ctl_backend_block.c revision 269122
1/*- 2 * Copyright (c) 2003 Silicon Graphics International Corp. 3 * Copyright (c) 2009-2011 Spectra Logic Corporation 4 * Copyright (c) 2012 The FreeBSD Foundation 5 * All rights reserved. 6 * 7 * Portions of this software were developed by Edward Tomasz Napierala 8 * under sponsorship from the FreeBSD Foundation. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions, and the following disclaimer, 15 * without modification. 16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17 * substantially similar to the "NO WARRANTY" disclaimer below 18 * ("Disclaimer") and any redistribution must be conditioned upon 19 * including a substantially similar Disclaimer requirement for further 20 * binary redistribution. 21 * 22 * NO WARRANTY 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGES. 34 * 35 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36 */ 37/* 38 * CAM Target Layer driver backend for block devices. 39 * 40 * Author: Ken Merry <ken@FreeBSD.org> 41 */ 42#include <sys/cdefs.h> 43__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl_backend_block.c 269122 2014-07-26 13:43:25Z mav $"); 44 45#include <sys/param.h> 46#include <sys/systm.h> 47#include <sys/kernel.h> 48#include <sys/types.h> 49#include <sys/kthread.h> 50#include <sys/bio.h> 51#include <sys/fcntl.h> 52#include <sys/limits.h> 53#include <sys/lock.h> 54#include <sys/mutex.h> 55#include <sys/condvar.h> 56#include <sys/malloc.h> 57#include <sys/conf.h> 58#include <sys/ioccom.h> 59#include <sys/queue.h> 60#include <sys/sbuf.h> 61#include <sys/endian.h> 62#include <sys/uio.h> 63#include <sys/buf.h> 64#include <sys/taskqueue.h> 65#include <sys/vnode.h> 66#include <sys/namei.h> 67#include <sys/mount.h> 68#include <sys/disk.h> 69#include <sys/fcntl.h> 70#include <sys/filedesc.h> 71#include <sys/proc.h> 72#include <sys/pcpu.h> 73#include <sys/module.h> 74#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_HALF_IO_SIZE (512 * 1024) 96#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 97#define CTLBLK_MAX_SEG MAXPHYS 98#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 99#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 100 101#ifdef CTLBLK_DEBUG 102#define DPRINTF(fmt, args...) \ 103 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 104#else 105#define DPRINTF(fmt, args...) do {} while(0) 106#endif 107 108#define PRIV(io) \ 109 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 110#define ARGS(io) \ 111 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 112 113SDT_PROVIDER_DEFINE(cbb); 114 115typedef enum { 116 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 117 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 118 CTL_BE_BLOCK_LUN_WAITING = 0x04, 119 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 120} ctl_be_block_lun_flags; 121 122typedef enum { 123 CTL_BE_BLOCK_NONE, 124 CTL_BE_BLOCK_DEV, 125 CTL_BE_BLOCK_FILE 126} ctl_be_block_type; 127 128struct ctl_be_block_devdata { 129 struct cdev *cdev; 130 struct cdevsw *csw; 131 int dev_ref; 132}; 133 134struct ctl_be_block_filedata { 135 struct ucred *cred; 136}; 137 138union ctl_be_block_bedata { 139 struct ctl_be_block_devdata dev; 140 struct ctl_be_block_filedata file; 141}; 142 143struct ctl_be_block_io; 144struct ctl_be_block_lun; 145 146typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 147 struct ctl_be_block_io *beio); 148 149/* 150 * Backend LUN structure. There is a 1:1 mapping between a block device 151 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 152 */ 153struct ctl_be_block_lun { 154 struct ctl_block_disk *disk; 155 char lunname[32]; 156 char *dev_path; 157 ctl_be_block_type dev_type; 158 struct vnode *vn; 159 union ctl_be_block_bedata backend; 160 cbb_dispatch_t dispatch; 161 cbb_dispatch_t lun_flush; 162 cbb_dispatch_t unmap; 163 uma_zone_t lun_zone; 164 uint64_t size_blocks; 165 uint64_t size_bytes; 166 uint32_t blocksize; 167 int blocksize_shift; 168 uint16_t pblockexp; 169 uint16_t pblockoff; 170 struct ctl_be_block_softc *softc; 171 struct devstat *disk_stats; 172 ctl_be_block_lun_flags flags; 173 STAILQ_ENTRY(ctl_be_block_lun) links; 174 struct ctl_be_lun ctl_be_lun; 175 struct taskqueue *io_taskqueue; 176 struct task io_task; 177 int num_threads; 178 STAILQ_HEAD(, ctl_io_hdr) input_queue; 179 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 180 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 181 struct mtx_padalign io_lock; 182 struct mtx_padalign queue_lock; 183}; 184 185/* 186 * Overall softc structure for the block backend module. 187 */ 188struct ctl_be_block_softc { 189 struct mtx lock; 190 int num_disks; 191 STAILQ_HEAD(, ctl_block_disk) disk_list; 192 int num_luns; 193 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 194}; 195 196static struct ctl_be_block_softc backend_block_softc; 197 198/* 199 * Per-I/O information. 200 */ 201struct ctl_be_block_io { 202 union ctl_io *io; 203 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 204 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 205 int bio_cmd; 206 int bio_flags; 207 int num_segs; 208 int num_bios_sent; 209 int num_bios_done; 210 int send_complete; 211 int num_errors; 212 struct bintime ds_t0; 213 devstat_tag_type ds_tag_type; 214 devstat_trans_flags ds_trans_type; 215 uint64_t io_len; 216 uint64_t io_offset; 217 struct ctl_be_block_softc *softc; 218 struct ctl_be_block_lun *lun; 219 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 220}; 221 222static int cbb_num_threads = 14; 223SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 224 "CAM Target Layer Block Backend"); 225SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 226 &cbb_num_threads, 0, "Number of threads per backing file"); 227 228static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 229static void ctl_free_beio(struct ctl_be_block_io *beio); 230static void ctl_complete_beio(struct ctl_be_block_io *beio); 231static int ctl_be_block_move_done(union ctl_io *io); 232static void ctl_be_block_biodone(struct bio *bio); 233static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 234 struct ctl_be_block_io *beio); 235static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 236 struct ctl_be_block_io *beio); 237static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 238 struct ctl_be_block_io *beio); 239static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 240 struct ctl_be_block_io *beio); 241static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 242 struct ctl_be_block_io *beio); 243static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 244 union ctl_io *io); 245static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 246 union ctl_io *io); 247static void ctl_be_block_worker(void *context, int pending); 248static int ctl_be_block_submit(union ctl_io *io); 249static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 250 int flag, struct thread *td); 251static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 252 struct ctl_lun_req *req); 253static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 254 struct ctl_lun_req *req); 255static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 256static int ctl_be_block_open(struct ctl_be_block_softc *softc, 257 struct ctl_be_block_lun *be_lun, 258 struct ctl_lun_req *req); 259static int ctl_be_block_create(struct ctl_be_block_softc *softc, 260 struct ctl_lun_req *req); 261static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 262 struct ctl_lun_req *req); 263static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 264 struct ctl_lun_req *req); 265static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 266 struct ctl_lun_req *req); 267static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 268 struct ctl_lun_req *req); 269static void ctl_be_block_lun_shutdown(void *be_lun); 270static void ctl_be_block_lun_config_status(void *be_lun, 271 ctl_lun_config_status status); 272static int ctl_be_block_config_write(union ctl_io *io); 273static int ctl_be_block_config_read(union ctl_io *io); 274static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 275int ctl_be_block_init(void); 276 277static struct ctl_backend_driver ctl_be_block_driver = 278{ 279 .name = "block", 280 .flags = CTL_BE_FLAG_HAS_CONFIG, 281 .init = ctl_be_block_init, 282 .data_submit = ctl_be_block_submit, 283 .data_move_done = ctl_be_block_move_done, 284 .config_read = ctl_be_block_config_read, 285 .config_write = ctl_be_block_config_write, 286 .ioctl = ctl_be_block_ioctl, 287 .lun_info = ctl_be_block_lun_info 288}; 289 290MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 291CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 292 293static uma_zone_t beio_zone; 294 295static struct ctl_be_block_io * 296ctl_alloc_beio(struct ctl_be_block_softc *softc) 297{ 298 struct ctl_be_block_io *beio; 299 300 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 301 beio->softc = softc; 302 return (beio); 303} 304 305static void 306ctl_free_beio(struct ctl_be_block_io *beio) 307{ 308 int duplicate_free; 309 int i; 310 311 duplicate_free = 0; 312 313 for (i = 0; i < beio->num_segs; i++) { 314 if (beio->sg_segs[i].addr == NULL) 315 duplicate_free++; 316 317 