xref: /freebsd/sys/cam/ctl/ctl_backend_ramdisk.c (revision c697fb7f)
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_CONFIG_ERR	= 0x02,
106 	CTL_BE_RAMDISK_LUN_WAITING	= 0x04
107 } ctl_be_ramdisk_lun_flags;
108 
109 struct ctl_be_ramdisk_lun {
110 	struct ctl_lun_create_params params;
111 	char			lunname[32];
112 	int			indir;
113 	uint8_t			**pages;
114 	uint8_t			*zero_page;
115 	struct sx		page_lock;
116 	u_int			pblocksize;
117 	u_int			pblockmul;
118 	uint64_t		size_bytes;
119 	uint64_t		size_blocks;
120 	uint64_t		cap_bytes;
121 	uint64_t		cap_used;
122 	struct ctl_be_ramdisk_softc *softc;
123 	ctl_be_ramdisk_lun_flags flags;
124 	STAILQ_ENTRY(ctl_be_ramdisk_lun) links;
125 	struct ctl_be_lun	cbe_lun;
126 	struct taskqueue	*io_taskqueue;
127 	struct task		io_task;
128 	STAILQ_HEAD(, ctl_io_hdr) cont_queue;
129 	struct mtx_padalign	queue_lock;
130 };
131 
132 struct ctl_be_ramdisk_softc {
133 	struct mtx lock;
134 	int num_luns;
135 	STAILQ_HEAD(, ctl_be_ramdisk_lun) lun_list;
136 };
137 
138 static struct ctl_be_ramdisk_softc rd_softc;
139 extern struct ctl_softc *control_softc;
140 
141 static int ctl_backend_ramdisk_init(void);
142 static int ctl_backend_ramdisk_shutdown(void);
143 static int ctl_backend_ramdisk_move_done(union ctl_io *io);
144 static void ctl_backend_ramdisk_compare(union ctl_io *io);
145 static void ctl_backend_ramdisk_rw(union ctl_io *io);
146 static int ctl_backend_ramdisk_submit(union ctl_io *io);
147 static void ctl_backend_ramdisk_worker(void *context, int pending);
148 static int ctl_backend_ramdisk_config_read(union ctl_io *io);
149 static int ctl_backend_ramdisk_config_write(union ctl_io *io);
150 static uint64_t ctl_backend_ramdisk_lun_attr(void *be_lun, const char *attrname);
151 static int ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd,
152 				     caddr_t addr, int flag, struct thread *td);
153 static int ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
154 				  struct ctl_lun_req *req);
155 static int ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
156 				      struct ctl_lun_req *req);
157 static int ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
158 				  struct ctl_lun_req *req);
159 static void ctl_backend_ramdisk_lun_shutdown(void *be_lun);
160 static void ctl_backend_ramdisk_lun_config_status(void *be_lun,
161 						  ctl_lun_config_status status);
162 
163 static struct ctl_backend_driver ctl_be_ramdisk_driver =
164 {
165 	.name = "ramdisk",
166 	.flags = CTL_BE_FLAG_HAS_CONFIG,
167 	.init = ctl_backend_ramdisk_init,
168 	.shutdown = ctl_backend_ramdisk_shutdown,
169 	.data_submit = ctl_backend_ramdisk_submit,
170 	.data_move_done = ctl_backend_ramdisk_move_done,
171 	.config_read = ctl_backend_ramdisk_config_read,
172 	.config_write = ctl_backend_ramdisk_config_write,
173 	.ioctl = ctl_backend_ramdisk_ioctl,
174 	.lun_attr = ctl_backend_ramdisk_lun_attr,
175 };
176 
177 MALLOC_DEFINE(M_RAMDISK, "ramdisk", "Memory used for CTL RAMdisk");
178 CTL_BACKEND_DECLARE(cbr, ctl_be_ramdisk_driver);
179 
180 static int
181 ctl_backend_ramdisk_init(void)
182 {
183 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
184 
185 	memset(softc, 0, sizeof(*softc));
186 	mtx_init(&softc->lock, "ctlramdisk", NULL, MTX_DEF);
187 	STAILQ_INIT(&softc->lun_list);
188 	return (0);
189 }
190 
191 static int
192 ctl_backend_ramdisk_shutdown(void)
193 {
194 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
195 	struct ctl_be_ramdisk_lun *lun, *next_lun;
196 
197 	mtx_lock(&softc->lock);
198 	STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun) {
199 		/*
200 		 * Drop our lock here.  Since ctl_invalidate_lun() can call
201 		 * back into us, this could potentially lead to a recursive
202 		 * lock of the same mutex, which would cause a hang.
