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_ioctl.h>
88#include <cam/ctl/ctl_ha.h>
89#include <cam/ctl/ctl_scsi_all.h>
90#include <cam/ctl/ctl_private.h>
91#include <cam/ctl/ctl_error.h>
92
93/*
94 * The idea here is that we'll allocate enough S/G space to hold a 1MB
95 * I/O. If we get an I/O larger than that, we'll split it.
96 */
97#define CTLBLK_HALF_IO_SIZE (512 * 1024)
98#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2)
99#define CTLBLK_MAX_SEG MAXPHYS
100#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1)
101#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2)
102
103#ifdef CTLBLK_DEBUG
104#define DPRINTF(fmt, args...) \
105 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
106#else
107#define DPRINTF(fmt, args...) do {} while(0)
108#endif
109
110#define PRIV(io) \
111 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
112#define ARGS(io) \
113 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
114
115SDT_PROVIDER_DEFINE(cbb);
116
117typedef enum {
118 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01,
119 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02,
120 CTL_BE_BLOCK_LUN_WAITING = 0x04,
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_filedata {
130 struct ucred *cred;
131};
132
133union ctl_be_block_bedata {
134 struct ctl_be_block_filedata file;
135};
136
137struct ctl_be_block_io;
138struct ctl_be_block_lun;
139
140typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
141 struct ctl_be_block_io *beio);
142typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
143 const char *attrname);
144
145/*
146 * Backend LUN structure. There is a 1:1 mapping between a block device
147 * and a backend block LUN, and between a backend block LUN and a CTL LUN.
148 */
149struct ctl_be_block_lun {
150 struct ctl_lun_create_params params;
151 char lunname[32];
152 char *dev_path;
153 ctl_be_block_type dev_type;
154 struct vnode *vn;
155 union ctl_be_block_bedata backend;
156 cbb_dispatch_t dispatch;
157 cbb_dispatch_t lun_flush;
158 cbb_dispatch_t unmap;
159 cbb_dispatch_t get_lba_status;
160 cbb_getattr_t getattr;
161 uma_zone_t lun_zone;
162 uint64_t size_blocks;
163 uint64_t size_bytes;
164 struct ctl_be_block_softc *softc;
165 struct devstat *disk_stats;
166 ctl_be_block_lun_flags flags;
167 STAILQ_ENTRY(ctl_be_block_lun) links;
168 struct ctl_be_lun cbe_lun;
169 struct taskqueue *io_taskqueue;
170 struct task io_task;
171 int num_threads;
172 STAILQ_HEAD(, ctl_io_hdr) input_queue;
173 STAILQ_HEAD(, ctl_io_hdr) config_read_queue;
174 STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
175 STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
176 struct mtx_padalign io_lock;
177 struct mtx_padalign queue_lock;
178};
179
180/*
181 * Overall softc structure for the block backend module.
182 */
183struct ctl_be_block_softc {
184 struct mtx lock;
185 int num_luns;
186 STAILQ_HEAD(, ctl_be_block_lun) lun_list;
187};
188
189static struct ctl_be_block_softc backend_block_softc;
190
191/*
192 * Per-I/O information.
193 */
194struct ctl_be_block_io {
195 union ctl_io *io;
196 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS];
197 struct iovec xiovecs[CTLBLK_MAX_SEGS];
198 int bio_cmd;
199 int num_segs;
200 int num_bios_sent;
201 int num_bios_done;
202 int send_complete;
203 int num_errors;
204 struct bintime ds_t0;
205 devstat_tag_type ds_tag_type;
206 devstat_trans_flags ds_trans_type;
207 uint64_t io_len;
208 uint64_t io_offset;
209 int io_arg;
210 struct ctl_be_block_softc *softc;
211 struct ctl_be_block_lun *lun;
212 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
213};
214
215extern struct ctl_softc *control_softc;
216
217static int cbb_num_threads = 14;
218SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0,
219 "CAM Target Layer Block Backend");
220SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
221 &cbb_num_threads, 0, "Number of threads per backing file");
222
223static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
224static void ctl_free_beio(struct ctl_be_block_io *beio);
225static void ctl_complete_beio(struct ctl_be_block_io *beio);
226static int ctl_be_block_move_done(union ctl_io *io);
227static void ctl_be_block_biodone(struct bio *bio);
228static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
229 struct ctl_be_block_io *beio);
230static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
231 struct ctl_be_block_io *beio);
232static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
233 struct ctl_be_block_io *beio);
234static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
235 const char *attrname);
236static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
237 struct ctl_be_block_io *beio);
238static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
239 struct ctl_be_block_io *beio);
240static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
241 struct ctl_be_block_io *beio);
242static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
243 const char *attrname);
244static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
245 union ctl_io *io);
246static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
247 union ctl_io *io);
248static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
249 union ctl_io *io);
250static void ctl_be_block_worker(void *context, int pending);
251static int ctl_be_block_submit(union ctl_io *io);
252static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
253 int flag, struct thread *td);
254static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
255 struct ctl_lun_req *req);
256static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
257 struct ctl_lun_req *req);
258static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
259static int ctl_be_block_open(struct ctl_be_block_softc *softc,
260 struct ctl_be_block_lun *be_lun,
261 struct ctl_lun_req *req);
262static int ctl_be_block_create(struct ctl_be_block_softc *softc,
263 struct ctl_lun_req *req);
264static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
265 struct ctl_lun_req *req);
266static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
267 struct ctl_lun_req *req);
268static void ctl_be_block_lun_shutdown(void *be_lun);
269static void ctl_be_block_lun_config_status(void *be_lun,
270 ctl_lun_config_status status);
271static int ctl_be_block_config_write(union ctl_io *io);
272static int ctl_be_block_config_read(union ctl_io *io);
273static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb);
274static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname);
275int ctl_be_block_init(void);
276
277static struct ctl_backend_driver ctl_be_block_driver =
278{
279 .name = "block",
280 .flags = CTL_BE_FLAG_HAS_CONFIG,
281 .init = ctl_be_block_init,
282 .data_submit = ctl_be_block_submit,
283 .data_move_done = ctl_be_block_move_done,
284 .config_read = ctl_be_block_config_read,
285 .config_write = ctl_be_block_config_write,
286 .ioctl = ctl_be_block_ioctl,
287 .lun_info = ctl_be_block_lun_info,
288 .lun_attr = ctl_be_block_lun_attr
289};
290
291MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend");
292CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
293
294static uma_zone_t beio_zone;
295
296static struct ctl_be_block_io *
297ctl_alloc_beio(struct ctl_be_block_softc *softc)
298{
299 struct ctl_be_block_io *beio;
300
301 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO);
302 beio->softc = softc;
303 return (beio);
304}
305
306static void
307ctl_free_beio(struct ctl_be_block_io *beio)
308{
309 int duplicate_free;
310 int i;
311
312 duplicate_free = 0;
313
314 for (i = 0; i < beio->num_segs; i++) {
315 if (beio->sg_segs[i].addr == NULL)
316 duplicate_free++;
317
318 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr);
319 beio->sg_segs[i].addr = NULL;
320
321 /* For compare we had two equal S/G lists. */
322 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) {
323 uma_zfree(beio->lun->lun_zone,
324 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr);
325 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL;
326 }
327 }
328
329 if (duplicate_free > 0) {
330 printf("%s: %d duplicate frees out of %d segments\n", __func__,
331 duplicate_free, beio->num_segs);
332 }
333
334 uma_zfree(beio_zone, beio);
335}
336
337static void
338ctl_complete_beio(struct ctl_be_block_io *beio)
339{
340 union ctl_io *io = beio->io;
341
342 if (beio->beio_cont != NULL) {
343 beio->beio_cont(beio);
344 } else {
345 ctl_free_beio(beio);
346 ctl_data_submit_done(io);
347 }
348}
349
350static size_t
351cmp(uint8_t *a, uint8_t *b, size_t size)
352{
353 size_t i;
354
355 for (i = 0; i < size; i++) {
356 if (a[i] != b[i])
357 break;
358 }
359 return (i);
360}
361
362static void
363ctl_be_block_compare(union ctl_io *io)
364{
365 struct ctl_be_block_io *beio;
366 uint64_t off, res;
367 int i;
368 uint8_t info[8];
369
370 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
371 off = 0;
372 for (i = 0; i < beio->num_segs; i++) {
373 res = cmp(beio->sg_segs[i].addr,
374 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
375 beio->sg_segs[i].len);
376 off += res;
377 if (res < beio->sg_segs[i].len)
378 break;
379 }
380 if (i < beio->num_segs) {
381 scsi_u64to8b(off, info);
382 ctl_set_sense(&io->scsiio, /*current_error*/ 1,
383 /*sense_key*/ SSD_KEY_MISCOMPARE,
384 /*asc*/ 0x1D, /*ascq*/ 0x00,
385 /*type*/ SSD_ELEM_INFO,
386 /*size*/ sizeof(info), /*data*/ &info,
387 /*type*/ SSD_ELEM_NONE);
388 } else
389 ctl_set_success(&io->scsiio);
390}
391
392static int
393ctl_be_block_move_done(union ctl_io *io)
394{
395 struct ctl_be_block_io *beio;
396 struct ctl_be_block_lun *be_lun;
397 struct ctl_lba_len_flags *lbalen;
398#ifdef CTL_TIME_IO
399 struct bintime cur_bt;
400#endif
401
402 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
403 be_lun = beio->lun;
404
405 DPRINTF("entered\n");
406
407#ifdef CTL_TIME_IO
408 getbinuptime(&cur_bt);
409 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
410 bintime_add(&io->io_hdr.dma_bt, &cur_bt);
411#endif
412 io->io_hdr.num_dmas++;
413 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
414
415 /*
416 * We set status at this point for read commands, and write
417 * commands with errors.
418 */
419 if (io->io_hdr.flags & CTL_FLAG_ABORT) {
420 ;
421 } else if ((io->io_hdr.port_status == 0) &&
422 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
423 lbalen = ARGS(beio->io);
424 if (lbalen->flags & CTL_LLF_READ) {
425 ctl_set_success(&io->scsiio);
426 } else if (lbalen->flags & CTL_LLF_COMPARE) {
427 /* We have two data blocks ready for comparison. */
428 ctl_be_block_compare(io);
429 }
430 } else if ((io->io_hdr.port_status != 0) &&
431 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
432 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
433 /*
434 * For hardware error sense keys, the sense key
435 * specific value is defined to be a retry count,
436 * but we use it to pass back an internal FETD
437 * error code. XXX KDM Hopefully the FETD is only
438 * using 16 bits for an error code, since that's
439 * all the space we have in the sks field.
440 */
441 ctl_set_internal_failure(&io->scsiio,
442 /*sks_valid*/ 1,
443 /*retry_count*/
444 io->io_hdr.port_status);
445 }
446
447 /*
448 * If this is a read, or a write with errors, it is done.
449 */
450 if ((beio->bio_cmd == BIO_READ)
451 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
452 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
453 ctl_complete_beio(beio);
454 return (0);
455 }
456
457 /*
458 * At this point, we have a write and the DMA completed
459 * successfully. We now have to queue it to the task queue to
460 * execute the backend I/O. That is because we do blocking
461 * memory allocations, and in the file backing case, blocking I/O.
462 * This move done routine is generally called in the SIM's
463 * interrupt context, and therefore we cannot block.
464 */
465 mtx_lock(&be_lun->queue_lock);
466 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
467 mtx_unlock(&be_lun->queue_lock);
468 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
469
470 return (0);
471}
472
473static void
474ctl_be_block_biodone(struct bio *bio)
475{
476 struct ctl_be_block_io *beio;
477 struct ctl_be_block_lun *be_lun;
478 union ctl_io *io;
479 int error;
480
481 beio = bio->bio_caller1;
482 be_lun = beio->lun;
483 io = beio->io;
484
485 DPRINTF("entered\n");
486
487 error = bio->bio_error;
488 mtx_lock(&be_lun->io_lock);
489 if (error != 0)
490 beio->num_errors++;
491
492 beio->num_bios_done++;
493
494 /*
495 * XXX KDM will this cause WITNESS to complain? Holding a lock
496 * during the free might cause it to complain.
497 */
498 g_destroy_bio(bio);
499
500 /*
501 * If the send complete bit isn't set, or we aren't the last I/O to
502 * complete, then we're done.
503 */
504 if ((beio->send_complete == 0)
505 || (beio->num_bios_done < beio->num_bios_sent)) {
506 mtx_unlock(&be_lun->io_lock);
507 return;
508 }
509
510 /*
511 * At this point, we've verified that we are the last I/O to
512 * complete, so it's safe to drop the lock.
513 */
514 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
515 beio->ds_tag_type, beio->ds_trans_type,
516 /*now*/ NULL, /*then*/&beio->ds_t0);
517 mtx_unlock(&be_lun->io_lock);
518
519 /*
520 * If there are any errors from the backing device, we fail the
521 * entire I/O with a medium error.
522 */
523 if (beio->num_errors > 0) {
524 if (error == EOPNOTSUPP) {
525 ctl_set_invalid_opcode(&io->scsiio);
526 } else if (error == ENOSPC || error == EDQUOT) {
527 ctl_set_space_alloc_fail(&io->scsiio);
528 } else if (error == EROFS || error == EACCES) {
529 ctl_set_hw_write_protected(&io->scsiio);
530 } else if (beio->bio_cmd == BIO_FLUSH) {
531 /* XXX KDM is there is a better error here? */
532 ctl_set_internal_failure(&io->scsiio,
533 /*sks_valid*/ 1,
534 /*retry_count*/ 0xbad2);
535 } else {
536 ctl_set_medium_error(&io->scsiio,
537 beio->bio_cmd == BIO_READ);
538 }
539 ctl_complete_beio(beio);
540 return;
541 }
542
543 /*
544 * If this is a write, a flush, a delete or verify, we're all done.
545 * If this is a read, we can now send the data to the user.
546 */
547 if ((beio->bio_cmd == BIO_WRITE)
548 || (beio->bio_cmd == BIO_FLUSH)
549 || (beio->bio_cmd == BIO_DELETE)
550 || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
551 ctl_set_success(&io->scsiio);
552 ctl_complete_beio(beio);
553 } else {
554 if ((ARGS(io)->flags & CTL_LLF_READ) &&
555 beio->beio_cont == NULL) {
556 ctl_set_success(&io->scsiio);
557 ctl_serseq_done(io);
558 }
559#ifdef CTL_TIME_IO
560 getbinuptime(&io->io_hdr.dma_start_bt);
561#endif
562 ctl_datamove(io);
563 }
564}
565
566static void
567ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
568 struct ctl_be_block_io *beio)
569{
570 union ctl_io *io = beio->io;
571 struct mount *mountpoint;
572 int error, lock_flags;
573
574 DPRINTF("entered\n");
575
576 binuptime(&beio->ds_t0);
577 mtx_lock(&be_lun->io_lock);
578 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
579 mtx_unlock(&be_lun->io_lock);
580
581 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
582
583 if (MNT_SHARED_WRITES(mountpoint) ||
584 ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
585 lock_flags = LK_SHARED;
586 else
587 lock_flags = LK_EXCLUSIVE;
588 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
589 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT,
590 curthread);
591 VOP_UNLOCK(be_lun->vn, 0);
592
593 vn_finished_write(mountpoint);
594
595 mtx_lock(&be_lun->io_lock);
596 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
597 beio->ds_tag_type, beio->ds_trans_type,
598 /*now*/ NULL, /*then*/&beio->ds_t0);
599 mtx_unlock(&be_lun->io_lock);
600
601 if (error == 0)
602 ctl_set_success(&io->scsiio);
603 else {
604 /* XXX KDM is there is a better error here? */
605 ctl_set_internal_failure(&io->scsiio,
606 /*sks_valid*/ 1,
607 /*retry_count*/ 0xbad1);
608 }
609
610 ctl_complete_beio(beio);
611}
612
613SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t");
614SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t");
615SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t");
616SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t");
617
618static void
619ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
620 struct ctl_be_block_io *beio)
621{
622 struct ctl_be_block_filedata *file_data;
623 union ctl_io *io;
624 struct uio xuio;
625 struct iovec *xiovec;
626 size_t s;
627 int error, flags, i;
628
629 DPRINTF("entered\n");
630
631 file_data = &be_lun->backend.file;
632 io = beio->io;
633 flags = 0;
634 if (ARGS(io)->flags & CTL_LLF_DPO)
635 flags |= IO_DIRECT;
636 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
637 flags |= IO_SYNC;
638
639 bzero(&xuio, sizeof(xuio));
640 if (beio->bio_cmd == BIO_READ) {
641 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
642 xuio.uio_rw = UIO_READ;
643 } else {
644 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
645 xuio.uio_rw = UIO_WRITE;
646 }
647 xuio.uio_offset = beio->io_offset;
648 xuio.uio_resid = beio->io_len;
649 xuio.uio_segflg = UIO_SYSSPACE;
650 xuio.uio_iov = beio->xiovecs;
651 xuio.uio_iovcnt = beio->num_segs;
652 xuio.uio_td = curthread;
653
654 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
655 xiovec->iov_base = beio->sg_segs[i].addr;
656 xiovec->iov_len = beio->sg_segs[i].len;
657 }
658
659 binuptime(&beio->ds_t0);
660 mtx_lock(&be_lun->io_lock);
661 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
662 mtx_unlock(&be_lun->io_lock);
663
664 if (beio->bio_cmd == BIO_READ) {
665 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
666
667 /*
668 * UFS pays attention to IO_DIRECT for reads. If the
669 * DIRECTIO option is configured into the kernel, it calls
670 * ffs_rawread(). But that only works for single-segment
671 * uios with user space addresses. In our case, with a
672 * kernel uio, it still reads into the buffer cache, but it
673 * will just try to release the buffer from the cache later
674 * on in ffs_read().