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 318 beio->sg_segs[i].addr = NULL; 319 320 /* For compare we had two equal S/G lists. */ 321 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 322 uma_zfree(beio->lun->lun_zone, 323 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 324 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 325 } 326 } 327 328 if (duplicate_free > 0) { 329 printf("%s: %d duplicate frees out of %d segments\n", __func__, 330 duplicate_free, beio->num_segs); 331 } 332 333 uma_zfree(beio_zone, beio); 334} 335 336static void 337ctl_complete_beio(struct ctl_be_block_io *beio) 338{ 339 union ctl_io *io = beio->io; 340 341 if (beio->beio_cont != NULL) { 342 beio->beio_cont(beio); 343 } else { 344 ctl_free_beio(beio); 345 ctl_data_submit_done(io); 346 } 347} 348 349static int 350ctl_be_block_move_done(union ctl_io *io) 351{ 352 struct ctl_be_block_io *beio; 353 struct ctl_be_block_lun *be_lun; 354 struct ctl_lba_len_flags *lbalen; 355#ifdef CTL_TIME_IO 356 struct bintime cur_bt; 357#endif 358 int i; 359 360 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 361 be_lun = beio->lun; 362 363 DPRINTF("entered\n"); 364 365#ifdef CTL_TIME_IO 366 getbintime(&cur_bt); 367 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 368 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 369 io->io_hdr.num_dmas++; 370#endif 371 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 372 373 /* 374 * We set status at this point for read commands, and write 375 * commands with errors. 376 */ 377 if ((io->io_hdr.port_status == 0) && 378 ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 379 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 380 lbalen = ARGS(beio->io); 381 if (lbalen->flags & CTL_LLF_READ) { 382 ctl_set_success(&io->scsiio); 383 } else if (lbalen->flags & CTL_LLF_COMPARE) { 384 /* We have two data blocks ready for comparison. */ 385 for (i = 0; i < beio->num_segs; i++) { 386 if (memcmp(beio->sg_segs[i].addr, 387 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 388 beio->sg_segs[i].len) != 0) 389 break; 390 } 391 if (i < beio->num_segs) 392 ctl_set_sense(&io->scsiio, 393 /*current_error*/ 1, 394 /*sense_key*/ SSD_KEY_MISCOMPARE, 395 /*asc*/ 0x1D, 396 /*ascq*/ 0x00, 397 SSD_ELEM_NONE); 398 else 399 ctl_set_success(&io->scsiio); 400 } 401 } 402 else if ((io->io_hdr.port_status != 0) 403 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 404 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 405 /* 406 * For hardware error sense keys, the sense key 407 * specific value is defined to be a retry count, 408 * but we use it to pass back an internal FETD 409 * error code. XXX KDM Hopefully the FETD is only 410 * using 16 bits for an error code, since that's 411 * all the space we have in the sks field. 412 */ 413 ctl_set_internal_failure(&io->scsiio, 414 /*sks_valid*/ 1, 415 /*retry_count*/ 416 io->io_hdr.port_status); 417 } 418 419 /* 420 * If this is a read, or a write with errors, it is done. 421 */ 422 if ((beio->bio_cmd == BIO_READ) 423 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 424 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 425 ctl_complete_beio(beio); 426 return (0); 427 } 428 429 /* 430 * At this point, we have a write and the DMA completed 431 * successfully. We now have to queue it to the task queue to 432 * execute the backend I/O. That is because we do blocking 433 * memory allocations, and in the file backing case, blocking I/O. 434 * This move done routine is generally called in the SIM's 435 * interrupt context, and therefore we cannot block. 436 */ 437 mtx_lock(&be_lun->queue_lock); 438 /* 439 * XXX KDM make sure that links is okay to use at this point. 440 * Otherwise, we either need to add another field to ctl_io_hdr, 441 * or deal with resource allocation here. 442 */ 443 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 444 mtx_unlock(&be_lun->queue_lock); 445 446 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 447 448 return (0); 449} 450 451static void 452ctl_be_block_biodone(struct bio *bio) 453{ 454 struct ctl_be_block_io *beio; 455 struct ctl_be_block_lun *be_lun; 456 union ctl_io *io; 457 int error; 458 459 beio = bio->bio_caller1; 460 be_lun = beio->lun; 461 io = beio->io; 462 463 DPRINTF("entered\n"); 464 465 error = bio->bio_error; 466 mtx_lock(&be_lun->io_lock); 467 if (error != 0) 468 beio->num_errors++; 469 470 beio->num_bios_done++; 471 472 /* 473 * XXX KDM will this cause WITNESS to complain? Holding a lock 474 * during the free might cause it to complain. 475 */ 476 g_destroy_bio(bio); 477 478 /* 479 * If the send complete bit isn't set, or we aren't the last I/O to 480 * complete, then we're done. 481 */ 482 if ((beio->send_complete == 0) 483 || (beio->num_bios_done < beio->num_bios_sent)) { 484 mtx_unlock(&be_lun->io_lock); 485 return; 486 } 487 488 /* 489 * At this point, we've verified that we are the last I/O to 490 * complete, so it's safe to drop the lock. 491 */ 492 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 493 beio->ds_tag_type, beio->ds_trans_type, 494 /*now*/ NULL, /*then*/&beio->ds_t0); 495 mtx_unlock(&be_lun->io_lock); 496 497 /* 498 * If there are any errors from the backing device, we fail the 499 * entire I/O with a medium error. 500 */ 501 if (beio->num_errors > 0) { 502 if (error == EOPNOTSUPP) { 503 ctl_set_invalid_opcode(&io->scsiio); 504 } else if (beio->bio_cmd == BIO_FLUSH) { 505 /* XXX KDM is there is a better error here? */ 506 ctl_set_internal_failure(&io->scsiio, 507 /*sks_valid*/ 1, 508 /*retry_count*/ 0xbad2); 509 } else 510 ctl_set_medium_error(&io->scsiio); 511 ctl_complete_beio(beio); 512 return; 513 } 514 515 /* 516 * If this is a write, a flush, a delete or verify, we're all done. 517 * If this is a read, we can now send the data to the user. 518 */ 519 if ((beio->bio_cmd == BIO_WRITE) 520 || (beio->bio_cmd == BIO_FLUSH) 521 || (beio->bio_cmd == BIO_DELETE) 522 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 523 ctl_set_success(&io->scsiio); 524 ctl_complete_beio(beio); 525 } else { 526#ifdef CTL_TIME_IO 527 getbintime(&io->io_hdr.dma_start_bt); 528#endif 529 ctl_datamove(io); 530 } 531} 532 533static void 534ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 535 struct ctl_be_block_io *beio) 536{ 537 union ctl_io *io = beio->io; 538 struct mount *mountpoint; 539 int error, lock_flags; 540 541 DPRINTF("entered\n"); 542 543 binuptime(&beio->ds_t0); 544 mtx_lock(&be_lun->io_lock); 545 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 546 mtx_unlock(&be_lun->io_lock); 547 548 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 549 550 if (MNT_SHARED_WRITES(mountpoint) 551 || ((mountpoint == NULL) 552 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 553 lock_flags = LK_SHARED; 554 else 555 lock_flags = LK_EXCLUSIVE; 556 557 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 558 559 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 560 VOP_UNLOCK(be_lun->vn, 0); 561 562 vn_finished_write(mountpoint); 563 564 mtx_lock(&be_lun->io_lock); 565 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 566 beio->ds_tag_type, beio->ds_trans_type, 567 /*now*/ NULL, /*then*/&beio->ds_t0); 568 mtx_unlock(&be_lun->io_lock); 569 570 if (error == 0) 571 ctl_set_success(&io->scsiio); 572 else { 573 /* XXX KDM is there is a better error here? */ 574 ctl_set_internal_failure(&io->scsiio, 575 /*sks_valid*/ 1, 576 /*retry_count*/ 0xbad1); 577 } 578 579 ctl_complete_beio(beio); 580} 581 582SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 583SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 584SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 585SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 586 587static void 588ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 589 struct ctl_be_block_io *beio) 590{ 591 struct ctl_be_block_filedata *file_data; 592 union ctl_io *io; 593 struct uio xuio; 594 struct iovec *xiovec; 595 int flags; 596 int error, i; 597 598 DPRINTF("entered\n"); 599 600 file_data = &be_lun->backend.file; 601 io = beio->io; 602 flags = beio->bio_flags; 603 604 bzero(&xuio, sizeof(xuio)); 605 if (beio->bio_cmd == BIO_READ) { 606 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 607 xuio.uio_rw = UIO_READ; 608 } else { 609 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 610 xuio.uio_rw = UIO_WRITE; 611 } 612 xuio.uio_offset = beio->io_offset; 613 xuio.uio_resid = beio->io_len; 614 xuio.uio_segflg = UIO_SYSSPACE; 615 xuio.uio_iov = beio->xiovecs; 616 xuio.uio_iovcnt = beio->num_segs; 617 xuio.uio_td = curthread; 618 619 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 620 xiovec->iov_base = beio->sg_segs[i].addr; 621 xiovec->iov_len = beio->sg_segs[i].len; 622 } 623 624 binuptime(&beio->ds_t0); 625 mtx_lock(&be_lun->io_lock); 626 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 627 mtx_unlock(&be_lun->io_lock); 628 629 if (beio->bio_cmd == BIO_READ) { 630 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 631 632 /* 633 * UFS pays attention to IO_DIRECT for reads. If the 634 * DIRECTIO option is configured into the kernel, it calls 635 * ffs_rawread(). But that only works for single-segment 636 * uios with user space addresses. In our case, with a 637 * kernel uio, it still reads into the buffer cache, but it 638 * will just try to release the buffer from the cache later 639 * on in ffs_read(). 