ctl_backend_block.c revision 269226
1/*- 2 * Copyright (c) 2003 Silicon Graphics International Corp. 3 * Copyright (c) 2009-2011 Spectra Logic Corporation 4 * Copyright (c) 2012 The FreeBSD Foundation 5 * All rights reserved. 6 * 7 * Portions of this software were developed by Edward Tomasz Napierala 8 * under sponsorship from the FreeBSD Foundation. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions, and the following disclaimer, 15 * without modification. 16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17 * substantially similar to the "NO WARRANTY" disclaimer below 18 * ("Disclaimer") and any redistribution must be conditioned upon 19 * including a substantially similar Disclaimer requirement for further 20 * binary redistribution. 21 * 22 * NO WARRANTY 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGES. 34 * 35 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36 */ 37/* 38 * CAM Target Layer driver backend for block devices. 39 * 40 * Author: Ken Merry <ken@FreeBSD.org> 41 */ 42#include <sys/cdefs.h> 43__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl_backend_block.c 269226 2014-07-29 07:40:14Z mav $"); 44 45#include <opt_kdtrace.h> 46 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/kernel.h> 50#include <sys/types.h> 51#include <sys/kthread.h> 52#include <sys/bio.h> 53#include <sys/fcntl.h> 54#include <sys/limits.h> 55#include <sys/lock.h> 56#include <sys/mutex.h> 57#include <sys/condvar.h> 58#include <sys/malloc.h> 59#include <sys/conf.h> 60#include <sys/ioccom.h> 61#include <sys/queue.h> 62#include <sys/sbuf.h> 63#include <sys/endian.h> 64#include <sys/uio.h> 65#include <sys/buf.h> 66#include <sys/taskqueue.h> 67#include <sys/vnode.h> 68#include <sys/namei.h> 69#include <sys/mount.h> 70#include <sys/disk.h> 71#include <sys/fcntl.h> 72#include <sys/filedesc.h> 73#include <sys/proc.h> 74#include <sys/pcpu.h> 75#include <sys/module.h> 76#include <sys/sdt.h> 77#include <sys/devicestat.h> 78#include <sys/sysctl.h> 79 80#include <geom/geom.h> 81 82#include <cam/cam.h> 83#include <cam/scsi/scsi_all.h> 84#include <cam/scsi/scsi_da.h> 85#include <cam/ctl/ctl_io.h> 86#include <cam/ctl/ctl.h> 87#include <cam/ctl/ctl_backend.h> 88#include <cam/ctl/ctl_frontend_internal.h> 89#include <cam/ctl/ctl_ioctl.h> 90#include <cam/ctl/ctl_scsi_all.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_MULTI_THREAD = 0x08 122} ctl_be_block_lun_flags; 123 124typedef enum { 125 CTL_BE_BLOCK_NONE, 126 CTL_BE_BLOCK_DEV, 127 CTL_BE_BLOCK_FILE 128} ctl_be_block_type; 129 130struct ctl_be_block_devdata { 131 struct cdev *cdev; 132 struct cdevsw *csw; 133 int dev_ref; 134}; 135 136struct ctl_be_block_filedata { 137 struct ucred *cred; 138}; 139 140union ctl_be_block_bedata { 141 struct ctl_be_block_devdata dev; 142 struct ctl_be_block_filedata file; 143}; 144 145struct ctl_be_block_io; 146struct ctl_be_block_lun; 147 148typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 149 struct ctl_be_block_io *beio); 150 151/* 152 * Backend LUN structure. There is a 1:1 mapping between a block device 153 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 154 */ 155struct ctl_be_block_lun { 156 struct ctl_block_disk *disk; 157 char lunname[32]; 158 char *dev_path; 159 ctl_be_block_type dev_type; 160 struct vnode *vn; 161 union ctl_be_block_bedata backend; 162 cbb_dispatch_t dispatch; 163 cbb_dispatch_t lun_flush; 164 cbb_dispatch_t unmap; 165 uma_zone_t lun_zone; 166 uint64_t size_blocks; 167 uint64_t size_bytes; 168 uint32_t blocksize; 169 int blocksize_shift; 170 uint16_t pblockexp; 171 uint16_t pblockoff; 172 struct ctl_be_block_softc *softc; 173 struct devstat *disk_stats; 174 ctl_be_block_lun_flags flags; 175 STAILQ_ENTRY(ctl_be_block_lun) links; 176 struct ctl_be_lun ctl_be_lun; 177 struct taskqueue *io_taskqueue; 178 struct task io_task; 179 int num_threads; 180 STAILQ_HEAD(, ctl_io_hdr) input_queue; 181 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 182 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 183 struct mtx_padalign io_lock; 184 struct mtx_padalign queue_lock; 185}; 186 187/* 188 * Overall softc structure for the block backend module. 189 */ 190struct ctl_be_block_softc { 191 struct mtx lock; 192 int num_disks; 193 STAILQ_HEAD(, ctl_block_disk) disk_list; 194 int num_luns; 195 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 196}; 197 198static struct ctl_be_block_softc backend_block_softc; 199 200/* 201 * Per-I/O information. 202 */ 203struct ctl_be_block_io { 204 union ctl_io *io; 205 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 206 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 207 int bio_cmd; 208 int bio_flags; 209 int num_segs; 210 int num_bios_sent; 211 int num_bios_done; 212 int send_complete; 213 int num_errors; 214 struct bintime ds_t0; 215 devstat_tag_type ds_tag_type; 216 devstat_trans_flags ds_trans_type; 217 uint64_t io_len; 218 uint64_t io_offset; 219 struct ctl_be_block_softc *softc; 220 struct ctl_be_block_lun *lun; 221 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 222}; 223 224static int cbb_num_threads = 14; 225TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 226SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 227 "CAM Target Layer Block Backend"); 228SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 229 &cbb_num_threads, 0, "Number of threads per backing file"); 230 231static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 232static void ctl_free_beio(struct ctl_be_block_io *beio); 233static void ctl_complete_beio(struct ctl_be_block_io *beio); 234static int ctl_be_block_move_done(union ctl_io *io); 235static void ctl_be_block_biodone(struct bio *bio); 236static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 237 struct ctl_be_block_io *beio); 238static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 239 struct ctl_be_block_io *beio); 240static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 241 struct ctl_be_block_io *beio); 242static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 243 struct ctl_be_block_io *beio); 244static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 245 struct ctl_be_block_io *beio); 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_file(struct ctl_be_block_lun *be_lun, 267 struct ctl_lun_req *req); 268static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 269 struct ctl_lun_req *req); 270static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 271 struct ctl_lun_req *req); 272static void ctl_be_block_lun_shutdown(void *be_lun); 273static void ctl_be_block_lun_config_status(void *be_lun, 274 ctl_lun_config_status status); 275static int ctl_be_block_config_write(union ctl_io *io); 276static int ctl_be_block_config_read(union ctl_io *io); 277static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 278int ctl_be_block_init(void); 279 280static struct ctl_backend_driver ctl_be_block_driver = 281{ 282 .name = "block", 283 .flags = CTL_BE_FLAG_HAS_CONFIG, 284 .init = ctl_be_block_init, 285 .data_submit = ctl_be_block_submit, 286 .data_move_done = ctl_be_block_move_done, 287 .config_read = ctl_be_block_config_read, 288 .config_write = ctl_be_block_config_write, 289 .ioctl = ctl_be_block_ioctl, 290 .lun_info = ctl_be_block_lun_info 291}; 292 293MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 294CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 295 296static uma_zone_t beio_zone; 297 298static struct ctl_be_block_io * 299ctl_alloc_beio(struct ctl_be_block_softc *softc) 300{ 301 struct ctl_be_block_io *beio; 302 303 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 304 beio->softc = softc; 305 return (beio); 306} 307 308static void 309ctl_free_beio(struct ctl_be_block_io *beio) 310{ 311 int duplicate_free; 312 int i; 313 314 duplicate_free = 0; 315 316 for (i = 0; i < beio->num_segs; i++) { 317 if (beio->sg_segs[i].addr == NULL) 318 duplicate_free++; 319 320 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 321 beio->sg_segs[i].addr = NULL; 322 323 /* For compare we had two equal S/G lists. */ 324 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 325 uma_zfree(beio->lun->lun_zone, 326 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 327 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 328 } 329 } 330 331 if (duplicate_free > 0) { 332 printf("%s: %d duplicate frees out of %d segments\n", __func__, 333 duplicate_free, beio->num_segs); 334 } 335 336 uma_zfree(beio_zone, beio); 337} 338 339static void 340ctl_complete_beio(struct ctl_be_block_io *beio) 341{ 342 union ctl_io *io = beio->io; 343 344 if (beio->beio_cont != NULL) { 345 beio->beio_cont(beio); 346 } else { 347 ctl_free_beio(beio); 348 ctl_data_submit_done(io); 349 } 350} 351 352static int 353ctl_be_block_move_done(union ctl_io *io) 354{ 355 struct ctl_be_block_io *beio; 356 struct ctl_be_block_lun *be_lun; 357 struct ctl_lba_len_flags *lbalen; 358#ifdef CTL_TIME_IO 359 struct bintime cur_bt; 360#endif 361 int i; 362 363 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 364 be_lun = beio->lun; 365 366 DPRINTF("entered\n"); 367 368#ifdef CTL_TIME_IO 369 getbintime(&cur_bt); 370 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 371 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 372 io->io_hdr.num_dmas++; 373#endif 374 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 375 376 /* 377 * We set status at this point for read commands, and write 378 * commands with errors. 