203 		 */
204 		mtx_unlock(&softc->lock);
205 		ctl_disable_lun(&lun->cbe_lun);
206 		ctl_invalidate_lun(&lun->cbe_lun);
207 		mtx_lock(&softc->lock);
208 	}
209 	mtx_unlock(&softc->lock);
210 	mtx_destroy(&softc->lock);
211 	return (0);
212 }
213 
214 static uint8_t *
215 ctl_backend_ramdisk_getpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn,
216     getpage_op_t op)
217 {
218 	uint8_t **p, ***pp;
219 	off_t i;
220 	int s;
221 
222 	if (be_lun->cap_bytes == 0) {
223 		switch (op) {
224 		case GP_READ:
225 			return (be_lun->zero_page);
226 		case GP_WRITE:
227 			return ((uint8_t *)be_lun->pages);
228 		case GP_ANCHOR:
229 			return (P_ANCHORED);
230 		default:
231 			return (P_UNMAPPED);
232 		}
233 	}
234 	if (op == GP_WRITE || op == GP_ANCHOR) {
235 		sx_xlock(&be_lun->page_lock);
236 		pp = &be_lun->pages;
237 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
238 			if (*pp == NULL) {
239 				*pp = malloc(PAGE_SIZE, M_RAMDISK,
240 				    M_WAITOK|M_ZERO);
241 			}
242 			i = pn >> s;
243 			pp = (uint8_t ***)&(*pp)[i];
244 			pn -= i << s;
245 		}
246 		if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
247 			if (op == GP_WRITE) {
248 				*pp = malloc(be_lun->pblocksize, M_RAMDISK,
249 				    M_WAITOK|M_ZERO);
250 			} else
251 				*pp = P_ANCHORED;
252 			be_lun->cap_used += be_lun->pblocksize;
253 		} else if (*pp == P_ANCHORED && op == GP_WRITE) {
254 			*pp = malloc(be_lun->pblocksize, M_RAMDISK,
255 			    M_WAITOK|M_ZERO);
256 		}
257 		sx_xunlock(&be_lun->page_lock);
258 		return ((uint8_t *)*pp);
259 	} else {
260 		sx_slock(&be_lun->page_lock);
261 		p = be_lun->pages;
262 		for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
263 			if (p == NULL)
264 				break;
265 			i = pn >> s;
266 			p = (uint8_t **)p[i];
267 			pn -= i << s;
268 		}
269 		sx_sunlock(&be_lun->page_lock);
270 		if ((p == P_UNMAPPED || p == P_ANCHORED) && op == GP_READ)
271 			return (be_lun->zero_page);
272 		return ((uint8_t *)p);
273 	}
274 };
275 
276 static void
277 ctl_backend_ramdisk_unmappage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
278 {
279 	uint8_t ***pp;
280 	off_t i;
281 	int s;
282 
283 	if (be_lun->cap_bytes == 0)
284 		return;
285 	sx_xlock(&be_lun->page_lock);
286 	pp = &be_lun->pages;
287 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
288 		if (*pp == NULL)
289 			goto noindir;
290 		i = pn >> s;
291 		pp = (uint8_t ***)&(*pp)[i];
292 		pn -= i << s;
293 	}
294 	if (*pp == P_ANCHORED) {
295 		be_lun->cap_used -= be_lun->pblocksize;
296 		*pp = P_UNMAPPED;
297 	} else if (*pp != P_UNMAPPED) {
298 		free(*pp, M_RAMDISK);
299 		be_lun->cap_used -= be_lun->pblocksize;
300 		*pp = P_UNMAPPED;
301 	}
302 noindir:
303 	sx_xunlock(&be_lun->page_lock);
304 };
305 
306 static void
307 ctl_backend_ramdisk_anchorpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn)
308 {
309 	uint8_t ***pp;
310 	off_t i;
311 	int s;
312 
313 	if (be_lun->cap_bytes == 0)
314 		return;
315 	sx_xlock(&be_lun->page_lock);
316 	pp = &be_lun->pages;
317 	for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) {
318 		if (*pp == NULL)
319 			goto noindir;
320 		i = pn >> s;
321 		pp = (uint8_t ***)&(*pp)[i];
322 		pn -= i << s;
323 	}
324 	if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) {
325 		be_lun->cap_used += be_lun->pblocksize;
326 		*pp = P_ANCHORED;
327 	} else if (*pp != P_ANCHORED) {
328 		free(*pp, M_RAMDISK);
329 		*pp = P_ANCHORED;
330 	}
331 noindir:
332 	sx_xunlock(&be_lun->page_lock);
333 };
334 
335 static void
336 ctl_backend_ramdisk_freeallpages(uint8_t **p, int indir)
337 {
338 	int i;
339 
340 	if (p == NULL)
341 		return;
342 	if (indir == 0) {
343 		free(p, M_RAMDISK);
344 		return;
345 	}
346 	for (i = 0; i < PPP; i++) {
347 		if (p[i] == NULL)
348 			continue;
349 		ctl_backend_ramdisk_freeallpages((uint8_t **)p[i], indir - 1);
350 	}
351 	free(p, M_RAMDISK);
352 };
353 
354 static size_t
355 cmp(uint8_t *a, uint8_t *b, size_t size)
356 {
357 	size_t i;
358 
359 	for (i = 0; i < size; i++) {
360 		if (a[i] != b[i])
361 			break;
362 	}
363 	return (i);
364 }
365 
366 static int
367 ctl_backend_ramdisk_cmp(union ctl_io *io)
368 {
369 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
370 	struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun;
371 	uint8_t *page;
372 	uint8_t info[8];
373 	uint64_t lba;
374 	u_int lbaoff, lbas, res, off;
375 
376 	lbas = io->scsiio.kern_data_len / cbe_lun->blocksize;
377 	lba = ARGS(io)->lba + PRIV(io)->len - lbas;
378 	off = 0;
379 	for (; lbas > 0; lbas--, lba++) {
380 		page = ctl_backend_ramdisk_getpage(be_lun,
381 		    lba >> cbe_lun->pblockexp, GP_READ);
382 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
383 		page += lbaoff * cbe_lun->blocksize;
384 		res = cmp(io->scsiio.kern_data_ptr + off, page,
385 		    cbe_lun->blocksize);
386 		off += res;
387 		if (res < cbe_lun->blocksize)
388 			break;
389 	}
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)
406 {
407 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
408 	struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun;
409 #ifdef CTL_TIME_IO
410 	struct bintime cur_bt;
411 #endif
412 
413 	CTL_DEBUG_PRINT(("ctl_backend_ramdisk_move_done\n"));
414 #ifdef CTL_TIME_IO
415 	getbinuptime(&cur_bt);
416 	bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
417 	bintime_add(&io->io_hdr.dma_bt, &cur_bt);
418 #endif
419 	io->io_hdr.num_dmas++;
420 	if (io->scsiio.kern_sg_entries > 0)
421 		free(io->scsiio.kern_data_ptr, M_RAMDISK);
422 	io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
