1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2003 Silicon Graphics International Corp.
5 * Copyright (c) 2009-2011 Spectra Logic Corporation
6 * Copyright (c) 2012,2021 The FreeBSD Foundation
7 * Copyright (c) 2014-2021 Alexander Motin <mav@FreeBSD.org>
8 * All rights reserved.
9 *
10 * Portions of this software were developed by Edward Tomasz Napierala
11 * under sponsorship from the FreeBSD Foundation.
12 *
13 * Portions of this software were developed by Ka Ho Ng <khng@FreeBSD.org>
14 * under sponsorship from the FreeBSD Foundation.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions, and the following disclaimer,
21 * without modification.
22 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
23 * substantially similar to the "NO WARRANTY" disclaimer below
24 * ("Disclaimer") and any redistribution must be conditioned upon
25 * including a substantially similar Disclaimer requirement for further
26 * binary redistribution.
27 *
28 * NO WARRANTY
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
37 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
38 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
39 * POSSIBILITY OF SUCH DAMAGES.
40 *
41 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $
42 */
43 /*
44 * CAM Target Layer driver backend for block devices.
45 *
46 * Author: Ken Merry <ken@FreeBSD.org>
47 */
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/types.h>
52 #include <sys/kthread.h>
53 #include <sys/bio.h>
54 #include <sys/fcntl.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/condvar.h>
59 #include <sys/malloc.h>
60 #include <sys/conf.h>
61 #include <sys/ioccom.h>
62 #include <sys/queue.h>
63 #include <sys/sbuf.h>
64 #include <sys/endian.h>
65 #include <sys/uio.h>
66 #include <sys/buf.h>
67 #include <sys/taskqueue.h>
68 #include <sys/vnode.h>
69 #include <sys/namei.h>
70 #include <sys/mount.h>
71 #include <sys/disk.h>
72 #include <sys/fcntl.h>
73 #include <sys/filedesc.h>
74 #include <sys/filio.h>
75 #include <sys/proc.h>
76 #include <sys/pcpu.h>
77 #include <sys/module.h>
78 #include <sys/sdt.h>
79 #include <sys/devicestat.h>
80 #include <sys/sysctl.h>
81 #include <sys/nv.h>
82 #include <sys/dnv.h>
83 #include <sys/sx.h>
84 #include <sys/unistd.h>
85
86 #include <geom/geom.h>
87
88 #include <cam/cam.h>
89 #include <cam/scsi/scsi_all.h>
90 #include <cam/scsi/scsi_da.h>
91 #include <cam/ctl/ctl_io.h>
92 #include <cam/ctl/ctl.h>
93 #include <cam/ctl/ctl_backend.h>
94 #include <cam/ctl/ctl_ioctl.h>
95 #include <cam/ctl/ctl_ha.h>
96 #include <cam/ctl/ctl_scsi_all.h>
97 #include <cam/ctl/ctl_private.h>
98 #include <cam/ctl/ctl_error.h>
99
100 /*
101 * The idea here is to allocate enough S/G space to handle at least 1MB I/Os.
102 * On systems with small maxphys it can be 8 128KB segments. On large systems
103 * it can be up to 8 1MB segments. I/Os larger than that we'll split.
104 */
105 #define CTLBLK_MAX_SEGS 8
106 #define CTLBLK_HALF_SEGS (CTLBLK_MAX_SEGS / 2)
107 #define CTLBLK_MIN_SEG (128 * 1024)
108 #define CTLBLK_MAX_SEG MIN(1024 * 1024, MAX(CTLBLK_MIN_SEG, maxphys))
109 #define CTLBLK_MAX_IO_SIZE (CTLBLK_MAX_SEG * CTLBLK_MAX_SEGS)
110
111 #ifdef CTLBLK_DEBUG
112 #define DPRINTF(fmt, args...) \
113 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
114 #else
115 #define DPRINTF(fmt, args...) do {} while(0)
116 #endif
117
118 #define PRIV(io) \
119 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
120 #define ARGS(io) \
121 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
122 #define DSM_RANGE(io) ((io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].integer)
123
124 SDT_PROVIDER_DEFINE(cbb);
125
126 typedef enum {
127 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01,
128 CTL_BE_BLOCK_LUN_WAITING = 0x04,
129 } ctl_be_block_lun_flags;
130
131 typedef enum {
132 CTL_BE_BLOCK_NONE,
133 CTL_BE_BLOCK_DEV,
134 CTL_BE_BLOCK_FILE
135 } ctl_be_block_type;
136
137 struct ctl_be_block_filedata {
138 struct ucred *cred;
139 };
140
141 union ctl_be_block_bedata {
142 struct ctl_be_block_filedata file;
143 };
144
145 struct ctl_be_block_io;
146 struct ctl_be_block_lun;
147
148 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
149 struct ctl_be_block_io *beio);
150 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
151 const char *attrname);
152
153 /*
154 * Backend LUN structure. There is a 1:1 mapping between a block device
155 * and a backend block LUN, and between a backend block LUN and a CTL LUN.
156 */
157 struct ctl_be_block_lun {
158 struct ctl_be_lun cbe_lun; /* Must be first element. */
159 struct ctl_lun_create_params params;
160 char *dev_path;
161 ctl_be_block_type dev_type;
162 struct vnode *vn;
163 union ctl_be_block_bedata backend;
164 cbb_dispatch_t dispatch;
165 cbb_dispatch_t lun_flush;
166 cbb_dispatch_t unmap;
167 cbb_dispatch_t get_lba_status;
168 cbb_getattr_t getattr;
169 uint64_t size_blocks;
170 uint64_t size_bytes;
171 struct ctl_be_block_softc *softc;
172 struct devstat *disk_stats;
173 ctl_be_block_lun_flags flags;
174 SLIST_ENTRY(ctl_be_block_lun) links;
175 struct taskqueue *io_taskqueue;
176 struct task io_task;
177 int num_threads;
178 STAILQ_HEAD(, ctl_io_hdr) input_queue;
179 STAILQ_HEAD(, ctl_io_hdr) config_read_queue;
180 STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
181 STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
182 struct mtx_padalign io_lock;
183 struct mtx_padalign queue_lock;
184 };
185
186 /*
187 * Overall softc structure for the block backend module.
188 */
189 struct ctl_be_block_softc {
190 struct sx modify_lock;
191 struct mtx lock;
192 int num_luns;
193 SLIST_HEAD(, ctl_be_block_lun) lun_list;
194 uma_zone_t beio_zone;
195 uma_zone_t bufmin_zone;
196 uma_zone_t bufmax_zone;
197 };
198
199 static struct ctl_be_block_softc backend_block_softc;
200
201 /*
202 * Per-I/O information.
203 */
204 struct ctl_be_block_io {
205 union ctl_io *io;
206 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS];
207 struct iovec xiovecs[CTLBLK_MAX_SEGS];
208 int refcnt;
209 int bio_cmd;
210 int two_sglists;
211 int num_segs;
212 int num_bios_sent;
213 int num_bios_done;
214 int send_complete;
215 int first_error;
216 uint64_t first_error_offset;
217 struct bintime ds_t0;
218 devstat_tag_type ds_tag_type;
219 devstat_trans_flags ds_trans_type;
220 uint64_t io_len;
221 uint64_t io_offset;
222 int io_arg;
223 struct ctl_be_block_softc *softc;
224 struct ctl_be_block_lun *lun;
225 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
226 };
227
228 extern struct ctl_softc *control_softc;
229
230 static int cbb_num_threads = 32;
231 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
232 "CAM Target Layer Block Backend");
233 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
234 &cbb_num_threads, 0, "Number of threads per backing file");
235
236 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
237 static void ctl_free_beio(struct ctl_be_block_io *beio);
238 static void ctl_complete_beio(struct ctl_be_block_io *beio);
239 static int ctl_be_block_move_done(union ctl_io *io, bool samethr);
240 static void ctl_be_block_biodone(struct bio *bio);
241 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
242 struct ctl_be_block_io *beio);
243 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
244 struct ctl_be_block_io *beio);
245 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
246 struct ctl_be_block_io *beio);
247 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
248 const char *attrname);
249 static void ctl_be_block_unmap_file(struct ctl_be_block_lun *be_lun,
250 struct ctl_be_block_io *beio);
251 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
252 struct ctl_be_block_io *beio);
253 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
254 struct ctl_be_block_io *beio);
255 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
256 struct ctl_be_block_io *beio);
257 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
258 const char *attrname);
259 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
260 union ctl_io *io);
261 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
262 union ctl_io *io);
263 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
264 union ctl_io *io);
265 static void ctl_be_block_worker(void *context, int pending);
266 static int ctl_be_block_submit(union ctl_io *io);
267 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
268 int flag, struct thread *td);
269 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
270 struct ctl_lun_req *req);
271 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
272 struct ctl_lun_req *req);
273 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
274 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun,
275 struct ctl_lun_req *req);
276 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
277 struct ctl_lun_req *req);
278 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
279 struct ctl_lun_req *req);
280 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
281 struct ctl_lun_req *req);
282 static void ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun);
283 static int ctl_be_block_config_write(union ctl_io *io);
284 static int ctl_be_block_config_read(union ctl_io *io);
285 static int ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb);
286 static uint64_t ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname);
287 static int ctl_be_block_init(void);
288 static int ctl_be_block_shutdown(void);
289
290 static struct ctl_backend_driver ctl_be_block_driver =
291 {
292 .name = "block",
293 .flags = CTL_BE_FLAG_HAS_CONFIG,
294 .init = ctl_be_block_init,
295 .shutdown = ctl_be_block_shutdown,
296 .data_submit = ctl_be_block_submit,
297 .config_read = ctl_be_block_config_read,
298 .config_write = ctl_be_block_config_write,
299 .ioctl = ctl_be_block_ioctl,
300 .lun_info = ctl_be_block_lun_info,
301 .lun_attr = ctl_be_block_lun_attr
302 };
303
304 MALLOC_DEFINE(M_CTLBLK, "ctlblock", "Memory used for CTL block backend");
305 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
306
307 static void
ctl_alloc_seg(struct ctl_be_block_softc * softc,struct ctl_sg_entry * sg,size_t len)308 ctl_alloc_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg,
309 size_t len)
310 {
311
312 if (len <= CTLBLK_MIN_SEG) {
313 sg->addr = uma_zalloc(softc->bufmin_zone, M_WAITOK);
314 } else {
315 KASSERT(len <= CTLBLK_MAX_SEG,
316 ("Too large alloc %zu > %lu", len, CTLBLK_MAX_SEG));
317 sg->addr = uma_zalloc(softc->bufmax_zone, M_WAITOK);
318 }
319 sg->len = len;
320 }
321
322 static void
ctl_free_seg(struct ctl_be_block_softc * softc,struct ctl_sg_entry * sg)323 ctl_free_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg)
324 {
325
326 if (sg->len <= CTLBLK_MIN_SEG) {
327 uma_zfree(softc->bufmin_zone, sg->addr);
328 } else {
329 KASSERT(sg->len <= CTLBLK_MAX_SEG,
330 ("Too large free %zu > %lu", sg->len, CTLBLK_MAX_SEG));
331 uma_zfree(softc->bufmax_zone, sg->addr);
332 }
333 }
334
335 static struct ctl_be_block_io *
ctl_alloc_beio(struct ctl_be_block_softc * softc)336 ctl_alloc_beio(struct ctl_be_block_softc *softc)
337 {
338 struct ctl_be_block_io *beio;
339
340 beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO);
341 beio->softc = softc;
342 beio->refcnt = 1;
343 return (beio);
344 }
345
346 static void
ctl_real_free_beio(struct ctl_be_block_io * beio)347 ctl_real_free_beio(struct ctl_be_block_io *beio)
348 {
349 struct ctl_be_block_softc *softc = beio->softc;
350 int i;
351
352 for (i = 0; i < beio->num_segs; i++) {
353 ctl_free_seg(softc, &beio->sg_segs[i]);
354
355 /* For compare we had two equal S/G lists. */
356 if (beio->two_sglists) {
357 ctl_free_seg(softc,
358 &beio->sg_segs[i + CTLBLK_HALF_SEGS]);
359 }
360 }
361
362 uma_zfree(softc->beio_zone, beio);
363 }
364
365 static void
ctl_refcnt_beio(void * arg,int diff)366 ctl_refcnt_beio(void *arg, int diff)
367 {
368 struct ctl_be_block_io *beio = arg;
369
370 if (atomic_fetchadd_int(&beio->refcnt, diff) + diff == 0)
371 ctl_real_free_beio(beio);
372 }
373
374 static void
ctl_free_beio(struct ctl_be_block_io * beio)375 ctl_free_beio(struct ctl_be_block_io *beio)
376 {
377
378 ctl_refcnt_beio(beio, -1);
379 }
380
381 static void
ctl_complete_beio(struct ctl_be_block_io * beio)382 ctl_complete_beio(struct ctl_be_block_io *beio)
383 {
384 union ctl_io *io = beio->io;
385
386 if (beio->beio_cont != NULL) {
387 beio->beio_cont(beio);
388 } else {
389 ctl_free_beio(beio);
390 ctl_data_submit_done(io);
391 }
392 }
393
394 static void
ctl_be_block_io_error(union ctl_io * io,int bio_cmd,uint16_t retry_count)395 ctl_be_block_io_error(union ctl_io *io, int bio_cmd, uint16_t retry_count)
396 {
397 switch (io->io_hdr.io_type) {
398 case CTL_IO_SCSI:
399 if (bio_cmd == BIO_FLUSH) {
400 /* XXX KDM is there is a better error here? */
401 ctl_set_internal_failure(&io->scsiio,
402 /*sks_valid*/ 1,
403 retry_count);
404 } else {
405 ctl_set_medium_error(&io->scsiio, bio_cmd == BIO_READ);
406 }
407 break;
408 case CTL_IO_NVME:
409 switch (bio_cmd) {
410 case BIO_FLUSH:
411 case BIO_WRITE:
412 ctl_nvme_set_write_fault(&io->nvmeio);
413 break;
414 case BIO_READ:
415 ctl_nvme_set_unrecoverable_read_error(&io->nvmeio);
416 break;
417 default:
418 ctl_nvme_set_internal_error(&io->nvmeio);
419 break;
420 }
421 break;
422 default:
423 __assert_unreachable();
424 }
425 }
426
427 static size_t
cmp(uint8_t * a,uint8_t * b,size_t size)428 cmp(uint8_t *a, uint8_t *b, size_t size)
429 {
430 size_t i;
431
432 for (i = 0; i < size; i++) {
433 if (a[i] != b[i])
434 break;
435 }
436 return (i);
437 }
438
439 static void
ctl_be_block_compare(union ctl_io * io)440 ctl_be_block_compare(union ctl_io *io)
441 {
442 struct ctl_be_block_io *beio;
443 uint64_t off, res;
444 int i;
445
446 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
447 off = 0;
448 for (i = 0; i < beio->num_segs; i++) {
449 res = cmp(beio->sg_segs[i].addr,
450 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
451 beio->sg_segs[i].len);
452 off += res;
453 if (res < beio->sg_segs[i].len)
454 break;
455 }
456 if (i < beio->num_segs) {
457 ctl_io_set_compare_failure(io, off);
458 } else
459 ctl_io_set_success(io);
460 }
461
462 static int
ctl_be_block_move_done(union ctl_io * io,bool samethr)463 ctl_be_block_move_done(union ctl_io *io, bool samethr)
464 {
465 struct ctl_be_block_io *beio;
466 struct ctl_be_block_lun *be_lun;
467 struct ctl_lba_len_flags *lbalen;
468
469 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
470
471 DPRINTF("entered\n");
472 ctl_add_kern_rel_offset(io, ctl_kern_data_len(io));
473
474 /*
475 * We set status at this point for read and compare commands.