379 */ 380 if ((io->io_hdr.port_status == 0) && 381 ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 382 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 383 lbalen = ARGS(beio->io); 384 if (lbalen->flags & CTL_LLF_READ) { 385 ctl_set_success(&io->scsiio); 386 } else if (lbalen->flags & CTL_LLF_COMPARE) { 387 /* We have two data blocks ready for comparison. */ 388 for (i = 0; i < beio->num_segs; i++) { 389 if (memcmp(beio->sg_segs[i].addr, 390 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 391 beio->sg_segs[i].len) != 0) 392 break; 393 } 394 if (i < beio->num_segs) 395 ctl_set_sense(&io->scsiio, 396 /*current_error*/ 1, 397 /*sense_key*/ SSD_KEY_MISCOMPARE, 398 /*asc*/ 0x1D, 399 /*ascq*/ 0x00, 400 SSD_ELEM_NONE); 401 else 402 ctl_set_success(&io->scsiio); 403 } 404 } 405 else if ((io->io_hdr.port_status != 0) 406 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 407 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 408 /* 409 * For hardware error sense keys, the sense key 410 * specific value is defined to be a retry count, 411 * but we use it to pass back an internal FETD 412 * error code. XXX KDM Hopefully the FETD is only 413 * using 16 bits for an error code, since that's 414 * all the space we have in the sks field. 415 */ 416 ctl_set_internal_failure(&io->scsiio, 417 /*sks_valid*/ 1, 418 /*retry_count*/ 419 io->io_hdr.port_status); 420 } 421 422 /* 423 * If this is a read, or a write with errors, it is done. 424 */ 425 if ((beio->bio_cmd == BIO_READ) 426 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 427 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 428 ctl_complete_beio(beio); 429 return (0); 430 } 431 432 /* 433 * At this point, we have a write and the DMA completed 434 * successfully. We now have to queue it to the task queue to 435 * execute the backend I/O. That is because we do blocking 436 * memory allocations, and in the file backing case, blocking I/O. 437 * This move done routine is generally called in the SIM's 438 * interrupt context, and therefore we cannot block. 439 */ 440 mtx_lock(&be_lun->queue_lock); 441 /* 442 * XXX KDM make sure that links is okay to use at this point. 443 * Otherwise, we either need to add another field to ctl_io_hdr, 444 * or deal with resource allocation here. 445 */ 446 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 447 mtx_unlock(&be_lun->queue_lock); 448 449 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 450 451 return (0); 452} 453 454static void 455ctl_be_block_biodone(struct bio *bio) 456{ 457 struct ctl_be_block_io *beio; 458 struct ctl_be_block_lun *be_lun; 459 union ctl_io *io; 460 int error; 461 462 beio = bio->bio_caller1; 463 be_lun = beio->lun; 464 io = beio->io; 465 466 DPRINTF("entered\n"); 467 468 error = bio->bio_error; 469 mtx_lock(&be_lun->io_lock); 470 if (error != 0) 471 beio->num_errors++; 472 473 beio->num_bios_done++; 474 475 /* 476 * XXX KDM will this cause WITNESS to complain? Holding a lock 477 * during the free might cause it to complain. 478 */ 479 g_destroy_bio(bio); 480 481 /* 482 * If the send complete bit isn't set, or we aren't the last I/O to 483 * complete, then we're done. 484 */ 485 if ((beio->send_complete == 0) 486 || (beio->num_bios_done < beio->num_bios_sent)) { 487 mtx_unlock(&be_lun->io_lock); 488 return; 489 } 490 491 /* 492 * At this point, we've verified that we are the last I/O to 493 * complete, so it's safe to drop the lock. 494 */ 495 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 496 beio->ds_tag_type, beio->ds_trans_type, 497 /*now*/ NULL, /*then*/&beio->ds_t0); 498 mtx_unlock(&be_lun->io_lock); 499 500 /* 501 * If there are any errors from the backing device, we fail the 502 * entire I/O with a medium error. 503 */ 504 if (beio->num_errors > 0) { 505 if (error == EOPNOTSUPP) { 506 ctl_set_invalid_opcode(&io->scsiio); 507 } else if (beio->bio_cmd == BIO_FLUSH) { 508 /* XXX KDM is there is a better error here? */ 509 ctl_set_internal_failure(&io->scsiio, 510 /*sks_valid*/ 1, 511 /*retry_count*/ 0xbad2); 512 } else 513 ctl_set_medium_error(&io->scsiio); 514 ctl_complete_beio(beio); 515 return; 516 } 517 518 /* 519 * If this is a write, a flush, a delete or verify, we're all done. 520 * If this is a read, we can now send the data to the user. 521 */ 522 if ((beio->bio_cmd == BIO_WRITE) 523 || (beio->bio_cmd == BIO_FLUSH) 524 || (beio->bio_cmd == BIO_DELETE) 525 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 526 ctl_set_success(&io->scsiio); 527 ctl_complete_beio(beio); 528 } else { 529#ifdef CTL_TIME_IO 530 getbintime(&io->io_hdr.dma_start_bt); 531#endif 532 ctl_datamove(io); 533 } 534} 535 536static void 537ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 538 struct ctl_be_block_io *beio) 539{ 540 union ctl_io *io = beio->io; 541 struct mount *mountpoint; 542 int error, lock_flags; 543 544 DPRINTF("entered\n"); 545 546 binuptime(&beio->ds_t0); 547 mtx_lock(&be_lun->io_lock); 548 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 549 mtx_unlock(&be_lun->io_lock); 550 551 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 552 553 if (MNT_SHARED_WRITES(mountpoint) 554 || ((mountpoint == NULL) 555 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 556 lock_flags = LK_SHARED; 557 else 558 lock_flags = LK_EXCLUSIVE; 559 560 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 561 562 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 563 VOP_UNLOCK(be_lun->vn, 0); 564 565 vn_finished_write(mountpoint); 566 567 mtx_lock(&be_lun->io_lock); 568 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 569 beio->ds_tag_type, beio->ds_trans_type, 570 /*now*/ NULL, /*then*/&beio->ds_t0); 571 mtx_unlock(&be_lun->io_lock); 572 573 if (error == 0) 574 ctl_set_success(&io->scsiio); 575 else { 576 /* XXX KDM is there is a better error here? */ 577 ctl_set_internal_failure(&io->scsiio, 578 /*sks_valid*/ 1, 579 /*retry_count*/ 0xbad1); 580 } 581 582 ctl_complete_beio(beio); 583} 584 585SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 586SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 587SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 588SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 589 590static void 591ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 592 struct ctl_be_block_io *beio) 593{ 594 struct ctl_be_block_filedata *file_data; 595 union ctl_io *io; 596 struct uio xuio; 597 struct iovec *xiovec; 598 int flags; 599 int error, i; 600 601 DPRINTF("entered\n"); 602 603 file_data = &be_lun->backend.file; 604 io = beio->io; 605 flags = beio->bio_flags; 606 607 bzero(&xuio, sizeof(xuio)); 608 if (beio->bio_cmd == BIO_READ) { 609 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 610 xuio.uio_rw = UIO_READ; 611 } else { 612 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 613 xuio.uio_rw = UIO_WRITE; 614 } 615 xuio.uio_offset = beio->io_offset; 616 xuio.uio_resid = beio->io_len; 617 xuio.uio_segflg = UIO_SYSSPACE; 618 xuio.uio_iov = beio->xiovecs; 619 xuio.uio_iovcnt = beio->num_segs; 620 xuio.uio_td = curthread; 621 622 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 623 xiovec->iov_base = beio->sg_segs[i].addr; 624 xiovec->iov_len = beio->sg_segs[i].len; 625 } 626 627 binuptime(&beio->ds_t0); 628 mtx_lock(&be_lun->io_lock); 629 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 630 mtx_unlock(&be_lun->io_lock); 631 632 if (beio->bio_cmd == BIO_READ) { 633 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 634 635 /* 636 * UFS pays attention to IO_DIRECT for reads. If the 637 * DIRECTIO option is configured into the kernel, it calls 638 * ffs_rawread(). But that only works for single-segment 639 * uios with user space addresses. In our case, with a 640 * kernel uio, it still reads into the buffer cache, but it 641 * will just try to release the buffer from the cache later 642 * on in ffs_read(). 643 * 644 * ZFS does not pay attention to IO_DIRECT for reads. 645 * 646 * UFS does not pay attention to IO_SYNC for reads. 647 * 648 * ZFS pays attention to IO_SYNC (which translates into the 649 * Solaris define FRSYNC for zfs_read()) for reads. It 650 * attempts to sync the file before reading. 651 * 652 * So, to attempt to provide some barrier semantics in the 653 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 654 */ 655 error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 656 (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 657 658 VOP_UNLOCK(be_lun->vn, 0); 659 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 660 } else { 661 struct mount *mountpoint; 662 int lock_flags; 663 664 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 665 666 if (MNT_SHARED_WRITES(mountpoint) 667 || ((mountpoint == NULL) 668 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 669 lock_flags = LK_SHARED; 670 else 671 lock_flags = LK_EXCLUSIVE; 672 673 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 674 675 /* 676 * UFS pays attention to IO_DIRECT for writes. The write 677 * is done asynchronously. (Normally the write would just 678 * get put into cache. 