423 	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
424 		;
425 	} else if (io->io_hdr.port_status != 0 &&
426 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
427 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
428 		ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
429 		    /*retry_count*/ io->io_hdr.port_status);
430 	} else if (io->scsiio.kern_data_resid != 0 &&
431 	    (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
432 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
433 	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
434 		ctl_set_invalid_field_ciu(&io->scsiio);
435 	} else if ((io->io_hdr.port_status == 0) &&
436 	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
437 		if (ARGS(io)->flags & CTL_LLF_COMPARE) {
438 			/* We have data block ready for comparison. */
439 			if (ctl_backend_ramdisk_cmp(io))
440 				goto done;
441 		}
442 		if (ARGS(io)->len > PRIV(io)->len) {
443 			mtx_lock(&be_lun->queue_lock);
444 			STAILQ_INSERT_TAIL(&be_lun->cont_queue,
445 			    &io->io_hdr, links);
446 			mtx_unlock(&be_lun->queue_lock);
447 			taskqueue_enqueue(be_lun->io_taskqueue,
448 			    &be_lun->io_task);
449 			return (0);
450 		}
451 		ctl_set_success(&io->scsiio);
452 	}
453 done:
454 	ctl_data_submit_done(io);
455 	return(0);
456 }
457 
458 static void
459 ctl_backend_ramdisk_compare(union ctl_io *io)
460 {
461 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
462 	u_int lbas, len;
463 
464 	lbas = ARGS(io)->len - PRIV(io)->len;
465 	lbas = MIN(lbas, 131072 / cbe_lun->blocksize);
466 	len = lbas * cbe_lun->blocksize;
467 
468 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
469 	io->scsiio.kern_data_ptr = malloc(len, M_RAMDISK, M_WAITOK);
470 	io->scsiio.kern_data_len = len;
471 	io->scsiio.kern_sg_entries = 0;
472 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
473 	PRIV(io)->len += lbas;
474 #ifdef CTL_TIME_IO
475 	getbinuptime(&io->io_hdr.dma_start_bt);
476 #endif
477 	ctl_datamove(io);
478 }
479 
480 static void
481 ctl_backend_ramdisk_rw(union ctl_io *io)
482 {
483 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
484 	struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun;
485 	struct ctl_sg_entry *sg_entries;
486 	uint8_t *page;
487 	uint64_t lba;
488 	u_int i, len, lbaoff, lbas, sgs, off;
489 	getpage_op_t op;
490 
491 	lba = ARGS(io)->lba + PRIV(io)->len;
492 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
493 	lbas = ARGS(io)->len - PRIV(io)->len;
494 	lbas = MIN(lbas, (SGPP << cbe_lun->pblockexp) - lbaoff);
495 	sgs = (lbas + lbaoff + be_lun->pblockmul - 1) >> cbe_lun->pblockexp;
496 	off = lbaoff * cbe_lun->blocksize;
497 	op = (ARGS(io)->flags & CTL_LLF_WRITE) ? GP_WRITE : GP_READ;
498 	if (sgs > 1) {
499 		io->scsiio.kern_data_ptr = malloc(sizeof(struct ctl_sg_entry) *
500 		    sgs, M_RAMDISK, M_WAITOK);
501 		sg_entries = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
502 		len = lbas * cbe_lun->blocksize;
503 		for (i = 0; i < sgs; i++) {
504 			page = ctl_backend_ramdisk_getpage(be_lun,
505 			    (lba >> cbe_lun->pblockexp) + i, op);
506 			if (page == P_UNMAPPED || page == P_ANCHORED) {
507 				free(io->scsiio.kern_data_ptr, M_RAMDISK);
508 nospc:
509 				ctl_set_space_alloc_fail(&io->scsiio);
510 				ctl_data_submit_done(io);
511 				return;
512 			}
513 			sg_entries[i].addr = page + off;
514 			sg_entries[i].len = MIN(len, be_lun->pblocksize - off);
515 			len -= sg_entries[i].len;
516 			off = 0;
517 		}
518 	} else {
519 		page = ctl_backend_ramdisk_getpage(be_lun,
520 		    lba >> cbe_lun->pblockexp, op);
521 		if (page == P_UNMAPPED || page == P_ANCHORED)
522 			goto nospc;
523 		sgs = 0;
524 		io->scsiio.kern_data_ptr = page + off;
525 	}
526 
527 	io->scsiio.be_move_done = ctl_backend_ramdisk_move_done;
528 	io->scsiio.kern_data_len = lbas * cbe_lun->blocksize;
529 	io->scsiio.kern_sg_entries = sgs;
530 	io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
531 	PRIV(io)->len += lbas;
532 	if ((ARGS(io)->flags & CTL_LLF_READ) &&
533 	    ARGS(io)->len <= PRIV(io)->len) {
534 		ctl_set_success(&io->scsiio);
535 		ctl_serseq_done(io);
536 	}
537 #ifdef CTL_TIME_IO
538 	getbinuptime(&io->io_hdr.dma_start_bt);
539 #endif
540 	ctl_datamove(io);
541 }
542 
543 static int
544 ctl_backend_ramdisk_submit(union ctl_io *io)
545 {
546 	struct ctl_lba_len_flags *lbalen = ARGS(io);
547 
548 	if (lbalen->flags & CTL_LLF_VERIFY) {
549 		ctl_set_success(&io->scsiio);
550 		ctl_data_submit_done(io);
551 		return (CTL_RETVAL_COMPLETE);
552 	}
553 	PRIV(io)->len = 0;
554 	if (lbalen->flags & CTL_LLF_COMPARE)
555 		ctl_backend_ramdisk_compare(io);
556 	else
557 		ctl_backend_ramdisk_rw(io);
558 	return (CTL_RETVAL_COMPLETE);
559 }
560 
561 static void
562 ctl_backend_ramdisk_worker(void *context, int pending)
563 {
564 	struct ctl_be_ramdisk_lun *be_lun;
565 	union ctl_io *io;
566 
567 	be_lun = (struct ctl_be_ramdisk_lun *)context;
568 	mtx_lock(&be_lun->queue_lock);
569 	for (;;) {
570 		io = (union ctl_io *)STAILQ_FIRST(&be_lun->cont_queue);
571 		if (io != NULL) {
572 			STAILQ_REMOVE(&be_lun->cont_queue, &io->io_hdr,
573 				      ctl_io_hdr, links);
574 			mtx_unlock(&be_lun->queue_lock);
575 			if (ARGS(io)->flags & CTL_LLF_COMPARE)
576 				ctl_backend_ramdisk_compare(io);
577 			else
578 				ctl_backend_ramdisk_rw(io);
579 			mtx_lock(&be_lun->queue_lock);
580 			continue;
581 		}
582 
583 		/*
584 		 * If we get here, there is no work left in the queues, so
585 		 * just break out and let the task queue go to sleep.