675 *
676 * ZFS does not pay attention to IO_DIRECT for reads.
677 *
678 * UFS does not pay attention to IO_SYNC for reads.
679 *
680 * ZFS pays attention to IO_SYNC (which translates into the
681 * Solaris define FRSYNC for zfs_read()) for reads. It
682 * attempts to sync the file before reading.
683 */
684 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
685
686 VOP_UNLOCK(be_lun->vn, 0);
687 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
688 if (error == 0 && xuio.uio_resid > 0) {
689 /*
690 * If we red less then requested (EOF), then
691 * we should clean the rest of the buffer.
692 */
693 s = beio->io_len - xuio.uio_resid;
694 for (i = 0; i < beio->num_segs; i++) {
695 if (s >= beio->sg_segs[i].len) {
696 s -= beio->sg_segs[i].len;
697 continue;
698 }
699 bzero((uint8_t *)beio->sg_segs[i].addr + s,
700 beio->sg_segs[i].len - s);
701 s = 0;
702 }
703 }
704 } else {
705 struct mount *mountpoint;
706 int lock_flags;
707
708 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
709
710 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL)
711 && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
712 lock_flags = LK_SHARED;
713 else
714 lock_flags = LK_EXCLUSIVE;
715 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
716
717 /*
718 * UFS pays attention to IO_DIRECT for writes. The write
719 * is done asynchronously. (Normally the write would just
720 * get put into cache.
721 *
722 * UFS pays attention to IO_SYNC for writes. It will
723 * attempt to write the buffer out synchronously if that
724 * flag is set.
725 *
726 * ZFS does not pay attention to IO_DIRECT for writes.
727 *
728 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
729 * for writes. It will flush the transaction from the
730 * cache before returning.
731 */
732 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
733 VOP_UNLOCK(be_lun->vn, 0);
734
735 vn_finished_write(mountpoint);
736 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
737 }
738
739 mtx_lock(&be_lun->io_lock);
740 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
741 beio->ds_tag_type, beio->ds_trans_type,
742 /*now*/ NULL, /*then*/&beio->ds_t0);
743 mtx_unlock(&be_lun->io_lock);
744
745 /*
746 * If we got an error, set the sense data to "MEDIUM ERROR" and
747 * return the I/O to the user.
748 */
749 if (error != 0) {
750 if (error == ENOSPC || error == EDQUOT) {
751 ctl_set_space_alloc_fail(&io->scsiio);
752 } else if (error == EROFS || error == EACCES) {
753 ctl_set_hw_write_protected(&io->scsiio);
754 } else {
755 ctl_set_medium_error(&io->scsiio,
756 beio->bio_cmd == BIO_READ);
757 }
758 ctl_complete_beio(beio);
759 return;
760 }
761
762 /*
763 * If this is a write or a verify, we're all done.
764 * If this is a read, we can now send the data to the user.
765 */
766 if ((beio->bio_cmd == BIO_WRITE) ||
767 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
768 ctl_set_success(&io->scsiio);
769 ctl_complete_beio(beio);
770 } else {
771 if ((ARGS(io)->flags & CTL_LLF_READ) &&
772 beio->beio_cont == NULL) {
773 ctl_set_success(&io->scsiio);
774 ctl_serseq_done(io);
775 }
776#ifdef CTL_TIME_IO
777 getbinuptime(&io->io_hdr.dma_start_bt);
778#endif
779 ctl_datamove(io);
780 }
781}
782
783static void
784ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
785 struct ctl_be_block_io *beio)
786{
787 union ctl_io *io = beio->io;
788 struct ctl_lba_len_flags *lbalen = ARGS(io);
789 struct scsi_get_lba_status_data *data;
790 off_t roff, off;
791 int error, status;
792
793 DPRINTF("entered\n");
794
795 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
796 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
797 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
798 0, curthread->td_ucred, curthread);
799 if (error == 0 && off > roff)
800 status = 0; /* mapped up to off */
801 else {
802 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
803 0, curthread->td_ucred, curthread);
804 if (error == 0 && off > roff)
805 status = 1; /* deallocated up to off */
806 else {
807 status = 0; /* unknown up to the end */
808 off = be_lun->size_bytes;
809 }
810 }
811 VOP_UNLOCK(be_lun->vn, 0);
812
813 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
814 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
815 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
816 lbalen->lba), data->descr[0].length);
817 data->descr[0].status = status;
818
819 ctl_complete_beio(beio);
820}
821
822static uint64_t
823ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
824{
825 struct vattr vattr;
826 struct statfs statfs;
827 uint64_t val;
828 int error;
829
830 val = UINT64_MAX;
831 if (be_lun->vn == NULL)
832 return (val);
833 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
834 if (strcmp(attrname, "blocksused") == 0) {
835 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
836 if (error == 0)
837 val = vattr.va_bytes / be_lun->cbe_lun.blocksize;
838 }
839 if (strcmp(attrname, "blocksavail") == 0 &&
840 (be_lun->vn->v_iflag & VI_DOOMED) == 0) {
841 error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
842 if (error == 0)
843 val = statfs.f_bavail * statfs.f_bsize /
844 be_lun->cbe_lun.blocksize;
845 }
846 VOP_UNLOCK(be_lun->vn, 0);
847 return (val);
848}
849
850static void
851ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
852 struct ctl_be_block_io *beio)
853{
854 union ctl_io *io;
855 struct cdevsw *csw;
856 struct cdev *dev;
857 struct uio xuio;
858 struct iovec *xiovec;
859 int error, flags, i, ref;
860
861 DPRINTF("entered\n");
862
863 io = beio->io;
864 flags = 0;
865 if (ARGS(io)->flags & CTL_LLF_DPO)
866 flags |= IO_DIRECT;
867 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
868 flags |= IO_SYNC;
869
870 bzero(&xuio, sizeof(xuio));
871 if (beio->bio_cmd == BIO_READ) {
872 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
873 xuio.uio_rw = UIO_READ;
874 } else {
875 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
876 xuio.uio_rw = UIO_WRITE;
877 }
878 xuio.uio_offset = beio->io_offset;
879 xuio.uio_resid = beio->io_len;
880 xuio.uio_segflg = UIO_SYSSPACE;
881 xuio.uio_iov = beio->xiovecs;
882 xuio.uio_iovcnt = beio->num_segs;
883 xuio.uio_td = curthread;
884
885 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
886 xiovec->iov_base = beio->sg_segs[i].addr;
887 xiovec->iov_len = beio->sg_segs[i].len;
888 }
889
890 binuptime(&beio->ds_t0);
891 mtx_lock(&be_lun->io_lock);
892 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
893 mtx_unlock(&be_lun->io_lock);
894
895 csw = devvn_refthread(be_lun->vn, &dev, &ref);
896 if (csw) {
897 if (beio->bio_cmd == BIO_READ)
898 error = csw->d_read(dev, &xuio, flags);
899 else
900 error = csw->d_write(dev, &xuio, flags);
901 dev_relthread(dev, ref);
902 } else
903 error = ENXIO;
904
905 if (beio->bio_cmd == BIO_READ)
906 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
907 else
908 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
909
910 mtx_lock(&be_lun->io_lock);
911 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
912 beio->ds_tag_type, beio->ds_trans_type,
913 /*now*/ NULL, /*then*/&beio->ds_t0);
914 mtx_unlock(&be_lun->io_lock);
915
916 /*
917 * If we got an error, set the sense data to "MEDIUM ERROR" and
918 * return the I/O to the user.
919 */
920 if (error != 0) {
921 if (error == ENOSPC || error == EDQUOT) {
922 ctl_set_space_alloc_fail(&io->scsiio);
923 } else if (error == EROFS || error == EACCES) {
924 ctl_set_hw_write_protected(&io->scsiio);
925 } else {
926 ctl_set_medium_error(&io->scsiio,
927 beio->bio_cmd == BIO_READ);
928 }
929 ctl_complete_beio(beio);
930 return;
931 }
932
933 /*
934 * If this is a write or a verify, we're all done.
935 * If this is a read, we can now send the data to the user.
936 */
937 if ((beio->bio_cmd == BIO_WRITE) ||
938 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
939 ctl_set_success(&io->scsiio);
940 ctl_complete_beio(beio);
941 } else {
942 if ((ARGS(io)->flags & CTL_LLF_READ) &&
943 beio->beio_cont == NULL) {
944 ctl_set_success(&io->scsiio);
945 ctl_serseq_done(io);
946 }
947#ifdef CTL_TIME_IO
948 getbinuptime(&io->io_hdr.dma_start_bt);
949#endif
950 ctl_datamove(io);
951 }
952}
953
954static void
955ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
956 struct ctl_be_block_io *beio)
957{
958 union ctl_io *io = beio->io;
959 struct cdevsw *csw;
960 struct cdev *dev;
961 struct ctl_lba_len_flags *lbalen = ARGS(io);
962 struct scsi_get_lba_status_data *data;
963 off_t roff, off;
964 int error, ref, status;
965
966 DPRINTF("entered\n");
967
968 csw = devvn_refthread(be_lun->vn, &dev, &ref);
969 if (csw == NULL) {
970 status = 0; /* unknown up to the end */
971 off = be_lun->size_bytes;
972 goto done;
973 }
974 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
975 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD,
976 curthread);
977 if (error == 0 && off > roff)
978 status = 0; /* mapped up to off */
979 else {
980 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD,
981 curthread);
982 if (error == 0 && off > roff)
983 status = 1; /* deallocated up to off */
984 else {
985 status = 0; /* unknown up to the end */
986 off = be_lun->size_bytes;
987 }
988 }
989 dev_relthread(dev, ref);
990
991done:
992 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
993 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
994 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
995 lbalen->lba), data->descr[0].length);
996 data->descr[0].status = status;
997
998 ctl_complete_beio(beio);
999}
1000
1001static void
1002ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
1003 struct ctl_be_block_io *beio)
1004{
1005 struct bio *bio;
1006 struct cdevsw *csw;
1007 struct cdev *dev;
1008 int ref;
1009
1010 DPRINTF("entered\n");
1011
1012 /* This can't fail, it's a blocking allocation. */
1013 bio = g_alloc_bio();
1014
1015 bio->bio_cmd = BIO_FLUSH;
1016 bio->bio_offset = 0;
1017 bio->bio_data = 0;
1018 bio->bio_done = ctl_be_block_biodone;
1019 bio->bio_caller1 = beio;
1020 bio->bio_pblkno = 0;
1021
1022 /*
1023 * We don't need to acquire the LUN lock here, because we are only
1024 * sending one bio, and so there is no other context to synchronize
1025 * with.
1026 */
1027 beio->num_bios_sent = 1;
1028 beio->send_complete = 1;
1029
1030 binuptime(&beio->ds_t0);
1031 mtx_lock(&be_lun->io_lock);
1032 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1033 mtx_unlock(&be_lun->io_lock);
1034
1035 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1036 if (csw) {
1037 bio->bio_dev = dev;
1038 csw->d_strategy(bio);
1039 dev_relthread(dev, ref);
1040 } else {
1041 bio->bio_error = ENXIO;
1042 ctl_be_block_biodone(bio);
1043 }
1044}
1045
1046static void
1047ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1048 struct ctl_be_block_io *beio,
1049 uint64_t off, uint64_t len, int last)
1050{
1051 struct bio *bio;
1052 uint64_t maxlen;
1053 struct cdevsw *csw;
1054 struct cdev *dev;
1055 int ref;
1056
1057 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1058 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize);
1059 while (len > 0) {
1060 bio = g_alloc_bio();
1061 bio->bio_cmd = BIO_DELETE;
1062 bio->bio_dev = dev;
1063 bio->bio_offset = off;
1064 bio->bio_length = MIN(len, maxlen);
1065 bio->bio_data = 0;
1066 bio->bio_done = ctl_be_block_biodone;
1067 bio->bio_caller1 = beio;
1068 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize;
1069
1070 off += bio->bio_length;
1071 len -= bio->bio_length;
1072
1073 mtx_lock(&be_lun->io_lock);
1074 beio->num_bios_sent++;
1075 if (last && len == 0)
1076 beio->send_complete = 1;
1077 mtx_unlock(&be_lun->io_lock);
1078
1079 if (csw) {
1080 csw->d_strategy(bio);
1081 } else {
1082 bio->bio_error = ENXIO;
1083 ctl_be_block_biodone(bio);
1084 }
1085 }
1086 if (csw)
1087 dev_relthread(dev, ref);
1088}
1089
1090static void
1091ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1092 struct ctl_be_block_io *beio)
1093{
1094 union ctl_io *io;
1095 struct ctl_ptr_len_flags *ptrlen;
1096 struct scsi_unmap_desc *buf, *end;
1097 uint64_t len;
1098
1099 io = beio->io;
1100
1101 DPRINTF("entered\n");
1102
1103 binuptime(&beio->ds_t0);
1104 mtx_lock(&be_lun->io_lock);
1105 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1106 mtx_unlock(&be_lun->io_lock);
1107
1108 if (beio->io_offset == -1) {
1109 beio->io_len = 0;
1110 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1111 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1112 end = buf + ptrlen->len / sizeof(*buf);
1113 for (; buf < end; buf++) {
1114 len = (uint64_t)scsi_4btoul(buf->length) *
1115 be_lun->cbe_lun.blocksize;
1116 beio->io_len += len;
1117 ctl_be_block_unmap_dev_range(be_lun, beio,
1118 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize,
1119 len, (end - buf < 2) ? TRUE : FALSE);
1120 }
1121 } else
1122 ctl_be_block_unmap_dev_range(be_lun, beio,
1123 beio->io_offset, beio->io_len, TRUE);
1124}
1125
1126static void
1127ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1128 struct ctl_be_block_io *beio)
1129{
1130 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1131 struct bio *bio;
1132 struct cdevsw *csw;
1133 struct cdev *dev;
1134 off_t cur_offset;
1135 int i, max_iosize, ref;
1136
1137 DPRINTF("entered\n");
1138 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1139
1140 /*
1141 * We have to limit our I/O size to the maximum supported by the
1142 * backend device. Hopefully it is MAXPHYS. If the driver doesn't
1143 * set it properly, use DFLTPHYS.
1144 */
1145 if (csw) {
1146 max_iosize = dev->si_iosize_max;
1147 if (max_iosize < PAGE_SIZE)
1148 max_iosize = DFLTPHYS;
1149 } else
1150 max_iosize = DFLTPHYS;
1151
1152 cur_offset = beio->io_offset;
1153 for (i = 0; i < beio->num_segs; i++) {
1154 size_t cur_size;
1155 uint8_t *cur_ptr;
1156
1157 cur_size = beio->sg_segs[i].len;
1158 cur_ptr = beio->sg_segs[i].addr;
1159
1160 while (cur_size > 0) {
1161 /* This can't fail, it's a blocking allocation. */
1162 bio = g_alloc_bio();
1163
1164 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1165
1166 bio->bio_cmd = beio->bio_cmd;
1167 bio->bio_dev = dev;
1168 bio->bio_caller1 = beio;
1169 bio->bio_length = min(cur_size, max_iosize);
1170 bio->bio_offset = cur_offset;
1171 bio->bio_data = cur_ptr;
1172 bio->bio_done = ctl_be_block_biodone;
1173 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize;
1174
1175 cur_offset += bio->bio_length;
1176 cur_ptr += bio->bio_length;
1177 cur_size -= bio->bio_length;
1178
1179 TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1180 beio->num_bios_sent++;
1181 }
1182 }
1183 binuptime(&beio->ds_t0);
1184 mtx_lock(&be_lun->io_lock);
1185 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1186 beio->send_complete = 1;
1187 mtx_unlock(&be_lun->io_lock);
1188
1189 /*
1190 * Fire off all allocated requests!