679 * 680 * UFS pays attention to IO_SYNC for writes. It will 681 * attempt to write the buffer out synchronously if that 682 * flag is set. 683 * 684 * ZFS does not pay attention to IO_DIRECT for writes. 685 * 686 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 687 * for writes. It will flush the transaction from the 688 * cache before returning. 689 * 690 * So if we've got the BIO_ORDERED flag set, we want 691 * IO_SYNC in either the UFS or ZFS case. 692 */ 693 error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 694 IO_SYNC : 0, file_data->cred); 695 VOP_UNLOCK(be_lun->vn, 0); 696 697 vn_finished_write(mountpoint); 698 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 699 } 700 701 mtx_lock(&be_lun->io_lock); 702 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 703 beio->ds_tag_type, beio->ds_trans_type, 704 /*now*/ NULL, /*then*/&beio->ds_t0); 705 mtx_unlock(&be_lun->io_lock); 706 707 /* 708 * If we got an error, set the sense data to "MEDIUM ERROR" and 709 * return the I/O to the user. 710 */ 711 if (error != 0) { 712 char path_str[32]; 713 714 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 715 /* 716 * XXX KDM ZFS returns ENOSPC when the underlying 717 * filesystem fills up. What kind of SCSI error should we 718 * return for that? 719 */ 720 printf("%s%s command returned errno %d\n", path_str, 721 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 722 ctl_set_medium_error(&io->scsiio); 723 ctl_complete_beio(beio); 724 return; 725 } 726 727 /* 728 * If this is a write or a verify, we're all done. 729 * If this is a read, we can now send the data to the user. 730 */ 731 if ((beio->bio_cmd == BIO_WRITE) || 732 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 733 ctl_set_success(&io->scsiio); 734 ctl_complete_beio(beio); 735 } else { 736#ifdef CTL_TIME_IO 737 getbintime(&io->io_hdr.dma_start_bt); 738#endif 739 ctl_datamove(io); 740 } 741} 742 743static void 744ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 745 struct ctl_be_block_io *beio) 746{ 747 struct bio *bio; 748 union ctl_io *io; 749 struct ctl_be_block_devdata *dev_data; 750 751 dev_data = &be_lun->backend.dev; 752 io = beio->io; 753 754 DPRINTF("entered\n"); 755 756 /* This can't fail, it's a blocking allocation. */ 757 bio = g_alloc_bio(); 758 759 bio->bio_cmd = BIO_FLUSH; 760 bio->bio_flags |= BIO_ORDERED; 761 bio->bio_dev = dev_data->cdev; 762 bio->bio_offset = 0; 763 bio->bio_data = 0; 764 bio->bio_done = ctl_be_block_biodone; 765 bio->bio_caller1 = beio; 766 bio->bio_pblkno = 0; 767 768 /* 769 * We don't need to acquire the LUN lock here, because we are only 770 * sending one bio, and so there is no other context to synchronize 771 * with. 772 */ 773 beio->num_bios_sent = 1; 774 beio->send_complete = 1; 775 776 binuptime(&beio->ds_t0); 777 mtx_lock(&be_lun->io_lock); 778 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 779 mtx_unlock(&be_lun->io_lock); 780 781 (*dev_data->csw->d_strategy)(bio); 782} 783 784static void 785ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 786 struct ctl_be_block_io *beio, 787 uint64_t off, uint64_t len, int last) 788{ 789 struct bio *bio; 790 struct ctl_be_block_devdata *dev_data; 791 uint64_t maxlen; 792 793 dev_data = &be_lun->backend.dev; 794 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 795 while (len > 0) { 796 bio = g_alloc_bio(); 797 bio->bio_cmd = BIO_DELETE; 798 bio->bio_flags |= beio->bio_flags; 799 bio->bio_dev = dev_data->cdev; 800 bio->bio_offset = off; 801 bio->bio_length = MIN(len, maxlen); 802 bio->bio_data = 0; 803 bio->bio_done = ctl_be_block_biodone; 804 bio->bio_caller1 = beio; 805 bio->bio_pblkno = off / be_lun->blocksize; 806 807 off += bio->bio_length; 808 len -= bio->bio_length; 809 810 mtx_lock(&be_lun->io_lock); 811 beio->num_bios_sent++; 812 if (last && len == 0) 813 beio->send_complete = 1; 814 mtx_unlock(&be_lun->io_lock); 815 816 (*dev_data->csw->d_strategy)(bio); 817 } 818} 819 820static void 821ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 822 struct ctl_be_block_io *beio) 823{ 824 union ctl_io *io; 825 struct ctl_be_block_devdata *dev_data; 826 struct ctl_ptr_len_flags *ptrlen; 827 struct scsi_unmap_desc *buf, *end; 828 uint64_t len; 829 830 dev_data = &be_lun->backend.dev; 831 io = beio->io; 832 833 DPRINTF("entered\n"); 834 835 binuptime(&beio->ds_t0); 836 mtx_lock(&be_lun->io_lock); 837 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 838 mtx_unlock(&be_lun->io_lock); 839 840 if (beio->io_offset == -1) { 841 beio->io_len = 0; 842 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 843 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 844 end = buf + ptrlen->len / sizeof(*buf); 845 for (; buf < end; buf++) { 846 len = (uint64_t)scsi_4btoul(buf->length) * 847 be_lun->blocksize; 848 beio->io_len += len; 849 ctl_be_block_unmap_dev_range(be_lun, beio, 850 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 851 (end - buf < 2) ? TRUE : FALSE); 852 } 853 } else 854 ctl_be_block_unmap_dev_range(be_lun, beio, 855 beio->io_offset, beio->io_len, TRUE); 856} 857 858static void 859ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 860 struct ctl_be_block_io *beio) 861{ 862 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 863 int i; 864 struct bio *bio; 865 struct ctl_be_block_devdata *dev_data; 866 off_t cur_offset; 867 int max_iosize; 868 869 DPRINTF("entered\n"); 870 871 dev_data = &be_lun->backend.dev; 872 873 /* 874 * We have to limit our I/O size to the maximum supported by the 875 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 876 * set it properly, use DFLTPHYS. 877 */ 878 max_iosize = dev_data->cdev->si_iosize_max; 879 if (max_iosize < PAGE_SIZE) 880 max_iosize = DFLTPHYS; 881 882 cur_offset = beio->io_offset; 883 for (i = 0; i < beio->num_segs; i++) { 884 size_t cur_size; 885 uint8_t *cur_ptr; 886 887 cur_size = beio->sg_segs[i].len; 888 cur_ptr = beio->sg_segs[i].addr; 889 890 while (cur_size > 0) { 891 /* This can't fail, it's a blocking allocation. */ 892 bio = g_alloc_bio(); 893 894 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 895 896 bio->bio_cmd = beio->bio_cmd; 897 bio->bio_flags |= beio->bio_flags; 898 bio->bio_dev = dev_data->cdev; 899 bio->bio_caller1 = beio; 900 bio->bio_length = min(cur_size, max_iosize); 901 bio->bio_offset = cur_offset; 902 bio->bio_data = cur_ptr; 903 bio->bio_done = ctl_be_block_biodone; 904 bio->bio_pblkno = cur_offset / be_lun->blocksize; 905 906 cur_offset += bio->bio_length; 907 cur_ptr += bio->bio_length; 908 cur_size -= bio->bio_length; 909 910 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 911 beio->num_bios_sent++; 912 } 913 } 914 binuptime(&beio->ds_t0); 915 mtx_lock(&be_lun->io_lock); 916 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 917 beio->send_complete = 1; 918 mtx_unlock(&be_lun->io_lock); 919 920 /* 921 * Fire off all allocated requests! 922 */ 923 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 924 TAILQ_REMOVE(&queue, bio, bio_queue); 925 (*dev_data->csw->d_strategy)(bio); 926 } 927} 928 929static void 930ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 931{ 932 union ctl_io *io; 933 934 io = beio->io; 935 ctl_free_beio(beio); 936 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 937 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 938 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 939 ctl_config_write_done(io); 940 return; 941 } 942 943 ctl_be_block_config_write(io); 944} 945 946static void 947ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 948 union ctl_io *io) 949{ 950 struct ctl_be_block_io *beio; 951 struct ctl_be_block_softc *softc; 952 struct ctl_lba_len_flags *lbalen; 953 uint64_t len_left, lba; 954 int i, seglen; 955 uint8_t *buf, *end; 956 957 DPRINTF("entered\n"); 958 959 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 960 softc = be_lun->softc; 961 lbalen = ARGS(beio->io); 962 963 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP) || 964 (lbalen->flags & SWS_UNMAP && be_lun->unmap == NULL)) { 965 ctl_free_beio(beio); 966 ctl_set_invalid_field(&io->scsiio, 967 /*sks_valid*/ 1, 968 /*command*/ 1, 969 /*field*/ 1, 970 /*bit_valid*/ 0, 971 /*bit*/ 0); 972 ctl_config_write_done(io); 973 return; 974 } 975 976 /* 977 * If the I/O came down with an ordered or head of queue tag, set 978 * the BIO_ORDERED attribute. For head of queue tags, that's 979 * pretty much the best we can do. 980 */ 981 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 982 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 983 beio->bio_flags = BIO_ORDERED; 984 985 switch (io->scsiio.tag_type) { 986 case CTL_TAG_ORDERED: 987 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 988 break; 989 case CTL_TAG_HEAD_OF_QUEUE: 990 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 991 break; 992 case CTL_TAG_UNTAGGED: 993 case CTL_TAG_SIMPLE: 994 case CTL_TAG_ACA: 995 default: 996 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 997 break; 998 } 999 1000 if (lbalen->flags & SWS_UNMAP) { 1001 beio->io_offset = lbalen->lba * be_lun->blocksize; 1002 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1003 beio->bio_cmd = BIO_DELETE; 1004 beio->ds_trans_type = DEVSTAT_FREE; 1005 1006 be_lun->unmap(be_lun, beio); 1007 return; 1008 } 1009 1010 beio->bio_cmd = BIO_WRITE; 1011 beio->ds_trans_type = DEVSTAT_WRITE; 1012 1013 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1014 (uintmax_t)lbalen->lba, lbalen->len); 1015 1016 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1017 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1018 1019 /* 1020 * Setup the S/G entry for this chunk. 