1/*-
2 * Copyright (c) 2003 Silicon Graphics International Corp.
3 * Copyright (c) 2009-2011 Spectra Logic Corporation
4 * Copyright (c) 2012 The FreeBSD Foundation
5 * All rights reserved.
6 *
7 * Portions of this software were developed by Edward Tomasz Napierala
8 * under sponsorship from the FreeBSD Foundation.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions, and the following disclaimer,
15 * without modification.
16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
17 * substantially similar to the "NO WARRANTY" disclaimer below
18 * ("Disclaimer") and any redistribution must be conditioned upon
19 * including a substantially similar Disclaimer requirement for further
20 * binary redistribution.
21 *
22 * NO WARRANTY
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33 * POSSIBILITY OF SUCH DAMAGES.
34 *
35 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $
36 */
37/*
38 * CAM Target Layer driver backend for block devices.
39 *
40 * Author: Ken Merry <ken@FreeBSD.org>
41 */
42#include <sys/cdefs.h>
43__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl_backend_block.c 275865 2014-12-17 17:30:54Z mav $");
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/kernel.h>
48#include <sys/types.h>
49#include <sys/kthread.h>
50#include <sys/bio.h>
51#include <sys/fcntl.h>
52#include <sys/limits.h>
53#include <sys/lock.h>
54#include <sys/mutex.h>
55#include <sys/condvar.h>
56#include <sys/malloc.h>
57#include <sys/conf.h>
58#include <sys/ioccom.h>
59#include <sys/queue.h>
60#include <sys/sbuf.h>
61#include <sys/endian.h>
62#include <sys/uio.h>
63#include <sys/buf.h>
64#include <sys/taskqueue.h>
65#include <sys/vnode.h>
66#include <sys/namei.h>
67#include <sys/mount.h>
68#include <sys/disk.h>
69#include <sys/fcntl.h>
70#include <sys/filedesc.h>
71#include <sys/filio.h>
72#include <sys/proc.h>
73#include <sys/pcpu.h>
74#include <sys/module.h>
75#include <sys/sdt.h>
76#include <sys/devicestat.h>
77#include <sys/sysctl.h>
78
79#include <geom/geom.h>
80
81#include <cam/cam.h>
82#include <cam/scsi/scsi_all.h>
83#include <cam/scsi/scsi_da.h>
84#include <cam/ctl/ctl_io.h>
85#include <cam/ctl/ctl.h>
86#include <cam/ctl/ctl_backend.h>
87#include <cam/ctl/ctl_frontend_internal.h>
88#include <cam/ctl/ctl_ioctl.h>
89#include <cam/ctl/ctl_scsi_all.h>
90#include <cam/ctl/ctl_error.h>
91
92/*
93 * The idea here is that we'll allocate enough S/G space to hold a 1MB
94 * I/O. If we get an I/O larger than that, we'll split it.
95 */
96#define CTLBLK_HALF_IO_SIZE (512 * 1024)
97#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2)
98#define CTLBLK_MAX_SEG MAXPHYS
99#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1)
100#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2)
101
102#ifdef CTLBLK_DEBUG
103#define DPRINTF(fmt, args...) \
104 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
105#else
106#define DPRINTF(fmt, args...) do {} while(0)
107#endif
108
109#define PRIV(io) \
110 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
111#define ARGS(io) \
112 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
113
114SDT_PROVIDER_DEFINE(cbb);
115
116typedef enum {
117 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01,
118 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02,
119 CTL_BE_BLOCK_LUN_WAITING = 0x04,
120 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08
121} ctl_be_block_lun_flags;
122
123typedef enum {
124 CTL_BE_BLOCK_NONE,
125 CTL_BE_BLOCK_DEV,
126 CTL_BE_BLOCK_FILE
127} ctl_be_block_type;
128
129struct ctl_be_block_devdata {
130 struct cdev *cdev;
131 struct cdevsw *csw;
132 int dev_ref;
133};
134
135struct ctl_be_block_filedata {
136 struct ucred *cred;
137};
138
139union ctl_be_block_bedata {
140 struct ctl_be_block_devdata dev;
141 struct ctl_be_block_filedata file;
142};
143
144struct ctl_be_block_io;
145struct ctl_be_block_lun;
146
147typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
148 struct ctl_be_block_io *beio);
149typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
150 const char *attrname);
151
152/*
153 * Backend LUN structure. There is a 1:1 mapping between a block device
154 * and a backend block LUN, and between a backend block LUN and a CTL LUN.
155 */
156struct ctl_be_block_lun {
157 struct ctl_lun_create_params params;
158 struct ctl_block_disk *disk;
159 char lunname[32];
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 uma_zone_t lun_zone;
170 uint64_t size_blocks;
171 uint64_t size_bytes;
172 uint32_t blocksize;
173 int blocksize_shift;
174 uint16_t pblockexp;
175 uint16_t pblockoff;
176 uint16_t ublockexp;
177 uint16_t ublockoff;
178 struct ctl_be_block_softc *softc;
179 struct devstat *disk_stats;
180 ctl_be_block_lun_flags flags;
181 STAILQ_ENTRY(ctl_be_block_lun) links;
182 struct ctl_be_lun ctl_be_lun;
183 struct taskqueue *io_taskqueue;
184 struct task io_task;
185 int num_threads;
186 STAILQ_HEAD(, ctl_io_hdr) input_queue;
187 STAILQ_HEAD(, ctl_io_hdr) config_read_queue;
188 STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
189 STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
190 struct mtx_padalign io_lock;
191 struct mtx_padalign queue_lock;
192};
193
194/*
195 * Overall softc structure for the block backend module.
196 */
197struct ctl_be_block_softc {
198 struct mtx lock;
199 int num_disks;
200 STAILQ_HEAD(, ctl_block_disk) disk_list;
201 int num_luns;
202 STAILQ_HEAD(, ctl_be_block_lun) lun_list;
203};
204
205static struct ctl_be_block_softc backend_block_softc;
206
207/*
208 * Per-I/O information.
209 */
210struct ctl_be_block_io {
211 union ctl_io *io;
212 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS];
213 struct iovec xiovecs[CTLBLK_MAX_SEGS];
214 int bio_cmd;
215 int num_segs;
216 int num_bios_sent;
217 int num_bios_done;
218 int send_complete;
219 int num_errors;
220 struct bintime ds_t0;
221 devstat_tag_type ds_tag_type;
222 devstat_trans_flags ds_trans_type;
223 uint64_t io_len;
224 uint64_t io_offset;
225 struct ctl_be_block_softc *softc;
226 struct ctl_be_block_lun *lun;
227 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
228};
229
230static int cbb_num_threads = 14;
231SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0,
232 "CAM Target Layer Block Backend");
233SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
234 &cbb_num_threads, 0, "Number of threads per backing file");
235
236static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
237static void ctl_free_beio(struct ctl_be_block_io *beio);
238static void ctl_complete_beio(struct ctl_be_block_io *beio);
239static int ctl_be_block_move_done(union ctl_io *io);
240static void ctl_be_block_biodone(struct bio *bio);
241static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
242 struct ctl_be_block_io *beio);
243static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
244 struct ctl_be_block_io *beio);
245static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
246 struct ctl_be_block_io *beio);
247static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
248 const char *attrname);
249static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
250 struct ctl_be_block_io *beio);
251static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
252 struct ctl_be_block_io *beio);
253static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
254 struct ctl_be_block_io *beio);
255static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
256 const char *attrname);
257static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
258 union ctl_io *io);
259static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
260 union ctl_io *io);
261static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
262 union ctl_io *io);
263static void ctl_be_block_worker(void *context, int pending);
264static int ctl_be_block_submit(union ctl_io *io);
265static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
266 int flag, struct thread *td);
267static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
268 struct ctl_lun_req *req);
269static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
270 struct ctl_lun_req *req);
271static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
272static int ctl_be_block_open(struct ctl_be_block_softc *softc,
273 struct ctl_be_block_lun *be_lun,
274 struct ctl_lun_req *req);
275static int ctl_be_block_create(struct ctl_be_block_softc *softc,
276 struct ctl_lun_req *req);
277static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
278 struct ctl_lun_req *req);
279static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
280 struct ctl_lun_req *req);
281static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
282 struct ctl_lun_req *req);
283static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
284 struct ctl_lun_req *req);
285static void ctl_be_block_lun_shutdown(void *be_lun);
286static void ctl_be_block_lun_config_status(void *be_lun,
287 ctl_lun_config_status status);
288static int ctl_be_block_config_write(union ctl_io *io);
289static int ctl_be_block_config_read(union ctl_io *io);
290static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb);
291static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname);
292int ctl_be_block_init(void);
293
294static struct ctl_backend_driver ctl_be_block_driver =
295{
296 .name = "block",
297 .flags = CTL_BE_FLAG_HAS_CONFIG,
298 .init = ctl_be_block_init,
299 .data_submit = ctl_be_block_submit,
300 .data_move_done = ctl_be_block_move_done,
301 .config_read = ctl_be_block_config_read,
302 .config_write = ctl_be_block_config_write,
303 .ioctl = ctl_be_block_ioctl,
304 .lun_info = ctl_be_block_lun_info,
305 .lun_attr = ctl_be_block_lun_attr
306};
307
308MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend");
309CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
310
311static uma_zone_t beio_zone;
312
313static struct ctl_be_block_io *
314ctl_alloc_beio(struct ctl_be_block_softc *softc)
315{
316 struct ctl_be_block_io *beio;
317
318 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO);
319 beio->softc = softc;
320 return (beio);
321}
322
323static void
324ctl_free_beio(struct ctl_be_block_io *beio)
325{
326 int duplicate_free;
327 int i;
328
329 duplicate_free = 0;
330
331 for (i = 0; i < beio->num_segs; i++) {
332 if (beio->sg_segs[i].addr == NULL)
333 duplicate_free++;
334
335 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr);
336 beio->sg_segs[i].addr = NULL;
337
338 /* For compare we had two equal S/G lists. */
339 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) {
340 uma_zfree(beio->lun->lun_zone,
341 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr);
342 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL;
343 }
344 }
345
346 if (duplicate_free > 0) {
347 printf("%s: %d duplicate frees out of %d segments\n", __func__,
348 duplicate_free, beio->num_segs);
349 }
350
351 uma_zfree(beio_zone, beio);
352}
353
354static void
355ctl_complete_beio(struct ctl_be_block_io *beio)
356{
357 union ctl_io *io = beio->io;
358
359 if (beio->beio_cont != NULL) {
360 beio->beio_cont(beio);
361 } else {
362 ctl_free_beio(beio);
363 ctl_data_submit_done(io);
364 }
365}
366
367static int
368ctl_be_block_move_done(union ctl_io *io)
369{
370 struct ctl_be_block_io *beio;
371 struct ctl_be_block_lun *be_lun;
372 struct ctl_lba_len_flags *lbalen;
373#ifdef CTL_TIME_IO
374 struct bintime cur_bt;
375#endif
376 int i;
377
378 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
379 be_lun = beio->lun;
380
381 DPRINTF("entered\n");
382
383#ifdef CTL_TIME_IO
384 getbintime(&cur_bt);
385 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
386 bintime_add(&io->io_hdr.dma_bt, &cur_bt);
387 io->io_hdr.num_dmas++;
388#endif
389 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
390
391 /*
392 * We set status at this point for read commands, and write
393 * commands with errors.
