ctl_backend_block.c revision 288774
1/*- 2 * Copyright (c) 2003 Silicon Graphics International Corp. 3 * Copyright (c) 2009-2011 Spectra Logic Corporation 4 * Copyright (c) 2012 The FreeBSD Foundation 5 * All rights reserved. 6 * 7 * Portions of this software were developed by Edward Tomasz Napierala 8 * under sponsorship from the FreeBSD Foundation. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions, and the following disclaimer, 15 * without modification. 16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17 * substantially similar to the "NO WARRANTY" disclaimer below 18 * ("Disclaimer") and any redistribution must be conditioned upon 19 * including a substantially similar Disclaimer requirement for further 20 * binary redistribution. 21 * 22 * NO WARRANTY 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGES. 34 * 35 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36 */ 37/* 38 * CAM Target Layer driver backend for block devices. 39 * 40 * Author: Ken Merry <ken@FreeBSD.org> 41 */ 42#include <sys/cdefs.h> 43__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl_backend_block.c 288774 2015-10-05 10:41:08Z mav $"); 44 45#include <opt_kdtrace.h> 46 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/kernel.h> 50#include <sys/types.h> 51#include <sys/kthread.h> 52#include <sys/bio.h> 53#include <sys/fcntl.h> 54#include <sys/limits.h> 55#include <sys/lock.h> 56#include <sys/mutex.h> 57#include <sys/condvar.h> 58#include <sys/malloc.h> 59#include <sys/conf.h> 60#include <sys/ioccom.h> 61#include <sys/queue.h> 62#include <sys/sbuf.h> 63#include <sys/endian.h> 64#include <sys/uio.h> 65#include <sys/buf.h> 66#include <sys/taskqueue.h> 67#include <sys/vnode.h> 68#include <sys/namei.h> 69#include <sys/mount.h> 70#include <sys/disk.h> 71#include <sys/fcntl.h> 72#include <sys/filedesc.h> 73#include <sys/filio.h> 74#include <sys/proc.h> 75#include <sys/pcpu.h> 76#include <sys/module.h> 77#include <sys/sdt.h> 78#include <sys/devicestat.h> 79#include <sys/sysctl.h> 80 81#include <geom/geom.h> 82 83#include <cam/cam.h> 84#include <cam/scsi/scsi_all.h> 85#include <cam/scsi/scsi_da.h> 86#include <cam/ctl/ctl_io.h> 87#include <cam/ctl/ctl.h> 88#include <cam/ctl/ctl_backend.h> 89#include <cam/ctl/ctl_ioctl.h> 90#include <cam/ctl/ctl_ha.h> 91#include <cam/ctl/ctl_scsi_all.h> 92#include <cam/ctl/ctl_private.h> 93#include <cam/ctl/ctl_error.h> 94 95/* 96 * The idea here is that we'll allocate enough S/G space to hold a 1MB 97 * I/O. If we get an I/O larger than that, we'll split it. 98 */ 99#define CTLBLK_HALF_IO_SIZE (512 * 1024) 100#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 101#define CTLBLK_MAX_SEG MAXPHYS 102#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 103#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 104 105#ifdef CTLBLK_DEBUG 106#define DPRINTF(fmt, args...) \ 107 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 108#else 109#define DPRINTF(fmt, args...) do {} while(0) 110#endif 111 112#define PRIV(io) \ 113 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 114#define ARGS(io) \ 115 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 116 117SDT_PROVIDER_DEFINE(cbb); 118 119typedef enum { 120 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 121 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 122 CTL_BE_BLOCK_LUN_WAITING = 0x04, 123} ctl_be_block_lun_flags; 124 125typedef enum { 126 CTL_BE_BLOCK_NONE, 127 CTL_BE_BLOCK_DEV, 128 CTL_BE_BLOCK_FILE 129} ctl_be_block_type; 130 131struct ctl_be_block_filedata { 132 struct ucred *cred; 133}; 134 135union ctl_be_block_bedata { 136 struct ctl_be_block_filedata file; 137}; 138 139struct ctl_be_block_io; 140struct ctl_be_block_lun; 141 142typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 143 struct ctl_be_block_io *beio); 144typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 145 const char *attrname); 146 147/* 148 * Backend LUN structure. There is a 1:1 mapping between a block device 149 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 150 */ 151struct ctl_be_block_lun { 152 struct ctl_lun_create_params params; 153 char lunname[32]; 154 char *dev_path; 155 ctl_be_block_type dev_type; 156 struct vnode *vn; 157 union ctl_be_block_bedata