476 */
477 if ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
478 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE) {
479 lbalen = ARGS(io);
480 if (lbalen->flags & CTL_LLF_READ) {
481 ctl_io_set_success(io);
482 } else if (lbalen->flags & CTL_LLF_COMPARE) {
483 /* We have two data blocks ready for comparison. */
484 ctl_be_block_compare(io);
485 }
486 }
487
488 /*
489 * If this is a read, or a write with errors, it is done.
490 */
491 if ((beio->bio_cmd == BIO_READ)
492 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
493 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
494 ctl_complete_beio(beio);
495 return (0);
496 }
497
498 /*
499 * At this point, we have a write and the DMA completed successfully.
500 * If we were called synchronously in the original thread then just
501 * dispatch, otherwise we now have to queue it to the task queue to
502 * execute the backend I/O. That is because we do blocking
503 * memory allocations, and in the file backing case, blocking I/O.
504 * This move done routine is generally called in the SIM's
505 * interrupt context, and therefore we cannot block.
506 */
507 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
508 if (samethr) {
509 be_lun->dispatch(be_lun, beio);
510 } else {
511 mtx_lock(&be_lun->queue_lock);
512 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
513 mtx_unlock(&be_lun->queue_lock);
514 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
515 }
516 return (0);
517 }
518
519 static void
ctl_be_block_biodone(struct bio * bio)520 ctl_be_block_biodone(struct bio *bio)
521 {
522 struct ctl_be_block_io *beio = bio->bio_caller1;
523 struct ctl_be_block_lun *be_lun = beio->lun;
524 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
525 union ctl_io *io;
526 int error;
527
528 io = beio->io;
529
530 DPRINTF("entered\n");
531
532 error = bio->bio_error;
533 mtx_lock(&be_lun->io_lock);
534 if (error != 0 &&
535 (beio->first_error == 0 ||
536 bio->bio_offset < beio->first_error_offset)) {
537 beio->first_error = error;
538 beio->first_error_offset = bio->bio_offset;
539 }
540
541 beio->num_bios_done++;
542
543 /*
544 * XXX KDM will this cause WITNESS to complain? Holding a lock
545 * during the free might cause it to complain.
546 */
547 g_destroy_bio(bio);
548
549 /*
550 * If the send complete bit isn't set, or we aren't the last I/O to
551 * complete, then we're done.
552 */
553 if ((beio->send_complete == 0)
554 || (beio->num_bios_done < beio->num_bios_sent)) {
555 mtx_unlock(&be_lun->io_lock);
556 return;
557 }
558
559 /*
560 * At this point, we've verified that we are the last I/O to
561 * complete, so it's safe to drop the lock.
562 */
563 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
564 beio->ds_tag_type, beio->ds_trans_type,
565 /*now*/ NULL, /*then*/&beio->ds_t0);
566 mtx_unlock(&be_lun->io_lock);
567
568 /*
569 * If there are any errors from the backing device, we fail the
570 * entire I/O with a medium error.
571 */
572 error = beio->first_error;
573 if (error != 0) {
574 if (error == EOPNOTSUPP) {
575 ctl_io_set_invalid_opcode(io);
576 } else if (error == ENOSPC || error == EDQUOT) {
577 ctl_io_set_space_alloc_fail(io);
578 } else if (error == EROFS || error == EACCES) {
579 ctl_io_set_hw_write_protected(io);
580 } else {
581 ctl_be_block_io_error(io, beio->bio_cmd,
582 /*retry_count*/ 0xbad2);
583 }
584 ctl_complete_beio(beio);
585 return;
586 }
587
588 /*
589 * If this is a write, a flush, a delete or verify, we're all done.
590 * If this is a read, we can now send the data to the user.
591 */
592 if ((beio->bio_cmd == BIO_WRITE)
593 || (beio->bio_cmd == BIO_FLUSH)
594 || (beio->bio_cmd == BIO_DELETE)
595 || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
596 ctl_io_set_success(io);
597 ctl_complete_beio(beio);
598 } else {
599 if ((ARGS(io)->flags & CTL_LLF_READ) &&
600 beio->beio_cont == NULL) {
601 ctl_io_set_success(io);
602 if (cbe_lun->serseq >= CTL_LUN_SERSEQ_SOFT)
603 ctl_serseq_done(io);
604 }
605 ctl_datamove(io);
606 }
607 }
608
609 static void
ctl_be_block_flush_file(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)610 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
611 struct ctl_be_block_io *beio)
612 {
613 union ctl_io *io = beio->io;
614 struct mount *mountpoint;
615 int error;
616
617 DPRINTF("entered\n");
618
619 binuptime(&beio->ds_t0);
620 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
621
622 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
623
624 vn_lock(be_lun->vn, vn_lktype_write(mountpoint, be_lun->vn) |
625 LK_RETRY);
626 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT,
627 curthread);
628 VOP_UNLOCK(be_lun->vn);
629
630 vn_finished_write(mountpoint);
631
632 mtx_lock(&be_lun->io_lock);
633 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
634 beio->ds_tag_type, beio->ds_trans_type,
635 /*now*/ NULL, /*then*/&beio->ds_t0);
636 mtx_unlock(&be_lun->io_lock);
637
638 if (error == 0)
639 ctl_io_set_success(io);
640 else {
641 ctl_be_block_io_error(io, BIO_FLUSH,
642 /*retry_count*/ 0xbad1);
643 }
644
645 ctl_complete_beio(beio);
646 }
647
648 SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t");
649 SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t");
650 SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t");
651 SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t");
652
653 static void
ctl_be_block_dispatch_file(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)654 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
655 struct ctl_be_block_io *beio)
656 {
657 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
658 struct ctl_be_block_filedata *file_data;
659 union ctl_io *io;
660 struct uio xuio;
661 struct iovec *xiovec;
662 size_t s;
663 int error, flags, i;
664
665 DPRINTF("entered\n");
666
667 file_data = &be_lun->backend.file;
668 io = beio->io;
669 flags = 0;
670 if (ARGS(io)->flags & CTL_LLF_DPO)
671 flags |= IO_DIRECT;
672 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
673 flags |= IO_SYNC;
674
675 bzero(&xuio, sizeof(xuio));
676 if (beio->bio_cmd == BIO_READ) {
677 SDT_PROBE0(cbb, , read, file_start);
678 xuio.uio_rw = UIO_READ;
679 } else {
680 SDT_PROBE0(cbb, , write, file_start);
681 xuio.uio_rw = UIO_WRITE;
682 }
683 xuio.uio_offset = beio->io_offset;
684 xuio.uio_resid = beio->io_len;
685 xuio.uio_segflg = UIO_SYSSPACE;
686 xuio.uio_iov = beio->xiovecs;
687 xuio.uio_iovcnt = beio->num_segs;
688 xuio.uio_td = curthread;
689
690 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
691 xiovec->iov_base = beio->sg_segs[i].addr;
692 xiovec->iov_len = beio->sg_segs[i].len;
693 }
694
695 binuptime(&beio->ds_t0);
696 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
697
698 if (beio->bio_cmd == BIO_READ) {
699 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
700
701 if (beio->beio_cont == NULL &&
702 cbe_lun->serseq == CTL_LUN_SERSEQ_SOFT)
703 ctl_serseq_done(io);
704 /*
705 * UFS pays attention to IO_DIRECT for reads. If the
706 * DIRECTIO option is configured into the kernel, it calls
707 * ffs_rawread(). But that only works for single-segment
708 * uios with user space addresses. In our case, with a
709 * kernel uio, it still reads into the buffer cache, but it
710 * will just try to release the buffer from the cache later
711 * on in ffs_read().
712 *
713 * ZFS does not pay attention to IO_DIRECT for reads.
714 *
715 * UFS does not pay attention to IO_SYNC for reads.
716 *
717 * ZFS pays attention to IO_SYNC (which translates into the
718 * Solaris define FRSYNC for zfs_read()) for reads. It
719 * attempts to sync the file before reading.
720 */
721 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
722
723 VOP_UNLOCK(be_lun->vn);
724 SDT_PROBE0(cbb, , read, file_done);
725 if (error == 0 && xuio.uio_resid > 0) {
726 /*
727 * If we read less then requested (EOF), then
728 * we should clean the rest of the buffer.
729 */
730 s = beio->io_len - xuio.uio_resid;
731 for (i = 0; i < beio->num_segs; i++) {
732 if (s >= beio->sg_segs[i].len) {
733 s -= beio->sg_segs[i].len;
734 continue;
735 }
736 bzero((uint8_t *)beio->sg_segs[i].addr + s,
737 beio->sg_segs[i].len - s);
738 s = 0;
739 }
740 }
741 } else {
742 struct mount *mountpoint;
743
744 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
745 vn_lock(be_lun->vn, vn_lktype_write(mountpoint,
746 be_lun->vn) | LK_RETRY);
747
748 /*
749 * UFS pays attention to IO_DIRECT for writes. The write
750 * is done asynchronously. (Normally the write would just
751 * get put into cache.
752 *
753 * UFS pays attention to IO_SYNC for writes. It will
754 * attempt to write the buffer out synchronously if that
755 * flag is set.
756 *
757 * ZFS does not pay attention to IO_DIRECT for writes.
758 *
759 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
760 * for writes. It will flush the transaction from the
761 * cache before returning.
762 */
763 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
764 VOP_UNLOCK(be_lun->vn);
765
766 vn_finished_write(mountpoint);
767 SDT_PROBE0(cbb, , write, file_done);
768 }
769
770 mtx_lock(&be_lun->io_lock);
771 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
772 beio->ds_tag_type, beio->ds_trans_type,
773 /*now*/ NULL, /*then*/&beio->ds_t0);
774 mtx_unlock(&be_lun->io_lock);
775
776 /*
777 * If we got an error, set the sense data to "MEDIUM ERROR" and
778 * return the I/O to the user.
779 */
780 if (error != 0) {
781 if (error == ENOSPC || error == EDQUOT) {
782 ctl_io_set_space_alloc_fail(io);
783 } else if (error == EROFS || error == EACCES) {
784 ctl_io_set_hw_write_protected(io);
785 } else {
786 ctl_be_block_io_error(io, beio->bio_cmd, 0);
787 }
788 ctl_complete_beio(beio);
789 return;
790 }
791
792 /*
793 * If this is a write or a verify, we're all done.
794 * If this is a read, we can now send the data to the user.
