xref: /freebsd/sys/cam/ctl/ctl_backend_ramdisk.c (revision e32fecd0c2c3ee37c47ee100f169e7eb0282a873)
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 	free(io->scsiio.kern_data_ptr, M_RAMDISK);
390 	if (lbas > 0) {
391 		off += io->scsiio.kern_rel_offset - io->scsiio.kern_data_len;
392 		scsi_u64to8b(off, info);
393 		ctl_set_sense(&io->scsiio, /*current_error*/ 1,
394 		    /*sense_key*/ SSD_KEY_MISCOMPARE,
395 		    /*asc*/ 0x1D, /*ascq*/ 0x00,
396 		    /*type*/ SSD_ELEM_INFO,
397 		    /*size*/ sizeof(info), /*data*/ &info,
398 		    /*type*/ SSD_ELEM_NONE);
399 		return (1);
400 	}
401 	return (0);
402 }
403 
404 static int
405 ctl_backend_ramdisk_move_done(union ctl_io *io, bool samethr)
406 {
407 	struct ctl_be_ramdisk_lun *be_lun =
408 	    (struct ctl_be_ramdisk_lun *)CTL_BACKEND_LUN(io);
409 
410 	CTL_DEBUG_PRINT(("ctl_backend_ramdisk_move_done\n"));
411 	if (io->scsiio.kern_sg_entries > 0)
412 		free(io->scsiio.kern_data_ptr, M_RAMDISK);
413 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
414 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
415 	    (io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE) {
416 		if (ARGS(io)->flags & CTL_LLF_COMPARE) {
417 			/* We have data block ready for comparison. */
418 			if (ctl_backend_ramdisk_cmp(io))
419 				goto done;
420 		}
421 		if (ARGS(io)->len > PRIV(io)->len) {
422 			mtx_lock(&be_lun->queue_lock);
423 			STAILQ_INSERT_TAIL(&be_lun->cont_queue,
424 			    &io->io_hdr, links);
425 			mtx_unlock(&be_lun->queue_lock);
426 			taskqueue_enqueue(be_lun->io_taskqueue,
427 			    &be_lun->io_task);
428 			return (0);
429 		}
430 		ctl_set_success(&io->scsiio);
431 	}
432 done:
433 	ctl_data_submit_done(io);
434 	return(0);
435 }
436 
437 static void
438 ctl_backend_ramdisk_compare(union ctl_io *io)
439 {
440 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
441 	u_int lbas, len;
442 
443 	lbas = ARGS(io)->len - PRIV(io)->len;
444 	lbas = MIN(lbas, 131072 / cbe_lun->blocksize);
445 	len = lbas * cbe_lun->blocksize;
446 
447 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
448 	io->scsiio.kern_data_ptr = malloc(len, M_RAMDISK, M_WAITOK);
449 	io->scsiio.kern_data_len = len;
450 	io->scsiio.kern_sg_entries = 0;
451 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
452 	PRIV(io)->len += lbas;
453 	ctl_datamove(io);
454 }
455 
456 static void
457 ctl_backend_ramdisk_rw(union ctl_io *io)
458 {
459 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
460 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
461 	struct ctl_sg_entry *sg_entries;
462 	uint8_t *page;
463 	uint64_t lba;
464 	u_int i, len, lbaoff, lbas, sgs, off;
465 	getpage_op_t op;
466 
467 	lba = ARGS(io)->lba + PRIV(io)->len;
468 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
469 	lbas = ARGS(io)->len - PRIV(io)->len;
470 	lbas = MIN(lbas, (SGPP << cbe_lun->pblockexp) - lbaoff);