backend; 158 cbb_dispatch_t dispatch; 159 cbb_dispatch_t lun_flush; 160 cbb_dispatch_t unmap; 161 cbb_dispatch_t get_lba_status; 162 cbb_getattr_t getattr; 163 uma_zone_t lun_zone; 164 uint64_t size_blocks; 165 uint64_t size_bytes; 166 struct ctl_be_block_softc *softc; 167 struct devstat *disk_stats; 168 ctl_be_block_lun_flags flags; 169 STAILQ_ENTRY(ctl_be_block_lun) links; 170 struct ctl_be_lun cbe_lun; 171 struct taskqueue *io_taskqueue; 172 struct task io_task; 173 int num_threads; 174 STAILQ_HEAD(, ctl_io_hdr) input_queue; 175 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 176 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 177 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 178 struct mtx_padalign io_lock; 179 struct mtx_padalign queue_lock; 180}; 181 182/* 183 * Overall softc structure for the block backend module. 184 */ 185struct ctl_be_block_softc { 186 struct mtx lock; 187 int num_luns; 188 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 189}; 190 191static struct ctl_be_block_softc backend_block_softc; 192 193/* 194 * Per-I/O information. 195 */ 196struct ctl_be_block_io { 197 union ctl_io *io; 198 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 199 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 200 int bio_cmd; 201 int num_segs; 202 int num_bios_sent; 203 int num_bios_done; 204 int send_complete; 205 int num_errors; 206 struct bintime ds_t0; 207 devstat_tag_type ds_tag_type; 208 devstat_trans_flags ds_trans_type; 209 uint64_t io_len; 210 uint64_t io_offset; 211 int io_arg; 212 struct ctl_be_block_softc *softc; 213 struct ctl_be_block_lun *lun; 214 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 215}; 216 217extern struct ctl_softc *control_softc; 218 219static int cbb_num_threads = 14; 220TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 221SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 222 "CAM Target Layer Block Backend"); 223SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 224 &cbb_num_threads, 0, "Number of threads per backing file"); 225 226static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 227static void ctl_free_beio(struct ctl_be_block_io *beio); 228static void ctl_complete_beio(struct ctl_be_block_io *beio); 229static int ctl_be_block_move_done(union ctl_io *io); 230static void ctl_be_block_biodone(struct bio *bio); 231static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 232 struct ctl_be_block_io *beio); 233static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 234 struct ctl_be_block_io *beio); 235static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 236 struct ctl_be_block_io *beio); 237static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 238 const char *attrname); 239static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 240 struct ctl_be_block_io *beio); 241static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 242 struct ctl_be_block_io *beio); 243static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 244 struct ctl_be_block_io *beio); 245static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 246 const char *attrname); 247static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 248 union ctl_io *io); 249static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 250 union ctl_io *io); 251static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 252 union ctl_io *io); 253static void ctl_be_block_worker(void *context, int pending); 254static int ctl_be_block_submit(union ctl_io *io); 255static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 256 int flag, struct thread *td); 257static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 258 struct ctl_lun_req *req); 259static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 260 struct ctl_lun_req *req); 261static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 262static int ctl_be_block_open(struct ctl_be_block_softc *softc, 263 struct ctl_be_block_lun *be_lun, 264 struct ctl_lun_req *req); 265static int ctl_be_block_create(struct ctl_be_block_softc *softc, 266 struct ctl_lun_req *req); 267static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 268 struct ctl_lun_req *req); 269static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 