1021 */ 1022 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1023 seglen -= seglen % be_lun->blocksize; 1024 beio->sg_segs[i].len = seglen; 1025 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1026 1027 DPRINTF("segment %d addr %p len %zd\n", i, 1028 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1029 1030 beio->num_segs++; 1031 len_left -= seglen; 1032 1033 buf = beio->sg_segs[i].addr; 1034 end = buf + seglen; 1035 for (; buf < end; buf += be_lun->blocksize) { 1036 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1037 if (lbalen->flags & SWS_LBDATA) 1038 scsi_ulto4b(lbalen->lba + lba, buf); 1039 lba++; 1040 } 1041 } 1042 1043 beio->io_offset = lbalen->lba * be_lun->blocksize; 1044 beio->io_len = lba * be_lun->blocksize; 1045 1046 /* We can not do all in one run. Correct and schedule rerun. */ 1047 if (len_left > 0) { 1048 lbalen->lba += lba; 1049 lbalen->len -= lba; 1050 beio->beio_cont = ctl_be_block_cw_done_ws; 1051 } 1052 1053 be_lun->dispatch(be_lun, beio); 1054} 1055 1056static void 1057ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1058 union ctl_io *io) 1059{ 1060 struct ctl_be_block_io *beio; 1061 struct ctl_be_block_softc *softc; 1062 struct ctl_ptr_len_flags *ptrlen; 1063 1064 DPRINTF("entered\n"); 1065 1066 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1067 softc = be_lun->softc; 1068 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1069 1070 if (ptrlen->flags != 0 || be_lun->unmap == NULL) { 1071 ctl_free_beio(beio); 1072 ctl_set_invalid_field(&io->scsiio, 1073 /*sks_valid*/ 0, 1074 /*command*/ 1, 1075 /*field*/ 0, 1076 /*bit_valid*/ 0, 1077 /*bit*/ 0); 1078 ctl_config_write_done(io); 1079 return; 1080 } 1081 1082 /* 1083 * If the I/O came down with an ordered or head of queue tag, set 1084 * the BIO_ORDERED attribute. For head of queue tags, that's 1085 * pretty much the best we can do. 1086 */ 1087 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1088 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1089 beio->bio_flags = BIO_ORDERED; 1090 1091 switch (io->scsiio.tag_type) { 1092 case CTL_TAG_ORDERED: 1093 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1094 break; 1095 case CTL_TAG_HEAD_OF_QUEUE: 1096 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1097 break; 1098 case CTL_TAG_UNTAGGED: 1099 case CTL_TAG_SIMPLE: 1100 case CTL_TAG_ACA: 1101 default: 1102 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1103 break; 1104 } 1105 1106 beio->io_len = 0; 1107 beio->io_offset = -1; 1108 1109 beio->bio_cmd = BIO_DELETE; 1110 beio->ds_trans_type = DEVSTAT_FREE; 1111 1112 DPRINTF("UNMAP\n"); 1113 1114 be_lun->unmap(be_lun, beio); 1115} 1116 1117static void 1118ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1119{ 1120 union ctl_io *io; 1121 1122 io = beio->io; 1123 ctl_free_beio(beio); 1124 ctl_config_write_done(io); 1125} 1126 1127static void 1128ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1129 union ctl_io *io) 1130{ 1131 struct ctl_be_block_io *beio; 1132 struct ctl_be_block_softc *softc; 1133 1134 DPRINTF("entered\n"); 1135 1136 softc = be_lun->softc; 1137 beio = ctl_alloc_beio(softc); 1138 beio->io = io; 1139 beio->lun = be_lun; 1140 beio->beio_cont = ctl_be_block_cw_done; 1141 PRIV(io)->ptr = (void *)beio; 1142 1143 switch (io->scsiio.cdb[0]) { 1144 case SYNCHRONIZE_CACHE: 1145 case SYNCHRONIZE_CACHE_16: 1146 beio->bio_cmd = BIO_FLUSH; 1147 beio->ds_trans_type = DEVSTAT_NO_DATA; 1148 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1149 beio->io_len = 0; 1150 be_lun->lun_flush(be_lun, beio); 1151 break; 1152 case WRITE_SAME_10: 1153 case WRITE_SAME_16: 1154 ctl_be_block_cw_dispatch_ws(be_lun, io); 1155 break; 1156 case UNMAP: 1157 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1158 break; 1159 default: 1160 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1161 break; 1162 } 1163} 1164 1165SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1166SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1167SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1168SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1169 1170static void 1171ctl_be_block_next(struct ctl_be_block_io *beio) 1172{ 1173 struct ctl_be_block_lun *be_lun; 1174 union ctl_io *io; 1175 1176 io = beio->io; 1177 be_lun = beio->lun; 1178 ctl_free_beio(beio); 1179 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1180 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1181 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1182 ctl_data_submit_done(io); 1183 return; 1184 } 1185 1186 io->io_hdr.status &= ~CTL_STATUS_MASK; 1187 io->io_hdr.status |= CTL_STATUS_NONE; 1188 1189 mtx_lock(&be_lun->queue_lock); 1190 /* 1191 * XXX KDM make sure that links is okay to use at this point. 1192 * Otherwise, we either need to add another field to ctl_io_hdr, 1193 * or deal with resource allocation here. 1194 */ 1195 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1196 mtx_unlock(&be_lun->queue_lock); 1197 1198 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1199} 1200 1201static void 1202ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1203 union ctl_io *io) 1204{ 1205 struct ctl_be_block_io *beio; 1206 struct ctl_be_block_softc *softc; 1207 struct ctl_lba_len_flags *lbalen; 1208 struct ctl_ptr_len_flags *bptrlen; 1209 uint64_t len_left, lbas; 1210 int i; 1211 1212 softc = be_lun->softc; 1213 1214 DPRINTF("entered\n"); 1215 1216 lbalen = ARGS(io); 1217 if (lbalen->flags & CTL_LLF_WRITE) { 1218 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1219 } else { 1220 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1221 } 1222 1223 beio = ctl_alloc_beio(softc); 1224 beio->io = io; 1225 beio->lun = be_lun; 1226 bptrlen = PRIV(io); 1227 bptrlen->ptr = (void *)beio; 1228 1229 /* 1230 * If the I/O came down with an ordered or head of queue tag, set 1231 * the BIO_ORDERED attribute. For head of queue tags, that's 1232 * pretty much the best we can do. 1233 * 1234 * XXX KDM we don't have a great way to easily know about the FUA 1235 * bit right now (it is decoded in ctl_read_write(), but we don't 1236 * pass that knowledge to the backend), and in any case we would 1237 * need to determine how to handle it. 1238 */ 1239 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1240 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1241 beio->bio_flags = BIO_ORDERED; 1242 1243 switch (io->scsiio.tag_type) { 1244 case CTL_TAG_ORDERED: 1245 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1246 break; 1247 case CTL_TAG_HEAD_OF_QUEUE: 1248 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1249 break; 1250 case CTL_TAG_UNTAGGED: 1251 case CTL_TAG_SIMPLE: 1252 case CTL_TAG_ACA: 1253 default: 1254 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1255 break; 1256 } 1257 1258 if (lbalen->flags & CTL_LLF_WRITE) { 1259 beio->bio_cmd = BIO_WRITE; 1260 beio->ds_trans_type = DEVSTAT_WRITE; 1261 } else { 1262 beio->bio_cmd = BIO_READ; 1263 beio->ds_trans_type = DEVSTAT_READ; 1264 } 1265 1266 DPRINTF("%s at LBA %jx len %u @%ju\n", 1267 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1268 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1269 if (lbalen->flags & CTL_LLF_COMPARE) 1270 lbas = CTLBLK_HALF_IO_SIZE; 1271 else 1272 lbas = CTLBLK_MAX_IO_SIZE; 1273 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1274 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1275 beio->io_len = lbas * be_lun->blocksize; 1276 bptrlen->len += lbas; 1277 1278 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1279 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1280 i, CTLBLK_MAX_SEGS)); 1281 1282 /* 1283 * Setup the S/G entry for this chunk. 