471 	sgs = (lbas + lbaoff + be_lun->pblockmul - 1) >> cbe_lun->pblockexp;
472 	off = lbaoff * cbe_lun->blocksize;
473 	op = (ARGS(io)->flags & CTL_LLF_WRITE) ? GP_WRITE : GP_READ;
474 	if (sgs > 1) {
475 		io->scsiio.kern_data_ptr = malloc(sizeof(struct ctl_sg_entry) *
476 		    sgs, M_RAMDISK, M_WAITOK);
477 		sg_entries = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
478 		len = lbas * cbe_lun->blocksize;
479 		for (i = 0; i < sgs; i++) {
480 			page = ctl_backend_ramdisk_getpage(be_lun,
481 			    (lba >> cbe_lun->pblockexp) + i, op);
482 			if (page == P_UNMAPPED || page == P_ANCHORED) {
483 				free(io->scsiio.kern_data_ptr, M_RAMDISK);
484 nospc:
485 				ctl_set_space_alloc_fail(&io->scsiio);
486 				ctl_data_submit_done(io);
487 				return;
488 			}
489 			sg_entries[i].addr = page + off;
490 			sg_entries[i].len = MIN(len, be_lun->pblocksize - off);
491 			len -= sg_entries[i].len;
492 			off = 0;
493 		}
494 	} else {
495 		page = ctl_backend_ramdisk_getpage(be_lun,
496 		    lba >> cbe_lun->pblockexp, op);
497 		if (page == P_UNMAPPED || page == P_ANCHORED)
498 			goto nospc;
499 		sgs = 0;
500 		io->scsiio.kern_data_ptr = page + off;
501 	}
502 
503 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
504 	io->scsiio.kern_data_len = lbas * cbe_lun->blocksize;
505 	io->scsiio.kern_sg_entries = sgs;
506 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
507 	PRIV(io)->len += lbas;
508 	if ((ARGS(io)->flags & CTL_LLF_READ) &&
509 	    ARGS(io)->len <= PRIV(io)->len) {
510 		ctl_set_success(&io->scsiio);
511 		if (cbe_lun->serseq >= CTL_LUN_SERSEQ_SOFT)
512 			ctl_serseq_done(io);
513 	}
514 	ctl_datamove(io);
515 }
516 
517 static int
518 ctl_backend_ramdisk_submit(union ctl_io *io)
519 {
520 	struct ctl_lba_len_flags *lbalen = ARGS(io);
521 
522 	if (lbalen->flags & CTL_LLF_VERIFY) {
523 		ctl_set_success(&io->scsiio);
524 		ctl_data_submit_done(io);
525 		return (CTL_RETVAL_COMPLETE);
526 	}
527 	PRIV(io)->len = 0;
528 	if (lbalen->flags & CTL_LLF_COMPARE)
529 		ctl_backend_ramdisk_compare(io);
530 	else
531 		ctl_backend_ramdisk_rw(io);
532 	return (CTL_RETVAL_COMPLETE);
533 }
534 
535 static void
536 ctl_backend_ramdisk_worker(void *context, int pending)
537 {
538 	struct ctl_be_ramdisk_lun *be_lun;
539 	union ctl_io *io;
540 
541 	be_lun = (struct ctl_be_ramdisk_lun *)context;
542 	mtx_lock(&be_lun->queue_lock);
543 	for (;;) {
544 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->cont_queue);
545 		if (io != NULL) {
546 			STAILQ_REMOVE_HEAD(&be_lun->cont_queue, links);
547 			mtx_unlock(&be_lun->queue_lock);
548 			if (ARGS(io)->flags & CTL_LLF_COMPARE)
549 				ctl_backend_ramdisk_compare(io);
550 			else
551 				ctl_backend_ramdisk_rw(io);
552 			mtx_lock(&be_lun->queue_lock);
553 			continue;
554 		}
555 
556 		/*
557 		 * If we get here, there is no work left in the queues, so
558 		 * just break out and let the task queue go to sleep.