795 */
796 if ((beio->bio_cmd == BIO_WRITE) ||
797 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
798 ctl_io_set_success(io);
799 ctl_complete_beio(beio);
800 } else {
801 if ((ARGS(io)->flags & CTL_LLF_READ) &&
802 beio->beio_cont == NULL) {
803 ctl_io_set_success(io);
804 if (cbe_lun->serseq > CTL_LUN_SERSEQ_SOFT)
805 ctl_serseq_done(io);
806 }
807 ctl_datamove(io);
808 }
809 }
810
811 static void
ctl_be_block_gls_file(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)812 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
813 struct ctl_be_block_io *beio)
814 {
815 union ctl_io *io = beio->io;
816 struct ctl_lba_len_flags *lbalen = ARGS(io);
817 struct scsi_get_lba_status_data *data;
818 off_t roff, off;
819 int error, status;
820
821 DPRINTF("entered\n");
822
823 CTL_IO_ASSERT(io, SCSI);
824
825 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
826 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
827 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
828 0, curthread->td_ucred, curthread);
829 if (error == 0 && off > roff)
830 status = 0; /* mapped up to off */
831 else {
832 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
833 0, curthread->td_ucred, curthread);
834 if (error == 0 && off > roff)
835 status = 1; /* deallocated up to off */
836 else {
837 status = 0; /* unknown up to the end */
838 off = be_lun->size_bytes;
839 }
840 }
841 VOP_UNLOCK(be_lun->vn);
842
843 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
844 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
845 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
846 lbalen->lba), data->descr[0].length);
847 data->descr[0].status = status;
848
849 ctl_complete_beio(beio);
850 }
851
852 static uint64_t
ctl_be_block_getattr_file(struct ctl_be_block_lun * be_lun,const char * attrname)853 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
854 {
855 struct vattr vattr;
856 struct statfs statfs;
857 uint64_t val;
858 int error;
859
860 val = UINT64_MAX;
861 if (be_lun->vn == NULL)
862 return (val);
863 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
864 if (strcmp(attrname, "blocksused") == 0) {
865 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
866 if (error == 0)
867 val = vattr.va_bytes / be_lun->cbe_lun.blocksize;
868 }
869 if (strcmp(attrname, "blocksavail") == 0 &&
870 !VN_IS_DOOMED(be_lun->vn)) {
871 error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
872 if (error == 0)
873 val = statfs.f_bavail * statfs.f_bsize /
874 be_lun->cbe_lun.blocksize;
875 }
876 VOP_UNLOCK(be_lun->vn);
877 return (val);
878 }
879
880 static void
ctl_be_block_unmap_file(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)881 ctl_be_block_unmap_file(struct ctl_be_block_lun *be_lun,
882 struct ctl_be_block_io *beio)
883 {
884 struct ctl_be_block_filedata *file_data;
885 union ctl_io *io;
886 struct ctl_ptr_len_flags *ptrlen;
887 struct scsi_unmap_desc *buf, *end;
888 struct mount *mp;
889 off_t off, len;
890 int error;
891
892 io = beio->io;
893 file_data = &be_lun->backend.file;
894 mp = NULL;
895 error = 0;
896
897 binuptime(&beio->ds_t0);
898 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
899
900 (void)vn_start_write(be_lun->vn, &mp, V_WAIT);
901 vn_lock(be_lun->vn, vn_lktype_write(mp, be_lun->vn) | LK_RETRY);
902 if (beio->io_offset == -1) {
903 beio->io_len = 0;
904 ptrlen = (struct ctl_ptr_len_flags *)
905 &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
906 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
907 end = buf + ptrlen->len / sizeof(*buf);
908 for (; buf < end; buf++) {
909 off = (off_t)scsi_8btou64(buf->lba) *
910 be_lun->cbe_lun.blocksize;
911 len = (off_t)scsi_4btoul(buf->length) *
912 be_lun->cbe_lun.blocksize;
913 beio->io_len += len;
914 error = vn_deallocate(be_lun->vn, &off, &len,
915 0, IO_NOMACCHECK | IO_NODELOCKED, file_data->cred,
916 NOCRED);
917 if (error != 0)
918 break;
919 }
920 } else {
921 /* WRITE_SAME */
922 off = beio->io_offset;
923 len = beio->io_len;
924 error = vn_deallocate(be_lun->vn, &off, &len, 0,
925 IO_NOMACCHECK | IO_NODELOCKED, file_data->cred, NOCRED);
926 }
927 VOP_UNLOCK(be_lun->vn);
928 vn_finished_write(mp);
929
930 mtx_lock(&be_lun->io_lock);
931 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
932 beio->ds_tag_type, beio->ds_trans_type,
933 /*now*/ NULL, /*then*/&beio->ds_t0);
934 mtx_unlock(&be_lun->io_lock);
935
936 /*
937 * If we got an error, set the sense data to "MEDIUM ERROR" and
938 * return the I/O to the user.
939 */
940 switch (error) {
941 case 0:
942 ctl_io_set_success(io);
943 break;
944 case ENOSPC:
945 case EDQUOT:
946 ctl_io_set_space_alloc_fail(io);
947 break;
948 case EROFS:
949 case EACCES:
950 ctl_io_set_hw_write_protected(io);
951 break;
952 default:
953 ctl_be_block_io_error(io, BIO_DELETE, 0);
954 }
955 ctl_complete_beio(beio);
956 }
957
958 static void
ctl_be_block_dispatch_zvol(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)959 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
960 struct ctl_be_block_io *beio)
961 {
962 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
963 union ctl_io *io;
964 struct cdevsw *csw;
965 struct cdev *dev;
966 struct uio xuio;
967 struct iovec *xiovec;
968 int error, flags, i, ref;
969
970 DPRINTF("entered\n");
971
972 io = beio->io;
973 flags = 0;
974 if (ARGS(io)->flags & CTL_LLF_DPO)
975 flags |= IO_DIRECT;
976 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
977 flags |= IO_SYNC;
978
979 bzero(&xuio, sizeof(xuio));
980 if (beio->bio_cmd == BIO_READ) {
981 SDT_PROBE0(cbb, , read, file_start);
982 xuio.uio_rw = UIO_READ;
983 } else {
984 SDT_PROBE0(cbb, , write, file_start);
985 xuio.uio_rw = UIO_WRITE;
986 }
987 xuio.uio_offset = beio->io_offset;
988 xuio.uio_resid = beio->io_len;
989 xuio.uio_segflg = UIO_SYSSPACE;
990 xuio.uio_iov = beio->xiovecs;
991 xuio.uio_iovcnt = beio->num_segs;
992 xuio.uio_td = curthread;
993
994 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
995 xiovec->iov_base = beio->sg_segs[i].addr;
996 xiovec->iov_len = beio->sg_segs[i].len;
997 }
998
999 binuptime(&beio->ds_t0);
1000 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
1001
1002 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1003 if (csw) {
1004 if (beio->bio_cmd == BIO_READ) {
1005 if (beio->beio_cont == NULL &&
1006 cbe_lun->serseq == CTL_LUN_SERSEQ_SOFT)
1007 ctl_serseq_done(io);
1008 error = csw->d_read(dev, &xuio, flags);
1009 } else
1010 error = csw->d_write(dev, &xuio, flags);
1011 dev_relthread(dev, ref);
1012 } else
1013 error = ENXIO;
1014
1015 if (beio->bio_cmd == BIO_READ)
1016 SDT_PROBE0(cbb, , read, file_done);
1017 else
1018 SDT_PROBE0(cbb, , write, file_done);
1019
1020 mtx_lock(&be_lun->io_lock);
1021 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
1022 beio->ds_tag_type, beio->ds_trans_type,
1023 /*now*/ NULL, /*then*/&beio->ds_t0);
1024 mtx_unlock(&be_lun->io_lock);
1025
1026 /*
1027 * If we got an error, set the sense data to "MEDIUM ERROR" and
1028 * return the I/O to the user.
1029 */
1030 if (error != 0) {
1031 if (error == ENOSPC || error == EDQUOT) {
1032 ctl_io_set_space_alloc_fail(io);
1033 } else if (error == EROFS || error == EACCES) {
1034 ctl_io_set_hw_write_protected(io);
1035 } else {
1036 ctl_be_block_io_error(io, beio->bio_cmd, 0);
1037 }
1038 ctl_complete_beio(beio);
1039 return;
1040 }
1041
1042 /*
1043 * If this is a write or a verify, we're all done.
1044 * If this is a read, we can now send the data to the user.
1045 */
1046 if ((beio->bio_cmd == BIO_WRITE) ||
1047 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
1048 ctl_io_set_success(io);
1049 ctl_complete_beio(beio);
1050 } else {
1051 if ((ARGS(io)->flags & CTL_LLF_READ) &&
1052 beio->beio_cont == NULL) {
1053 ctl_io_set_success(io);
1054 if (cbe_lun->serseq > CTL_LUN_SERSEQ_SOFT)
1055 ctl_serseq_done(io);
1056 }
1057 ctl_datamove(io);
1058 }
1059 }
1060
1061 static void
ctl_be_block_gls_zvol(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)1062 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
1063 struct ctl_be_block_io *beio)
1064 {
1065 union ctl_io *io = beio->io;
1066 struct cdevsw *csw;
1067 struct cdev *dev;
1068 struct ctl_lba_len_flags *lbalen = ARGS(io);
1069 struct scsi_get_lba_status_data *data;
1070 off_t roff, off;
1071 int error, ref, status;
1072
1073 DPRINTF("entered\n");
1074
1075 CTL_IO_ASSERT(io, SCSI);
1076
1077 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1078 if (csw == NULL) {
1079 status = 0; /* unknown up to the end */
1080 off = be_lun->size_bytes;
1081 goto done;
1082 }
1083 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
1084 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD,
1085 curthread);
1086 if (error == 0 && off > roff)
1087 status = 0; /* mapped up to off */
1088 else {
1089 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD,
1090 curthread);
1091 if (error == 0 && off > roff)
1092 status = 1; /* deallocated up to off */
1093 else {
1094 status = 0; /* unknown up to the end */
1095 off = be_lun->size_bytes;
1096 }
1097 }
1098 dev_relthread(dev, ref);
1099
1100 done:
1101 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
1102 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
1103 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
1104 lbalen->lba), data->descr[0].length);
1105 data->descr[0].status = status;
1106
1107 ctl_complete_beio(beio);
1108 }
1109
1110 static void
ctl_be_block_flush_dev(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)1111 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
1112 struct ctl_be_block_io *beio)
1113 {
1114 struct bio *bio;
1115 struct cdevsw *csw;
1116 struct cdev *dev;
1117 int ref;
1118
1119 DPRINTF("entered\n");
1120
1121 /* This can't fail, it's a blocking allocation. */
1122 bio = g_alloc_bio();
1123
1124 bio->bio_cmd = BIO_FLUSH;
1125 bio->bio_offset = 0;
1126 bio->bio_data = 0;
1127 bio->bio_done = ctl_be_block_biodone;
1128 bio->bio_caller1 = beio;
1129 bio->bio_pblkno = 0;
1130
1131 /*
1132 * We don't need to acquire the LUN lock here, because we are only
1133 * sending one bio, and so there is no other context to synchronize
1134 * with.
1135 */
1136 beio->num_bios_sent = 1;
1137 beio->send_complete = 1;
1138
1139 binuptime(&beio->ds_t0);
1140 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1141
1142 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1143 if (csw) {
1144 bio->bio_dev = dev;
1145 csw->d_strategy(bio);
1146 dev_relthread(dev, ref);
1147 } else {
1148 bio->bio_error = ENXIO;
1149 ctl_be_block_biodone(bio);
1150 }
1151 }
1152
1153 static void
ctl_be_block_unmap_dev_range(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio,uint64_t off,uint64_t len,int last)1154 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1155 struct ctl_be_block_io *beio,
1156 uint64_t off, uint64_t len, int last)
1157 {
1158 struct bio *bio;
1159 uint64_t maxlen;
1160 struct cdevsw *csw;
1161 struct cdev *dev;
1162 int ref;
1163
1164 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1165 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize);
1166 while (len > 0) {
1167 bio = g_alloc_bio();
1168 bio->bio_cmd = BIO_DELETE;
1169 bio->bio_dev = dev;
1170 bio->bio_offset = off;
1171 bio->bio_length = MIN(len, maxlen);
1172 bio->bio_data = 0;
1173 bio->bio_done = ctl_be_block_biodone;
1174 bio->bio_caller1 = beio;
1175 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize;
1176
1177 off += bio->bio_length;
1178 len -= bio->bio_length;
1179
1180 mtx_lock(&be_lun->io_lock);
1181 beio->num_bios_sent++;
1182 if (last && len == 0)
1183 beio->send_complete = 1;
1184 mtx_unlock(&be_lun->io_lock);
1185
1186 if (csw) {
1187 csw->d_strategy(bio);
1188 } else {
1189 bio->bio_error = ENXIO;
1190 ctl_be_block_biodone(bio);
1191 }
1192 }
1193 if (csw)
1194 dev_relthread(dev, ref);
1195 }
1196
1197 static void
ctl_be_block_unmap_dev(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)1198 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1199 struct ctl_be_block_io *beio)
1200 {
1201 union ctl_io *io;
1202 struct ctl_ptr_len_flags *ptrlen;
1203 struct scsi_unmap_desc *buf, *end;
1204 uint64_t len;
1205
1206 io = beio->io;
1207
1208 DPRINTF("entered\n");
1209
1210 binuptime(&beio->ds_t0);
1211 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1212
1213 if (beio->io_offset == -1) {
1214 beio->io_len = 0;
1215 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1216 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1217 end = buf + ptrlen->len / sizeof(*buf);
1218 for (; buf < end; buf++) {
1219 len = (uint64_t)scsi_4btoul(buf->length) *
1220 be_lun->cbe_lun.blocksize;
1221 beio->io_len += len;
1222 ctl_be_block_unmap_dev_range(be_lun, beio,
1223 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize,
1224 len, (end - buf < 2) ? TRUE : FALSE);
1225 }
1226 } else
1227 ctl_be_block_unmap_dev_range(be_lun, beio,
1228 beio->io_offset, beio->io_len, TRUE);
1229 }
1230
1231 static void
ctl_be_block_dispatch_dev(struct ctl_be_block_lun * be_lun,struct ctl_be_block_io * beio)1232 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1233 struct ctl_be_block_io *beio)
1234 {
1235 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1236 struct bio *bio;
1237 struct cdevsw *csw;
1238 struct cdev *dev;
1239 off_t cur_offset;
1240 int i, max_iosize, ref;
1241
1242 DPRINTF("entered\n");
1243 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1244
1245 /*
1246 * We have to limit our I/O size to the maximum supported by the
1247 * backend device.
1248 */
1249 if (csw) {
1250 max_iosize = dev->si_iosize_max;
1251 if (max_iosize <= 0)
1252 max_iosize = DFLTPHYS;
1253 } else
1254 max_iosize = maxphys;
1255
1256 cur_offset = beio->io_offset;
1257 for (i = 0; i < beio->num_segs; i++) {
1258 size_t cur_size;
1259 uint8_t *cur_ptr;
1260
1261 cur_size = beio->sg_segs[i].len;
1262 cur_ptr = beio->sg_segs[i].addr;
1263
1264 while (cur_size > 0) {
1265 /* This can't fail, it's a blocking allocation. */
1266 bio = g_alloc_bio();
1267
1268 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1269
1270 bio->bio_cmd = beio->bio_cmd;
1271 bio->bio_dev = dev;
1272 bio->bio_caller1 = beio;
1273 bio->bio_length = min(cur_size, max_iosize);
1274 bio->bio_offset = cur_offset;
1275 bio->bio_data = cur_ptr;
1276 bio->bio_done = ctl_be_block_biodone;
1277 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize;
1278
1279 cur_offset += bio->bio_length;
1280 cur_ptr += bio->bio_length;
1281 cur_size -= bio->bio_length;
1282
1283 TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1284 beio->num_bios_sent++;
1285 }
1286 }
1287 beio->send_complete = 1;
1288 binuptime(&beio->ds_t0);
1289 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1290
1291 /*
1292 * Fire off all allocated requests!
