ctl_backend_block.c revision 267514
1178476Sjb/*- 2178476Sjb * Copyright (c) 2003 Silicon Graphics International Corp. 3178476Sjb * Copyright (c) 2009-2011 Spectra Logic Corporation 4178476Sjb * Copyright (c) 2012 The FreeBSD Foundation 5178476Sjb * All rights reserved. 6178476Sjb * 7178476Sjb * Portions of this software were developed by Edward Tomasz Napierala 8178476Sjb * under sponsorship from the FreeBSD Foundation. 9178476Sjb * 10178476Sjb * Redistribution and use in source and binary forms, with or without 11178476Sjb * modification, are permitted provided that the following conditions 12178476Sjb * are met: 13178476Sjb * 1. Redistributions of source code must retain the above copyright 14178476Sjb * notice, this list of conditions, and the following disclaimer, 15178476Sjb * without modification. 16178476Sjb * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17178476Sjb * substantially similar to the "NO WARRANTY" disclaimer below 18178476Sjb * ("Disclaimer") and any redistribution must be conditioned upon 19178476Sjb * including a substantially similar Disclaimer requirement for further 20178476Sjb * binary redistribution. 21178476Sjb * 22178476Sjb * NO WARRANTY 23178476Sjb * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24178476Sjb * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25178476Sjb * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26178476Sjb * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27178476Sjb * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28178476Sjb * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29178476Sjb * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30178476Sjb * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31178476Sjb * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32178476Sjb * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33178476Sjb * POSSIBILITY OF SUCH DAMAGES. 34178476Sjb * 35178476Sjb * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36178476Sjb */ 37178476Sjb/* 38178476Sjb * CAM Target Layer driver backend for block devices. 39178476Sjb * 40178476Sjb * Author: Ken Merry <ken@FreeBSD.org> 41178476Sjb */ 42178476Sjb#include <sys/cdefs.h> 43178476Sjb__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl_backend_block.c 267514 2014-06-15 17:14:52Z mav $"); 44178476Sjb 45178476Sjb#include <sys/param.h> 46178476Sjb#include <sys/systm.h> 47178476Sjb#include <sys/kernel.h> 48178476Sjb#include <sys/types.h> 49178476Sjb#include <sys/kthread.h> 50178476Sjb#include <sys/bio.h> 51178476Sjb#include <sys/fcntl.h> 52178476Sjb#include <sys/limits.h> 53178476Sjb#include <sys/lock.h> 54178476Sjb#include <sys/mutex.h> 55178476Sjb#include <sys/condvar.h> 56178476Sjb#include <sys/malloc.h> 57178476Sjb#include <sys/conf.h> 58178476Sjb#include <sys/ioccom.h> 59178476Sjb#include <sys/queue.h> 60178476Sjb#include <sys/sbuf.h> 61178476Sjb#include <sys/endian.h> 62178476Sjb#include <sys/uio.h> 63178476Sjb#include <sys/buf.h> 64178476Sjb#include <sys/taskqueue.h> 65178476Sjb#include <sys/vnode.h> 66178476Sjb#include <sys/namei.h> 67178476Sjb#include <sys/mount.h> 68178476Sjb#include <sys/disk.h> 69178476Sjb#include <sys/fcntl.h> 70178476Sjb#include <sys/filedesc.h> 71178476Sjb#include <sys/proc.h> 72178476Sjb#include <sys/pcpu.h> 73178476Sjb#include <sys/module.h> 74178476Sjb#include <sys/sdt.h> 75#include <sys/devicestat.h> 76#include <sys/sysctl.h> 77 78#include <geom/geom.h> 79 80#include <cam/cam.h> 81#include <cam/scsi/scsi_all.h> 82#include <cam/scsi/scsi_da.h> 83#include <cam/ctl/ctl_io.h> 84#include <cam/ctl/ctl.h> 85#include <cam/ctl/ctl_backend.h> 86#include <cam/ctl/ctl_frontend_internal.h> 87#include <cam/ctl/ctl_ioctl.h> 88#include <cam/ctl/ctl_scsi_all.h> 89#include <cam/ctl/ctl_error.h> 90 91/* 92 * The idea here is that we'll allocate enough S/G space to hold a 1MB 93 * I/O. If we get an I/O larger than that, we'll split it. 94 */ 95#define CTLBLK_MAX_IO_SIZE (1024 * 1024) 96#define CTLBLK_MAX_SEG MAXPHYS 97#define CTLBLK_MAX_SEGS MAX(CTLBLK_MAX_IO_SIZE / CTLBLK_MAX_SEG, 1) 98 99#ifdef CTLBLK_DEBUG 100#define DPRINTF(fmt, args...) \ 101 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 102#else 103#define DPRINTF(fmt, args...) do {} while(0) 104#endif 105 106SDT_PROVIDER_DEFINE(cbb); 107 108typedef enum { 109 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 110 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 111 CTL_BE_BLOCK_LUN_WAITING = 0x04, 112 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 113} ctl_be_block_lun_flags; 114 115typedef enum { 116 CTL_BE_BLOCK_NONE, 117 CTL_BE_BLOCK_DEV, 118 CTL_BE_BLOCK_FILE 119} ctl_be_block_type; 120 121struct ctl_be_block_devdata { 122 struct cdev *cdev; 123 struct cdevsw *csw; 124 int dev_ref; 125}; 126 127struct ctl_be_block_filedata { 128 struct ucred *cred; 129}; 130 131union ctl_be_block_bedata { 132 struct ctl_be_block_devdata dev; 133 struct ctl_be_block_filedata file; 134}; 135 136struct ctl_be_block_io; 137struct ctl_be_block_lun; 138 139typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 140 struct ctl_be_block_io *beio); 141 142/* 143 * Backend LUN structure. There is a 1:1 mapping between a block device 144 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 145 */ 146struct ctl_be_block_lun { 147 struct ctl_block_disk *disk; 148 char lunname[32]; 149 char *dev_path; 150 ctl_be_block_type dev_type; 151 struct vnode *vn; 152 union ctl_be_block_bedata backend; 153 cbb_dispatch_t dispatch; 154 cbb_dispatch_t lun_flush; 155 cbb_dispatch_t unmap; 156 struct mtx lock; 157 uma_zone_t lun_zone; 158 uint64_t size_blocks; 159 uint64_t size_bytes; 160 uint32_t blocksize; 161 int blocksize_shift; 162 uint16_t pblockexp; 163 uint16_t pblockoff; 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 ctl_be_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_write_queue; 174 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 175}; 176 177/* 178 * Overall softc structure for the block backend module. 179 */ 180struct ctl_be_block_softc { 181 struct mtx lock; 182 int num_disks; 183 STAILQ_HEAD(, ctl_block_disk) disk_list; 184 int num_luns; 185 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 186}; 187 188static struct ctl_be_block_softc backend_block_softc; 189 190/* 191 * Per-I/O information. 192 */ 193struct ctl_be_block_io { 194 union ctl_io *io; 195 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 196 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 197 int bio_cmd; 198 int bio_flags; 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 struct ctl_be_block_softc *softc; 210 struct ctl_be_block_lun *lun; 211 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 212}; 213 214static int cbb_num_threads = 14; 215TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 216SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 217 "CAM Target Layer Block Backend"); 218SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 219 &cbb_num_threads, 0, "Number of threads per backing file"); 220 221static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 222static void ctl_free_beio(struct ctl_be_block_io *beio); 223static void ctl_complete_beio(struct ctl_be_block_io *beio); 224static int ctl_be_block_move_done(union ctl_io *io); 225static void ctl_be_block_biodone(struct bio *bio); 226static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 227 struct ctl_be_block_io *beio); 228static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 229 struct ctl_be_block_io *beio); 230static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 231 struct ctl_be_block_io *beio); 232static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 233 struct ctl_be_block_io *beio); 234static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 235 struct ctl_be_block_io *beio); 236static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 237 union ctl_io *io); 238static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 239 union ctl_io *io); 240static void ctl_be_block_worker(void *context, int pending); 241static int ctl_be_block_submit(union ctl_io *io); 242static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 243 int flag, struct thread *td); 244static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 245 struct ctl_lun_req *req); 246static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 247 struct ctl_lun_req *req); 248static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 249static int ctl_be_block_open(struct ctl_be_block_softc *softc, 250 struct ctl_be_block_lun *be_lun, 251 struct ctl_lun_req *req); 252static int ctl_be_block_create(struct ctl_be_block_softc *softc, 253 struct ctl_lun_req *req); 254static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 255 struct ctl_lun_req *req); 256static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 257 struct ctl_lun_req *req); 258static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 259 struct ctl_lun_req *req); 260static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 261 struct ctl_lun_req *req); 262static void ctl_be_block_lun_shutdown(void *be_lun); 263static void ctl_be_block_lun_config_status(void *be_lun, 264 