xref: /freebsd/sys/cam/ctl/ctl_backend_ramdisk.c (revision f81cdf24ba5436367377f7c8e8f51f6df2a75ca7)
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/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/condvar.h>
49 #include <sys/types.h>
50 #include <sys/limits.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/malloc.h>
54 #include <sys/sx.h>
55 #include <sys/taskqueue.h>
56 #include <sys/time.h>
57 #include <sys/queue.h>
58 #include <sys/conf.h>
59 #include <sys/ioccom.h>
60 #include <sys/module.h>
61 #include <sys/sysctl.h>
62 #include <sys/nv.h>
63 #include <sys/dnv.h>
64 
65 #include <cam/scsi/scsi_all.h>
66 #include <cam/scsi/scsi_da.h>
67 #include <cam/ctl/ctl_io.h>
68 #include <cam/ctl/ctl.h>
69 #include <cam/ctl/ctl_util.h>
70 #include <cam/ctl/ctl_backend.h>
71 #include <cam/ctl/ctl_debug.h>
72 #include <cam/ctl/ctl_ioctl.h>
73 #include <cam/ctl/ctl_ha.h>
74 #include <cam/ctl/ctl_private.h>
75 #include <cam/ctl/ctl_error.h>
76 
77 #define PRIV(io)	\
78     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
79 #define ARGS(io)	\
80     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
81 
82 #define	PPP	(PAGE_SIZE / sizeof(uint8_t **))
83 #ifdef __LP64__
84 #define	PPPS	(PAGE_SHIFT - 3)
85 #else
86 #define	PPPS	(PAGE_SHIFT - 2)
87 #endif
88 #define	SGPP	(PAGE_SIZE / sizeof(struct ctl_sg_entry))
89 
90 #define	P_UNMAPPED	NULL			/* Page is unmapped. */
91 #define	P_ANCHORED	((void *)(uintptr_t)1)	/* Page is anchored. */
92 
93 typedef enum {
94 	GP_READ,	/* Return data page or zero page. */
95 	GP_WRITE,	/* Return data page, try allocate if none. */
96 	GP_ANCHOR,	/* Return data page, try anchor if none. */
97 	GP_OTHER,	/* Return what present, do not allocate/anchor. */
98 } getpage_op_t;
99 
100 typedef enum {
101 	CTL_BE_RAMDISK_LUN_UNCONFIGURED	= 0x01,
102 	CTL_BE_RAMDISK_LUN_WAITING	= 0x04
103 } ctl_be_ramdisk_lun_flags;
104 
105 struct ctl_be_ramdisk_lun {
106 	struct ctl_be_lun	cbe_lun;	/* Must be first element. */
107 	struct ctl_lun_create_params params;
108 	int			indir;
109 	uint8_t			**pages;
110 	uint8_t			*zero_page;
111 	struct sx		page_lock;
112 	u_int			pblocksize;
113 	u_int			pblockmul;
114 	uint64_t		size_bytes;
115 	uint64_t		size_blocks;
116 	uint64_t		cap_bytes;
117 	uint64_t		cap_used;
118 	struct ctl_be_ramdisk_softc *softc;
119 	ctl_be_ramdisk_lun_flags flags;
120 	SLIST_ENTRY(ctl_be_ramdisk_lun) links;
121 	struct taskqueue	*io_taskqueue;
122 	struct task		io_task;
123 	STAILQ_HEAD(, ctl_io_hdr) cont_queue;
124 	struct mtx_padalign	queue_lock;
125 };
126 
127 struct ctl_be_ramdisk_softc {
128 	struct sx modify_lock;
129 	struct mtx lock;
130 	int num_luns;
131 	SLIST_HEAD(, ctl_be_ramdisk_lun) lun_list;
132 };
133 
134 static struct ctl_be_ramdisk_softc rd_softc;
135 extern struct ctl_softc *control_softc;
136 
137 static int ctl_backend_ramdisk_init(void);
138 static int ctl_backend_ramdisk_shutdown(void);
139 static int ctl_backend_ramdisk_move_done(union ctl_io *io, bool samethr);
140 static void ctl_backend_ramdisk_compare(union ctl_io *io);
141 static void ctl_backend_ramdisk_rw(union ctl_io *io);
142 static int ctl_backend_ramdisk_submit(union ctl_io *io);
143 static void ctl_backend_ramdisk_worker(void *context, int pending);
144 static int ctl_backend_ramdisk_config_read(union ctl_io *io);
145 static int ctl_backend_ramdisk_config_write(union ctl_io *io);
146 static uint64_t ctl_backend_ramdisk_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname);
147 static int ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd,
148 				     caddr_t addr, int flag, struct thread *td);
149 static int ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
150 				  struct ctl_lun_req *req);
151 static int ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
152 				      struct ctl_lun_req *req);
153 static int ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
154 				  struct ctl_lun_req *req);
155 static void ctl_backend_ramdisk_lun_shutdown(struct ctl_be_lun *cbe_lun);
156 
157 static struct ctl_backend_driver ctl_be_ramdisk_driver =
158 {
159 	.name = "ramdisk",
160 	.flags = CTL_BE_FLAG_HAS_CONFIG,
161 	.init = ctl_backend_ramdisk_init,
162 	.shutdown = ctl_backend_ramdisk_shutdown,
163 	.data_submit = ctl_backend_ramdisk_submit,
164 	.config_read = ctl_backend_ramdisk_config_read,
165 	.config_write = ctl_backend_ramdisk_config_write,
166 	.ioctl = ctl_backend_ramdisk_ioctl,
167 	.lun_attr = ctl_backend_ramdisk_lun_attr,
168 };
169 
170 MALLOC_DEFINE(M_RAMDISK, "ctlramdisk", "Memory used for CTL RAMdisk");
171 CTL_BACKEND_DECLARE(cbr, ctl_be_ramdisk_driver);
172 
173 static int
174 ctl_backend_ramdisk_init(void)
175 {
176 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
177 
178 	memset(softc, 0, sizeof(*softc));
179 	sx_init(&softc->modify_lock, "ctlrammod");
180 	mtx_init(&softc->lock, "ctlram", NULL, MTX_DEF);
181 	SLIST_INIT(&softc->lun_list);
182 	return (0);
183 }
184 
185 static int
186 ctl_backend_ramdisk_shutdown(void)
187 {
188 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
189 	struct ctl_be_ramdisk_lun *lun;
190 
191 	mtx_lock(&softc->lock);
192 	while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
193 		SLIST_REMOVE_HEAD(&softc->lun_list, links);
194 		softc->num_luns--;
195 		/*
196 		 * Drop our lock here.  Since ctl_remove_lun() can call
197 		 * back into us, this could potentially lead to a recursive
198 		 * lock of the same mutex, which would cause a hang.