640 * 641 * ZFS does not pay attention to IO_DIRECT for reads. 642 * 643 * UFS does not pay attention to IO_SYNC for reads. 644 * 645 * ZFS pays attention to IO_SYNC (which translates into the 646 * Solaris define FRSYNC for zfs_read()) for reads. It 647 * attempts to sync the file before reading. 648 * 649 * So, to attempt to provide some barrier semantics in the 650 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 651 */ 652 error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 653 (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 654 655 VOP_UNLOCK(be_lun->vn, 0); 656 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 657 } else { 658 struct mount *mountpoint; 659 int lock_flags; 660 661 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 662 663 if (MNT_SHARED_WRITES(mountpoint) 664 || ((mountpoint == NULL) 665 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 666 lock_flags = LK_SHARED; 667 else 668 lock_flags = LK_EXCLUSIVE; 669 670 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 671 672 /* 673 * UFS pays attention to IO_DIRECT for writes. The write 674 * is done asynchronously. (Normally the write would just 675 * get put into cache. 676 * 677 * UFS pays attention to IO_SYNC for writes. It will 678 * attempt to write the buffer out synchronously if that 679 * flag is set. 680 * 681 * ZFS does not pay attention to IO_DIRECT for writes. 682 * 683 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 684 * for writes. It will flush the transaction from the 685 * cache before returning. 686 * 687 * So if we've got the BIO_ORDERED flag set, we want 688 * IO_SYNC in either the UFS or ZFS case. 689 */ 690 error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 691 IO_SYNC : 0, file_data->cred); 692 VOP_UNLOCK(be_lun->vn, 0); 693 694 vn_finished_write(mountpoint); 695 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 696 } 697 698 mtx_lock(&be_lun->io_lock); 699 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 700 beio->ds_tag_type, beio->ds_trans_type, 701 /*now*/ NULL, /*then*/&beio->ds_t0); 702 mtx_unlock(&be_lun->io_lock); 703 704 /* 705 * If we got an error, set the sense data to "MEDIUM ERROR" and 706 * return the I/O to the user. 707 */ 708 if (error != 0) { 709 char path_str[32]; 710 711 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 712 /* 713 * XXX KDM ZFS returns ENOSPC when the underlying 714 * filesystem fills up. What kind of SCSI error should we 715 * return for that? 716 */ 717 printf("%s%s command returned errno %d\n", path_str, 718 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 719 ctl_set_medium_error(&io->scsiio); 720 ctl_complete_beio(beio); 721 return; 722 } 723 724 /* 725 * If this is a write or a verify, we're all done. 726 * If this is a read, we can now send the data to the user. 727 */ 728 if ((beio->bio_cmd == BIO_WRITE) || 729 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 730 ctl_set_success(&io->scsiio); 731 ctl_complete_beio(beio); 732 } else { 733#ifdef CTL_TIME_IO 734 getbintime(&io->io_hdr.dma_start_bt); 735#endif 736 ctl_datamove(io); 737 } 738} 739 740static void 741ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 742 struct ctl_be_block_io *beio) 743{ 744 struct bio *bio; 745 union ctl_io *io; 746 struct ctl_be_block_devdata *dev_data; 747 748 dev_data = &be_lun->backend.dev; 749 io = beio->io; 750 751 DPRINTF("entered\n"); 752 753 /* This can't fail, it's a blocking allocation. */ 754 bio = g_alloc_bio(); 755 756 bio->bio_cmd = BIO_FLUSH; 757 bio->bio_flags |= BIO_ORDERED; 758 bio->bio_dev = dev_data->cdev; 759 bio->bio_offset = 0; 760 bio->bio_data = 0; 761 bio->bio_done = ctl_be_block_biodone; 762 bio->bio_caller1 = beio; 763 bio->bio_pblkno = 0; 764 765 /* 766 * We don't need to acquire the LUN lock here, because we are only 767 * sending one bio, and so there is no other context to synchronize 768 * with. 769 */ 770 beio->num_bios_sent = 1; 771 beio->send_complete = 1; 772 773 binuptime(&beio->ds_t0); 774 mtx_lock(&be_lun->io_lock); 775 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 776 mtx_unlock(&be_lun->io_lock); 777 778 (*dev_data->csw->d_strategy)(bio); 779} 780 781static void 782ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 783 struct ctl_be_block_io *beio, 784 uint64_t off, uint64_t len, int last) 785{ 786 struct bio *bio; 787 struct ctl_be_block_devdata *dev_data; 788 uint64_t maxlen; 789 790 dev_data = &be_lun->backend.dev; 791 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 792 while (len > 0) { 793 bio = g_alloc_bio(); 794 bio->bio_cmd = BIO_DELETE; 795 bio->bio_flags |= beio->bio_flags; 796 bio->bio_dev = dev_data->cdev; 797 bio->bio_offset = off; 798 bio->bio_length = MIN(len, maxlen); 799 bio->bio_data = 0; 800 bio->bio_done = ctl_be_block_biodone; 801 bio->bio_caller1 = beio; 802 bio->bio_pblkno = off / be_lun->blocksize; 803 804 off += bio->bio_length; 805 len -= bio->bio_length; 806 807 mtx_lock(&be_lun->io_lock); 808 beio->num_bios_sent++; 809 if (last && len == 0) 810 beio->send_complete = 1; 811 mtx_unlock(&be_lun->io_lock); 812 813 (*dev_data->csw->d_strategy)(bio); 814 } 815} 816 817static void 818ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 819 struct ctl_be_block_io *beio) 820{ 821 union ctl_io *io; 822 struct ctl_be_block_devdata *dev_data; 823 struct ctl_ptr_len_flags *ptrlen; 824 struct scsi_unmap_desc *buf, *end; 825 uint64_t len; 826 827 dev_data = &be_lun->backend.dev; 828 io = beio->io; 829 830 DPRINTF("entered\n"); 831 832 binuptime(&beio->ds_t0); 833 mtx_lock(&be_lun->io_lock); 834 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 835 mtx_unlock(&be_lun->io_lock); 836 837 if (beio->io_offset == -1) { 838 beio->io_len = 0; 839 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 840 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 841 end = buf + ptrlen->len / sizeof(*buf); 842 for (; buf < end; buf++) { 843 len = (uint64_t)scsi_4btoul(buf->length) * 844 be_lun->blocksize; 845 beio->io_len += len; 846 ctl_be_block_unmap_dev_range(be_lun, beio, 847 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 848 (end - buf < 2) ? TRUE : FALSE); 849 } 850 } else 851 ctl_be_block_unmap_dev_range(be_lun, beio, 852 beio->io_offset, beio->io_len, TRUE); 853} 854 855static void 856ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 857 struct ctl_be_block_io *beio) 858{ 859 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 860 int i; 861 struct bio *bio; 862 struct ctl_be_block_devdata *dev_data; 863 off_t cur_offset; 864 int max_iosize; 865 866 DPRINTF("entered\n"); 867 868 dev_data = &be_lun->backend.dev; 869 870 /* 871 * We have to limit our I/O size to the maximum supported by the 872 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 873 * set it properly, use DFLTPHYS. 874 */ 875 max_iosize = dev_data->cdev->si_iosize_max; 876 if (max_iosize < PAGE_SIZE) 877 max_iosize = DFLTPHYS; 878 879 cur_offset = beio->io_offset; 880 for (i = 0; i < beio->num_segs; i++) { 881 size_t cur_size; 882 uint8_t *cur_ptr; 883 884 cur_size = beio->sg_segs[i].len; 885 cur_ptr = beio->sg_segs[i].addr; 886 887 while (cur_size > 0) { 888 /* This can't fail, it's a blocking allocation. */ 889 bio = g_alloc_bio(); 890 891 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 892 893 bio->bio_cmd = beio->bio_cmd; 894 bio->bio_flags |= beio->bio_flags; 895 bio->bio_dev = dev_data->cdev; 896 bio->bio_caller1 = beio; 897 bio->bio_length = min(cur_size, max_iosize); 898 bio->bio_offset = cur_offset; 899 bio->bio_data = cur_ptr; 900 bio->bio_done = ctl_be_block_biodone; 901 bio->bio_pblkno = cur_offset / be_lun->blocksize; 902 903 cur_offset += bio->bio_length; 904 cur_ptr += bio->bio_length; 905 cur_size -= bio->bio_length; 906 907 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 908 beio->num_bios_sent++; 909 } 910 } 911 binuptime(&beio->ds_t0); 912 mtx_lock(&be_lun->io_lock); 913 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 914 beio->send_complete = 1; 915 mtx_unlock(&be_lun->io_lock); 916 917 /* 918 * Fire off all allocated requests! 