1293 */
1294 while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1295 TAILQ_REMOVE(&queue, bio, bio_queue);
1296 if (csw)
1297 csw->d_strategy(bio);
1298 else {
1299 bio->bio_error = ENXIO;
1300 ctl_be_block_biodone(bio);
1301 }
1302 }
1303 if (csw)
1304 dev_relthread(dev, ref);
1305 }
1306
1307 static uint64_t
ctl_be_block_getattr_dev(struct ctl_be_block_lun * be_lun,const char * attrname)1308 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1309 {
1310 struct diocgattr_arg arg;
1311 struct cdevsw *csw;
1312 struct cdev *dev;
1313 int error, ref;
1314
1315 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1316 if (csw == NULL)
1317 return (UINT64_MAX);
1318 strlcpy(arg.name, attrname, sizeof(arg.name));
1319 arg.len = sizeof(arg.value.off);
1320 if (csw->d_ioctl) {
1321 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
1322 curthread);
1323 } else
1324 error = ENODEV;
1325 dev_relthread(dev, ref);
1326 if (error != 0)
1327 return (UINT64_MAX);
1328 return (arg.value.off);
1329 }
1330
1331 static void
ctl_be_block_namespace_data(struct ctl_be_block_lun * be_lun,union ctl_io * io)1332 ctl_be_block_namespace_data(struct ctl_be_block_lun *be_lun,
1333 union ctl_io *io)
1334 {
1335 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1336 struct nvme_namespace_data *nsdata;
1337
1338 nsdata = (struct nvme_namespace_data *)io->nvmeio.kern_data_ptr;
1339 memset(nsdata, 0, sizeof(*nsdata));
1340 nsdata->nsze = htole64(be_lun->size_blocks);
1341 nsdata->ncap = nsdata->nsze;
1342 nsdata->nuse = nsdata->nsze;
1343 nsdata->nlbaf = 1 - 1;
1344 nsdata->dlfeat = NVMEM(NVME_NS_DATA_DLFEAT_DWZ) |
1345 NVMEF(NVME_NS_DATA_DLFEAT_READ, NVME_NS_DATA_DLFEAT_READ_00);
1346 nsdata->flbas = NVMEF(NVME_NS_DATA_FLBAS_FORMAT, 0);
1347 nsdata->lbaf[0] = NVMEF(NVME_NS_DATA_LBAF_LBADS,
1348 ffs(cbe_lun->blocksize) - 1);
1349
1350 ctl_lun_nsdata_ids(cbe_lun, nsdata);
1351 ctl_config_read_done(io);
1352 }
1353
1354 static void
ctl_be_block_nvme_ids(struct ctl_be_block_lun * be_lun,union ctl_io * io)1355 ctl_be_block_nvme_ids(struct ctl_be_block_lun *be_lun,
1356 union ctl_io *io)
1357 {
1358 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1359
1360 ctl_lun_nvme_ids(cbe_lun, io->nvmeio.kern_data_ptr);
1361 ctl_config_read_done(io);
1362 }
1363
1364 static void
ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun * be_lun,union ctl_io * io)1365 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun,
1366 union ctl_io *io)
1367 {
1368 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1369 struct ctl_be_block_io *beio;
1370 struct ctl_lba_len_flags *lbalen;
1371
1372 DPRINTF("entered\n");
1373 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1374 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1375
1376 beio->io_len = lbalen->len * cbe_lun->blocksize;
1377 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1378 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0;
1379 beio->bio_cmd = BIO_FLUSH;
1380 beio->ds_trans_type = DEVSTAT_NO_DATA;
1381 DPRINTF("SYNC\n");
1382 be_lun->lun_flush(be_lun, beio);
1383 }
1384
1385 static void
ctl_be_block_cw_done_ws(struct ctl_be_block_io * beio)1386 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1387 {
1388 union ctl_io *io;
1389
1390 io = beio->io;
1391 ctl_free_beio(beio);
1392 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1393 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1394 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1395 ctl_config_write_done(io);
1396 return;
1397 }
1398
1399 ctl_be_block_config_write(io);
1400 }
1401
1402 static void
ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun * be_lun,union ctl_io * io)1403 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1404 union ctl_io *io)
1405 {
1406 struct ctl_be_block_softc *softc = be_lun->softc;
1407 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1408 struct ctl_be_block_io *beio;
1409 struct ctl_lba_len_flags *lbalen;
1410 uint64_t len_left, lba;
1411 uint32_t pb, pbo, adj;
1412 int i, seglen;
1413 uint8_t *buf, *end;
1414
1415 DPRINTF("entered\n");
1416
1417 CTL_IO_ASSERT(io, SCSI);
1418
1419 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1420 lbalen = ARGS(io);
1421
1422 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1423 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1424 ctl_free_beio(beio);
1425 ctl_set_invalid_field(&io->scsiio,
1426 /*sks_valid*/ 1,
1427 /*command*/ 1,
1428 /*field*/ 1,
1429 /*bit_valid*/ 0,
1430 /*bit*/ 0);
1431 ctl_config_write_done(io);
1432 return;
1433 }
1434
1435 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1436 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1437 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1438 beio->bio_cmd = BIO_DELETE;
1439 beio->ds_trans_type = DEVSTAT_FREE;
1440
1441 be_lun->unmap(be_lun, beio);
1442 return;
1443 }
1444
1445 beio->bio_cmd = BIO_WRITE;
1446 beio->ds_trans_type = DEVSTAT_WRITE;
1447
1448 DPRINTF("WRITE SAME at LBA %jx len %u\n",
1449 (uintmax_t)lbalen->lba, lbalen->len);
1450
1451 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1452 if (be_lun->cbe_lun.pblockoff > 0)
1453 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1454 else
1455 pbo = 0;
1456 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1457 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1458 /*
1459 * Setup the S/G entry for this chunk.
1460 */
1461 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1462 if (pb > cbe_lun->blocksize) {
1463 adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1464 seglen - pbo) % pb;
1465 if (seglen > adj)
1466 seglen -= adj;
1467 else
1468 seglen -= seglen % cbe_lun->blocksize;
1469 } else
1470 seglen -= seglen % cbe_lun->blocksize;
1471 ctl_alloc_seg(softc, &beio->sg_segs[i], seglen);
1472
1473 DPRINTF("segment %d addr %p len %zd\n", i,
1474 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1475
1476 beio->num_segs++;
1477 len_left -= seglen;
1478
1479 buf = beio->sg_segs[i].addr;
1480 end = buf + seglen;
1481 for (; buf < end; buf += cbe_lun->blocksize) {
1482 if (lbalen->flags & SWS_NDOB) {
1483 memset(buf, 0, cbe_lun->blocksize);
1484 } else {
1485 memcpy(buf, io->scsiio.kern_data_ptr,
1486 cbe_lun->blocksize);
1487 }
1488 if (lbalen->flags & SWS_LBDATA)
1489 scsi_ulto4b(lbalen->lba + lba, buf);
1490 lba++;
1491 }
1492 }
1493
1494 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1495 beio->io_len = lba * cbe_lun->blocksize;
1496
1497 /* We can not do all in one run. Correct and schedule rerun. */
1498 if (len_left > 0) {
1499 lbalen->lba += lba;
1500 lbalen->len -= lba;
1501 beio->beio_cont = ctl_be_block_cw_done_ws;
1502 }
1503
1504 be_lun->dispatch(be_lun, beio);
1505 }
1506
1507 static void
ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun * be_lun,union ctl_io * io)1508 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1509 union ctl_io *io)
1510 {
1511 struct ctl_be_block_io *beio;
1512 struct ctl_ptr_len_flags *ptrlen;
1513
1514 DPRINTF("entered\n");
1515
1516 CTL_IO_ASSERT(io, SCSI);
1517
1518 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1519 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1520
1521 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1522 ctl_free_beio(beio);
1523 ctl_set_invalid_field(&io->scsiio,
1524 /*sks_valid*/ 0,
1525 /*command*/ 1,
1526 /*field*/ 0,
1527 /*bit_valid*/ 0,
1528 /*bit*/ 0);
1529 ctl_config_write_done(io);
1530 return;
1531 }
1532
1533 beio->io_len = 0;
1534 beio->io_offset = -1;
1535 beio->bio_cmd = BIO_DELETE;
1536 beio->ds_trans_type = DEVSTAT_FREE;
1537 DPRINTF("UNMAP\n");
1538 be_lun->unmap(be_lun, beio);
1539 }
1540
1541 static void
ctl_be_block_cw_dispatch_flush(struct ctl_be_block_lun * be_lun,union ctl_io * io)1542 ctl_be_block_cw_dispatch_flush(struct ctl_be_block_lun *be_lun,
1543 union ctl_io *io)
1544 {
1545 struct ctl_be_block_io *beio;
1546
1547 DPRINTF("entered\n");
1548 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1549
1550 beio->io_len = be_lun->size_bytes;
1551 beio->io_offset = 0;
1552 beio->io_arg = 1;
1553 beio->bio_cmd = BIO_FLUSH;
1554 beio->ds_trans_type = DEVSTAT_NO_DATA;
1555 DPRINTF("FLUSH\n");
1556 be_lun->lun_flush(be_lun, beio);
1557 }
1558
1559 static void
ctl_be_block_cw_dispatch_wu(struct ctl_be_block_lun * be_lun,union ctl_io * io)1560 ctl_be_block_cw_dispatch_wu(struct ctl_be_block_lun *be_lun,
1561 union ctl_io *io)
1562 {
1563 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1564 struct ctl_be_block_io *beio;
1565 struct ctl_lba_len_flags *lbalen;
1566
1567 CTL_IO_ASSERT(io, NVME);
1568
1569 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1570 lbalen = ARGS(io);
1571
1572 /*
1573 * XXX: Not quite right as reads will return zeroes rather
1574 * than failing.
1575 */
1576 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1577 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1578 beio->bio_cmd = BIO_DELETE;
1579 beio->ds_trans_type = DEVSTAT_FREE;
1580
1581 be_lun->unmap(be_lun, beio);
1582 }
1583
1584 static void
ctl_be_block_cw_dispatch_wz(struct ctl_be_block_lun * be_lun,union ctl_io * io)1585 ctl_be_block_cw_dispatch_wz(struct ctl_be_block_lun *be_lun,
1586 union ctl_io *io)
1587 {
1588 struct ctl_be_block_softc *softc = be_lun->softc;
1589 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1590 struct ctl_be_block_io *beio;
1591 struct ctl_lba_len_flags *lbalen;
1592 uint64_t len_left, lba;
1593 uint32_t pb, pbo, adj;
1594 int i, seglen;
1595
1596 DPRINTF("entered\n");
1597
1598 CTL_IO_ASSERT(io, NVME);
1599
1600 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1601 lbalen = ARGS(io);
1602
1603 if ((le32toh(io->nvmeio.cmd.cdw12) & (1U << 25)) != 0 &&
1604 be_lun->unmap != NULL) {
1605 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1606 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1607 beio->bio_cmd = BIO_DELETE;
1608 beio->ds_trans_type = DEVSTAT_FREE;
1609
1610 be_lun->unmap(be_lun, beio);
1611 return;
1612 }
1613
1614 beio->bio_cmd = BIO_WRITE;
1615 beio->ds_trans_type = DEVSTAT_WRITE;
1616
1617 DPRINTF("WRITE ZEROES at LBA %jx len %u\n",
1618 (uintmax_t)lbalen->lba, lbalen->len);
1619
1620 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1621 if (be_lun->cbe_lun.pblockoff > 0)
1622 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1623 else
1624 pbo = 0;
1625 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1626 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1627 /*
1628 * Setup the S/G entry for this chunk.