394 */
395 if (io->io_hdr.flags & CTL_FLAG_ABORT) {
396 ;
397 } else if ((io->io_hdr.port_status == 0) &&
398 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
399 lbalen = ARGS(beio->io);
400 if (lbalen->flags & CTL_LLF_READ) {
401 ctl_set_success(&io->scsiio);
402 } else if (lbalen->flags & CTL_LLF_COMPARE) {
403 /* We have two data blocks ready for comparison. */
404 for (i = 0; i < beio->num_segs; i++) {
405 if (memcmp(beio->sg_segs[i].addr,
406 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
407 beio->sg_segs[i].len) != 0)
408 break;
409 }
410 if (i < beio->num_segs)
411 ctl_set_sense(&io->scsiio,
412 /*current_error*/ 1,
413 /*sense_key*/ SSD_KEY_MISCOMPARE,
414 /*asc*/ 0x1D,
415 /*ascq*/ 0x00,
416 SSD_ELEM_NONE);
417 else
418 ctl_set_success(&io->scsiio);
419 }
420 } else if ((io->io_hdr.port_status != 0) &&
421 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
422 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
423 /*
424 * For hardware error sense keys, the sense key
425 * specific value is defined to be a retry count,
426 * but we use it to pass back an internal FETD
427 * error code. XXX KDM Hopefully the FETD is only
428 * using 16 bits for an error code, since that's
429 * all the space we have in the sks field.
430 */
431 ctl_set_internal_failure(&io->scsiio,
432 /*sks_valid*/ 1,
433 /*retry_count*/
434 io->io_hdr.port_status);
435 }
436
437 /*
438 * If this is a read, or a write with errors, it is done.
439 */
440 if ((beio->bio_cmd == BIO_READ)
441 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
442 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
443 ctl_complete_beio(beio);
444 return (0);
445 }
446
447 /*
448 * At this point, we have a write and the DMA completed
449 * successfully. We now have to queue it to the task queue to
450 * execute the backend I/O. That is because we do blocking
451 * memory allocations, and in the file backing case, blocking I/O.
452 * This move done routine is generally called in the SIM's
453 * interrupt context, and therefore we cannot block.
454 */
455 mtx_lock(&be_lun->queue_lock);
456 /*
457 * XXX KDM make sure that links is okay to use at this point.
458 * Otherwise, we either need to add another field to ctl_io_hdr,
459 * or deal with resource allocation here.
460 */
461 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
462 mtx_unlock(&be_lun->queue_lock);
463
464 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
465
466 return (0);
467}
468
469static void
470ctl_be_block_biodone(struct bio *bio)
471{
472 struct ctl_be_block_io *beio;
473 struct ctl_be_block_lun *be_lun;
474 union ctl_io *io;
475 int error;
476
477 beio = bio->bio_caller1;
478 be_lun = beio->lun;
479 io = beio->io;
480
481 DPRINTF("entered\n");
482
483 error = bio->bio_error;
484 mtx_lock(&be_lun->io_lock);
485 if (error != 0)
486 beio->num_errors++;
487
488 beio->num_bios_done++;
489
490 /*
491 * XXX KDM will this cause WITNESS to complain? Holding a lock
492 * during the free might cause it to complain.
493 */
494 g_destroy_bio(bio);
495
496 /*
497 * If the send complete bit isn't set, or we aren't the last I/O to
498 * complete, then we're done.
499 */
500 if ((beio->send_complete == 0)
501 || (beio->num_bios_done < beio->num_bios_sent)) {
502 mtx_unlock(&be_lun->io_lock);
503 return;
504 }
505
506 /*
507 * At this point, we've verified that we are the last I/O to
508 * complete, so it's safe to drop the lock.
509 */
510 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
511 beio->ds_tag_type, beio->ds_trans_type,
512 /*now*/ NULL, /*then*/&beio->ds_t0);
513 mtx_unlock(&be_lun->io_lock);
514
515 /*
516 * If there are any errors from the backing device, we fail the
517 * entire I/O with a medium error.
518 */
519 if (beio->num_errors > 0) {
520 if (error == EOPNOTSUPP) {
521 ctl_set_invalid_opcode(&io->scsiio);
522 } else if (error == ENOSPC) {
523 ctl_set_space_alloc_fail(&io->scsiio);
524 } else if (beio->bio_cmd == BIO_FLUSH) {
525 /* XXX KDM is there is a better error here? */
526 ctl_set_internal_failure(&io->scsiio,
527 /*sks_valid*/ 1,
528 /*retry_count*/ 0xbad2);
529 } else
530 ctl_set_medium_error(&io->scsiio);
531 ctl_complete_beio(beio);
532 return;
533 }
534
535 /*
536 * If this is a write, a flush, a delete or verify, we're all done.
537 * If this is a read, we can now send the data to the user.
538 */
539 if ((beio->bio_cmd == BIO_WRITE)
540 || (beio->bio_cmd == BIO_FLUSH)
541 || (beio->bio_cmd == BIO_DELETE)
542 || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
543 ctl_set_success(&io->scsiio);
544 ctl_complete_beio(beio);
545 } else {
546 if ((ARGS(io)->flags & CTL_LLF_READ) &&
547 beio->beio_cont == NULL)
548 ctl_set_success(&io->scsiio);
549#ifdef CTL_TIME_IO
550 getbintime(&io->io_hdr.dma_start_bt);
551#endif
552 ctl_datamove(io);
553 }
554}
555
556static void
557ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
558 struct ctl_be_block_io *beio)
559{
560 union ctl_io *io = beio->io;
561 struct mount *mountpoint;
562 int error, lock_flags;
563
564 DPRINTF("entered\n");
565
566 binuptime(&beio->ds_t0);
567 mtx_lock(&be_lun->io_lock);
568 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
569 mtx_unlock(&be_lun->io_lock);
570
571 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
572
573 if (MNT_SHARED_WRITES(mountpoint)
574 || ((mountpoint == NULL)
575 && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
576 lock_flags = LK_SHARED;
577 else
578 lock_flags = LK_EXCLUSIVE;
579
580 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
581
582 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread);
583 VOP_UNLOCK(be_lun->vn, 0);
584
585 vn_finished_write(mountpoint);
586
587 mtx_lock(&be_lun->io_lock);
588 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
589 beio->ds_tag_type, beio->ds_trans_type,
590 /*now*/ NULL, /*then*/&beio->ds_t0);
591 mtx_unlock(&be_lun->io_lock);
592
593 if (error == 0)
594 ctl_set_success(&io->scsiio);
595 else {
596 /* XXX KDM is there is a better error here? */
597 ctl_set_internal_failure(&io->scsiio,
598 /*sks_valid*/ 1,
599 /*retry_count*/ 0xbad1);
600 }
601
602 ctl_complete_beio(beio);
603}
604
605SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t");
606SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t");
607SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t");
608SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t");
609
610static void
611ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
612 struct ctl_be_block_io *beio)
613{
614 struct ctl_be_block_filedata *file_data;
615 union ctl_io *io;
616 struct uio xuio;
617 struct iovec *xiovec;
618 int flags;
619 int error, i;
620
621 DPRINTF("entered\n");
622
623 file_data = &be_lun->backend.file;
624 io = beio->io;
625 flags = 0;
626 if (ARGS(io)->flags & CTL_LLF_DPO)
627 flags |= IO_DIRECT;
628 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
629 flags |= IO_SYNC;
630
631 bzero(&xuio, sizeof(xuio));
632 if (beio->bio_cmd == BIO_READ) {
633 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
634 xuio.uio_rw = UIO_READ;
635 } else {
636 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
637 xuio.uio_rw = UIO_WRITE;
638 }
639 xuio.uio_offset = beio->io_offset;
640 xuio.uio_resid = beio->io_len;
641 xuio.uio_segflg = UIO_SYSSPACE;
642 xuio.uio_iov = beio->xiovecs;
643 xuio.uio_iovcnt = beio->num_segs;
644 xuio.uio_td = curthread;
645
646 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
647 xiovec->iov_base = beio->sg_segs[i].addr;
648 xiovec->iov_len = beio->sg_segs[i].len;
649 }
650
651 binuptime(&beio->ds_t0);
652 mtx_lock(&be_lun->io_lock);
653 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
654 mtx_unlock(&be_lun->io_lock);
655
656 if (beio->bio_cmd == BIO_READ) {
657 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
658
659 /*
660 * UFS pays attention to IO_DIRECT for reads. If the
661 * DIRECTIO option is configured into the kernel, it calls
662 * ffs_rawread(). But that only works for single-segment
663 * uios with user space addresses. In our case, with a
664 * kernel uio, it still reads into the buffer cache, but it
665 * will just try to release the buffer from the cache later
666 * on in ffs_read().
667 *
668 * ZFS does not pay attention to IO_DIRECT for reads.
669 *
670 * UFS does not pay attention to IO_SYNC for reads.
671 *
672 * ZFS pays attention to IO_SYNC (which translates into the
673 * Solaris define FRSYNC for zfs_read()) for reads. It
674 * attempts to sync the file before reading.
675 *
676 * So, to attempt to provide some barrier semantics in the
677 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC.
678 */
679 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
680
681 VOP_UNLOCK(be_lun->vn, 0);
682 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
683 } else {
684 struct mount *mountpoint;
685 int lock_flags;
686
687 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
688
689 if (MNT_SHARED_WRITES(mountpoint)
690 || ((mountpoint == NULL)
691 && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
692 lock_flags = LK_SHARED;
693 else
694 lock_flags = LK_EXCLUSIVE;
695
696 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
697
698 /*
699 * UFS pays attention to IO_DIRECT for writes. The write
700 * is done asynchronously. (Normally the write would just
701 * get put into cache.
702 *
703 * UFS pays attention to IO_SYNC for writes. It will
704 * attempt to write the buffer out synchronously if that
705 * flag is set.
706 *
707 * ZFS does not pay attention to IO_DIRECT for writes.
708 *
709 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
710 * for writes. It will flush the transaction from the
711 * cache before returning.
712 *
713 * So if we've got the BIO_ORDERED flag set, we want
714 * IO_SYNC in either the UFS or ZFS case.
715 */
716 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
717 VOP_UNLOCK(be_lun->vn, 0);
718
719 vn_finished_write(mountpoint);
720 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
721 }
722
723 mtx_lock(&be_lun->io_lock);
724 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
725 beio->ds_tag_type, beio->ds_trans_type,
726 /*now*/ NULL, /*then*/&beio->ds_t0);
727 mtx_unlock(&be_lun->io_lock);
728
729 /*
730 * If we got an error, set the sense data to "MEDIUM ERROR" and
731 * return the I/O to the user.
732 */
733 if (error != 0) {
734 char path_str[32];
735
736 ctl_scsi_path_string(io, path_str, sizeof(path_str));
737 printf("%s%s command returned errno %d\n", path_str,
738 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error);
739 if (error == ENOSPC) {
740 ctl_set_space_alloc_fail(&io->scsiio);
741 } else
742 ctl_set_medium_error(&io->scsiio);
743 ctl_complete_beio(beio);
744 return;
745 }
746
747 /*
748 * If this is a write or a verify, we're all done.
749 * If this is a read, we can now send the data to the user.