ctl_lun_config_status status); 265static int ctl_be_block_config_write(union ctl_io *io); 266static int ctl_be_block_config_read(union ctl_io *io); 267static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 268int ctl_be_block_init(void); 269 270static struct ctl_backend_driver ctl_be_block_driver = 271{ 272 .name = "block", 273 .flags = CTL_BE_FLAG_HAS_CONFIG, 274 .init = ctl_be_block_init, 275 .data_submit = ctl_be_block_submit, 276 .data_move_done = ctl_be_block_move_done, 277 .config_read = ctl_be_block_config_read, 278 .config_write = ctl_be_block_config_write, 279 .ioctl = ctl_be_block_ioctl, 280 .lun_info = ctl_be_block_lun_info 281}; 282 283MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 284CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 285 286static uma_zone_t beio_zone; 287 288static struct ctl_be_block_io * 289ctl_alloc_beio(struct ctl_be_block_softc *softc) 290{ 291 struct ctl_be_block_io *beio; 292 293 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 294 beio->softc = softc; 295 return (beio); 296} 297 298static void 299ctl_free_beio(struct ctl_be_block_io *beio) 300{ 301 int duplicate_free; 302 int i; 303 304 duplicate_free = 0; 305 306 for (i = 0; i < beio->num_segs; i++) { 307 if (beio->sg_segs[i].addr == NULL) 308 duplicate_free++; 309 310 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 311 beio->sg_segs[i].addr = NULL; 312 } 313 314 if (duplicate_free > 0) { 315 printf("%s: %d duplicate frees out of %d segments\n", __func__, 316 duplicate_free, beio->num_segs); 317 } 318 319 uma_zfree(beio_zone, beio); 320} 321 322static void 323ctl_complete_beio(struct ctl_be_block_io *beio) 324{ 325 union ctl_io *io; 326 int io_len; 327 328 io = beio->io; 329 330 if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) 331 io_len = beio->io_len; 332 else 333 io_len = 0; 334 335 devstat_end_transaction(beio->lun->disk_stats, 336 /*bytes*/ io_len, 337 beio->ds_tag_type, 338 beio->ds_trans_type, 339 /*now*/ NULL, 340 /*then*/&beio->ds_t0); 341 342 if (beio->beio_cont != NULL) { 343 beio->beio_cont(beio); 344 } else { 345 ctl_free_beio(beio); 346 ctl_done(io); 347 } 348} 349 350static int 351ctl_be_block_move_done(union ctl_io *io) 352{ 353 struct ctl_be_block_io *beio; 354 struct ctl_be_block_lun *be_lun; 355#ifdef CTL_TIME_IO 356 struct bintime cur_bt; 357#endif 358 359 beio = (struct ctl_be_block_io *) 360 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 361 362 be_lun = beio->lun; 363 364 DPRINTF("entered\n"); 365 366#ifdef CTL_TIME_IO 367 getbintime(&cur_bt); 368 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 369 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 370 io->io_hdr.num_dmas++; 371#endif 372 373 /* 374 * We set status at this point for read commands, and write 375 * commands with errors. 376 */ 377 if ((beio->bio_cmd == BIO_READ) 378 && (io->io_hdr.port_status == 0) 379 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 380 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) 381 ctl_set_success(&io->scsiio); 382 else if ((io->io_hdr.port_status != 0) 383 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 384 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 385 /* 386 * For hardware error sense keys, the sense key 387 * specific value is defined to be a retry count, 388 * but we use it to pass back an internal FETD 389 * error code. XXX KDM Hopefully the FETD is only 390 * using 16 bits for an error code, since that's 391 * all the space we have in the sks field. 392 */ 393 ctl_set_internal_failure(&io->scsiio, 394 /*sks_valid*/ 1, 395 /*retry_count*/ 396 io->io_hdr.port_status); 397 } 398 399 /* 400 * If this is a read, or a write with errors, it is done. 401 */ 402 if ((beio->bio_cmd == BIO_READ) 403 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 404 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 405 ctl_complete_beio(beio); 406 return (0); 407 } 408 409 /* 410 * At this point, we have a write and the DMA completed 411 * successfully. We now have to queue it to the task queue to 412 * execute the backend I/O. That is because we do blocking 413 * memory allocations, and in the file backing case, blocking I/O. 414 * This move done routine is generally called in the SIM's 415 * interrupt context, and therefore we cannot block. 416 */ 417 mtx_lock(&be_lun->lock); 418 /* 419 * XXX KDM make sure that links is okay to use at this point. 420 * Otherwise, we either need to add another field to ctl_io_hdr, 421 * or deal with resource allocation here. 422 */ 423 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 424 mtx_unlock(&be_lun->lock); 425 426 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 427 428 return (0); 429} 430 431static void 432ctl_be_block_biodone(struct bio *bio) 433{ 434 struct ctl_be_block_io *beio; 435 struct ctl_be_block_lun *be_lun; 436 union ctl_io *io; 437 int error; 438 439 beio = bio->bio_caller1; 440 be_lun = beio->lun; 441 io = beio->io; 442 443 DPRINTF("entered\n"); 444 445 error = bio->bio_error; 446 mtx_lock(&be_lun->lock); 447 if (error != 0) 448 beio->num_errors++; 449 450 beio->num_bios_done++; 451 452 /* 453 * XXX KDM will this cause WITNESS to complain? Holding a lock 454 * during the free might cause it to complain. 455 */ 456 g_destroy_bio(bio); 457 458 /* 459 * If the send complete bit isn't set, or we aren't the last I/O to 460 * complete, then we're done. 461 */ 462 if ((beio->send_complete == 0) 463 || (beio->num_bios_done < beio->num_bios_sent)) { 464 mtx_unlock(&be_lun->lock); 465 return; 466 } 467 468 /* 469 * At this point, we've verified that we are the last I/O to 470 * complete, so it's safe to drop the lock. 471 */ 472 mtx_unlock(&be_lun->lock); 473 474 /* 475 * If there are any errors from the backing device, we fail the 476 * entire I/O with a medium error. 477 */ 478 if (beio->num_errors > 0) { 479 if (error == EOPNOTSUPP) { 480 ctl_set_invalid_opcode(&io->scsiio); 481 } else if (beio->bio_cmd == BIO_FLUSH) { 482 /* XXX KDM is there is a better error here? */ 483 ctl_set_internal_failure(&io->scsiio, 484 /*sks_valid*/ 1, 485 /*retry_count*/ 0xbad2); 486 } else 487 ctl_set_medium_error(&io->scsiio); 488 ctl_complete_beio(beio); 489 return; 490 } 491 492 /* 493 * If this is a write, a flush or a delete, we're all done. 494 * If this is a read, we can now send the data to the user. 495 */ 496 if ((beio->bio_cmd == BIO_WRITE) 497 || (beio->bio_cmd == BIO_FLUSH) 498 || (beio->bio_cmd == BIO_DELETE)) { 499 ctl_set_success(&io->scsiio); 500 ctl_complete_beio(beio); 501 } else { 502#ifdef CTL_TIME_IO 503 getbintime(&io->io_hdr.dma_start_bt); 504#endif 505 ctl_datamove(io); 506 } 507} 508 509static void 510ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 511 struct ctl_be_block_io *beio) 512{ 513 union ctl_io *io; 514 struct mount *mountpoint; 515 int error, lock_flags; 516 517 DPRINTF("entered\n"); 518 519 io = beio->io; 520 521 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 522 523 if (MNT_SHARED_WRITES(mountpoint) 524 || ((mountpoint == NULL) 525 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 526 lock_flags = LK_SHARED; 527 else 528 lock_flags = LK_EXCLUSIVE; 529 530 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 531 532 binuptime(&beio->ds_t0); 533 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 534 535 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 536 VOP_UNLOCK(be_lun->vn, 0); 537 538 vn_finished_write(mountpoint); 539 540 if (error == 0) 541 ctl_set_success(&io->scsiio); 542 else { 543 /* XXX KDM is there is a better error here? */ 544 ctl_set_internal_failure(&io->scsiio, 545 /*sks_valid*/ 1, 546 /*retry_count*/ 0xbad1); 547 } 548 549 ctl_complete_beio(beio); 550} 551 552SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 553SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 554SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 555SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 556 557static void 558ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 559 struct ctl_be_block_io *beio) 560{ 561 struct ctl_be_block_filedata *file_data; 562 union ctl_io *io; 563 struct uio xuio; 564 struct iovec *xiovec; 565 int flags; 566 int error, i; 567 568 DPRINTF("entered\n"); 569 570 file_data = &be_lun->backend.file; 571 io = beio->io; 572 flags = beio->bio_flags; 573 574 if (beio->bio_cmd == BIO_READ) { 575 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 576 } else { 577 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 578 } 579 580 bzero(&xuio, sizeof(xuio)); 581 if (beio->bio_cmd == BIO_READ) 582 xuio.uio_rw = UIO_READ; 583 else 584 xuio.uio_rw = UIO_WRITE; 585 586 xuio.uio_offset = beio->io_offset; 587 xuio.uio_resid = beio->io_len; 588 xuio.uio_segflg = UIO_SYSSPACE; 589 xuio.uio_iov = beio->xiovecs; 590 xuio.uio_iovcnt = beio->num_segs; 591 xuio.uio_td = curthread; 592 593 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 594 xiovec->iov_base = beio->sg_segs[i].addr; 595 xiovec->iov_len = beio->sg_segs[i].len; 596 } 597 598 if (beio->bio_cmd == BIO_READ) { 599 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 600 601 binuptime(&beio->ds_t0); 602 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 603 604 /* 605 * UFS pays attention to IO_DIRECT for reads. If the 606 * DIRECTIO option is configured into the kernel, it calls 607 * ffs_rawread(). But that only works for single-segment 608 * uios with user space addresses. In our case, with a 609 * kernel uio, it still reads into the buffer cache, but it 610 * will just try to release the buffer from the cache later 611 * on in ffs_read(). 