1191 */
1192 while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1193 TAILQ_REMOVE(&queue, bio, bio_queue);
1194 if (csw)
1195 csw->d_strategy(bio);
1196 else {
1197 bio->bio_error = ENXIO;
1198 ctl_be_block_biodone(bio);
1199 }
1200 }
1201 if (csw)
1202 dev_relthread(dev, ref);
1203}
1204
1205static uint64_t
1206ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1207{
1208 struct diocgattr_arg arg;
1209 struct cdevsw *csw;
1210 struct cdev *dev;
1211 int error, ref;
1212
1213 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1214 if (csw == NULL)
1215 return (UINT64_MAX);
1216 strlcpy(arg.name, attrname, sizeof(arg.name));
1217 arg.len = sizeof(arg.value.off);
1218 if (csw->d_ioctl) {
1219 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
1220 curthread);
1221 } else
1222 error = ENODEV;
1223 dev_relthread(dev, ref);
1224 if (error != 0)
1225 return (UINT64_MAX);
1226 return (arg.value.off);
1227}
1228
1229static void
1230ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun,
1231 union ctl_io *io)
1232{
1233 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1234 struct ctl_be_block_io *beio;
1235 struct ctl_lba_len_flags *lbalen;
1236
1237 DPRINTF("entered\n");
1238 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1239 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1240
1241 beio->io_len = lbalen->len * cbe_lun->blocksize;
1242 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1243 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0;
1244 beio->bio_cmd = BIO_FLUSH;
1245 beio->ds_trans_type = DEVSTAT_NO_DATA;
1246 DPRINTF("SYNC\n");
1247 be_lun->lun_flush(be_lun, beio);
1248}
1249
1250static void
1251ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1252{
1253 union ctl_io *io;
1254
1255 io = beio->io;
1256 ctl_free_beio(beio);
1257 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1258 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1259 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1260 ctl_config_write_done(io);
1261 return;
1262 }
1263
1264 ctl_be_block_config_write(io);
1265}
1266
1267static void
1268ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1269 union ctl_io *io)
1270{
1271 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1272 struct ctl_be_block_io *beio;
1273 struct ctl_lba_len_flags *lbalen;
1274 uint64_t len_left, lba;
1275 uint32_t pb, pbo, adj;
1276 int i, seglen;
1277 uint8_t *buf, *end;
1278
1279 DPRINTF("entered\n");
1280
1281 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1282 lbalen = ARGS(beio->io);
1283
1284 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1285 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1286 ctl_free_beio(beio);
1287 ctl_set_invalid_field(&io->scsiio,
1288 /*sks_valid*/ 1,
1289 /*command*/ 1,
1290 /*field*/ 1,
1291 /*bit_valid*/ 0,
1292 /*bit*/ 0);
1293 ctl_config_write_done(io);
1294 return;
1295 }
1296
1297 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1298 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1299 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1300 beio->bio_cmd = BIO_DELETE;
1301 beio->ds_trans_type = DEVSTAT_FREE;
1302
1303 be_lun->unmap(be_lun, beio);
1304 return;
1305 }
1306
1307 beio->bio_cmd = BIO_WRITE;
1308 beio->ds_trans_type = DEVSTAT_WRITE;
1309
1310 DPRINTF("WRITE SAME at LBA %jx len %u\n",
1311 (uintmax_t)lbalen->lba, lbalen->len);
1312
1313 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1314 if (be_lun->cbe_lun.pblockoff > 0)
1315 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1316 else
1317 pbo = 0;
1318 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1319 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1320
1321 /*
1322 * Setup the S/G entry for this chunk.
1323 */
1324 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1325 if (pb > cbe_lun->blocksize) {
1326 adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1327 seglen - pbo) % pb;
1328 if (seglen > adj)
1329 seglen -= adj;
1330 else
1331 seglen -= seglen % cbe_lun->blocksize;
1332 } else
1333 seglen -= seglen % cbe_lun->blocksize;
1334 beio->sg_segs[i].len = seglen;
1335 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1336
1337 DPRINTF("segment %d addr %p len %zd\n", i,
1338 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1339
1340 beio->num_segs++;
1341 len_left -= seglen;
1342
1343 buf = beio->sg_segs[i].addr;
1344 end = buf + seglen;
1345 for (; buf < end; buf += cbe_lun->blocksize) {
1346 if (lbalen->flags & SWS_NDOB) {
1347 memset(buf, 0, cbe_lun->blocksize);
1348 } else {
1349 memcpy(buf, io->scsiio.kern_data_ptr,
1350 cbe_lun->blocksize);
1351 }
1352 if (lbalen->flags & SWS_LBDATA)
1353 scsi_ulto4b(lbalen->lba + lba, buf);
1354 lba++;
1355 }
1356 }
1357
1358 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1359 beio->io_len = lba * cbe_lun->blocksize;
1360
1361 /* We can not do all in one run. Correct and schedule rerun. */
1362 if (len_left > 0) {
1363 lbalen->lba += lba;
1364 lbalen->len -= lba;
1365 beio->beio_cont = ctl_be_block_cw_done_ws;
1366 }
1367
1368 be_lun->dispatch(be_lun, beio);
1369}
1370
1371static void
1372ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1373 union ctl_io *io)
1374{
1375 struct ctl_be_block_io *beio;
1376 struct ctl_ptr_len_flags *ptrlen;
1377
1378 DPRINTF("entered\n");
1379
1380 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1381 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1382
1383 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1384 ctl_free_beio(beio);
1385 ctl_set_invalid_field(&io->scsiio,
1386 /*sks_valid*/ 0,
1387 /*command*/ 1,
1388 /*field*/ 0,
1389 /*bit_valid*/ 0,
1390 /*bit*/ 0);
1391 ctl_config_write_done(io);
1392 return;
1393 }
1394
1395 beio->io_len = 0;
1396 beio->io_offset = -1;
1397 beio->bio_cmd = BIO_DELETE;
1398 beio->ds_trans_type = DEVSTAT_FREE;
1399 DPRINTF("UNMAP\n");
1400 be_lun->unmap(be_lun, beio);
1401}
1402
1403static void
1404ctl_be_block_cr_done(struct ctl_be_block_io *beio)
1405{
1406 union ctl_io *io;
1407
1408 io = beio->io;
1409 ctl_free_beio(beio);
1410 ctl_config_read_done(io);
1411}
1412
1413static void
1414ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1415 union ctl_io *io)
1416{
1417 struct ctl_be_block_io *beio;
1418 struct ctl_be_block_softc *softc;
1419
1420 DPRINTF("entered\n");
1421
1422 softc = be_lun->softc;
1423 beio = ctl_alloc_beio(softc);
1424 beio->io = io;
1425 beio->lun = be_lun;
1426 beio->beio_cont = ctl_be_block_cr_done;
1427 PRIV(io)->ptr = (void *)beio;
1428
1429 switch (io->scsiio.cdb[0]) {
1430 case SERVICE_ACTION_IN: /* GET LBA STATUS */
1431 beio->bio_cmd = -1;
1432 beio->ds_trans_type = DEVSTAT_NO_DATA;
1433 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1434 beio->io_len = 0;
1435 if (be_lun->get_lba_status)
1436 be_lun->get_lba_status(be_lun, beio);
1437 else
1438 ctl_be_block_cr_done(beio);
1439 break;
1440 default:
1441 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1442 break;
1443 }
1444}
1445
1446static void
1447ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1448{
1449 union ctl_io *io;
1450
1451 io = beio->io;
1452 ctl_free_beio(beio);
1453 ctl_config_write_done(io);
1454}
1455
1456static void
1457ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1458 union ctl_io *io)
1459{
1460 struct ctl_be_block_io *beio;
1461 struct ctl_be_block_softc *softc;
1462
1463 DPRINTF("entered\n");
1464
1465 softc = be_lun->softc;
1466 beio = ctl_alloc_beio(softc);
1467 beio->io = io;
1468 beio->lun = be_lun;
1469 beio->beio_cont = ctl_be_block_cw_done;
1470 switch (io->scsiio.tag_type) {
1471 case CTL_TAG_ORDERED:
1472 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1473 break;
1474 case CTL_TAG_HEAD_OF_QUEUE:
1475 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1476 break;
1477 case CTL_TAG_UNTAGGED:
1478 case CTL_TAG_SIMPLE:
1479 case CTL_TAG_ACA:
1480 default:
1481 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1482 break;
1483 }
1484 PRIV(io)->ptr = (void *)beio;
1485
1486 switch (io->scsiio.cdb[0]) {
1487 case SYNCHRONIZE_CACHE:
1488 case SYNCHRONIZE_CACHE_16:
1489 ctl_be_block_cw_dispatch_sync(be_lun, io);
1490 break;
1491 case WRITE_SAME_10:
1492 case WRITE_SAME_16:
1493 ctl_be_block_cw_dispatch_ws(be_lun, io);
1494 break;
1495 case UNMAP:
1496 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1497 break;
1498 default:
1499 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1500 break;
1501 }
1502}
1503
1504SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t");
1505SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t");
1506SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t");
1507SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t");
1508
1509static void
1510ctl_be_block_next(struct ctl_be_block_io *beio)
1511{
1512 struct ctl_be_block_lun *be_lun;
1513 union ctl_io *io;
1514
1515 io = beio->io;
1516 be_lun = beio->lun;
1517 ctl_free_beio(beio);
1518 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1519 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1520 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1521 ctl_data_submit_done(io);
1522 return;
1523 }
1524
1525 io->io_hdr.status &= ~CTL_STATUS_MASK;
1526 io->io_hdr.status |= CTL_STATUS_NONE;
1527
1528 mtx_lock(&be_lun->queue_lock);
1529 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1530 mtx_unlock(&be_lun->queue_lock);
1531 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1532}
1533
1534static void
1535ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1536 union ctl_io *io)
1537{
1538 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1539 struct ctl_be_block_io *beio;
1540 struct ctl_be_block_softc *softc;
1541 struct ctl_lba_len_flags *lbalen;
1542 struct ctl_ptr_len_flags *bptrlen;
1543 uint64_t len_left, lbas;
1544 int i;
1545
1546 softc = be_lun->softc;
1547
1548 DPRINTF("entered\n");
1549
1550 lbalen = ARGS(io);
1551 if (lbalen->flags & CTL_LLF_WRITE) {
1552 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0);
1553 } else {
1554 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0);
1555 }
1556
1557 beio = ctl_alloc_beio(softc);
1558 beio->io = io;
1559 beio->lun = be_lun;
1560 bptrlen = PRIV(io);
1561 bptrlen->ptr = (void *)beio;
1562
1563 switch (io->scsiio.tag_type) {
1564 case CTL_TAG_ORDERED:
1565 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1566 break;
1567 case CTL_TAG_HEAD_OF_QUEUE:
1568 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1569 break;
1570 case CTL_TAG_UNTAGGED:
1571 case CTL_TAG_SIMPLE:
1572 case CTL_TAG_ACA:
1573 default:
1574 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1575 break;
1576 }
1577
1578 if (lbalen->flags & CTL_LLF_WRITE) {
1579 beio->bio_cmd = BIO_WRITE;
1580 beio->ds_trans_type = DEVSTAT_WRITE;
1581 } else {
1582 beio->bio_cmd = BIO_READ;
1583 beio->ds_trans_type = DEVSTAT_READ;
1584 }
1585
1586 DPRINTF("%s at LBA %jx len %u @%ju\n",
1587 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1588 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1589 if (lbalen->flags & CTL_LLF_COMPARE)
1590 lbas = CTLBLK_HALF_IO_SIZE;
1591 else
1592 lbas = CTLBLK_MAX_IO_SIZE;
1593 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize);
1594 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize;
1595 beio->io_len = lbas * cbe_lun->blocksize;
1596 bptrlen->len += lbas;
1597
1598 for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1599 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1600 i, CTLBLK_MAX_SEGS));
1601
1602 /*
1603 * Setup the S/G entry for this chunk.
1604 */
1605 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1606 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1607
1608 DPRINTF("segment %d addr %p len %zd\n", i,
1609 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1610
1611 /* Set up second segment for compare operation. */
1612 if (lbalen->flags & CTL_LLF_COMPARE) {
1613 beio->sg_segs[i + CTLBLK_HALF_SEGS].len =
1614 beio->sg_segs[i].len;
1615 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr =
1616 uma_zalloc(be_lun->lun_zone, M_WAITOK);
1617 }
1618
1619 beio->num_segs++;
1620 len_left -= beio->sg_segs[i].len;
1621 }
1622 if (bptrlen->len < lbalen->len)
1623 beio->beio_cont = ctl_be_block_next;
1624 io->scsiio.be_move_done = ctl_be_block_move_done;
1625 /* For compare we have separate S/G lists for read and datamove. */
1626 if (lbalen->flags & CTL_LLF_COMPARE)
1627 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1628 else
1629 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1630 io->scsiio.kern_data_len = beio->io_len;
1631 io->scsiio.kern_data_resid = 0;
1632 io->scsiio.kern_sg_entries = beio->num_segs;
1633 io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
1634
1635 /*
1636 * For the read case, we need to read the data into our buffers and
1637 * then we can send it back to the user. For the write case, we
1638 * need to get the data from the user first.
1639 */
1640 if (beio->bio_cmd == BIO_READ) {
1641 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0);
1642 be_lun->dispatch(be_lun, beio);
1643 } else {
1644 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0);
1645#ifdef CTL_TIME_IO
1646 getbinuptime(&io->io_hdr.dma_start_bt);
1647#endif
1648 ctl_datamove(io);
1649 }
1650}
1651
1652static void
1653ctl_be_block_worker(void *context, int pending)
1654{
1655 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context;
1656 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1657 union ctl_io *io;
1658 struct ctl_be_block_io *beio;
1659
1660 DPRINTF("entered\n");
1661 /*
1662 * Fetch and process I/Os from all queues. If we detect LUN
1663 * CTL_LUN_FLAG_OFFLINE status here -- it is result of a race,
1664 * so make response maximally opaque to not confuse initiator.
1665 */
1666 for (;;) {
1667 mtx_lock(&be_lun->queue_lock);
1668 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1669 if (io != NULL) {
1670 DPRINTF("datamove queue\n");
1671 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1672 ctl_io_hdr, links);
1673 mtx_unlock(&be_lun->queue_lock);
1674 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1675 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1676 ctl_set_busy(&io->scsiio);
1677 ctl_complete_beio(beio);
1678 return;
1679 }
1680 be_lun->dispatch(be_lun, beio);
1681 continue;
1682 }
1683 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1684 if (io != NULL) {
1685 DPRINTF("config write queue\n");
1686 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1687 ctl_io_hdr, links);
1688 mtx_unlock(&be_lun->queue_lock);
1689 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1690 ctl_set_busy(&io->scsiio);
1691 ctl_config_write_done(io);
1692 return;
1693 }
1694 ctl_be_block_cw_dispatch(be_lun, io);
1695 continue;
1696 }
1697 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
1698 if (io != NULL) {
1699 DPRINTF("config read queue\n");
1700 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr,
1701 ctl_io_hdr, links);
1702 mtx_unlock(&be_lun->queue_lock);
1703 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1704 ctl_set_busy(&io->scsiio);
1705 ctl_config_read_done(io);
1706 return;
1707 }
1708 ctl_be_block_cr_dispatch(be_lun, io);
1709 continue;
1710 }
1711 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1712 if (io != NULL) {
1713 DPRINTF("input queue\n");
1714 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1715 ctl_io_hdr, links);
1716 mtx_unlock(&be_lun->queue_lock);
1717 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1718 ctl_set_busy(&io->scsiio);
1719 ctl_data_submit_done(io);
1720 return;
1721 }
1722 ctl_be_block_dispatch(be_lun, io);
1723 continue;
1724 }
1725
1726 /*
1727 * If we get here, there is no work left in the queues, so
1728 * just break out and let the task queue go to sleep.
1729 */
1730 mtx_unlock(&be_lun->queue_lock);
1731 break;
1732 }
1733}
1734
1735/*
1736 * Entry point from CTL to the backend for I/O. We queue everything to a
1737 * work thread, so this just puts the I/O on a queue and wakes up the
1738 * thread.
1739 */
1740static int
1741ctl_be_block_submit(union ctl_io *io)
1742{
1743 struct ctl_be_block_lun *be_lun;
1744 struct ctl_be_lun *cbe_lun;
1745
1746 DPRINTF("entered\n");
1747
1748 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
1749 CTL_PRIV_BACKEND_LUN].ptr;
1750 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun;
1751
1752 /*
1753 * Make sure we only get SCSI I/O.
1754 */
1755 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1756 "%#x) encountered", io->io_hdr.io_type));
1757
1758 PRIV(io)->len = 0;
1759
1760 mtx_lock(&be_lun->queue_lock);
1761 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1762 mtx_unlock(&be_lun->queue_lock);
1763 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1764
1765 return (CTL_RETVAL_COMPLETE);
1766}
1767
1768static int
1769ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1770 int flag, struct thread *td)
1771{
1772 struct ctl_be_block_softc *softc;
1773 int error;
1774
1775 softc = &backend_block_softc;
1776
1777 error = 0;
1778
1779 switch (cmd) {
1780 case CTL_LUN_REQ: {
1781 struct ctl_lun_req *lun_req;
1782
1783 lun_req = (struct ctl_lun_req *)addr;
1784
1785 switch (lun_req->reqtype) {
1786 case CTL_LUNREQ_CREATE:
1787 error = ctl_be_block_create(softc, lun_req);
1788 break;
1789 case CTL_LUNREQ_RM:
1790 error = ctl_be_block_rm(softc, lun_req);
1791 break;
1792 case CTL_LUNREQ_MODIFY:
1793 error = ctl_be_block_modify(softc, lun_req);
1794 break;
1795 default:
1796 lun_req->status = CTL_LUN_ERROR;
1797 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1798 "invalid LUN request type %d",
1799 lun_req->reqtype);
1800 break;
1801 }
1802 break;
1803 }
1804 default:
1805 error = ENOTTY;
1806 break;
1807 }
1808
1809 return (error);
1810}
1811
1812static int
1813ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1814{
1815 struct ctl_be_lun *cbe_lun;
1816 struct ctl_be_block_filedata *file_data;
1817 struct ctl_lun_create_params *params;
1818 char *value;
1819 struct vattr vattr;
1820 off_t ps, pss, po, pos, us, uss, uo, uos;
1821 int error;
1822
1823 cbe_lun = &be_lun->cbe_lun;
1824 file_data = &be_lun->backend.file;
1825 params = &be_lun->params;
1826
1827 be_lun->dev_type = CTL_BE_BLOCK_FILE;
1828 be_lun->dispatch = ctl_be_block_dispatch_file;
1829 be_lun->lun_flush = ctl_be_block_flush_file;
1830 be_lun->get_lba_status = ctl_be_block_gls_file;
1831 be_lun->getattr = ctl_be_block_getattr_file;
1832 be_lun->unmap = NULL;
1833 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
1834
1835 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1836 if (error != 0) {
1837 snprintf(req->error_str, sizeof(req->error_str),
1838 "error calling VOP_GETATTR() for file %s",
1839 be_lun->dev_path);
1840 return (error);
1841 }
1842
1843 /*
1844 * Verify that we have the ability to upgrade to exclusive
1845 * access on this file so we can trap errors at open instead
1846 * of reporting them during first access.