919 */ 920 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 921 TAILQ_REMOVE(&queue, bio, bio_queue); 922 (*dev_data->csw->d_strategy)(bio); 923 } 924} 925 926static void 927ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 928{ 929 union ctl_io *io; 930 931 io = beio->io; 932 ctl_free_beio(beio); 933 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 934 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 935 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 936 ctl_config_write_done(io); 937 return; 938 } 939 940 ctl_be_block_config_write(io); 941} 942 943static void 944ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 945 union ctl_io *io) 946{ 947 struct ctl_be_block_io *beio; 948 struct ctl_be_block_softc *softc; 949 struct ctl_lba_len_flags *lbalen; 950 uint64_t len_left, lba; 951 int i, seglen; 952 uint8_t *buf, *end; 953 954 DPRINTF("entered\n"); 955 956 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 957 softc = be_lun->softc; 958 lbalen = ARGS(beio->io); 959 960 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP) || 961 (lbalen->flags & SWS_UNMAP && be_lun->unmap == NULL)) { 962 ctl_free_beio(beio); 963 ctl_set_invalid_field(&io->scsiio, 964 /*sks_valid*/ 1, 965 /*command*/ 1, 966 /*field*/ 1, 967 /*bit_valid*/ 0, 968 /*bit*/ 0); 969 ctl_config_write_done(io); 970 return; 971 } 972 973 /* 974 * If the I/O came down with an ordered or head of queue tag, set 975 * the BIO_ORDERED attribute. For head of queue tags, that's 976 * pretty much the best we can do. 977 */ 978 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 979 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 980 beio->bio_flags = BIO_ORDERED; 981 982 switch (io->scsiio.tag_type) { 983 case CTL_TAG_ORDERED: 984 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 985 break; 986 case CTL_TAG_HEAD_OF_QUEUE: 987 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 988 break; 989 case CTL_TAG_UNTAGGED: 990 case CTL_TAG_SIMPLE: 991 case CTL_TAG_ACA: 992 default: 993 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 994 break; 995 } 996 997 if (lbalen->flags & SWS_UNMAP) { 998 beio->io_offset = lbalen->lba * be_lun->blocksize; 999 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1000 beio->bio_cmd = BIO_DELETE; 1001 beio->ds_trans_type = DEVSTAT_FREE; 1002 1003 be_lun->unmap(be_lun, beio); 1004 return; 1005 } 1006 1007 beio->bio_cmd = BIO_WRITE; 1008 beio->ds_trans_type = DEVSTAT_WRITE; 1009 1010 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1011 (uintmax_t)lbalen->lba, lbalen->len); 1012 1013 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1014 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1015 1016 /* 1017 * Setup the S/G entry for this chunk. 1018 */ 1019 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1020 seglen -= seglen % be_lun->blocksize; 1021 beio->sg_segs[i].len = seglen; 1022 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1023 1024 DPRINTF("segment %d addr %p len %zd\n", i, 1025 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1026 1027 beio->num_segs++; 1028 len_left -= seglen; 1029 1030 buf = beio->sg_segs[i].addr; 1031 end = buf + seglen; 1032 for (; buf < end; buf += be_lun->blocksize) { 1033 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1034 if (lbalen->flags & SWS_LBDATA) 1035 scsi_ulto4b(lbalen->lba + lba, buf); 1036 lba++; 1037 } 1038 } 1039 1040 beio->io_offset = lbalen->lba * be_lun->blocksize; 1041 beio->io_len = lba * be_lun->blocksize; 1042 1043 /* We can not do all in one run. Correct and schedule rerun. */ 1044 if (len_left > 0) { 1045 lbalen->lba += lba; 1046 lbalen->len -= lba; 1047 beio->beio_cont = ctl_be_block_cw_done_ws; 1048 } 1049 1050 be_lun->dispatch(be_lun, beio); 1051} 1052 1053static void 1054ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1055 union ctl_io *io) 1056{ 1057 struct ctl_be_block_io *beio; 1058 struct ctl_be_block_softc *softc; 1059 struct ctl_ptr_len_flags *ptrlen; 1060 1061 DPRINTF("entered\n"); 1062 1063 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1064 softc = be_lun->softc; 1065 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1066 1067 if (ptrlen->flags != 0 || be_lun->unmap == NULL) { 1068 ctl_free_beio(beio); 1069 ctl_set_invalid_field(&io->scsiio, 1070 /*sks_valid*/ 0, 1071 /*command*/ 1, 1072 /*field*/ 0, 1073 /*bit_valid*/ 0, 1074 /*bit*/ 0); 1075 ctl_config_write_done(io); 1076 return; 1077 } 1078 1079 /* 1080 * If the I/O came down with an ordered or head of queue tag, set 1081 * the BIO_ORDERED attribute. For head of queue tags, that's 1082 * pretty much the best we can do. 1083 */ 1084 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1085 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1086 beio->bio_flags = BIO_ORDERED; 1087 1088 switch (io->scsiio.tag_type) { 1089 case CTL_TAG_ORDERED: 1090 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1091 break; 1092 case CTL_TAG_HEAD_OF_QUEUE: 1093 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1094 break; 1095 case CTL_TAG_UNTAGGED: 1096 case CTL_TAG_SIMPLE: 1097 case CTL_TAG_ACA: 1098 default: 1099 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1100 break; 1101 } 1102 1103 beio->io_len = 0; 1104 beio->io_offset = -1; 1105 1106 beio->bio_cmd = BIO_DELETE; 1107 beio->ds_trans_type = DEVSTAT_FREE; 1108 1109 DPRINTF("UNMAP\n"); 1110 1111 be_lun->unmap(be_lun, beio); 1112} 1113 1114static void 1115ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1116{ 1117 union ctl_io *io; 1118 1119 io = beio->io; 1120 ctl_free_beio(beio); 1121 ctl_config_write_done(io); 1122} 1123 1124static void 1125ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1126 union ctl_io *io) 1127{ 1128 struct ctl_be_block_io *beio; 1129 struct ctl_be_block_softc *softc; 1130 1131 DPRINTF("entered\n"); 1132 1133 softc = be_lun->softc; 1134 beio = ctl_alloc_beio(softc); 1135 beio->io = io; 1136 beio->lun = be_lun; 1137 beio->beio_cont = ctl_be_block_cw_done; 1138 PRIV(io)->ptr = (void *)beio; 1139 1140 switch (io->scsiio.cdb[0]) { 1141 case SYNCHRONIZE_CACHE: 1142 case SYNCHRONIZE_CACHE_16: 1143 beio->bio_cmd = BIO_FLUSH; 1144 beio->ds_trans_type = DEVSTAT_NO_DATA; 1145 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1146 beio->io_len = 0; 1147 be_lun->lun_flush(be_lun, beio); 1148 break; 1149 case WRITE_SAME_10: 1150 case WRITE_SAME_16: 1151 ctl_be_block_cw_dispatch_ws(be_lun, io); 1152 break; 1153 case UNMAP: 1154 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1155 break; 1156 default: 1157 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1158 break; 1159 } 1160} 1161 1162SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1163SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1164SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1165SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1166 1167static void 1168ctl_be_block_next(struct ctl_be_block_io *beio) 1169{ 1170 struct ctl_be_block_lun *be_lun; 1171 union ctl_io *io; 1172 1173 io = beio->io; 1174 be_lun = beio->lun; 1175 ctl_free_beio(beio); 1176 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1177 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1178 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1179 ctl_data_submit_done(io); 1180 return; 1181 } 1182 1183 io->io_hdr.status &= ~CTL_STATUS_MASK; 1184 io->io_hdr.status |= CTL_STATUS_NONE; 1185 1186 mtx_lock(&be_lun->queue_lock); 1187 /* 1188 * XXX KDM make sure that links is okay to use at this point. 1189 * Otherwise, we either need to add another field to ctl_io_hdr, 1190 * or deal with resource allocation here. 1191 */ 1192 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1193 mtx_unlock(&be_lun->queue_lock); 1194 1195 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1196} 1197 1198static void 1199ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1200 union ctl_io *io) 1201{ 1202 struct ctl_be_block_io *beio; 1203 struct ctl_be_block_softc *softc; 1204 struct ctl_lba_len_flags *lbalen; 1205 struct ctl_ptr_len_flags *bptrlen; 1206 uint64_t len_left, lbas; 1207 int i; 1208 1209 softc = be_lun->softc; 1210 1211 DPRINTF("entered\n"); 1212 1213 lbalen = ARGS(io); 1214 if (lbalen->flags & CTL_LLF_WRITE) { 1215 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1216 } else { 1217 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1218 } 1219 1220 beio = ctl_alloc_beio(softc); 1221 beio->io = io; 1222 beio->lun = be_lun; 1223 bptrlen = PRIV(io); 1224 bptrlen->ptr = (void *)beio; 1225 1226 /* 1227 * If the I/O came down with an ordered or head of queue tag, set 1228 * the BIO_ORDERED attribute. For head of queue tags, that's 1229 * pretty much the best we can do. 1230 * 1231 * XXX KDM we don't have a great way to easily know about the FUA 1232 * bit right now (it is decoded in ctl_read_write(), but we don't 1233 * pass that knowledge to the backend), and in any case we would 1234 * need to determine how to handle it. 1235 */ 1236 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1237 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1238 beio->bio_flags = BIO_ORDERED; 1239 1240 switch (io->scsiio.tag_type) { 1241 case CTL_TAG_ORDERED: 1242 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1243 break; 1244 case CTL_TAG_HEAD_OF_QUEUE: 1245 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1246 break; 1247 case CTL_TAG_UNTAGGED: 1248 case CTL_TAG_SIMPLE: 1249 case CTL_TAG_ACA: 1250 default: 1251 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1252 break; 1253 } 1254 1255 if (lbalen->flags & CTL_LLF_WRITE) { 1256 beio->bio_cmd = BIO_WRITE; 1257 beio->ds_trans_type = DEVSTAT_WRITE; 1258 } else { 1259 beio->bio_cmd = BIO_READ; 1260 beio->ds_trans_type = DEVSTAT_READ; 1261 } 1262 1263 DPRINTF("%s at LBA %jx len %u @%ju\n", 1264 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1265 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1266 if (lbalen->flags & CTL_LLF_COMPARE) 1267 lbas = CTLBLK_HALF_IO_SIZE; 1268 else 1269 lbas = CTLBLK_MAX_IO_SIZE; 1270 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1271 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1272 beio->io_len = lbas * be_lun->blocksize; 1273 bptrlen->len += lbas; 1274 1275 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1276 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1277 i, CTLBLK_MAX_SEGS)); 1278 1279 /* 1280 * Setup the S/G entry for this chunk. 