1284 */ 1285 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1286 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1287 1288 DPRINTF("segment %d addr %p len %zd\n", i, 1289 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1290 1291 /* Set up second segment for compare operation. */ 1292 if (lbalen->flags & CTL_LLF_COMPARE) { 1293 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1294 beio->sg_segs[i].len; 1295 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1296 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1297 } 1298 1299 beio->num_segs++; 1300 len_left -= beio->sg_segs[i].len; 1301 } 1302 if (bptrlen->len < lbalen->len) 1303 beio->beio_cont = ctl_be_block_next; 1304 io->scsiio.be_move_done = ctl_be_block_move_done; 1305 /* For compare we have separate S/G lists for read and datamove. */ 1306 if (lbalen->flags & CTL_LLF_COMPARE) 1307 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1308 else 1309 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1310 io->scsiio.kern_data_len = beio->io_len; 1311 io->scsiio.kern_data_resid = 0; 1312 io->scsiio.kern_sg_entries = beio->num_segs; 1313 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1314 1315 /* 1316 * For the read case, we need to read the data into our buffers and 1317 * then we can send it back to the user. For the write case, we 1318 * need to get the data from the user first. 1319 */ 1320 if (beio->bio_cmd == BIO_READ) { 1321 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1322 be_lun->dispatch(be_lun, beio); 1323 } else { 1324 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1325#ifdef CTL_TIME_IO 1326 getbintime(&io->io_hdr.dma_start_bt); 1327#endif 1328 ctl_datamove(io); 1329 } 1330} 1331 1332static void 1333ctl_be_block_worker(void *context, int pending) 1334{ 1335 struct ctl_be_block_lun *be_lun; 1336 struct ctl_be_block_softc *softc; 1337 union ctl_io *io; 1338 1339 be_lun = (struct ctl_be_block_lun *)context; 1340 softc = be_lun->softc; 1341 1342 DPRINTF("entered\n"); 1343 1344 mtx_lock(&be_lun->queue_lock); 1345 for (;;) { 1346 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1347 if (io != NULL) { 1348 struct ctl_be_block_io *beio; 1349 1350 DPRINTF("datamove queue\n"); 1351 1352 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1353 ctl_io_hdr, links); 1354 1355 mtx_unlock(&be_lun->queue_lock); 1356 1357 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1358 1359 be_lun->dispatch(be_lun, beio); 1360 1361 mtx_lock(&be_lun->queue_lock); 1362 continue; 1363 } 1364 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1365 if (io != NULL) { 1366 1367 DPRINTF("config write queue\n"); 1368 1369 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1370 ctl_io_hdr, links); 1371 1372 mtx_unlock(&be_lun->queue_lock); 1373 1374 ctl_be_block_cw_dispatch(be_lun, io); 1375 1376 mtx_lock(&be_lun->queue_lock); 1377 continue; 1378 } 1379 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1380 if (io != NULL) { 1381 DPRINTF("input queue\n"); 1382 1383 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1384 ctl_io_hdr, links); 1385 mtx_unlock(&be_lun->queue_lock); 1386 1387 /* 1388 * We must drop the lock, since this routine and 1389 * its children may sleep. 1390 */ 1391 ctl_be_block_dispatch(be_lun, io); 1392 1393 mtx_lock(&be_lun->queue_lock); 1394 continue; 1395 } 1396 1397 /* 1398 * If we get here, there is no work left in the queues, so 1399 * just break out and let the task queue go to sleep. 1400 */ 1401 break; 1402 } 1403 mtx_unlock(&be_lun->queue_lock); 1404} 1405 1406/* 1407 * Entry point from CTL to the backend for I/O. We queue everything to a 1408 * work thread, so this just puts the I/O on a queue and wakes up the 1409 * thread. 1410 */ 1411static int 1412ctl_be_block_submit(union ctl_io *io) 1413{ 1414 struct ctl_be_block_lun *be_lun; 1415 struct ctl_be_lun *ctl_be_lun; 1416 1417 DPRINTF("entered\n"); 1418 1419 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1420 CTL_PRIV_BACKEND_LUN].ptr; 1421 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1422 1423 /* 1424 * Make sure we only get SCSI I/O. 1425 */ 1426 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1427 "%#x) encountered", io->io_hdr.io_type)); 1428 1429 PRIV(io)->len = 0; 1430 1431 mtx_lock(&be_lun->queue_lock); 1432 /* 1433 * XXX KDM make sure that links is okay to use at this point. 1434 * Otherwise, we either need to add another field to ctl_io_hdr, 1435 * or deal with resource allocation here. 1436 */ 1437 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1438 mtx_unlock(&be_lun->queue_lock); 1439 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1440 1441 return (CTL_RETVAL_COMPLETE); 1442} 1443 1444static int 1445ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1446 int flag, struct thread *td) 1447{ 1448 struct ctl_be_block_softc *softc; 1449 int error; 1450 1451 softc = &backend_block_softc; 1452 1453 error = 0; 1454 1455 switch (cmd) { 1456 case CTL_LUN_REQ: { 1457 struct ctl_lun_req *lun_req; 1458 1459 lun_req = (struct ctl_lun_req *)addr; 1460 1461 switch (lun_req->reqtype) { 1462 case CTL_LUNREQ_CREATE: 1463 error = ctl_be_block_create(softc, lun_req); 1464 break; 1465 case CTL_LUNREQ_RM: 1466 error = ctl_be_block_rm(softc, lun_req); 1467 break; 1468 case CTL_LUNREQ_MODIFY: 1469 error = ctl_be_block_modify(softc, lun_req); 1470 break; 1471 default: 1472 lun_req->status = CTL_LUN_ERROR; 1473 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1474 "%s: invalid LUN request type %d", __func__, 1475 lun_req->reqtype); 1476 break; 1477 } 1478 break; 1479 } 1480 default: 1481 error = ENOTTY; 1482 break; 1483 } 1484 1485 return (error); 1486} 1487 1488static int 1489ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1490{ 1491 struct ctl_be_block_filedata *file_data; 1492 struct ctl_lun_create_params *params; 1493 struct vattr vattr; 1494 int error; 1495 1496 error = 0; 1497 file_data = &be_lun->backend.file; 1498 params = &req->reqdata.create; 1499 1500 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1501 be_lun->dispatch = ctl_be_block_dispatch_file; 1502 be_lun->lun_flush = ctl_be_block_flush_file; 1503 1504 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1505 if (error != 0) { 1506 snprintf(req->error_str, sizeof(req->error_str), 1507 "error calling VOP_GETATTR() for file %s", 1508 be_lun->dev_path); 1509 return (error); 1510 } 1511 1512 /* 1513 * Verify that we have the ability to upgrade to exclusive 1514 * access on this file so we can trap errors at open instead 1515 * of reporting them during first access. 1516 */ 1517 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1518 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1519 if (be_lun->vn->v_iflag & VI_DOOMED) { 1520 error = EBADF; 1521 snprintf(req->error_str, sizeof(req->error_str), 1522 "error locking file %s", be_lun->dev_path); 1523 return (error); 1524 } 1525 } 1526 1527 1528 file_data->cred = crhold(curthread->td_ucred); 1529 if (params->lun_size_bytes != 0) 1530 be_lun->size_bytes = params->lun_size_bytes; 1531 else 1532 be_lun->size_bytes = vattr.va_size; 1533 /* 1534 * We set the multi thread flag for file operations because all 1535 * filesystems (in theory) are capable of allowing multiple readers 1536 * of a file at once. So we want to get the maximum possible 1537 * concurrency. 1538 */ 1539 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1540 1541 /* 1542 * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1543 * With ZFS, it is 131072 bytes. Block sizes that large don't work 1544 * with disklabel and UFS on FreeBSD at least. Large block sizes 1545 * may not work with other OSes as well. So just export a sector 1546 * size of 512 bytes, which should work with any OS or 1547 * application. Since our backing is a file, any block size will 1548 * work fine for the backing store. 1549 */ 1550#if 0 1551 be_lun->blocksize= vattr.va_blocksize; 1552#endif 1553 if (params->blocksize_bytes != 0) 1554 be_lun->blocksize = params->blocksize_bytes; 1555 else 1556 be_lun->blocksize = 512; 1557 1558 /* 1559 * Sanity check. The media size has to be at least one 1560 * sector long. 1561 */ 1562 if (be_lun->size_bytes < be_lun->blocksize) { 1563 error = EINVAL; 1564 snprintf(req->error_str, sizeof(req->error_str), 1565 "file %s size %ju < block size %u", be_lun->dev_path, 1566 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1567 } 1568 return (error); 1569} 1570 1571static int 1572ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1573{ 1574 struct ctl_lun_create_params *params; 1575 struct vattr vattr; 1576 struct cdev *dev; 1577 struct cdevsw *devsw; 1578 int error; 1579 off_t ps, pss, po, pos; 1580 1581 params = &req->reqdata.create; 1582 1583 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1584 be_lun->dispatch = ctl_be_block_dispatch_dev; 1585 be_lun->lun_flush = ctl_be_block_flush_dev; 1586 be_lun->unmap = ctl_be_block_unmap_dev; 1587 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1588 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1589 &be_lun->backend.dev.dev_ref); 1590 if (be_lun->backend.dev.csw == NULL) 1591 panic("Unable to retrieve device switch"); 1592 1593 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1594 if (error) { 1595 snprintf(req->error_str, sizeof(req->error_str), 1596 "%s: error getting vnode attributes for device %s", 1597 __func__, be_lun->dev_path); 1598 return (error); 1599 } 1600 1601 dev = be_lun->vn->v_rdev; 1602 devsw = dev->si_devsw; 1603 if (!devsw->d_ioctl) { 1604 snprintf(req->error_str, sizeof(req->error_str), 1605 "%s: no d_ioctl for device %s!", __func__, 1606 be_lun->dev_path); 1607 return (ENODEV); 1608 } 1609 1610 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1611 (caddr_t)&be_lun->blocksize, FREAD, 1612 curthread); 1613 if (error) { 1614 snprintf(req->error_str, sizeof(req->error_str), 1615 "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1616 "on %s!", __func__, error, be_lun->dev_path); 1617 return (error); 1618 } 1619 1620 /* 1621 * If the user has asked for a blocksize that is greater than the 1622 * backing device's blocksize, we can do it only if the blocksize 1623 * the user is asking for is an even multiple of the underlying 1624 * device's blocksize. 