559 		 */
560 		break;
561 	}
562 	mtx_unlock(&be_lun->queue_lock);
563 }
564 
565 static int
566 ctl_backend_ramdisk_gls(union ctl_io *io)
567 {
568 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
569 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
570 	struct scsi_get_lba_status_data *data;
571 	uint8_t *page;
572 	u_int lbaoff;
573 
574 	data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
575 	scsi_u64to8b(ARGS(io)->lba, data->descr[0].addr);
576 	lbaoff = ARGS(io)->lba & ~(UINT_MAX << cbe_lun->pblockexp);
577 	scsi_ulto4b(be_lun->pblockmul - lbaoff, data->descr[0].length);
578 	page = ctl_backend_ramdisk_getpage(be_lun,
579 	    ARGS(io)->lba >> cbe_lun->pblockexp, GP_OTHER);
580 	if (page == P_UNMAPPED)
581 		data->descr[0].status = 1;
582 	else if (page == P_ANCHORED)
583 		data->descr[0].status = 2;
584 	else
585 		data->descr[0].status = 0;
586 	ctl_config_read_done(io);
587 	return (CTL_RETVAL_COMPLETE);
588 }
589 
590 static int
591 ctl_backend_ramdisk_config_read(union ctl_io *io)
592 {
593 	int retval = 0;
594 
595 	switch (io->scsiio.cdb[0]) {
596 	case SERVICE_ACTION_IN:
597 		if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
598 			retval = ctl_backend_ramdisk_gls(io);
599 			break;
600 		}
601 		ctl_set_invalid_field(&io->scsiio,
602 				      /*sks_valid*/ 1,
603 				      /*command*/ 1,
604 				      /*field*/ 1,
605 				      /*bit_valid*/ 1,
606 				      /*bit*/ 4);
607 		ctl_config_read_done(io);
608 		retval = CTL_RETVAL_COMPLETE;
609 		break;
610 	default:
611 		ctl_set_invalid_opcode(&io->scsiio);
612 		ctl_config_read_done(io);
613 		retval = CTL_RETVAL_COMPLETE;
614 		break;
615 	}
616 	return (retval);
617 }
618 
619 static void
620 ctl_backend_ramdisk_delete(struct ctl_be_lun *cbe_lun, off_t lba, off_t len,
621     int anchor)
622 {
623 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
624 	uint8_t *page;
625 	uint64_t p, lp;
626 	u_int lbaoff;
627 	getpage_op_t op = anchor ? GP_ANCHOR : GP_OTHER;
628 
629 	/* Partially zero first partial page. */
630 	p = lba >> cbe_lun->pblockexp;
631 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
632 	if (lbaoff != 0) {
633 		page = ctl_backend_ramdisk_getpage(be_lun, p, op);
634 		if (page != P_UNMAPPED && page != P_ANCHORED) {
635 			memset(page + lbaoff * cbe_lun->blocksize, 0,
636 			    min(len, be_lun->pblockmul - lbaoff) *
637 			    cbe_lun->blocksize);
638 		}
639 		p++;
640 	}
641 
642 	/* Partially zero last partial page. */
643 	lp = (lba + len) >> cbe_lun->pblockexp;
644 	lbaoff = (lba + len) & ~(UINT_MAX << cbe_lun->pblockexp);
645 	if (p <= lp && lbaoff != 0) {
646 		page = ctl_backend_ramdisk_getpage(be_lun, lp, op);
647 		if (page != P_UNMAPPED && page != P_ANCHORED)
648 			memset(page, 0, lbaoff * cbe_lun->blocksize);
649 	}
650 
651 	/* Delete remaining full pages. */
652 	if (anchor) {
653 		for (; p < lp; p++)
654 			ctl_backend_ramdisk_anchorpage(be_lun, p);
655 	} else {
656 		for (; p < lp; p++)
657 			ctl_backend_ramdisk_unmappage(be_lun, p);
658 	}
659 }
660 
661 static void
662 ctl_backend_ramdisk_ws(union ctl_io *io)
663 {
664 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
665 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
666 	struct ctl_lba_len_flags *lbalen = ARGS(io);
667 	uint8_t *page;
668 	uint64_t lba;
669 	u_int lbaoff, lbas;
670 
671 	if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB)) {
672 		ctl_set_invalid_field(&io->scsiio,
673 				      /*sks_valid*/ 1,
674 				      /*command*/ 1,
675 				      /*field*/ 1,
676 				      /*bit_valid*/ 0,
677 				      /*bit*/ 0);
678 		ctl_config_write_done(io);
679 		return;
680 	}
681 	if (lbalen->flags & SWS_UNMAP) {
682 		ctl_backend_ramdisk_delete(cbe_lun, lbalen->lba, lbalen->len,
683 		    (lbalen->flags & SWS_ANCHOR) != 0);