199 		 */
200 		mtx_unlock(&softc->lock);
201 		ctl_remove_lun(&lun->cbe_lun);
202 		mtx_lock(&softc->lock);
203 	}
204 	mtx_unlock(&softc->lock);
205 	mtx_destroy(&softc->lock);
206 	sx_destroy(&softc->modify_lock);
207 	return (0);
208 }
209 
210 static uint8_t *
211 ctl_backend_ramdisk_getpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn,
212     getpage_op_t op)
213 {
214 	uint8_t **p, ***pp;
215 	off_t i;
216 	int s;
217 
218 	if (be_lun->cap_bytes == 0) {
219 		switch (op) {
220 		case GP_READ:
221 			return (be_lun->zero_page);
222 		case GP_WRITE:
223 			return ((uint8_t *)be_lun->pages);
224 		case GP_ANCHOR:
225 			return (P_ANCHORED);
226 		default:
227 			return (P_UNMAPPED);
228 		}
229 	}
230 	if (op == GP_WRITE || op == GP_ANCHOR) {
231 		sx_xlock(&be_lun->page_lock);
232 		pp = &be_lun->pages;
233 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
234 			if (*pp == NULL) {
235 				*pp = malloc(PAGE_SIZE, M_RAMDISK,
236 				    M_WAITOK|M_ZERO);
237 			}
238 			i = pn >> s;
239 			pp = (uint8_t ***)&(*pp)[i];
240 			pn -= i << s;
241 		}
242 		if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
243 			if (op == GP_WRITE) {
244 				*pp = malloc(be_lun->pblocksize, M_RAMDISK,
245 				    M_WAITOK|M_ZERO);
246 			} else
247 				*pp = P_ANCHORED;
248 			be_lun->cap_used += be_lun->pblocksize;
249 		} else if (*pp == P_ANCHORED && op == GP_WRITE) {
250 			*pp = malloc(be_lun->pblocksize, M_RAMDISK,
251 			    M_WAITOK|M_ZERO);
252 		}
253 		sx_xunlock(&be_lun->page_lock);
254 		return ((uint8_t *)*pp);
255 	} else {
256 		sx_slock(&be_lun->page_lock);
257 		p = be_lun->pages;
258 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
259 			if (p == NULL)
260 				break;
261 			i = pn >> s;
262 			p = (uint8_t **)p[i];
263 			pn -= i << s;
264 		}
265 		sx_sunlock(&be_lun->page_lock);
266 		if ((p == P_UNMAPPED || p == P_ANCHORED) && op == GP_READ)
267 			return (be_lun->zero_page);
268 		return ((uint8_t *)p);
269 	}
270 };
271 
272 static void
273 ctl_backend_ramdisk_unmappage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
274 {
275 	uint8_t ***pp;
276 	off_t i;
277 	int s;
278 
279 	if (be_lun->cap_bytes == 0)
280 		return;
281 	sx_xlock(&be_lun->page_lock);
282 	pp = &be_lun->pages;
283 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
284 		if (*pp == NULL)
285 			goto noindir;
286 		i = pn >> s;
287 		pp = (uint8_t ***)&(*pp)[i];
288 		pn -= i << s;
289 	}
290 	if (*pp == P_ANCHORED) {
291 		be_lun->cap_used -= be_lun->pblocksize;
292 		*pp = P_UNMAPPED;
293 	} else if (*pp != P_UNMAPPED) {
294 		free(*pp, M_RAMDISK);
295 		be_lun->cap_used -= be_lun->pblocksize;
296 		*pp = P_UNMAPPED;
297 	}
298 noindir:
299 	sx_xunlock(&be_lun->page_lock);
300 };
301 
302 static void
303 ctl_backend_ramdisk_anchorpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
304 {
305 	uint8_t ***pp;
306 	off_t i;
307 	int s;
308 
309 	if (be_lun->cap_bytes == 0)
310 		return;
311 	sx_xlock(&be_lun->page_lock);
312 	pp = &be_lun->pages;
313 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
314 		if (*pp == NULL)
315 			goto noindir;
316 		i = pn >> s;
317 		pp = (uint8_t ***)&(*pp)[i];
318 		pn -= i << s;
319 	}
320 	if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
321 		be_lun->cap_used += be_lun->pblocksize;
322 		*pp = P_ANCHORED;
323 	} else if (*pp != P_ANCHORED) {
324 		free(*pp, M_RAMDISK);
325 		*pp = P_ANCHORED;
326 	}
327 noindir:
328 	sx_xunlock(&be_lun->page_lock);
329 };
330 
331 static void
332 ctl_backend_ramdisk_freeallpages(uint8_t **p, int indir)
333 {
334 	int i;
335 
336 	if (p == NULL)
337 		return;
338 	if (indir == 0) {
339 		free(p, M_RAMDISK);
340 		return;
341 	}
342 	for (i = 0; i < PPP; i++) {
343 		if (p[i] == NULL)
344 			continue;
345 		ctl_backend_ramdisk_freeallpages((uint8_t **)p[i], indir - 1);
346 	}
347 	free(p, M_RAMDISK);
348 };
349 
350 static size_t
351 cmp(uint8_t *a, uint8_t *b, size_t size)
352 {
353 	size_t i;
354 
355 	for (i = 0; i < size; i++) {
356 		if (a[i] != b[i])
357 			break;
358 	}
359 	return (i);
360 }
361 
362 static int
363 ctl_backend_ramdisk_cmp(union ctl_io *io)
364 {
365 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
366 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
367 	uint8_t *page;
368 	uint8_t info[8];
369 	uint64_t lba;
370 	u_int lbaoff, lbas, res, off;
371 
372 	lbas = io->scsiio.kern_data_len / cbe_lun->blocksize;
373 	lba = ARGS(io)->lba + PRIV(io)->len - lbas;
374 	off = 0;
375 	for (; lbas > 0; lbas--, lba++) {
376 		page = ctl_backend_ramdisk_getpage(be_lun,
377 		    lba >> cbe_lun->pblockexp, GP_READ);
378 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
379 		page += lbaoff * cbe_lun->blocksize;
380 		res = cmp(io->scsiio.kern_data_ptr + off, page,