1281 */ 1282 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1283 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1284 1285 DPRINTF("segment %d addr %p len %zd\n", i, 1286 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1287 1288 /* Set up second segment for compare operation. */ 1289 if (lbalen->flags & CTL_LLF_COMPARE) { 1290 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1291 beio->sg_segs[i].len; 1292 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1293 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1294 } 1295 1296 beio->num_segs++; 1297 len_left -= beio->sg_segs[i].len; 1298 } 1299 if (bptrlen->len < lbalen->len) 1300 beio->beio_cont = ctl_be_block_next; 1301 io->scsiio.be_move_done = ctl_be_block_move_done; 1302 /* For compare we have separate S/G lists for read and datamove. */ 1303 if (lbalen->flags & CTL_LLF_COMPARE) 1304 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1305 else 1306 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1307 io->scsiio.kern_data_len = beio->io_len; 1308 io->scsiio.kern_data_resid = 0; 1309 io->scsiio.kern_sg_entries = beio->num_segs; 1310 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1311 1312 /* 1313 * For the read case, we need to read the data into our buffers and 1314 * then we can send it back to the user. For the write case, we 1315 * need to get the data from the user first. 1316 */ 1317 if (beio->bio_cmd == BIO_READ) { 1318 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1319 be_lun->dispatch(be_lun, beio); 1320 } else { 1321 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1322#ifdef CTL_TIME_IO 1323 getbintime(&io->io_hdr.dma_start_bt); 1324#endif 1325 ctl_datamove(io); 1326 } 1327} 1328 1329static void 1330ctl_be_block_worker(void *context, int pending) 1331{ 1332 struct ctl_be_block_lun *be_lun; 1333 struct ctl_be_block_softc *softc; 1334 union ctl_io *io; 1335 1336 be_lun = (struct ctl_be_block_lun *)context; 1337 softc = be_lun->softc; 1338 1339 DPRINTF("entered\n"); 1340 1341 mtx_lock(&be_lun->queue_lock); 1342 for (;;) { 1343 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1344 if (io != NULL) { 1345 struct ctl_be_block_io *beio; 1346 1347 DPRINTF("datamove queue\n"); 1348 1349 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1350 ctl_io_hdr, links); 1351 1352 mtx_unlock(&be_lun->queue_lock); 1353 1354 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1355 1356 be_lun->dispatch(be_lun, beio); 1357 1358 mtx_lock(&be_lun->queue_lock); 1359 continue; 1360 } 1361 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1362 if (io != NULL) { 1363 1364 DPRINTF("config write queue\n"); 1365 1366 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1367 ctl_io_hdr, links); 1368 1369 mtx_unlock(&be_lun->queue_lock); 1370 1371 ctl_be_block_cw_dispatch(be_lun, io); 1372 1373 mtx_lock(&be_lun->queue_lock); 1374 continue; 1375 } 1376 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1377 if (io != NULL) { 1378 DPRINTF("input queue\n"); 1379 1380 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1381 ctl_io_hdr, links); 1382 mtx_unlock(&be_lun->queue_lock); 1383 1384 /* 1385 * We must drop the lock, since this routine and 1386 * its children may sleep. 1387 */ 1388 ctl_be_block_dispatch(be_lun, io); 1389 1390 mtx_lock(&be_lun->queue_lock); 1391 continue; 1392 } 1393 1394 /* 1395 * If we get here, there is no work left in the queues, so 1396 * just break out and let the task queue go to sleep. 1397 */ 1398 break; 1399 } 1400 mtx_unlock(&be_lun->queue_lock); 1401} 1402 1403/* 1404 * Entry point from CTL to the backend for I/O. We queue everything to a 1405 * work thread, so this just puts the I/O on a queue and wakes up the 1406 * thread. 1407 */ 1408static int 1409ctl_be_block_submit(union ctl_io *io) 1410{ 1411 struct ctl_be_block_lun *be_lun; 1412 struct ctl_be_lun *ctl_be_lun; 1413 1414 DPRINTF("entered\n"); 1415 1416 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1417 CTL_PRIV_BACKEND_LUN].ptr; 1418 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1419 1420 /* 1421 * Make sure we only get SCSI I/O. 1422 */ 1423 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1424 "%#x) encountered", io->io_hdr.io_type)); 1425 1426 PRIV(io)->len = 0; 1427 1428 mtx_lock(&be_lun->queue_lock); 1429 /* 1430 * XXX KDM make sure that links is okay to use at this point. 1431 * Otherwise, we either need to add another field to ctl_io_hdr, 1432 * or deal with resource allocation here. 1433 */ 1434 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1435 mtx_unlock(&be_lun->queue_lock); 1436 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1437 1438 return (CTL_RETVAL_COMPLETE); 1439} 1440 1441static int 1442ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1443 int flag, struct thread *td) 1444{ 1445 struct ctl_be_block_softc *softc; 1446 int error; 1447 1448 softc = &backend_block_softc; 1449 1450 error = 0; 1451 1452 switch (cmd) { 1453 case CTL_LUN_REQ: { 1454 struct ctl_lun_req *lun_req; 1455 1456 lun_req = (struct ctl_lun_req *)addr; 1457 1458 switch (lun_req->reqtype) { 1459 case CTL_LUNREQ_CREATE: 1460 error = ctl_be_block_create(softc, lun_req); 1461 break; 1462 case CTL_LUNREQ_RM: 1463 error = ctl_be_block_rm(softc, lun_req); 1464 break; 1465 case CTL_LUNREQ_MODIFY: 1466 error = ctl_be_block_modify(softc, lun_req); 1467 break; 1468 default: 1469 lun_req->status = CTL_LUN_ERROR; 1470 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1471 "%s: invalid LUN request type %d", __func__, 1472 lun_req->reqtype); 1473 break; 1474 } 1475 break; 1476 } 1477 default: 1478 error = ENOTTY; 1479 break; 1480 } 1481 1482 return (error); 1483} 1484 1485static int 1486ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1487{ 1488 struct ctl_be_block_filedata *file_data; 1489 struct ctl_lun_create_params *params; 1490 struct vattr vattr; 1491 int error; 1492 1493 error = 0; 1494 file_data = &be_lun->backend.file; 1495 params = &req->reqdata.create; 1496 1497 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1498 be_lun->dispatch = ctl_be_block_dispatch_file; 1499 be_lun->lun_flush = ctl_be_block_flush_file; 1500 1501 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1502 if (error != 0) { 1503 snprintf(req->error_str, sizeof(req->error_str), 1504 "error calling VOP_GETATTR() for file %s", 1505 be_lun->dev_path); 1506 return (error); 1507 } 1508 1509 /* 1510 * Verify that we have the ability to upgrade to exclusive 1511 * access on this file so we can trap errors at open instead 1512 * of reporting them during first access. 1513 */ 1514 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1515 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1516 if (be_lun->vn->v_iflag & VI_DOOMED) { 1517 error = EBADF; 1518 snprintf(req->error_str, sizeof(req->error_str), 1519 "error locking file %s", be_lun->dev_path); 1520 return (error); 1521 } 1522 } 1523 1524 1525 file_data->cred = crhold(curthread->td_ucred); 1526 if (params->lun_size_bytes != 0) 1527 be_lun->size_bytes = params->lun_size_bytes; 1528 else 1529 be_lun->size_bytes = vattr.va_size; 1530 /* 1531 * We set the multi thread flag for file operations because all 1532 * filesystems (in theory) are capable of allowing multiple readers 1533 * of a file at once. So we want to get the maximum possible 1534 * concurrency. 1535 */ 1536 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1537 1538 /* 1539 * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1540 * With ZFS, it is 131072 bytes. Block sizes that large don't work 1541 * with disklabel and UFS on FreeBSD at least. Large block sizes 1542 * may not work with other OSes as well. So just export a sector 1543 * size of 512 bytes, which should work with any OS or 1544 * application. Since our backing is a file, any block size will 1545 * work fine for the backing store. 