684 		ctl_set_success(&io->scsiio);
685 		ctl_config_write_done(io);
686 		return;
687 	}
688 
689 	for (lba = lbalen->lba, lbas = lbalen->len; lbas > 0; lba++, lbas--) {
690 		page = ctl_backend_ramdisk_getpage(be_lun,
691 		    lba >> cbe_lun->pblockexp, GP_WRITE);
692 		if (page == P_UNMAPPED || page == P_ANCHORED) {
693 			ctl_set_space_alloc_fail(&io->scsiio);
694 			ctl_data_submit_done(io);
695 			return;
696 		}
697 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
698 		page += lbaoff * cbe_lun->blocksize;
699 		if (lbalen->flags & SWS_NDOB) {
700 			memset(page, 0, cbe_lun->blocksize);
701 		} else {
702 			memcpy(page, io->scsiio.kern_data_ptr,
703 			    cbe_lun->blocksize);
704 		}
705 		if (lbalen->flags & SWS_LBDATA)
706 			scsi_ulto4b(lba, page);
707 	}
708 	ctl_set_success(&io->scsiio);
709 	ctl_config_write_done(io);
710 }
711 
712 static void
713 ctl_backend_ramdisk_unmap(union ctl_io *io)
714 {
715 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
716 	struct ctl_ptr_len_flags *ptrlen = (struct ctl_ptr_len_flags *)ARGS(io);
717 	struct scsi_unmap_desc *buf, *end;
718 
719 	if ((ptrlen->flags & ~SU_ANCHOR) != 0) {
720 		ctl_set_invalid_field(&io->scsiio,
721 				      /*sks_valid*/ 0,
722 				      /*command*/ 0,
723 				      /*field*/ 0,
724 				      /*bit_valid*/ 0,
725 				      /*bit*/ 0);
726 		ctl_config_write_done(io);
727 		return;
728 	}
729 
730 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
731 	end = buf + ptrlen->len / sizeof(*buf);
732 	for (; buf < end; buf++) {
733 		ctl_backend_ramdisk_delete(cbe_lun,
734 		    scsi_8btou64(buf->lba), scsi_4btoul(buf->length),
735 		    (ptrlen->flags & SU_ANCHOR) != 0);
736 	}
737 
738 	ctl_set_success(&io->scsiio);
739 	ctl_config_write_done(io);
740 }
741 
742 static int
743 ctl_backend_ramdisk_config_write(union ctl_io *io)
744 {
745 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
746 	int retval = 0;
747 
748 	switch (io->scsiio.cdb[0]) {
749 	case SYNCHRONIZE_CACHE:
750 	case SYNCHRONIZE_CACHE_16:
751 		/* We have no cache to flush. */
752 		ctl_set_success(&io->scsiio);
753 		ctl_config_write_done(io);
754 		break;
755 	case START_STOP_UNIT: {
756 		struct scsi_start_stop_unit *cdb;
757 
758 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
759 		if ((cdb->how & SSS_PC_MASK) != 0) {
760 			ctl_set_success(&io->scsiio);
761 			ctl_config_write_done(io);
762 			break;
763 		}
764 		if (cdb->how & SSS_START) {
765 			if (cdb->how & SSS_LOEJ)
766 				ctl_lun_has_media(cbe_lun);
767 			ctl_start_lun(cbe_lun);
768 		} else {
769 			ctl_stop_lun(cbe_lun);
770 			if (cdb->how & SSS_LOEJ)
771 				ctl_lun_ejected(cbe_lun);
772 		}
773 		ctl_set_success(&io->scsiio);
774 		ctl_config_write_done(io);
775 		break;
776 	}
777 	case PREVENT_ALLOW:
778 		ctl_set_success(&io->scsiio);
779 		ctl_config_write_done(io);
780 		break;
781 	case WRITE_SAME_10:
782 	case WRITE_SAME_16:
783 		ctl_backend_ramdisk_ws(io);
784 		break;
785 	case UNMAP:
786 		ctl_backend_ramdisk_unmap(io);
787 		break;
788 	default:
789 		ctl_set_invalid_opcode(&io->scsiio);
790 		ctl_config_write_done(io);
791 		retval = CTL_RETVAL_COMPLETE;
792 		break;
793 	}
794 
795 	return (retval);
796 }
797 
798 static uint64_t
799 ctl_backend_ramdisk_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
800 {
801 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
802 	uint64_t		val;
803 
804 	val = UINT64_MAX;
805 	if (be_lun->cap_bytes == 0)
806 		return (val);
807 	sx_slock(&be_lun->page_lock);
808 	if (strcmp(attrname, "blocksused") == 0) {
809 		val = be_lun->cap_used / be_lun->cbe_lun.blocksize;
810 	} else if (strcmp(attrname, "blocksavail") == 0) {
811 		val = (be_lun->cap_bytes - be_lun->cap_used) /
812 		    be_lun->cbe_lun.blocksize;
