_lun->indir = 0;
1085 		be_lun->pages = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK);
1086 	}
1087 	be_lun->zero_page = malloc(be_lun->pblocksize, M_RAMDISK,
1088 	    M_WAITOK|M_ZERO);
1089 	mtx_init(&be_lun->queue_lock, "ctlram queue", NULL, MTX_DEF);
1090 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_backend_ramdisk_worker,
1091 	    be_lun);
1092 
1093 	be_lun->io_taskqueue = taskqueue_create("ctlramtq", M_WAITOK,
1094 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
1095 	if (be_lun->io_taskqueue == NULL) {
1096 		snprintf(req->error_str, sizeof(req->error_str),
1097 			 "%s: Unable to create taskqueue", __func__);
1098 		goto bailout_error;
1099 	}
1100 
1101 	retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
1102 					 /*num threads*/1,
1103 					 /*priority*/PUSER,
1104 					 /*proc*/control_softc->ctl_proc,
1105 					 /*thread name*/"ramdisk");
1106 	if (retval != 0)
1107 		goto bailout_error;
1108 
1109 	retval = ctl_add_lun(&be_lun->cbe_lun);
1110 	if (retval != 0) {
1111 		snprintf(req->error_str, sizeof(req->error_str),
1112 			 "%s: ctl_add_lun() returned error %d, see dmesg for "
1113 			"details", __func__, retval);
1114 		retval = 0;
1115 		goto bailout_error;
1116 	}
1117 
1118 	mtx_lock(&softc->lock);
1119 	softc->num_luns++;
1120 	SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
1121 	mtx_unlock(&softc->lock);
1122 
1123 	params->req_lun_id = cbe_lun->lun_id;
1124 
1125 	req->status = CTL_LUN_OK;
1126 	return (retval);
1127 
1128 bailout_error:
1129 	req->status = CTL_LUN_ERROR;
1130 	if (be_lun != NULL) {
1131 		if (be_lun->io_taskqueue != NULL)
1132 			taskqueue_free(be_lun->io_taskqueue);
1133 		nvlist_destroy(cbe_lun->options);
1134 		free(be_lun->zero_page, M_RAMDISK);
1135 		ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1136 		sx_destroy(&be_lun->page_lock);
1137 		mtx_destroy(&be_lun->queue_lock);
1138 		free(be_lun, M_RAMDISK);
1139 	}
1140 	return (retval);
1141 }
1142 
1143 static int
1144 ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
1145 		       struct ctl_lun_req *req)
1146 {
1147 	struct ctl_be_ramdisk_lun *be_lun;
1148 	struct ctl_be_lun *cbe_lun;
1149 	struct ctl_lun_modify_params *params;
1150 	const char *value;
1151 	uint32_t blocksize;
1152 	int wasprim;
1153 
1154 	params = &req->reqdata.modify;
1155 	sx_xlock(&softc->modify_lock);
1156 	mtx_lock(&softc->lock);
1157 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
1158 		if (be_lun->cbe_lun.lun_id == params->lun_id)
1159 			break;
1160 	}
1161 	mtx_unlock(&softc->lock);
1162 	if (be_lun == NULL) {
1163 		snprintf(req->error_str, sizeof(req->error_str),
1164 			 "%s: LUN %u is not managed by the ramdisk backend",
1165 			 __func__, params->lun_id);
1166 		goto bailout_error;
1167 	}
1168 	cbe_lun = &be_lun->cbe_lun;
1169 
1170 	if (params->lun_size_bytes != 0)
1171 		be_lun->params.lun_size_bytes = params->lun_size_bytes;
1172 
1173 	if (req->args_nvl != NULL) {
1174 		nvlist_destroy(cbe_lun->options);
1175 		cbe_lun->options = nvlist_clone(req->args_nvl);
1176 	}
1177 
1178 	wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
1179 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
1180 	if (value != NULL) {
1181 		if (strcmp(value, "primary") == 0)
1182 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1183 		else
1184 			cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1185 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
1186 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1187 	else
1188 		cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1189 	if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
1190 		if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
1191 			ctl_lun_primary(cbe_lun);
1192 		else
1193 			ctl_lun_secondary(cbe_lun);
1194 	}
1195 
1196 	blocksize = be_lun->cbe_lun.blocksize;
1197 	if (be_lun->params.lun_size_bytes < blocksize) {
1198 		snprintf(req->error_str, sizeof(req->error_str),
1199 			"%s: LUN size %ju < blocksize %u", __func__,
1200 			be_lun->params.lun_size_bytes, blocksize);
1201 		goto bailout_error;
1202 	}
1203 	be_lun->size_blocks = be_lun->params.lun_size_bytes / blocksize;
1204 	be_lun->size_bytes = be_lun->size_blocks * blocksize;
1205 	be_lun->cbe_lun.maxlba = be_lun->size_blocks - 1;
1206 	ctl_lun_capacity_changed(&be_lun->cbe_lun);
1207 
1208 	/* Tell the user the exact size we ended up using */
1209 	params->lun_size_bytes = be_lun->size_bytes;
1210 
1211 	sx_xunlock(&softc->modify_lock);
1212 	req->status = CTL_LUN_OK;
1213 	return (0);
1214 
1215 bailout_error:
1216 	sx_xunlock(&softc->modify_lock);
1217 	req->status = CTL_LUN_ERROR;
1218 	return (0);
1219 }
1220 
1221 static void
1222 ctl_backend_ramdisk_lun_shutdown(struct ctl_be_lun *cbe_lun)
1223 {
1224 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
1225 	struct ctl_be_ramdisk_softc *softc = be_lun->softc;
1226 
1227 	taskqueue_drain_all(be_lun->io_taskqueue);
1228 	taskqueue_free(be_lun->io_taskqueue);
1229 	nvlist_destroy(be_lun->cbe_lun.options);
1230 	free(be_lun->zero_page, M_RAMDISK);
1231 	ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1232 	sx_destroy(&be_lun->page_lock);
1233 	mtx_destroy(&be_lun->queue_lock);
1234 
1235 	mtx_lock(&softc->lock);
1236 	be_lun->flags |= CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1237 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_WAITING)
1238 		wakeup(be_lun);
1239 	else
1240 		free(be_lun, M_RAMDISK);
1241 	mtx_unlock(&softc->lock);
1242 }
1243