1546 */ 1547#if 0 1548 be_lun->blocksize= vattr.va_blocksize; 1549#endif 1550 if (params->blocksize_bytes != 0) 1551 be_lun->blocksize = params->blocksize_bytes; 1552 else 1553 be_lun->blocksize = 512; 1554 1555 /* 1556 * Sanity check. The media size has to be at least one 1557 * sector long. 1558 */ 1559 if (be_lun->size_bytes < be_lun->blocksize) { 1560 error = EINVAL; 1561 snprintf(req->error_str, sizeof(req->error_str), 1562 "file %s size %ju < block size %u", be_lun->dev_path, 1563 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1564 } 1565 return (error); 1566} 1567 1568static int 1569ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1570{ 1571 struct ctl_lun_create_params *params; 1572 struct vattr vattr; 1573 struct cdev *dev; 1574 struct cdevsw *devsw; 1575 int error; 1576 off_t ps, pss, po, pos; 1577 1578 params = &req->reqdata.create; 1579 1580 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1581 be_lun->dispatch = ctl_be_block_dispatch_dev; 1582 be_lun->lun_flush = ctl_be_block_flush_dev; 1583 be_lun->unmap = ctl_be_block_unmap_dev; 1584 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1585 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1586 &be_lun->backend.dev.dev_ref); 1587 if (be_lun->backend.dev.csw == NULL) 1588 panic("Unable to retrieve device switch"); 1589 1590 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1591 if (error) { 1592 snprintf(req->error_str, sizeof(req->error_str), 1593 "%s: error getting vnode attributes for device %s", 1594 __func__, be_lun->dev_path); 1595 return (error); 1596 } 1597 1598 dev = be_lun->vn->v_rdev; 1599 devsw = dev->si_devsw; 1600 if (!devsw->d_ioctl) { 1601 snprintf(req->error_str, sizeof(req->error_str), 1602 "%s: no d_ioctl for device %s!", __func__, 1603 be_lun->dev_path); 1604 return (ENODEV); 1605 } 1606 1607 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1608 (caddr_t)&be_lun->blocksize, FREAD, 1609 curthread); 1610 if (error) { 1611 snprintf(req->error_str, sizeof(req->error_str), 1612 "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1613 "on %s!", __func__, error, be_lun->dev_path); 1614 return (error); 1615 } 1616 1617 /* 1618 * If the user has asked for a blocksize that is greater than the 1619 * backing device's blocksize, we can do it only if the blocksize 1620 * the user is asking for is an even multiple of the underlying 1621 * device's blocksize. 1622 */ 1623 if ((params->blocksize_bytes != 0) 1624 && (params->blocksize_bytes > be_lun->blocksize)) { 1625 uint32_t bs_multiple, tmp_blocksize; 1626 1627 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1628 1629 tmp_blocksize = bs_multiple * be_lun->blocksize; 1630 1631 if (tmp_blocksize == params->blocksize_bytes) { 1632 be_lun->blocksize = params->blocksize_bytes; 1633 } else { 1634 snprintf(req->error_str, sizeof(req->error_str), 1635 "%s: requested blocksize %u is not an even " 1636 "multiple of backing device blocksize %u", 1637 __func__, params->blocksize_bytes, 1638 be_lun->blocksize); 1639 return (EINVAL); 1640 1641 } 1642 } else if ((params->blocksize_bytes != 0) 1643 && (params->blocksize_bytes != be_lun->blocksize)) { 1644 snprintf(req->error_str, sizeof(req->error_str), 1645 "%s: requested blocksize %u < backing device " 1646 "blocksize %u", __func__, params->blocksize_bytes, 1647 be_lun->blocksize); 1648 return (EINVAL); 1649 } 1650 1651 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1652 (caddr_t)&be_lun->size_bytes, FREAD, 1653 curthread); 1654 if (error) { 1655 snprintf(req->error_str, sizeof(req->error_str), 1656 "%s: error %d returned for DIOCGMEDIASIZE " 1657 " ioctl on %s!", __func__, error, 1658 be_lun->dev_path); 1659 return (error); 1660 } 1661 1662 if (params->lun_size_bytes != 0) { 1663 if (params->lun_size_bytes > be_lun->size_bytes) { 1664 snprintf(req->error_str, sizeof(req->error_str), 1665 "%s: requested LUN size %ju > backing device " 1666 "size %ju", __func__, 1667 (uintmax_t)params->lun_size_bytes, 1668 (uintmax_t)be_lun->size_bytes); 1669 return (EINVAL); 1670 } 1671 1672 be_lun->size_bytes = params->lun_size_bytes; 1673 } 1674 1675 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1676 (caddr_t)&ps, FREAD, curthread); 1677 if (error) 1678 ps = po = 0; 1679 else { 1680 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1681 (caddr_t)&po, FREAD, curthread); 1682 if (error) 1683 po = 0; 1684 } 1685 pss = ps / be_lun->blocksize; 1686 pos = po / be_lun->blocksize; 1687 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1688 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1689 be_lun->pblockexp = fls(pss) - 1; 1690 be_lun->pblockoff = (pss - pos) % pss; 1691 } 1692 1693 return (0); 1694} 1695 1696static int 1697ctl_be_block_close(struct ctl_be_block_lun *be_lun) 1698{ 1699 DROP_GIANT(); 1700 if (be_lun->vn) { 1701 int flags = FREAD | FWRITE; 1702 1703 switch (be_lun->dev_type) { 1704 case CTL_BE_BLOCK_DEV: 1705 if (be_lun->backend.dev.csw) { 1706 dev_relthread(be_lun->backend.dev.cdev, 1707 be_lun->backend.dev.dev_ref); 1708 be_lun->backend.dev.csw = NULL; 1709 be_lun->backend.dev.cdev = NULL; 1710 } 1711 break; 1712 case CTL_BE_BLOCK_FILE: 1713 break; 1714 case CTL_BE_BLOCK_NONE: 1715 break; 1716 default: 1717 panic("Unexpected backend type."); 1718 break; 1719 } 1720 1721 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1722 be_lun->vn = NULL; 1723 1724 switch (be_lun->dev_type) { 1725 case CTL_BE_BLOCK_DEV: 1726 break; 1727 case CTL_BE_BLOCK_FILE: 1728 if (be_lun->backend.file.cred != NULL) { 1729 crfree(be_lun->backend.file.cred); 1730 be_lun->backend.file.cred = NULL; 1731 } 1732 break; 1733 case CTL_BE_BLOCK_NONE: 1734 break; 1735 default: 1736 panic("Unexpected backend type."); 1737 break; 1738 } 1739 } 1740 PICKUP_GIANT(); 1741 1742 return (0); 1743} 1744 1745static int 1746ctl_be_block_open(struct ctl_be_block_softc *softc, 1747 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1748{ 1749 struct nameidata nd; 1750 int flags; 1751 int error; 1752 1753 /* 1754 * XXX KDM allow a read-only option? 1755 */ 1756 flags = FREAD | FWRITE; 1757 error = 0; 1758 1759 if (rootvnode == NULL) { 1760 snprintf(req->error_str, sizeof(req->error_str), 1761 "%s: Root filesystem is not mounted", __func__); 1762 return (1); 1763 } 1764 1765 if (!curthread->td_proc->p_fd->fd_cdir) { 1766 curthread->td_proc->p_fd->fd_cdir = rootvnode; 1767 VREF(rootvnode); 1768 } 1769 if (!curthread->td_proc->p_fd->fd_rdir) { 1770 curthread->td_proc->p_fd->fd_rdir = rootvnode; 1771 VREF(rootvnode); 1772 } 1773 if (!curthread->td_proc->p_fd->fd_jdir) { 1774 curthread->td_proc->p_fd->fd_jdir = rootvnode; 1775 VREF(rootvnode); 1776 } 1777 1778 again: 1779 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1780 error = vn_open(&nd, &flags, 0, NULL); 1781 if (error) { 1782 /* 1783 * This is the only reasonable guess we can make as far as 1784 * path if the user doesn't give us a fully qualified path. 1785 * If they want to specify a file, they need to specify the 1786 * full path. 1787 */ 1788 if (be_lun->dev_path[0] != '/') { 1789 char *dev_path = "/dev/"; 1790 char *dev_name; 1791 1792 /* Try adding device path at beginning of name */ 1793 dev_name = malloc(strlen(be_lun->dev_path) 1794 + strlen(dev_path) + 1, 1795 M_CTLBLK, M_WAITOK); 1796 if (dev_name) { 1797 sprintf(dev_name, "%s%s", dev_path, 1798 be_lun->dev_path); 1799 free(be_lun->dev_path, M_CTLBLK); 1800 be_lun->dev_path = dev_name; 1801 goto again; 1802 } 1803 } 1804 snprintf(req->error_str, sizeof(req->error_str), 1805 "%s: error opening %s", __func__, be_lun->dev_path); 1806 return (error); 1807 } 1808 1809 NDFREE(&nd, NDF_ONLY_PNBUF); 1810 1811 be_lun->vn = nd.ni_vp; 1812 1813 /* We only support disks and files. */ 1814 if (vn_isdisk(be_lun->vn, &error)) { 1815 error = ctl_be_block_open_dev(be_lun, req); 1816 } else if (be_lun->vn->v_type == VREG) { 1817 error = ctl_be_block_open_file(be_lun, req); 1818 } else { 1819 error = EINVAL; 1820 snprintf(req->error_str, sizeof(req->error_str), 1821 "%s is not a disk or plain file", be_lun->dev_path); 1822 } 1823 VOP_UNLOCK(be_lun->vn, 0); 1824 1825 if (error != 0) { 1826 ctl_be_block_close(be_lun); 1827 return (error); 1828 } 1829 1830 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1831 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1832 1833 return (0); 1834} 1835 1836static int 1837ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1838{ 1839 struct ctl_be_block_lun *be_lun; 1840 struct ctl_lun_create_params *params; 1841 char num_thread_str[16]; 1842 char tmpstr[32]; 1843 char *value; 1844 int retval, num_threads, unmap; 1845 int tmp_num_threads; 1846 1847 params = &req->reqdata.create; 1848 retval = 0; 1849 1850 num_threads = cbb_num_threads; 1851 1852 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1853 1854 be_lun->softc = softc; 1855 STAILQ_INIT(&be_lun->input_queue); 1856 STAILQ_INIT(&be_lun->config_write_queue); 1857 STAILQ_INIT(&be_lun->datamove_queue); 1858 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1859 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1860 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1861 ctl_init_opts(&be_lun->ctl_be_lun.options, 1862 req->num_be_args, req->kern_be_args); 1863 1864 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1865 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1866 1867 if (be_lun->lun_zone == NULL) { 1868 snprintf(req->error_str, sizeof(req->error_str), 1869 "%s: error allocating UMA zone", __func__); 1870 goto bailout_error; 1871 } 1872 1873 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1874 be_lun->ctl_be_lun.lun_type = params->device_type; 1875 else 1876 be_lun->ctl_be_lun.lun_type = T_DIRECT; 1877 1878 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1879 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1880 if (value == NULL) { 1881 snprintf(req->error_str, sizeof(req->error_str), 1882 "%s: no file argument specified", __func__); 1883 goto bailout_error; 1884 } 1885 be_lun->dev_path = strdup(value, M_CTLBLK); 1886 1887 retval = ctl_be_block_open(softc, be_lun, req); 1888 if (retval != 0) { 1889 retval = 0; 1890 goto bailout_error; 1891 } 1892 1893 /* 1894 * Tell the user the size of the file/device. 1895 */ 1896 params->lun_size_bytes = be_lun->size_bytes; 1897 1898 /* 1899 * The maximum LBA is the size - 1. 