813 	}
814 	sx_sunlock(&be_lun->page_lock);
815 	return (val);
816 }
817 
818 static int
819 ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
820 			  int flag, struct thread *td)
821 {
822 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
823 	struct ctl_lun_req *lun_req;
824 	int retval;
825 
826 	retval = 0;
827 	switch (cmd) {
828 	case CTL_LUN_REQ:
829 		lun_req = (struct ctl_lun_req *)addr;
830 		switch (lun_req->reqtype) {
831 		case CTL_LUNREQ_CREATE:
832 			retval = ctl_backend_ramdisk_create(softc, lun_req);
833 			break;
834 		case CTL_LUNREQ_RM:
835 			retval = ctl_backend_ramdisk_rm(softc, lun_req);
836 			break;
837 		case CTL_LUNREQ_MODIFY:
838 			retval = ctl_backend_ramdisk_modify(softc, lun_req);
839 			break;
840 		default:
841 			lun_req->status = CTL_LUN_ERROR;
842 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
843 				 "%s: invalid LUN request type %d", __func__,
844 				 lun_req->reqtype);
845 			break;
846 		}
847 		break;
848 	default:
849 		retval = ENOTTY;
850 		break;
851 	}
852 
853 	return (retval);
854 }
855 
856 static int
857 ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
858 		       struct ctl_lun_req *req)
859 {
860 	struct ctl_be_ramdisk_lun *be_lun;
861 	struct ctl_lun_rm_params *params;
862 	int retval;
863 
864 	params = &req->reqdata.rm;
865 	sx_xlock(&softc->modify_lock);
866 	mtx_lock(&softc->lock);
867 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
868 		if (be_lun->cbe_lun.lun_id == params->lun_id) {
869 			SLIST_REMOVE(&softc->lun_list, be_lun,
870 			    ctl_be_ramdisk_lun, links);
871 			softc->num_luns--;
872 			break;
873 		}
874 	}
875 	mtx_unlock(&softc->lock);
876 	sx_xunlock(&softc->modify_lock);
877 	if (be_lun == NULL) {
878 		snprintf(req->error_str, sizeof(req->error_str),
879 			 "%s: LUN %u is not managed by the ramdisk backend",
880 			 __func__, params->lun_id);
881 		goto bailout_error;
882 	}
883 
884 	/*
885 	 * Set the waiting flag before we invalidate the LUN.  Our shutdown
886 	 * routine can be called any time after we invalidate the LUN,
887 	 * and can be called from our context.
888 	 *
889 	 * This tells the shutdown routine that we're waiting, or we're
890 	 * going to wait for the shutdown to happen.
891 	 */
892 	mtx_lock(&softc->lock);
893 	be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING;
894 	mtx_unlock(&softc->lock);
895 
896 	retval = ctl_remove_lun(&be_lun->cbe_lun);
897 	if (retval != 0) {
898 		snprintf(req->error_str, sizeof(req->error_str),
899 			 "%s: error %d returned from ctl_remove_lun() for "
900 			 "LUN %d", __func__, retval, params->lun_id);
901 		mtx_lock(&softc->lock);
902 		be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
903 		mtx_unlock(&softc->lock);
904 		goto bailout_error;
905 	}
906 
907 	mtx_lock(&softc->lock);
908 	while ((be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) == 0) {
909 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlramrm", 0);
910 		if (retval == EINTR)
911 			break;
912 	}
913 	be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
914 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) {
915 		mtx_unlock(&softc->lock);
916 		free(be_lun, M_RAMDISK);
917 	} else {
918 		mtx_unlock(&softc->lock);
919 		return (EINTR);
920 	}
921 
922 	req->status = CTL_LUN_OK;
923 	return (retval);
924 
925 bailout_error:
926 	req->status = CTL_LUN_ERROR;
927 	return (0);
928 }
929 
930 static int
931 ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
932 			   struct ctl_lun_req *req)
933 {
934 	struct ctl_be_ramdisk_lun *be_lun;
935 	struct ctl_be_lun *cbe_lun;
936 	struct ctl_lun_create_params *params;
937 	const char *value;
938 	char tmpstr[32];
939 	uint64_t t;
940 	int retval;
941 
942 	retval = 0;
943 	params = &req->reqdata.create;