1847 */
1848 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) {
1849 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY);
1850 if (be_lun->vn->v_iflag & VI_DOOMED) {
1851 error = EBADF;
1852 snprintf(req->error_str, sizeof(req->error_str),
1853 "error locking file %s", be_lun->dev_path);
1854 return (error);
1855 }
1856 }
1857
1858 file_data->cred = crhold(curthread->td_ucred);
1859 if (params->lun_size_bytes != 0)
1860 be_lun->size_bytes = params->lun_size_bytes;
1861 else
1862 be_lun->size_bytes = vattr.va_size;
1863
1864 /*
1865 * For files we can use any logical block size. Prefer 512 bytes
1866 * for compatibility reasons. If file's vattr.va_blocksize
1867 * (preferred I/O block size) is bigger and multiple to chosen
1868 * logical block size -- report it as physical block size.
1869 */
1870 if (params->blocksize_bytes != 0)
1871 cbe_lun->blocksize = params->blocksize_bytes;
| 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_ioctl.h>
88#include <cam/ctl/ctl_ha.h>
89#include <cam/ctl/ctl_scsi_all.h>
90#include <cam/ctl/ctl_private.h>
91#include <cam/ctl/ctl_error.h>
92
93/*
94 * The idea here is that we'll allocate enough S/G space to hold a 1MB
95 * I/O. If we get an I/O larger than that, we'll split it.
96 */
97#define CTLBLK_HALF_IO_SIZE (512 * 1024)
98#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2)
99#define CTLBLK_MAX_SEG MAXPHYS
100#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1)
101#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2)
102
103#ifdef CTLBLK_DEBUG
104#define DPRINTF(fmt, args...) \
105 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
106#else
107#define DPRINTF(fmt, args...) do {} while(0)
108#endif
109
110#define PRIV(io) \
111 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
112#define ARGS(io) \
113 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
114
115SDT_PROVIDER_DEFINE(cbb);
116
117typedef enum {
118 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01,
119 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02,
120 CTL_BE_BLOCK_LUN_WAITING = 0x04,
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_filedata {
130 struct ucred *cred;
131};
132
133union ctl_be_block_bedata {
134 struct ctl_be_block_filedata file;
135};
136
137struct ctl_be_block_io;
138struct ctl_be_block_lun;
139
140typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
141 struct ctl_be_block_io *beio);
142typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
143 const char *attrname);
144
145/*
146 * Backend LUN structure. There is a 1:1 mapping between a block device
147 * and a backend block LUN, and between a backend block LUN and a CTL LUN.
148 */
149struct ctl_be_block_lun {
150 struct ctl_lun_create_params params;
151 char lunname[32];
152 char *dev_path;
153 ctl_be_block_type dev_type;
154 struct vnode *vn;
155 union ctl_be_block_bedata backend;
156 cbb_dispatch_t dispatch;
157 cbb_dispatch_t lun_flush;
158 cbb_dispatch_t unmap;
159 cbb_dispatch_t get_lba_status;
160 cbb_getattr_t getattr;
161 uma_zone_t lun_zone;
162 uint64_t size_blocks;
163 uint64_t size_bytes;
164 struct ctl_be_block_softc *softc;
165 struct devstat *disk_stats;
166 ctl_be_block_lun_flags flags;
167 STAILQ_ENTRY(ctl_be_block_lun) links;
168 struct ctl_be_lun cbe_lun;
169 struct taskqueue *io_taskqueue;
170 struct task io_task;
171 int num_threads;
172 STAILQ_HEAD(, ctl_io_hdr) input_queue;
173 STAILQ_HEAD(, ctl_io_hdr) config_read_queue;
174 STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
175 STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
176 struct mtx_padalign io_lock;
177 struct mtx_padalign queue_lock;
178};
179
180/*
181 * Overall softc structure for the block backend module.
182 */
183struct ctl_be_block_softc {
184 struct mtx lock;
185 int num_luns;
186 STAILQ_HEAD(, ctl_be_block_lun) lun_list;
187};
188
189static struct ctl_be_block_softc backend_block_softc;
190
191/*
192 * Per-I/O information.
193 */
194struct ctl_be_block_io {
195 union ctl_io *io;
196 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS];
197 struct iovec xiovecs[CTLBLK_MAX_SEGS];
198 int bio_cmd;
199 int num_segs;
200 int num_bios_sent;
201 int num_bios_done;
202 int send_complete;
203 int num_errors;
204 struct bintime ds_t0;
205 devstat_tag_type ds_tag_type;
206 devstat_trans_flags ds_trans_type;
207 uint64_t io_len;
208 uint64_t io_offset;
209 int io_arg;
210 struct ctl_be_block_softc *softc;
211 struct ctl_be_block_lun *lun;
212 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
213};
214
215extern struct ctl_softc *control_softc;
216
217static int cbb_num_threads = 14;
218SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0,
219 "CAM Target Layer Block Backend");
220SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
221 &cbb_num_threads, 0, "Number of threads per backing file");
222
223static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
224static void ctl_free_beio(struct ctl_be_block_io *beio);
225static void ctl_complete_beio(struct ctl_be_block_io *beio);
226static int ctl_be_block_move_done(union ctl_io *io);
227static void ctl_be_block_biodone(struct bio *bio);
228static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
229 struct ctl_be_block_io *beio);
230static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
231 struct ctl_be_block_io *beio);
232static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
233 struct ctl_be_block_io *beio);
234static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
235 const char *attrname);
236static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
237 struct ctl_be_block_io *beio);
238static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
239 struct ctl_be_block_io *beio);
240static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
241 struct ctl_be_block_io *beio);
242static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
243 const char *attrname);
244static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
245 union ctl_io *io);
246static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
247 union ctl_io *io);
248static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
249 union ctl_io *io);
250static void ctl_be_block_worker(void *context, int pending);
251static int ctl_be_block_submit(union ctl_io *io);
252static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
253 int flag, struct thread *td);
254static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
255 struct ctl_lun_req *req);
256static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
257 struct ctl_lun_req *req);
258static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
259static int ctl_be_block_open(struct ctl_be_block_softc *softc,
260 struct ctl_be_block_lun *be_lun,
261 struct ctl_lun_req *req);
262static int ctl_be_block_create(struct ctl_be_block_softc *softc,
263 struct ctl_lun_req *req);
264static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
265 struct ctl_lun_req *req);
266static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
267 struct ctl_lun_req *req);
268static void ctl_be_block_lun_shutdown(void *be_lun);
269static void ctl_be_block_lun_config_status(void *be_lun,
270 ctl_lun_config_status status);
271static int ctl_be_block_config_write(union ctl_io *io);
272static int ctl_be_block_config_read(union ctl_io *io);
273static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb);
274static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname);
275int ctl_be_block_init(void);
276
277static struct ctl_backend_driver ctl_be_block_driver =
278{
279 .name = "block",
280 .flags = CTL_BE_FLAG_HAS_CONFIG,
281 .init = ctl_be_block_init,
282 .data_submit = ctl_be_block_submit,
283 .data_move_done = ctl_be_block_move_done,
284 .config_read = ctl_be_block_config_read,
285 .config_write = ctl_be_block_config_write,
286 .ioctl = ctl_be_block_ioctl,
287 .lun_info = ctl_be_block_lun_info,
288 .lun_attr = ctl_be_block_lun_attr
289};
290
291MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend");
292CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
293
294static uma_zone_t beio_zone;
295
296static struct ctl_be_block_io *
297ctl_alloc_beio(struct ctl_be_block_softc *softc)
298{
299 struct ctl_be_block_io *beio;
300
301 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO);
302 beio->softc = softc;
303 return (beio);
304}
305
306static void
307ctl_free_beio(struct ctl_be_block_io *beio)
308{
309 int duplicate_free;
310 int i;
311
312 duplicate_free = 0;
313
314 for (i = 0; i < beio->num_segs; i++) {
315 if (beio->sg_segs[i].addr == NULL)
316 duplicate_free++;
317
318 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr);
319 beio->sg_segs[i].addr = NULL;
320
321 /* For compare we had two equal S/G lists. */
322 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) {
323 uma_zfree(beio->lun->lun_zone,
324 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr);
325 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL;
326 }
327 }
328
329 if (duplicate_free > 0) {
330 printf("%s: %d duplicate frees out of %d segments\n", __func__,
331 duplicate_free, beio->num_segs);
332 }
333
334 uma_zfree(beio_zone, beio);
335}
336
337static void
338ctl_complete_beio(struct ctl_be_block_io *beio)
339{
340 union ctl_io *io = beio->io;
341
342 if (beio->beio_cont != NULL) {
343 beio->beio_cont(beio);
344 } else {
345 ctl_free_beio(beio);
346 ctl_data_submit_done(io);
347 }
348}
349
350static size_t
351cmp(uint8_t *a, uint8_t *b, size_t size)
352{
353 size_t i;
354
355 for (i = 0; i < size; i++) {
356 if (a[i] != b[i])
357 break;
358 }
359 return (i);
360}
361
362static void
363ctl_be_block_compare(union ctl_io *io)
364{
365 struct ctl_be_block_io *beio;
366 uint64_t off, res;
367 int i;
368 uint8_t info[8];
369
370 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
371 off = 0;
372 for (i = 0; i < beio->num_segs; i++) {
373 res = cmp(beio->sg_segs[i].addr,
374 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
375 beio->sg_segs[i].len);
376 off += res;
377 if (res < beio->sg_segs[i].len)
378 break;
379 }
380 if (i < beio->num_segs) {
381 scsi_u64to8b(off, info);
382 ctl_set_sense(&io->scsiio, /*current_error*/ 1,
383 /*sense_key*/ SSD_KEY_MISCOMPARE,
384 /*asc*/ 0x1D, /*ascq*/ 0x00,
385 /*type*/ SSD_ELEM_INFO,
386 /*size*/ sizeof(info), /*data*/ &info,
387 /*type*/ SSD_ELEM_NONE);
388 } else
389 ctl_set_success(&io->scsiio);
390}
391
392static int
393ctl_be_block_move_done(union ctl_io *io)
394{
395 struct ctl_be_block_io *beio;
396 struct ctl_be_block_lun *be_lun;
397 struct ctl_lba_len_flags *lbalen;
398#ifdef CTL_TIME_IO
399 struct bintime cur_bt;
400#endif
401
402 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
403 be_lun = beio->lun;
404
405 DPRINTF("entered\n");
406
407#ifdef CTL_TIME_IO
408 getbinuptime(&cur_bt);
409 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
410 bintime_add(&io->io_hdr.dma_bt, &cur_bt);
411#endif
412 io->io_hdr.num_dmas++;
413 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
414
415 /*
416 * We set status at this point for read commands, and write
417 * commands with errors.
418 */
419 if (io->io_hdr.flags & CTL_FLAG_ABORT) {
420 ;
421 } else if ((io->io_hdr.port_status == 0) &&
422 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
423 lbalen = ARGS(beio->io);
424 if (lbalen->flags & CTL_LLF_READ) {
425 ctl_set_success(&io->scsiio);
426 } else if (lbalen->flags & CTL_LLF_COMPARE) {
427 /* We have two data blocks ready for comparison. */
428 ctl_be_block_compare(io);
429 }
430 } else if ((io->io_hdr.port_status != 0) &&
431 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
432 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
433 /*
434 * For hardware error sense keys, the sense key
435 * specific value is defined to be a retry count,
436 * but we use it to pass back an internal FETD
437 * error code. XXX KDM Hopefully the FETD is only
438 * using 16 bits for an error code, since that's
439 * all the space we have in the sks field.
440 */
441 ctl_set_internal_failure(&io->scsiio,
442 /*sks_valid*/ 1,
443 /*retry_count*/
444 io->io_hdr.port_status);
445 }
446
447 /*
448 * If this is a read, or a write with errors, it is done.
449 */
450 if ((beio->bio_cmd == BIO_READ)
451 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
452 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
453 ctl_complete_beio(beio);
454 return (0);
455 }
456
457 /*
458 * At this point, we have a write and the DMA completed
459 * successfully. We now have to queue it to the task queue to
460 * execute the backend I/O. That is because we do blocking
461 * memory allocations, and in the file backing case, blocking I/O.
462 * This move done routine is generally called in the SIM's
463 * interrupt context, and therefore we cannot block.
464 */
465 mtx_lock(&be_lun->queue_lock);
466 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
467 mtx_unlock(&be_lun->queue_lock);
468 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
469
470 return (0);
471}
472
473static void
474ctl_be_block_biodone(struct bio *bio)
475{
476 struct ctl_be_block_io *beio;
477 struct ctl_be_block_lun *be_lun;
478 union ctl_io *io;
479 int error;
480
481 beio = bio->bio_caller1;
482 be_lun = beio->lun;
483 io = beio->io;
484
485 DPRINTF("entered\n");
486
487 error = bio->bio_error;
488 mtx_lock(&be_lun->io_lock);
489 if (error != 0)
490 beio->num_errors++;
491
492 beio->num_bios_done++;
493
494 /*
495 * XXX KDM will this cause WITNESS to complain? Holding a lock
496 * during the free might cause it to complain.
497 */
498 g_destroy_bio(bio);
499
500 /*
501 * If the send complete bit isn't set, or we aren't the last I/O to
502 * complete, then we're done.
503 */
504 if ((beio->send_complete == 0)
505 || (beio->num_bios_done < beio->num_bios_sent)) {
506 mtx_unlock(&be_lun->io_lock);
507 return;
508 }
509
510 /*
511 * At this point, we've verified that we are the last I/O to
512 * complete, so it's safe to drop the lock.
513 */
514 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
515 beio->ds_tag_type, beio->ds_trans_type,
516 /*now*/ NULL, /*then*/&beio->ds_t0);
517 mtx_unlock(&be_lun->io_lock);
518
519 /*
520 * If there are any errors from the backing device, we fail the
521 * entire I/O with a medium error.
522 */
523 if (beio->num_errors > 0) {
524 if (error == EOPNOTSUPP) {
525 ctl_set_invalid_opcode(&io->scsiio);
526 } else if (error == ENOSPC || error == EDQUOT) {
527 ctl_set_space_alloc_fail(&io->scsiio);
528 } else if (error == EROFS || error == EACCES) {
529 ctl_set_hw_write_protected(&io->scsiio);
530 } else if (beio->bio_cmd == BIO_FLUSH) {
531 /* XXX KDM is there is a better error here? */
532 ctl_set_internal_failure(&io->scsiio,
533 /*sks_valid*/ 1,
534 /*retry_count*/ 0xbad2);
535 } else {
536 ctl_set_medium_error(&io->scsiio,
537 beio->bio_cmd == BIO_READ);
538 }
539 ctl_complete_beio(beio);
540 return;
541 }
542
543 /*
544 * If this is a write, a flush, a delete or verify, we're all done.
545 * If this is a read, we can now send the data to the user.