586 		 */
587 		break;
588 	}
589 	mtx_unlock(&be_lun->queue_lock);
590 }
591 
592 static int
593 ctl_backend_ramdisk_gls(union ctl_io *io)
594 {
595 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
596 	struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun;
597 	struct scsi_get_lba_status_data *data;
598 	uint8_t *page;
599 	u_int lbaoff;
600 
601 	data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
602 	scsi_u64to8b(ARGS(io)->lba, data->descr[0].addr);
603 	lbaoff = ARGS(io)->lba & ~(UINT_MAX << cbe_lun->pblockexp);
604 	scsi_ulto4b(be_lun->pblockmul - lbaoff, data->descr[0].length);
605 	page = ctl_backend_ramdisk_getpage(be_lun,
606 	    ARGS(io)->lba >> cbe_lun->pblockexp, GP_OTHER);
607 	if (page == P_UNMAPPED)
608 		data->descr[0].status = 1;
609 	else if (page == P_ANCHORED)
610 		data->descr[0].status = 2;
611 	else
612 		data->descr[0].status = 0;
613 	ctl_config_read_done(io);
614 	return (CTL_RETVAL_COMPLETE);
615 }
616 
617 static int
618 ctl_backend_ramdisk_config_read(union ctl_io *io)
619 {
620 	int retval = 0;
621 
622 	switch (io->scsiio.cdb[0]) {
623 	case SERVICE_ACTION_IN:
624 		if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
625 			retval = ctl_backend_ramdisk_gls(io);
626 			break;
627 		}
628 		ctl_set_invalid_field(&io->scsiio,
629 				      /*sks_valid*/ 1,
630 				      /*command*/ 1,
631 				      /*field*/ 1,
632 				      /*bit_valid*/ 1,
633 				      /*bit*/ 4);
634 		ctl_config_read_done(io);
635 		retval = CTL_RETVAL_COMPLETE;
636 		break;
637 	default:
638 		ctl_set_invalid_opcode(&io->scsiio);
639 		ctl_config_read_done(io);
640 		retval = CTL_RETVAL_COMPLETE;
641 		break;
642 	}
643 	return (retval);
644 }
645 
646 static void
647 ctl_backend_ramdisk_delete(struct ctl_be_lun *cbe_lun, off_t lba, off_t len,
648     int anchor)
649 {
650 	struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun;
651 	uint8_t *page;
652 	uint64_t p, lp;
653 	u_int lbaoff;
654 	getpage_op_t op = anchor ? GP_ANCHOR : GP_OTHER;
655 
656 	/* Partially zero first partial page. */
657 	p = lba >> cbe_lun->pblockexp;
658 	lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
659 	if (lbaoff != 0) {
660 		page = ctl_backend_ramdisk_getpage(be_lun, p, op);
661 		if (page != P_UNMAPPED && page != P_ANCHORED) {
662 			memset(page + lbaoff * cbe_lun->blocksize, 0,
663 			    min(len, be_lun->pblockmul - lbaoff) *
664 			    cbe_lun->blocksize);
665 		}
666 		p++;
667 	}
668 
669 	/* Partially zero last partial page. */
670 	lp = (lba + len) >> cbe_lun->pblockexp;
671 	lbaoff = (lba + len) & ~(UINT_MAX << cbe_lun->pblockexp);
672 	if (p <= lp && lbaoff != 0) {
673 		page = ctl_backend_ramdisk_getpage(be_lun, lp, op);
674 		if (page != P_UNMAPPED && page != P_ANCHORED)
675 			memset(page, 0, lbaoff * cbe_lun->blocksize);
676 	}
677 
678 	/* Delete remaining full pages. */
679 	if (anchor) {
680 		for (; p < lp; p++)
681 			ctl_backend_ramdisk_anchorpage(be_lun, p);
682 	} else {
683 		for (; p < lp; p++)
684 			ctl_backend_ramdisk_unmappage(be_lun, p);
685 	}
686 }
687 
688 static void
689 ctl_backend_ramdisk_ws(union ctl_io *io)
690 {
691 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
692 	struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun;
693 	struct ctl_lba_len_flags *lbalen = ARGS(io);
694 	uint8_t *page;
695 	uint64_t lba;
696 	u_int lbaoff, lbas;
697 
698 	if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB)) {
699 		ctl_set_invalid_field(&io->scsiio,
700 				      /*sks_valid*/ 1,
701 				      /*command*/ 1,
702 				      /*field*/ 1,
703 				      /*bit_valid*/ 0,
704 				      /*bit*/ 0);
705 		ctl_config_write_done(io);
706 		return;
707 	}
708 	if (lbalen->flags & SWS_UNMAP) {
709 		ctl_backend_ramdisk_delete(cbe_lun, lbalen->lba, lbalen->len,
710 		    (lbalen->flags & SWS_ANCHOR) != 0);
711 		ctl_set_success(&io->scsiio);
712 		ctl_config_write_done(io);
713 		return;
714 	}
715 
716 	for (lba = lbalen->lba, lbas = lbalen->len; lbas > 0; lba++, lbas--) {
717 		page = ctl_backend_ramdisk_getpage(be_lun,
718 		    lba >> cbe_lun->pblockexp, GP_WRITE);
719 		if (page == P_UNMAPPED || page == P_ANCHORED) {
720 			ctl_set_space_alloc_fail(&io->scsiio);
721 			ctl_data_submit_done(io);
722 			return;
723 		}
724 		lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp);
725 		page += lbaoff * cbe_lun->blocksize;
726 		if (lbalen->flags & SWS_NDOB) {
727 			memset(page, 0, cbe_lun->blocksize);
728 		} else {
729 			memcpy(page, io->scsiio.kern_data_ptr,
730 			    cbe_lun->blocksize);
731 		}
732 		if (lbalen->flags & SWS_LBDATA)
733 			scsi_ulto4b(lba, page);
734 	}
735 	ctl_set_success(&io->scsiio);
736 	ctl_config_write_done(io);
737 }
738 
739 static void
740 ctl_backend_ramdisk_unmap(union ctl_io *io)
741 {
742 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
743 	struct ctl_ptr_len_flags *ptrlen = (struct ctl_ptr_len_flags *)ARGS(io);