381 		    cbe_lun->blocksize);
382 		off += res;
383 		if (res < cbe_lun->blocksize)
384 			break;
385 	}
386 	free(io->scsiio.kern_data_ptr, M_RAMDISK);
387 	if (lbas > 0) {
388 		off += io->scsiio.kern_rel_offset - io->scsiio.kern_data_len;
389 		scsi_u64to8b(off, info);
390 		ctl_set_sense(&io->scsiio, /*current_error*/ 1,
391 		    /*sense_key*/ SSD_KEY_MISCOMPARE,
392 		    /*asc*/ 0x1D, /*ascq*/ 0x00,
393 		    /*type*/ SSD_ELEM_INFO,
394 		    /*size*/ sizeof(info), /*data*/ &info,
395 		    /*type*/ SSD_ELEM_NONE);
396 		return (1);
397 	}
398 	return (0);
399 }
400 
401 static int
402 ctl_backend_ramdisk_move_done(union ctl_io *io, bool samethr)
403 {
404 	struct ctl_be_ramdisk_lun *be_lun =
405 	    (struct ctl_be_ramdisk_lun *)CTL_BACKEND_LUN(io);
406 
407 	CTL_DEBUG_PRINT(("ctl_backend_ramdisk_move_done\n"));
408 	if (io->scsiio.kern_sg_entries > 0)
409 		free(io->scsiio.kern_data_ptr, M_RAMDISK);
410 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
411 	if ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
412 	    (io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE) {
413 		if (ARGS(io)->flags & CTL_LLF_COMPARE) {
414 			/* We have data block ready for comparison. */
415 			if (ctl_backend_ramdisk_cmp(io))
416 				goto done;
417 		}
418 		if (ARGS(io)->len > PRIV(io)->len) {
419 			mtx_lock(&be_lun->queue_lock);
420 			STAILQ_INSERT_TAIL(&be_lun->cont_queue,
421 			    &io->io_hdr, links);
422 			mtx_unlock(&be_lun->queue_lock);
423 			taskqueue_enqueue(be_lun->io_taskqueue,
424 			    &be_lun->io_task);
425 			return (0);
426 		}
427 		ctl_set_success(&io->scsiio);
428 	}
429 done:
430 	ctl_data_submit_done(io);
431 	return(0);
432 }
433 
434 static void
435 ctl_backend_ramdisk_compare(union ctl_io *io)
436 {
437 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
438 	u_int lbas, len;
439 
440 	lbas = ARGS(io)->len - PRIV(io)->len;
441 	lbas = MIN(lbas, 131072 / cbe_lun->blocksize);
442 	len = lbas * cbe_lun->blocksize;
443 
444 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
445 	io->scsiio.kern_data_ptr = malloc(len, M_RAMDISK, M_WAITOK);
446 	io->scsiio.kern_data_len = len;
447 	io->scsiio.kern_sg_entries = 0;
448 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
449 	PRIV(io)->len += lbas;
450 	ctl_datamove(io);
451 }
452 
453 static void
454 ctl_backend_ramdisk_rw(union ctl_io *io)
455 {
456 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
457 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
458 	struct ctl_sg_entry *sg_entries;
459 	uint8_t *page;
460 	uint64_t lba;
461 	u_int i, len, lbaoff, lbas, sgs, off;
462 	getpage_op_t op;
463 
464 	lba = ARGS(io)->lba + PRIV(io)->len;
465 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
466 	lbas = ARGS(io)->len - PRIV(io)->len;
467 	lbas = MIN(lbas, (SGPP << cbe_lun->pblockexp) - lbaoff);
468 	sgs = (lbas + lbaoff + be_lun->pblockmul - 1) >> cbe_lun->pblockexp;
469 	off = lbaoff * cbe_lun->blocksize;
470 	op = (ARGS(io)->flags & CTL_LLF_WRITE) ? GP_WRITE : GP_READ;
471 	if (sgs > 1) {
472 		io->scsiio.kern_data_ptr = malloc(sizeof(struct ctl_sg_entry) *
473 		    sgs, M_RAMDISK, M_WAITOK);
474 		sg_entries = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
475 		len = lbas * cbe_lun->blocksize;
476 		for (i = 0; i < sgs; i++) {
477 			page = ctl_backend_ramdisk_getpage(be_lun,
478 			    (lba >> cbe_lun->pblockexp) + i, op);
479 			if (page == P_UNMAPPED || page == P_ANCHORED) {
480 				free(io->scsiio.kern_data_ptr, M_RAMDISK);
481 nospc:
482 				ctl_set_space_alloc_fail(&io->scsiio);
483 				ctl_data_submit_done(io);
484 				return;
485 			}
486 			sg_entries[i].addr = page + off;
487 			sg_entries[i].len = MIN(len, be_lun->pblocksize - off);
488 			len -= sg_entries[i].len;
489 			off = 0;
490 		}
491 	} else {
492 		page = ctl_backend_ramdisk_getpage(be_lun,
493 		    lba >> cbe_lun->pblockexp, op);
494 		if (page == P_UNMAPPED || page == P_ANCHORED)
495 			goto nospc;
496 		sgs = 0;
497 		io->scsiio.kern_data_ptr = page + off;
498 	}
499 
500 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
501 	io->scsiio.kern_data_len = lbas * cbe_lun->blocksize;
502 	io->scsiio.kern_sg_entries = sgs;
503 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
504 	PRIV(io)->len += lbas;
505 	if ((ARGS(io)->flags & CTL_LLF_READ) &&
506 	    ARGS(io)->len <= PRIV(io)->len) {
507 		ctl_set_success(&io->scsiio);
508 		if (cbe_lun->serseq >= CTL_LUN_SERSEQ_SOFT)
509 			ctl_serseq_done(io);
510 	}
511 	ctl_datamove(io);
512 }
513 
514 static int
515 ctl_backend_ramdisk_submit(union ctl_io *io)
516 {
517 	struct ctl_lba_len_flags *lbalen = ARGS(io);
518 
519 	if (lbalen->flags & CTL_LLF_VERIFY) {
520 		ctl_set_success(&io->scsiio);
521 		ctl_data_submit_done(io);