612 * 613 * ZFS does not pay attention to IO_DIRECT for reads. 614 * 615 * UFS does not pay attention to IO_SYNC for reads. 616 * 617 * ZFS pays attention to IO_SYNC (which translates into the 618 * Solaris define FRSYNC for zfs_read()) for reads. It 619 * attempts to sync the file before reading. 620 * 621 * So, to attempt to provide some barrier semantics in the 622 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 623 */ 624 error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 625 (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 626 627 VOP_UNLOCK(be_lun->vn, 0); 628 } else { 629 struct mount *mountpoint; 630 int lock_flags; 631 632 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 633 634 if (MNT_SHARED_WRITES(mountpoint) 635 || ((mountpoint == NULL) 636 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 637 lock_flags = LK_SHARED; 638 else 639 lock_flags = LK_EXCLUSIVE; 640 641 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 642 643 binuptime(&beio->ds_t0); 644 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 645 646 /* 647 * UFS pays attention to IO_DIRECT for writes. The write 648 * is done asynchronously. (Normally the write would just 649 * get put into cache. 650 * 651 * UFS pays attention to IO_SYNC for writes. It will 652 * attempt to write the buffer out synchronously if that 653 * flag is set. 654 * 655 * ZFS does not pay attention to IO_DIRECT for writes. 656 * 657 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 658 * for writes. It will flush the transaction from the 659 * cache before returning. 660 * 661 * So if we've got the BIO_ORDERED flag set, we want 662 * IO_SYNC in either the UFS or ZFS case. 663 */ 664 error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 665 IO_SYNC : 0, file_data->cred); 666 VOP_UNLOCK(be_lun->vn, 0); 667 668 vn_finished_write(mountpoint); 669 } 670 671 /* 672 * If we got an error, set the sense data to "MEDIUM ERROR" and 673 * return the I/O to the user. 674 */ 675 if (error != 0) { 676 char path_str[32]; 677 678 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 679 /* 680 * XXX KDM ZFS returns ENOSPC when the underlying 681 * filesystem fills up. What kind of SCSI error should we 682 * return for that? 683 */ 684 printf("%s%s command returned errno %d\n", path_str, 685 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 686 ctl_set_medium_error(&io->scsiio); 687 ctl_complete_beio(beio); 688 return; 689 } 690 691 /* 692 * If this is a write, we're all done. 693 * If this is a read, we can now send the data to the user. 694 */ 695 if (beio->bio_cmd == BIO_WRITE) { 696 ctl_set_success(&io->scsiio); 697 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 698 ctl_complete_beio(beio); 699 } else { 700 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 701#ifdef CTL_TIME_IO 702 getbintime(&io->io_hdr.dma_start_bt); 703#endif 704 ctl_datamove(io); 705 } 706} 707 708static void 709ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 710 struct ctl_be_block_io *beio) 711{ 712 struct bio *bio; 713 union ctl_io *io; 714 struct ctl_be_block_devdata *dev_data; 715 716 dev_data = &be_lun->backend.dev; 717 io = beio->io; 718 719 DPRINTF("entered\n"); 720 721 /* This can't fail, it's a blocking allocation. */ 722 bio = g_alloc_bio(); 723 724 bio->bio_cmd = BIO_FLUSH; 725 bio->bio_flags |= BIO_ORDERED; 726 bio->bio_dev = dev_data->cdev; 727 bio->bio_offset = 0; 728 bio->bio_data = 0; 729 bio->bio_done = ctl_be_block_biodone; 730 bio->bio_caller1 = beio; 731 bio->bio_pblkno = 0; 732 733 /* 734 * We don't need to acquire the LUN lock here, because we are only 735 * sending one bio, and so there is no other context to synchronize 736 * with. 737 */ 738 beio->num_bios_sent = 1; 739 beio->send_complete = 1; 740 741 binuptime(&beio->ds_t0); 742 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 743 744 (*dev_data->csw->d_strategy)(bio); 745} 746 747static void 748ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 749 struct ctl_be_block_io *beio, 750 uint64_t off, uint64_t len, int last) 751{ 752 struct bio *bio; 753 struct ctl_be_block_devdata *dev_data; 754 uint64_t maxlen; 755 756 dev_data = &be_lun->backend.dev; 757 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 758 while (len > 0) { 759 bio = g_alloc_bio(); 760 bio->bio_cmd = BIO_DELETE; 761 bio->bio_flags |= beio->bio_flags; 762 bio->bio_dev = dev_data->cdev; 763 bio->bio_offset = off; 764 bio->bio_length = MIN(len, maxlen); 765 bio->bio_data = 0; 766 bio->bio_done = ctl_be_block_biodone; 767 bio->bio_caller1 = beio; 768 bio->bio_pblkno = off / be_lun->blocksize; 769 770 off += bio->bio_length; 771 len -= bio->bio_length; 772 773 mtx_lock(&be_lun->lock); 774 beio->num_bios_sent++; 775 if (last && len == 0) 776 beio->send_complete = 1; 777 mtx_unlock(&be_lun->lock); 778 779 (*dev_data->csw->d_strategy)(bio); 780 } 781} 782 783static void 784ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 785 struct ctl_be_block_io *beio) 786{ 787 union ctl_io *io; 788 struct ctl_be_block_devdata *dev_data; 789 struct ctl_ptr_len_flags ptrlen; 790 struct scsi_unmap_desc *buf, *end; 791 uint64_t len; 792 793 dev_data = &be_lun->backend.dev; 794 io = beio->io; 795 796 DPRINTF("entered\n"); 797 798 binuptime(&beio->ds_t0); 799 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 800 801 if (beio->io_offset == -1) { 802 beio->io_len = 0; 803 memcpy(&ptrlen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 804 sizeof(ptrlen)); 805 buf = (struct scsi_unmap_desc *)ptrlen.ptr; 806 end = buf + ptrlen.len / sizeof(*buf); 807 for (; buf < end; buf++) { 808 len = (uint64_t)scsi_4btoul(buf->length) * 809 be_lun->blocksize; 810 beio->io_len += len; 811 ctl_be_block_unmap_dev_range(be_lun, beio, 812 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 813 (end - buf < 2) ? TRUE : FALSE); 814 } 815 } else 816 ctl_be_block_unmap_dev_range(be_lun, beio, 817 beio->io_offset, beio->io_len, TRUE); 818} 819 820static void 821ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 822 struct ctl_be_block_io *beio) 823{ 824 int i; 825 struct bio *bio; 826 struct ctl_be_block_devdata *dev_data; 827 off_t cur_offset; 828 int max_iosize; 829 830 DPRINTF("entered\n"); 831 832 dev_data = &be_lun->backend.dev; 833 834 /* 835 * We have to limit our I/O size to the maximum supported by the 836 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 837 * set it properly, use DFLTPHYS. 838 */ 839 max_iosize = dev_data->cdev->si_iosize_max; 840 if (max_iosize < PAGE_SIZE) 841 max_iosize = DFLTPHYS; 842 843 cur_offset = beio->io_offset; 844 845 /* 846 * XXX KDM need to accurately reflect the number of I/Os outstanding 847 * to a device. 848 */ 849 binuptime(&beio->ds_t0); 850 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 851 852 for (i = 0; i < beio->num_segs; i++) { 853 size_t cur_size; 854 uint8_t *cur_ptr; 855 856 cur_size = beio->sg_segs[i].len; 857 cur_ptr = beio->sg_segs[i].addr; 858 859 while (cur_size > 0) { 860 /* This can't fail, it's a blocking allocation. */ 861 bio = g_alloc_bio(); 862 863 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 864 865 bio->bio_cmd = beio->bio_cmd; 866 bio->bio_flags |= beio->bio_flags; 867 bio->bio_dev = dev_data->cdev; 868 bio->bio_caller1 = beio; 869 bio->bio_length = min(cur_size, max_iosize); 870 bio->bio_offset = cur_offset; 871 bio->bio_data = cur_ptr; 872 bio->bio_done = ctl_be_block_biodone; 873 bio->bio_pblkno = cur_offset / be_lun->blocksize; 874 875 cur_offset += bio->bio_length; 876 cur_ptr += bio->bio_length; 877 cur_size -= bio->bio_length; 878 879 /* 880 * Make sure we set the complete bit just before we 881 * issue the last bio so we don't wind up with a 882 * race. 883 * 884 * Use the LUN mutex here instead of a combination 885 * of atomic variables for simplicity. 