744 	struct scsi_unmap_desc *buf, *end;
745 
746 	if ((ptrlen->flags & ~SU_ANCHOR) != 0) {
747 		ctl_set_invalid_field(&io->scsiio,
748 				      /*sks_valid*/ 0,
749 				      /*command*/ 0,
750 				      /*field*/ 0,
751 				      /*bit_valid*/ 0,
752 				      /*bit*/ 0);
753 		ctl_config_write_done(io);
754 		return;
755 	}
756 
757 	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
758 	end = buf + ptrlen->len / sizeof(*buf);
759 	for (; buf < end; buf++) {
760 		ctl_backend_ramdisk_delete(cbe_lun,
761 		    scsi_8btou64(buf->lba), scsi_4btoul(buf->length),
762 		    (ptrlen->flags & SU_ANCHOR) != 0);
763 	}
764 
765 	ctl_set_success(&io->scsiio);
766 	ctl_config_write_done(io);
767 }
768 
769 static int
770 ctl_backend_ramdisk_config_write(union ctl_io *io)
771 {
772 	struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io);
773 	int retval = 0;
774 
775 	switch (io->scsiio.cdb[0]) {
776 	case SYNCHRONIZE_CACHE:
777 	case SYNCHRONIZE_CACHE_16:
778 		/* We have no cache to flush. */
779 		ctl_set_success(&io->scsiio);
780 		ctl_config_write_done(io);
781 		break;
782 	case START_STOP_UNIT: {
783 		struct scsi_start_stop_unit *cdb;
784 
785 		cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
786 		if ((cdb->how & SSS_PC_MASK) != 0) {
787 			ctl_set_success(&io->scsiio);
788 			ctl_config_write_done(io);
789 			break;
790 		}
791 		if (cdb->how & SSS_START) {
792 			if (cdb->how & SSS_LOEJ)
793 				ctl_lun_has_media(cbe_lun);
794 			ctl_start_lun(cbe_lun);
795 		} else {
796 			ctl_stop_lun(cbe_lun);
797 			if (cdb->how & SSS_LOEJ)
798 				ctl_lun_ejected(cbe_lun);
799 		}
800 		ctl_set_success(&io->scsiio);
801 		ctl_config_write_done(io);
802 		break;
803 	}
804 	case PREVENT_ALLOW:
805 		ctl_set_success(&io->scsiio);
806 		ctl_config_write_done(io);
807 		break;
808 	case WRITE_SAME_10:
809 	case WRITE_SAME_16:
810 		ctl_backend_ramdisk_ws(io);
811 		break;
812 	case UNMAP:
813 		ctl_backend_ramdisk_unmap(io);
814 		break;
815 	default:
816 		ctl_set_invalid_opcode(&io->scsiio);
817 		ctl_config_write_done(io);
818 		retval = CTL_RETVAL_COMPLETE;
819 		break;
820 	}
821 
822 	return (retval);
823 }
824 
825 static uint64_t
826 ctl_backend_ramdisk_lun_attr(void *arg, const char *attrname)
827 {
828 	struct ctl_be_ramdisk_lun *be_lun = arg;
829 	uint64_t		val;
830 
831 	val = UINT64_MAX;
832 	if (be_lun->cap_bytes == 0)
833 		return (val);
834 	sx_slock(&be_lun->page_lock);
835 	if (strcmp(attrname, "blocksused") == 0) {
836 		val = be_lun->cap_used / be_lun->cbe_lun.blocksize;
837 	} else if (strcmp(attrname, "blocksavail") == 0) {
838 		val = (be_lun->cap_bytes - be_lun->cap_used) /
839 		    be_lun->cbe_lun.blocksize;
840 	}
841 	sx_sunlock(&be_lun->page_lock);
842 	return (val);
843 }
844 
845 static int
846 ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
847 			  int flag, struct thread *td)
848 {
849 	struct ctl_be_ramdisk_softc *softc = &rd_softc;
850 	struct ctl_lun_req *lun_req;
851 	int retval;
852 
853 	retval = 0;
854 	switch (cmd) {
855 	case CTL_LUN_REQ:
856 		lun_req = (struct ctl_lun_req *)addr;
857 		switch (lun_req->reqtype) {
858 		case CTL_LUNREQ_CREATE:
859 			retval = ctl_backend_ramdisk_create(softc, lun_req);
860 			break;
861 		case CTL_LUNREQ_RM:
862 			retval = ctl_backend_ramdisk_rm(softc, lun_req);
863 			break;
864 		case CTL_LUNREQ_MODIFY:
865 			retval = ctl_backend_ramdisk_modify(softc, lun_req);
866 			break;
867 		default:
868 			lun_req->status = CTL_LUN_ERROR;
869 			snprintf(lun_req->error_str, sizeof(lun_req->error_str),
870 				 "%s: invalid LUN request type %d", __func__,
871 				 lun_req->reqtype);
872 			break;
873 		}
874 		break;
875 	default:
876 		retval = ENOTTY;
877 		break;
878 	}
879 
880 	return (retval);
881 }
882 
883 static int
884 ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc,
885 		       struct ctl_lun_req *req)
886 {
887 	struct ctl_be_ramdisk_lun *be_lun;
888 	struct ctl_lun_rm_params *params;
889 	int retval;
890 
891 	params = &req->reqdata.rm;
892 	mtx_lock(&softc->lock);
893 	STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
894 		if (be_lun->cbe_lun.lun_id == params->lun_id)
895 			break;
896 	}
897 	mtx_unlock(&softc->lock);
898 	if (be_lun == NULL) {
899 		snprintf(req->error_str, sizeof(req->error_str),
900 			 "%s: LUN %u is not managed by the ramdisk backend",
901 			 __func__, params->lun_id);
902 		goto bailout_error;
903 	}
904 
905 	retval = ctl_disable_lun(&be_lun->cbe_lun);
906 	if (retval != 0) {
907 		snprintf(req->error_str, sizeof(req->error_str),
908 			 "%s: error %d returned from ctl_disable_lun() for "
909 			 "LUN %d", __func__, retval, params->lun_id);
910 		goto bailout_error;
911 	}
912 
913 	/*
914 	 * Set the waiting flag before we invalidate the LUN.  Our shutdown
915 	 * routine can be called any time after we invalidate the LUN,
916 	 * and can be called from our context.