546 */
547 if ((beio->bio_cmd == BIO_WRITE)
548 || (beio->bio_cmd == BIO_FLUSH)
549 || (beio->bio_cmd == BIO_DELETE)
550 || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
551 ctl_set_success(&io->scsiio);
552 ctl_complete_beio(beio);
553 } else {
554 if ((ARGS(io)->flags & CTL_LLF_READ) &&
555 beio->beio_cont == NULL) {
556 ctl_set_success(&io->scsiio);
557 ctl_serseq_done(io);
558 }
559#ifdef CTL_TIME_IO
560 getbinuptime(&io->io_hdr.dma_start_bt);
561#endif
562 ctl_datamove(io);
563 }
564}
565
566static void
567ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
568 struct ctl_be_block_io *beio)
569{
570 union ctl_io *io = beio->io;
571 struct mount *mountpoint;
572 int error, lock_flags;
573
574 DPRINTF("entered\n");
575
576 binuptime(&beio->ds_t0);
577 mtx_lock(&be_lun->io_lock);
578 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
579 mtx_unlock(&be_lun->io_lock);
580
581 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
582
583 if (MNT_SHARED_WRITES(mountpoint) ||
584 ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
585 lock_flags = LK_SHARED;
586 else
587 lock_flags = LK_EXCLUSIVE;
588 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
589 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT,
590 curthread);
591 VOP_UNLOCK(be_lun->vn, 0);
592
593 vn_finished_write(mountpoint);
594
595 mtx_lock(&be_lun->io_lock);
596 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
597 beio->ds_tag_type, beio->ds_trans_type,
598 /*now*/ NULL, /*then*/&beio->ds_t0);
599 mtx_unlock(&be_lun->io_lock);
600
601 if (error == 0)
602 ctl_set_success(&io->scsiio);
603 else {
604 /* XXX KDM is there is a better error here? */
605 ctl_set_internal_failure(&io->scsiio,
606 /*sks_valid*/ 1,
607 /*retry_count*/ 0xbad1);
608 }
609
610 ctl_complete_beio(beio);
611}
612
613SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t");
614SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t");
615SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t");
616SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t");
617
618static void
619ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
620 struct ctl_be_block_io *beio)
621{
622 struct ctl_be_block_filedata *file_data;
623 union ctl_io *io;
624 struct uio xuio;
625 struct iovec *xiovec;
626 size_t s;
627 int error, flags, i;
628
629 DPRINTF("entered\n");
630
631 file_data = &be_lun->backend.file;
632 io = beio->io;
633 flags = 0;
634 if (ARGS(io)->flags & CTL_LLF_DPO)
635 flags |= IO_DIRECT;
636 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
637 flags |= IO_SYNC;
638
639 bzero(&xuio, sizeof(xuio));
640 if (beio->bio_cmd == BIO_READ) {
641 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
642 xuio.uio_rw = UIO_READ;
643 } else {
644 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
645 xuio.uio_rw = UIO_WRITE;
646 }
647 xuio.uio_offset = beio->io_offset;
648 xuio.uio_resid = beio->io_len;
649 xuio.uio_segflg = UIO_SYSSPACE;
650 xuio.uio_iov = beio->xiovecs;
651 xuio.uio_iovcnt = beio->num_segs;
652 xuio.uio_td = curthread;
653
654 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
655 xiovec->iov_base = beio->sg_segs[i].addr;
656 xiovec->iov_len = beio->sg_segs[i].len;
657 }
658
659 binuptime(&beio->ds_t0);
660 mtx_lock(&be_lun->io_lock);
661 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
662 mtx_unlock(&be_lun->io_lock);
663
664 if (beio->bio_cmd == BIO_READ) {
665 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
666
667 /*
668 * UFS pays attention to IO_DIRECT for reads. If the
669 * DIRECTIO option is configured into the kernel, it calls
670 * ffs_rawread(). But that only works for single-segment
671 * uios with user space addresses. In our case, with a
672 * kernel uio, it still reads into the buffer cache, but it
673 * will just try to release the buffer from the cache later
674 * on in ffs_read().
675 *
676 * ZFS does not pay attention to IO_DIRECT for reads.
677 *
678 * UFS does not pay attention to IO_SYNC for reads.
679 *
680 * ZFS pays attention to IO_SYNC (which translates into the
681 * Solaris define FRSYNC for zfs_read()) for reads. It
682 * attempts to sync the file before reading.
683 */
684 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
685
686 VOP_UNLOCK(be_lun->vn, 0);
687 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
688 if (error == 0 && xuio.uio_resid > 0) {
689 /*
690 * If we red less then requested (EOF), then
691 * we should clean the rest of the buffer.
692 */
693 s = beio->io_len - xuio.uio_resid;
694 for (i = 0; i < beio->num_segs; i++) {
695 if (s >= beio->sg_segs[i].len) {
696 s -= beio->sg_segs[i].len;
697 continue;
698 }
699 bzero((uint8_t *)beio->sg_segs[i].addr + s,
700 beio->sg_segs[i].len - s);
701 s = 0;
702 }
703 }
704 } else {
705 struct mount *mountpoint;
706 int lock_flags;
707
708 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
709
710 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL)
711 && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
712 lock_flags = LK_SHARED;
713 else
714 lock_flags = LK_EXCLUSIVE;
715 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
716
717 /*
718 * UFS pays attention to IO_DIRECT for writes. The write
719 * is done asynchronously. (Normally the write would just
720 * get put into cache.
721 *
722 * UFS pays attention to IO_SYNC for writes. It will
723 * attempt to write the buffer out synchronously if that
724 * flag is set.
725 *
726 * ZFS does not pay attention to IO_DIRECT for writes.
727 *
728 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
729 * for writes. It will flush the transaction from the
730 * cache before returning.
731 */
732 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
733 VOP_UNLOCK(be_lun->vn, 0);
734
735 vn_finished_write(mountpoint);
736 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
737 }
738
739 mtx_lock(&be_lun->io_lock);
740 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
741 beio->ds_tag_type, beio->ds_trans_type,
742 /*now*/ NULL, /*then*/&beio->ds_t0);
743 mtx_unlock(&be_lun->io_lock);
744
745 /*
746 * If we got an error, set the sense data to "MEDIUM ERROR" and
747 * return the I/O to the user.
748 */
749 if (error != 0) {
750 if (error == ENOSPC || error == EDQUOT) {
751 ctl_set_space_alloc_fail(&io->scsiio);
752 } else if (error == EROFS || error == EACCES) {
753 ctl_set_hw_write_protected(&io->scsiio);
754 } else {
755 ctl_set_medium_error(&io->scsiio,
756 beio->bio_cmd == BIO_READ);
757 }
758 ctl_complete_beio(beio);
759 return;
760 }
761
762 /*
763 * If this is a write or a verify, we're all done.
764 * If this is a read, we can now send the data to the user.
765 */
766 if ((beio->bio_cmd == BIO_WRITE) ||
767 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
768 ctl_set_success(&io->scsiio);
769 ctl_complete_beio(beio);
770 } else {
771 if ((ARGS(io)->flags & CTL_LLF_READ) &&
772 beio->beio_cont == NULL) {
773 ctl_set_success(&io->scsiio);
774 ctl_serseq_done(io);
775 }
776#ifdef CTL_TIME_IO
777 getbinuptime(&io->io_hdr.dma_start_bt);
778#endif
779 ctl_datamove(io);
780 }
781}
782
783static void
784ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
785 struct ctl_be_block_io *beio)
786{
787 union ctl_io *io = beio->io;
788 struct ctl_lba_len_flags *lbalen = ARGS(io);
789 struct scsi_get_lba_status_data *data;
790 off_t roff, off;
791 int error, status;
792
793 DPRINTF("entered\n");
794
795 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
796 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
797 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
798 0, curthread->td_ucred, curthread);
799 if (error == 0 && off > roff)
800 status = 0; /* mapped up to off */
801 else {
802 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
803 0, curthread->td_ucred, curthread);
804 if (error == 0 && off > roff)
805 status = 1; /* deallocated up to off */
806 else {
807 status = 0; /* unknown up to the end */
808 off = be_lun->size_bytes;
809 }
810 }
811 VOP_UNLOCK(be_lun->vn, 0);
812
813 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
814 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
815 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
816 lbalen->lba), data->descr[0].length);
817 data->descr[0].status = status;
818
819 ctl_complete_beio(beio);
820}
821
822static uint64_t
823ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
824{
825 struct vattr vattr;
826 struct statfs statfs;
827 uint64_t val;
828 int error;
829
830 val = UINT64_MAX;
831 if (be_lun->vn == NULL)
832 return (val);
833 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
834 if (strcmp(attrname, "blocksused") == 0) {
835 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
836 if (error == 0)
837 val = vattr.va_bytes / be_lun->cbe_lun.blocksize;
838 }
839 if (strcmp(attrname, "blocksavail") == 0 &&
840 (be_lun->vn->v_iflag & VI_DOOMED) == 0) {
841 error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
842 if (error == 0)
843 val = statfs.f_bavail * statfs.f_bsize /
844 be_lun->cbe_lun.blocksize;
845 }
846 VOP_UNLOCK(be_lun->vn, 0);
847 return (val);
848}
849
850static void
851ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
852 struct ctl_be_block_io *beio)
853{
854 union ctl_io *io;
855 struct cdevsw *csw;
856 struct cdev *dev;
857 struct uio xuio;
858 struct iovec *xiovec;
859 int error, flags, i, ref;
860
861 DPRINTF("entered\n");
862
863 io = beio->io;
864 flags = 0;
865 if (ARGS(io)->flags & CTL_LLF_DPO)
866 flags |= IO_DIRECT;
867 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
868 flags |= IO_SYNC;
869
870 bzero(&xuio, sizeof(xuio));
871 if (beio->bio_cmd == BIO_READ) {
872 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
873 xuio.uio_rw = UIO_READ;
874 } else {
875 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
876 xuio.uio_rw = UIO_WRITE;
877 }
878 xuio.uio_offset = beio->io_offset;
879 xuio.uio_resid = beio->io_len;
880 xuio.uio_segflg = UIO_SYSSPACE;
881 xuio.uio_iov = beio->xiovecs;
882 xuio.uio_iovcnt = beio->num_segs;
883 xuio.uio_td = curthread;
884
885 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
886 xiovec->iov_base = beio->sg_segs[i].addr;
887 xiovec->iov_len = beio->sg_segs[i].len;
888 }
889
890 binuptime(&beio->ds_t0);
891 mtx_lock(&be_lun->io_lock);
892 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
893 mtx_unlock(&be_lun->io_lock);
894
895 csw = devvn_refthread(be_lun->vn, &dev, &ref);
896 if (csw) {
897 if (beio->bio_cmd == BIO_READ)
898 error = csw->d_read(dev, &xuio, flags);
899 else
900 error = csw->d_write(dev, &xuio, flags);
901 dev_relthread(dev, ref);
902 } else
903 error = ENXIO;
904
905 if (beio->bio_cmd == BIO_READ)
906 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
907 else
908 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
909
910 mtx_lock(&be_lun->io_lock);
911 devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
912 beio->ds_tag_type, beio->ds_trans_type,
913 /*now*/ NULL, /*then*/&beio->ds_t0);
914 mtx_unlock(&be_lun->io_lock);
915
916 /*
917 * If we got an error, set the sense data to "MEDIUM ERROR" and
918 * return the I/O to the user.
919 */
920 if (error != 0) {
921 if (error == ENOSPC || error == EDQUOT) {
922 ctl_set_space_alloc_fail(&io->scsiio);
923 } else if (error == EROFS || error == EACCES) {
924 ctl_set_hw_write_protected(&io->scsiio);
925 } else {
926 ctl_set_medium_error(&io->scsiio,
927 beio->bio_cmd == BIO_READ);
928 }
929 ctl_complete_beio(beio);
930 return;
931 }
932
933 /*
934 * If this is a write or a verify, we're all done.
935 * If this is a read, we can now send the data to the user.
936 */
937 if ((beio->bio_cmd == BIO_WRITE) ||
938 (ARGS(io)->flags & CTL_LLF_VERIFY)) {
939 ctl_set_success(&io->scsiio);
940 ctl_complete_beio(beio);
941 } else {
942 if ((ARGS(io)->flags & CTL_LLF_READ) &&
943 beio->beio_cont == NULL) {
944 ctl_set_success(&io->scsiio);
945 ctl_serseq_done(io);
946 }
947#ifdef CTL_TIME_IO
948 getbinuptime(&io->io_hdr.dma_start_bt);
949#endif
950 ctl_datamove(io);
951 }
952}
953
954static void
955ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
956 struct ctl_be_block_io *beio)
957{
958 union ctl_io *io = beio->io;
959 struct cdevsw *csw;
960 struct cdev *dev;
961 struct ctl_lba_len_flags *lbalen = ARGS(io);
962 struct scsi_get_lba_status_data *data;
963 off_t roff, off;
964 int error, ref, status;
965
966 DPRINTF("entered\n");
967
968 csw = devvn_refthread(be_lun->vn, &dev, &ref);
969 if (csw == NULL) {
970 status = 0; /* unknown up to the end */
971 off = be_lun->size_bytes;
972 goto done;
973 }
974 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
975 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD,
976 curthread);
977 if (error == 0 && off > roff)
978 status = 0; /* mapped up to off */
979 else {
980 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD,
981 curthread);
982 if (error == 0 && off > roff)
983 status = 1; /* deallocated up to off */
984 else {
985 status = 0; /* unknown up to the end */
986 off = be_lun->size_bytes;
987 }
988 }
989 dev_relthread(dev, ref);
990
991done:
992 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
993 scsi_u64to8b(lbalen->lba, data->descr[0].addr);
994 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
995 lbalen->lba), data->descr[0].length);
996 data->descr[0].status = status;
997
998 ctl_complete_beio(beio);
999}
1000
1001static void
1002ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
1003 struct ctl_be_block_io *beio)
1004{
1005 struct bio *bio;
1006 struct cdevsw *csw;
1007 struct cdev *dev;
1008 int ref;
1009
1010 DPRINTF("entered\n");
1011
1012 /* This can't fail, it's a blocking allocation. */
1013 bio = g_alloc_bio();
1014
1015 bio->bio_cmd = BIO_FLUSH;
1016 bio->bio_offset = 0;
1017 bio->bio_data = 0;
1018 bio->bio_done = ctl_be_block_biodone;
1019 bio->bio_caller1 = beio;
1020 bio->bio_pblkno = 0;
1021
1022 /*
1023 * We don't need to acquire the LUN lock here, because we are only
1024 * sending one bio, and so there is no other context to synchronize
1025 * with.
1026 */
1027 beio->num_bios_sent = 1;
1028 beio->send_complete = 1;
1029
1030 binuptime(&beio->ds_t0);
1031 mtx_lock(&be_lun->io_lock);
1032 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1033 mtx_unlock(&be_lun->io_lock);
1034
1035 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1036 if (csw) {
1037 bio->bio_dev = dev;
1038 csw->d_strategy(bio);
1039 dev_relthread(dev, ref);
1040 } else {
1041 bio->bio_error = ENXIO;
1042 ctl_be_block_biodone(bio);
1043 }
1044}
1045
1046static void
1047ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1048 struct ctl_be_block_io *beio,
1049 uint64_t off, uint64_t len, int last)
1050{
1051 struct bio *bio;
1052 uint64_t maxlen;
1053 struct cdevsw *csw;
1054 struct cdev *dev;
1055 int ref;
1056
1057 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1058 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize);
1059 while (len > 0) {
1060 bio = g_alloc_bio();
1061 bio->bio_cmd = BIO_DELETE;
1062 bio->bio_dev = dev;
1063 bio->bio_offset = off;
1064 bio->bio_length = MIN(len, maxlen);
1065 bio->bio_data = 0;
1066 bio->bio_done = ctl_be_block_biodone;
1067 bio->bio_caller1 = beio;
1068 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize;
1069
1070 off += bio->bio_length;
1071 len -= bio->bio_length;
1072
1073 mtx_lock(&be_lun->io_lock);
1074 beio->num_bios_sent++;
1075 if (last && len == 0)
1076 beio->send_complete = 1;
1077 mtx_unlock(&be_lun->io_lock);
1078
1079 if (csw) {
1080 csw->d_strategy(bio);
1081 } else {
1082 bio->bio_error = ENXIO;
1083 ctl_be_block_biodone(bio);
1084 }
1085 }
1086 if (csw)
1087 dev_relthread(dev, ref);
1088}
1089
1090static void
1091ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1092 struct ctl_be_block_io *beio)
1093{
1094 union ctl_io *io;
1095 struct ctl_ptr_len_flags *ptrlen;
1096 struct scsi_unmap_desc *buf, *end;
1097 uint64_t len;
1098
1099 io = beio->io;
1100
1101 DPRINTF("entered\n");
1102
1103 binuptime(&beio->ds_t0);
1104 mtx_lock(&be_lun->io_lock);
1105 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1106 mtx_unlock(&be_lun->io_lock);
1107
1108 if (beio->io_offset == -1) {
1109 beio->io_len = 0;
1110 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1111 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1112 end = buf + ptrlen->len / sizeof(*buf);
1113 for (; buf < end; buf++) {
1114 len = (uint64_t)scsi_4btoul(buf->length) *
1115 be_lun->cbe_lun.blocksize;
1116 beio->io_len += len;
1117 ctl_be_block_unmap_dev_range(be_lun, beio,
1118 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize,
1119 len, (end - buf < 2) ? TRUE : FALSE);
1120 }
1121 } else
1122 ctl_be_block_unmap_dev_range(be_lun, beio,
1123 beio->io_offset, beio->io_len, TRUE);
1124}
1125
1126static void
1127ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1128 struct ctl_be_block_io *beio)
1129{
1130 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1131 struct bio *bio;
1132 struct cdevsw *csw;
1133 struct cdev *dev;
1134 off_t cur_offset;
1135 int i, max_iosize, ref;
1136
1137 DPRINTF("entered\n");
1138 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1139
1140 /*
1141 * We have to limit our I/O size to the maximum supported by the
1142 * backend device. Hopefully it is MAXPHYS. If the driver doesn't
1143 * set it properly, use DFLTPHYS.
1144 */
1145 if (csw) {
1146 max_iosize = dev->si_iosize_max;
1147 if (max_iosize < PAGE_SIZE)
1148 max_iosize = DFLTPHYS;
1149 } else
1150 max_iosize = DFLTPHYS;
1151
1152 cur_offset = beio->io_offset;
1153 for (i = 0; i < beio->num_segs; i++) {
1154 size_t cur_size;
1155 uint8_t *cur_ptr;
1156
1157 cur_size = beio->sg_segs[i].len;
1158 cur_ptr = beio->sg_segs[i].addr;
1159
1160 while (cur_size > 0) {
1161 /* This can't fail, it's a blocking allocation. */
1162 bio = g_alloc_bio();
1163
1164 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1165
1166 bio->bio_cmd = beio->bio_cmd;
1167 bio->bio_dev = dev;
1168 bio->bio_caller1 = beio;
1169 bio->bio_length = min(cur_size, max_iosize);
1170 bio->bio_offset = cur_offset;
1171 bio->bio_data = cur_ptr;
1172 bio->bio_done = ctl_be_block_biodone;
1173 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize;
1174
1175 cur_offset += bio->bio_length;
1176 cur_ptr += bio->bio_length;
1177 cur_size -= bio->bio_length;
1178
1179 TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1180 beio->num_bios_sent++;
1181 }
1182 }
1183 binuptime(&beio->ds_t0);
1184 mtx_lock(&be_lun->io_lock);
1185 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1186 beio->send_complete = 1;
1187 mtx_unlock(&be_lun->io_lock);
1188
1189 /*
1190 * Fire off all allocated requests!