886 * 887 * XXX KDM we could have a per-IO lock, but that 888 * would cause additional per-IO setup and teardown 889 * overhead. Hopefully there won't be too much 890 * contention on the LUN lock. 891 */ 892 mtx_lock(&be_lun->lock); 893 894 beio->num_bios_sent++; 895 896 if ((i == beio->num_segs - 1) 897 && (cur_size == 0)) 898 beio->send_complete = 1; 899 900 mtx_unlock(&be_lun->lock); 901 902 (*dev_data->csw->d_strategy)(bio); 903 } 904 } 905} 906 907static void 908ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 909{ 910 union ctl_io *io; 911 912 io = beio->io; 913 ctl_free_beio(beio); 914 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) 915 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 916 ctl_config_write_done(io); 917 return; 918 } 919 920 ctl_be_block_config_write(io); 921} 922 923static void 924ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 925 union ctl_io *io) 926{ 927 struct ctl_be_block_io *beio; 928 struct ctl_be_block_softc *softc; 929 struct ctl_lba_len_flags lbalen; 930 uint64_t len_left, lba; 931 int i, seglen; 932 uint8_t *buf, *end; 933 934 DPRINTF("entered\n"); 935 936 beio = io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 937 softc = be_lun->softc; 938 memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 939 sizeof(lbalen)); 940 941 if (lbalen.flags & ~(SWS_LBDATA | SWS_UNMAP) || 942 (lbalen.flags & SWS_UNMAP && be_lun->unmap == NULL)) { 943 ctl_free_beio(beio); 944 ctl_set_invalid_field(&io->scsiio, 945 /*sks_valid*/ 1, 946 /*command*/ 1, 947 /*field*/ 1, 948 /*bit_valid*/ 0, 949 /*bit*/ 0); 950 ctl_config_write_done(io); 951 return; 952 } 953 954 /* 955 * If the I/O came down with an ordered or head of queue tag, set 956 * the BIO_ORDERED attribute. For head of queue tags, that's 957 * pretty much the best we can do. 958 */ 959 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 960 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 961 beio->bio_flags = BIO_ORDERED; 962 963 switch (io->scsiio.tag_type) { 964 case CTL_TAG_ORDERED: 965 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 966 break; 967 case CTL_TAG_HEAD_OF_QUEUE: 968 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 969 break; 970 case CTL_TAG_UNTAGGED: 971 case CTL_TAG_SIMPLE: 972 case CTL_TAG_ACA: 973 default: 974 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 975 break; 976 } 977 978 if (lbalen.flags & SWS_UNMAP) { 979 beio->io_offset = lbalen.lba * be_lun->blocksize; 980 beio->io_len = (uint64_t)lbalen.len * be_lun->blocksize; 981 beio->bio_cmd = BIO_DELETE; 982 beio->ds_trans_type = DEVSTAT_FREE; 983 984 be_lun->unmap(be_lun, beio); 985 return; 986 } 987 988 beio->bio_cmd = BIO_WRITE; 989 beio->ds_trans_type = DEVSTAT_WRITE; 990 991 DPRINTF("WRITE SAME at LBA %jx len %u\n", 992 (uintmax_t)lbalen.lba, lbalen.len); 993 994 len_left = (uint64_t)lbalen.len * be_lun->blocksize; 995 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 996 997 /* 998 * Setup the S/G entry for this chunk. 999 */ 1000 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1001 seglen -= seglen % be_lun->blocksize; 1002 beio->sg_segs[i].len = seglen; 1003 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1004 1005 DPRINTF("segment %d addr %p len %zd\n", i, 1006 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1007 1008 beio->num_segs++; 1009 len_left -= seglen; 1010 1011 buf = beio->sg_segs[i].addr; 1012 end = buf + seglen; 1013 for (; buf < end; buf += be_lun->blocksize) { 1014 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1015 if (lbalen.flags & SWS_LBDATA) 1016 scsi_ulto4b(lbalen.lba + lba, buf); 1017 lba++; 1018 } 1019 } 1020 1021 beio->io_offset = lbalen.lba * be_lun->blocksize; 1022 beio->io_len = lba * be_lun->blocksize; 1023 1024 /* We can not do all in one run. Correct and schedule rerun. */ 1025 if (len_left > 0) { 1026 lbalen.lba += lba; 1027 lbalen.len -= lba; 1028 memcpy(io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen, 1029 sizeof(lbalen)); 1030 beio->beio_cont = ctl_be_block_cw_done_ws; 1031 } 1032 1033 be_lun->dispatch(be_lun, beio); 1034} 1035 1036static void 1037ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1038 union ctl_io *io) 1039{ 1040 struct ctl_be_block_io *beio; 1041 struct ctl_be_block_softc *softc; 1042 struct ctl_ptr_len_flags ptrlen; 1043 1044 DPRINTF("entered\n"); 1045 1046 beio = io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1047 softc = be_lun->softc; 1048 memcpy(&ptrlen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 1049 sizeof(ptrlen)); 1050 1051 if (ptrlen.flags != 0 || be_lun->unmap == NULL) { 1052 ctl_free_beio(beio); 1053 ctl_set_invalid_field(&io->scsiio, 1054 /*sks_valid*/ 0, 1055 /*command*/ 1, 1056 /*field*/ 0, 1057 /*bit_valid*/ 0, 1058 /*bit*/ 0); 1059 ctl_config_write_done(io); 1060 return; 1061 } 1062 1063 /* 1064 * If the I/O came down with an ordered or head of queue tag, set 1065 * the BIO_ORDERED attribute. For head of queue tags, that's 1066 * pretty much the best we can do. 1067 */ 1068 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1069 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1070 beio->bio_flags = BIO_ORDERED; 1071 1072 switch (io->scsiio.tag_type) { 1073 case CTL_TAG_ORDERED: 1074 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1075 break; 1076 case CTL_TAG_HEAD_OF_QUEUE: 1077 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1078 break; 1079 case CTL_TAG_UNTAGGED: 1080 case CTL_TAG_SIMPLE: 1081 case CTL_TAG_ACA: 1082 default: 1083 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1084 break; 1085 } 1086 1087 beio->io_len = 0; 1088 beio->io_offset = -1; 1089 1090 beio->bio_cmd = BIO_DELETE; 1091 beio->ds_trans_type = DEVSTAT_FREE; 1092 1093 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1094 (uintmax_t)lbalen.lba, lbalen.len); 1095 1096 be_lun->unmap(be_lun, beio); 1097} 1098 1099static void 1100ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1101{ 1102 union ctl_io *io; 1103 1104 io = beio->io; 1105 ctl_free_beio(beio); 1106 ctl_config_write_done(io); 1107} 1108 1109static void 1110ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1111 union ctl_io *io) 1112{ 1113 struct ctl_be_block_io *beio; 1114 struct ctl_be_block_softc *softc; 1115 1116 DPRINTF("entered\n"); 1117 1118 softc = be_lun->softc; 1119 beio = ctl_alloc_beio(softc); 1120 beio->io = io; 1121 beio->lun = be_lun; 1122 beio->beio_cont = ctl_be_block_cw_done; 1123 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 1124 1125 switch (io->scsiio.cdb[0]) { 1126 case SYNCHRONIZE_CACHE: 1127 case SYNCHRONIZE_CACHE_16: 1128 beio->bio_cmd = BIO_FLUSH; 1129 beio->ds_trans_type = DEVSTAT_NO_DATA; 1130 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1131 beio->io_len = 0; 1132 be_lun->lun_flush(be_lun, beio); 1133 break; 1134 case WRITE_SAME_10: 1135 case WRITE_SAME_16: 1136 ctl_be_block_cw_dispatch_ws(be_lun, io); 1137 break; 1138 case UNMAP: 1139 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1140 break; 1141 default: 1142 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1143 break; 1144 } 1145} 1146 1147SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1148SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1149SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1150SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1151 1152static void 1153ctl_be_block_next(struct ctl_be_block_io *beio) 1154{ 1155 struct ctl_be_block_lun *be_lun; 1156 union ctl_io *io; 1157 1158 io = beio->io; 1159 be_lun = beio->lun; 1160 ctl_free_beio(beio); 1161 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) 1162 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1163 ctl_done(io); 1164 return; 1165 } 1166 1167 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 1168 io->io_hdr.status &= ~CTL_STATUS_MASK; 1169 io->io_hdr.status |= CTL_STATUS_NONE; 1170 1171 mtx_lock(&be_lun->lock); 1172 /* 1173 * XXX KDM make sure that links is okay to use at this point. 1174 * Otherwise, we either need to add another field to ctl_io_hdr, 1175 * or deal with resource allocation here. 1176 */ 1177 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1178 mtx_unlock(&be_lun->lock); 1179 1180 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1181} 1182 1183static void 1184ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1185 union ctl_io *io) 1186{ 1187 struct ctl_be_block_io *beio; 1188 struct ctl_be_block_softc *softc; 1189 struct ctl_lba_len lbalen; 1190 uint64_t len_left, lbaoff; 1191 int i; 1192 1193 softc = be_lun->softc; 1194 1195 DPRINTF("entered\n"); 1196 1197 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 1198 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1199 } else { 1200 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1201 } 1202 1203 beio = ctl_alloc_beio(softc); 1204 beio->io = io; 1205 beio->lun = be_lun; 1206 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 1207 1208 /* 1209 * If the I/O came down with an ordered or head of queue tag, set 1210 * the BIO_ORDERED attribute. For head of queue tags, that's 1211 * pretty much the best we can do. 1212 * 1213 * XXX KDM we don't have a great way to easily know about the FUA 1214 * bit right now (it is decoded in ctl_read_write(), but we don't 1215 * pass that knowledge to the backend), and in any case we would 1216 * need to determine how to handle it. 1217 */ 1218 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1219 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1220 beio->bio_flags = BIO_ORDERED; 1221 1222 switch (io->scsiio.tag_type) { 1223 case CTL_TAG_ORDERED: 1224 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1225 break; 1226 case CTL_TAG_HEAD_OF_QUEUE: 1227 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1228 break; 1229 case CTL_TAG_UNTAGGED: 1230 case CTL_TAG_SIMPLE: 1231 case CTL_TAG_ACA: 1232 default: 1233 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1234 break; 1235 } 1236 1237 /* 1238 * This path handles read and write only. The config write path 1239 * handles flush operations. 