750 */
751 if ((beio->bio_cmd == BIO_WRITE) ||
752 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
753 ctl_set_success(&io->scsiio);
754 ctl_complete_beio(beio);
755 } else {
756 if ((ARGS(io)->flags & CTL_LLF_READ) &&
757 beio->beio_cont == NULL)
758 ctl_set_success(&io->scsiio);
759#ifdef CTL_TIME_IO
760 getbintime(&io->io_hdr.dma_start_bt);
761#endif
762 ctl_datamove(io);
763 }
764}
765
766static void
767ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
768 struct ctl_be_block_io *beio)
769{
770 union ctl_io *io = beio->io;
771 struct ctl_lba_len_flags *lbalen = ARGS(io);
772 struct scsi_get_lba_status_data *data;
773 off_t roff, off;
774 int error, status;
775
776 DPRINTF("entered\n");
777
778 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift;
779 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
780 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
781 0, curthread->td_ucred, curthread);
782 if (error == 0 && off > roff)
783 status = 0; /* mapped up to off */
784 else {
785 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
786 0, curthread->td_ucred, curthread);
787 if (error == 0 && off > roff)
788 status = 1; /* deallocated up to off */
789 else {
790 status = 0; /* unknown up to the end */
791 off = be_lun->size_bytes;
792 }
793 }
794 VOP_UNLOCK(be_lun->vn, 0);
795
796 off >>= be_lun->blocksize_shift;
797 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
798 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
799 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba),
800 data->descr[0].length);
801 data->descr[0].status = status;
802
803 ctl_complete_beio(beio);
804}
805
806static uint64_t
807ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
808{
809 struct vattr vattr;
810 struct statfs statfs;
811 int error;
812
813 if (be_lun->vn == NULL)
814 return (UINT64_MAX);
815 if (strcmp(attrname, "blocksused") == 0) {
816 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
817 if (error != 0)
818 return (UINT64_MAX);
819 return (vattr.va_bytes >> be_lun->blocksize_shift);
820 }
821 if (strcmp(attrname, "blocksavail") == 0) {
822 error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
823 if (error != 0)
824 return (UINT64_MAX);
825 return ((statfs.f_bavail * statfs.f_bsize) >>
826 be_lun->blocksize_shift);
827 }
828 return (UINT64_MAX);
829}
830
831static void
832ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
833 struct ctl_be_block_io *beio)
834{
835 struct ctl_be_block_devdata *dev_data;
836 union ctl_io *io;
837 struct uio xuio;
838 struct iovec *xiovec;
839 int flags;
840 int error, i;
841
842 DPRINTF("entered\n");
843
844 dev_data = &be_lun->backend.dev;
845 io = beio->io;
846 flags = 0;
847 if (ARGS(io)->flags & CTL_LLF_DPO)
848 flags |= IO_DIRECT;
849 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
850 flags |= IO_SYNC;
851
852 bzero(&xuio, sizeof(xuio));
853 if (beio->bio_cmd == BIO_READ) {
854 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
855 xuio.uio_rw = UIO_READ;
856 } else {
857 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
858 xuio.uio_rw = UIO_WRITE;
859 }
860 xuio.uio_offset = beio->io_offset;
861 xuio.uio_resid = beio->io_len;
862 xuio.uio_segflg = UIO_SYSSPACE;
863 xuio.uio_iov = beio->xiovecs;
864 xuio.uio_iovcnt = beio->num_segs;
865 xuio.uio_td = curthread;
866
867 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
868 xiovec->iov_base = beio->sg_segs[i].addr;
869 xiovec->iov_len = beio->sg_segs[i].len;
870 }
871
872 binuptime(&beio->ds_t0);
873 mtx_lock(&be_lun->io_lock);
874 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
875 mtx_unlock(&be_lun->io_lock);
876
877 if (beio->bio_cmd == BIO_READ) {
878 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags);
879 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
880 } else {
881 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags);
882 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
883 }
884
885 mtx_lock(&be_lun->io_lock);
886 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
887 beio->ds_tag_type, beio->ds_trans_type,
888 /*now*/ NULL, /*then*/&beio->ds_t0);
889 mtx_unlock(&be_lun->io_lock);
890
891 /*
892 * If we got an error, set the sense data to "MEDIUM ERROR" and
893 * return the I/O to the user.
894 */
895 if (error != 0) {
896 if (error == ENOSPC) {
897 ctl_set_space_alloc_fail(&io->scsiio);
898 } else
899 ctl_set_medium_error(&io->scsiio);
900 ctl_complete_beio(beio);
901 return;
902 }
903
904 /*
905 * If this is a write or a verify, we're all done.
906 * If this is a read, we can now send the data to the user.
907 */
908 if ((beio->bio_cmd == BIO_WRITE) ||
909 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
910 ctl_set_success(&io->scsiio);
911 ctl_complete_beio(beio);
912 } else {
913 if ((ARGS(io)->flags & CTL_LLF_READ) &&
914 beio->beio_cont == NULL)
915 ctl_set_success(&io->scsiio);
916#ifdef CTL_TIME_IO
917 getbintime(&io->io_hdr.dma_start_bt);
918#endif
919 ctl_datamove(io);
920 }
921}
922
923static void
924ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
925 struct ctl_be_block_io *beio)
926{
927 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev;
928 union ctl_io *io = beio->io;
929 struct ctl_lba_len_flags *lbalen = ARGS(io);
930 struct scsi_get_lba_status_data *data;
931 off_t roff, off;
932 int error, status;
933
934 DPRINTF("entered\n");
935
936 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift;
937 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE,
938 (caddr_t)&off, FREAD, curthread);
939 if (error == 0 && off > roff)
940 status = 0; /* mapped up to off */
941 else {
942 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA,
943 (caddr_t)&off, FREAD, curthread);
944 if (error == 0 && off > roff)
945 status = 1; /* deallocated up to off */
946 else {
947 status = 0; /* unknown up to the end */
948 off = be_lun->size_bytes;
949 }
950 }
951
952 off >>= be_lun->blocksize_shift;
953 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
954 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
955 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba),
956 data->descr[0].length);
957 data->descr[0].status = status;
958
959 ctl_complete_beio(beio);
960}
961
962static void
963ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
964 struct ctl_be_block_io *beio)
965{
966 struct bio *bio;
967 union ctl_io *io;
968 struct ctl_be_block_devdata *dev_data;
969
970 dev_data = &be_lun->backend.dev;
971 io = beio->io;
972
973 DPRINTF("entered\n");
974
975 /* This can't fail, it's a blocking allocation. */
976 bio = g_alloc_bio();
977
978 bio->bio_cmd = BIO_FLUSH;
979 bio->bio_flags |= BIO_ORDERED;
980 bio->bio_dev = dev_data->cdev;
981 bio->bio_offset = 0;
982 bio->bio_data = 0;
983 bio->bio_done = ctl_be_block_biodone;
984 bio->bio_caller1 = beio;
985 bio->bio_pblkno = 0;
986
987 /*
988 * We don't need to acquire the LUN lock here, because we are only
989 * sending one bio, and so there is no other context to synchronize
990 * with.
991 */
992 beio->num_bios_sent = 1;
993 beio->send_complete = 1;
994
995 binuptime(&beio->ds_t0);
996 mtx_lock(&be_lun->io_lock);
997 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
998 mtx_unlock(&be_lun->io_lock);
999
1000 (*dev_data->csw->d_strategy)(bio);
1001}
1002
1003static void
1004ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1005 struct ctl_be_block_io *beio,
1006 uint64_t off, uint64_t len, int last)
1007{
1008 struct bio *bio;
1009 struct ctl_be_block_devdata *dev_data;
1010 uint64_t maxlen;
1011
1012 dev_data = &be_lun->backend.dev;
1013 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize);
1014 while (len > 0) {
1015 bio = g_alloc_bio();
1016 bio->bio_cmd = BIO_DELETE;
1017 bio->bio_dev = dev_data->cdev;
1018 bio->bio_offset = off;
1019 bio->bio_length = MIN(len, maxlen);
1020 bio->bio_data = 0;
1021 bio->bio_done = ctl_be_block_biodone;
1022 bio->bio_caller1 = beio;
1023 bio->bio_pblkno = off / be_lun->blocksize;
1024
1025 off += bio->bio_length;
1026 len -= bio->bio_length;
1027
1028 mtx_lock(&be_lun->io_lock);
1029 beio->num_bios_sent++;
1030 if (last && len == 0)
1031 beio->send_complete = 1;
1032 mtx_unlock(&be_lun->io_lock);
1033
1034 (*dev_data->csw->d_strategy)(bio);
1035 }
1036}
1037
1038static void
1039ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1040 struct ctl_be_block_io *beio)
1041{
1042 union ctl_io *io;
1043 struct ctl_be_block_devdata *dev_data;
1044 struct ctl_ptr_len_flags *ptrlen;
1045 struct scsi_unmap_desc *buf, *end;
1046 uint64_t len;
1047
1048 dev_data = &be_lun->backend.dev;
1049 io = beio->io;
1050
1051 DPRINTF("entered\n");
1052
1053 binuptime(&beio->ds_t0);
1054 mtx_lock(&be_lun->io_lock);
1055 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1056 mtx_unlock(&be_lun->io_lock);
1057
1058 if (beio->io_offset == -1) {
1059 beio->io_len = 0;
1060 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1061 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1062 end = buf + ptrlen->len / sizeof(*buf);
1063 for (; buf < end; buf++) {
1064 len = (uint64_t)scsi_4btoul(buf->length) *
1065 be_lun->blocksize;
1066 beio->io_len += len;
1067 ctl_be_block_unmap_dev_range(be_lun, beio,
1068 scsi_8btou64(buf->lba) * be_lun->blocksize, len,
1069 (end - buf < 2) ? TRUE : FALSE);
1070 }
1071 } else
1072 ctl_be_block_unmap_dev_range(be_lun, beio,
1073 beio->io_offset, beio->io_len, TRUE);
1074}
1075
1076static void
1077ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1078 struct ctl_be_block_io *beio)
1079{
1080 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1081 int i;
1082 struct bio *bio;
1083 struct ctl_be_block_devdata *dev_data;
1084 off_t cur_offset;
1085 int max_iosize;
1086
1087 DPRINTF("entered\n");
1088
1089 dev_data = &be_lun->backend.dev;
1090
1091 /*
1092 * We have to limit our I/O size to the maximum supported by the
1093 * backend device. Hopefully it is MAXPHYS. If the driver doesn't
1094 * set it properly, use DFLTPHYS.
1095 */
1096 max_iosize = dev_data->cdev->si_iosize_max;
1097 if (max_iosize < PAGE_SIZE)
1098 max_iosize = DFLTPHYS;
1099
1100 cur_offset = beio->io_offset;
1101 for (i = 0; i < beio->num_segs; i++) {
1102 size_t cur_size;
1103 uint8_t *cur_ptr;
1104
1105 cur_size = beio->sg_segs[i].len;
1106 cur_ptr = beio->sg_segs[i].addr;
1107
1108 while (cur_size > 0) {
1109 /* This can't fail, it's a blocking allocation. */
1110 bio = g_alloc_bio();
1111
1112 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1113
1114 bio->bio_cmd = beio->bio_cmd;
1115 bio->bio_dev = dev_data->cdev;
1116 bio->bio_caller1 = beio;
1117 bio->bio_length = min(cur_size, max_iosize);
1118 bio->bio_offset = cur_offset;
1119 bio->bio_data = cur_ptr;
1120 bio->bio_done = ctl_be_block_biodone;
1121 bio->bio_pblkno = cur_offset / be_lun->blocksize;
1122
1123 cur_offset += bio->bio_length;
1124 cur_ptr += bio->bio_length;
1125 cur_size -= bio->bio_length;
1126
1127 TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1128 beio->num_bios_sent++;
1129 }
1130 }
1131 binuptime(&beio->ds_t0);
1132 mtx_lock(&be_lun->io_lock);
1133 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1134 beio->send_complete = 1;
1135 mtx_unlock(&be_lun->io_lock);
1136
1137 /*
1138 * Fire off all allocated requests!