1191 */
1192 while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1193 TAILQ_REMOVE(&queue, bio, bio_queue);
1194 if (csw)
1195 csw->d_strategy(bio);
1196 else {
1197 bio->bio_error = ENXIO;
1198 ctl_be_block_biodone(bio);
1199 }
1200 }
1201 if (csw)
1202 dev_relthread(dev, ref);
1203}
1204
1205static uint64_t
1206ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1207{
1208 struct diocgattr_arg arg;
1209 struct cdevsw *csw;
1210 struct cdev *dev;
1211 int error, ref;
1212
1213 csw = devvn_refthread(be_lun->vn, &dev, &ref);
1214 if (csw == NULL)
1215 return (UINT64_MAX);
1216 strlcpy(arg.name, attrname, sizeof(arg.name));
1217 arg.len = sizeof(arg.value.off);
1218 if (csw->d_ioctl) {
1219 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
1220 curthread);
1221 } else
1222 error = ENODEV;
1223 dev_relthread(dev, ref);
1224 if (error != 0)
1225 return (UINT64_MAX);
1226 return (arg.value.off);
1227}
1228
1229static void
1230ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun,
1231 union ctl_io *io)
1232{
1233 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1234 struct ctl_be_block_io *beio;
1235 struct ctl_lba_len_flags *lbalen;
1236
1237 DPRINTF("entered\n");
1238 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1239 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1240
1241 beio->io_len = lbalen->len * cbe_lun->blocksize;
1242 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1243 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0;
1244 beio->bio_cmd = BIO_FLUSH;
1245 beio->ds_trans_type = DEVSTAT_NO_DATA;
1246 DPRINTF("SYNC\n");
1247 be_lun->lun_flush(be_lun, beio);
1248}
1249
1250static void
1251ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1252{
1253 union ctl_io *io;
1254
1255 io = beio->io;
1256 ctl_free_beio(beio);
1257 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1258 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1259 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1260 ctl_config_write_done(io);
1261 return;
1262 }
1263
1264 ctl_be_block_config_write(io);
1265}
1266
1267static void
1268ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1269 union ctl_io *io)
1270{
1271 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1272 struct ctl_be_block_io *beio;
1273 struct ctl_lba_len_flags *lbalen;
1274 uint64_t len_left, lba;
1275 uint32_t pb, pbo, adj;
1276 int i, seglen;
1277 uint8_t *buf, *end;
1278
1279 DPRINTF("entered\n");
1280
1281 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1282 lbalen = ARGS(beio->io);
1283
1284 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1285 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1286 ctl_free_beio(beio);
1287 ctl_set_invalid_field(&io->scsiio,
1288 /*sks_valid*/ 1,
1289 /*command*/ 1,
1290 /*field*/ 1,
1291 /*bit_valid*/ 0,
1292 /*bit*/ 0);
1293 ctl_config_write_done(io);
1294 return;
1295 }
1296
1297 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1298 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1299 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1300 beio->bio_cmd = BIO_DELETE;
1301 beio->ds_trans_type = DEVSTAT_FREE;
1302
1303 be_lun->unmap(be_lun, beio);
1304 return;
1305 }
1306
1307 beio->bio_cmd = BIO_WRITE;
1308 beio->ds_trans_type = DEVSTAT_WRITE;
1309
1310 DPRINTF("WRITE SAME at LBA %jx len %u\n",
1311 (uintmax_t)lbalen->lba, lbalen->len);
1312
1313 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1314 if (be_lun->cbe_lun.pblockoff > 0)
1315 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1316 else
1317 pbo = 0;
1318 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1319 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1320
1321 /*
1322 * Setup the S/G entry for this chunk.
1323 */
1324 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1325 if (pb > cbe_lun->blocksize) {
1326 adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1327 seglen - pbo) % pb;
1328 if (seglen > adj)
1329 seglen -= adj;
1330 else
1331 seglen -= seglen % cbe_lun->blocksize;
1332 } else
1333 seglen -= seglen % cbe_lun->blocksize;
1334 beio->sg_segs[i].len = seglen;
1335 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1336
1337 DPRINTF("segment %d addr %p len %zd\n", i,
1338 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1339
1340 beio->num_segs++;
1341 len_left -= seglen;
1342
1343 buf = beio->sg_segs[i].addr;
1344 end = buf + seglen;
1345 for (; buf < end; buf += cbe_lun->blocksize) {
1346 if (lbalen->flags & SWS_NDOB) {
1347 memset(buf, 0, cbe_lun->blocksize);
1348 } else {
1349 memcpy(buf, io->scsiio.kern_data_ptr,
1350 cbe_lun->blocksize);
1351 }
1352 if (lbalen->flags & SWS_LBDATA)
1353 scsi_ulto4b(lbalen->lba + lba, buf);
1354 lba++;
1355 }
1356 }
1357
1358 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1359 beio->io_len = lba * cbe_lun->blocksize;
1360
1361 /* We can not do all in one run. Correct and schedule rerun. */
1362 if (len_left > 0) {
1363 lbalen->lba += lba;
1364 lbalen->len -= lba;
1365 beio->beio_cont = ctl_be_block_cw_done_ws;
1366 }
1367
1368 be_lun->dispatch(be_lun, beio);
1369}
1370
1371static void
1372ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1373 union ctl_io *io)
1374{
1375 struct ctl_be_block_io *beio;
1376 struct ctl_ptr_len_flags *ptrlen;
1377
1378 DPRINTF("entered\n");
1379
1380 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1381 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1382
1383 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1384 ctl_free_beio(beio);
1385 ctl_set_invalid_field(&io->scsiio,
1386 /*sks_valid*/ 0,
1387 /*command*/ 1,
1388 /*field*/ 0,
1389 /*bit_valid*/ 0,
1390 /*bit*/ 0);
1391 ctl_config_write_done(io);
1392 return;
1393 }
1394
1395 beio->io_len = 0;
1396 beio->io_offset = -1;
1397 beio->bio_cmd = BIO_DELETE;
1398 beio->ds_trans_type = DEVSTAT_FREE;
1399 DPRINTF("UNMAP\n");
1400 be_lun->unmap(be_lun, beio);
1401}
1402
1403static void
1404ctl_be_block_cr_done(struct ctl_be_block_io *beio)
1405{
1406 union ctl_io *io;
1407
1408 io = beio->io;
1409 ctl_free_beio(beio);
1410 ctl_config_read_done(io);
1411}
1412
1413static void
1414ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1415 union ctl_io *io)
1416{
1417 struct ctl_be_block_io *beio;
1418 struct ctl_be_block_softc *softc;
1419
1420 DPRINTF("entered\n");
1421
1422 softc = be_lun->softc;
1423 beio = ctl_alloc_beio(softc);
1424 beio->io = io;
1425 beio->lun = be_lun;
1426 beio->beio_cont = ctl_be_block_cr_done;
1427 PRIV(io)->ptr = (void *)beio;
1428
1429 switch (io->scsiio.cdb[0]) {
1430 case SERVICE_ACTION_IN: /* GET LBA STATUS */
1431 beio->bio_cmd = -1;
1432 beio->ds_trans_type = DEVSTAT_NO_DATA;
1433 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1434 beio->io_len = 0;
1435 if (be_lun->get_lba_status)
1436 be_lun->get_lba_status(be_lun, beio);
1437 else
1438 ctl_be_block_cr_done(beio);
1439 break;
1440 default:
1441 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1442 break;
1443 }
1444}
1445
1446static void
1447ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1448{
1449 union ctl_io *io;
1450
1451 io = beio->io;
1452 ctl_free_beio(beio);
1453 ctl_config_write_done(io);
1454}
1455
1456static void
1457ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1458 union ctl_io *io)
1459{
1460 struct ctl_be_block_io *beio;
1461 struct ctl_be_block_softc *softc;
1462
1463 DPRINTF("entered\n");
1464
1465 softc = be_lun->softc;
1466 beio = ctl_alloc_beio(softc);
1467 beio->io = io;
1468 beio->lun = be_lun;
1469 beio->beio_cont = ctl_be_block_cw_done;
1470 switch (io->scsiio.tag_type) {
1471 case CTL_TAG_ORDERED:
1472 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1473 break;
1474 case CTL_TAG_HEAD_OF_QUEUE:
1475 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1476 break;
1477 case CTL_TAG_UNTAGGED:
1478 case CTL_TAG_SIMPLE:
1479 case CTL_TAG_ACA:
1480 default:
1481 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1482 break;
1483 }
1484 PRIV(io)->ptr = (void *)beio;
1485
1486 switch (io->scsiio.cdb[0]) {
1487 case SYNCHRONIZE_CACHE:
1488 case SYNCHRONIZE_CACHE_16:
1489 ctl_be_block_cw_dispatch_sync(be_lun, io);
1490 break;
1491 case WRITE_SAME_10:
1492 case WRITE_SAME_16:
1493 ctl_be_block_cw_dispatch_ws(be_lun, io);
1494 break;
1495 case UNMAP:
1496 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1497 break;
1498 default:
1499 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1500 break;
1501 }
1502}
1503
1504SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t");
1505SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t");
1506SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t");
1507SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t");
1508
1509static void
1510ctl_be_block_next(struct ctl_be_block_io *beio)
1511{
1512 struct ctl_be_block_lun *be_lun;
1513 union ctl_io *io;
1514
1515 io = beio->io;
1516 be_lun = beio->lun;
1517 ctl_free_beio(beio);
1518 if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1519 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1520 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1521 ctl_data_submit_done(io);
1522 return;
1523 }
1524
1525 io->io_hdr.status &= ~CTL_STATUS_MASK;
1526 io->io_hdr.status |= CTL_STATUS_NONE;
1527
1528 mtx_lock(&be_lun->queue_lock);
1529 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1530 mtx_unlock(&be_lun->queue_lock);
1531 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1532}
1533
1534static void
1535ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1536 union ctl_io *io)
1537{
1538 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1539 struct ctl_be_block_io *beio;
1540 struct ctl_be_block_softc *softc;
1541 struct ctl_lba_len_flags *lbalen;
1542 struct ctl_ptr_len_flags *bptrlen;
1543 uint64_t len_left, lbas;
1544 int i;
1545
1546 softc = be_lun->softc;
1547
1548 DPRINTF("entered\n");
1549
1550 lbalen = ARGS(io);
1551 if (lbalen->flags & CTL_LLF_WRITE) {
1552 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0);
1553 } else {
1554 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0);
1555 }
1556
1557 beio = ctl_alloc_beio(softc);
1558 beio->io = io;
1559 beio->lun = be_lun;
1560 bptrlen = PRIV(io);
1561 bptrlen->ptr = (void *)beio;
1562
1563 switch (io->scsiio.tag_type) {
1564 case CTL_TAG_ORDERED:
1565 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1566 break;
1567 case CTL_TAG_HEAD_OF_QUEUE:
1568 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1569 break;
1570 case CTL_TAG_UNTAGGED:
1571 case CTL_TAG_SIMPLE:
1572 case CTL_TAG_ACA:
1573 default:
1574 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1575 break;
1576 }
1577
1578 if (lbalen->flags & CTL_LLF_WRITE) {
1579 beio->bio_cmd = BIO_WRITE;
1580 beio->ds_trans_type = DEVSTAT_WRITE;
1581 } else {
1582 beio->bio_cmd = BIO_READ;
1583 beio->ds_trans_type = DEVSTAT_READ;
1584 }
1585
1586 DPRINTF("%s at LBA %jx len %u @%ju\n",
1587 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1588 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1589 if (lbalen->flags & CTL_LLF_COMPARE)
1590 lbas = CTLBLK_HALF_IO_SIZE;
1591 else
1592 lbas = CTLBLK_MAX_IO_SIZE;
1593 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize);
1594 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize;
1595 beio->io_len = lbas * cbe_lun->blocksize;
1596 bptrlen->len += lbas;
1597
1598 for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1599 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1600 i, CTLBLK_MAX_SEGS));
1601
1602 /*
1603 * Setup the S/G entry for this chunk.
1604 */
1605 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1606 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1607
1608 DPRINTF("segment %d addr %p len %zd\n", i,
1609 beio->sg_segs[i].addr, beio->sg_segs[i].len);
1610
1611 /* Set up second segment for compare operation. */
1612 if (lbalen->flags & CTL_LLF_COMPARE) {
1613 beio->sg_segs[i + CTLBLK_HALF_SEGS].len =
1614 beio->sg_segs[i].len;
1615 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr =
1616 uma_zalloc(be_lun->lun_zone, M_WAITOK);
1617 }
1618
1619 beio->num_segs++;
1620 len_left -= beio->sg_segs[i].len;
1621 }
1622 if (bptrlen->len < lbalen->len)
1623 beio->beio_cont = ctl_be_block_next;
1624 io->scsiio.be_move_done = ctl_be_block_move_done;
1625 /* For compare we have separate S/G lists for read and datamove. */
1626 if (lbalen->flags & CTL_LLF_COMPARE)
1627 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1628 else
1629 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1630 io->scsiio.kern_data_len = beio->io_len;
1631 io->scsiio.kern_data_resid = 0;
1632 io->scsiio.kern_sg_entries = beio->num_segs;
1633 io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
1634
1635 /*
1636 * For the read case, we need to read the data into our buffers and
1637 * then we can send it back to the user. For the write case, we
1638 * need to get the data from the user first.
1639 */
1640 if (beio->bio_cmd == BIO_READ) {
1641 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0);
1642 be_lun->dispatch(be_lun, beio);
1643 } else {
1644 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0);
1645#ifdef CTL_TIME_IO
1646 getbinuptime(&io->io_hdr.dma_start_bt);
1647#endif
1648 ctl_datamove(io);
1649 }
1650}
1651
1652static void
1653ctl_be_block_worker(void *context, int pending)
1654{
1655 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context;
1656 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1657 union ctl_io *io;
1658 struct ctl_be_block_io *beio;
1659
1660 DPRINTF("entered\n");
1661 /*
1662 * Fetch and process I/Os from all queues. If we detect LUN
1663 * CTL_LUN_FLAG_OFFLINE status here -- it is result of a race,
1664 * so make response maximally opaque to not confuse initiator.
1665 */
1666 for (;;) {
1667 mtx_lock(&be_lun->queue_lock);
1668 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1669 if (io != NULL) {
1670 DPRINTF("datamove queue\n");
1671 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1672 ctl_io_hdr, links);
1673 mtx_unlock(&be_lun->queue_lock);
1674 beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1675 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1676 ctl_set_busy(&io->scsiio);
1677 ctl_complete_beio(beio);
1678 return;
1679 }
1680 be_lun->dispatch(be_lun, beio);
1681 continue;
1682 }
1683 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1684 if (io != NULL) {
1685 DPRINTF("config write queue\n");
1686 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1687 ctl_io_hdr, links);
1688 mtx_unlock(&be_lun->queue_lock);
1689 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1690 ctl_set_busy(&io->scsiio);
1691 ctl_config_write_done(io);
1692 return;
1693 }
1694 ctl_be_block_cw_dispatch(be_lun, io);
1695 continue;
1696 }
1697 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
1698 if (io != NULL) {
1699 DPRINTF("config read queue\n");
1700 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr,
1701 ctl_io_hdr, links);
1702 mtx_unlock(&be_lun->queue_lock);
1703 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1704 ctl_set_busy(&io->scsiio);
1705 ctl_config_read_done(io);
1706 return;
1707 }
1708 ctl_be_block_cr_dispatch(be_lun, io);
1709 continue;
1710 }
1711 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1712 if (io != NULL) {
1713 DPRINTF("input queue\n");
1714 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1715 ctl_io_hdr, links);
1716 mtx_unlock(&be_lun->queue_lock);
1717 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) {
1718 ctl_set_busy(&io->scsiio);
1719 ctl_data_submit_done(io);
1720 return;
1721 }
1722 ctl_be_block_dispatch(be_lun, io);
1723 continue;
1724 }
1725
1726 /*
1727 * If we get here, there is no work left in the queues, so
1728 * just break out and let the task queue go to sleep.
1729 */
1730 mtx_unlock(&be_lun->queue_lock);
1731 break;
1732 }
1733}
1734
1735/*
1736 * Entry point from CTL to the backend for I/O. We queue everything to a
1737 * work thread, so this just puts the I/O on a queue and wakes up the
1738 * thread.
1739 */
1740static int
1741ctl_be_block_submit(union ctl_io *io)
1742{
1743 struct ctl_be_block_lun *be_lun;
1744 struct ctl_be_lun *cbe_lun;
1745
1746 DPRINTF("entered\n");
1747
1748 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
1749 CTL_PRIV_BACKEND_LUN].ptr;
1750 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun;
1751
1752 /*
1753 * Make sure we only get SCSI I/O.