1240 */ 1241 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 1242 beio->bio_cmd = BIO_READ; 1243 beio->ds_trans_type = DEVSTAT_READ; 1244 } else { 1245 beio->bio_cmd = BIO_WRITE; 1246 beio->ds_trans_type = DEVSTAT_WRITE; 1247 } 1248 1249 memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 1250 sizeof(lbalen)); 1251 DPRINTF("%s at LBA %jx len %u @%ju\n", 1252 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1253 (uintmax_t)lbalen.lba, lbalen.len, lbaoff); 1254 lbaoff = io->scsiio.kern_rel_offset / be_lun->blocksize; 1255 beio->io_offset = (lbalen.lba + lbaoff) * be_lun->blocksize; 1256 beio->io_len = MIN((lbalen.len - lbaoff) * be_lun->blocksize, 1257 CTLBLK_MAX_IO_SIZE); 1258 beio->io_len -= beio->io_len % be_lun->blocksize; 1259 1260 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1261 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1262 i, CTLBLK_MAX_SEGS)); 1263 1264 /* 1265 * Setup the S/G entry for this chunk. 1266 */ 1267 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1268 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1269 1270 DPRINTF("segment %d addr %p len %zd\n", i, 1271 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1272 1273 beio->num_segs++; 1274 len_left -= beio->sg_segs[i].len; 1275 } 1276 if (io->scsiio.kern_rel_offset + beio->io_len < 1277 io->scsiio.kern_total_len) 1278 beio->beio_cont = ctl_be_block_next; 1279 io->scsiio.be_move_done = ctl_be_block_move_done; 1280 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1281 io->scsiio.kern_data_len = beio->io_len; 1282 io->scsiio.kern_data_resid = 0; 1283 io->scsiio.kern_sg_entries = beio->num_segs; 1284 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1285 1286 /* 1287 * For the read case, we need to read the data into our buffers and 1288 * then we can send it back to the user. For the write case, we 1289 * need to get the data from the user first. 1290 */ 1291 if (beio->bio_cmd == BIO_READ) { 1292 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1293 be_lun->dispatch(be_lun, beio); 1294 } else { 1295 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1296#ifdef CTL_TIME_IO 1297 getbintime(&io->io_hdr.dma_start_bt); 1298#endif 1299 ctl_datamove(io); 1300 } 1301} 1302 1303static void 1304ctl_be_block_worker(void *context, int pending) 1305{ 1306 struct ctl_be_block_lun *be_lun; 1307 struct ctl_be_block_softc *softc; 1308 union ctl_io *io; 1309 1310 be_lun = (struct ctl_be_block_lun *)context; 1311 softc = be_lun->softc; 1312 1313 DPRINTF("entered\n"); 1314 1315 mtx_lock(&be_lun->lock); 1316 for (;;) { 1317 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1318 if (io != NULL) { 1319 struct ctl_be_block_io *beio; 1320 1321 DPRINTF("datamove queue\n"); 1322 1323 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1324 ctl_io_hdr, links); 1325 1326 mtx_unlock(&be_lun->lock); 1327 1328 beio = (struct ctl_be_block_io *) 1329 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1330 1331 be_lun->dispatch(be_lun, beio); 1332 1333 mtx_lock(&be_lun->lock); 1334 continue; 1335 } 1336 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1337 if (io != NULL) { 1338 1339 DPRINTF("config write queue\n"); 1340 1341 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1342 ctl_io_hdr, links); 1343 1344 mtx_unlock(&be_lun->lock); 1345 1346 ctl_be_block_cw_dispatch(be_lun, io); 1347 1348 mtx_lock(&be_lun->lock); 1349 continue; 1350 } 1351 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1352 if (io != NULL) { 1353 DPRINTF("input queue\n"); 1354 1355 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1356 ctl_io_hdr, links); 1357 mtx_unlock(&be_lun->lock); 1358 1359 /* 1360 * We must drop the lock, since this routine and 1361 * its children may sleep. 1362 */ 1363 ctl_be_block_dispatch(be_lun, io); 1364 1365 mtx_lock(&be_lun->lock); 1366 continue; 1367 } 1368 1369 /* 1370 * If we get here, there is no work left in the queues, so 1371 * just break out and let the task queue go to sleep. 1372 */ 1373 break; 1374 } 1375 mtx_unlock(&be_lun->lock); 1376} 1377 1378/* 1379 * Entry point from CTL to the backend for I/O. We queue everything to a 1380 * work thread, so this just puts the I/O on a queue and wakes up the 1381 * thread. 1382 */ 1383static int 1384ctl_be_block_submit(union ctl_io *io) 1385{ 1386 struct ctl_be_block_lun *be_lun; 1387 struct ctl_be_lun *ctl_be_lun; 1388 1389 DPRINTF("entered\n"); 1390 1391 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1392 CTL_PRIV_BACKEND_LUN].ptr; 1393 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1394 1395 /* 1396 * Make sure we only get SCSI I/O. 1397 */ 1398 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1399 "%#x) encountered", io->io_hdr.io_type)); 1400 1401 mtx_lock(&be_lun->lock); 1402 /* 1403 * XXX KDM make sure that links is okay to use at this point. 1404 * Otherwise, we either need to add another field to ctl_io_hdr, 1405 * or deal with resource allocation here. 1406 */ 1407 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1408 mtx_unlock(&be_lun->lock); 1409 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1410 1411 return (CTL_RETVAL_COMPLETE); 1412} 1413 1414static int 1415ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1416 int flag, struct thread *td) 1417{ 1418 struct ctl_be_block_softc *softc; 1419 int error; 1420 1421 softc = &backend_block_softc; 1422 1423 error = 0; 1424 1425 switch (cmd) { 1426 case CTL_LUN_REQ: { 1427 struct ctl_lun_req *lun_req; 1428 1429 lun_req = (struct ctl_lun_req *)addr; 1430 1431 switch (lun_req->reqtype) { 1432 case CTL_LUNREQ_CREATE: 1433 error = ctl_be_block_create(softc, lun_req); 1434 break; 1435 case CTL_LUNREQ_RM: 1436 error = ctl_be_block_rm(softc, lun_req); 1437 break; 1438 case CTL_LUNREQ_MODIFY: 1439 error = ctl_be_block_modify(softc, lun_req); 1440 break; 1441 default: 1442 lun_req->status = CTL_LUN_ERROR; 1443 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1444 "%s: invalid LUN request type %d", __func__, 1445 lun_req->reqtype); 1446 break; 1447 } 1448 break; 1449 } 1450 default: 1451 error = ENOTTY; 1452 break; 1453 } 1454 1455 return (error); 1456} 1457 1458static int 1459ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1460{ 1461 struct ctl_be_block_filedata *file_data; 1462 struct ctl_lun_create_params *params; 1463 struct vattr vattr; 1464 int error; 1465 1466 error = 0; 1467 file_data = &be_lun->backend.file; 1468 params = &req->reqdata.create; 1469 1470 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1471 be_lun->dispatch = ctl_be_block_dispatch_file; 1472 be_lun->lun_flush = ctl_be_block_flush_file; 1473 1474 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1475 if (error != 0) { 1476 snprintf(req->error_str, sizeof(req->error_str), 1477 "error calling VOP_GETATTR() for file %s", 1478 be_lun->dev_path); 1479 return (error); 1480 } 1481 1482 /* 1483 * Verify that we have the ability to upgrade to exclusive 1484 * access on this file so we can trap errors at open instead 1485 * of reporting them during first access. 1486 */ 1487 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1488 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1489 if (be_lun->vn->v_iflag & VI_DOOMED) { 1490 error = EBADF; 1491 snprintf(req->error_str, sizeof(req->error_str), 1492 "error locking file %s", be_lun->dev_path); 1493 return (error); 1494 } 1495 } 1496 1497 1498 file_data->cred = crhold(curthread->td_ucred); 1499 if (params->lun_size_bytes != 0) 1500 be_lun->size_bytes = params->lun_size_bytes; 1501 else 1502 be_lun->size_bytes = vattr.va_size; 1503 /* 1504 * We set the multi thread flag for file operations because all 1505 * filesystems (in theory) are capable of allowing multiple readers 1506 * of a file at once. So we want to get the maximum possible 1507 * concurrency. 1508 */ 1509 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1510 1511 /* 1512 * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1513 * With ZFS, it is 131072 bytes. Block sizes that large don't work 1514 * with disklabel and UFS on FreeBSD at least. Large block sizes 1515 * may not work with other OSes as well. So just export a sector 1516 * size of 512 bytes, which should work with any OS or 1517 * application. Since our backing is a file, any block size will 1518 * work fine for the backing store. 