917 	 *
918 	 * This tells the shutdown routine that we're waiting, or we're
919 	 * going to wait for the shutdown to happen.
920 	 */
921 	mtx_lock(&softc->lock);
922 	be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING;
923 	mtx_unlock(&softc->lock);
924 
925 	retval = ctl_invalidate_lun(&be_lun->cbe_lun);
926 	if (retval != 0) {
927 		snprintf(req->error_str, sizeof(req->error_str),
928 			 "%s: error %d returned from ctl_invalidate_lun() for "
929 			 "LUN %d", __func__, retval, params->lun_id);
930 		mtx_lock(&softc->lock);
931 		be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
932 		mtx_unlock(&softc->lock);
933 		goto bailout_error;
934 	}
935 
936 	mtx_lock(&softc->lock);
937 	while ((be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) == 0) {
938 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlram", 0);
939 		if (retval == EINTR)
940 			break;
941 	}
942 	be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
943 
944 	/*
945 	 * We only remove this LUN from the list and free it (below) if
946 	 * retval == 0.  If the user interrupted the wait, we just bail out
947 	 * without actually freeing the LUN.  We let the shutdown routine
948 	 * free the LUN if that happens.
949 	 */
950 	if (retval == 0) {
951 		STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_ramdisk_lun,
952 			      links);
953 		softc->num_luns--;
954 	}
955 
956 	mtx_unlock(&softc->lock);
957 
958 	if (retval == 0) {
959 		taskqueue_drain_all(be_lun->io_taskqueue);
960 		taskqueue_free(be_lun->io_taskqueue);
961 		nvlist_destroy(be_lun->cbe_lun.options);
962 		free(be_lun->zero_page, M_RAMDISK);
963 		ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
964 		sx_destroy(&be_lun->page_lock);
965 		mtx_destroy(&be_lun->queue_lock);
966 		free(be_lun, M_RAMDISK);
967 	}
968 
969 	req->status = CTL_LUN_OK;
970 	return (retval);
971 
972 bailout_error:
973 	req->status = CTL_LUN_ERROR;
974 	return (0);
975 }
976 
977 static int
978 ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc,
979 			   struct ctl_lun_req *req)
980 {
981 	struct ctl_be_ramdisk_lun *be_lun;
982 	struct ctl_be_lun *cbe_lun;
983 	struct ctl_lun_create_params *params;
984 	const char *value;
985 	char tmpstr[32];
986 	uint64_t t;
987 	int retval;
988 
989 	retval = 0;
990 	params = &req->reqdata.create;
991 
992 	be_lun = malloc(sizeof(*be_lun), M_RAMDISK, M_ZERO | M_WAITOK);
993 	cbe_lun = &be_lun->cbe_lun;
994 	cbe_lun->be_lun = be_lun;
995 	cbe_lun->options = nvlist_clone(req->args_nvl);
996 	be_lun->params = req->reqdata.create;
997 	be_lun->softc = softc;
998 	sprintf(be_lun->lunname, "cram%d", softc->num_luns);
999 
1000 	if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
1001 		cbe_lun->lun_type = params->device_type;
1002 	else
1003 		cbe_lun->lun_type = T_DIRECT;
1004 	be_lun->flags = CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1005 	cbe_lun->flags = 0;
1006 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
1007 	if (value != NULL) {
1008 		if (strcmp(value, "primary") == 0)
1009 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1010 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
1011 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1012 
1013 	be_lun->pblocksize = PAGE_SIZE;
1014 	value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
1015 	if (value != NULL) {
1016 		ctl_expand_number(value, &t);
1017 		be_lun->pblocksize = t;
1018 	}
1019 	if (be_lun->pblocksize < 512 || be_lun->pblocksize > 131072) {
1020 		snprintf(req->error_str, sizeof(req->error_str),
1021 			 "%s: unsupported pblocksize %u", __func__,
1022 			 be_lun->pblocksize);
1023 		goto bailout_error;
1024 	}
1025 
1026 	if (cbe_lun->lun_type == T_DIRECT ||
1027 	    cbe_lun->lun_type == T_CDROM) {
1028 		if (params->blocksize_bytes != 0)
1029 			cbe_lun->blocksize = params->blocksize_bytes;
1030 		else if (cbe_lun->lun_type == T_CDROM)
1031 			cbe_lun->blocksize = 2048;
1032 		else
1033 			cbe_lun->blocksize = 512;
1034 		be_lun->pblockmul = be_lun->pblocksize / cbe_lun->blocksize;
1035 		if (be_lun->pblockmul < 1 || !powerof2(be_lun->pblockmul)) {
1036 			snprintf(req->error_str, sizeof(req->error_str),
1037 				 "%s: pblocksize %u not exp2 of blocksize %u",
1038 				 __func__,
1039 				 be_lun->pblocksize, cbe_lun->blocksize);
1040 			goto bailout_error;
1041 		}
1042 		if (params->lun_size_bytes < cbe_lun->blocksize) {
1043 			snprintf(req->error_str, sizeof(req->error_str),
1044 				 "%s: LUN size %ju < blocksize %u", __func__,
1045 				 params->lun_size_bytes, cbe_lun->blocksize);
1046 			goto bailout_error;
1047 		}
1048 		be_lun->size_blocks = params->lun_size_bytes / cbe_lun->blocksize;
1049 		be_lun->size_bytes = be_lun->size_blocks * cbe_lun->blocksize;
1050 		be_lun->indir = 0;
1051 		t = be_lun->size_bytes / be_lun->pblocksize;
1052 		while (t > 1) {
1053 			t /= PPP;
1054 			be_lun->indir++;
1055 		}
1056 		cbe_lun->maxlba = be_lun->size_blocks - 1;
1057 		cbe_lun->pblockexp = fls(be_lun->pblockmul) - 1;
1058 		cbe_lun->pblockoff = 0;
1059 		cbe_lun->ublockexp = cbe_lun->pblockexp;