522 		return (CTL_RETVAL_COMPLETE);
523 	}
524 	PRIV(io)->len = 0;
525 	if (lbalen->flags & CTL_LLF_COMPARE)
526 		ctl_backend_ramdisk_compare(io);
527 	else
528 		ctl_backend_ramdisk_rw(io);
529 	return (CTL_RETVAL_COMPLETE);
530 }
531 
532 static void
533 ctl_backend_ramdisk_worker(void *context, int pending)
534 {
535 	struct ctl_be_ramdisk_lun *be_lun;
536 	union ctl_io *io;
537 
538 	be_lun = (struct ctl_be_ramdisk_lun *)context;
539 	mtx_lock(&be_lun->queue_lock);
540 	for (;;) {
541 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->cont_queue);
542 		if (io != NULL) {
543 			STAILQ_REMOVE_HEAD(&be_lun->cont_queue, links);
544 			mtx_unlock(&be_lun->queue_lock);
545 			if (ARGS(io)->flags & CTL_LLF_COMPARE)
546 				ctl_backend_ramdisk_compare(io);
547 			else
548 				ctl_backend_ramdisk_rw(io);
549 			mtx_lock(&be_lun->queue_lock);
550 			continue;
551 		}
552 
553 		/*
554 		 * If we get here, there is no work left in the queues, so
555 		 * just break out and let the task queue go to sleep.
556 		 */
557 		break;
558 	}
559 	mtx_unlock(&be_lun->queue_lock);
560 }
561 
562 static int
563 ctl_backend_ramdisk_gls(union ctl_io *io)
564 {
565 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
566 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
567 	struct scsi_get_lba_status_data *data;
568 	uint8_t *page;
569 	u_int lbaoff;
570 
571 	data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
572 	scsi_u64to8b(ARGS(io)->lba, data->descr[0].addr);
573 	lbaoff = ARGS(io)->lba & ~(UINT_MAX << cbe_lun->pblockexp);
574 	scsi_ulto4b(be_lun->pblockmul - lbaoff, data->descr[0].length);
575 	page = ctl_backend_ramdisk_getpage(be_lun,
576 	    ARGS(io)->lba >> cbe_lun->pblockexp, GP_OTHER);
577 	if (page == P_UNMAPPED)
578 		data->descr[0].status = 1;
579 	else if (page == P_ANCHORED)
580 		data->descr[0].status = 2;
581 	else
582 		data->descr[0].status = 0;
583 	ctl_config_read_done(io);
584 	return (CTL_RETVAL_COMPLETE);
585 }
586 
587 static int
588 ctl_backend_ramdisk_config_read(union ctl_io *io)
589 {
590 	int retval = 0;
591 
592 	switch (io->scsiio.cdb[0]) {
593 	case SERVICE_ACTION_IN:
594 		if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
595 			retval = ctl_backend_ramdisk_gls(io);
596 			break;
597 		}
598 		ctl_set_invalid_field(&io->scsiio,
599 				      /*sks_valid*/ 1,
600 				      /*command*/ 1,
601 				      /*field*/ 1,
602 				      /*bit_valid*/ 1,
603 				      /*bit*/ 4);
604 		ctl_config_read_done(io);
605 		retval = CTL_RETVAL_COMPLETE;
606 		break;
607 	default:
608 		ctl_set_invalid_opcode(&io->scsiio);
609 		ctl_config_read_done(io);
610 		retval = CTL_RETVAL_COMPLETE;
611 		break;
612 	}
613 	return (retval);
614 }
615 
616 static void
617 ctl_backend_ramdisk_delete(struct ctl_be_lun *cbe_lun, off_t lba, off_t len,
618     int anchor)
619 {
620 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
621 	uint8_t *page;
622 	uint64_t p, lp;
623 	u_int lbaoff;
624 	getpage_op_t op = anchor ? GP_ANCHOR : GP_OTHER;
625 
626 	/* Partially zero first partial page. */
627 	p = lba >> cbe_lun->pblockexp;
628 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
629 	if (lbaoff != 0) {
630 		page = ctl_backend_ramdisk_getpage(be_lun, p, op);
631 		if (page != P_UNMAPPED && page != P_ANCHORED) {
632 			memset(page + lbaoff * cbe_lun->blocksize, 0,
633 			    min(len, be_lun->pblockmul - lbaoff) *
634 			    cbe_lun->blocksize);
635 		}
636 		p++;
637 	}
638 
639 	/* Partially zero last partial page. */
640 	lp = (lba + len) >> cbe_lun->pblockexp;
641 	lbaoff = (lba + len) & ~(UINT_MAX << cbe_lun->pblockexp);
642 	if (p <= lp && lbaoff != 0) {
643 		page = ctl_backend_ramdisk_getpage(be_lun, lp, op);
644 		if (page != P_UNMAPPED && page != P_ANCHORED)
645 			memset(page, 0, lbaoff * cbe_lun->blocksize);
646 	}
647 
648 	/* Delete remaining full pages. */
649 	if (anchor) {
650 		for (; p < lp; p++)
651 			ctl_backend_ramdisk_anchorpage(be_lun, p);
652 	} else {
653 		for (; p < lp; p++)
654 			ctl_backend_ramdisk_unmappage(be_lun, p);
655 	}
656 }
657 
658 static void
659 ctl_backend_ramdisk_ws(union ctl_io *io)
660 {
661 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
662 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
663 	struct ctl_lba_len_flags *lbalen = ARGS(io);
664 	uint8_t *page;
665 	uint64_t lba;
666 	u_int lbaoff, lbas;
667 
668 	if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB)) {
669 		ctl_set_invalid_field(&io->scsiio,
670 				      /*sks_valid*/ 1,
671 				      /*command*/ 1,
672 				      /*field*/ 1,
673 				      /*bit_valid*/ 0,
674 				      /*bit*/ 0);
675 		ctl_config_write_done(io);
676 		return;
677 	}
678 	if (lbalen->flags & SWS_UNMAP) {