1519 */ 1520#if 0 1521 be_lun->blocksize= vattr.va_blocksize; 1522#endif 1523 if (params->blocksize_bytes != 0) 1524 be_lun->blocksize = params->blocksize_bytes; 1525 else 1526 be_lun->blocksize = 512; 1527 1528 /* 1529 * Sanity check. The media size has to be at least one 1530 * sector long. 1531 */ 1532 if (be_lun->size_bytes < be_lun->blocksize) { 1533 error = EINVAL; 1534 snprintf(req->error_str, sizeof(req->error_str), 1535 "file %s size %ju < block size %u", be_lun->dev_path, 1536 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1537 } 1538 return (error); 1539} 1540 1541static int 1542ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1543{ 1544 struct ctl_lun_create_params *params; 1545 struct vattr vattr; 1546 struct cdev *dev; 1547 struct cdevsw *devsw; 1548 int error; 1549 off_t ps, pss, po, pos; 1550 1551 params = &req->reqdata.create; 1552 1553 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1554 be_lun->dispatch = ctl_be_block_dispatch_dev; 1555 be_lun->lun_flush = ctl_be_block_flush_dev; 1556 be_lun->unmap = ctl_be_block_unmap_dev; 1557 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1558 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1559 &be_lun->backend.dev.dev_ref); 1560 if (be_lun->backend.dev.csw == NULL) 1561 panic("Unable to retrieve device switch"); 1562 1563 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1564 if (error) { 1565 snprintf(req->error_str, sizeof(req->error_str), 1566 "%s: error getting vnode attributes for device %s", 1567 __func__, be_lun->dev_path); 1568 return (error); 1569 } 1570 1571 dev = be_lun->vn->v_rdev; 1572 devsw = dev->si_devsw; 1573 if (!devsw->d_ioctl) { 1574 snprintf(req->error_str, sizeof(req->error_str), 1575 "%s: no d_ioctl for device %s!", __func__, 1576 be_lun->dev_path); 1577 return (ENODEV); 1578 } 1579 1580 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1581 (caddr_t)&be_lun->blocksize, FREAD, 1582 curthread); 1583 if (error) { 1584 snprintf(req->error_str, sizeof(req->error_str), 1585 "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1586 "on %s!", __func__, error, be_lun->dev_path); 1587 return (error); 1588 } 1589 1590 /* 1591 * If the user has asked for a blocksize that is greater than the 1592 * backing device's blocksize, we can do it only if the blocksize 1593 * the user is asking for is an even multiple of the underlying 1594 * device's blocksize. 1595 */ 1596 if ((params->blocksize_bytes != 0) 1597 && (params->blocksize_bytes > be_lun->blocksize)) { 1598 uint32_t bs_multiple, tmp_blocksize; 1599 1600 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1601 1602 tmp_blocksize = bs_multiple * be_lun->blocksize; 1603 1604 if (tmp_blocksize == params->blocksize_bytes) { 1605 be_lun->blocksize = params->blocksize_bytes; 1606 } else { 1607 snprintf(req->error_str, sizeof(req->error_str), 1608 "%s: requested blocksize %u is not an even " 1609 "multiple of backing device blocksize %u", 1610 __func__, params->blocksize_bytes, 1611 be_lun->blocksize); 1612 return (EINVAL); 1613 1614 } 1615 } else if ((params->blocksize_bytes != 0) 1616 && (params->blocksize_bytes != be_lun->blocksize)) { 1617 snprintf(req->error_str, sizeof(req->error_str), 1618 "%s: requested blocksize %u < backing device " 1619 "blocksize %u", __func__, params->blocksize_bytes, 1620 be_lun->blocksize); 1621 return (EINVAL); 1622 } 1623 1624 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1625 (caddr_t)&be_lun->size_bytes, FREAD, 1626 curthread); 1627 if (error) { 1628 snprintf(req->error_str, sizeof(req->error_str), 1629 "%s: error %d returned for DIOCGMEDIASIZE " 1630 " ioctl on %s!", __func__, error, 1631 be_lun->dev_path); 1632 return (error); 1633 } 1634 1635 if (params->lun_size_bytes != 0) { 1636 if (params->lun_size_bytes > be_lun->size_bytes) { 1637 snprintf(req->error_str, sizeof(req->error_str), 1638 "%s: requested LUN size %ju > backing device " 1639 "size %ju", __func__, 1640 (uintmax_t)params->lun_size_bytes, 1641 (uintmax_t)be_lun->size_bytes); 1642 return (EINVAL); 1643 } 1644 1645 be_lun->size_bytes = params->lun_size_bytes; 1646 } 1647 1648 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1649 (caddr_t)&ps, FREAD, curthread); 1650 if (error) 1651 ps = po = 0; 1652 else { 1653 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1654 (caddr_t)&po, FREAD, curthread); 1655 if (error) 1656 po = 0; 1657 } 1658 pss = ps / be_lun->blocksize; 1659 pos = po / be_lun->blocksize; 1660 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1661 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1662 be_lun->pblockexp = fls(pss) - 1; 1663 be_lun->pblockoff = (pss - pos) % pss; 1664 } 1665 1666 return (0); 1667} 1668 1669static int 1670ctl_be_block_close(struct ctl_be_block_lun *be_lun) 1671{ 1672 DROP_GIANT(); 1673 if (be_lun->vn) { 1674 int flags = FREAD | FWRITE; 1675 1676 switch (be_lun->dev_type) { 1677 case CTL_BE_BLOCK_DEV: 1678 if (be_lun->backend.dev.csw) { 1679 dev_relthread(be_lun->backend.dev.cdev, 1680 be_lun->backend.dev.dev_ref); 1681 be_lun->backend.dev.csw = NULL; 1682 be_lun->backend.dev.cdev = NULL; 1683 } 1684 break; 1685 case CTL_BE_BLOCK_FILE: 1686 break; 1687 case CTL_BE_BLOCK_NONE: 1688 break; 1689 default: 1690 panic("Unexpected backend type."); 1691 break; 1692 } 1693 1694 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1695 be_lun->vn = NULL; 1696 1697 switch (be_lun->dev_type) { 1698 case CTL_BE_BLOCK_DEV: 1699 break; 1700 case CTL_BE_BLOCK_FILE: 1701 if (be_lun->backend.file.cred != NULL) { 1702 crfree(be_lun->backend.file.cred); 1703 be_lun->backend.file.cred = NULL; 1704 } 1705 break; 1706 case CTL_BE_BLOCK_NONE: 1707 break; 1708 default: 1709 panic("Unexpected backend type."); 1710 break; 1711 } 1712 } 1713 PICKUP_GIANT(); 1714 1715 return (0); 1716} 1717 1718static int 1719ctl_be_block_open(struct ctl_be_block_softc *softc, 1720 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1721{ 1722 struct nameidata nd; 1723 int flags; 1724 int error; 1725 1726 /* 1727 * XXX KDM allow a read-only option? 1728 */ 1729 flags = FREAD | FWRITE; 1730 error = 0; 1731 1732 if (rootvnode == NULL) { 1733 snprintf(req->error_str, sizeof(req->error_str), 1734 "%s: Root filesystem is not mounted", __func__); 1735 return (1); 1736 } 1737 1738 if (!curthread->td_proc->p_fd->fd_cdir) { 1739 curthread->td_proc->p_fd->fd_cdir = rootvnode; 1740 VREF(rootvnode); 1741 } 1742 if (!curthread->td_proc->p_fd->fd_rdir) { 1743 curthread->td_proc->p_fd->fd_rdir = rootvnode; 1744 VREF(rootvnode); 1745 } 1746 if (!curthread->td_proc->p_fd->fd_jdir) { 1747 curthread->td_proc->p_fd->fd_jdir = rootvnode; 1748 VREF(rootvnode); 1749 } 1750 1751 again: 1752 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1753 error = vn_open(&nd, &flags, 0, NULL); 1754 if (error) { 1755 /* 1756 * This is the only reasonable guess we can make as far as 1757 * path if the user doesn't give us a fully qualified path. 1758 * If they want to specify a file, they need to specify the 1759 * full path. 1760 */ 1761 if (be_lun->dev_path[0] != '/') { 1762 char *dev_path = "/dev/"; 1763 char *dev_name; 1764 1765 /* Try adding device path at beginning of name */ 1766 dev_name = malloc(strlen(be_lun->dev_path) 1767 + strlen(dev_path) + 1, 1768 M_CTLBLK, M_WAITOK); 1769 if (dev_name) { 1770 sprintf(dev_name, "%s%s", dev_path, 1771 be_lun->dev_path); 1772 free(be_lun->dev_path, M_CTLBLK); 1773 be_lun->dev_path = dev_name; 1774 goto again; 1775 } 1776 } 1777 snprintf(req->error_str, sizeof(req->error_str), 1778 "%s: error opening %s", __func__, be_lun->dev_path); 1779 return (error); 1780 } 1781 1782 NDFREE(&nd, NDF_ONLY_PNBUF); 1783 1784 be_lun->vn = nd.ni_vp; 1785 1786 /* We only support disks and files. */ 1787 if (vn_isdisk(be_lun->vn, &error)) { 1788 error = ctl_be_block_open_dev(be_lun, req); 1789 } else if (be_lun->vn->v_type == VREG) { 1790 error = ctl_be_block_open_file(be_lun, req); 1791 } else { 1792 error = EINVAL; 1793 snprintf(req->error_str, sizeof(req->error_str), 1794 "%s is not a disk or plain file", be_lun->dev_path); 1795 } 1796 VOP_UNLOCK(be_lun->vn, 0); 1797 1798 if (error != 0) { 1799 ctl_be_block_close(be_lun); 1800 return (error); 1801 } 1802 1803 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1804 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1805 1806 return (0); 1807} 1808 1809static int 1810ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1811{ 1812 struct ctl_be_block_lun *be_lun; 1813 struct ctl_lun_create_params *params; 1814 char num_thread_str[16]; 1815 char tmpstr[32]; 1816 char *value; 1817 int retval, num_threads, unmap; 1818 int tmp_num_threads; 1819 1820 params = &req->reqdata.create; 1821 retval = 0; 1822 1823 num_threads = cbb_num_threads; 1824 1825 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1826 1827 be_lun->softc = softc; 1828 STAILQ_INIT(&be_lun->input_queue); 1829 STAILQ_INIT(&be_lun->config_write_queue); 1830 STAILQ_INIT(&be_lun->datamove_queue); 1831 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1832 mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF); 1833 ctl_init_opts(&be_lun->ctl_be_lun, req); 1834 1835 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1836 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1837 1838 if (be_lun->lun_zone == NULL) { 1839 snprintf(req->error_str, sizeof(req->error_str), 1840 "%s: error allocating UMA zone", __func__); 1841 goto bailout_error; 1842 } 1843 1844 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1845 be_lun->ctl_be_lun.lun_type = params->device_type; 1846 else 1847 be_lun->ctl_be_lun.lun_type = T_DIRECT; 1848 1849 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1850 value = ctl_get_opt(&be_lun->ctl_be_lun, "file"); 1851 if (value == NULL) { 1852 snprintf(req->error_str, sizeof(req->error_str), 1853 "%s: no file argument specified", __func__); 1854 goto bailout_error; 1855 } 1856 be_lun->dev_path = strdup(value, M_CTLBLK); 1857 1858 retval = ctl_be_block_open(softc, be_lun, req); 1859 if (retval != 0) { 1860 retval = 0; 1861 goto bailout_error; 1862 } 1863 1864 /* 1865 * Tell the user the size of the file/device. 