270 struct ctl_lun_req *req); 271static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 272 struct ctl_lun_req *req); 273static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 274 struct ctl_lun_req *req); 275static void ctl_be_block_lun_shutdown(void *be_lun); 276static void ctl_be_block_lun_config_status(void *be_lun, 277 ctl_lun_config_status status); 278static int ctl_be_block_config_write(union ctl_io *io); 279static int ctl_be_block_config_read(union ctl_io *io); 280static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 281static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 282int ctl_be_block_init(void); 283 284static struct ctl_backend_driver ctl_be_block_driver = 285{ 286 .name = "block", 287 .flags = CTL_BE_FLAG_HAS_CONFIG, 288 .init = ctl_be_block_init, 289 .data_submit = ctl_be_block_submit, 290 .data_move_done = ctl_be_block_move_done, 291 .config_read = ctl_be_block_config_read, 292 .config_write = ctl_be_block_config_write, 293 .ioctl = ctl_be_block_ioctl, 294 .lun_info = ctl_be_block_lun_info, 295 .lun_attr = ctl_be_block_lun_attr 296}; 297 298MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 299CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 300 301static uma_zone_t beio_zone; 302 303static struct ctl_be_block_io * 304ctl_alloc_beio(struct ctl_be_block_softc *softc) 305{ 306 struct ctl_be_block_io *beio; 307 308 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 309 beio->softc = softc; 310 return (beio); 311} 312 313static void 314ctl_free_beio(struct ctl_be_block_io *beio) 315{ 316 int duplicate_free; 317 int i; 318 319 duplicate_free = 0; 320 321 for (i = 0; i < beio->num_segs; i++) { 322 if (beio->sg_segs[i].addr == NULL) 323 duplicate_free++; 324 325 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 326 beio->sg_segs[i].addr = NULL; 327 328 /* For compare we had two equal S/G lists. */ 329 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 330 uma_zfree(beio->lun->lun_zone, 331 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 332 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 333 } 334 } 335 336 if (duplicate_free > 0) { 337 printf("%s: %d duplicate frees out of %d segments\n", __func__, 338 duplicate_free, beio->num_segs); 339 } 340 341 uma_zfree(beio_zone, beio); 342} 343 344static void 345ctl_complete_beio(struct ctl_be_block_io *beio) 346{ 347 union ctl_io *io = beio->io; 348 349 if (beio->beio_cont != NULL) { 350 beio->beio_cont(beio); 351 } else { 352 ctl_free_beio(beio); 353 ctl_data_submit_done(io); 354 } 355} 356 357static size_t 358cmp(uint8_t *a, uint8_t *b, size_t size) 359{ 360 size_t i; 361 362 for (i = 0; i < size; i++) { 363 if (a[i] != b[i]) 364 break; 365 } 366 return (i); 367} 368 369static void 370ctl_be_block_compare(union ctl_io *io) 371{ 372 struct ctl_be_block_io *beio; 373 uint64_t off, res; 374 int i; 375 uint8_t info[8]; 376 377 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 378 off = 0; 379 for (i = 0; i < beio->num_segs; i++) { 380 res = cmp(beio->sg_segs[i].addr, 381 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 382 beio->sg_segs[i].len); 383 off += res; 384 if (res < beio->sg_segs[i].len) 385 break; 386 } 387 if (i < beio->num_segs) { 388 scsi_u64to8b(off, info); 389 ctl_set_sense(&io->scsiio, /*current_error*/ 1, 390 /*sense_key*/ SSD_KEY_MISCOMPARE, 391 /*asc*/ 0x1D, /*ascq*/ 0x00, 392 /*type*/ SSD_ELEM_INFO, 393 /*size*/ sizeof(info), /*data*/ &info, 394 /*type*/ SSD_ELEM_NONE); 395 } else 396 ctl_set_success(&io->scsiio); 397} 398 399static int 400ctl_be_block_move_done(union ctl_io *io) 401{ 402 struct ctl_be_block_io *beio; 403 struct ctl_be_block_lun *be_lun; 404 struct ctl_lba_len_flags *lbalen; 405#ifdef CTL_TIME_IO 406 struct bintime cur_bt; 407#endif 408 409 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 410 be_lun = beio->lun; 411 412 DPRINTF("entered\n"); 413 414#ifdef CTL_TIME_IO 415 getbintime(&cur_bt); 416 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 417 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 418 io->io_hdr.num_dmas++; 419#endif 420 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 421 422 /* 423 * We set status at this point for read commands, and write 424 * commands with errors. 