1060 		cbe_lun->ublockoff = 0;
1061 		cbe_lun->atomicblock = be_lun->pblocksize;
1062 		cbe_lun->opttxferlen = SGPP * be_lun->pblocksize;
1063 		value = dnvlist_get_string(cbe_lun->options, "capacity", NULL);
1064 		if (value != NULL)
1065 			ctl_expand_number(value, &be_lun->cap_bytes);
1066 	} else {
1067 		be_lun->pblockmul = 1;
1068 		cbe_lun->pblockexp = 0;
1069 	}
1070 
1071 	/* Tell the user the blocksize we ended up using */
1072 	params->blocksize_bytes = cbe_lun->blocksize;
1073 	params->lun_size_bytes = be_lun->size_bytes;
1074 
1075 	value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
1076 	if (value == NULL || strcmp(value, "off") != 0)
1077 		cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
1078 	value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
1079 	if (value != NULL) {
1080 		if (strcmp(value, "on") == 0)
1081 			cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1082 	} else if (cbe_lun->lun_type != T_DIRECT)
1083 		cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
1084 	cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1085 	value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
1086 	if (value != NULL && strcmp(value, "on") == 0)
1087 		cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
1088 	else if (value != NULL && strcmp(value, "read") == 0)
1089 		cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
1090 	else if (value != NULL && strcmp(value, "off") == 0)
1091 		cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
1092 
1093 	if (params->flags & CTL_LUN_FLAG_ID_REQ) {
1094 		cbe_lun->req_lun_id = params->req_lun_id;
1095 		cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
1096 	} else
1097 		cbe_lun->req_lun_id = 0;
1098 
1099 	cbe_lun->lun_shutdown = ctl_backend_ramdisk_lun_shutdown;
1100 	cbe_lun->lun_config_status = ctl_backend_ramdisk_lun_config_status;
1101 	cbe_lun->be = &ctl_be_ramdisk_driver;
1102 	if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
1103 		snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
1104 			 softc->num_luns);
1105 		strncpy((char *)cbe_lun->serial_num, tmpstr,
1106 			MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
1107 
1108 		/* Tell the user what we used for a serial number */
1109 		strncpy((char *)params->serial_num, tmpstr,
1110 			MIN(sizeof(params->serial_num), sizeof(tmpstr)));
1111 	} else {
1112 		strncpy((char *)cbe_lun->serial_num, params->serial_num,
1113 			MIN(sizeof(cbe_lun->serial_num),
1114 			    sizeof(params->serial_num)));
1115 	}
1116 	if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
1117 		snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
1118 		strncpy((char *)cbe_lun->device_id, tmpstr,
1119 			MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
1120 
1121 		/* Tell the user what we used for a device ID */
1122 		strncpy((char *)params->device_id, tmpstr,
1123 			MIN(sizeof(params->device_id), sizeof(tmpstr)));
1124 	} else {
1125 		strncpy((char *)cbe_lun->device_id, params->device_id,
1126 			MIN(sizeof(cbe_lun->device_id),
1127 			    sizeof(params->device_id)));
1128 	}
1129 
1130 	STAILQ_INIT(&be_lun->cont_queue);
1131 	sx_init(&be_lun->page_lock, "cram page lock");
1132 	if (be_lun->cap_bytes == 0) {
1133 		be_lun->indir = 0;
1134 		be_lun->pages = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK);
1135 	}
1136 	be_lun->zero_page = malloc(be_lun->pblocksize, M_RAMDISK,
1137 	    M_WAITOK|M_ZERO);
1138 	mtx_init(&be_lun->queue_lock, "cram queue lock", NULL, MTX_DEF);
1139 	TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_backend_ramdisk_worker,
1140 	    be_lun);
1141 
1142 	be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK,
1143 	    taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
1144 	if (be_lun->io_taskqueue == NULL) {
1145 		snprintf(req->error_str, sizeof(req->error_str),
1146 			 "%s: Unable to create taskqueue", __func__);
1147 		goto bailout_error;
1148 	}
1149 
1150 	retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
1151 					 /*num threads*/1,
1152 					 /*priority*/PUSER,
1153 					 /*proc*/control_softc->ctl_proc,
1154 					 /*thread name*/
1155 					 "%s taskq", be_lun->lunname);
1156 	if (retval != 0)
1157 		goto bailout_error;
1158 
1159 	mtx_lock(&softc->lock);
1160 	softc->num_luns++;
1161 	STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links);
1162 	mtx_unlock(&softc->lock);
1163 
1164 	retval = ctl_add_lun(&be_lun->cbe_lun);
1165 	if (retval != 0) {
1166 		mtx_lock(&softc->lock);
1167 		STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_ramdisk_lun,
1168 			      links);
1169 		softc->num_luns--;
1170 		mtx_unlock(&softc->lock);
1171 		snprintf(req->error_str, sizeof(req->error_str),
1172 			 "%s: ctl_add_lun() returned error %d, see dmesg for "
1173 			"details", __func__, retval);
1174 		retval = 0;
1175 		goto bailout_error;
1176 	}
1177 
1178 	mtx_lock(&softc->lock);
1179 
1180 	/*
1181 	 * Tell the config_status routine that we're waiting so it won't
1182 	 * clean up the LUN in the event of an error.