1625 */ 1626 if ((params->blocksize_bytes != 0) 1627 && (params->blocksize_bytes > be_lun->blocksize)) { 1628 uint32_t bs_multiple, tmp_blocksize; 1629 1630 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1631 1632 tmp_blocksize = bs_multiple * be_lun->blocksize; 1633 1634 if (tmp_blocksize == params->blocksize_bytes) { 1635 be_lun->blocksize = params->blocksize_bytes; 1636 } else { 1637 snprintf(req->error_str, sizeof(req->error_str), 1638 "%s: requested blocksize %u is not an even " 1639 "multiple of backing device blocksize %u", 1640 __func__, params->blocksize_bytes, 1641 be_lun->blocksize); 1642 return (EINVAL); 1643 1644 } 1645 } else if ((params->blocksize_bytes != 0) 1646 && (params->blocksize_bytes != be_lun->blocksize)) { 1647 snprintf(req->error_str, sizeof(req->error_str), 1648 "%s: requested blocksize %u < backing device " 1649 "blocksize %u", __func__, params->blocksize_bytes, 1650 be_lun->blocksize); 1651 return (EINVAL); 1652 } 1653 1654 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1655 (caddr_t)&be_lun->size_bytes, FREAD, 1656 curthread); 1657 if (error) { 1658 snprintf(req->error_str, sizeof(req->error_str), 1659 "%s: error %d returned for DIOCGMEDIASIZE " 1660 " ioctl on %s!", __func__, error, 1661 be_lun->dev_path); 1662 return (error); 1663 } 1664 1665 if (params->lun_size_bytes != 0) { 1666 if (params->lun_size_bytes > be_lun->size_bytes) { 1667 snprintf(req->error_str, sizeof(req->error_str), 1668 "%s: requested LUN size %ju > backing device " 1669 "size %ju", __func__, 1670 (uintmax_t)params->lun_size_bytes, 1671 (uintmax_t)be_lun->size_bytes); 1672 return (EINVAL); 1673 } 1674 1675 be_lun->size_bytes = params->lun_size_bytes; 1676 } 1677 1678 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1679 (caddr_t)&ps, FREAD, curthread); 1680 if (error) 1681 ps = po = 0; 1682 else { 1683 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1684 (caddr_t)&po, FREAD, curthread); 1685 if (error) 1686 po = 0; 1687 } 1688 pss = ps / be_lun->blocksize; 1689 pos = po / be_lun->blocksize; 1690 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1691 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1692 be_lun->pblockexp = fls(pss) - 1; 1693 be_lun->pblockoff = (pss - pos) % pss; 1694 } 1695 1696 return (0); 1697} 1698 1699static int 1700ctl_be_block_close(struct ctl_be_block_lun *be_lun) 1701{ 1702 DROP_GIANT(); 1703 if (be_lun->vn) { 1704 int flags = FREAD | FWRITE; 1705 1706 switch (be_lun->dev_type) { 1707 case CTL_BE_BLOCK_DEV: 1708 if (be_lun->backend.dev.csw) { 1709 dev_relthread(be_lun->backend.dev.cdev, 1710 be_lun->backend.dev.dev_ref); 1711 be_lun->backend.dev.csw = NULL; 1712 be_lun->backend.dev.cdev = NULL; 1713 } 1714 break; 1715 case CTL_BE_BLOCK_FILE: 1716 break; 1717 case CTL_BE_BLOCK_NONE: 1718 break; 1719 default: 1720 panic("Unexpected backend type."); 1721 break; 1722 } 1723 1724 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1725 be_lun->vn = NULL; 1726 1727 switch (be_lun->dev_type) { 1728 case CTL_BE_BLOCK_DEV: 1729 break; 1730 case CTL_BE_BLOCK_FILE: 1731 if (be_lun->backend.file.cred != NULL) { 1732 crfree(be_lun->backend.file.cred); 1733 be_lun->backend.file.cred = NULL; 1734 } 1735 break; 1736 case CTL_BE_BLOCK_NONE: 1737 break; 1738 default: 1739 panic("Unexpected backend type."); 1740 break; 1741 } 1742 } 1743 PICKUP_GIANT(); 1744 1745 return (0); 1746} 1747 1748static int 1749ctl_be_block_open(struct ctl_be_block_softc *softc, 1750 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1751{ 1752 struct nameidata nd; 1753 int flags; 1754 int error; 1755 1756 /* 1757 * XXX KDM allow a read-only option? 1758 */ 1759 flags = FREAD | FWRITE; 1760 error = 0; 1761 1762 if (rootvnode == NULL) { 1763 snprintf(req->error_str, sizeof(req->error_str), 1764 "%s: Root filesystem is not mounted", __func__); 1765 return (1); 1766 } 1767 1768 if (!curthread->td_proc->p_fd->fd_cdir) { 1769 curthread->td_proc->p_fd->fd_cdir = rootvnode; 1770 VREF(rootvnode); 1771 } 1772 if (!curthread->td_proc->p_fd->fd_rdir) { 1773 curthread->td_proc->p_fd->fd_rdir = rootvnode; 1774 VREF(rootvnode); 1775 } 1776 if (!curthread->td_proc->p_fd->fd_jdir) { 1777 curthread->td_proc->p_fd->fd_jdir = rootvnode; 1778 VREF(rootvnode); 1779 } 1780 1781 again: 1782 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1783 error = vn_open(&nd, &flags, 0, NULL); 1784 if (error) { 1785 /* 1786 * This is the only reasonable guess we can make as far as 1787 * path if the user doesn't give us a fully qualified path. 1788 * If they want to specify a file, they need to specify the 1789 * full path. 1790 */ 1791 if (be_lun->dev_path[0] != '/') { 1792 char *dev_path = "/dev/"; 1793 char *dev_name; 1794 1795 /* Try adding device path at beginning of name */ 1796 dev_name = malloc(strlen(be_lun->dev_path) 1797 + strlen(dev_path) + 1, 1798 M_CTLBLK, M_WAITOK); 1799 if (dev_name) { 1800 sprintf(dev_name, "%s%s", dev_path, 1801 be_lun->dev_path); 1802 free(be_lun->dev_path, M_CTLBLK); 1803 be_lun->dev_path = dev_name; 1804 goto again; 1805 } 1806 } 1807 snprintf(req->error_str, sizeof(req->error_str), 1808 "%s: error opening %s", __func__, be_lun->dev_path); 1809 return (error); 1810 } 1811 1812 NDFREE(&nd, NDF_ONLY_PNBUF); 1813 1814 be_lun->vn = nd.ni_vp; 1815 1816 /* We only support disks and files. */ 1817 if (vn_isdisk(be_lun->vn, &error)) { 1818 error = ctl_be_block_open_dev(be_lun, req); 1819 } else if (be_lun->vn->v_type == VREG) { 1820 error = ctl_be_block_open_file(be_lun, req); 1821 } else { 1822 error = EINVAL; 1823 snprintf(req->error_str, sizeof(req->error_str), 1824 "%s is not a disk or plain file", be_lun->dev_path); 1825 } 1826 VOP_UNLOCK(be_lun->vn, 0); 1827 1828 if (error != 0) { 1829 ctl_be_block_close(be_lun); 1830 return (error); 1831 } 1832 1833 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1834 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1835 1836 return (0); 1837} 1838 1839static int 1840ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1841{ 1842 struct ctl_be_block_lun *be_lun; 1843 struct ctl_lun_create_params *params; 1844 char num_thread_str[16]; 1845 char tmpstr[32]; 1846 char *value; 1847 int retval, num_threads, unmap; 1848 int tmp_num_threads; 1849 1850 params = &req->reqdata.create; 1851 retval = 0; 1852 1853 num_threads = cbb_num_threads; 1854 1855 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1856 1857 be_lun->softc = softc; 1858 STAILQ_INIT(&be_lun->input_queue); 1859 STAILQ_INIT(&be_lun->config_write_queue); 1860 STAILQ_INIT(&be_lun->datamove_queue); 1861 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1862 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1863 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1864 ctl_init_opts(&be_lun->ctl_be_lun.options, 1865 req->num_be_args, req->kern_be_args); 1866 1867 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1868 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1869 1870 if (be_lun->lun_zone == NULL) { 1871 snprintf(req->error_str, sizeof(req->error_str), 1872 "%s: error allocating UMA zone", __func__); 1873 goto bailout_error; 1874 } 1875 1876 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1877 be_lun->ctl_be_lun.lun_type = params->device_type; 1878 else 1879 be_lun->ctl_be_lun.lun_type = T_DIRECT; 1880 1881 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1882 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1883 if (value == NULL) { 1884 snprintf(req->error_str, sizeof(req->error_str), 1885 "%s: no file argument specified", __func__); 1886 goto bailout_error; 1887 } 1888 be_lun->dev_path = strdup(value, M_CTLBLK); 1889 1890 retval = ctl_be_block_open(softc, be_lun, req); 1891 if (retval != 0) { 1892 retval = 0; 1893 goto bailout_error; 1894 } 1895 1896 /* 1897 * Tell the user the size of the file/device. 1898 */ 1899 params->lun_size_bytes = be_lun->size_bytes; 1900 1901 /* 1902 * The maximum LBA is the size - 1. 1903 */ 1904 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1905 } else { 1906 /* 1907 * For processor devices, we don't have any size. 1908 */ 1909 be_lun->blocksize = 0; 1910 be_lun->pblockexp = 0; 1911 be_lun->pblockoff = 0; 1912 be_lun->size_blocks = 0; 1913 be_lun->size_bytes = 0; 1914 be_lun->ctl_be_lun.maxlba = 0; 1915 params->lun_size_bytes = 0; 1916 1917 /* 1918 * Default to just 1 thread for processor devices. 