1900 */ 1901 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1902 } else { 1903 /* 1904 * For processor devices, we don't have any size. 1905 */ 1906 be_lun->blocksize = 0; 1907 be_lun->pblockexp = 0; 1908 be_lun->pblockoff = 0; 1909 be_lun->size_blocks = 0; 1910 be_lun->size_bytes = 0; 1911 be_lun->ctl_be_lun.maxlba = 0; 1912 params->lun_size_bytes = 0; 1913 1914 /* 1915 * Default to just 1 thread for processor devices. 1916 */ 1917 num_threads = 1; 1918 } 1919 1920 /* 1921 * XXX This searching loop might be refactored to be combined with 1922 * the loop above, 1923 */ 1924 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 1925 if (value != NULL) { 1926 tmp_num_threads = strtol(value, NULL, 0); 1927 1928 /* 1929 * We don't let the user specify less than one 1930 * thread, but hope he's clueful enough not to 1931 * specify 1000 threads. 1932 */ 1933 if (tmp_num_threads < 1) { 1934 snprintf(req->error_str, sizeof(req->error_str), 1935 "%s: invalid number of threads %s", 1936 __func__, num_thread_str); 1937 goto bailout_error; 1938 } 1939 num_threads = tmp_num_threads; 1940 } 1941 unmap = 0; 1942 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 1943 if (value != NULL && strcmp(value, "on") == 0) 1944 unmap = 1; 1945 1946 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1947 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1948 if (unmap) 1949 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 1950 be_lun->ctl_be_lun.be_lun = be_lun; 1951 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1952 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 1953 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 1954 /* Tell the user the blocksize we ended up using */ 1955 params->blocksize_bytes = be_lun->blocksize; 1956 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1957 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1958 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1959 } else 1960 be_lun->ctl_be_lun.req_lun_id = 0; 1961 1962 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1963 be_lun->ctl_be_lun.lun_config_status = 1964 ctl_be_block_lun_config_status; 1965 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1966 1967 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1968 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1969 softc->num_luns); 1970 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1971 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1972 sizeof(tmpstr))); 1973 1974 /* Tell the user what we used for a serial number */ 1975 strncpy((char *)params->serial_num, tmpstr, 1976 ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1977 } else { 1978 strncpy((char *)be_lun->ctl_be_lun.serial_num, 1979 params->serial_num, 1980 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1981 sizeof(params->serial_num))); 1982 } 1983 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1984 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1985 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1986 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1987 sizeof(tmpstr))); 1988 1989 /* Tell the user what we used for a device ID */ 1990 strncpy((char *)params->device_id, tmpstr, 1991 ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1992 } else { 1993 strncpy((char *)be_lun->ctl_be_lun.device_id, 1994 params->device_id, 1995 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1996 sizeof(params->device_id))); 1997 } 1998 1999 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2000 2001 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2002 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2003 2004 if (be_lun->io_taskqueue == NULL) { 2005 snprintf(req->error_str, sizeof(req->error_str), 2006 "%s: Unable to create taskqueue", __func__); 2007 goto bailout_error; 2008 } 2009 2010 /* 2011 * Note that we start the same number of threads by default for 2012 * both the file case and the block device case. For the file 2013 * case, we need multiple threads to allow concurrency, because the 2014 * vnode interface is designed to be a blocking interface. For the 2015 * block device case, ZFS zvols at least will block the caller's 2016 * context in many instances, and so we need multiple threads to 2017 * overcome that problem. Other block devices don't need as many 2018 * threads, but they shouldn't cause too many problems. 2019 * 2020 * If the user wants to just have a single thread for a block 2021 * device, he can specify that when the LUN is created, or change 2022 * the tunable/sysctl to alter the default number of threads. 2023 */ 2024 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2025 /*num threads*/num_threads, 2026 /*priority*/PWAIT, 2027 /*thread name*/ 2028 "%s taskq", be_lun->lunname); 2029 2030 if (retval != 0) 2031 goto bailout_error; 2032 2033 be_lun->num_threads = num_threads; 2034 2035 mtx_lock(&softc->lock); 2036 softc->num_luns++; 2037 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2038 2039 mtx_unlock(&softc->lock); 2040 2041 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2042 if (retval != 0) { 2043 mtx_lock(&softc->lock); 2044 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2045 links); 2046 softc->num_luns--; 2047 mtx_unlock(&softc->lock); 2048 snprintf(req->error_str, sizeof(req->error_str), 2049 "%s: ctl_add_lun() returned error %d, see dmesg for " 2050 "details", __func__, retval); 2051 retval = 0; 2052 goto bailout_error; 2053 } 2054 2055 mtx_lock(&softc->lock); 2056 2057 /* 2058 * Tell the config_status routine that we're waiting so it won't 2059 * clean up the LUN in the event of an error. 2060 */ 2061 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2062 2063 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2064 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2065 if (retval == EINTR) 2066 break; 2067 } 2068 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2069 2070 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2071 snprintf(req->error_str, sizeof(req->error_str), 2072 "%s: LUN configuration error, see dmesg for details", 2073 __func__); 2074 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2075 links); 2076 softc->num_luns--; 2077 mtx_unlock(&softc->lock); 2078 goto bailout_error; 2079 } else { 2080 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2081 } 2082 2083 mtx_unlock(&softc->lock); 2084 2085 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2086 be_lun->blocksize, 2087 DEVSTAT_ALL_SUPPORTED, 2088 be_lun->ctl_be_lun.lun_type 2089 | DEVSTAT_TYPE_IF_OTHER, 2090 DEVSTAT_PRIORITY_OTHER); 2091 2092 2093 req->status = CTL_LUN_OK; 2094 2095 return (retval); 2096 2097bailout_error: 2098 req->status = CTL_LUN_ERROR; 2099 2100 if (be_lun->io_taskqueue != NULL) 2101 taskqueue_free(be_lun->io_taskqueue); 2102 ctl_be_block_close(be_lun); 2103 if (be_lun->dev_path != NULL) 2104 free(be_lun->dev_path, M_CTLBLK); 2105 if (be_lun->lun_zone != NULL) 2106 uma_zdestroy(be_lun->lun_zone); 2107 ctl_free_opts(&be_lun->ctl_be_lun.options); 2108 mtx_destroy(&be_lun->queue_lock); 2109 mtx_destroy(&be_lun->io_lock); 2110 free(be_lun, M_CTLBLK); 2111 2112 return (retval); 2113} 2114 2115static int 2116ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2117{ 2118 struct ctl_lun_rm_params *params; 2119 struct ctl_be_block_lun *be_lun; 2120 int retval; 2121 2122 params = &req->reqdata.rm; 2123 2124 mtx_lock(&softc->lock); 2125 2126 be_lun = NULL; 2127 2128 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2129 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2130 break; 2131 } 2132 mtx_unlock(&softc->lock); 2133 2134 if (be_lun == NULL) { 2135 snprintf(req->error_str, sizeof(req->error_str), 2136 "%s: LUN %u is not managed by the block backend", 2137 __func__, params->lun_id); 2138 goto bailout_error; 2139 } 2140 2141 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2142 2143 if (retval != 0) { 2144 snprintf(req->error_str, sizeof(req->error_str), 2145 "%s: error %d returned from ctl_disable_lun() for " 2146 "LUN %d", __func__, retval, params->lun_id); 2147 goto bailout_error; 2148 2149 } 2150 2151 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2152 if (retval != 0) { 2153 snprintf(req->error_str, sizeof(req->error_str), 2154 "%s: error %d returned from ctl_invalidate_lun() for " 2155 "LUN %d", __func__, retval, params->lun_id); 2156 goto bailout_error; 2157 } 2158 2159 mtx_lock(&softc->lock); 2160 2161 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2162 2163 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2164 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2165 if (retval == EINTR) 2166 break; 2167 } 2168 2169 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2170 2171 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2172 snprintf(req->error_str, sizeof(req->error_str), 2173 "%s: interrupted waiting for LUN to be freed", 2174 __func__); 2175 mtx_unlock(&softc->lock); 2176 goto bailout_error; 2177 } 2178 2179 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2180 2181 softc->num_luns--; 2182 mtx_unlock(&softc->lock); 2183 2184 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2185 2186 taskqueue_free(be_lun->io_taskqueue); 