1866 */ 1867 params->lun_size_bytes = be_lun->size_bytes; 1868 1869 /* 1870 * The maximum LBA is the size - 1. 1871 */ 1872 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1873 } else { 1874 /* 1875 * For processor devices, we don't have any size. 1876 */ 1877 be_lun->blocksize = 0; 1878 be_lun->pblockexp = 0; 1879 be_lun->pblockoff = 0; 1880 be_lun->size_blocks = 0; 1881 be_lun->size_bytes = 0; 1882 be_lun->ctl_be_lun.maxlba = 0; 1883 params->lun_size_bytes = 0; 1884 1885 /* 1886 * Default to just 1 thread for processor devices. 1887 */ 1888 num_threads = 1; 1889 } 1890 1891 /* 1892 * XXX This searching loop might be refactored to be combined with 1893 * the loop above, 1894 */ 1895 value = ctl_get_opt(&be_lun->ctl_be_lun, "num_threads"); 1896 if (value != NULL) { 1897 tmp_num_threads = strtol(value, NULL, 0); 1898 1899 /* 1900 * We don't let the user specify less than one 1901 * thread, but hope he's clueful enough not to 1902 * specify 1000 threads. 1903 */ 1904 if (tmp_num_threads < 1) { 1905 snprintf(req->error_str, sizeof(req->error_str), 1906 "%s: invalid number of threads %s", 1907 __func__, num_thread_str); 1908 goto bailout_error; 1909 } 1910 num_threads = tmp_num_threads; 1911 } 1912 unmap = 0; 1913 value = ctl_get_opt(&be_lun->ctl_be_lun, "unmap"); 1914 if (value != NULL && strcmp(value, "on") == 0) 1915 unmap = 1; 1916 1917 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1918 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1919 if (unmap) 1920 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 1921 be_lun->ctl_be_lun.be_lun = be_lun; 1922 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1923 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 1924 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 1925 /* Tell the user the blocksize we ended up using */ 1926 params->blocksize_bytes = be_lun->blocksize; 1927 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1928 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1929 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1930 } else 1931 be_lun->ctl_be_lun.req_lun_id = 0; 1932 1933 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1934 be_lun->ctl_be_lun.lun_config_status = 1935 ctl_be_block_lun_config_status; 1936 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1937 1938 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1939 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1940 softc->num_luns); 1941 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1942 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1943 sizeof(tmpstr))); 1944 1945 /* Tell the user what we used for a serial number */ 1946 strncpy((char *)params->serial_num, tmpstr, 1947 ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1948 } else { 1949 strncpy((char *)be_lun->ctl_be_lun.serial_num, 1950 params->serial_num, 1951 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1952 sizeof(params->serial_num))); 1953 } 1954 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1955 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1956 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1957 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1958 sizeof(tmpstr))); 1959 1960 /* Tell the user what we used for a device ID */ 1961 strncpy((char *)params->device_id, tmpstr, 1962 ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1963 } else { 1964 strncpy((char *)be_lun->ctl_be_lun.device_id, 1965 params->device_id, 1966 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1967 sizeof(params->device_id))); 1968 } 1969 1970 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 1971 1972 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 1973 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 1974 1975 if (be_lun->io_taskqueue == NULL) { 1976 snprintf(req->error_str, sizeof(req->error_str), 1977 "%s: Unable to create taskqueue", __func__); 1978 goto bailout_error; 1979 } 1980 1981 /* 1982 * Note that we start the same number of threads by default for 1983 * both the file case and the block device case. For the file 1984 * case, we need multiple threads to allow concurrency, because the 1985 * vnode interface is designed to be a blocking interface. For the 1986 * block device case, ZFS zvols at least will block the caller's 1987 * context in many instances, and so we need multiple threads to 1988 * overcome that problem. Other block devices don't need as many 1989 * threads, but they shouldn't cause too many problems. 1990 * 1991 * If the user wants to just have a single thread for a block 1992 * device, he can specify that when the LUN is created, or change 1993 * the tunable/sysctl to alter the default number of threads. 1994 */ 1995 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 1996 /*num threads*/num_threads, 1997 /*priority*/PWAIT, 1998 /*thread name*/ 1999 "%s taskq", be_lun->lunname); 2000 2001 if (retval != 0) 2002 goto bailout_error; 2003 2004 be_lun->num_threads = num_threads; 2005 2006 mtx_lock(&softc->lock); 2007 softc->num_luns++; 2008 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2009 2010 mtx_unlock(&softc->lock); 2011 2012 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2013 if (retval != 0) { 2014 mtx_lock(&softc->lock); 2015 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2016 links); 2017 softc->num_luns--; 2018 mtx_unlock(&softc->lock); 2019 snprintf(req->error_str, sizeof(req->error_str), 2020 "%s: ctl_add_lun() returned error %d, see dmesg for " 2021 "details", __func__, retval); 2022 retval = 0; 2023 goto bailout_error; 2024 } 2025 2026 mtx_lock(&softc->lock); 2027 2028 /* 2029 * Tell the config_status routine that we're waiting so it won't 2030 * clean up the LUN in the event of an error. 2031 */ 2032 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2033 2034 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2035 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2036 if (retval == EINTR) 2037 break; 2038 } 2039 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2040 2041 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2042 snprintf(req->error_str, sizeof(req->error_str), 2043 "%s: LUN configuration error, see dmesg for details", 2044 __func__); 2045 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2046 links); 2047 softc->num_luns--; 2048 mtx_unlock(&softc->lock); 2049 goto bailout_error; 2050 } else { 2051 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2052 } 2053 2054 mtx_unlock(&softc->lock); 2055 2056 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2057 be_lun->blocksize, 2058 DEVSTAT_ALL_SUPPORTED, 2059 be_lun->ctl_be_lun.lun_type 2060 | DEVSTAT_TYPE_IF_OTHER, 2061 DEVSTAT_PRIORITY_OTHER); 2062 2063 2064 req->status = CTL_LUN_OK; 2065 2066 return (retval); 2067 2068bailout_error: 2069 req->status = CTL_LUN_ERROR; 2070 2071 if (be_lun->io_taskqueue != NULL) 2072 taskqueue_free(be_lun->io_taskqueue); 2073 ctl_be_block_close(be_lun); 2074 if (be_lun->dev_path != NULL) 2075 free(be_lun->dev_path, M_CTLBLK); 2076 if (be_lun->lun_zone != NULL) 2077 uma_zdestroy(be_lun->lun_zone); 2078 ctl_free_opts(&be_lun->ctl_be_lun); 2079 mtx_destroy(&be_lun->lock); 2080 free(be_lun, M_CTLBLK); 2081 2082 return (retval); 2083} 2084 2085static int 2086ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2087{ 2088 struct ctl_lun_rm_params *params; 2089 struct ctl_be_block_lun *be_lun; 2090 int retval; 2091 2092 params = &req->reqdata.rm; 2093 2094 mtx_lock(&softc->lock); 2095 2096 be_lun = NULL; 2097 2098 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2099 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2100 break; 2101 } 2102 mtx_unlock(&softc->lock); 2103 2104 if (be_lun == NULL) { 2105 snprintf(req->error_str, sizeof(req->error_str), 2106 "%s: LUN %u is not managed by the block backend", 2107 __func__, params->lun_id); 2108 goto bailout_error; 2109 } 2110 2111 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2112 2113 if (retval != 0) { 2114 snprintf(req->error_str, sizeof(req->error_str), 2115 "%s: error %d returned from ctl_disable_lun() for " 2116 "LUN %d", __func__, retval, params->lun_id); 2117 goto bailout_error; 2118 2119 } 2120 2121 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2122 if (retval != 0) { 2123 snprintf(req->error_str, sizeof(req->error_str), 2124 "%s: error %d returned from ctl_invalidate_lun() for " 2125 "LUN %d", __func__, retval, params->lun_id); 2126 goto bailout_error; 2127 } 2128 2129 mtx_lock(&softc->lock); 2130 2131 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2132 2133 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2134 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2135 if (retval == EINTR) 2136 break; 2137 } 2138 2139 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2140 2141 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2142 snprintf(req->error_str, sizeof(req->error_str), 2143 "%s: interrupted waiting for LUN to be freed", 2144 __func__); 2145 mtx_unlock(&softc->lock); 2146 goto bailout_error; 2147 } 2148 2149 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2150 2151 softc->num_luns--; 2152 mtx_unlock(&softc->lock); 2153 2154 