1139 */
1140 while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1141 TAILQ_REMOVE(&queue, bio, bio_queue);
1142 (*dev_data->csw->d_strategy)(bio);
1143 }
1144}
1145
1146static uint64_t
1147ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1148{
1149 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev;
1150 struct diocgattr_arg arg;
1151 int error;
1152
1153 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL)
1154 return (UINT64_MAX);
1155 strlcpy(arg.name, attrname, sizeof(arg.name));
1156 arg.len = sizeof(arg.value.off);
1157 error = dev_data->csw->d_ioctl(dev_data->cdev,
1158 DIOCGATTR, (caddr_t)&arg, FREAD, curthread);
1159 if (error != 0)
1160 return (UINT64_MAX);
1161 return (arg.value.off);
1162}
1163
1164static void
1165ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1166{
1167 union ctl_io *io;
1168
1169 io = beio->io;
1170 ctl_free_beio(beio);
1171 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1172 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1173 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1174 ctl_config_write_done(io);
1175 return;
1176 }
1177
1178 ctl_be_block_config_write(io);
1179}
1180
1181static void
1182ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1183 union ctl_io *io)
1184{
1185 struct ctl_be_block_io *beio;
1186 struct ctl_be_block_softc *softc;
1187 struct ctl_lba_len_flags *lbalen;
1188 uint64_t len_left, lba;
1189 int i, seglen;
1190 uint8_t *buf, *end;
1191
1192 DPRINTF("entered\n");
1193
1194 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1195 softc = be_lun->softc;
1196 lbalen = ARGS(beio->io);
1197
1198 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1199 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1200 ctl_free_beio(beio);
1201 ctl_set_invalid_field(&io->scsiio,
1202 /*sks_valid*/ 1,
1203 /*command*/ 1,
1204 /*field*/ 1,
1205 /*bit_valid*/ 0,
1206 /*bit*/ 0);
1207 ctl_config_write_done(io);
1208 return;
1209 }
1210
1211 switch (io->scsiio.tag_type) {
1212 case CTL_TAG_ORDERED:
1213 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1214 break;
1215 case CTL_TAG_HEAD_OF_QUEUE:
1216 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1217 break;
1218 case CTL_TAG_UNTAGGED:
1219 case CTL_TAG_SIMPLE:
1220 case CTL_TAG_ACA:
1221 default:
1222 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1223 break;
1224 }
1225
1226 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1227 beio->io_offset = lbalen->lba * be_lun->blocksize;
1228 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize;
1229 beio->bio_cmd = BIO_DELETE;
1230 beio->ds_trans_type = DEVSTAT_FREE;
1231
1232 be_lun->unmap(be_lun, beio);
1233 return;
1234 }
1235
1236 beio->bio_cmd = BIO_WRITE;
1237 beio->ds_trans_type = DEVSTAT_WRITE;
1238
1239 DPRINTF("WRITE SAME at LBA %jx len %u\n",
1240 (uintmax_t)lbalen->lba, lbalen->len);
1241
1242 len_left = (uint64_t)lbalen->len * be_lun->blocksize;
1243 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1244
1245 /*
1246 * Setup the S/G entry for this chunk.
1247 */
1248 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1249 seglen -= seglen % be_lun->blocksize;
1250 beio->sg_segs[i].len = seglen;
1251 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1252
1253 DPRINTF("segment %d addr %p len %zd\n", i,
1254 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1255
1256 beio->num_segs++;
1257 len_left -= seglen;
1258
1259 buf = beio->sg_segs[i].addr;
1260 end = buf + seglen;
1261 for (; buf < end; buf += be_lun->blocksize) {
1262 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize);
1263 if (lbalen->flags & SWS_LBDATA)
1264 scsi_ulto4b(lbalen->lba + lba, buf);
1265 lba++;
1266 }
1267 }
1268
1269 beio->io_offset = lbalen->lba * be_lun->blocksize;
1270 beio->io_len = lba * be_lun->blocksize;
1271
1272 /* We can not do all in one run. Correct and schedule rerun. */
1273 if (len_left > 0) {
1274 lbalen->lba += lba;
1275 lbalen->len -= lba;
1276 beio->beio_cont = ctl_be_block_cw_done_ws;
1277 }
1278
1279 be_lun->dispatch(be_lun, beio);
1280}
1281
1282static void
1283ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1284 union ctl_io *io)
1285{
1286 struct ctl_be_block_io *beio;
1287 struct ctl_be_block_softc *softc;
1288 struct ctl_ptr_len_flags *ptrlen;
1289
1290 DPRINTF("entered\n");
1291
1292 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1293 softc = be_lun->softc;
1294 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1295
1296 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1297 ctl_free_beio(beio);
1298 ctl_set_invalid_field(&io->scsiio,
1299 /*sks_valid*/ 0,
1300 /*command*/ 1,
1301 /*field*/ 0,
1302 /*bit_valid*/ 0,
1303 /*bit*/ 0);
1304 ctl_config_write_done(io);
1305 return;
1306 }
1307
1308 switch (io->scsiio.tag_type) {
1309 case CTL_TAG_ORDERED:
1310 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1311 break;
1312 case CTL_TAG_HEAD_OF_QUEUE:
1313 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1314 break;
1315 case CTL_TAG_UNTAGGED:
1316 case CTL_TAG_SIMPLE:
1317 case CTL_TAG_ACA:
1318 default:
1319 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1320 break;
1321 }
1322
1323 beio->io_len = 0;
1324 beio->io_offset = -1;
1325
1326 beio->bio_cmd = BIO_DELETE;
1327 beio->ds_trans_type = DEVSTAT_FREE;
1328
1329 DPRINTF("UNMAP\n");
1330
1331 be_lun->unmap(be_lun, beio);
1332}
1333
1334static void
1335ctl_be_block_cr_done(struct ctl_be_block_io *beio)
1336{
1337 union ctl_io *io;
1338
1339 io = beio->io;
1340 ctl_free_beio(beio);
1341 ctl_config_read_done(io);
1342}
1343
1344static void
1345ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1346 union ctl_io *io)
1347{
1348 struct ctl_be_block_io *beio;
1349 struct ctl_be_block_softc *softc;
1350
1351 DPRINTF("entered\n");
1352
1353 softc = be_lun->softc;
1354 beio = ctl_alloc_beio(softc);
1355 beio->io = io;
1356 beio->lun = be_lun;
1357 beio->beio_cont = ctl_be_block_cr_done;
1358 PRIV(io)->ptr = (void *)beio;
1359
1360 switch (io->scsiio.cdb[0]) {
1361 case SERVICE_ACTION_IN: /* GET LBA STATUS */
1362 beio->bio_cmd = -1;
1363 beio->ds_trans_type = DEVSTAT_NO_DATA;
1364 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1365 beio->io_len = 0;
1366 if (be_lun->get_lba_status)
1367 be_lun->get_lba_status(be_lun, beio);
1368 else
1369 ctl_be_block_cr_done(beio);
1370 break;
1371 default:
1372 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1373 break;
1374 }
1375}
1376
1377static void
1378ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1379{
1380 union ctl_io *io;
1381
1382 io = beio->io;
1383 ctl_free_beio(beio);
1384 ctl_config_write_done(io);
1385}
1386
1387static void
1388ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1389 union ctl_io *io)
1390{
1391 struct ctl_be_block_io *beio;
1392 struct ctl_be_block_softc *softc;
1393
1394 DPRINTF("entered\n");
1395
1396 softc = be_lun->softc;
1397 beio = ctl_alloc_beio(softc);
1398 beio->io = io;
1399 beio->lun = be_lun;
1400 beio->beio_cont = ctl_be_block_cw_done;
1401 PRIV(io)->ptr = (void *)beio;
1402
1403 switch (io->scsiio.cdb[0]) {
1404 case SYNCHRONIZE_CACHE:
1405 case SYNCHRONIZE_CACHE_16:
1406 beio->bio_cmd = BIO_FLUSH;
1407 beio->ds_trans_type = DEVSTAT_NO_DATA;
1408 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1409 beio->io_len = 0;
1410 be_lun->lun_flush(be_lun, beio);
1411 break;
1412 case WRITE_SAME_10:
1413 case WRITE_SAME_16:
1414 ctl_be_block_cw_dispatch_ws(be_lun, io);
1415 break;
1416 case UNMAP:
1417 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1418 break;
1419 default:
1420 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1421 break;
1422 }
1423}
1424
1425SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t");
1426SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t");
1427SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t");
1428SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t");
1429
1430static void
1431ctl_be_block_next(struct ctl_be_block_io *beio)
1432{
1433 struct ctl_be_block_lun *be_lun;
1434 union ctl_io *io;
1435
1436 io = beio->io;
1437 be_lun = beio->lun;
1438 ctl_free_beio(beio);
1439 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1440 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1441 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1442 ctl_data_submit_done(io);
1443 return;
1444 }
1445
1446 io->io_hdr.status &= ~CTL_STATUS_MASK;
1447 io->io_hdr.status |= CTL_STATUS_NONE;
1448
1449 mtx_lock(&be_lun->queue_lock);
1450 /*
1451 * XXX KDM make sure that links is okay to use at this point.
1452 * Otherwise, we either need to add another field to ctl_io_hdr,
1453 * or deal with resource allocation here.
1454 */
1455 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1456 mtx_unlock(&be_lun->queue_lock);
1457
1458 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1459}
1460
1461static void
1462ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1463 union ctl_io *io)
1464{
1465 struct ctl_be_block_io *beio;
1466 struct ctl_be_block_softc *softc;
1467 struct ctl_lba_len_flags *lbalen;
1468 struct ctl_ptr_len_flags *bptrlen;
1469 uint64_t len_left, lbas;
1470 int i;
1471
1472 softc = be_lun->softc;
1473
1474 DPRINTF("entered\n");
1475
1476 lbalen = ARGS(io);
1477 if (lbalen->flags & CTL_LLF_WRITE) {
1478 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0);
1479 } else {
1480 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0);
1481 }
1482
1483 beio = ctl_alloc_beio(softc);
1484 beio->io = io;
1485 beio->lun = be_lun;
1486 bptrlen = PRIV(io);
1487 bptrlen->ptr = (void *)beio;
1488
1489 switch (io->scsiio.tag_type) {
1490 case CTL_TAG_ORDERED:
1491 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1492 break;
1493 case CTL_TAG_HEAD_OF_QUEUE:
1494 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1495 break;
1496 case CTL_TAG_UNTAGGED:
1497 case CTL_TAG_SIMPLE:
1498 case CTL_TAG_ACA:
1499 default:
1500 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1501 break;
1502 }
1503
1504 if (lbalen->flags & CTL_LLF_WRITE) {
1505 beio->bio_cmd = BIO_WRITE;
1506 beio->ds_trans_type = DEVSTAT_WRITE;
1507 } else {
1508 beio->bio_cmd = BIO_READ;
1509 beio->ds_trans_type = DEVSTAT_READ;
1510 }
1511
1512 DPRINTF("%s at LBA %jx len %u @%ju\n",
1513 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1514 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1515 if (lbalen->flags & CTL_LLF_COMPARE)
1516 lbas = CTLBLK_HALF_IO_SIZE;
1517 else
1518 lbas = CTLBLK_MAX_IO_SIZE;
1519 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize);
1520 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize;
1521 beio->io_len = lbas * be_lun->blocksize;
1522 bptrlen->len += lbas;
1523
1524 for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1525 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1526 i, CTLBLK_MAX_SEGS));
1527
1528 /*
1529 * Setup the S/G entry for this chunk.