679 		ctl_backend_ramdisk_delete(cbe_lun, lbalen->lba, lbalen->len,
680 		    (lbalen->flags & SWS_ANCHOR) != 0);
681 		ctl_set_success(&io->scsiio);
682 		ctl_config_write_done(io);
683 		return;
684 	}
685 
686 	for (lba = lbalen->lba, lbas = lbalen->len; lbas > 0; lba++, lbas--) {
687 		page = ctl_backend_ramdisk_getpage(be_lun,
688 		    lba >> cbe_lun->pblockexp, GP_WRITE);
689 		if (page == P_UNMAPPED || page == P_ANCHORED) {
690 			ctl_set_space_alloc_fail(&io->scsiio);
691 			ctl_data_submit_done(io);
692 			return;
693 		}
694 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
695 		page += lbaoff * cbe_lun->blocksize;
696 		if (lbalen->flags & SWS_NDOB) {
697 			memset(page, 0, cbe_lun->blocksize);
698 		} else {
699 			memcpy(page, io->scsiio.kern_data_ptr,
700 			    cbe_lun->blocksize);
701 		}
702 		if (lbalen->flags & SWS_LBDATA)
703 			scsi_ulto4b(lba, page);
704 	}
705 	ctl_set_success(&io->scsiio);
706 	ctl_config_write_done(io);
707 }
708 
709 static void
710 ctl_backend_ramdisk_unmap(union ctl_io *io)
711 {
712 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
713 	struct ctl_ptr_len_flags *ptrlen = (struct ctl_ptr_len_flags *)ARGS(io);
714 	struct scsi_unmap_desc *buf, *end;
715 
716 	if ((ptrlen->flags & ~SU_ANCHOR) != 0) {
717 		ctl_set_invalid_field(&io->scsiio,
718 				      /*sks_valid*/ 0,
719 				      /*command*/ 0,
720 				      /*field*/ 0,
721 				      /*bit_valid*/ 0,
722 				      /*bit*/ 0);
723 		ctl_config_write_done(io);
724 		return;
725 	}
726 
727 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
728 	end = buf + ptrlen->len / sizeof(*buf);
729 	for (; buf < end; buf++) {
730 		ctl_backend_ramdisk_delete(cbe_lun,
731 		    scsi_8btou64(buf->lba), scsi_4btoul(buf->length),
732 		    (ptrlen->flags & SU_ANCHOR) != 0);
733 	}
734 
735 	ctl_set_success(&io->scsiio);
736 	ctl_config_write_done(io);
737 }
738 
739 static int
740 ctl_backend_ramdisk_config_write(union ctl_io *io)
741 {
742 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
743 	int retval = 0;
744 
745 	switch (io->scsiio.cdb[0]) {
746 	case SYNCHRONIZE_CACHE:
747 	case SYNCHRONIZE_CACHE_16:
748 		/* We have no cache to flush. */
749 		ctl_set_success(&io->scsiio);
750 		ctl_config_write_done(io);
751 		break;
752 	case START_STOP_UNIT: {
753 		struct scsi_start_stop_unit *cdb;
754 
755 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
756 		if ((cdb->how & SSS_PC_MASK) != 0) {
757 			ctl_set_success(&io->scsiio);
758 			ctl_config_write_done(io);
759 			break;
760 		}
761 		if (cdb->how & SSS_START) {
762 			if (cdb->how & SSS_LOEJ)
763 				ctl_lun_has_media(cbe_lun);
764 			ctl_start_lun(cbe_lun);
765 		} else {
766 			ctl_stop_lun(cbe_lun);
767 			if (cdb->how & SSS_LOEJ)
768 				ctl_lun_ejected(cbe_lun);
769 		}
770 		ctl_set_success(&io->scsiio);
771 		ctl_config_write_done(io);
772 		break;
773 	}
774 	case PREVENT_ALLOW:
775 		ctl_set_success(&io->scsiio);
776 		ctl_config_write_done(io);
777 		break;
778 	case WRITE_SAME_10:
779 	case WRITE_SAME_16:
780 		ctl_backend_ramdisk_ws(io);
781 		break;
782 	case UNMAP:
783 		ctl_backend_ramdisk_unmap(io);
784 		break;
785 	default:
786 		ctl_set_invalid_opcode(&io->scsiio);
787 		ctl_config_write_done(io);
788 		retval = CTL_RETVAL_COMPLETE;
789 		break;
790 	}
791 
792 	return (retval);
793 }
794 
795 static uint64_t
796 ctl_backend_ramdisk_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
797 {
798 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
799 	uint64_t		val;
800 
801 	val = UINT64_MAX;
802 	if (be_lun->cap_bytes == 0)
803 		return (val);
804 	sx_slock(&be_lun->page_lock);
805 	if (strcmp(attrname, "blocksused") == 0) {
806 		val = be_lun->cap_used / be_lun->cbe_lun.blocksize;
807 	} else if (strcmp(attrname, "blocksavail") == 0) {
808 		val = (be_lun->cap_bytes - be_lun->cap_used) /
809 		    be_lun->cbe_lun.blocksize;
810 	}
811 	sx_sunlock(&be_lun->page_lock);
812 	return (val);
813 }
814 
815 static int
816 ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
817 			  int flag, struct thread *td)
818 {
819 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
820 	struct ctl_lun_req *lun_req;
821 	int retval;
822 
823 	retval = 0;
824 	switch (cmd) {
825 	case CTL_LUN_REQ:
826 		lun_req = (struct ctl_lun_req *)addr;
827 		switch (lun_req->reqtype) {
828 		case CTL_LUNREQ_CREATE:
829 			retval = ctl_backend_ramdisk_create(softc, lun_req);
830 			break;
831 		case CTL_LUNREQ_RM:
832 			retval = ctl_backend_ramdisk_rm(softc, lun_req);
833 			break;
834 		case CTL_LUNREQ_MODIFY:
835 			retval = ctl_backend_ramdisk_modify(softc, lun_req);
836 			break;
837 		default:
838 			lun_req->status = CTL_LUN_ERROR;
839 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
840 				 "%s: invalid LUN request type %d", __func__,