1754 */
1755 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1756 "%#x) encountered", io->io_hdr.io_type));
1757
1758 PRIV(io)->len = 0;
1759
1760 mtx_lock(&be_lun->queue_lock);
1761 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1762 mtx_unlock(&be_lun->queue_lock);
1763 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1764
1765 return (CTL_RETVAL_COMPLETE);
1766}
1767
1768static int
1769ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1770 int flag, struct thread *td)
1771{
1772 struct ctl_be_block_softc *softc;
1773 int error;
1774
1775 softc = &backend_block_softc;
1776
1777 error = 0;
1778
1779 switch (cmd) {
1780 case CTL_LUN_REQ: {
1781 struct ctl_lun_req *lun_req;
1782
1783 lun_req = (struct ctl_lun_req *)addr;
1784
1785 switch (lun_req->reqtype) {
1786 case CTL_LUNREQ_CREATE:
1787 error = ctl_be_block_create(softc, lun_req);
1788 break;
1789 case CTL_LUNREQ_RM:
1790 error = ctl_be_block_rm(softc, lun_req);
1791 break;
1792 case CTL_LUNREQ_MODIFY:
1793 error = ctl_be_block_modify(softc, lun_req);
1794 break;
1795 default:
1796 lun_req->status = CTL_LUN_ERROR;
1797 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1798 "invalid LUN request type %d",
1799 lun_req->reqtype);
1800 break;
1801 }
1802 break;
1803 }
1804 default:
1805 error = ENOTTY;
1806 break;
1807 }
1808
1809 return (error);
1810}
1811
1812static int
1813ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1814{
1815 struct ctl_be_lun *cbe_lun;
1816 struct ctl_be_block_filedata *file_data;
1817 struct ctl_lun_create_params *params;
1818 char *value;
1819 struct vattr vattr;
1820 off_t ps, pss, po, pos, us, uss, uo, uos;
1821 int error;
1822
1823 cbe_lun = &be_lun->cbe_lun;
1824 file_data = &be_lun->backend.file;
1825 params = &be_lun->params;
1826
1827 be_lun->dev_type = CTL_BE_BLOCK_FILE;
1828 be_lun->dispatch = ctl_be_block_dispatch_file;
1829 be_lun->lun_flush = ctl_be_block_flush_file;
1830 be_lun->get_lba_status = ctl_be_block_gls_file;
1831 be_lun->getattr = ctl_be_block_getattr_file;
1832 be_lun->unmap = NULL;
1833 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
1834
1835 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1836 if (error != 0) {
1837 snprintf(req->error_str, sizeof(req->error_str),
1838 "error calling VOP_GETATTR() for file %s",
1839 be_lun->dev_path);
1840 return (error);
1841 }
1842
1843 /*
1844 * Verify that we have the ability to upgrade to exclusive
1845 * access on this file so we can trap errors at open instead
1846 * of reporting them during first access.
1847 */
1848 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) {
1849 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY);
1850 if (be_lun->vn->v_iflag & VI_DOOMED) {
1851 error = EBADF;
1852 snprintf(req->error_str, sizeof(req->error_str),
1853 "error locking file %s", be_lun->dev_path);
1854 return (error);
1855 }
1856 }
1857
1858 file_data->cred = crhold(curthread->td_ucred);
1859 if (params->lun_size_bytes != 0)
1860 be_lun->size_bytes = params->lun_size_bytes;
1861 else
1862 be_lun->size_bytes = vattr.va_size;
1863
1864 /*
1865 * For files we can use any logical block size. Prefer 512 bytes
1866 * for compatibility reasons. If file's vattr.va_blocksize
1867 * (preferred I/O block size) is bigger and multiple to chosen
1868 * logical block size -- report it as physical block size.
1869 */
1870 if (params->blocksize_bytes != 0)
1871 cbe_lun->blocksize = params->blocksize_bytes;
|
2280 else
2281 cbe_lun->blocksize = 512;
2282 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2283 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2284 0 : (be_lun->size_blocks - 1);
2285
2286 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2287 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2288 retval = ctl_be_block_open(softc, be_lun, req);
2289 if (retval != 0) {
2290 retval = 0;
2291 req->status = CTL_LUN_WARNING;
2292 }
2293 }
2294 num_threads = cbb_num_threads;
2295 } else {
2296 num_threads = 1;
2297 }
2298
2299 value = ctl_get_opt(&cbe_lun->options, "num_threads");
2300 if (value != NULL) {
2301 tmp_num_threads = strtol(value, NULL, 0);
2302
2303 /*
2304 * We don't let the user specify less than one
2305 * thread, but hope he's clueful enough not to
2306 * specify 1000 threads.
2307 */
2308 if (tmp_num_threads < 1) {
2309 snprintf(req->error_str, sizeof(req->error_str),
2310 "invalid number of threads %s",
2311 num_thread_str);
2312 goto bailout_error;
2313 }
2314 num_threads = tmp_num_threads;
2315 }
2316
2317 if (be_lun->vn == NULL)
2318 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE;
2319 /* Tell the user the blocksize we ended up using */
2320 params->lun_size_bytes = be_lun->size_bytes;
2321 params->blocksize_bytes = cbe_lun->blocksize;
2322 if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2323 cbe_lun->req_lun_id = params->req_lun_id;
2324 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
2325 } else
2326 cbe_lun->req_lun_id = 0;
2327
2328 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown;
2329 cbe_lun->lun_config_status = ctl_be_block_lun_config_status;
2330 cbe_lun->be = &ctl_be_block_driver;
2331
2332 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2333 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d",
2334 softc->num_luns);
2335 strncpy((char *)cbe_lun->serial_num, tmpstr,
2336 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
2337
2338 /* Tell the user what we used for a serial number */
2339 strncpy((char *)params->serial_num, tmpstr,
2340 MIN(sizeof(params->serial_num), sizeof(tmpstr)));
2341 } else {
2342 strncpy((char *)cbe_lun->serial_num, params->serial_num,
2343 MIN(sizeof(cbe_lun->serial_num),
2344 sizeof(params->serial_num)));
2345 }
2346 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2347 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns);
2348 strncpy((char *)cbe_lun->device_id, tmpstr,
2349 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
2350
2351 /* Tell the user what we used for a device ID */
2352 strncpy((char *)params->device_id, tmpstr,
2353 MIN(sizeof(params->device_id), sizeof(tmpstr)));
2354 } else {
2355 strncpy((char *)cbe_lun->device_id, params->device_id,
2356 MIN(sizeof(cbe_lun->device_id),
2357 sizeof(params->device_id)));
2358 }
2359
2360 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2361
2362 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK,
2363 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2364
2365 if (be_lun->io_taskqueue == NULL) {
2366 snprintf(req->error_str, sizeof(req->error_str),
2367 "unable to create taskqueue");
2368 goto bailout_error;
2369 }
2370
2371 /*
2372 * Note that we start the same number of threads by default for
2373 * both the file case and the block device case. For the file
2374 * case, we need multiple threads to allow concurrency, because the
2375 * vnode interface is designed to be a blocking interface. For the
2376 * block device case, ZFS zvols at least will block the caller's
2377 * context in many instances, and so we need multiple threads to
2378 * overcome that problem. Other block devices don't need as many
2379 * threads, but they shouldn't cause too many problems.
2380 *
2381 * If the user wants to just have a single thread for a block
2382 * device, he can specify that when the LUN is created, or change
2383 * the tunable/sysctl to alter the default number of threads.
2384 */
2385 retval = taskqueue_start_threads(&be_lun->io_taskqueue,
2386 /*num threads*/num_threads,
2387 /*priority*/PWAIT,
2388 /*thread name*/
2389 "%s taskq", be_lun->lunname);
2390
2391 if (retval != 0)
2392 goto bailout_error;
2393
2394 be_lun->num_threads = num_threads;
2395
2396 mtx_lock(&softc->lock);
2397 softc->num_luns++;
2398 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links);
2399
2400 mtx_unlock(&softc->lock);
2401
2402 retval = ctl_add_lun(&be_lun->cbe_lun);
2403 if (retval != 0) {
2404 mtx_lock(&softc->lock);
2405 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2406 links);
2407 softc->num_luns--;
2408 mtx_unlock(&softc->lock);
2409 snprintf(req->error_str, sizeof(req->error_str),
2410 "ctl_add_lun() returned error %d, see dmesg for "
2411 "details", retval);
2412 retval = 0;
2413 goto bailout_error;
2414 }
2415
2416 mtx_lock(&softc->lock);
2417
2418 /*
2419 * Tell the config_status routine that we're waiting so it won't
2420 * clean up the LUN in the event of an error.
2421 */
2422 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2423
2424 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2425 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2426 if (retval == EINTR)
2427 break;
2428 }
2429 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2430
2431 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) {
2432 snprintf(req->error_str, sizeof(req->error_str),
2433 "LUN configuration error, see dmesg for details");
2434 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2435 links);
2436 softc->num_luns--;
2437 mtx_unlock(&softc->lock);
2438 goto bailout_error;
2439 } else {
2440 params->req_lun_id = cbe_lun->lun_id;
2441 }
2442
2443 mtx_unlock(&softc->lock);
2444
2445 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id,
2446 cbe_lun->blocksize,
2447 DEVSTAT_ALL_SUPPORTED,
2448 cbe_lun->lun_type
2449 | DEVSTAT_TYPE_IF_OTHER,
2450 DEVSTAT_PRIORITY_OTHER);
2451
2452 return (retval);
2453
2454bailout_error:
2455 req->status = CTL_LUN_ERROR;
2456
2457 if (be_lun->io_taskqueue != NULL)
2458 taskqueue_free(be_lun->io_taskqueue);
2459 ctl_be_block_close(be_lun);
2460 if (be_lun->dev_path != NULL)
2461 free(be_lun->dev_path, M_CTLBLK);
2462 if (be_lun->lun_zone != NULL)
2463 uma_zdestroy(be_lun->lun_zone);
2464 ctl_free_opts(&cbe_lun->options);
2465 mtx_destroy(&be_lun->queue_lock);
2466 mtx_destroy(&be_lun->io_lock);
2467 free(be_lun, M_CTLBLK);
2468
2469 return (retval);
2470}
2471
2472static int
2473ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2474{
2475 struct ctl_lun_rm_params *params;
2476 struct ctl_be_block_lun *be_lun;
2477 struct ctl_be_lun *cbe_lun;
2478 int retval;
2479
2480 params = &req->reqdata.rm;
2481
2482 mtx_lock(&softc->lock);
2483 STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2484 if (be_lun->cbe_lun.lun_id == params->lun_id)
2485 break;
2486 }
2487 mtx_unlock(&softc->lock);
2488 if (be_lun == NULL) {
2489 snprintf(req->error_str, sizeof(req->error_str),
2490 "LUN %u is not managed by the block backend",
2491 params->lun_id);
2492 goto bailout_error;
2493 }
2494 cbe_lun = &be_lun->cbe_lun;
2495
2496 retval = ctl_disable_lun(cbe_lun);
2497 if (retval != 0) {
2498 snprintf(req->error_str, sizeof(req->error_str),
2499 "error %d returned from ctl_disable_lun() for "
2500 "LUN %d", retval, params->lun_id);
2501 goto bailout_error;
2502 }
2503
2504 if (be_lun->vn != NULL) {
2505 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE;
2506 ctl_lun_offline(cbe_lun);
2507 taskqueue_drain_all(be_lun->io_taskqueue);
2508 ctl_be_block_close(be_lun);
2509 }
2510
2511 retval = ctl_invalidate_lun(cbe_lun);
2512 if (retval != 0) {
2513 snprintf(req->error_str, sizeof(req->error_str),
2514 "error %d returned from ctl_invalidate_lun() for "
2515 "LUN %d", retval, params->lun_id);
2516 goto bailout_error;
2517 }
2518
2519 mtx_lock(&softc->lock);
2520 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2521 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2522 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2523 if (retval == EINTR)
2524 break;
2525 }
2526 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2527
2528 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2529 snprintf(req->error_str, sizeof(req->error_str),
2530 "interrupted waiting for LUN to be freed");
2531 mtx_unlock(&softc->lock);
2532 goto bailout_error;
2533 }
2534
2535 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links);
2536
2537 softc->num_luns--;
2538 mtx_unlock(&softc->lock);
2539
2540 taskqueue_drain_all(be_lun->io_taskqueue);
2541 taskqueue_free(be_lun->io_taskqueue);
2542
2543 if (be_lun->disk_stats != NULL)
2544 devstat_remove_entry(be_lun->disk_stats);
2545
2546 uma_zdestroy(be_lun->lun_zone);
2547
2548 ctl_free_opts(&cbe_lun->options);
2549 free(be_lun->dev_path, M_CTLBLK);
2550 mtx_destroy(&be_lun->queue_lock);
2551 mtx_destroy(&be_lun->io_lock);
2552 free(be_lun, M_CTLBLK);
2553
2554 req->status = CTL_LUN_OK;
2555 return (0);
2556
2557bailout_error:
2558 req->status = CTL_LUN_ERROR;
2559 return (0);
2560}
2561
2562static int
2563ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2564{
2565 struct ctl_lun_modify_params *params;
2566 struct ctl_be_block_lun *be_lun;
2567 struct ctl_be_lun *cbe_lun;
2568 char *value;
2569 uint64_t oldsize;
2570 int error, wasprim;
2571
2572 params = &req->reqdata.modify;
2573
2574 mtx_lock(&softc->lock);
2575 STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2576 if (be_lun->cbe_lun.lun_id == params->lun_id)
2577 break;
2578 }
2579 mtx_unlock(&softc->lock);
2580 if (be_lun == NULL) {
2581 snprintf(req->error_str, sizeof(req->error_str),
2582 "LUN %u is not managed by the block backend",
2583 params->lun_id);
2584 goto bailout_error;
2585 }
2586 cbe_lun = &be_lun->cbe_lun;
2587
2588 if (params->lun_size_bytes != 0)
2589 be_lun->params.lun_size_bytes = params->lun_size_bytes;
2590 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args);
2591
2592 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
2593 value = ctl_get_opt(&cbe_lun->options, "ha_role");
2594 if (value != NULL) {
2595 if (strcmp(value, "primary") == 0)
2596 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2597 else
2598 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2599 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2600 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2601 else
2602 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2603 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
2604 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
2605 ctl_lun_primary(cbe_lun);
2606 else
2607 ctl_lun_secondary(cbe_lun);
2608 }
2609
2610 oldsize = be_lun->size_blocks;
2611 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2612 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2613 if (be_lun->vn == NULL)
2614 error = ctl_be_block_open(softc, be_lun, req);
2615 else if (vn_isdisk(be_lun->vn, &error))
2616 error = ctl_be_block_open_dev(be_lun, req);
2617 else if (be_lun->vn->v_type == VREG)
2618 error = ctl_be_block_open_file(be_lun, req);
2619 else
2620 error = EINVAL;
2621 if ((cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) &&
2622 be_lun->vn != NULL) {
2623 cbe_lun->flags &= ~CTL_LUN_FLAG_OFFLINE;
2624 ctl_lun_online(cbe_lun);
2625 }
2626 } else {
2627 if (be_lun->vn != NULL) {
2628 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE;
2629 ctl_lun_offline(cbe_lun);
2630 taskqueue_drain_all(be_lun->io_taskqueue);
2631 error = ctl_be_block_close(be_lun);
2632 } else
2633 error = 0;
2634 }
2635 if (be_lun->size_blocks != oldsize)
2636 ctl_lun_capacity_changed(cbe_lun);
2637
2638 /* Tell the user the exact size we ended up using */
2639 params->lun_size_bytes = be_lun->size_bytes;
2640
2641 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2642 return (0);
2643
2644bailout_error:
2645 req->status = CTL_LUN_ERROR;
2646 return (0);
2647}
2648
2649static void
2650ctl_be_block_lun_shutdown(void *be_lun)
2651{
2652 struct ctl_be_block_lun *lun;
2653 struct ctl_be_block_softc *softc;
2654
2655 lun = (struct ctl_be_block_lun *)be_lun;
2656 softc = lun->softc;
2657
2658 mtx_lock(&softc->lock);
2659 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2660 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2661 wakeup(lun);
2662 mtx_unlock(&softc->lock);
2663}
2664
2665static void
2666ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status)
2667{
2668 struct ctl_be_block_lun *lun;
2669 struct ctl_be_block_softc *softc;
2670
2671 lun = (struct ctl_be_block_lun *)be_lun;
2672 softc = lun->softc;
2673
2674 if (status == CTL_LUN_CONFIG_OK) {
2675 mtx_lock(&softc->lock);
2676 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2677 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2678 wakeup(lun);
2679 mtx_unlock(&softc->lock);
2680
2681 /*
2682 * We successfully added the LUN, attempt to enable it.