1530 */
1531 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1532 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1533
1534 DPRINTF("segment %d addr %p len %zd\n", i,
1535 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1536
1537 /* Set up second segment for compare operation. */
1538 if (lbalen->flags & CTL_LLF_COMPARE) {
1539 beio->sg_segs[i + CTLBLK_HALF_SEGS].len =
1540 beio->sg_segs[i].len;
1541 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr =
1542 uma_zalloc(be_lun->lun_zone, M_WAITOK);
1543 }
1544
1545 beio->num_segs++;
1546 len_left -= beio->sg_segs[i].len;
1547 }
1548 if (bptrlen->len < lbalen->len)
1549 beio->beio_cont = ctl_be_block_next;
1550 io->scsiio.be_move_done = ctl_be_block_move_done;
1551 /* For compare we have separate S/G lists for read and datamove. */
1552 if (lbalen->flags & CTL_LLF_COMPARE)
1553 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1554 else
1555 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1556 io->scsiio.kern_data_len = beio->io_len;
1557 io->scsiio.kern_data_resid = 0;
1558 io->scsiio.kern_sg_entries = beio->num_segs;
1559 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST;
1560
1561 /*
1562 * For the read case, we need to read the data into our buffers and
1563 * then we can send it back to the user. For the write case, we
1564 * need to get the data from the user first.
1565 */
1566 if (beio->bio_cmd == BIO_READ) {
1567 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0);
1568 be_lun->dispatch(be_lun, beio);
1569 } else {
1570 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0);
1571#ifdef CTL_TIME_IO
1572 getbintime(&io->io_hdr.dma_start_bt);
1573#endif
1574 ctl_datamove(io);
1575 }
1576}
1577
1578static void
1579ctl_be_block_worker(void *context, int pending)
1580{
1581 struct ctl_be_block_lun *be_lun;
1582 struct ctl_be_block_softc *softc;
1583 union ctl_io *io;
1584
1585 be_lun = (struct ctl_be_block_lun *)context;
1586 softc = be_lun->softc;
1587
1588 DPRINTF("entered\n");
1589
1590 mtx_lock(&be_lun->queue_lock);
1591 for (;;) {
1592 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1593 if (io != NULL) {
1594 struct ctl_be_block_io *beio;
1595
1596 DPRINTF("datamove queue\n");
1597
1598 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1599 ctl_io_hdr, links);
1600
1601 mtx_unlock(&be_lun->queue_lock);
1602
1603 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1604
1605 be_lun->dispatch(be_lun, beio);
1606
1607 mtx_lock(&be_lun->queue_lock);
1608 continue;
1609 }
1610 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1611 if (io != NULL) {
1612 DPRINTF("config write queue\n");
1613 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1614 ctl_io_hdr, links);
1615 mtx_unlock(&be_lun->queue_lock);
1616 ctl_be_block_cw_dispatch(be_lun, io);
1617 mtx_lock(&be_lun->queue_lock);
1618 continue;
1619 }
1620 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
1621 if (io != NULL) {
1622 DPRINTF("config read queue\n");
1623 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr,
1624 ctl_io_hdr, links);
1625 mtx_unlock(&be_lun->queue_lock);
1626 ctl_be_block_cr_dispatch(be_lun, io);
1627 mtx_lock(&be_lun->queue_lock);
1628 continue;
1629 }
1630 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1631 if (io != NULL) {
1632 DPRINTF("input queue\n");
1633
1634 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1635 ctl_io_hdr, links);
1636 mtx_unlock(&be_lun->queue_lock);
1637
1638 /*
1639 * We must drop the lock, since this routine and
1640 * its children may sleep.
1641 */
1642 ctl_be_block_dispatch(be_lun, io);
1643
1644 mtx_lock(&be_lun->queue_lock);
1645 continue;
1646 }
1647
1648 /*
1649 * If we get here, there is no work left in the queues, so
1650 * just break out and let the task queue go to sleep.
1651 */
1652 break;
1653 }
1654 mtx_unlock(&be_lun->queue_lock);
1655}
1656
1657/*
1658 * Entry point from CTL to the backend for I/O. We queue everything to a
1659 * work thread, so this just puts the I/O on a queue and wakes up the
1660 * thread.
1661 */
1662static int
1663ctl_be_block_submit(union ctl_io *io)
1664{
1665 struct ctl_be_block_lun *be_lun;
1666 struct ctl_be_lun *ctl_be_lun;
1667
1668 DPRINTF("entered\n");
1669
1670 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
1671 CTL_PRIV_BACKEND_LUN].ptr;
1672 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
1673
1674 /*
1675 * Make sure we only get SCSI I/O.
1676 */
1677 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1678 "%#x) encountered", io->io_hdr.io_type));
1679
1680 PRIV(io)->len = 0;
1681
1682 mtx_lock(&be_lun->queue_lock);
1683 /*
1684 * XXX KDM make sure that links is okay to use at this point.
1685 * Otherwise, we either need to add another field to ctl_io_hdr,
1686 * or deal with resource allocation here.
1687 */
1688 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1689 mtx_unlock(&be_lun->queue_lock);
1690 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1691
1692 return (CTL_RETVAL_COMPLETE);
1693}
1694
1695static int
1696ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1697 int flag, struct thread *td)
1698{
1699 struct ctl_be_block_softc *softc;
1700 int error;
1701
1702 softc = &backend_block_softc;
1703
1704 error = 0;
1705
1706 switch (cmd) {
1707 case CTL_LUN_REQ: {
1708 struct ctl_lun_req *lun_req;
1709
1710 lun_req = (struct ctl_lun_req *)addr;
1711
1712 switch (lun_req->reqtype) {
1713 case CTL_LUNREQ_CREATE:
1714 error = ctl_be_block_create(softc, lun_req);
1715 break;
1716 case CTL_LUNREQ_RM:
1717 error = ctl_be_block_rm(softc, lun_req);
1718 break;
1719 case CTL_LUNREQ_MODIFY:
1720 error = ctl_be_block_modify(softc, lun_req);
1721 break;
1722 default:
1723 lun_req->status = CTL_LUN_ERROR;
1724 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1725 "invalid LUN request type %d",
1726 lun_req->reqtype);
1727 break;
1728 }
1729 break;
1730 }
1731 default:
1732 error = ENOTTY;
1733 break;
1734 }
1735
1736 return (error);
1737}
1738
1739static int
1740ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1741{
1742 struct ctl_be_block_filedata *file_data;
1743 struct ctl_lun_create_params *params;
1744 char *value;
1745 struct vattr vattr;
1746 off_t ps, pss, po, pos, us, uss, uo, uos;
1747 int error;
1748
1749 error = 0;
1750 file_data = &be_lun->backend.file;
1751 params = &be_lun->params;
1752
1753 be_lun->dev_type = CTL_BE_BLOCK_FILE;
1754 be_lun->dispatch = ctl_be_block_dispatch_file;
1755 be_lun->lun_flush = ctl_be_block_flush_file;
1756 be_lun->get_lba_status = ctl_be_block_gls_file;
1757 be_lun->getattr = ctl_be_block_getattr_file;
1758
1759 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1760 if (error != 0) {
1761 snprintf(req->error_str, sizeof(req->error_str),
1762 "error calling VOP_GETATTR() for file %s",
1763 be_lun->dev_path);
1764 return (error);
1765 }
1766
1767 /*
1768 * Verify that we have the ability to upgrade to exclusive
1769 * access on this file so we can trap errors at open instead
1770 * of reporting them during first access.
1771 */
1772 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) {
1773 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY);
1774 if (be_lun->vn->v_iflag & VI_DOOMED) {
1775 error = EBADF;
1776 snprintf(req->error_str, sizeof(req->error_str),
1777 "error locking file %s", be_lun->dev_path);
1778 return (error);
1779 }
1780 }
1781
1782
1783 file_data->cred = crhold(curthread->td_ucred);
1784 if (params->lun_size_bytes != 0)
1785 be_lun->size_bytes = params->lun_size_bytes;
1786 else
1787 be_lun->size_bytes = vattr.va_size;
1788 /*
1789 * We set the multi thread flag for file operations because all
1790 * filesystems (in theory) are capable of allowing multiple readers
1791 * of a file at once. So we want to get the maximum possible
1792 * concurrency.
1793 */
1794 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD;
1795
1796 /*
1797 * For files we can use any logical block size. Prefer 512 bytes
1798 * for compatibility reasons. If file's vattr.va_blocksize
1799 * (preferred I/O block size) is bigger and multiple to chosen
1800 * logical block size -- report it as physical block size.
1801 */
1802 if (params->blocksize_bytes != 0)
1803 be_lun->blocksize = params->blocksize_bytes;
1804 else
1805 be_lun->blocksize = 512;
1806
1807 us = ps = vattr.va_blocksize;
1808 uo = po = 0;
1809
1810 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize");
1811 if (value != NULL)
1812 ctl_expand_number(value, &ps);
1813 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset");
1814 if (value != NULL)
1815 ctl_expand_number(value, &po);
1816 pss = ps / be_lun->blocksize;
1817 pos = po / be_lun->blocksize;
1818 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) &&
1819 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) {
1820 be_lun->pblockexp = fls(pss) - 1;
1821 be_lun->pblockoff = (pss - pos) % pss;
1822 }
1823
1824 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize");
1825 if (value != NULL)
1826 ctl_expand_number(value, &us);
1827 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset");
1828 if (value != NULL)
1829 ctl_expand_number(value, &uo);
1830 uss = us / be_lun->blocksize;
1831 uos = uo / be_lun->blocksize;
1832 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) &&
1833 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) {
1834 be_lun->ublockexp = fls(uss) - 1;
1835 be_lun->ublockoff = (uss - uos) % uss;
1836 }
1837
1838 /*
1839 * Sanity check. The media size has to be at least one
1840 * sector long.