841 				 lun_req->reqtype);
842 			break;
843 		}
844 		break;
845 	default:
846 		retval = ENOTTY;
847 		break;
848 	}
849 
850 	return (retval);
851 }
852 
853 static int
854 ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
855 		       struct ctl_lun_req *req)
856 {
857 	struct ctl_be_ramdisk_lun *be_lun;
858 	struct ctl_lun_rm_params *params;
859 	int retval;
860 
861 	params = &req->reqdata.rm;
862 	sx_xlock(&softc->modify_lock);
863 	mtx_lock(&softc->lock);
864 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
865 		if (be_lun->cbe_lun.lun_id == params->lun_id) {
866 			SLIST_REMOVE(&softc->lun_list, be_lun,
867 			    ctl_be_ramdisk_lun, links);
868 			softc->num_luns--;
869 			break;
870 		}
871 	}
872 	mtx_unlock(&softc->lock);
873 	sx_xunlock(&softc->modify_lock);
874 	if (be_lun == NULL) {
875 		snprintf(req->error_str, sizeof(req->error_str),
876 			 "%s: LUN %u is not managed by the ramdisk backend",
877 			 __func__, params->lun_id);
878 		goto bailout_error;
879 	}
880 
881 	/*
882 	 * Set the waiting flag before we invalidate the LUN.  Our shutdown
883 	 * routine can be called any time after we invalidate the LUN,
884 	 * and can be called from our context.
885 	 *
886 	 * This tells the shutdown routine that we're waiting, or we're
887 	 * going to wait for the shutdown to happen.
888 	 */
889 	mtx_lock(&softc->lock);
890 	be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING;
891 	mtx_unlock(&softc->lock);
892 
893 	retval = ctl_remove_lun(&be_lun->cbe_lun);
894 	if (retval != 0) {
895 		snprintf(req->error_str, sizeof(req->error_str),
896 			 "%s: error %d returned from ctl_remove_lun() for "
897 			 "LUN %d", __func__, retval, params->lun_id);
898 		mtx_lock(&softc->lock);
899 		be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
900 		mtx_unlock(&softc->lock);
901 		goto bailout_error;
902 	}
903 
904 	mtx_lock(&softc->lock);
905 	while ((be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) == 0) {
906 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlramrm", 0);
907 		if (retval == EINTR)
908 			break;
909 	}
910 	be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
911 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) {
912 		mtx_unlock(&softc->lock);
913 		free(be_lun, M_RAMDISK);
914 	} else {
915 		mtx_unlock(&softc->lock);
916 		return (EINTR);
917 	}
918 
919 	req->status = CTL_LUN_OK;
920 	return (retval);
921 
922 bailout_error:
923 	req->status = CTL_LUN_ERROR;
924 	return (0);
925 }
926 
927 static int
928 ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
929 			   struct ctl_lun_req *req)
930 {
931 	struct ctl_be_ramdisk_lun *be_lun;
932 	struct ctl_be_lun *cbe_lun;
933 	struct ctl_lun_create_params *params;
934 	const char *value;
935 	char tmpstr[32];
936 	uint64_t t;
937 	int retval;
938 
939 	retval = 0;
940 	params = &req->reqdata.create;
941 
942 	be_lun = malloc(sizeof(*be_lun), M_RAMDISK, M_ZERO | M_WAITOK);
943 	cbe_lun = &be_lun->cbe_lun;
944 	cbe_lun->options = nvlist_clone(req->args_nvl);
945 	be_lun->params = req->reqdata.create;
946 	be_lun->softc = softc;
947 
948 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
949 		cbe_lun->lun_type = params->device_type;
950 	else
951 		cbe_lun->lun_type = T_DIRECT;
952 	be_lun->flags = 0;
953 	cbe_lun->flags = 0;
954 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
955 	if (value != NULL) {
956 		if (strcmp(value, "primary") == 0)
957 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
958 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
959 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
960 
961 	be_lun->pblocksize = PAGE_SIZE;
962 	value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
963 	if (value != NULL) {
964 		ctl_expand_number(value, &t);
965 		be_lun->pblocksize = t;
966 	}
967 	if (be_lun->pblocksize < 512 || be_lun->pblocksize > 131072) {
968 		snprintf(req->error_str, sizeof(req->error_str),
969 			 "%s: unsupported pblocksize %u", __func__,
970 			 be_lun->pblocksize);
971 		goto bailout_error;
972 	}
973 
974 	if (cbe_lun->lun_type == T_DIRECT ||
975 	    cbe_lun->lun_type == T_CDROM) {
976 		if (params->blocksize_bytes != 0)
977 			cbe_lun->blocksize = params->blocksize_bytes;
978 		else if (cbe_lun->lun_type == T_CDROM)
979 			cbe_lun->blocksize = 2048;
980 		else
981 			cbe_lun->blocksize = 512;
982 		be_lun->pblockmul = be_lun->pblocksize / cbe_lun->blocksize;
983 		if (be_lun->pblockmul < 1 || !powerof2(be_lun->pblockmul)) {
984 			snprintf(req->error_str, sizeof(req->error_str),
985 				 "%s: pblocksize %u not exp2 of blocksize %u",
986 				 __func__,
987 				 be_lun->pblocksize, cbe_lun->blocksize);
988 			goto bailout_error;
989 		}
990 		if (params->lun_size_bytes < cbe_lun->blocksize) {