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2155 2156 taskqueue_free(be_lun->io_taskqueue); 2157 2158 ctl_be_block_close(be_lun); 2159 2160 if (be_lun->disk_stats != NULL) 2161 devstat_remove_entry(be_lun->disk_stats); 2162 2163 uma_zdestroy(be_lun->lun_zone); 2164 2165 ctl_free_opts(&be_lun->ctl_be_lun); 2166 free(be_lun->dev_path, M_CTLBLK); 2167 2168 free(be_lun, M_CTLBLK); 2169 2170 req->status = CTL_LUN_OK; 2171 2172 return (0); 2173 2174bailout_error: 2175 2176 req->status = CTL_LUN_ERROR; 2177 2178 return (0); 2179} 2180 2181static int 2182ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2183 struct ctl_lun_req *req) 2184{ 2185 struct vattr vattr; 2186 int error; 2187 struct ctl_lun_modify_params *params; 2188 2189 params = &req->reqdata.modify; 2190 2191 if (params->lun_size_bytes != 0) { 2192 be_lun->size_bytes = params->lun_size_bytes; 2193 } else { 2194 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2195 if (error != 0) { 2196 snprintf(req->error_str, sizeof(req->error_str), 2197 "error calling VOP_GETATTR() for file %s", 2198 be_lun->dev_path); 2199 return (error); 2200 } 2201 2202 be_lun->size_bytes = vattr.va_size; 2203 } 2204 2205 return (0); 2206} 2207 2208static int 2209ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2210 struct ctl_lun_req *req) 2211{ 2212 struct cdev *dev; 2213 struct cdevsw *devsw; 2214 int error; 2215 struct ctl_lun_modify_params *params; 2216 uint64_t size_bytes; 2217 2218 params = &req->reqdata.modify; 2219 2220 dev = be_lun->vn->v_rdev; 2221 devsw = dev->si_devsw; 2222 if (!devsw->d_ioctl) { 2223 snprintf(req->error_str, sizeof(req->error_str), 2224 "%s: no d_ioctl for device %s!", __func__, 2225 be_lun->dev_path); 2226 return (ENODEV); 2227 } 2228 2229 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2230 (caddr_t)&size_bytes, FREAD, 2231 curthread); 2232 if (error) { 2233 snprintf(req->error_str, sizeof(req->error_str), 2234 "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2235 "on %s!", __func__, error, be_lun->dev_path); 2236 return (error); 2237 } 2238 2239 if (params->lun_size_bytes != 0) { 2240 if (params->lun_size_bytes > size_bytes) { 2241 snprintf(req->error_str, sizeof(req->error_str), 2242 "%s: requested LUN size %ju > backing device " 2243 "size %ju", __func__, 2244 (uintmax_t)params->lun_size_bytes, 2245 (uintmax_t)size_bytes); 2246 return (EINVAL); 2247 } 2248 2249 be_lun->size_bytes = params->lun_size_bytes; 2250 } else { 2251 be_lun->size_bytes = size_bytes; 2252 } 2253 2254 return (0); 2255} 2256 2257static int 2258ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2259{ 2260 struct ctl_lun_modify_params *params; 2261 struct ctl_be_block_lun *be_lun; 2262 int error; 2263 2264 params = &req->reqdata.modify; 2265 2266 mtx_lock(&softc->lock); 2267 2268 be_lun = NULL; 2269 2270 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2271 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2272 break; 2273 } 2274 mtx_unlock(&softc->lock); 2275 2276 if (be_lun == NULL) { 2277 snprintf(req->error_str, sizeof(req->error_str), 2278 "%s: LUN %u is not managed by the block backend", 2279 __func__, params->lun_id); 2280 goto bailout_error; 2281 } 2282 2283 if (params->lun_size_bytes != 0) { 2284 if (params->lun_size_bytes < be_lun->blocksize) { 2285 snprintf(req->error_str, sizeof(req->error_str), 2286 "%s: LUN size %ju < blocksize %u", __func__, 2287 params->lun_size_bytes, be_lun->blocksize); 2288 goto bailout_error; 2289 } 2290 } 2291 2292 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2293 2294 if (be_lun->vn->v_type == VREG) 2295 error = ctl_be_block_modify_file(be_lun, req); 2296 else 2297 error = ctl_be_block_modify_dev(be_lun, req); 2298 2299 VOP_UNLOCK(be_lun->vn, 0); 2300 2301 if (error != 0) 2302 goto bailout_error; 2303 2304 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2305 2306 /* 2307 * The maximum LBA is the size - 1. 2308 * 2309 * XXX: Note that this field is being updated without locking, 2310 * which might cause problems on 32-bit architectures. 2311 */ 2312 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2313 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2314 2315 /* Tell the user the exact size we ended up using */ 2316 params->lun_size_bytes = be_lun->size_bytes; 2317 2318 req->status = CTL_LUN_OK; 2319 2320 return (0); 2321 2322bailout_error: 2323 req->status = CTL_LUN_ERROR; 2324 2325 return (0); 2326} 2327 2328static void 2329ctl_be_block_lun_shutdown(void *be_lun) 2330{ 2331 struct ctl_be_block_lun *lun; 2332 struct ctl_be_block_softc *softc; 2333 2334 lun = (struct ctl_be_block_lun *)be_lun; 2335 2336 softc = lun->softc; 2337 2338 mtx_lock(&softc->lock); 2339 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2340 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2341 wakeup(lun); 2342 mtx_unlock(&softc->lock); 2343 2344} 2345 2346static void 2347ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2348{ 2349 struct ctl_be_block_lun *lun; 2350 struct ctl_be_block_softc *softc; 2351 2352 lun = (struct ctl_be_block_lun *)be_lun; 2353 softc = lun->softc; 2354 2355 if (status == CTL_LUN_CONFIG_OK) { 2356 mtx_lock(&softc->lock); 2357 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2358 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2359 wakeup(lun); 2360 mtx_unlock(&softc->lock); 2361 2362 /* 2363 * We successfully added the LUN, attempt to enable it. 2364 */ 2365 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2366 printf("%s: ctl_enable_lun() failed!\n", __func__); 2367 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2368 printf("%s: ctl_invalidate_lun() failed!\n", 2369 __func__); 2370 } 2371 } 2372 2373 return; 2374 } 2375 2376 2377 mtx_lock(&softc->lock); 2378 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2379 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2380 wakeup(lun); 2381 mtx_unlock(&softc->lock); 2382} 2383 2384 2385static int 2386ctl_be_block_config_write(union ctl_io *io) 2387{ 2388 struct ctl_be_block_lun *be_lun; 2389 struct ctl_be_lun *ctl_be_lun; 2390 int retval; 2391 2392 retval = 0; 2393 2394 DPRINTF("entered\n"); 2395 2396 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2397 CTL_PRIV_BACKEND_LUN].ptr; 2398 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2399 2400 switch (io->scsiio.cdb[0]) { 2401 case SYNCHRONIZE_CACHE: 2402 case SYNCHRONIZE_CACHE_16: 2403 case WRITE_SAME_10: 2404 case WRITE_SAME_16: 2405 case UNMAP: 2406 /* 2407 * The upper level CTL code will filter out any CDBs with 2408 * the immediate bit set and return the proper error. 2409 * 2410 * We don't really need to worry about what LBA range the 2411 * user asked to be synced out. When they issue a sync 2412 * cache command, we'll sync out the whole thing. 2413 */ 2414 mtx_lock(&be_lun->lock); 2415 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2416 links); 2417 mtx_unlock(&be_lun->lock); 2418 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2419 break; 2420 case START_STOP_UNIT: { 2421 struct scsi_start_stop_unit *cdb; 2422 2423 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2424 2425 if (cdb->how & SSS_START) 2426 retval = ctl_start_lun(ctl_be_lun); 2427 else { 2428 retval = ctl_stop_lun(ctl_be_lun); 2429 /* 2430 * XXX KDM Copan-specific offline behavior. 2431 * Figure out a reasonable way to port this? 2432 */ 2433#ifdef NEEDTOPORT 2434 if ((retval == 0) 2435 && (cdb->byte2 & SSS_ONOFFLINE)) 2436 retval = ctl_lun_offline(ctl_be_lun); 2437#endif 2438 } 2439 2440 /* 2441 * In general, the above routines should not fail. They 2442 * just set state for the LUN. So we've got something 2443 * pretty wrong here if we can't start or stop the LUN. 2444 */ 2445 if (retval != 0) { 2446 ctl_set_internal_failure(&io->scsiio, 2447 /*sks_valid*/ 1, 2448 /*retry_count*/ 0xf051); 2449 retval = CTL_RETVAL_COMPLETE; 2450 } else { 2451 ctl_set_success(&io->scsiio); 2452 } 2453 ctl_config_write_done(io); 2454 break; 2455 } 2456 default: 2457 ctl_set_invalid_opcode(&io->scsiio); 2458 ctl_config_write_done(io); 2459 retval = CTL_RETVAL_COMPLETE; 2460 break; 2461 } 2462 2463 return (retval); 2464 2465} 2466 2467static int 2468ctl_be_block_config_read(union ctl_io *io) 2469{ 2470 return (0); 2471} 2472 2473static int 2474ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2475{ 2476 struct ctl_be_block_lun *lun; 2477 int retval; 2478 2479 lun = (struct ctl_be_block_lun *)be_lun; 2480 retval = 0; 2481 2482 retval = sbuf_printf(sb, "<num_threads>"); 2483 2484 if (retval != 0) 2485 goto bailout; 2486 2487 retval = sbuf_printf(sb, "%d", lun->num_threads); 2488 2489 if (retval != 0) 2490 goto bailout; 2491 2492 retval = sbuf_printf(sb, "</num_threads>"); 2493 2494bailout: 2495 2496 return (retval); 2497} 2498 2499int 2500ctl_be_block_init(void) 2501{ 2502 struct ctl_be_block_softc *softc; 2503 int retval; 2504 2505 softc = &backend_block_softc; 2506 retval = 0; 2507 2508 mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF); 2509 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2510 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2511 STAILQ_INIT(&softc->disk_list); 2512 STAILQ_INIT(&softc->lun_list); 2513 2514 return (retval); 2515} 2516