1841 */
1842 if (be_lun->size_bytes < be_lun->blocksize) {
1843 error = EINVAL;
1844 snprintf(req->error_str, sizeof(req->error_str),
1845 "file %s size %ju < block size %u", be_lun->dev_path,
1846 (uintmax_t)be_lun->size_bytes, be_lun->blocksize);
1847 }
1848 return (error);
1849}
1850
1851static int
1852ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1853{
1854 struct ctl_lun_create_params *params;
1855 struct vattr vattr;
1856 struct cdev *dev;
1857 struct cdevsw *devsw;
1858 char *value;
1859 int error;
1860 off_t ps, pss, po, pos, us, uss, uo, uos;
1861
1862 params = &be_lun->params;
1863
1864 be_lun->dev_type = CTL_BE_BLOCK_DEV;
1865 be_lun->backend.dev.cdev = be_lun->vn->v_rdev;
1866 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev,
1867 &be_lun->backend.dev.dev_ref);
1868 if (be_lun->backend.dev.csw == NULL)
1869 panic("Unable to retrieve device switch");
1870 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) {
1871 be_lun->dispatch = ctl_be_block_dispatch_zvol;
1872 be_lun->get_lba_status = ctl_be_block_gls_zvol;
1873 } else
1874 be_lun->dispatch = ctl_be_block_dispatch_dev;
1875 be_lun->lun_flush = ctl_be_block_flush_dev;
1876 be_lun->unmap = ctl_be_block_unmap_dev;
1877 be_lun->getattr = ctl_be_block_getattr_dev;
1878
1879 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED);
1880 if (error) {
1881 snprintf(req->error_str, sizeof(req->error_str),
1882 "error getting vnode attributes for device %s",
1883 be_lun->dev_path);
1884 return (error);
1885 }
1886
1887 dev = be_lun->vn->v_rdev;
1888 devsw = dev->si_devsw;
1889 if (!devsw->d_ioctl) {
1890 snprintf(req->error_str, sizeof(req->error_str),
1891 "no d_ioctl for device %s!",
1892 be_lun->dev_path);
1893 return (ENODEV);
1894 }
1895
1896 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE,
1897 (caddr_t)&be_lun->blocksize, FREAD,
1898 curthread);
1899 if (error) {
1900 snprintf(req->error_str, sizeof(req->error_str),
1901 "error %d returned for DIOCGSECTORSIZE ioctl "
1902 "on %s!", error, be_lun->dev_path);
1903 return (error);
1904 }
1905
1906 /*
1907 * If the user has asked for a blocksize that is greater than the
1908 * backing device's blocksize, we can do it only if the blocksize
1909 * the user is asking for is an even multiple of the underlying
1910 * device's blocksize.
1911 */
1912 if ((params->blocksize_bytes != 0)
1913 && (params->blocksize_bytes > be_lun->blocksize)) {
1914 uint32_t bs_multiple, tmp_blocksize;
1915
1916 bs_multiple = params->blocksize_bytes / be_lun->blocksize;
1917
1918 tmp_blocksize = bs_multiple * be_lun->blocksize;
1919
1920 if (tmp_blocksize == params->blocksize_bytes) {
1921 be_lun->blocksize = params->blocksize_bytes;
1922 } else {
1923 snprintf(req->error_str, sizeof(req->error_str),
1924 "requested blocksize %u is not an even "
1925 "multiple of backing device blocksize %u",
1926 params->blocksize_bytes,
1927 be_lun->blocksize);
1928 return (EINVAL);
1929
1930 }
1931 } else if ((params->blocksize_bytes != 0)
1932 && (params->blocksize_bytes != be_lun->blocksize)) {
1933 snprintf(req->error_str, sizeof(req->error_str),
1934 "requested blocksize %u < backing device "
1935 "blocksize %u", params->blocksize_bytes,
1936 be_lun->blocksize);
1937 return (EINVAL);
1938 }
1939
1940 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE,
1941 (caddr_t)&be_lun->size_bytes, FREAD,
1942 curthread);
1943 if (error) {
1944 snprintf(req->error_str, sizeof(req->error_str),
1945 "error %d returned for DIOCGMEDIASIZE "
1946 " ioctl on %s!", error,
1947 be_lun->dev_path);
1948 return (error);
1949 }
1950
1951 if (params->lun_size_bytes != 0) {
1952 if (params->lun_size_bytes > be_lun->size_bytes) {
1953 snprintf(req->error_str, sizeof(req->error_str),
1954 "requested LUN size %ju > backing device "
1955 "size %ju",
1956 (uintmax_t)params->lun_size_bytes,
1957 (uintmax_t)be_lun->size_bytes);
1958 return (EINVAL);
1959 }
1960
1961 be_lun->size_bytes = params->lun_size_bytes;
1962 }
1963
1964 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE,
1965 (caddr_t)&ps, FREAD, curthread);
1966 if (error)
1967 ps = po = 0;
1968 else {
1969 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET,
1970 (caddr_t)&po, FREAD, curthread);
1971 if (error)
1972 po = 0;
1973 }
1974 us = ps;
1975 uo = po;
1976
1977 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize");
1978 if (value != NULL)
1979 ctl_expand_number(value, &ps);
1980 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset");
1981 if (value != NULL)
1982 ctl_expand_number(value, &po);
1983 pss = ps / be_lun->blocksize;
1984 pos = po / be_lun->blocksize;
1985 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) &&
1986 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) {
1987 be_lun->pblockexp = fls(pss) - 1;
1988 be_lun->pblockoff = (pss - pos) % pss;
1989 }
1990
1991 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize");
1992 if (value != NULL)
1993 ctl_expand_number(value, &us);
1994 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset");
1995 if (value != NULL)
1996 ctl_expand_number(value, &uo);
1997 uss = us / be_lun->blocksize;
1998 uos = uo / be_lun->blocksize;
1999 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) &&
2000 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) {
2001 be_lun->ublockexp = fls(uss) - 1;
2002 be_lun->ublockoff = (uss - uos) % uss;
2003 }
2004
2005 return (0);
2006}
2007
2008static int
2009ctl_be_block_close(struct ctl_be_block_lun *be_lun)
2010{
2011 DROP_GIANT();
2012 if (be_lun->vn) {
2013 int flags = FREAD | FWRITE;
2014
2015 switch (be_lun->dev_type) {
2016 case CTL_BE_BLOCK_DEV:
2017 if (be_lun->backend.dev.csw) {
2018 dev_relthread(be_lun->backend.dev.cdev,
2019 be_lun->backend.dev.dev_ref);
2020 be_lun->backend.dev.csw = NULL;
2021 be_lun->backend.dev.cdev = NULL;
2022 }
2023 break;
2024 case CTL_BE_BLOCK_FILE:
2025 break;
2026 case CTL_BE_BLOCK_NONE:
2027 break;
2028 default:
2029 panic("Unexpected backend type.");
2030 break;
2031 }
2032
2033 (void)vn_close(be_lun->vn, flags, NOCRED, curthread);
2034 be_lun->vn = NULL;
2035
2036 switch (be_lun->dev_type) {
2037 case CTL_BE_BLOCK_DEV:
2038 break;
2039 case CTL_BE_BLOCK_FILE:
2040 if (be_lun->backend.file.cred != NULL) {
2041 crfree(be_lun->backend.file.cred);
2042 be_lun->backend.file.cred = NULL;
2043 }
2044 break;
2045 case CTL_BE_BLOCK_NONE:
2046 break;
2047 default:
2048 panic("Unexpected backend type.");
2049 break;
2050 }
2051 be_lun->dev_type = CTL_BE_BLOCK_NONE;
2052 }
2053 PICKUP_GIANT();
2054
2055 return (0);
2056}
2057
2058static int
2059ctl_be_block_open(struct ctl_be_block_softc *softc,
2060 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2061{
2062 struct nameidata nd;
2063 int flags;
2064 int error;
2065
2066 /*
2067 * XXX KDM allow a read-only option?
2068 */
2069 flags = FREAD | FWRITE;
2070 error = 0;
2071
2072 if (rootvnode == NULL) {
2073 snprintf(req->error_str, sizeof(req->error_str),
2074 "Root filesystem is not mounted");
2075 return (1);
2076 }
2077
2078 if (!curthread->td_proc->p_fd->fd_cdir) {
2079 curthread->td_proc->p_fd->fd_cdir = rootvnode;
2080 VREF(rootvnode);
2081 }
2082 if (!curthread->td_proc->p_fd->fd_rdir) {
2083 curthread->td_proc->p_fd->fd_rdir = rootvnode;
2084 VREF(rootvnode);
2085 }
2086 if (!curthread->td_proc->p_fd->fd_jdir) {
2087 curthread->td_proc->p_fd->fd_jdir = rootvnode;
2088 VREF(rootvnode);
2089 }
2090
2091 again:
2092 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread);
2093 error = vn_open(&nd, &flags, 0, NULL);
2094 if (error) {
2095 /*
2096 * This is the only reasonable guess we can make as far as
2097 * path if the user doesn't give us a fully qualified path.
2098 * If they want to specify a file, they need to specify the
2099 * full path.
2100 */
2101 if (be_lun->dev_path[0] != '/') {
2102 char *dev_path = "/dev/";
2103 char *dev_name;
2104
2105 /* Try adding device path at beginning of name */
2106 dev_name = malloc(strlen(be_lun->dev_path)
2107 + strlen(dev_path) + 1,
2108 M_CTLBLK, M_WAITOK);
2109 if (dev_name) {
2110 sprintf(dev_name, "%s%s", dev_path,
2111 be_lun->dev_path);
2112 free(be_lun->dev_path, M_CTLBLK);
2113 be_lun->dev_path = dev_name;
2114 goto again;
2115 }
2116 }
2117 snprintf(req->error_str, sizeof(req->error_str),
2118 "error opening %s: %d", be_lun->dev_path, error);
2119 return (error);
2120 }
2121
2122 NDFREE(&nd, NDF_ONLY_PNBUF);
2123
2124 be_lun->vn = nd.ni_vp;
2125
2126 /* We only support disks and files. */
2127 if (vn_isdisk(be_lun->vn, &error)) {
2128 error = ctl_be_block_open_dev(be_lun, req);
2129 } else if (be_lun->vn->v_type == VREG) {
2130 error = ctl_be_block_open_file(be_lun, req);
2131 } else {
2132 error = EINVAL;
2133 snprintf(req->error_str, sizeof(req->error_str),
2134 "%s is not a disk or plain file", be_lun->dev_path);
2135 }
2136 VOP_UNLOCK(be_lun->vn, 0);
2137
2138 if (error != 0) {
2139 ctl_be_block_close(be_lun);
2140 return (error);
2141 }
2142
2143 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1;
2144 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift;
2145
2146 return (0);
2147}
2148
2149static int
2150ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2151{
2152 struct ctl_be_block_lun *be_lun;
2153 struct ctl_lun_create_params *params;
2154 char num_thread_str[16];
2155 char tmpstr[32];
2156 char *value;
2157 int retval, num_threads, unmap;
2158 int tmp_num_threads;
2159
2160 params = &req->reqdata.create;
2161 retval = 0;
2162 req->status = CTL_LUN_OK;
2163
2164 num_threads = cbb_num_threads;
2165
2166 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
2167
2168 be_lun->params = req->reqdata.create;
2169 be_lun->softc = softc;
2170 STAILQ_INIT(&be_lun->input_queue);
2171 STAILQ_INIT(&be_lun->config_read_queue);
2172 STAILQ_INIT(&be_lun->config_write_queue);
2173 STAILQ_INIT(&be_lun->datamove_queue);
2174 sprintf(be_lun->lunname, "cblk%d", softc->num_luns);
2175 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF);
2176 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF);
2177 ctl_init_opts(&be_lun->ctl_be_lun.options,
2178 req->num_be_args, req->kern_be_args);
2179
2180 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG,
2181 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
2182
2183 if (be_lun->lun_zone == NULL) {
2184 snprintf(req->error_str, sizeof(req->error_str),
2185 "error allocating UMA zone");
2186 goto bailout_error;
2187 }
2188
2189 if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
2190 be_lun->ctl_be_lun.lun_type = params->device_type;
2191 else
2192 be_lun->ctl_be_lun.lun_type = T_DIRECT;
2193
2194 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) {
2195 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file");
2196 if (value == NULL) {
2197 snprintf(req->error_str, sizeof(req->error_str),
2198 "no file argument specified");
2199 goto bailout_error;
2200 }
2201 be_lun->dev_path = strdup(value, M_CTLBLK);
2202 be_lun->blocksize = 512;
2203 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1;
2204
2205 retval = ctl_be_block_open(softc, be_lun, req);
2206 if (retval != 0) {
2207 retval = 0;
2208 req->status = CTL_LUN_WARNING;
2209 }
2210 } else {
2211 /*
2212 * For processor devices, we don't have any size.