1919 */ 1920 num_threads = 1; 1921 } 1922 1923 /* 1924 * XXX This searching loop might be refactored to be combined with 1925 * the loop above, 1926 */ 1927 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 1928 if (value != NULL) { 1929 tmp_num_threads = strtol(value, NULL, 0); 1930 1931 /* 1932 * We don't let the user specify less than one 1933 * thread, but hope he's clueful enough not to 1934 * specify 1000 threads. 1935 */ 1936 if (tmp_num_threads < 1) { 1937 snprintf(req->error_str, sizeof(req->error_str), 1938 "%s: invalid number of threads %s", 1939 __func__, num_thread_str); 1940 goto bailout_error; 1941 } 1942 num_threads = tmp_num_threads; 1943 } 1944 unmap = 0; 1945 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 1946 if (value != NULL && strcmp(value, "on") == 0) 1947 unmap = 1; 1948 1949 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1950 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1951 if (unmap) 1952 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 1953 be_lun->ctl_be_lun.be_lun = be_lun; 1954 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1955 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 1956 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 1957 /* Tell the user the blocksize we ended up using */ 1958 params->blocksize_bytes = be_lun->blocksize; 1959 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1960 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1961 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1962 } else 1963 be_lun->ctl_be_lun.req_lun_id = 0; 1964 1965 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1966 be_lun->ctl_be_lun.lun_config_status = 1967 ctl_be_block_lun_config_status; 1968 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1969 1970 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1971 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1972 softc->num_luns); 1973 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1974 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1975 sizeof(tmpstr))); 1976 1977 /* Tell the user what we used for a serial number */ 1978 strncpy((char *)params->serial_num, tmpstr, 1979 ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1980 } else { 1981 strncpy((char *)be_lun->ctl_be_lun.serial_num, 1982 params->serial_num, 1983 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1984 sizeof(params->serial_num))); 1985 } 1986 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1987 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1988 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1989 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1990 sizeof(tmpstr))); 1991 1992 /* Tell the user what we used for a device ID */ 1993 strncpy((char *)params->device_id, tmpstr, 1994 ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1995 } else { 1996 strncpy((char *)be_lun->ctl_be_lun.device_id, 1997 params->device_id, 1998 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1999 sizeof(params->device_id))); 2000 } 2001 2002 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2003 2004 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2005 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2006 2007 if (be_lun->io_taskqueue == NULL) { 2008 snprintf(req->error_str, sizeof(req->error_str), 2009 "%s: Unable to create taskqueue", __func__); 2010 goto bailout_error; 2011 } 2012 2013 /* 2014 * Note that we start the same number of threads by default for 2015 * both the file case and the block device case. For the file 2016 * case, we need multiple threads to allow concurrency, because the 2017 * vnode interface is designed to be a blocking interface. For the 2018 * block device case, ZFS zvols at least will block the caller's 2019 * context in many instances, and so we need multiple threads to 2020 * overcome that problem. Other block devices don't need as many 2021 * threads, but they shouldn't cause too many problems. 2022 * 2023 * If the user wants to just have a single thread for a block 2024 * device, he can specify that when the LUN is created, or change 2025 * the tunable/sysctl to alter the default number of threads. 2026 */ 2027 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2028 /*num threads*/num_threads, 2029 /*priority*/PWAIT, 2030 /*thread name*/ 2031 "%s taskq", be_lun->lunname); 2032 2033 if (retval != 0) 2034 goto bailout_error; 2035 2036 be_lun->num_threads = num_threads; 2037 2038 mtx_lock(&softc->lock); 2039 softc->num_luns++; 2040 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2041 2042 mtx_unlock(&softc->lock); 2043 2044 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2045 if (retval != 0) { 2046 mtx_lock(&softc->lock); 2047 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2048 links); 2049 softc->num_luns--; 2050 mtx_unlock(&softc->lock); 2051 snprintf(req->error_str, sizeof(req->error_str), 2052 "%s: ctl_add_lun() returned error %d, see dmesg for " 2053 "details", __func__, retval); 2054 retval = 0; 2055 goto bailout_error; 2056 } 2057 2058 mtx_lock(&softc->lock); 2059 2060 /* 2061 * Tell the config_status routine that we're waiting so it won't 2062 * clean up the LUN in the event of an error. 2063 */ 2064 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2065 2066 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2067 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2068 if (retval == EINTR) 2069 break; 2070 } 2071 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2072 2073 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2074 snprintf(req->error_str, sizeof(req->error_str), 2075 "%s: LUN configuration error, see dmesg for details", 2076 __func__); 2077 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2078 links); 2079 softc->num_luns--; 2080 mtx_unlock(&softc->lock); 2081 goto bailout_error; 2082 } else { 2083 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2084 } 2085 2086 mtx_unlock(&softc->lock); 2087 2088 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2089 be_lun->blocksize, 2090 DEVSTAT_ALL_SUPPORTED, 2091 be_lun->ctl_be_lun.lun_type 2092 | DEVSTAT_TYPE_IF_OTHER, 2093 DEVSTAT_PRIORITY_OTHER); 2094 2095 2096 req->status = CTL_LUN_OK; 2097 2098 return (retval); 2099 2100bailout_error: 2101 req->status = CTL_LUN_ERROR; 2102 2103 if (be_lun->io_taskqueue != NULL) 2104 taskqueue_free(be_lun->io_taskqueue); 2105 ctl_be_block_close(be_lun); 2106 if (be_lun->dev_path != NULL) 2107 free(be_lun->dev_path, M_CTLBLK); 2108 if (be_lun->lun_zone != NULL) 2109 uma_zdestroy(be_lun->lun_zone); 2110 ctl_free_opts(&be_lun->ctl_be_lun.options); 2111 mtx_destroy(&be_lun->queue_lock); 2112 mtx_destroy(&be_lun->io_lock); 2113 free(be_lun, M_CTLBLK); 2114 2115 return (retval); 2116} 2117 2118static int 2119ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2120{ 2121 struct ctl_lun_rm_params *params; 2122 struct ctl_be_block_lun *be_lun; 2123 int retval; 2124 2125 params = &req->reqdata.rm; 2126 2127 mtx_lock(&softc->lock); 2128 2129 be_lun = NULL; 2130 2131 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2132 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2133 break; 2134 } 2135 mtx_unlock(&softc->lock); 2136 2137 if (be_lun == NULL) { 2138 snprintf(req->error_str, sizeof(req->error_str), 2139 "%s: LUN %u is not managed by the block backend", 2140 __func__, params->lun_id); 2141 goto bailout_error; 2142 } 2143 2144 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2145 2146 if (retval != 0) { 2147 snprintf(req->error_str, sizeof(req->error_str), 2148 "%s: error %d returned from ctl_disable_lun() for " 2149 "LUN %d", __func__, retval, params->lun_id); 2150 goto bailout_error; 2151 2152 } 2153 2154 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2155 if (retval != 0) { 2156 snprintf(req->error_str, sizeof(req->error_str), 2157 "%s: error %d returned from ctl_invalidate_lun() for " 2158 "LUN %d", __func__, retval, params->lun_id); 2159 goto bailout_error; 2160 } 2161 2162 mtx_lock(&softc->lock); 2163 2164 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2165 2166 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2167 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2168 if (retval == EINTR) 2169 break; 2170 } 2171 2172 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2173 2174 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2175 snprintf(req->error_str, sizeof(req->error_str), 2176 "%s: interrupted waiting for LUN to be freed", 2177 __func__); 2178 mtx_unlock(&softc->lock); 2179 goto bailout_error; 2180 } 2181 2182 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2183 2184 softc->num_luns--; 2185 mtx_unlock(&softc->lock); 2186 2187 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2188 2189 taskqueue_free(be_lun->io_taskqueue); 2190 2191 ctl_be_block_close(be_lun); 2192 2193 if (be_lun->disk_stats != NULL) 2194 devstat_remove_entry(be_lun->disk_stats); 2195 2196 