991 			snprintf(req->error_str, sizeof(req->error_str),
992 				 "%s: LUN size %ju < blocksize %u", __func__,
993 				 params->lun_size_bytes, cbe_lun->blocksize);
994 			goto bailout_error;
995 		}
996 		be_lun->size_blocks = params->lun_size_bytes / cbe_lun->blocksize;
997 		be_lun->size_bytes = be_lun->size_blocks * cbe_lun->blocksize;
998 		be_lun->indir = 0;
999 		t = be_lun->size_bytes / be_lun->pblocksize;
1000 		while (t > 1) {
1001 			t /= PPP;
1002 			be_lun->indir++;
1003 		}
1004 		cbe_lun->maxlba = be_lun->size_blocks - 1;
1005 		cbe_lun->pblockexp = fls(be_lun->pblockmul) - 1;
1006 		cbe_lun->pblockoff = 0;
1007 		cbe_lun->ublockexp = cbe_lun->pblockexp;
1008 		cbe_lun->ublockoff = 0;
1009 		cbe_lun->atomicblock = be_lun->pblocksize;
1010 		cbe_lun->opttxferlen = SGPP * be_lun->pblocksize;
1011 		value = dnvlist_get_string(cbe_lun->options, "capacity", NULL);
1012 		if (value != NULL)
1013 			ctl_expand_number(value, &be_lun->cap_bytes);
1014 	} else {
1015 		be_lun->pblockmul = 1;
1016 		cbe_lun->pblockexp = 0;
1017 	}
1018 
1019 	/* Tell the user the blocksize we ended up using */
1020 	params->blocksize_bytes = cbe_lun->blocksize;
1021 	params->lun_size_bytes = be_lun->size_bytes;
1022 
1023 	value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
1024 	if (value == NULL || strcmp(value, "off") != 0)
1025 		cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
1026 	value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
1027 	if (value != NULL) {
1028 		if (strcmp(value, "on") == 0)
1029 			cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1030 	} else if (cbe_lun->lun_type != T_DIRECT)
1031 		cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1032 	cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1033 	value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
1034 	if (value != NULL && strcmp(value, "on") == 0)
1035 		cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
1036 	else if (value != NULL && strcmp(value, "read") == 0)
1037 		cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
1038 	else if (value != NULL && strcmp(value, "soft") == 0)
1039 		cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
1040 	else if (value != NULL && strcmp(value, "off") == 0)
1041 		cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1042 
1043 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
1044 		cbe_lun->req_lun_id = params->req_lun_id;
1045 		cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
1046 	} else
1047 		cbe_lun->req_lun_id = 0;
1048 
1049 	cbe_lun->lun_shutdown = ctl_backend_ramdisk_lun_shutdown;
1050 	cbe_lun->be = &ctl_be_ramdisk_driver;
1051 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
1052 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
1053 			 softc->num_luns);
1054 		strncpy((char *)cbe_lun->serial_num, tmpstr,
1055 			MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
1056 
1057 		/* Tell the user what we used for a serial number */
1058 		strncpy((char *)params->serial_num, tmpstr,
1059 			MIN(sizeof(params->serial_num), sizeof(tmpstr)));
1060 	} else {
1061 		strncpy((char *)cbe_lun->serial_num, params->serial_num,
1062 			MIN(sizeof(cbe_lun->serial_num),
1063 			    sizeof(params->serial_num)));
1064 	}
1065 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
1066 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
1067 		strncpy((char *)cbe_lun->device_id, tmpstr,
1068 			MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
1069 
1070 		/* Tell the user what we used for a device ID */
1071 		strncpy((char *)params->device_id, tmpstr,
1072 			MIN(sizeof(params->device_id), sizeof(tmpstr)));
1073 	} else {
1074 		strncpy((char *)cbe_lun->device_id, params->device_id,
1075 			MIN(sizeof(cbe_lun->device_id),
1076 			    sizeof(params->device_id)));
1077 	}
1078 
1079 	STAILQ_INIT(&be_lun->cont_queue);
1080 	sx_init(&be_lun->page_lock, "ctlram page");
1081 	if (be_lun->cap_bytes == 0) {
1082 		be_lun->indir = 0;
1083 		be_lun->pages = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK);
1084 	}
1085 	be_lun->zero_page = malloc(be_lun->pblocksize, M_RAMDISK,
1086 	    M_WAITOK|M_ZERO);
1087 	mtx_init(&be_lun->queue_lock, "ctlram queue", NULL, MTX_DEF);
1088 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_backend_ramdisk_worker,
1089 	    be_lun);
1090 
1091 	be_lun->io_taskqueue = taskqueue_create("ctlramtq", M_WAITOK,
1092 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
1093 	if (be_lun->io_taskqueue == NULL) {
1094 		snprintf(req->error_str, sizeof(req->error_str),
1095 			 "%s: Unable to create taskqueue", __func__);
1096 		goto bailout_error;
1097 	}
1098 
1099 	retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