2213 */
2214 be_lun->blocksize = 0;
2215 be_lun->pblockexp = 0;
2216 be_lun->pblockoff = 0;
2217 be_lun->ublockexp = 0;
2218 be_lun->ublockoff = 0;
2219 be_lun->size_blocks = 0;
2220 be_lun->size_bytes = 0;
2221 be_lun->ctl_be_lun.maxlba = 0;
2222
2223 /*
2224 * Default to just 1 thread for processor devices.
2225 */
2226 num_threads = 1;
2227 }
2228
2229 /*
2230 * XXX This searching loop might be refactored to be combined with
2231 * the loop above,
2232 */
2233 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads");
2234 if (value != NULL) {
2235 tmp_num_threads = strtol(value, NULL, 0);
2236
2237 /*
2238 * We don't let the user specify less than one
2239 * thread, but hope he's clueful enough not to
2240 * specify 1000 threads.
2241 */
2242 if (tmp_num_threads < 1) {
2243 snprintf(req->error_str, sizeof(req->error_str),
2244 "invalid number of threads %s",
2245 num_thread_str);
2246 goto bailout_error;
2247 }
2248 num_threads = tmp_num_threads;
2249 }
2250 unmap = (be_lun->dispatch == ctl_be_block_dispatch_zvol);
2251 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap");
2252 if (value != NULL)
2253 unmap = (strcmp(value, "on") == 0);
2254
2255 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED;
2256 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY;
2257 if (be_lun->vn == NULL)
2258 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE;
2259 if (unmap)
2260 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP;
2261 if (be_lun->dispatch != ctl_be_block_dispatch_dev)
2262 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_SERSEQ_READ;
2263 be_lun->ctl_be_lun.be_lun = be_lun;
2264 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ?
2265 0 : (be_lun->size_blocks - 1);
2266 be_lun->ctl_be_lun.blocksize = be_lun->blocksize;
2267 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp;
2268 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff;
2269 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp;
2270 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff;
2271 if (be_lun->dispatch == ctl_be_block_dispatch_zvol &&
2272 be_lun->blocksize != 0)
2273 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE /
2274 be_lun->blocksize;
2275 /* Tell the user the blocksize we ended up using */
2276 params->lun_size_bytes = be_lun->size_bytes;
2277 params->blocksize_bytes = be_lun->blocksize;
2278 if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2279 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id;
2280 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ;
2281 } else
2282 be_lun->ctl_be_lun.req_lun_id = 0;
2283
2284 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown;
2285 be_lun->ctl_be_lun.lun_config_status =
2286 ctl_be_block_lun_config_status;
2287 be_lun->ctl_be_lun.be = &ctl_be_block_driver;
2288
2289 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2290 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d",
2291 softc->num_luns);
2292 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr,
2293 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
2294 sizeof(tmpstr)));
2295
2296 /* Tell the user what we used for a serial number */
2297 strncpy((char *)params->serial_num, tmpstr,
2298 ctl_min(sizeof(params->serial_num), sizeof(tmpstr)));
2299 } else {
2300 strncpy((char *)be_lun->ctl_be_lun.serial_num,
2301 params->serial_num,
2302 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
2303 sizeof(params->serial_num)));
2304 }
2305 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2306 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns);
2307 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr,
2308 ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2309 sizeof(tmpstr)));
2310
2311 /* Tell the user what we used for a device ID */
2312 strncpy((char *)params->device_id, tmpstr,
2313 ctl_min(sizeof(params->device_id), sizeof(tmpstr)));
2314 } else {
2315 strncpy((char *)be_lun->ctl_be_lun.device_id,
2316 params->device_id,
2317 ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2318 sizeof(params->device_id)));
2319 }
2320
2321 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2322
2323 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK,
2324 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2325
2326 if (be_lun->io_taskqueue == NULL) {
2327 snprintf(req->error_str, sizeof(req->error_str),
2328 "unable to create taskqueue");
2329 goto bailout_error;
2330 }
2331
2332 /*
2333 * Note that we start the same number of threads by default for
2334 * both the file case and the block device case. For the file
2335 * case, we need multiple threads to allow concurrency, because the
2336 * vnode interface is designed to be a blocking interface. For the
2337 * block device case, ZFS zvols at least will block the caller's
2338 * context in many instances, and so we need multiple threads to
2339 * overcome that problem. Other block devices don't need as many
2340 * threads, but they shouldn't cause too many problems.
2341 *
2342 * If the user wants to just have a single thread for a block
2343 * device, he can specify that when the LUN is created, or change
2344 * the tunable/sysctl to alter the default number of threads.
2345 */
2346 retval = taskqueue_start_threads(&be_lun->io_taskqueue,
2347 /*num threads*/num_threads,
2348 /*priority*/PWAIT,
2349 /*thread name*/
2350 "%s taskq", be_lun->lunname);
2351
2352 if (retval != 0)
2353 goto bailout_error;
2354
2355 be_lun->num_threads = num_threads;
2356
2357 mtx_lock(&softc->lock);
2358 softc->num_luns++;
2359 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links);
2360
2361 mtx_unlock(&softc->lock);
2362
2363 retval = ctl_add_lun(&be_lun->ctl_be_lun);
2364 if (retval != 0) {
2365 mtx_lock(&softc->lock);
2366 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2367 links);
2368 softc->num_luns--;
2369 mtx_unlock(&softc->lock);
2370 snprintf(req->error_str, sizeof(req->error_str),
2371 "ctl_add_lun() returned error %d, see dmesg for "
2372 "details", retval);
2373 retval = 0;
2374 goto bailout_error;
2375 }
2376
2377 mtx_lock(&softc->lock);
2378
2379 /*
2380 * Tell the config_status routine that we're waiting so it won't
2381 * clean up the LUN in the event of an error.
2382 */
2383 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2384
2385 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2386 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2387 if (retval == EINTR)
2388 break;
2389 }
2390 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2391
2392 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) {
2393 snprintf(req->error_str, sizeof(req->error_str),
2394 "LUN configuration error, see dmesg for details");
2395 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2396 links);
2397 softc->num_luns--;
2398 mtx_unlock(&softc->lock);
2399 goto bailout_error;
2400 } else {
2401 params->req_lun_id = be_lun->ctl_be_lun.lun_id;
2402 }
2403
2404 mtx_unlock(&softc->lock);
2405
2406 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id,
2407 be_lun->blocksize,
2408 DEVSTAT_ALL_SUPPORTED,
2409 be_lun->ctl_be_lun.lun_type
2410 | DEVSTAT_TYPE_IF_OTHER,
2411 DEVSTAT_PRIORITY_OTHER);
2412
2413 return (retval);
2414
2415bailout_error:
2416 req->status = CTL_LUN_ERROR;
2417
2418 if (be_lun->io_taskqueue != NULL)
2419 taskqueue_free(be_lun->io_taskqueue);
2420 ctl_be_block_close(be_lun);
2421 if (be_lun->dev_path != NULL)
2422 free(be_lun->dev_path, M_CTLBLK);
2423 if (be_lun->lun_zone != NULL)
2424 uma_zdestroy(be_lun->lun_zone);
2425 ctl_free_opts(&be_lun->ctl_be_lun.options);
2426 mtx_destroy(&be_lun->queue_lock);
2427 mtx_destroy(&be_lun->io_lock);
2428 free(be_lun, M_CTLBLK);
2429
2430 return (retval);
2431}
2432
2433static int
2434ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2435{
2436 struct ctl_lun_rm_params *params;
2437 struct ctl_be_block_lun *be_lun;
2438 int retval;
2439
2440 params = &req->reqdata.rm;
2441
2442 mtx_lock(&softc->lock);
2443
2444 be_lun = NULL;
2445
2446 STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2447 if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2448 break;
2449 }
2450 mtx_unlock(&softc->lock);
2451
2452 if (be_lun == NULL) {
2453 snprintf(req->error_str, sizeof(req->error_str),
2454 "LUN %u is not managed by the block backend",
2455 params->lun_id);
2456 goto bailout_error;
2457 }
2458
2459 retval = ctl_disable_lun(&be_lun->ctl_be_lun);
2460
2461 if (retval != 0) {
2462 snprintf(req->error_str, sizeof(req->error_str),
2463 "error %d returned from ctl_disable_lun() for "
2464 "LUN %d", retval, params->lun_id);
2465 goto bailout_error;
2466
2467 }
2468
2469 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun);
2470 if (retval != 0) {
2471 snprintf(req->error_str, sizeof(req->error_str),
2472 "error %d returned from ctl_invalidate_lun() for "
2473 "LUN %d", retval, params->lun_id);
2474 goto bailout_error;
2475 }
2476
2477 mtx_lock(&softc->lock);
2478
2479 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2480
2481 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2482 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2483 if (retval == EINTR)
2484 break;
2485 }
2486
2487 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2488
2489 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2490 snprintf(req->error_str, sizeof(req->error_str),
2491 "interrupted waiting for LUN to be freed");
2492 mtx_unlock(&softc->lock);
2493 goto bailout_error;
2494 }
2495
2496 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links);
2497
2498 softc->num_luns--;
2499 mtx_unlock(&softc->lock);
2500
2501 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task);
2502
2503 taskqueue_free(be_lun->io_taskqueue);
2504
2505 ctl_be_block_close(be_lun);
2506
2507 if (be_lun->disk_stats != NULL)
2508 devstat_remove_entry(be_lun->disk_stats);
2509
2510 uma_zdestroy(be_lun->lun_zone);
2511
2512 ctl_free_opts(&be_lun->ctl_be_lun.options);
2513 free(be_lun->dev_path, M_CTLBLK);
2514 mtx_destroy(&be_lun->queue_lock);
2515 mtx_destroy(&be_lun->io_lock);
2516 free(be_lun, M_CTLBLK);
2517
2518 req->status = CTL_LUN_OK;
2519
2520 return (0);
2521
2522bailout_error:
2523
2524 req->status = CTL_LUN_ERROR;
2525
2526 return (0);
2527}
2528
2529static int
2530ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
2531 struct ctl_lun_req *req)
2532{
2533 struct vattr vattr;
2534 int error;
2535 struct ctl_lun_create_params *params = &be_lun->params;
2536
2537 if (params->lun_size_bytes != 0) {
2538 be_lun->size_bytes = params->lun_size_bytes;
2539 } else {
2540 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2541 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
2542 VOP_UNLOCK(be_lun->vn, 0);
2543 if (error != 0) {
2544 snprintf(req->error_str, sizeof(req->error_str),
2545 "error calling VOP_GETATTR() for file %s",
2546 be_lun->dev_path);
2547 return (error);
2548 }
2549
2550 be_lun->size_bytes = vattr.va_size;
2551 }
2552
2553 return (0);
2554}
2555
2556static int
2557ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
2558 struct ctl_lun_req *req)
2559{
2560 struct ctl_be_block_devdata *dev_data;
2561 int error;
2562 struct ctl_lun_create_params *params = &be_lun->params;
2563 uint64_t size_bytes;
2564
2565 dev_data = &be_lun->backend.dev;
2566 if (!dev_data->csw->d_ioctl) {
2567 snprintf(req->error_str, sizeof(req->error_str),
2568 "no d_ioctl for device %s!", be_lun->dev_path);
2569 return (ENODEV);
2570 }
2571
2572 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE,
2573 (caddr_t)&size_bytes, FREAD,
2574 curthread);
2575 if (error) {
2576 snprintf(req->error_str, sizeof(req->error_str),
2577 "error %d returned for DIOCGMEDIASIZE ioctl "
2578 "on %s!", error, be_lun->dev_path);
2579 return (error);
2580 }
2581
2582 if (params->lun_size_bytes != 0) {
2583 if (params->lun_size_bytes > size_bytes) {
2584 snprintf(req->error_str, sizeof(req->error_str),
2585 "requested LUN size %ju > backing device "
2586 "size %ju",
2587 (uintmax_t)params->lun_size_bytes,
2588 (uintmax_t)size_bytes);
2589 return (EINVAL);
2590 }
2591
2592 be_lun->size_bytes = params->lun_size_bytes;
2593 } else {
2594 be_lun->size_bytes = size_bytes;
2595 }
2596
2597 return (0);
2598}
2599
2600static int
2601ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2602{
2603 struct ctl_lun_modify_params *params;
2604 struct ctl_be_block_lun *be_lun;
2605 uint64_t oldsize;
2606 int error;
2607
2608 params = &req->reqdata.modify;
2609
2610 mtx_lock(&softc->lock);
2611 be_lun = NULL;
2612 STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2613 if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2614 break;
2615 }
2616 mtx_unlock(&softc->lock);
2617
2618 if (be_lun == NULL) {
2619 snprintf(req->error_str, sizeof(req->error_str),
2620 "LUN %u is not managed by the block backend",
2621 params->lun_id);
2622 goto bailout_error;
2623 }
2624
2625 be_lun->params.lun_size_bytes = params->lun_size_bytes;
2626
2627 oldsize = be_lun->size_bytes;
2628 if (be_lun->vn == NULL)
2629 error = ctl_be_block_open(softc, be_lun, req);
2630 else if (be_lun->vn->v_type == VREG)
2631 error = ctl_be_block_modify_file(be_lun, req);
2632 else
2633 error = ctl_be_block_modify_dev(be_lun, req);
2634
2635 if (error == 0 && be_lun->size_bytes != oldsize) {
2636 be_lun->size_blocks = be_lun->size_bytes >>
2637 be_lun->blocksize_shift;
2638
2639 /*
2640 * The maximum LBA is the size - 1.