425 */ 426 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 427 ; 428 } else if ((io->io_hdr.port_status == 0) && 429 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 430 lbalen = ARGS(beio->io); 431 if (lbalen->flags & CTL_LLF_READ) { 432 ctl_set_success(&io->scsiio); 433 } else if (lbalen->flags & CTL_LLF_COMPARE) { 434 /* We have two data blocks ready for comparison. */ 435 ctl_be_block_compare(io); 436 } 437 } else if ((io->io_hdr.port_status != 0) && 438 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 439 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 440 /* 441 * For hardware error sense keys, the sense key 442 * specific value is defined to be a retry count, 443 * but we use it to pass back an internal FETD 444 * error code. XXX KDM Hopefully the FETD is only 445 * using 16 bits for an error code, since that's 446 * all the space we have in the sks field. 447 */ 448 ctl_set_internal_failure(&io->scsiio, 449 /*sks_valid*/ 1, 450 /*retry_count*/ 451 io->io_hdr.port_status); 452 } 453 454 /* 455 * If this is a read, or a write with errors, it is done. 456 */ 457 if ((beio->bio_cmd == BIO_READ) 458 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 459 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 460 ctl_complete_beio(beio); 461 return (0); 462 } 463 464 /* 465 * At this point, we have a write and the DMA completed 466 * successfully. We now have to queue it to the task queue to 467 * execute the backend I/O. That is because we do blocking 468 * memory allocations, and in the file backing case, blocking I/O. 469 * This move done routine is generally called in the SIM's 470 * interrupt context, and therefore we cannot block. 471 */ 472 mtx_lock(&be_lun->queue_lock); 473 /* 474 * XXX KDM make sure that links is okay to use at this point. 475 * Otherwise, we either need to add another field to ctl_io_hdr, 476 * or deal with resource allocation here. 477 */ 478 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 479 mtx_unlock(&be_lun->queue_lock); 480 481 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 482 483 return (0); 484} 485 486static void 487ctl_be_block_biodone(struct bio *bio) 488{ 489 struct ctl_be_block_io *beio; 490 struct ctl_be_block_lun *be_lun; 491 union ctl_io *io; 492 int error; 493 494 beio = bio->bio_caller1; 495 be_lun = beio->lun; 496 io = beio->io; 497 498 DPRINTF("entered\n"); 499 500 error = bio->bio_error; 501 mtx_lock(&be_lun->io_lock); 502 if (error != 0) 503 beio->num_errors++; 504 505 beio->num_bios_done++; 506 507 /* 508 * XXX KDM will this cause WITNESS to complain? Holding a lock 509 * during the free might cause it to complain. 510 */ 511 g_destroy_bio(bio); 512 513 /* 514 * If the send complete bit isn't set, or we aren't the last I/O to 515 * complete, then we're done. 516 */ 517 if ((beio->send_complete == 0) 518 || (beio->num_bios_done < beio->num_bios_sent)) { 519 mtx_unlock(&be_lun->io_lock); 520 return; 521 } 522 523 /* 524 * At this point, we've verified that we are the last I/O to 525 * complete, so it's safe to drop the lock. 526 */ 527 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 528 beio->ds_tag_type, beio->ds_trans_type, 529 /*now*/ NULL, /*then*/&beio->ds_t0); 530 mtx_unlock(&be_lun->io_lock); 531 532 /* 533 * If there are any errors from the backing device, we fail the 534 * entire I/O with a medium error. 535 */ 536 if (beio->num_errors > 0) { 537 if (error == EOPNOTSUPP) { 538 ctl_set_invalid_opcode(&io->scsiio); 539 } else if (error == ENOSPC || error == EDQUOT) { 540 ctl_set_space_alloc_fail(&io->scsiio); 541 } else if (error == EROFS || error == EACCES) { 542 ctl_set_hw_write_protected(&io->scsiio); 543 } else if (beio->bio_cmd == BIO_FLUSH) { 544 /* XXX KDM is there is a better error here? */ 545 ctl_set_internal_failure(&io->scsiio, 546 /*sks_valid*/ 1, 547 /*retry_count*/ 0xbad2); 548 } else { 549 ctl_set_medium_error(&io->scsiio, 550 beio->bio_cmd == BIO_READ); 551 } 552 ctl_complete_beio(beio); 553 return; 554 } 555 556 /* 557 * If this is a write, a flush, a delete or verify, we're all done. 558 * If this is a read, we can now send the data to the user. 559 */ 560 if ((beio->bio_cmd == BIO_WRITE) 561 || (beio->bio_cmd == BIO_FLUSH) 