944 
945 	be_lun = malloc(sizeof(*be_lun), M_RAMDISK, M_ZERO | M_WAITOK);
946 	cbe_lun = &be_lun->cbe_lun;
947 	cbe_lun->options = nvlist_clone(req->args_nvl);
948 	be_lun->params = req->reqdata.create;
949 	be_lun->softc = softc;
950 
951 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
952 		cbe_lun->lun_type = params->device_type;
953 	else
954 		cbe_lun->lun_type = T_DIRECT;
955 	be_lun->flags = 0;
956 	cbe_lun->flags = 0;
957 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
958 	if (value != NULL) {
959 		if (strcmp(value, "primary") == 0)
960 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
961 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
962 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
963 
964 	be_lun->pblocksize = PAGE_SIZE;
965 	value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
966 	if (value != NULL) {
967 		ctl_expand_number(value, &t);
968 		be_lun->pblocksize = t;
969 	}
970 	if (be_lun->pblocksize < 512 || be_lun->pblocksize > 131072) {
971 		snprintf(req->error_str, sizeof(req->error_str),
972 			 "%s: unsupported pblocksize %u", __func__,
973 			 be_lun->pblocksize);
974 		goto bailout_error;
975 	}
976 
977 	if (cbe_lun->lun_type == T_DIRECT ||
978 	    cbe_lun->lun_type == T_CDROM) {
979 		if (params->blocksize_bytes != 0)
980 			cbe_lun->blocksize = params->blocksize_bytes;
981 		else if (cbe_lun->lun_type == T_CDROM)
982 			cbe_lun->blocksize = 2048;
983 		else
984 			cbe_lun->blocksize = 512;
985 		be_lun->pblockmul = be_lun->pblocksize / cbe_lun->blocksize;
986 		if (be_lun->pblockmul < 1 || !powerof2(be_lun->pblockmul)) {
987 			snprintf(req->error_str, sizeof(req->error_str),
988 				 "%s: pblocksize %u not exp2 of blocksize %u",
989 				 __func__,
990 				 be_lun->pblocksize, cbe_lun->blocksize);
991 			goto bailout_error;
992 		}
993 		if (params->lun_size_bytes < cbe_lun->blocksize) {
994 			snprintf(req->error_str, sizeof(req->error_str),
995 				 "%s: LUN size %ju < blocksize %u", __func__,
996 				 params->lun_size_bytes, cbe_lun->blocksize);
997 			goto bailout_error;
998 		}
999 		be_lun->size_blocks = params->lun_size_bytes / cbe_lun->blocksize;
1000 		be_lun->size_bytes = be_lun->size_blocks * cbe_lun->blocksize;
1001 		be_lun->indir = 0;
1002 		t = be_lun->size_bytes / be_lun->pblocksize;
1003 		while (t > 1) {
1004 			t /= PPP;
1005 			be_lun->indir++;
1006 		}
1007 		cbe_lun->maxlba = be_lun->size_blocks - 1;
1008 		cbe_lun->pblockexp = fls(be_lun->pblockmul) - 1;
1009 		cbe_lun->pblockoff = 0;
1010 		cbe_lun->ublockexp = cbe_lun->pblockexp;
1011 		cbe_lun->ublockoff = 0;
1012 		cbe_lun->atomicblock = be_lun->pblocksize;
1013 		cbe_lun->opttxferlen = SGPP * be_lun->pblocksize;
1014 		value = dnvlist_get_string(cbe_lun->options, "capacity", NULL);
1015 		if (value != NULL)
1016 			ctl_expand_number(value, &be_lun->cap_bytes);
1017 	} else {
1018 		be_lun->pblockmul = 1;
1019 		cbe_lun->pblockexp = 0;
1020 	}
1021 
1022 	/* Tell the user the blocksize we ended up using */
1023 	params->blocksize_bytes = cbe_lun->blocksize;
1024 	params->lun_size_bytes = be_lun->size_bytes;
1025 
1026 	value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
1027 	if (value == NULL || strcmp(value, "off") != 0)
1028 		cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
1029 	value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
1030 	if (value != NULL) {
1031 		if (strcmp(value, "on") == 0)
1032 			cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1033 	} else if (cbe_lun->lun_type != T_DIRECT)
1034 		cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1035 	cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1036 	value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
1037 	if (value != NULL && strcmp(value, "on") == 0)