2683 */
2684 if (ctl_enable_lun(&lun->cbe_lun) != 0) {
2685 printf("%s: ctl_enable_lun() failed!\n", __func__);
2686 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) {
2687 printf("%s: ctl_invalidate_lun() failed!\n",
2688 __func__);
2689 }
2690 }
2691
2692 return;
2693 }
2694
2695
2696 mtx_lock(&softc->lock);
2697 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2698 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR;
2699 wakeup(lun);
2700 mtx_unlock(&softc->lock);
2701}
2702
2703
2704static int
2705ctl_be_block_config_write(union ctl_io *io)
2706{
2707 struct ctl_be_block_lun *be_lun;
2708 struct ctl_be_lun *cbe_lun;
2709 int retval;
2710
2711 DPRINTF("entered\n");
2712
2713 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
2714 CTL_PRIV_BACKEND_LUN].ptr;
2715 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun;
2716
2717 retval = 0;
2718 switch (io->scsiio.cdb[0]) {
2719 case SYNCHRONIZE_CACHE:
2720 case SYNCHRONIZE_CACHE_16:
2721 case WRITE_SAME_10:
2722 case WRITE_SAME_16:
2723 case UNMAP:
2724 /*
2725 * The upper level CTL code will filter out any CDBs with
2726 * the immediate bit set and return the proper error.
2727 *
2728 * We don't really need to worry about what LBA range the
2729 * user asked to be synced out. When they issue a sync
2730 * cache command, we'll sync out the whole thing.
2731 */
2732 mtx_lock(&be_lun->queue_lock);
2733 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2734 links);
2735 mtx_unlock(&be_lun->queue_lock);
2736 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2737 break;
2738 case START_STOP_UNIT: {
2739 struct scsi_start_stop_unit *cdb;
2740
2741 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2742
2743 if (cdb->how & SSS_START)
2744 retval = ctl_start_lun(cbe_lun);
2745 else
2746 retval = ctl_stop_lun(cbe_lun);
2747
2748 /*
2749 * In general, the above routines should not fail. They
2750 * just set state for the LUN. So we've got something
2751 * pretty wrong here if we can't start or stop the LUN.
2752 */
2753 if (retval != 0) {
2754 ctl_set_internal_failure(&io->scsiio,
2755 /*sks_valid*/ 1,
2756 /*retry_count*/ 0xf051);
2757 retval = CTL_RETVAL_COMPLETE;
2758 } else {
2759 ctl_set_success(&io->scsiio);
2760 }
2761 ctl_config_write_done(io);
2762 break;
2763 }
| 2290 else
2291 cbe_lun->blocksize = 512;
2292 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2293 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2294 0 : (be_lun->size_blocks - 1);
2295
2296 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2297 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2298 retval = ctl_be_block_open(softc, be_lun, req);
2299 if (retval != 0) {
2300 retval = 0;
2301 req->status = CTL_LUN_WARNING;
2302 }
2303 }
2304 num_threads = cbb_num_threads;
2305 } else {
2306 num_threads = 1;
2307 }
2308
2309 value = ctl_get_opt(&cbe_lun->options, "num_threads");
2310 if (value != NULL) {
2311 tmp_num_threads = strtol(value, NULL, 0);
2312
2313 /*
2314 * We don't let the user specify less than one
2315 * thread, but hope he's clueful enough not to
2316 * specify 1000 threads.
2317 */
2318 if (tmp_num_threads < 1) {
2319 snprintf(req->error_str, sizeof(req->error_str),
2320 "invalid number of threads %s",
2321 num_thread_str);
2322 goto bailout_error;
2323 }
2324 num_threads = tmp_num_threads;
2325 }
2326
2327 if (be_lun->vn == NULL)
2328 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE;
2329 /* Tell the user the blocksize we ended up using */
2330 params->lun_size_bytes = be_lun->size_bytes;
2331 params->blocksize_bytes = cbe_lun->blocksize;
2332 if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2333 cbe_lun->req_lun_id = params->req_lun_id;
2334 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
2335 } else
2336 cbe_lun->req_lun_id = 0;
2337
2338 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown;
2339 cbe_lun->lun_config_status = ctl_be_block_lun_config_status;
2340 cbe_lun->be = &ctl_be_block_driver;
2341
2342 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2343 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d",
2344 softc->num_luns);
2345 strncpy((char *)cbe_lun->serial_num, tmpstr,
2346 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
2347
2348 /* Tell the user what we used for a serial number */
2349 strncpy((char *)params->serial_num, tmpstr,
2350 MIN(sizeof(params->serial_num), sizeof(tmpstr)));
2351 } else {
2352 strncpy((char *)cbe_lun->serial_num, params->serial_num,
2353 MIN(sizeof(cbe_lun->serial_num),
2354 sizeof(params->serial_num)));
2355 }
2356 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2357 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns);
2358 strncpy((char *)cbe_lun->device_id, tmpstr,
2359 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
2360
2361 /* Tell the user what we used for a device ID */
2362 strncpy((char *)params->device_id, tmpstr,
2363 MIN(sizeof(params->device_id), sizeof(tmpstr)));
2364 } else {
2365 strncpy((char *)cbe_lun->device_id, params->device_id,
2366 MIN(sizeof(cbe_lun->device_id),
2367 sizeof(params->device_id)));
2368 }
2369
2370 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2371
2372 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK,
2373 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2374
2375 if (be_lun->io_taskqueue == NULL) {
2376 snprintf(req->error_str, sizeof(req->error_str),
2377 "unable to create taskqueue");
2378 goto bailout_error;
2379 }
2380
2381 /*
2382 * Note that we start the same number of threads by default for
2383 * both the file case and the block device case. For the file
2384 * case, we need multiple threads to allow concurrency, because the
2385 * vnode interface is designed to be a blocking interface. For the
2386 * block device case, ZFS zvols at least will block the caller's
2387 * context in many instances, and so we need multiple threads to
2388 * overcome that problem. Other block devices don't need as many
2389 * threads, but they shouldn't cause too many problems.
2390 *
2391 * If the user wants to just have a single thread for a block
2392 * device, he can specify that when the LUN is created, or change
2393 * the tunable/sysctl to alter the default number of threads.
2394 */
2395 retval = taskqueue_start_threads(&be_lun->io_taskqueue,
2396 /*num threads*/num_threads,
2397 /*priority*/PWAIT,
2398 /*thread name*/
2399 "%s taskq", be_lun->lunname);
2400
2401 if (retval != 0)
2402 goto bailout_error;
2403
2404 be_lun->num_threads = num_threads;
2405
2406 mtx_lock(&softc->lock);
2407 softc->num_luns++;
2408 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links);
2409
2410 mtx_unlock(&softc->lock);
2411
2412 retval = ctl_add_lun(&be_lun->cbe_lun);
2413 if (retval != 0) {
2414 mtx_lock(&softc->lock);
2415 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2416 links);
2417 softc->num_luns--;
2418 mtx_unlock(&softc->lock);
2419 snprintf(req->error_str, sizeof(req->error_str),
2420 "ctl_add_lun() returned error %d, see dmesg for "
2421 "details", retval);
2422 retval = 0;
2423 goto bailout_error;
2424 }
2425
2426 mtx_lock(&softc->lock);
2427
2428 /*
2429 * Tell the config_status routine that we're waiting so it won't
2430 * clean up the LUN in the event of an error.
2431 */
2432 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2433
2434 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2435 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2436 if (retval == EINTR)
2437 break;
2438 }
2439 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2440
2441 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) {
2442 snprintf(req->error_str, sizeof(req->error_str),
2443 "LUN configuration error, see dmesg for details");
2444 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2445 links);
2446 softc->num_luns--;
2447 mtx_unlock(&softc->lock);
2448 goto bailout_error;
2449 } else {
2450 params->req_lun_id = cbe_lun->lun_id;
2451 }
2452
2453 mtx_unlock(&softc->lock);
2454
2455 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id,
2456 cbe_lun->blocksize,
2457 DEVSTAT_ALL_SUPPORTED,
2458 cbe_lun->lun_type
2459 | DEVSTAT_TYPE_IF_OTHER,
2460 DEVSTAT_PRIORITY_OTHER);
2461
2462 return (retval);
2463
2464bailout_error:
2465 req->status = CTL_LUN_ERROR;
2466
2467 if (be_lun->io_taskqueue != NULL)
2468 taskqueue_free(be_lun->io_taskqueue);
2469 ctl_be_block_close(be_lun);
2470 if (be_lun->dev_path != NULL)
2471 free(be_lun->dev_path, M_CTLBLK);
2472 if (be_lun->lun_zone != NULL)
2473 uma_zdestroy(be_lun->lun_zone);
2474 ctl_free_opts(&cbe_lun->options);
2475 mtx_destroy(&be_lun->queue_lock);
2476 mtx_destroy(&be_lun->io_lock);
2477 free(be_lun, M_CTLBLK);
2478
2479 return (retval);
2480}
2481
2482static int
2483ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2484{
2485 struct ctl_lun_rm_params *params;
2486 struct ctl_be_block_lun *be_lun;
2487 struct ctl_be_lun *cbe_lun;
2488 int retval;
2489
2490 params = &req->reqdata.rm;
2491
2492 mtx_lock(&softc->lock);
2493 STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2494 if (be_lun->cbe_lun.lun_id == params->lun_id)
2495 break;
2496 }
2497 mtx_unlock(&softc->lock);
2498 if (be_lun == NULL) {
2499 snprintf(req->error_str, sizeof(req->error_str),
2500 "LUN %u is not managed by the block backend",
2501 params->lun_id);
2502 goto bailout_error;
2503 }
2504 cbe_lun = &be_lun->cbe_lun;
2505
2506 retval = ctl_disable_lun(cbe_lun);
2507 if (retval != 0) {
2508 snprintf(req->error_str, sizeof(req->error_str),
2509 "error %d returned from ctl_disable_lun() for "
2510 "LUN %d", retval, params->lun_id);
2511 goto bailout_error;
2512 }
2513
2514 if (be_lun->vn != NULL) {
2515 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE;
2516 ctl_lun_offline(cbe_lun);
2517 taskqueue_drain_all(be_lun->io_taskqueue);
2518 ctl_be_block_close(be_lun);
2519 }
2520
2521 retval = ctl_invalidate_lun(cbe_lun);
2522 if (retval != 0) {
2523 snprintf(req->error_str, sizeof(req->error_str),
2524 "error %d returned from ctl_invalidate_lun() for "
2525 "LUN %d", retval, params->lun_id);
2526 goto bailout_error;
2527 }
2528
2529 mtx_lock(&softc->lock);
2530 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2531 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2532 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2533 if (retval == EINTR)
2534 break;
2535 }
2536 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2537
2538 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2539 snprintf(req->error_str, sizeof(req->error_str),
2540 "interrupted waiting for LUN to be freed");
2541 mtx_unlock(&softc->lock);
2542 goto bailout_error;
2543 }
2544
2545 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links);
2546
2547 softc->num_luns--;
2548 mtx_unlock(&softc->lock);
2549
2550 taskqueue_drain_all(be_lun->io_taskqueue);
2551 taskqueue_free(be_lun->io_taskqueue);
2552
2553 if (be_lun->disk_stats != NULL)
2554 devstat_remove_entry(be_lun->disk_stats);
2555
2556 uma_zdestroy(be_lun->lun_zone);
2557
2558 ctl_free_opts(&cbe_lun->options);
2559 free(be_lun->dev_path, M_CTLBLK);
2560 mtx_destroy(&be_lun->queue_lock);
2561 mtx_destroy(&be_lun->io_lock);
2562 free(be_lun, M_CTLBLK);
2563
2564 req->status = CTL_LUN_OK;
2565 return (0);
2566
2567bailout_error:
2568 req->status = CTL_LUN_ERROR;
2569 return (0);
2570}
2571
2572static int
2573ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2574{
2575 struct ctl_lun_modify_params *params;
2576 struct ctl_be_block_lun *be_lun;
2577 struct ctl_be_lun *cbe_lun;
2578 char *value;
2579 uint64_t oldsize;
2580 int error, wasprim;
2581
2582 params = &req->reqdata.modify;
2583
2584 mtx_lock(&softc->lock);
2585 STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2586 if (be_lun->cbe_lun.lun_id == params->lun_id)
2587 break;
2588 }
2589 mtx_unlock(&softc->lock);
2590 if (be_lun == NULL) {
2591 snprintf(req->error_str, sizeof(req->error_str),
2592 "LUN %u is not managed by the block backend",
2593 params->lun_id);
2594 goto bailout_error;
2595 }
2596 cbe_lun = &be_lun->cbe_lun;
2597
2598 if (params->lun_size_bytes != 0)
2599 be_lun->params.lun_size_bytes = params->lun_size_bytes;
2600 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args);
2601
2602 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
2603 value = ctl_get_opt(&cbe_lun->options, "ha_role");
2604 if (value != NULL) {
2605 if (strcmp(value, "primary") == 0)
2606 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2607 else
2608 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2609 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2610 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2611 else
2612 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2613 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
2614 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
2615 ctl_lun_primary(cbe_lun);
2616 else
2617 ctl_lun_secondary(cbe_lun);
2618 }
2619
2620 oldsize = be_lun->size_blocks;
2621 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2622 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2623 if (be_lun->vn == NULL)
2624 error = ctl_be_block_open(softc, be_lun, req);
2625 else if (vn_isdisk(be_lun->vn, &error))
2626 error = ctl_be_block_open_dev(be_lun, req);
2627 else if (be_lun->vn->v_type == VREG)
2628 error = ctl_be_block_open_file(be_lun, req);
2629 else
2630 error = EINVAL;
2631 if ((cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) &&
2632 be_lun->vn != NULL) {
2633 cbe_lun->flags &= ~CTL_LUN_FLAG_OFFLINE;
2634 ctl_lun_online(cbe_lun);
2635 }
2636 } else {
2637 if (be_lun->vn != NULL) {
2638 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE;
2639 ctl_lun_offline(cbe_lun);
2640 taskqueue_drain_all(be_lun->io_taskqueue);
2641 error = ctl_be_block_close(be_lun);
2642 } else
2643 error = 0;
2644 }
2645 if (be_lun->size_blocks != oldsize)
2646 ctl_lun_capacity_changed(cbe_lun);
2647
2648 /* Tell the user the exact size we ended up using */
2649 params->lun_size_bytes = be_lun->size_bytes;
2650
2651 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2652 return (0);
2653
2654bailout_error:
2655 req->status = CTL_LUN_ERROR;
2656 return (0);
2657}
2658
2659static void
2660ctl_be_block_lun_shutdown(void *be_lun)
2661{
2662 struct ctl_be_block_lun *lun;
2663 struct ctl_be_block_softc *softc;
2664
2665 lun = (struct ctl_be_block_lun *)be_lun;
2666 softc = lun->softc;
2667
2668 mtx_lock(&softc->lock);
2669 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2670 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2671 wakeup(lun);
2672 mtx_unlock(&softc->lock);
2673}
2674
2675static void
2676ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status)
2677{
2678 struct ctl_be_block_lun *lun;
2679 struct ctl_be_block_softc *softc;
2680
2681 lun = (struct ctl_be_block_lun *)be_lun;
2682 softc = lun->softc;
2683
2684 if (status == CTL_LUN_CONFIG_OK) {
2685 mtx_lock(&softc->lock);
2686 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2687 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2688 wakeup(lun);
2689 mtx_unlock(&softc->lock);
2690
2691 /*
2692 * We successfully added the LUN, attempt to enable it.
2693 */
2694 if (ctl_enable_lun(&lun->cbe_lun) != 0) {
2695 printf("%s: ctl_enable_lun() failed!\n", __func__);
2696 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) {
2697 printf("%s: ctl_invalidate_lun() failed!\n",
2698 __func__);
2699 }
2700 }
2701
2702 return;
2703 }
2704
2705
2706 mtx_lock(&softc->lock);
2707 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2708 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR;
2709 wakeup(lun);
2710 mtx_unlock(&softc->lock);
2711}
2712
2713
2714static int
2715ctl_be_block_config_write(union ctl_io *io)
2716{
2717 struct ctl_be_block_lun *be_lun;
2718 struct ctl_be_lun *cbe_lun;
2719 int retval;
2720
2721 DPRINTF("entered\n");
2722
2723 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
2724 CTL_PRIV_BACKEND_LUN].ptr;
2725 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun;
2726
2727 retval = 0;
2728 switch (io->scsiio.cdb[0]) {
2729 case SYNCHRONIZE_CACHE:
2730 case SYNCHRONIZE_CACHE_16:
2731 case WRITE_SAME_10:
2732 case WRITE_SAME_16:
2733 case UNMAP:
2734 /*
2735 * The upper level CTL code will filter out any CDBs with
2736 * the immediate bit set and return the proper error.
2737 *
2738 * We don't really need to worry about what LBA range the
2739 * user asked to be synced out. When they issue a sync
2740 * cache command, we'll sync out the whole thing.
2741 */
2742 mtx_lock(&be_lun->queue_lock);
2743 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2744 links);
2745 mtx_unlock(&be_lun->queue_lock);
2746 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2747 break;
2748 case START_STOP_UNIT: {
2749 struct scsi_start_stop_unit *cdb;
2750
2751 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2752
2753 if (cdb->how & SSS_START)
2754 retval = ctl_start_lun(cbe_lun);
2755 else
2756 retval = ctl_stop_lun(cbe_lun);
2757
2758 /*
2759 * In general, the above routines should not fail. They
2760 * just set state for the LUN. So we've got something
2761 * pretty wrong here if we can't start or stop the LUN.
2762 */
2763 if (retval != 0) {
2764 ctl_set_internal_failure(&io->scsiio,
2765 /*sks_valid*/ 1,
2766 /*retry_count*/ 0xf051);
2767 retval = CTL_RETVAL_COMPLETE;
2768 } else {
2769 ctl_set_success(&io->scsiio);
2770 }
2771 ctl_config_write_done(io);
2772 break;
2773 }
|