1629 */
1630 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1631 if (pb > cbe_lun->blocksize) {
1632 adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1633 seglen - pbo) % pb;
1634 if (seglen > adj)
1635 seglen -= adj;
1636 else
1637 seglen -= seglen % cbe_lun->blocksize;
1638 } else
1639 seglen -= seglen % cbe_lun->blocksize;
1640 ctl_alloc_seg(softc, &beio->sg_segs[i], seglen);
1641
1642 DPRINTF("segment %d addr %p len %zd\n", i,
1643 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1644
1645 beio->num_segs++;
1646 len_left -= seglen;
1647
1648 memset(beio->sg_segs[i].addr, 0, seglen);
1649 lba += seglen / cbe_lun->blocksize;
1650 }
1651
1652 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1653 beio->io_len = lba * cbe_lun->blocksize;
1654
1655 /* We can not do all in one run. Correct and schedule rerun. */
1656 if (len_left > 0) {
1657 lbalen->lba += lba;
1658 lbalen->len -= lba;
1659 beio->beio_cont = ctl_be_block_cw_done_ws;
1660 }
1661
1662 be_lun->dispatch(be_lun, beio);
1663 }
1664
1665 static void
ctl_be_block_cw_dispatch_dsm(struct ctl_be_block_lun * be_lun,union ctl_io * io)1666 ctl_be_block_cw_dispatch_dsm(struct ctl_be_block_lun *be_lun,
1667 union ctl_io *io)
1668 {
1669 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1670 struct ctl_be_block_io *beio;
1671 struct nvme_dsm_range *r;
1672 uint64_t lba;
1673 uint32_t num_blocks;
1674 u_int i, ranges;
1675
1676 CTL_IO_ASSERT(io, NVME);
1677
1678 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1679
1680 if (be_lun->unmap == NULL) {
1681 ctl_free_beio(beio);
1682 ctl_nvme_set_success(&io->nvmeio);
1683 ctl_config_write_done(io);
1684 return;
1685 }
1686
1687 ranges = le32toh(io->nvmeio.cmd.cdw10) & 0xff;
1688 r = (struct nvme_dsm_range *)io->nvmeio.kern_data_ptr;
1689
1690 /* Find the next range to delete. */
1691 for (i = DSM_RANGE(io); i < ranges; i++) {
1692 if ((le32toh(r[i].attributes) & (1U << 2)) != 0)
1693 break;
1694 }
1695
1696 /* If no range to delete, complete the operation. */
1697 if (i == ranges) {
1698 ctl_free_beio(beio);
1699 ctl_nvme_set_success(&io->nvmeio);
1700 ctl_config_write_done(io);
1701 return;
1702 }
1703
1704 /* If this is not the last range, request a rerun after this range. */
1705 if (i + 1 < ranges) {
1706 DSM_RANGE(io) = i + 1;
1707 beio->beio_cont = ctl_be_block_cw_done_ws;
1708 }
1709
1710 lba = le64toh(r[i].starting_lba);
1711 num_blocks = le32toh(r[i].length);
1712
1713 beio->io_offset = lba * cbe_lun->blocksize;
1714 beio->io_len = (uint64_t)num_blocks * cbe_lun->blocksize;
1715 beio->bio_cmd = BIO_DELETE;
1716 beio->ds_trans_type = DEVSTAT_FREE;
1717
1718 be_lun->unmap(be_lun, beio);
1719 }
1720
1721 static void
ctl_be_block_scsi_cr_done(struct ctl_be_block_io * beio)1722 ctl_be_block_scsi_cr_done(struct ctl_be_block_io *beio)
1723 {
1724 union ctl_io *io;
1725
1726 io = beio->io;
1727 ctl_free_beio(beio);
1728 ctl_config_read_done(io);
1729 }
1730
1731 static void
ctl_be_block_scsi_cr_dispatch(struct ctl_be_block_lun * be_lun,union ctl_io * io)1732 ctl_be_block_scsi_cr_dispatch(struct ctl_be_block_lun *be_lun,
1733 union ctl_io *io)
1734 {
1735 struct ctl_be_block_io *beio;
1736 struct ctl_be_block_softc *softc;
1737
1738 DPRINTF("entered\n");
1739
1740 softc = be_lun->softc;
1741 beio = ctl_alloc_beio(softc);
1742 beio->io = io;
1743 beio->lun = be_lun;
1744 beio->beio_cont = ctl_be_block_scsi_cr_done;
1745 PRIV(io)->ptr = (void *)beio;
1746
1747 switch (io->scsiio.cdb[0]) {
1748 case SERVICE_ACTION_IN: /* GET LBA STATUS */
1749 beio->bio_cmd = -1;
1750 beio->ds_trans_type = DEVSTAT_NO_DATA;
1751 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1752 beio->io_len = 0;
1753 if (be_lun->get_lba_status)
1754 be_lun->get_lba_status(be_lun, beio);
1755 else
1756 ctl_be_block_scsi_cr_done(beio);
1757 break;
1758 default:
1759 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1760 break;
1761 }
1762 }
1763
1764 static void
ctl_be_block_nvme_cr_dispatch(struct ctl_be_block_lun * be_lun,union ctl_io * io)1765 ctl_be_block_nvme_cr_dispatch(struct ctl_be_block_lun *be_lun,
1766 union ctl_io *io)
1767 {
1768 uint8_t cns;
1769
1770 DPRINTF("entered\n");
1771
1772 MPASS(io->nvmeio.cmd.opc == NVME_OPC_IDENTIFY);
1773
1774 cns = le32toh(io->nvmeio.cmd.cdw10) & 0xff;
1775 switch (cns) {
1776 case 0:
1777 ctl_be_block_namespace_data(be_lun, io);
1778 break;
1779 case 3:
1780 ctl_be_block_nvme_ids(be_lun, io);
1781 break;
1782 default:
1783 __assert_unreachable();
1784 }
1785 }
1786
1787 static void
ctl_be_block_cr_dispatch(struct ctl_be_block_lun * be_lun,union ctl_io * io)1788 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1789 union ctl_io *io)
1790 {
1791 switch (io->io_hdr.io_type) {
1792 case CTL_IO_SCSI:
1793 ctl_be_block_scsi_cr_dispatch(be_lun, io);
1794 break;
1795 case CTL_IO_NVME_ADMIN:
1796 ctl_be_block_nvme_cr_dispatch(be_lun, io);
1797 break;
1798 default:
1799 __assert_unreachable();
1800 }
1801 }
1802
1803 static void
ctl_be_block_cw_done(struct ctl_be_block_io * beio)1804 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1805 {
1806 union ctl_io *io;
1807
1808 io = beio->io;
1809 ctl_free_beio(beio);
1810 ctl_config_write_done(io);
1811 }
1812
1813 static void
ctl_be_block_scsi_cw_dispatch(struct ctl_be_block_lun * be_lun,union ctl_io * io)1814 ctl_be_block_scsi_cw_dispatch(struct ctl_be_block_lun *be_lun,
1815 union ctl_io *io)
1816 {
1817 struct ctl_be_block_io *beio;
1818
1819 DPRINTF("entered\n");
1820
1821 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1822
1823 switch (io->scsiio.tag_type) {
1824 case CTL_TAG_ORDERED:
1825 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1826 break;
1827 case CTL_TAG_HEAD_OF_QUEUE:
1828 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1829 break;
1830 case CTL_TAG_UNTAGGED:
1831 case CTL_TAG_SIMPLE:
1832 case CTL_TAG_ACA:
1833 default:
1834 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1835 break;
1836 }
1837
1838 switch (io->scsiio.cdb[0]) {
1839 case SYNCHRONIZE_CACHE:
1840 case SYNCHRONIZE_CACHE_16:
1841 ctl_be_block_cw_dispatch_sync(be_lun, io);
1842 break;
1843 case WRITE_SAME_10:
1844 case WRITE_SAME_16:
1845 ctl_be_block_cw_dispatch_ws(be_lun, io);
1846 break;
1847 case UNMAP:
1848 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1849 break;
1850 default:
1851 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1852 break;
1853 }
1854 }
1855
1856 static void
ctl_be_block_nvme_cw_dispatch(struct ctl_be_block_lun * be_lun,union ctl_io * io)1857 ctl_be_block_nvme_cw_dispatch(struct ctl_be_block_lun *be_lun,
1858 union ctl_io *io)
1859 {
1860 struct ctl_be_block_io *beio;
1861
1862 DPRINTF("entered\n");
1863
1864 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1865 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1866
1867 switch (io->nvmeio.cmd.opc) {
1868 case NVME_OPC_FLUSH:
1869 ctl_be_block_cw_dispatch_flush(be_lun, io);
1870 break;
1871 case NVME_OPC_WRITE_UNCORRECTABLE:
1872 ctl_be_block_cw_dispatch_wu(be_lun, io);
1873 break;
1874 case NVME_OPC_WRITE_ZEROES:
1875 ctl_be_block_cw_dispatch_wz(be_lun, io);
1876 break;
1877 case NVME_OPC_DATASET_MANAGEMENT:
1878 ctl_be_block_cw_dispatch_dsm(be_lun, io);
1879 break;
1880 default:
1881 __assert_unreachable();
1882 }
1883 }
1884
1885 static void
ctl_be_block_cw_dispatch(struct ctl_be_block_lun * be_lun,union ctl_io * io)1886 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1887 union ctl_io *io)
1888 {
1889 struct ctl_be_block_io *beio;
1890 struct ctl_be_block_softc *softc;
1891
1892 softc = be_lun->softc;
1893 beio = ctl_alloc_beio(softc);
1894 beio->io = io;
1895 beio->lun = be_lun;
1896 beio->beio_cont = ctl_be_block_cw_done;
1897 PRIV(io)->ptr = (void *)beio;
1898
1899 switch (io->io_hdr.io_type) {
1900 case CTL_IO_SCSI:
1901 ctl_be_block_scsi_cw_dispatch(be_lun, io);
1902 break;
1903 case CTL_IO_NVME:
1904 ctl_be_block_nvme_cw_dispatch(be_lun, io);
1905 break;
1906 default:
1907 __assert_unreachable();
1908 }
1909 }
1910
1911 SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t");
1912 SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t");
1913 SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t");
1914 SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t");
1915
1916 static void
ctl_be_block_next(struct ctl_be_block_io * beio)1917 ctl_be_block_next(struct ctl_be_block_io *beio)
1918 {
1919 struct ctl_be_block_lun *be_lun;
1920 union ctl_io *io;
1921
1922 io = beio->io;
1923 be_lun = beio->lun;
1924 ctl_free_beio(beio);
1925 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1926 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1927 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1928 ctl_data_submit_done(io);
1929 return;
1930 }
1931
1932 io->io_hdr.status &= ~CTL_STATUS_MASK;
1933 io->io_hdr.status |= CTL_STATUS_NONE;
1934
1935 mtx_lock(&be_lun->queue_lock);
1936 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1937 mtx_unlock(&be_lun->queue_lock);
1938 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1939 }
1940
1941 static void
ctl_be_block_dispatch(struct ctl_be_block_lun * be_lun,union ctl_io * io)1942 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1943 union ctl_io *io)
1944 {
1945 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1946 struct ctl_be_block_io *beio;
1947 struct ctl_be_block_softc *softc;
1948 struct ctl_lba_len_flags *lbalen;
1949 struct ctl_ptr_len_flags *bptrlen;
1950 uint64_t len_left, lbas;
1951 int i;
1952
1953 softc = be_lun->softc;
1954
1955 DPRINTF("entered\n");
1956
1957 lbalen = ARGS(io);
1958 if (lbalen->flags & CTL_LLF_WRITE) {
1959 SDT_PROBE0(cbb, , write, start);
1960 } else {
1961 SDT_PROBE0(cbb, , read, start);
1962 }
1963
1964 beio = ctl_alloc_beio(softc);
1965 beio->io = io;
1966 beio->lun = be_lun;
1967 bptrlen = PRIV(io);
1968 bptrlen->ptr = (void *)beio;
1969
1970 switch (io->io_hdr.io_type) {
1971 case CTL_IO_SCSI:
1972 switch (io->scsiio.tag_type) {
1973 case CTL_TAG_ORDERED:
1974 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1975 break;
1976 case CTL_TAG_HEAD_OF_QUEUE:
1977 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1978 break;
1979 case CTL_TAG_UNTAGGED:
1980 case CTL_TAG_SIMPLE:
1981 case CTL_TAG_ACA:
1982 default:
1983 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1984 break;
1985 }
1986 break;
1987 case CTL_IO_NVME:
1988 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1989 break;
1990 default:
1991 __assert_unreachable();
1992 }
1993
1994 if (lbalen->flags & CTL_LLF_WRITE) {
1995 beio->bio_cmd = BIO_WRITE;
1996 beio->ds_trans_type = DEVSTAT_WRITE;
1997 } else {
1998 beio->bio_cmd = BIO_READ;
1999 beio->ds_trans_type = DEVSTAT_READ;
2000 }
2001
2002 DPRINTF("%s at LBA %jx len %u @%ju\n",
2003 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
2004 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
2005 lbas = CTLBLK_MAX_IO_SIZE;
2006 if (lbalen->flags & CTL_LLF_COMPARE) {
2007 beio->two_sglists = 1;
2008 lbas /= 2;
2009 }
2010 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize);
2011 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize;
2012 beio->io_len = lbas * cbe_lun->blocksize;
2013 bptrlen->len += lbas;
2014
2015 for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
2016 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
2017 i, CTLBLK_MAX_SEGS));
2018
2019 /*
2020 * Setup the S/G entry for this chunk.
2021 */
2022 ctl_alloc_seg(softc, &beio->sg_segs[i],
2023 MIN(CTLBLK_MAX_SEG, len_left));
2024
2025 DPRINTF("segment %d addr %p len %zd\n", i,
2026 beio->sg_segs[i].addr, beio->sg_segs[i].len);
2027
2028 /* Set up second segment for compare operation. */
2029 if (beio->two_sglists) {
2030 ctl_alloc_seg(softc,
2031 &beio->sg_segs[i + CTLBLK_HALF_SEGS],
2032 beio->sg_segs[i].len);
2033 }
2034
2035 beio->num_segs++;
2036 len_left -= beio->sg_segs[i].len;
2037 }
2038 if (bptrlen->len < lbalen->len)
2039 beio->beio_cont = ctl_be_block_next;
2040 ctl_set_be_move_done(io, ctl_be_block_move_done);
2041 /* For compare we have separate S/G lists for read and datamove. */
2042 if (beio->two_sglists)
2043 ctl_set_kern_data_ptr(io, &beio->sg_segs[CTLBLK_HALF_SEGS]);
2044 else
2045 ctl_set_kern_data_ptr(io, beio->sg_segs);
2046 ctl_set_kern_data_len(io, beio->io_len);
2047 ctl_set_kern_sg_entries(io, beio->num_segs);
2048 ctl_set_kern_data_ref(io, ctl_refcnt_beio);
2049 ctl_set_kern_data_arg(io, beio);
2050 io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
2051
2052 /*
2053 * For the read case, we need to read the data into our buffers and
2054 * then we can send it back to the user. For the write case, we
2055 * need to get the data from the user first.
2056 */
2057 if (beio->bio_cmd == BIO_READ) {
2058 SDT_PROBE0(cbb, , read, alloc_done);
2059 be_lun->dispatch(be_lun, beio);
2060 } else {
2061 SDT_PROBE0(cbb, , write, alloc_done);
2062 ctl_datamove(io);
2063 }
2064 }
2065
2066 static void
ctl_be_block_worker(void * context,int pending)2067 ctl_be_block_worker(void *context, int pending)
2068 {
2069 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context;
2070 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2071 union ctl_io *io;
2072 struct ctl_be_block_io *beio;
2073
2074 DPRINTF("entered\n");
2075 /*
2076 * Fetch and process I/Os from all queues. If we detect LUN
2077 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race,
2078 * so make response maximally opaque to not confuse initiator.