uma_zdestroy(be_lun->lun_zone); 2197 2198 ctl_free_opts(&be_lun->ctl_be_lun.options); 2199 free(be_lun->dev_path, M_CTLBLK); 2200 mtx_destroy(&be_lun->queue_lock); 2201 mtx_destroy(&be_lun->io_lock); 2202 free(be_lun, M_CTLBLK); 2203 2204 req->status = CTL_LUN_OK; 2205 2206 return (0); 2207 2208bailout_error: 2209 2210 req->status = CTL_LUN_ERROR; 2211 2212 return (0); 2213} 2214 2215static int 2216ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2217 struct ctl_lun_req *req) 2218{ 2219 struct vattr vattr; 2220 int error; 2221 struct ctl_lun_modify_params *params; 2222 2223 params = &req->reqdata.modify; 2224 2225 if (params->lun_size_bytes != 0) { 2226 be_lun->size_bytes = params->lun_size_bytes; 2227 } else { 2228 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2229 if (error != 0) { 2230 snprintf(req->error_str, sizeof(req->error_str), 2231 "error calling VOP_GETATTR() for file %s", 2232 be_lun->dev_path); 2233 return (error); 2234 } 2235 2236 be_lun->size_bytes = vattr.va_size; 2237 } 2238 2239 return (0); 2240} 2241 2242static int 2243ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2244 struct ctl_lun_req *req) 2245{ 2246 struct cdev *dev; 2247 struct cdevsw *devsw; 2248 int error; 2249 struct ctl_lun_modify_params *params; 2250 uint64_t size_bytes; 2251 2252 params = &req->reqdata.modify; 2253 2254 dev = be_lun->vn->v_rdev; 2255 devsw = dev->si_devsw; 2256 if (!devsw->d_ioctl) { 2257 snprintf(req->error_str, sizeof(req->error_str), 2258 "%s: no d_ioctl for device %s!", __func__, 2259 be_lun->dev_path); 2260 return (ENODEV); 2261 } 2262 2263 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2264 (caddr_t)&size_bytes, FREAD, 2265 curthread); 2266 if (error) { 2267 snprintf(req->error_str, sizeof(req->error_str), 2268 "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2269 "on %s!", __func__, error, be_lun->dev_path); 2270 return (error); 2271 } 2272 2273 if (params->lun_size_bytes != 0) { 2274 if (params->lun_size_bytes > size_bytes) { 2275 snprintf(req->error_str, sizeof(req->error_str), 2276 "%s: requested LUN size %ju > backing device " 2277 "size %ju", __func__, 2278 (uintmax_t)params->lun_size_bytes, 2279 (uintmax_t)size_bytes); 2280 return (EINVAL); 2281 } 2282 2283 be_lun->size_bytes = params->lun_size_bytes; 2284 } else { 2285 be_lun->size_bytes = size_bytes; 2286 } 2287 2288 return (0); 2289} 2290 2291static int 2292ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2293{ 2294 struct ctl_lun_modify_params *params; 2295 struct ctl_be_block_lun *be_lun; 2296 int error; 2297 2298 params = &req->reqdata.modify; 2299 2300 mtx_lock(&softc->lock); 2301 2302 be_lun = NULL; 2303 2304 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2305 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2306 break; 2307 } 2308 mtx_unlock(&softc->lock); 2309 2310 if (be_lun == NULL) { 2311 snprintf(req->error_str, sizeof(req->error_str), 2312 "%s: LUN %u is not managed by the block backend", 2313 __func__, params->lun_id); 2314 goto bailout_error; 2315 } 2316 2317 if (params->lun_size_bytes != 0) { 2318 if (params->lun_size_bytes < be_lun->blocksize) { 2319 snprintf(req->error_str, sizeof(req->error_str), 2320 "%s: LUN size %ju < blocksize %u", __func__, 2321 params->lun_size_bytes, be_lun->blocksize); 2322 goto bailout_error; 2323 } 2324 } 2325 2326 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2327 2328 if (be_lun->vn->v_type == VREG) 2329 error = ctl_be_block_modify_file(be_lun, req); 2330 else 2331 error = ctl_be_block_modify_dev(be_lun, req); 2332 2333 VOP_UNLOCK(be_lun->vn, 0); 2334 2335 if (error != 0) 2336 goto bailout_error; 2337 2338 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2339 2340 /* 2341 * The maximum LBA is the size - 1. 2342 * 2343 * XXX: Note that this field is being updated without locking, 2344 * which might cause problems on 32-bit architectures. 2345 */ 2346 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2347 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2348 2349 /* Tell the user the exact size we ended up using */ 2350 params->lun_size_bytes = be_lun->size_bytes; 2351 2352 req->status = CTL_LUN_OK; 2353 2354 return (0); 2355 2356bailout_error: 2357 req->status = CTL_LUN_ERROR; 2358 2359 return (0); 2360} 2361 2362static void 2363ctl_be_block_lun_shutdown(void *be_lun) 2364{ 2365 struct ctl_be_block_lun *lun; 2366 struct ctl_be_block_softc *softc; 2367 2368 lun = (struct ctl_be_block_lun *)be_lun; 2369 2370 softc = lun->softc; 2371 2372 mtx_lock(&softc->lock); 2373 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2374 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2375 wakeup(lun); 2376 mtx_unlock(&softc->lock); 2377 2378} 2379 2380static void 2381ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2382{ 2383 struct ctl_be_block_lun *lun; 2384 struct ctl_be_block_softc *softc; 2385 2386 lun = (struct ctl_be_block_lun *)be_lun; 2387 softc = lun->softc; 2388 2389 if (status == CTL_LUN_CONFIG_OK) { 2390 mtx_lock(&softc->lock); 2391 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2392 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2393 wakeup(lun); 2394 mtx_unlock(&softc->lock); 2395 2396 /* 2397 * We successfully added the LUN, attempt to enable it. 2398 */ 2399 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2400 printf("%s: ctl_enable_lun() failed!\n", __func__); 2401 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2402 printf("%s: ctl_invalidate_lun() failed!\n", 2403 __func__); 2404 } 2405 } 2406 2407 return; 2408 } 2409 2410 2411 mtx_lock(&softc->lock); 2412 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2413 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2414 wakeup(lun); 2415 mtx_unlock(&softc->lock); 2416} 2417 2418 2419static int 2420ctl_be_block_config_write(union ctl_io *io) 2421{ 2422 struct ctl_be_block_lun *be_lun; 2423 struct ctl_be_lun *ctl_be_lun; 2424 int retval; 2425 2426 retval = 0; 2427 2428 DPRINTF("entered\n"); 2429 2430 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2431 CTL_PRIV_BACKEND_LUN].ptr; 2432 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2433 2434 switch (io->scsiio.cdb[0]) { 2435 case SYNCHRONIZE_CACHE: 2436 case SYNCHRONIZE_CACHE_16: 2437 case WRITE_SAME_10: 2438 case WRITE_SAME_16: 2439 case UNMAP: 2440 /* 2441 * The upper level CTL code will filter out any CDBs with 2442 * the immediate bit set and return the proper error. 2443 * 2444 * We don't really need to worry about what LBA range the 2445 * user asked to be synced out. When they issue a sync 2446 * cache command, we'll sync out the whole thing. 2447 */ 2448 mtx_lock(&be_lun->queue_lock); 2449 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2450 links); 2451 mtx_unlock(&be_lun->queue_lock); 2452 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2453 break; 2454 case START_STOP_UNIT: { 2455 struct scsi_start_stop_unit *cdb; 2456 2457 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2458 2459 if (cdb->how & SSS_START) 2460 retval = ctl_start_lun(ctl_be_lun); 2461 else { 2462 retval = ctl_stop_lun(ctl_be_lun); 2463 /* 2464 * XXX KDM Copan-specific offline behavior. 2465 * Figure out a reasonable way to port this? 2466 */ 2467#ifdef NEEDTOPORT 2468 if ((retval == 0) 2469 && (cdb->byte2 & SSS_ONOFFLINE)) 2470 retval = ctl_lun_offline(ctl_be_lun); 2471#endif 2472 } 2473 2474 /* 2475 * In general, the above routines should not fail. They 2476 * just set state for the LUN. So we've got something 2477 * pretty wrong here if we can't start or stop the LUN. 2478 */ 2479 if (retval != 0) { 2480 ctl_set_internal_failure(&io->scsiio, 2481 /*sks_valid*/ 1, 2482 /*retry_count*/ 0xf051); 2483 retval = CTL_RETVAL_COMPLETE; 2484 } else { 2485 ctl_set_success(&io->scsiio); 2486 } 2487 ctl_config_write_done(io); 2488 break; 2489 } 2490 default: 2491 ctl_set_invalid_opcode(&io->scsiio); 2492 ctl_config_write_done(io); 2493 retval = CTL_RETVAL_COMPLETE; 2494 break; 2495 } 2496 2497 return (retval); 2498 2499} 2500 2501static int 2502ctl_be_block_config_read(union ctl_io *io) 2503{ 2504 return (0); 2505} 2506 2507static int 2508ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2509{ 2510 struct ctl_be_block_lun *lun; 2511 int retval; 2512 2513 lun = (struct ctl_be_block_lun *)be_lun; 2514 retval = 0; 2515 2516 retval = sbuf_printf(sb, "\t<num_threads>"); 2517 2518 if (retval != 0) 2519 goto bailout; 2520 2521 retval = sbuf_printf(sb, "%d", lun->num_threads); 2522 2523 if (retval != 0) 2524 goto bailout; 2525 2526 retval = sbuf_printf(sb, "</num_threads>\n"); 2527 2528bailout: 2529 2530 return (retval); 2531} 2532 2533int 2534ctl_be_block_init(void) 2535{ 2536 struct ctl_be_block_softc *softc; 2537 int retval; 2538 2539 softc = &backend_block_softc; 2540 retval = 0; 2541 2542 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2543 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2544 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2545 STAILQ_INIT(&softc->disk_list); 2546 STAILQ_INIT(&softc->lun_list); 2547 2548 return (retval); 2549} 2550