1100 					 /*num threads*/1,
1101 					 /*priority*/PUSER,
1102 					 /*proc*/control_softc->ctl_proc,
1103 					 /*thread name*/"ramdisk");
1104 	if (retval != 0)
1105 		goto bailout_error;
1106 
1107 	retval = ctl_add_lun(&be_lun->cbe_lun);
1108 	if (retval != 0) {
1109 		snprintf(req->error_str, sizeof(req->error_str),
1110 			 "%s: ctl_add_lun() returned error %d, see dmesg for "
1111 			"details", __func__, retval);
1112 		retval = 0;
1113 		goto bailout_error;
1114 	}
1115 
1116 	mtx_lock(&softc->lock);
1117 	softc->num_luns++;
1118 	SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
1119 	mtx_unlock(&softc->lock);
1120 
1121 	params->req_lun_id = cbe_lun->lun_id;
1122 
1123 	req->status = CTL_LUN_OK;
1124 	return (retval);
1125 
1126 bailout_error:
1127 	req->status = CTL_LUN_ERROR;
1128 	if (be_lun != NULL) {
1129 		if (be_lun->io_taskqueue != NULL)
1130 			taskqueue_free(be_lun->io_taskqueue);
1131 		nvlist_destroy(cbe_lun->options);
1132 		free(be_lun->zero_page, M_RAMDISK);
1133 		ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1134 		sx_destroy(&be_lun->page_lock);
1135 		mtx_destroy(&be_lun->queue_lock);
1136 		free(be_lun, M_RAMDISK);
1137 	}
1138 	return (retval);
1139 }
1140 
1141 static int
1142 ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
1143 		       struct ctl_lun_req *req)
1144 {
1145 	struct ctl_be_ramdisk_lun *be_lun;
1146 	struct ctl_be_lun *cbe_lun;
1147 	struct ctl_lun_modify_params *params;
1148 	const char *value;
1149 	uint32_t blocksize;
1150 	int wasprim;
1151 
1152 	params = &req->reqdata.modify;
1153 	sx_xlock(&softc->modify_lock);
1154 	mtx_lock(&softc->lock);
1155 	SLIST_FOREACH(be_lun, &softc->lun_list, links) {
1156 		if (be_lun->cbe_lun.lun_id == params->lun_id)
1157 			break;
1158 	}
1159 	mtx_unlock(&softc->lock);
1160 	if (be_lun == NULL) {
1161 		snprintf(req->error_str, sizeof(req->error_str),
1162 			 "%s: LUN %u is not managed by the ramdisk backend",
1163 			 __func__, params->lun_id);
1164 		goto bailout_error;
1165 	}
1166 	cbe_lun = &be_lun->cbe_lun;
1167 
1168 	if (params->lun_size_bytes != 0)
1169 		be_lun->params.lun_size_bytes = params->lun_size_bytes;
1170 
1171 	if (req->args_nvl != NULL) {
1172 		nvlist_destroy(cbe_lun->options);
1173 		cbe_lun->options = nvlist_clone(req->args_nvl);
1174 	}
1175 
1176 	wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
1177 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
1178 	if (value != NULL) {
1179 		if (strcmp(value, "primary") == 0)
1180 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1181 		else
1182 			cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1183 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
1184 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1185 	else
1186 		cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1187 	if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
1188 		if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
1189 			ctl_lun_primary(cbe_lun);
1190 		else
1191 			ctl_lun_secondary(cbe_lun);
1192 	}
1193 
1194 	blocksize = be_lun->cbe_lun.blocksize;
1195 	if (be_lun->params.lun_size_bytes < blocksize) {
1196 		snprintf(req->error_str, sizeof(req->error_str),
1197 			"%s: LUN size %ju < blocksize %u", __func__,
1198 			be_lun->params.lun_size_bytes, blocksize);
1199 		goto bailout_error;
1200 	}
1201 	be_lun->size_blocks = be_lun->params.lun_size_bytes / blocksize;
1202 	be_lun->size_bytes = be_lun->size_blocks * blocksize;
1203 	be_lun->cbe_lun.maxlba = be_lun->size_blocks - 1;
1204 	ctl_lun_capacity_changed(&be_lun->cbe_lun);
1205 
1206 	/* Tell the user the exact size we ended up using */
1207 	params->lun_size_bytes = be_lun->size_bytes;
1208 
1209 	sx_xunlock(&softc->modify_lock);
1210 	req->status = CTL_LUN_OK;
1211 	return (0);
1212 
1213 bailout_error:
1214 	sx_xunlock(&softc->modify_lock);
1215 	req->status = CTL_LUN_ERROR;
1216 	return (0);
1217 }
1218 
1219 static void
1220 ctl_backend_ramdisk_lun_shutdown(struct ctl_be_lun *cbe_lun)
1221 {
1222 	struct ctl_be_ramdisk_lun *be_lun = (struct ctl_be_ramdisk_lun *)cbe_lun;
1223 	struct ctl_be_ramdisk_softc *softc = be_lun->softc;
1224 
1225 	taskqueue_drain_all(be_lun->io_taskqueue);
1226 	taskqueue_free(be_lun->io_taskqueue);
1227 	nvlist_destroy(be_lun->cbe_lun.options);
1228 	free(be_lun->zero_page, M_RAMDISK);
1229 	ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1230 	sx_destroy(&be_lun->page_lock);
1231 	mtx_destroy(&be_lun->queue_lock);
1232 
1233 	mtx_lock(&softc->lock);
1234 	be_lun->flags |= CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1235 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_WAITING)
1236 		wakeup(be_lun);
1237 	else
1238 		free(be_lun, M_RAMDISK);
1239 	mtx_unlock(&softc->lock);
1240 }
1241