1038 		cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
1039 	else if (value != NULL && strcmp(value, "read") == 0)
1040 		cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
1041 	else if (value != NULL && strcmp(value, "soft") == 0)
1042 		cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
1043 	else if (value != NULL && strcmp(value, "off") == 0)
1044 		cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1045 
1046 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
1047 		cbe_lun->req_lun_id = params->req_lun_id;
1048 		cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
1049 	} else
1050 		cbe_lun->req_lun_id = 0;
1051 
1052 	cbe_lun->lun_shutdown = ctl_backend_ramdisk_lun_shutdown;
1053 	cbe_lun->be = &ctl_be_ramdisk_driver;
1054 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
1055 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
1056 			 softc->num_luns);
1057 		strncpy((char *)cbe_lun->serial_num, tmpstr,
1058 			MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
1059 
1060 		/* Tell the user what we used for a serial number */
1061 		strncpy((char *)params->serial_num, tmpstr,
1062 			MIN(sizeof(params->serial_num), sizeof(tmpstr)));
1063 	} else {
1064 		strncpy((char *)cbe_lun->serial_num, params->serial_num,
1065 			MIN(sizeof(cbe_lun->serial_num),
1066 			    sizeof(params->serial_num)));
1067 	}
1068 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
1069 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
1070 		strncpy((char *)cbe_lun->device_id, tmpstr,
1071 			MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
1072 
1073 		/* Tell the user what we used for a device ID */
1074 		strncpy((char *)params->device_id, tmpstr,
1075 			MIN(sizeof(params->device_id), sizeof(tmpstr)));
1076 	} else {
1077 		strncpy((char *)cbe_lun->device_id, params->device_id,
1078 			MIN(sizeof(cbe_lun->device_id),
1079 			    sizeof(params->device_id)));
1080 	}
1081 
1082 	STAILQ_INIT(&be_lun->cont_queue);
1083 	sx_init(&be_lun->page_lock, "ctlram page");
1084 	if (be_lun->cap_bytes == 0) {
1085 		be_lun->indir = 0;
1086 		be_lun->pages = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK);
1087 	}
1088 	be_lun->zero_page = malloc(be_lun->pblocksize, M_RAMDISK,
1089 	    M_WAITOK|M_ZERO);
1090 	mtx_init(&be_lun->queue_lock, "ctlram queue", NULL, MTX_DEF);
1091 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_backend_ramdisk_worker,
1092 	    be_lun);
1093 
1094 	be_lun->io_taskqueue = taskqueue_create("ctlramtq", M_WAITOK,
1095 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
1096 	if (be_lun->io_taskqueue == NULL) {
1097 		snprintf(req->error_str, sizeof(req->error_str),
1098 			 "%s: Unable to create taskqueue", __func__);
1099 		goto bailout_error;
1100 	}
1101 
1102 	retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
1103 					 /*num threads*/1,
1104 					 /*priority*/PUSER,
1105 					 /*proc*/control_softc->ctl_proc,
1106 					 /*thread name*/"ramdisk");
1107 	if (retval != 0)
1108 		goto bailout_error;
1109 
1110 	retval = ctl_add_lun(&be_lun->cbe_lun);
1111 	if (retval != 0) {
1112 		snprintf(req->error_str, sizeof(req->error_str),
1113 			 "%s: ctl_add_lun() returned error %d, see dmesg for "
1114 			"details", __func__, retval);
1115 		retval = 0;
1116 		goto bailout_error;
1117 	}
1118 
1119 	mtx_lock(&softc->lock);
1120 	softc->num_luns++;
1121 	SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
1122 	mtx_unlock(&softc->lock);
1123 
1124 	params->req_lun_id = cbe_lun->lun_id;
1125 
1126 	req->status = CTL_LUN_OK;
1127 	return (retval);
1128 
1129 bailout_error:
1130 	req->status = CTL_LUN_ERROR;
1131 	if (be_lun != NULL) {
1132 		if (be_lun->io_taskqueue != NULL)
1133 			taskqueue_free(be_lun->io_taskqueue);
1134 		nvlist_destroy(cbe_lun->options);