2079 */
2080 for (;;) {
2081 mtx_lock(&be_lun->queue_lock);
2082 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
2083 if (io != NULL) {
2084 DPRINTF("datamove queue\n");
2085 STAILQ_REMOVE_HEAD(&be_lun->datamove_queue, links);
2086 mtx_unlock(&be_lun->queue_lock);
2087 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
2088 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
2089 ctl_io_set_busy(io);
2090 ctl_complete_beio(beio);
2091 continue;
2092 }
2093 be_lun->dispatch(be_lun, beio);
2094 continue;
2095 }
2096 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
2097 if (io != NULL) {
2098 DPRINTF("config write queue\n");
2099 STAILQ_REMOVE_HEAD(&be_lun->config_write_queue, links);
2100 mtx_unlock(&be_lun->queue_lock);
2101 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
2102 ctl_io_set_busy(io);
2103 ctl_config_write_done(io);
2104 continue;
2105 }
2106 ctl_be_block_cw_dispatch(be_lun, io);
2107 continue;
2108 }
2109 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
2110 if (io != NULL) {
2111 DPRINTF("config read queue\n");
2112 STAILQ_REMOVE_HEAD(&be_lun->config_read_queue, links);
2113 mtx_unlock(&be_lun->queue_lock);
2114 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
2115 ctl_io_set_busy(io);
2116 ctl_config_read_done(io);
2117 continue;
2118 }
2119 ctl_be_block_cr_dispatch(be_lun, io);
2120 continue;
2121 }
2122 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
2123 if (io != NULL) {
2124 DPRINTF("input queue\n");
2125 STAILQ_REMOVE_HEAD(&be_lun->input_queue, links);
2126 mtx_unlock(&be_lun->queue_lock);
2127 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
2128 ctl_io_set_busy(io);
2129 ctl_data_submit_done(io);
2130 continue;
2131 }
2132 ctl_be_block_dispatch(be_lun, io);
2133 continue;
2134 }
2135
2136 /*
2137 * If we get here, there is no work left in the queues, so
2138 * just break out and let the task queue go to sleep.
2139 */
2140 mtx_unlock(&be_lun->queue_lock);
2141 break;
2142 }
2143 }
2144
2145 /*
2146 * Entry point from CTL to the backend for I/O. We queue everything to a
2147 * work thread, so this just puts the I/O on a queue and wakes up the
2148 * thread.
2149 */
2150 static int
ctl_be_block_submit(union ctl_io * io)2151 ctl_be_block_submit(union ctl_io *io)
2152 {
2153 struct ctl_be_block_lun *be_lun;
2154
2155 DPRINTF("entered\n");
2156
2157 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
2158
2159 CTL_IO_ASSERT(io, SCSI, NVME);
2160
2161 PRIV(io)->len = 0;
2162
2163 mtx_lock(&be_lun->queue_lock);
2164 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
2165 mtx_unlock(&be_lun->queue_lock);
2166 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2167
2168 return (CTL_RETVAL_COMPLETE);
2169 }
2170
2171 static int
ctl_be_block_ioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flag,struct thread * td)2172 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
2173 int flag, struct thread *td)
2174 {
2175 struct ctl_be_block_softc *softc = &backend_block_softc;
2176 int error;
2177
2178 error = 0;
2179 switch (cmd) {
2180 case CTL_LUN_REQ: {
2181 struct ctl_lun_req *lun_req;
2182
2183 lun_req = (struct ctl_lun_req *)addr;
2184
2185 switch (lun_req->reqtype) {
2186 case CTL_LUNREQ_CREATE:
2187 error = ctl_be_block_create(softc, lun_req);
2188 break;
2189 case CTL_LUNREQ_RM:
2190 error = ctl_be_block_rm(softc, lun_req);
2191 break;
2192 case CTL_LUNREQ_MODIFY:
2193 error = ctl_be_block_modify(softc, lun_req);
2194 break;
2195 default:
2196 lun_req->status = CTL_LUN_ERROR;
2197 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2198 "invalid LUN request type %d",
2199 lun_req->reqtype);
2200 break;
2201 }
2202 break;
2203 }
2204 default:
2205 error = ENOTTY;
2206 break;
2207 }
2208
2209 return (error);
2210 }
2211
2212 static int
ctl_be_block_open_file(struct ctl_be_block_lun * be_lun,struct ctl_lun_req * req)2213 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2214 {
2215 struct ctl_be_lun *cbe_lun;
2216 struct ctl_be_block_filedata *file_data;
2217 struct ctl_lun_create_params *params;
2218 const char *value;
2219 struct vattr vattr;
2220 off_t ps, pss, po, pos, us, uss, uo, uos;
2221 int error;
2222 long pconf;
2223
2224 cbe_lun = &be_lun->cbe_lun;
2225 file_data = &be_lun->backend.file;
2226 params = &be_lun->params;
2227
2228 be_lun->dev_type = CTL_BE_BLOCK_FILE;
2229 be_lun->dispatch = ctl_be_block_dispatch_file;
2230 be_lun->lun_flush = ctl_be_block_flush_file;
2231 be_lun->get_lba_status = ctl_be_block_gls_file;
2232 be_lun->getattr = ctl_be_block_getattr_file;
2233 be_lun->unmap = ctl_be_block_unmap_file;
2234 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
2235
2236 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
2237 if (error != 0) {
2238 snprintf(req->error_str, sizeof(req->error_str),
2239 "error calling VOP_GETATTR() for file %s",
2240 be_lun->dev_path);
2241 return (error);
2242 }
2243
2244 error = VOP_PATHCONF(be_lun->vn, _PC_DEALLOC_PRESENT, &pconf);
2245 if (error != 0) {
2246 snprintf(req->error_str, sizeof(req->error_str),
2247 "error calling VOP_PATHCONF() for file %s",
2248 be_lun->dev_path);
2249 return (error);
2250 }
2251 if (pconf == 1)
2252 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
2253
2254 file_data->cred = crhold(curthread->td_ucred);
2255 if (params->lun_size_bytes != 0)
2256 be_lun->size_bytes = params->lun_size_bytes;
2257 else
2258 be_lun->size_bytes = vattr.va_size;
2259
2260 /*
2261 * For files we can use any logical block size. Prefer 512 bytes
2262 * for compatibility reasons. If file's vattr.va_blocksize
2263 * (preferred I/O block size) is bigger and multiple to chosen
2264 * logical block size -- report it as physical block size.
2265 */
2266 if (params->blocksize_bytes != 0)
2267 cbe_lun->blocksize = params->blocksize_bytes;
2268 else if (cbe_lun->lun_type == T_CDROM)
2269 cbe_lun->blocksize = 2048;
2270 else
2271 cbe_lun->blocksize = 512;
2272 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2273 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2274 0 : (be_lun->size_blocks - 1);
2275
2276 us = ps = vattr.va_blocksize;
2277 uo = po = 0;
2278
2279 value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
2280 if (value != NULL)
2281 ctl_expand_number(value, &ps);
2282 value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
2283 if (value != NULL)
2284 ctl_expand_number(value, &po);
2285 pss = ps / cbe_lun->blocksize;
2286 pos = po / cbe_lun->blocksize;
2287 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
2288 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
2289 cbe_lun->pblockexp = fls(pss) - 1;
2290 cbe_lun->pblockoff = (pss - pos) % pss;
2291 }
2292
2293 value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
2294 if (value != NULL)
2295 ctl_expand_number(value, &us);
2296 value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
2297 if (value != NULL)
2298 ctl_expand_number(value, &uo);
2299 uss = us / cbe_lun->blocksize;
2300 uos = uo / cbe_lun->blocksize;
2301 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
2302 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
2303 cbe_lun->ublockexp = fls(uss) - 1;
2304 cbe_lun->ublockoff = (uss - uos) % uss;
2305 }
2306
2307 /*
2308 * Sanity check. The media size has to be at least one
2309 * sector long.
2310 */
2311 if (be_lun->size_bytes < cbe_lun->blocksize) {
2312 error = EINVAL;
2313 snprintf(req->error_str, sizeof(req->error_str),
2314 "file %s size %ju < block size %u", be_lun->dev_path,
2315 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize);
2316 }
2317
2318 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize;
2319 return (error);
2320 }
2321
2322 static int
ctl_be_block_open_dev(struct ctl_be_block_lun * be_lun,struct ctl_lun_req * req)2323 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2324 {
2325 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2326 struct ctl_lun_create_params *params;
2327 struct cdevsw *csw;
2328 struct cdev *dev;
2329 const char *value;
2330 int error, atomic, maxio, ref, unmap, tmp;
2331 off_t ps, pss, po, pos, us, uss, uo, uos, otmp;
2332
2333 params = &be_lun->params;
2334
2335 be_lun->dev_type = CTL_BE_BLOCK_DEV;
2336 csw = devvn_refthread(be_lun->vn, &dev, &ref);
2337 if (csw == NULL)
2338 return (ENXIO);
2339 if (strcmp(csw->d_name, "zvol") == 0) {
2340 be_lun->dispatch = ctl_be_block_dispatch_zvol;
2341 be_lun->get_lba_status = ctl_be_block_gls_zvol;
2342 atomic = maxio = CTLBLK_MAX_IO_SIZE;
2343 } else {
2344 be_lun->dispatch = ctl_be_block_dispatch_dev;
2345 be_lun->get_lba_status = NULL;
2346 atomic = 0;
2347 maxio = dev->si_iosize_max;
2348 if (maxio <= 0)
2349 maxio = DFLTPHYS;
2350 if (maxio > CTLBLK_MAX_SEG)
2351 maxio = CTLBLK_MAX_SEG;
2352 }
2353 be_lun->lun_flush = ctl_be_block_flush_dev;
2354 be_lun->getattr = ctl_be_block_getattr_dev;
2355 be_lun->unmap = ctl_be_block_unmap_dev;
2356
2357 if (!csw->d_ioctl) {
2358 dev_relthread(dev, ref);
2359 snprintf(req->error_str, sizeof(req->error_str),
2360 "no d_ioctl for device %s!", be_lun->dev_path);
2361 return (ENODEV);
2362 }
2363
2364 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD,
2365 curthread);
2366 if (error) {
2367 dev_relthread(dev, ref);
2368 snprintf(req->error_str, sizeof(req->error_str),
2369 "error %d returned for DIOCGSECTORSIZE ioctl "
2370 "on %s!", error, be_lun->dev_path);
2371 return (error);
2372 }
2373
2374 /*
2375 * If the user has asked for a blocksize that is greater than the
2376 * backing device's blocksize, we can do it only if the blocksize
2377 * the user is asking for is an even multiple of the underlying
2378 * device's blocksize.
2379 */
2380 if ((params->blocksize_bytes != 0) &&
2381 (params->blocksize_bytes >= tmp)) {
2382 if (params->blocksize_bytes % tmp == 0) {
2383 cbe_lun->blocksize = params->blocksize_bytes;
2384 } else {
2385 dev_relthread(dev, ref);
2386 snprintf(req->error_str, sizeof(req->error_str),
2387 "requested blocksize %u is not an even "
2388 "multiple of backing device blocksize %u",
2389 params->blocksize_bytes, tmp);
2390 return (EINVAL);
2391 }
2392 } else if (params->blocksize_bytes != 0) {
2393 dev_relthread(dev, ref);
2394 snprintf(req->error_str, sizeof(req->error_str),
2395 "requested blocksize %u < backing device "
2396 "blocksize %u", params->blocksize_bytes, tmp);
2397 return (EINVAL);
2398 } else if (cbe_lun->lun_type == T_CDROM)
2399 cbe_lun->blocksize = MAX(tmp, 2048);
2400 else
2401 cbe_lun->blocksize = tmp;
2402
2403 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD,
2404 curthread);
2405 if (error) {
2406 dev_relthread(dev, ref);
2407 snprintf(req->error_str, sizeof(req->error_str),
2408 "error %d returned for DIOCGMEDIASIZE "
2409 " ioctl on %s!", error,
2410 be_lun->dev_path);
2411 return (error);
2412 }
2413
2414 if (params->lun_size_bytes != 0) {
2415 if (params->lun_size_bytes > otmp) {
2416 dev_relthread(dev, ref);
2417 snprintf(req->error_str, sizeof(req->error_str),
2418 "requested LUN size %ju > backing device "
2419 "size %ju",
2420 (uintmax_t)params->lun_size_bytes,
2421 (uintmax_t)otmp);
2422 return (EINVAL);
2423 }
2424
2425 be_lun->size_bytes = params->lun_size_bytes;
2426 } else
2427 be_lun->size_bytes = otmp;
2428 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2429 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2430 0 : (be_lun->size_blocks - 1);
2431
2432 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD,
2433 curthread);
2434 if (error)
2435 ps = po = 0;
2436 else {
2437 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po,
2438 FREAD, curthread);
2439 if (error)
2440 po = 0;
2441 }
2442 us = ps;
2443 uo = po;
2444
2445 value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
2446 if (value != NULL)
2447 ctl_expand_number(value, &ps);
2448 value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
2449 if (value != NULL)
2450 ctl_expand_number(value, &po);
2451 pss = ps / cbe_lun->blocksize;
2452 pos = po / cbe_lun->blocksize;
2453 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
2454 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
2455 cbe_lun->pblockexp = fls(pss) - 1;
2456 cbe_lun->pblockoff = (pss - pos) % pss;
2457 }
2458
2459 value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
2460 if (value != NULL)
2461 ctl_expand_number(value, &us);
2462 value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
2463 if (value != NULL)
2464 ctl_expand_number(value, &uo);
2465 uss = us / cbe_lun->blocksize;
2466 uos = uo / cbe_lun->blocksize;
2467 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
2468 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
2469 cbe_lun->ublockexp = fls(uss) - 1;
2470 cbe_lun->ublockoff = (uss - uos) % uss;
2471 }
2472
2473 cbe_lun->atomicblock = atomic / cbe_lun->blocksize;
2474 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize;
2475
2476 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) {
2477 unmap = 1;
2478 } else {
2479 struct diocgattr_arg arg;
2480
2481 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
2482 arg.len = sizeof(arg.value.i);
2483 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
2484 curthread);
2485 unmap = (error == 0) ? arg.value.i : 0;
2486 }
2487 value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
2488 if (value != NULL)
2489 unmap = (strcmp(value, "on") == 0);
2490 if (unmap)
2491 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
2492 else
2493 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
2494
2495 dev_relthread(dev, ref);
2496 return (0);
2497 }
2498
2499 static int
ctl_be_block_close(struct ctl_be_block_lun * be_lun)2500 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
2501 {
2502 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2503 int flags;
2504
2505 if (be_lun->vn) {
2506 flags = FREAD;
2507 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0)
2508 flags |= FWRITE;
2509 (void)vn_close(be_lun->vn, flags, NOCRED, curthread);
2510 be_lun->vn = NULL;
2511
2512 switch (be_lun->dev_type) {
2513 case CTL_BE_BLOCK_DEV:
2514 break;
2515 case CTL_BE_BLOCK_FILE:
2516 if (be_lun->backend.file.cred != NULL) {
2517 crfree(be_lun->backend.file.cred);
2518 be_lun->backend.file.cred = NULL;
2519 }
2520 break;
2521 case CTL_BE_BLOCK_NONE:
2522 break;
2523 default:
2524 panic("Unexpected backend type %d", be_lun->dev_type);
2525 break;
2526 }
2527 be_lun->dev_type = CTL_BE_BLOCK_NONE;
2528 }
2529 return (0);
2530 }
2531
2532 static int
ctl_be_block_open(struct ctl_be_block_lun * be_lun,struct ctl_lun_req * req)2533 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2534 {
2535 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2536 struct nameidata nd;
2537 const char *value;
2538 int error, flags;
2539
2540 error = 0;
2541 if (rootvnode == NULL) {
2542 snprintf(req->error_str, sizeof(req->error_str),
2543 "Root filesystem is not mounted");
2544 return (1);
2545 }
2546 pwd_ensure_dirs();
2547
2548 value = dnvlist_get_string(cbe_lun->options, "file", NULL);
2549 if (value == NULL) {
2550 snprintf(req->error_str, sizeof(req->error_str),
2551 "no file argument specified");
2552 return (1);
2553 }
2554 free(be_lun->dev_path, M_CTLBLK);
2555 be_lun->dev_path = strdup(value, M_CTLBLK);
2556
2557 flags = FREAD;
2558 value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
2559 if (value != NULL) {
2560 if (strcmp(value, "on") != 0)
2561 flags |= FWRITE;
2562 } else if (cbe_lun->lun_type == T_DIRECT)
2563 flags |= FWRITE;
2564
2565 again:
2566 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path);
2567 error = vn_open(&nd, &flags, 0, NULL);
2568 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) {
2569 flags &= ~FWRITE;
2570 goto again;
2571 }
2572 if (error) {
2573 /*
2574 * This is the only reasonable guess we can make as far as
2575 * path if the user doesn't give us a fully qualified path.