2187 2188 ctl_be_block_close(be_lun); 2189 2190 if (be_lun->disk_stats != NULL) 2191 devstat_remove_entry(be_lun->disk_stats); 2192 2193 uma_zdestroy(be_lun->lun_zone); 2194 2195 ctl_free_opts(&be_lun->ctl_be_lun.options); 2196 free(be_lun->dev_path, M_CTLBLK); 2197 mtx_destroy(&be_lun->queue_lock); 2198 mtx_destroy(&be_lun->io_lock); 2199 free(be_lun, M_CTLBLK); 2200 2201 req->status = CTL_LUN_OK; 2202 2203 return (0); 2204 2205bailout_error: 2206 2207 req->status = CTL_LUN_ERROR; 2208 2209 return (0); 2210} 2211 2212static int 2213ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2214 struct ctl_lun_req *req) 2215{ 2216 struct vattr vattr; 2217 int error; 2218 struct ctl_lun_modify_params *params; 2219 2220 params = &req->reqdata.modify; 2221 2222 if (params->lun_size_bytes != 0) { 2223 be_lun->size_bytes = params->lun_size_bytes; 2224 } else { 2225 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2226 if (error != 0) { 2227 snprintf(req->error_str, sizeof(req->error_str), 2228 "error calling VOP_GETATTR() for file %s", 2229 be_lun->dev_path); 2230 return (error); 2231 } 2232 2233 be_lun->size_bytes = vattr.va_size; 2234 } 2235 2236 return (0); 2237} 2238 2239static int 2240ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2241 struct ctl_lun_req *req) 2242{ 2243 struct cdev *dev; 2244 struct cdevsw *devsw; 2245 int error; 2246 struct ctl_lun_modify_params *params; 2247 uint64_t size_bytes; 2248 2249 params = &req->reqdata.modify; 2250 2251 dev = be_lun->vn->v_rdev; 2252 devsw = dev->si_devsw; 2253 if (!devsw->d_ioctl) { 2254 snprintf(req->error_str, sizeof(req->error_str), 2255 "%s: no d_ioctl for device %s!", __func__, 2256 be_lun->dev_path); 2257 return (ENODEV); 2258 } 2259 2260 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2261 (caddr_t)&size_bytes, FREAD, 2262 curthread); 2263 if (error) { 2264 snprintf(req->error_str, sizeof(req->error_str), 2265 "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2266 "on %s!", __func__, error, be_lun->dev_path); 2267 return (error); 2268 } 2269 2270 if (params->lun_size_bytes != 0) { 2271 if (params->lun_size_bytes > size_bytes) { 2272 snprintf(req->error_str, sizeof(req->error_str), 2273 "%s: requested LUN size %ju > backing device " 2274 "size %ju", __func__, 2275 (uintmax_t)params->lun_size_bytes, 2276 (uintmax_t)size_bytes); 2277 return (EINVAL); 2278 } 2279 2280 be_lun->size_bytes = params->lun_size_bytes; 2281 } else { 2282 be_lun->size_bytes = size_bytes; 2283 } 2284 2285 return (0); 2286} 2287 2288static int 2289ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2290{ 2291 struct ctl_lun_modify_params *params; 2292 struct ctl_be_block_lun *be_lun; 2293 int error; 2294 2295 params = &req->reqdata.modify; 2296 2297 mtx_lock(&softc->lock); 2298 2299 be_lun = NULL; 2300 2301 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2302 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2303 break; 2304 } 2305 mtx_unlock(&softc->lock); 2306 2307 if (be_lun == NULL) { 2308 snprintf(req->error_str, sizeof(req->error_str), 2309 "%s: LUN %u is not managed by the block backend", 2310 __func__, params->lun_id); 2311 goto bailout_error; 2312 } 2313 2314 if (params->lun_size_bytes != 0) { 2315 if (params->lun_size_bytes < be_lun->blocksize) { 2316 snprintf(req->error_str, sizeof(req->error_str), 2317 "%s: LUN size %ju < blocksize %u", __func__, 2318 params->lun_size_bytes, be_lun->blocksize); 2319 goto bailout_error; 2320 } 2321 } 2322 2323 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2324 2325 if (be_lun->vn->v_type == VREG) 2326 error = ctl_be_block_modify_file(be_lun, req); 2327 else 2328 error = ctl_be_block_modify_dev(be_lun, req); 2329 2330 VOP_UNLOCK(be_lun->vn, 0); 2331 2332 if (error != 0) 2333 goto bailout_error; 2334 2335 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2336 2337 /* 2338 * The maximum LBA is the size - 1. 2339 * 2340 * XXX: Note that this field is being updated without locking, 2341 * which might cause problems on 32-bit architectures. 2342 */ 2343 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2344 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2345 2346 /* Tell the user the exact size we ended up using */ 2347 params->lun_size_bytes = be_lun->size_bytes; 2348 2349 req->status = CTL_LUN_OK; 2350 2351 return (0); 2352 2353bailout_error: 2354 req->status = CTL_LUN_ERROR; 2355 2356 return (0); 2357} 2358 2359static void 2360ctl_be_block_lun_shutdown(void *be_lun) 2361{ 2362 struct ctl_be_block_lun *lun; 2363 struct ctl_be_block_softc *softc; 2364 2365 lun = (struct ctl_be_block_lun *)be_lun; 2366 2367 softc = lun->softc; 2368 2369 mtx_lock(&softc->lock); 2370 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2371 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2372 wakeup(lun); 2373 mtx_unlock(&softc->lock); 2374 2375} 2376 2377static void 2378ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2379{ 2380 struct ctl_be_block_lun *lun; 2381 struct ctl_be_block_softc *softc; 2382 2383 lun = (struct ctl_be_block_lun *)be_lun; 2384 softc = lun->softc; 2385 2386 if (status == CTL_LUN_CONFIG_OK) { 2387 mtx_lock(&softc->lock); 2388 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2389 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2390 wakeup(lun); 2391 mtx_unlock(&softc->lock); 2392 2393 /* 2394 * We successfully added the LUN, attempt to enable it. 2395 */ 2396 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2397 printf("%s: ctl_enable_lun() failed!\n", __func__); 2398 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2399 printf("%s: ctl_invalidate_lun() failed!\n", 2400 __func__); 2401 } 2402 } 2403 2404 return; 2405 } 2406 2407 2408 mtx_lock(&softc->lock); 2409 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2410 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2411 wakeup(lun); 2412 mtx_unlock(&softc->lock); 2413} 2414 2415 2416static int 2417ctl_be_block_config_write(union ctl_io *io) 2418{ 2419 struct ctl_be_block_lun *be_lun; 2420 struct ctl_be_lun *ctl_be_lun; 2421 int retval; 2422 2423 retval = 0; 2424 2425 DPRINTF("entered\n"); 2426 2427 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2428 CTL_PRIV_BACKEND_LUN].ptr; 2429 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2430 2431 switch (io->scsiio.cdb[0]) { 2432 case SYNCHRONIZE_CACHE: 2433 case SYNCHRONIZE_CACHE_16: 2434 case WRITE_SAME_10: 2435 case WRITE_SAME_16: 2436 case UNMAP: 2437 /* 2438 * The upper level CTL code will filter out any CDBs with 2439 * the immediate bit set and return the proper error. 2440 * 2441 * We don't really need to worry about what LBA range the 2442 * user asked to be synced out. When they issue a sync 2443 * cache command, we'll sync out the whole thing. 2444 */ 2445 mtx_lock(&be_lun->queue_lock); 2446 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2447 links); 2448 mtx_unlock(&be_lun->queue_lock); 2449 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2450 break; 2451 case START_STOP_UNIT: { 2452 struct scsi_start_stop_unit *cdb; 2453 2454 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2455 2456 if (cdb->how & SSS_START) 2457 retval = ctl_start_lun(ctl_be_lun); 2458 else { 2459 retval = ctl_stop_lun(ctl_be_lun); 2460 /* 2461 * XXX KDM Copan-specific offline behavior. 2462 * Figure out a reasonable way to port this? 2463 */ 2464#ifdef NEEDTOPORT 2465 if ((retval == 0) 2466 && (cdb->byte2 & SSS_ONOFFLINE)) 2467 retval = ctl_lun_offline(ctl_be_lun); 2468#endif 2469 } 2470 2471 /* 2472 * In general, the above routines should not fail. They 2473 * just set state for the LUN. So we've got something 2474 * pretty wrong here if we can't start or stop the LUN. 2475 */ 2476 if (retval != 0) { 2477 ctl_set_internal_failure(&io->scsiio, 2478 /*sks_valid*/ 1, 2479 /*retry_count*/ 0xf051); 2480 retval = CTL_RETVAL_COMPLETE; 2481 } else { 2482 ctl_set_success(&io->scsiio); 2483 } 2484 ctl_config_write_done(io); 2485 break; 2486 } 2487 default: 2488 ctl_set_invalid_opcode(&io->scsiio); 2489 ctl_config_write_done(io); 2490 retval = CTL_RETVAL_COMPLETE; 2491 break; 2492 } 2493 2494 return (retval); 2495 2496} 2497 2498static int 2499ctl_be_block_config_read(union ctl_io *io) 2500{ 2501 return (0); 2502} 2503 2504static int 2505ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2506{ 2507 struct ctl_be_block_lun *lun; 2508 int retval; 2509 2510 lun = (struct ctl_be_block_lun *)be_lun; 2511 retval = 0; 2512 2513 retval = sbuf_printf(sb, "\t<num_threads>"); 2514 2515 if (retval != 0) 2516 goto bailout; 2517 2518 retval = sbuf_printf(sb, "%d", lun->num_threads); 2519 2520 if (retval != 0) 2521 goto bailout; 2522 2523 retval = sbuf_printf(sb, "</num_threads>\n"); 2524 2525bailout: 2526 2527 return (retval); 2528} 2529 2530int 2531ctl_be_block_init(void) 2532{ 2533 struct ctl_be_block_softc *softc; 2534 int retval; 2535 2536 softc = &backend_block_softc; 2537 retval = 0; 2538 2539 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2540 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2541 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2542 STAILQ_INIT(&softc->disk_list); 2543 STAILQ_INIT(&softc->lun_list); 2544 2545 return (retval); 2546} 2547