1183 	 */
1184 	be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING;
1185 
1186 	while (be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) {
1187 		retval = msleep(be_lun, &softc->lock, PCATCH, "ctlram", 0);
1188 		if (retval == EINTR)
1189 			break;
1190 	}
1191 	be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING;
1192 
1193 	if (be_lun->flags & CTL_BE_RAMDISK_LUN_CONFIG_ERR) {
1194 		snprintf(req->error_str, sizeof(req->error_str),
1195 			 "%s: LUN configuration error, see dmesg for details",
1196 			 __func__);
1197 		STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_ramdisk_lun,
1198 			      links);
1199 		softc->num_luns--;
1200 		mtx_unlock(&softc->lock);
1201 		goto bailout_error;
1202 	} else {
1203 		params->req_lun_id = cbe_lun->lun_id;
1204 	}
1205 	mtx_unlock(&softc->lock);
1206 
1207 	req->status = CTL_LUN_OK;
1208 	return (retval);
1209 
1210 bailout_error:
1211 	req->status = CTL_LUN_ERROR;
1212 	if (be_lun != NULL) {
1213 		if (be_lun->io_taskqueue != NULL)
1214 			taskqueue_free(be_lun->io_taskqueue);
1215 		nvlist_destroy(cbe_lun->options);
1216 		free(be_lun->zero_page, M_RAMDISK);
1217 		ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir);
1218 		sx_destroy(&be_lun->page_lock);
1219 		mtx_destroy(&be_lun->queue_lock);
1220 		free(be_lun, M_RAMDISK);
1221 	}
1222 	return (retval);
1223 }
1224 
1225 static int
1226 ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc,
1227 		       struct ctl_lun_req *req)
1228 {
1229 	struct ctl_be_ramdisk_lun *be_lun;
1230 	struct ctl_be_lun *cbe_lun;
1231 	struct ctl_lun_modify_params *params;
1232 	const char *value;
1233 	uint32_t blocksize;
1234 	int wasprim;
1235 
1236 	params = &req->reqdata.modify;
1237 
1238 	mtx_lock(&softc->lock);
1239 	STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
1240 		if (be_lun->cbe_lun.lun_id == params->lun_id)
1241 			break;
1242 	}
1243 	mtx_unlock(&softc->lock);
1244 	if (be_lun == NULL) {
1245 		snprintf(req->error_str, sizeof(req->error_str),
1246 			 "%s: LUN %u is not managed by the ramdisk backend",
1247 			 __func__, params->lun_id);
1248 		goto bailout_error;
1249 	}
1250 	cbe_lun = &be_lun->cbe_lun;
1251 
1252 	if (params->lun_size_bytes != 0)
1253 		be_lun->params.lun_size_bytes = params->lun_size_bytes;
1254 
1255 	nvlist_destroy(cbe_lun->options);
1256 	cbe_lun->options = nvlist_clone(req->args_nvl);
1257 
1258 	wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
1259 	value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
1260 	if (value != NULL) {
1261 		if (strcmp(value, "primary") == 0)
1262 			cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1263 		else
1264 			cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1265 	} else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
1266 		cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
1267 	else
1268 		cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
1269 	if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
1270 		if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
1271 			ctl_lun_primary(cbe_lun);
1272 		else
1273 			ctl_lun_secondary(cbe_lun);
1274 	}
1275 
1276 	blocksize = be_lun->cbe_lun.blocksize;
1277 	if (be_lun->params.lun_size_bytes < blocksize) {
1278 		snprintf(req->error_str, sizeof(req->error_str),
1279 			"%s: LUN size %ju < blocksize %u", __func__,
1280 			be_lun->params.lun_size_bytes, blocksize);
1281 		goto bailout_error;
1282 	}
1283 	be_lun->size_blocks = be_lun->params.lun_size_bytes / blocksize;
1284 	be_lun->size_bytes = be_lun->size_blocks * blocksize;
1285 	be_lun->cbe_lun.maxlba = be_lun->size_blocks - 1;
1286 	ctl_lun_capacity_changed(&be_lun->cbe_lun);
1287 
1288 	/* Tell the user the exact size we ended up using */
1289 	params->lun_size_bytes = be_lun->size_bytes;
1290 
1291 	req->status = CTL_LUN_OK;
1292 	return (0);
1293 
1294 bailout_error:
1295 	req->status = CTL_LUN_ERROR;
1296 	return (0);
1297 }
1298 
1299 static void
1300 ctl_backend_ramdisk_lun_shutdown(void *be_lun)
1301 {
1302 	struct ctl_be_ramdisk_lun *lun = be_lun;
1303 	struct ctl_be_ramdisk_softc *softc = lun->softc;
1304 
1305 	mtx_lock(&softc->lock);
1306 	lun->flags |= CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1307 	if (lun->flags & CTL_BE_RAMDISK_LUN_WAITING) {
1308 		wakeup(lun);
1309 	} else {
1310 		STAILQ_REMOVE(&softc->lun_list, lun, ctl_be_ramdisk_lun,
1311 			      links);
1312 		softc->num_luns--;
1313 		free(be_lun, M_RAMDISK);
1314 	}
1315 	mtx_unlock(&softc->lock);
1316 }
1317 
1318 static void
1319 ctl_backend_ramdisk_lun_config_status(void *be_lun,
1320 				      ctl_lun_config_status status)
1321 {
1322 	struct ctl_be_ramdisk_lun *lun;
1323 	struct ctl_be_ramdisk_softc *softc;
1324 
1325 	lun = (struct ctl_be_ramdisk_lun *)be_lun;
1326 	softc = lun->softc;
1327 
1328 	if (status == CTL_LUN_CONFIG_OK) {
1329 		mtx_lock(&softc->lock);
1330 		lun->flags &= ~CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1331 		if (lun->flags & CTL_BE_RAMDISK_LUN_WAITING)
1332 			wakeup(lun);
1333 		mtx_unlock(&softc->lock);
1334 
1335 		/*
1336 		 * We successfully added the LUN, attempt to enable it.
1337 		 */
1338 		if (ctl_enable_lun(&lun->cbe_lun) != 0) {
1339 			printf("%s: ctl_enable_lun() failed!\n", __func__);
1340 			if (ctl_invalidate_lun(&lun->cbe_lun) != 0) {
1341 				printf("%s: ctl_invalidate_lun() failed!\n",
1342 				       __func__);
1343 			}
1344 		}
1345 
1346 		return;
1347 	}
1348 
1349 
1350 	mtx_lock(&softc->lock);
1351 	lun->flags &= ~CTL_BE_RAMDISK_LUN_UNCONFIGURED;
1352 
1353 	/*
1354 	 * If we have a user waiting, let him handle the cleanup.  If not,
1355 	 * clean things up here.
1356 	 */
1357 	if (lun->flags & CTL_BE_RAMDISK_LUN_WAITING) {
1358 		lun->flags |= CTL_BE_RAMDISK_LUN_CONFIG_ERR;
1359 		wakeup(lun);
1360 	} else {
1361 		STAILQ_REMOVE(&softc->lun_list, lun, ctl_be_ramdisk_lun,
1362 			      links);
1363 		softc->num_luns--;
1364 		free(lun, M_RAMDISK);
1365 	}
1366 	mtx_unlock(&softc->lock);
1367 }
1368