2576 * If they want to specify a file, they need to specify the
2577 * full path.
2578 */
2579 if (be_lun->dev_path[0] != '/') {
2580 char *dev_name;
2581
2582 asprintf(&dev_name, M_CTLBLK, "/dev/%s",
2583 be_lun->dev_path);
2584 free(be_lun->dev_path, M_CTLBLK);
2585 be_lun->dev_path = dev_name;
2586 goto again;
2587 }
2588 snprintf(req->error_str, sizeof(req->error_str),
2589 "error opening %s: %d", be_lun->dev_path, error);
2590 return (error);
2591 }
2592 if (flags & FWRITE)
2593 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY;
2594 else
2595 cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
2596
2597 NDFREE_PNBUF(&nd);
2598 be_lun->vn = nd.ni_vp;
2599
2600 /* We only support disks and files. */
2601 if (vn_isdisk_error(be_lun->vn, &error)) {
2602 error = ctl_be_block_open_dev(be_lun, req);
2603 } else if (be_lun->vn->v_type == VREG) {
2604 error = ctl_be_block_open_file(be_lun, req);
2605 } else {
2606 error = EINVAL;
2607 snprintf(req->error_str, sizeof(req->error_str),
2608 "%s is not a disk or plain file", be_lun->dev_path);
2609 }
2610 VOP_UNLOCK(be_lun->vn);
2611
2612 if (error != 0)
2613 ctl_be_block_close(be_lun);
2614 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2615 if (be_lun->dispatch != ctl_be_block_dispatch_dev)
2616 cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
2617 value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
2618 if (value != NULL && strcmp(value, "on") == 0)
2619 cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
2620 else if (value != NULL && strcmp(value, "read") == 0)
2621 cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2622 else if (value != NULL && strcmp(value, "soft") == 0)
2623 cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT;
2624 else if (value != NULL && strcmp(value, "off") == 0)
2625 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2626 return (0);
2627 }
2628
2629 static int
ctl_be_block_create(struct ctl_be_block_softc * softc,struct ctl_lun_req * req)2630 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2631 {
2632 struct ctl_be_lun *cbe_lun;
2633 struct ctl_be_block_lun *be_lun;
2634 struct ctl_lun_create_params *params;
2635 char num_thread_str[16];
2636 char tmpstr[32];
2637 const char *value;
2638 int retval, num_threads;
2639 int tmp_num_threads;
2640
2641 params = &req->reqdata.create;
2642 retval = 0;
2643 req->status = CTL_LUN_OK;
2644
2645 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
2646 cbe_lun = &be_lun->cbe_lun;
2647 be_lun->params = req->reqdata.create;
2648 be_lun->softc = softc;
2649 STAILQ_INIT(&be_lun->input_queue);
2650 STAILQ_INIT(&be_lun->config_read_queue);
2651 STAILQ_INIT(&be_lun->config_write_queue);
2652 STAILQ_INIT(&be_lun->datamove_queue);
2653 mtx_init(&be_lun->io_lock, "ctlblock io", NULL, MTX_DEF);
2654 mtx_init(&be_lun->queue_lock, "ctlblock queue", NULL, MTX_DEF);
2655 cbe_lun->options = nvlist_clone(req->args_nvl);
2656
2657 if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
2658 cbe_lun->lun_type = params->device_type;
2659 else
2660 cbe_lun->lun_type = T_DIRECT;
2661 be_lun->flags = 0;
2662 cbe_lun->flags = 0;
2663 value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2664 if (value != NULL) {
2665 if (strcmp(value, "primary") == 0)
2666 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2667 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2668 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2669
2670 if (cbe_lun->lun_type == T_DIRECT ||
2671 cbe_lun->lun_type == T_CDROM) {
2672 be_lun->size_bytes = params->lun_size_bytes;
2673 if (params->blocksize_bytes != 0)
2674 cbe_lun->blocksize = params->blocksize_bytes;
2675 else if (cbe_lun->lun_type == T_CDROM)
2676 cbe_lun->blocksize = 2048;
2677 else
2678 cbe_lun->blocksize = 512;
2679 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2680 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2681 0 : (be_lun->size_blocks - 1);
2682
2683 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2684 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2685 retval = ctl_be_block_open(be_lun, req);
2686 if (retval != 0) {
2687 retval = 0;
2688 req->status = CTL_LUN_WARNING;
2689 }
2690 }
2691 num_threads = cbb_num_threads;
2692 } else {
2693 num_threads = 1;
2694 }
2695
2696 value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL);
2697 if (value != NULL) {
2698 tmp_num_threads = strtol(value, NULL, 0);
2699
2700 /*
2701 * We don't let the user specify less than one
2702 * thread, but hope he's clueful enough not to
2703 * specify 1000 threads.
2704 */
2705 if (tmp_num_threads < 1) {
2706 snprintf(req->error_str, sizeof(req->error_str),
2707 "invalid number of threads %s",
2708 num_thread_str);
2709 goto bailout_error;
2710 }
2711 num_threads = tmp_num_threads;
2712 }
2713
2714 if (be_lun->vn == NULL)
2715 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2716 /* Tell the user the blocksize we ended up using */
2717 params->lun_size_bytes = be_lun->size_bytes;
2718 params->blocksize_bytes = cbe_lun->blocksize;
2719 if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2720 cbe_lun->req_lun_id = params->req_lun_id;
2721 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
2722 } else
2723 cbe_lun->req_lun_id = 0;
2724
2725 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown;
2726 cbe_lun->be = &ctl_be_block_driver;
2727
2728 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2729 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
2730 softc->num_luns);
2731 strncpy((char *)cbe_lun->serial_num, tmpstr,
2732 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
2733
2734 /* Tell the user what we used for a serial number */
2735 strncpy((char *)params->serial_num, tmpstr,
2736 MIN(sizeof(params->serial_num), sizeof(tmpstr)));
2737 } else {
2738 strncpy((char *)cbe_lun->serial_num, params->serial_num,
2739 MIN(sizeof(cbe_lun->serial_num),
2740 sizeof(params->serial_num)));
2741 }
2742 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2743 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
2744 strncpy((char *)cbe_lun->device_id, tmpstr,
2745 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
2746
2747 /* Tell the user what we used for a device ID */
2748 strncpy((char *)params->device_id, tmpstr,
2749 MIN(sizeof(params->device_id), sizeof(tmpstr)));
2750 } else {
2751 strncpy((char *)cbe_lun->device_id, params->device_id,
2752 MIN(sizeof(cbe_lun->device_id),
2753 sizeof(params->device_id)));
2754 }
2755
2756 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2757
2758 be_lun->io_taskqueue = taskqueue_create("ctlblocktq", M_WAITOK,
2759 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2760
2761 if (be_lun->io_taskqueue == NULL) {
2762 snprintf(req->error_str, sizeof(req->error_str),
2763 "unable to create taskqueue");
2764 goto bailout_error;
2765 }
2766
2767 /*
2768 * Note that we start the same number of threads by default for
2769 * both the file case and the block device case. For the file
2770 * case, we need multiple threads to allow concurrency, because the
2771 * vnode interface is designed to be a blocking interface. For the
2772 * block device case, ZFS zvols at least will block the caller's
2773 * context in many instances, and so we need multiple threads to
2774 * overcome that problem. Other block devices don't need as many
2775 * threads, but they shouldn't cause too many problems.
2776 *
2777 * If the user wants to just have a single thread for a block
2778 * device, he can specify that when the LUN is created, or change
2779 * the tunable/sysctl to alter the default number of threads.
2780 */
2781 retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
2782 /*num threads*/num_threads,
2783 /*priority*/PUSER,
2784 /*proc*/control_softc->ctl_proc,
2785 /*thread name*/"block");
2786
2787 if (retval != 0)
2788 goto bailout_error;
2789
2790 be_lun->num_threads = num_threads;
2791
2792 retval = ctl_add_lun(&be_lun->cbe_lun);
2793 if (retval != 0) {
2794 snprintf(req->error_str, sizeof(req->error_str),
2795 "ctl_add_lun() returned error %d, see dmesg for "
2796 "details", retval);
2797 retval = 0;
2798 goto bailout_error;
2799 }
2800
2801 be_lun->disk_stats = devstat_new_entry("cbb", cbe_lun->lun_id,
2802 cbe_lun->blocksize,
2803 DEVSTAT_ALL_SUPPORTED,
2804 cbe_lun->lun_type
2805 | DEVSTAT_TYPE_IF_OTHER,
2806 DEVSTAT_PRIORITY_OTHER);
2807
2808 mtx_lock(&softc->lock);
2809 softc->num_luns++;
2810 SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
2811 mtx_unlock(&softc->lock);
2812
2813 params->req_lun_id = cbe_lun->lun_id;
2814
2815 return (retval);
2816
2817 bailout_error:
2818 req->status = CTL_LUN_ERROR;
2819
2820 if (be_lun->io_taskqueue != NULL)
2821 taskqueue_free(be_lun->io_taskqueue);
2822 ctl_be_block_close(be_lun);
2823 if (be_lun->dev_path != NULL)
2824 free(be_lun->dev_path, M_CTLBLK);
2825 nvlist_destroy(cbe_lun->options);
2826 mtx_destroy(&be_lun->queue_lock);
2827 mtx_destroy(&be_lun->io_lock);
2828 free(be_lun, M_CTLBLK);
2829
2830 return (retval);
2831 }
2832
2833 static int
ctl_be_block_rm(struct ctl_be_block_softc * softc,struct ctl_lun_req * req)2834 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2835 {
2836 struct ctl_lun_rm_params *params;
2837 struct ctl_be_block_lun *be_lun;
2838 struct ctl_be_lun *cbe_lun;
2839 int retval;
2840
2841 params = &req->reqdata.rm;
2842
2843 sx_xlock(&softc->modify_lock);
2844 mtx_lock(&softc->lock);
2845 SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2846 if (be_lun->cbe_lun.lun_id == params->lun_id) {
2847 SLIST_REMOVE(&softc->lun_list, be_lun,
2848 ctl_be_block_lun, links);
2849 softc->num_luns--;
2850 break;
2851 }
2852 }
2853 mtx_unlock(&softc->lock);
2854 sx_xunlock(&softc->modify_lock);
2855 if (be_lun == NULL) {
2856 snprintf(req->error_str, sizeof(req->error_str),
2857 "LUN %u is not managed by the block backend",
2858 params->lun_id);
2859 goto bailout_error;
2860 }
2861 cbe_lun = &be_lun->cbe_lun;
2862
2863 if (be_lun->vn != NULL) {
2864 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2865 ctl_lun_no_media(cbe_lun);
2866 taskqueue_drain_all(be_lun->io_taskqueue);
2867 ctl_be_block_close(be_lun);
2868 }
2869
2870 mtx_lock(&softc->lock);
2871 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2872 mtx_unlock(&softc->lock);
2873
2874 retval = ctl_remove_lun(cbe_lun);
2875 if (retval != 0) {
2876 snprintf(req->error_str, sizeof(req->error_str),
2877 "error %d returned from ctl_remove_lun() for "
2878 "LUN %d", retval, params->lun_id);
2879 mtx_lock(&softc->lock);
2880 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2881 mtx_unlock(&softc->lock);
2882 goto bailout_error;
2883 }
2884
2885 mtx_lock(&softc->lock);
2886 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2887 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblockrm", 0);
2888 if (retval == EINTR)
2889 break;
2890 }
2891 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2892 if (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2893 mtx_unlock(&softc->lock);
2894 free(be_lun, M_CTLBLK);
2895 } else {
2896 mtx_unlock(&softc->lock);
2897 return (EINTR);
2898 }
2899
2900 req->status = CTL_LUN_OK;
2901 return (0);
2902
2903 bailout_error:
2904 req->status = CTL_LUN_ERROR;
2905 return (0);
2906 }
2907
2908 static int
ctl_be_block_modify(struct ctl_be_block_softc * softc,struct ctl_lun_req * req)2909 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2910 {
2911 struct ctl_lun_modify_params *params;
2912 struct ctl_be_block_lun *be_lun;
2913 struct ctl_be_lun *cbe_lun;
2914 const char *value;
2915 uint64_t oldsize;
2916 int error, wasprim;
2917
2918 params = &req->reqdata.modify;
2919
2920 sx_xlock(&softc->modify_lock);
2921 mtx_lock(&softc->lock);
2922 SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2923 if (be_lun->cbe_lun.lun_id == params->lun_id)
2924 break;
2925 }
2926 mtx_unlock(&softc->lock);
2927 if (be_lun == NULL) {
2928 snprintf(req->error_str, sizeof(req->error_str),
2929 "LUN %u is not managed by the block backend",
2930 params->lun_id);
2931 goto bailout_error;
2932 }
2933 cbe_lun = &be_lun->cbe_lun;
2934
2935 if (params->lun_size_bytes != 0)