1135 		free(be_lun->zero_page, M_RAMDISK);
1136 		ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1137 		sx_destroy(&be_lun->page_lock);
1138 		mtx_destroy(&be_lun->queue_lock);
1139 		free(be_lun, M_RAMDISK);
1140 	}
1141 	return (retval);
1142 }
1143 
1144 static int
1145 ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
1146 		       struct ctl_lun_req *req)
1147 {
1148 	struct ctl_be_ramdisk_lun *be_lun;
1149 	struct ctl_be_lun *cbe_lun;
1150 	struct ctl_lun_modify_params *params;
1151 	const char *value;
1152 	uint32_t blocksize;
1153 	int wasprim;
1154 
1155 	params = &req->reqdata.modify;
1156 	sx_xlock(&softc->modify_lock);
1157 	mtx_lock(&softc->lock);
1158 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
1159 		if (be_lun->cbe_lun.lun_id == params->lun_id)
1160 			break;
1161 	}
1162 	mtx_unlock(&softc->lock);
1163 	if (be_lun == NULL) {
1164 		snprintf(req->error_str, sizeof(req->error_str),
1165 			 "%s: LUN %u is not managed by the ramdisk backend",
1166 			 __func__, params->lun_id);
1167 		goto bailout_error;
1168 	}
1169 	cbe_lun = &be_lun->cbe_lun;
1170 
1171 	if (params->lun_size_bytes != 0)
1172 		be_lun->params.lun_size_bytes = params->lun_size_bytes;
1173 
1174 	if (req->args_nvl != NULL) {
1175 		nvlist_destroy(cbe_lun->options);
1176 		cbe_lun->options = nvlist_clone(req->args_nvl);
1177 	}
1178 
1179 	wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
1180 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
1181 	if (value != NULL) {
1182 		if (strcmp(value, "primary") == 0)
1183 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1184 		else
1185 			cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1186 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
1187 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1188 	else
1189 		cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1190 	if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
1191 		if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
1192 			ctl_lun_primary(cbe_lun);
1193 		else
1194 			ctl_lun_secondary(cbe_lun);
1195 	}
1196 
1197 	blocksize = be_lun->cbe_lun.blocksize;
1198 	if (be_lun->params.lun_size_bytes < blocksize) {
1199 		snprintf(req->error_str, sizeof(req->error_str),
1200 			"%s: LUN size %ju < blocksize %u", __func__,
1201 			be_lun->params.lun_size_bytes, blocksize);
1202 		goto bailout_error;
1203 	}
1204 	be_lun->size_blocks = be_lun->params.lun_size_bytes / blocksize;
1205 	be_lun->size_bytes = be_lun->size_blocks * blocksize;
1206 	be_lun->cbe_lun.maxlba = be_lun->size_blocks - 1;
1207 	ctl_lun_capacity_changed(&be_lun->cbe_lun);
1208 
1209 	/* Tell the user the exact size we ended up using */
1210 	params->lun_size_bytes = be_lun->size_bytes;
1211 
1212 	sx_xunlock(&softc->modify_lock);
1213 	req->status = CTL_LUN_OK;
1214 	return (0);
1215 
1216 bailout_error:
1217 	sx_xunlock(&softc->modify_lock);
1218 	req->status = CTL_LUN_ERROR;
1219 	return (0);
1220 }
1221 
1222 static void
1223 ctl_backend_ramdisk_lun_shutdown(struct ctl_be_lun *cbe_lun)
1224 {
1225 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
1226 	struct ctl_be_ramdisk_softc *softc = be_lun->softc;
1227 
1228 	taskqueue_drain_all(be_lun->io_taskqueue);
1229 	taskqueue_free(be_lun->io_taskqueue);
1230 	nvlist_destroy(be_lun->cbe_lun.options);
1231 	free(be_lun->zero_page, M_RAMDISK);
1232 	ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1233 	sx_destroy(&be_lun->page_lock);
1234 	mtx_destroy(&be_lun->queue_lock);
1235 
1236 	mtx_lock(&softc->lock);
1237 	be_lun->flags |= CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1238 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_WAITING)
1239 		wakeup(be_lun);
1240 	else
1241 		free(be_lun, M_RAMDISK);
1242 	mtx_unlock(&softc->lock);
1243 }
1244