562 || (beio->bio_cmd == BIO_DELETE) 563 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 564 ctl_set_success(&io->scsiio); 565 ctl_complete_beio(beio); 566 } else { 567 if ((ARGS(io)->flags & CTL_LLF_READ) && 568 beio->beio_cont == NULL) { 569 ctl_set_success(&io->scsiio); 570 ctl_serseq_done(io); 571 } 572#ifdef CTL_TIME_IO 573 getbintime(&io->io_hdr.dma_start_bt); 574#endif 575 ctl_datamove(io); 576 } 577} 578 579static void 580ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 581 struct ctl_be_block_io *beio) 582{ 583 union ctl_io *io = beio->io; 584 struct mount *mountpoint; 585 int error, lock_flags; 586 587 DPRINTF("entered\n"); 588 589 binuptime(&beio->ds_t0); 590 mtx_lock(&be_lun->io_lock); 591 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 592 mtx_unlock(&be_lun->io_lock); 593 594 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 595 596 if (MNT_SHARED_WRITES(mountpoint) 597 || ((mountpoint == NULL) 598 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 599 lock_flags = LK_SHARED; 600 else 601 lock_flags = LK_EXCLUSIVE; 602 603 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 604 605 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 606 curthread); 607 VOP_UNLOCK(be_lun->vn, 0); 608 609 vn_finished_write(mountpoint); 610 611 mtx_lock(&be_lun->io_lock); 612 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 613 beio->ds_tag_type, beio->ds_trans_type, 614 /*now*/ NULL, /*then*/&beio->ds_t0); 615 mtx_unlock(&be_lun->io_lock); 616 617 if (error == 0) 618 ctl_set_success(&io->scsiio); 619 else { 620 /* XXX KDM is there is a better error here? */ 621 ctl_set_internal_failure(&io->scsiio, 622 /*sks_valid*/ 1, 623 /*retry_count*/ 0xbad1); 624 } 625 626 ctl_complete_beio(beio); 627} 628 629SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 630SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 631SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 632SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 633 634static void 635ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 636 struct ctl_be_block_io *beio) 637{ 638 struct ctl_be_block_filedata *file_data; 639 union ctl_io *io; 640 struct uio xuio; 641 struct iovec *xiovec; 642 size_t s; 643 int error, flags, i; 644 645 DPRINTF("entered\n"); 646 647 file_data = &be_lun->backend.file; 648 io = beio->io; 649 flags = 0; 650 if (ARGS(io)->flags & CTL_LLF_DPO) 651 flags |= IO_DIRECT; 652 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 653 flags |= IO_SYNC; 654 655 bzero(&xuio, sizeof(xuio)); 656 if (beio->bio_cmd == BIO_READ) { 657 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 658 xuio.uio_rw = UIO_READ; 659 } else { 660 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 661 xuio.uio_rw = UIO_WRITE; 662 } 663 xuio.uio_offset = beio->io_offset; 664 xuio.uio_resid = beio->io_len; 665 xuio.uio_segflg = UIO_SYSSPACE; 666 xuio.uio_iov = beio->xiovecs; 667 xuio.uio_iovcnt = beio->num_segs; 668 xuio.uio_td = curthread; 669 670 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 671 xiovec->iov_base = beio->sg_segs[i].addr; 672 xiovec->iov_len = beio->sg_segs[i].len; 673 } 674 675 binuptime(&beio->ds_t0); 676 mtx_lock(&be_lun->io_lock); 677 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 678 mtx_unlock(&be_lun->io_lock); 679 680 if (beio->bio_cmd == BIO_READ) { 681 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 682 683 /* 684 * UFS pays attention to IO_DIRECT for reads. If the 685 * DIRECTIO option is configured into the kernel, it calls 686 * ffs_rawread(). But that only works for single-segment 687 * uios with user space addresses. In our case, with a 688 * kernel uio, it still reads into the buffer cache, but it 689 * will just try to release the buffer from the cache later 690 * on in ffs_read(). 691 * 692 * ZFS does not pay attention to IO_DIRECT for reads. 693 * 694 * UFS does not pay attention to IO_SYNC for reads. 695 * 696 * ZFS pays attention to IO_SYNC (which translates into the 697 * Solaris define FRSYNC for zfs_read()) for reads. It 698 * attempts to sync the file before reading. 