2936 be_lun->params.lun_size_bytes = params->lun_size_bytes;
2937
2938 if (req->args_nvl != NULL) {
2939 nvlist_destroy(cbe_lun->options);
2940 cbe_lun->options = nvlist_clone(req->args_nvl);
2941 }
2942
2943 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
2944 value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2945 if (value != NULL) {
2946 if (strcmp(value, "primary") == 0)
2947 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2948 else
2949 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2950 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2951 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2952 else
2953 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2954 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
2955 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
2956 ctl_lun_primary(cbe_lun);
2957 else
2958 ctl_lun_secondary(cbe_lun);
2959 }
2960
2961 oldsize = be_lun->size_blocks;
2962 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2963 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2964 if (be_lun->vn == NULL)
2965 error = ctl_be_block_open(be_lun, req);
2966 else if (vn_isdisk_error(be_lun->vn, &error))
2967 error = ctl_be_block_open_dev(be_lun, req);
2968 else if (be_lun->vn->v_type == VREG) {
2969 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2970 error = ctl_be_block_open_file(be_lun, req);
2971 VOP_UNLOCK(be_lun->vn);
2972 } else
2973 error = EINVAL;
2974 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) &&
2975 be_lun->vn != NULL) {
2976 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2977 ctl_lun_has_media(cbe_lun);
2978 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 &&
2979 be_lun->vn == NULL) {
2980 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2981 ctl_lun_no_media(cbe_lun);
2982 }
2983 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2984 } else {
2985 if (be_lun->vn != NULL) {
2986 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2987 ctl_lun_no_media(cbe_lun);
2988 taskqueue_drain_all(be_lun->io_taskqueue);
2989 error = ctl_be_block_close(be_lun);
2990 } else
2991 error = 0;
2992 }
2993 if (be_lun->size_blocks != oldsize)
2994 ctl_lun_capacity_changed(cbe_lun);
2995
2996 /* Tell the user the exact size we ended up using */
2997 params->lun_size_bytes = be_lun->size_bytes;
2998
2999 sx_xunlock(&softc->modify_lock);
3000 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
3001 return (0);
3002
3003 bailout_error:
3004 sx_xunlock(&softc->modify_lock);
3005 req->status = CTL_LUN_ERROR;
3006 return (0);
3007 }
3008
3009 static void
ctl_be_block_lun_shutdown(struct ctl_be_lun * cbe_lun)3010 ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun)
3011 {
3012 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)cbe_lun;
3013 struct ctl_be_block_softc *softc = be_lun->softc;
3014
3015 taskqueue_drain_all(be_lun->io_taskqueue);
3016 taskqueue_free(be_lun->io_taskqueue);
3017 if (be_lun->disk_stats != NULL)
3018 devstat_remove_entry(be_lun->disk_stats);
3019 nvlist_destroy(be_lun->cbe_lun.options);
3020 free(be_lun->dev_path, M_CTLBLK);
3021 mtx_destroy(&be_lun->queue_lock);
3022 mtx_destroy(&be_lun->io_lock);
3023
3024 mtx_lock(&softc->lock);
3025 be_lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
3026 if (be_lun->flags & CTL_BE_BLOCK_LUN_WAITING)
3027 wakeup(be_lun);
3028 else
3029 free(be_lun, M_CTLBLK);
3030 mtx_unlock(&softc->lock);
3031 }
3032
3033 static int
ctl_be_block_scsi_config_write(union ctl_io * io)3034 ctl_be_block_scsi_config_write(union ctl_io *io)
3035 {
3036 struct ctl_be_block_lun *be_lun;
3037 struct ctl_be_lun *cbe_lun;
3038 int retval;
3039
3040 DPRINTF("entered\n");
3041
3042 cbe_lun = CTL_BACKEND_LUN(io);
3043 be_lun = (struct ctl_be_block_lun *)cbe_lun;
3044
3045 retval = 0;
3046 switch (io->scsiio.cdb[0]) {
3047 case SYNCHRONIZE_CACHE:
3048 case SYNCHRONIZE_CACHE_16:
3049 case WRITE_SAME_10:
3050 case WRITE_SAME_16:
3051 case UNMAP:
3052 /*
3053 * The upper level CTL code will filter out any CDBs with
3054 * the immediate bit set and return the proper error.
3055 *
3056 * We don't really need to worry about what LBA range the
3057 * user asked to be synced out. When they issue a sync
3058 * cache command, we'll sync out the whole thing.
3059 */
3060 mtx_lock(&be_lun->queue_lock);
3061 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
3062 links);
3063 mtx_unlock(&be_lun->queue_lock);
3064 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
3065 break;
3066 case START_STOP_UNIT: {
3067 struct scsi_start_stop_unit *cdb;
3068 struct ctl_lun_req req;
3069
3070 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
3071 if ((cdb->how & SSS_PC_MASK) != 0) {
3072 ctl_set_success(&io->scsiio);
3073 ctl_config_write_done(io);
3074 break;
3075 }
3076 if (cdb->how & SSS_START) {
3077 if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) {
3078 retval = ctl_be_block_open(be_lun, &req);
3079 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
3080 if (retval == 0) {
3081 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
3082 ctl_lun_has_media(cbe_lun);
3083 } else {
3084 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
3085 ctl_lun_no_media(cbe_lun);
3086 }
3087 }
3088 ctl_start_lun(cbe_lun);
3089 } else {
3090 ctl_stop_lun(cbe_lun);
3091 if (cdb->how & SSS_LOEJ) {
3092 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
3093 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED;
3094 ctl_lun_ejected(cbe_lun);
3095 if (be_lun->vn != NULL)
3096 ctl_be_block_close(be_lun);
3097 }
3098 }
3099
3100 ctl_set_success(&io->scsiio);
3101 ctl_config_write_done(io);
3102 break;
3103 }
3104 case PREVENT_ALLOW:
3105 ctl_set_success(&io->scsiio);
3106 ctl_config_write_done(io);
3107 break;
3108 default:
3109 ctl_set_invalid_opcode(&io->scsiio);
3110 ctl_config_write_done(io);
3111 retval = CTL_RETVAL_COMPLETE;
3112 break;
3113 }
3114
3115 return (retval);
3116 }
3117
3118 static int
ctl_be_block_nvme_config_write(union ctl_io * io)3119 ctl_be_block_nvme_config_write(union ctl_io *io)
3120 {
3121 struct ctl_be_block_lun *be_lun;
3122
3123 DPRINTF("entered\n");
3124
3125 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
3126
3127 switch (io->nvmeio.cmd.opc) {
3128 case NVME_OPC_DATASET_MANAGEMENT:
3129 DSM_RANGE(io) = 0;
3130 /* FALLTHROUGH */
3131 case NVME_OPC_FLUSH:
3132 case NVME_OPC_WRITE_UNCORRECTABLE:
3133 case NVME_OPC_WRITE_ZEROES:
3134 mtx_lock(&be_lun->queue_lock);
3135 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
3136 links);
3137 mtx_unlock(&be_lun->queue_lock);
3138 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
3139 break;
3140 default:
3141 ctl_nvme_set_invalid_opcode(&io->nvmeio);
3142 ctl_config_write_done(io);
3143 break;
3144 }
3145 return (CTL_RETVAL_COMPLETE);
3146 }
3147
3148 static int
ctl_be_block_config_write(union ctl_io * io)3149 ctl_be_block_config_write(union ctl_io *io)
3150 {
3151 switch (io->io_hdr.io_type) {
3152 case CTL_IO_SCSI:
3153 return (ctl_be_block_scsi_config_write(io));
3154 case CTL_IO_NVME:
3155 return (ctl_be_block_nvme_config_write(io));
3156 default:
3157 __assert_unreachable();
3158 }
3159 }
3160
3161 static int
ctl_be_block_scsi_config_read(union ctl_io * io)3162 ctl_be_block_scsi_config_read(union ctl_io *io)
3163 {
3164 struct ctl_be_block_lun *be_lun;
3165 int retval = 0;
3166
3167 DPRINTF("entered\n");
3168
3169 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
3170
3171 switch (io->scsiio.cdb[0]) {
3172 case SERVICE_ACTION_IN:
3173 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
3174 mtx_lock(&be_lun->queue_lock);
3175 STAILQ_INSERT_TAIL(&be_lun->config_read_queue,
3176 &io->io_hdr, links);
3177 mtx_unlock(&be_lun->queue_lock);
3178 taskqueue_enqueue(be_lun->io_taskqueue,
3179 &be_lun->io_task);
3180 retval = CTL_RETVAL_QUEUED;
3181 break;
3182 }
3183 ctl_set_invalid_field(&io->scsiio,
3184 /*sks_valid*/ 1,
3185 /*command*/ 1,
3186 /*field*/ 1,
3187 /*bit_valid*/ 1,
3188 /*bit*/ 4);
3189 ctl_config_read_done(io);
3190 retval = CTL_RETVAL_COMPLETE;
3191 break;
3192 default:
3193 ctl_set_invalid_opcode(&io->scsiio);
3194 ctl_config_read_done(io);
3195 retval = CTL_RETVAL_COMPLETE;
3196 break;
3197 }
3198
3199 return (retval);
3200 }
3201
3202 static int
ctl_be_block_nvme_config_read(union ctl_io * io)3203 ctl_be_block_nvme_config_read(union ctl_io *io)
3204 {
3205 struct ctl_be_block_lun *be_lun;
3206
3207 DPRINTF("entered\n");
3208
3209 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
3210
3211 switch (io->nvmeio.cmd.opc) {
3212 case NVME_OPC_IDENTIFY:
3213 {
3214 uint8_t cns;
3215
3216 cns = le32toh(io->nvmeio.cmd.cdw10) & 0xff;
3217 switch (cns) {
3218 case 0:
3219 case 3:
3220 mtx_lock(&be_lun->queue_lock);
3221 STAILQ_INSERT_TAIL(&be_lun->config_read_queue,
3222 &io->io_hdr, links);
3223 mtx_unlock(&be_lun->queue_lock);
3224 taskqueue_enqueue(be_lun->io_taskqueue,
3225 &be_lun->io_task);
3226 return (CTL_RETVAL_QUEUED);
3227 default:
3228 ctl_nvme_set_invalid_field(&io->nvmeio);
3229 ctl_config_read_done(io);
3230 break;
3231 }
3232 break;
3233 }
3234 default:
3235 ctl_nvme_set_invalid_opcode(&io->nvmeio);
3236 ctl_config_read_done(io);
3237 break;
3238 }
3239 return (CTL_RETVAL_COMPLETE);
3240 }
3241
3242 static int
ctl_be_block_config_read(union ctl_io * io)3243 ctl_be_block_config_read(union ctl_io *io)
3244 {
3245 switch (io->io_hdr.io_type) {
3246 case CTL_IO_SCSI:
3247 return (ctl_be_block_scsi_config_read(io));
3248 case CTL_IO_NVME_ADMIN:
3249 return (ctl_be_block_nvme_config_read(io));
3250 default:
3251 __assert_unreachable();
3252 }
3253 }
3254
3255 static int
ctl_be_block_lun_info(struct ctl_be_lun * cbe_lun,struct sbuf * sb)3256 ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb)
3257 {
3258 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
3259 int retval;
3260
3261 retval = sbuf_cat(sb, "\t<num_threads>");
3262 if (retval != 0)
3263 goto bailout;
3264 retval = sbuf_printf(sb, "%d", lun->num_threads);
3265 if (retval != 0)
3266 goto bailout;
3267 retval = sbuf_cat(sb, "</num_threads>\n");
3268
3269 bailout:
3270 return (retval);
3271 }
3272
3273 static uint64_t
ctl_be_block_lun_attr(struct ctl_be_lun * cbe_lun,const char * attrname)3274 ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
3275 {
3276 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
3277
3278 if (lun->getattr == NULL)
3279 return (UINT64_MAX);
3280 return (lun->getattr(lun, attrname));
3281 }
3282
3283 static int
ctl_be_block_init(void)3284 ctl_be_block_init(void)
3285 {
3286 struct ctl_be_block_softc *softc = &backend_block_softc;
3287
3288 sx_init(&softc->modify_lock, "ctlblock modify");
3289 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
3290 softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
3291 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3292 softc->bufmin_zone = uma_zcreate("ctlblockmin", CTLBLK_MIN_SEG,
3293 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
3294 if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG)
3295 softc->bufmax_zone = uma_zcreate("ctlblockmax", CTLBLK_MAX_SEG,
3296 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
3297 SLIST_INIT(&softc->lun_list);
3298 return (0);
3299 }
3300
3301 static int
ctl_be_block_shutdown(void)3302 ctl_be_block_shutdown(void)
3303 {
3304 struct ctl_be_block_softc *softc = &backend_block_softc;
3305 struct ctl_be_block_lun *lun;
3306
3307 mtx_lock(&softc->lock);
3308 while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
3309 SLIST_REMOVE_HEAD(&softc->lun_list, links);
3310 softc->num_luns--;
3311 /*
3312 * Drop our lock here. Since ctl_remove_lun() can call
3313 * back into us, this could potentially lead to a recursive
3314 * lock of the same mutex, which would cause a hang.
3315 */
3316 mtx_unlock(&softc->lock);
3317 ctl_remove_lun(&lun->cbe_lun);
3318 mtx_lock(&softc->lock);
3319 }
3320 mtx_unlock(&softc->lock);
3321 uma_zdestroy(softc->bufmin_zone);
3322 if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG)
3323 uma_zdestroy(softc->bufmax_zone);
3324 uma_zdestroy(softc->beio_zone);
3325 mtx_destroy(&softc->lock);
3326 sx_destroy(&softc->modify_lock);
3327 return (0);
3328 }
3329