2641 *
2642 * XXX: Note that this field is being updated without locking,
2643 * which might cause problems on 32-bit architectures.
2644 */
2645 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ?
2646 0 : (be_lun->size_blocks - 1);
2647 be_lun->ctl_be_lun.blocksize = be_lun->blocksize;
2648 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp;
2649 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff;
2650 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp;
2651 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff;
2652 if (be_lun->dispatch == ctl_be_block_dispatch_zvol &&
2653 be_lun->blocksize != 0)
2654 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE /
2655 be_lun->blocksize;
2656 ctl_lun_capacity_changed(&be_lun->ctl_be_lun);
2657 if (oldsize == 0 && be_lun->size_blocks != 0)
2658 ctl_lun_online(&be_lun->ctl_be_lun);
2659 }
2660
2661 /* Tell the user the exact size we ended up using */
2662 params->lun_size_bytes = be_lun->size_bytes;
2663
2664 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2665
2666 return (0);
2667
2668bailout_error:
2669 req->status = CTL_LUN_ERROR;
2670
2671 return (0);
2672}
2673
2674static void
2675ctl_be_block_lun_shutdown(void *be_lun)
2676{
2677 struct ctl_be_block_lun *lun;
2678 struct ctl_be_block_softc *softc;
2679
2680 lun = (struct ctl_be_block_lun *)be_lun;
2681
2682 softc = lun->softc;
2683
2684 mtx_lock(&softc->lock);
2685 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2686 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2687 wakeup(lun);
2688 mtx_unlock(&softc->lock);
2689
2690}
2691
2692static void
2693ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status)
2694{
2695 struct ctl_be_block_lun *lun;
2696 struct ctl_be_block_softc *softc;
2697
2698 lun = (struct ctl_be_block_lun *)be_lun;
2699 softc = lun->softc;
2700
2701 if (status == CTL_LUN_CONFIG_OK) {
2702 mtx_lock(&softc->lock);
2703 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2704 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2705 wakeup(lun);
2706 mtx_unlock(&softc->lock);
2707
2708 /*
2709 * We successfully added the LUN, attempt to enable it.
2710 */
2711 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) {
2712 printf("%s: ctl_enable_lun() failed!\n", __func__);
2713 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) {
2714 printf("%s: ctl_invalidate_lun() failed!\n",
2715 __func__);
2716 }
2717 }
2718
2719 return;
2720 }
2721
2722
2723 mtx_lock(&softc->lock);
2724 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2725 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR;
2726 wakeup(lun);
2727 mtx_unlock(&softc->lock);
2728}
2729
2730
2731static int
2732ctl_be_block_config_write(union ctl_io *io)
2733{
2734 struct ctl_be_block_lun *be_lun;
2735 struct ctl_be_lun *ctl_be_lun;
2736 int retval;
2737
2738 retval = 0;
2739
2740 DPRINTF("entered\n");
2741
2742 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
2743 CTL_PRIV_BACKEND_LUN].ptr;
2744 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
2745
2746 switch (io->scsiio.cdb[0]) {
2747 case SYNCHRONIZE_CACHE:
2748 case SYNCHRONIZE_CACHE_16:
2749 case WRITE_SAME_10:
2750 case WRITE_SAME_16:
2751 case UNMAP:
2752 /*
2753 * The upper level CTL code will filter out any CDBs with
2754 * the immediate bit set and return the proper error.
2755 *
2756 * We don't really need to worry about what LBA range the
2757 * user asked to be synced out. When they issue a sync
2758 * cache command, we'll sync out the whole thing.
2759 */
2760 mtx_lock(&be_lun->queue_lock);
2761 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2762 links);
2763 mtx_unlock(&be_lun->queue_lock);
2764 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2765 break;
2766 case START_STOP_UNIT: {
2767 struct scsi_start_stop_unit *cdb;
2768
2769 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2770
2771 if (cdb->how & SSS_START)
2772 retval = ctl_start_lun(ctl_be_lun);
2773 else {
2774 retval = ctl_stop_lun(ctl_be_lun);
2775 /*
2776 * XXX KDM Copan-specific offline behavior.
2777 * Figure out a reasonable way to port this?
2778 */
2779#ifdef NEEDTOPORT
2780 if ((retval == 0)
2781 && (cdb->byte2 & SSS_ONOFFLINE))
2782 retval = ctl_lun_offline(ctl_be_lun);
2783#endif
2784 }
2785
2786 /*
2787 * In general, the above routines should not fail. They
2788 * just set state for the LUN. So we've got something
2789 * pretty wrong here if we can't start or stop the LUN.
2790 */
2791 if (retval != 0) {
2792 ctl_set_internal_failure(&io->scsiio,
2793 /*sks_valid*/ 1,
2794 /*retry_count*/ 0xf051);
2795 retval = CTL_RETVAL_COMPLETE;
2796 } else {
2797 ctl_set_success(&io->scsiio);
2798 }
2799 ctl_config_write_done(io);
2800 break;
2801 }
2802 default:
2803 ctl_set_invalid_opcode(&io->scsiio);
2804 ctl_config_write_done(io);
2805 retval = CTL_RETVAL_COMPLETE;
2806 break;
2807 }
2808
2809 return (retval);
2810}
2811
2812static int
2813ctl_be_block_config_read(union ctl_io *io)
2814{
2815 struct ctl_be_block_lun *be_lun;
2816 struct ctl_be_lun *ctl_be_lun;
2817 int retval = 0;
2818
2819 DPRINTF("entered\n");
2820
2821 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
2822 CTL_PRIV_BACKEND_LUN].ptr;
2823 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
2824
2825 switch (io->scsiio.cdb[0]) {
2826 case SERVICE_ACTION_IN:
2827 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
2828 mtx_lock(&be_lun->queue_lock);
2829 STAILQ_INSERT_TAIL(&be_lun->config_read_queue,
2830 &io->io_hdr, links);
2831 mtx_unlock(&be_lun->queue_lock);
2832 taskqueue_enqueue(be_lun->io_taskqueue,
2833 &be_lun->io_task);
2834 retval = CTL_RETVAL_QUEUED;
2835 break;
2836 }
2837 ctl_set_invalid_field(&io->scsiio,
2838 /*sks_valid*/ 1,
2839 /*command*/ 1,
2840 /*field*/ 1,
2841 /*bit_valid*/ 1,
2842 /*bit*/ 4);
2843 ctl_config_read_done(io);
2844 retval = CTL_RETVAL_COMPLETE;
2845 break;
2846 default:
2847 ctl_set_invalid_opcode(&io->scsiio);
2848 ctl_config_read_done(io);
2849 retval = CTL_RETVAL_COMPLETE;
2850 break;
2851 }
2852
2853 return (retval);
2854}
2855
2856static int
2857ctl_be_block_lun_info(void *be_lun, struct sbuf *sb)
2858{
2859 struct ctl_be_block_lun *lun;
2860 int retval;
2861
2862 lun = (struct ctl_be_block_lun *)be_lun;
2863 retval = 0;
2864
2865 retval = sbuf_printf(sb, "\t<num_threads>");
2866
2867 if (retval != 0)
2868 goto bailout;
2869
2870 retval = sbuf_printf(sb, "%d", lun->num_threads);
2871
2872 if (retval != 0)
2873 goto bailout;
2874
2875 retval = sbuf_printf(sb, "</num_threads>\n");
2876
2877bailout:
2878
2879 return (retval);
2880}
2881
2882static uint64_t
2883ctl_be_block_lun_attr(void *be_lun, const char *attrname)
2884{
2885 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun;
2886
2887 if (lun->getattr == NULL)
2888 return (UINT64_MAX);
2889 return (lun->getattr(lun, attrname));
2890}
2891
2892int
2893ctl_be_block_init(void)
2894{
2895 struct ctl_be_block_softc *softc;
2896 int retval;
2897
2898 softc = &backend_block_softc;
2899 retval = 0;
2900
2901 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
2902 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2903 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2904 STAILQ_INIT(&softc->disk_list);
2905 STAILQ_INIT(&softc->lun_list);
2906
2907 return (retval);
2908}