699 */ 700 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 701 702 VOP_UNLOCK(be_lun->vn, 0); 703 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 704 if (error == 0 && xuio.uio_resid > 0) { 705 /* 706 * If we red less then requested (EOF), then 707 * we should clean the rest of the buffer. 708 */ 709 s = beio->io_len - xuio.uio_resid; 710 for (i = 0; i < beio->num_segs; i++) { 711 if (s >= beio->sg_segs[i].len) { 712 s -= beio->sg_segs[i].len; 713 continue; 714 } 715 bzero((uint8_t *)beio->sg_segs[i].addr + s, 716 beio->sg_segs[i].len - s); 717 s = 0; 718 } 719 } 720 } else { 721 struct mount *mountpoint; 722 int lock_flags; 723 724 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 725 726 if (MNT_SHARED_WRITES(mountpoint) 727 || ((mountpoint == NULL) 728 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 729 lock_flags = LK_SHARED; 730 else 731 lock_flags = LK_EXCLUSIVE; 732 733 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 734 735 /* 736 * UFS pays attention to IO_DIRECT for writes. The write 737 * is done asynchronously. (Normally the write would just 738 * get put into cache. 739 * 740 * UFS pays attention to IO_SYNC for writes. It will 741 * attempt to write the buffer out synchronously if that 742 * flag is set. 743 * 744 * ZFS does not pay attention to IO_DIRECT for writes. 745 * 746 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 747 * for writes. It will flush the transaction from the 748 * cache before returning. 749 */ 750 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 751 VOP_UNLOCK(be_lun->vn, 0); 752 753 vn_finished_write(mountpoint); 754 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 755 } 756 757 mtx_lock(&be_lun->io_lock); 758 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 759 beio->ds_tag_type, beio->ds_trans_type, 760 /*now*/ NULL, /*then*/&beio->ds_t0); 761 mtx_unlock(&be_lun->io_lock); 762 763 /* 764 * If we got an error, set the sense data to "MEDIUM ERROR" and 765 * return the I/O to the user. 766 */ 767 if (error != 0) { 768 if (error == ENOSPC || error == EDQUOT) { 769 ctl_set_space_alloc_fail(&io->scsiio); 770 } else if (error == EROFS || error == EACCES) { 771 ctl_set_hw_write_protected(&io->scsiio); 772 } else { 773 ctl_set_medium_error(&io->scsiio, 774 beio->bio_cmd == BIO_READ); 775 } 776 ctl_complete_beio(beio); 777 return; 778 } 779 780 /* 781 * If this is a write or a verify, we're all done. 782 * If this is a read, we can now send the data to the user. 783 */ 784 if ((beio->bio_cmd == BIO_WRITE) || 785 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 786 ctl_set_success(&io->scsiio); 787 ctl_complete_beio(beio); 788 } else { 789 if ((ARGS(io)->flags & CTL_LLF_READ) && 790 beio->beio_cont == NULL) { 791 ctl_set_success(&io->scsiio); 792 ctl_serseq_done(io); 793 } 794#ifdef CTL_TIME_IO 795 getbintime(&io->io_hdr.dma_start_bt); 796#endif 797 ctl_datamove(io); 798 } 799} 800 801static void 802ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 803 struct ctl_be_block_io *beio) 804{ 805 union ctl_io *io = beio->io; 806 struct ctl_lba_len_flags *lbalen = ARGS(io); 807 struct scsi_get_lba_status_data *data; 808 off_t roff, off; 809 int error, status; 810 811 DPRINTF("entered\n"); 812 813 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 814 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 815 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 816 0, curthread->td_ucred, curthread); 817 if (error == 0 && off > roff) 818 status = 0; /* mapped up to off */ 819 else { 820 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 821 0, curthread->td_ucred, curthread); 822 if (error == 0 && off > roff) 823 status = 1; /* deallocated up to off */ 824 else { 825 status = 0; /* unknown up to the end */ 826 off = be_lun->size_bytes; 827 } 828 } 829 VOP_UNLOCK(be_lun->vn, 0); 830 831 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 832 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 833 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 834 lbalen->lba), data->descr[0].length); 835 data->descr[0].status = status; 836 837 ctl_complete_beio(beio); 838} 839 840static uint64_t 841ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 842{ 843 struct vattr vattr; 844 struct statfs statfs; 845 uint64_t val; 846 int error; 847 848 val = UINT64_MAX; 849 if (be_lun->vn == NULL) 850 return (val); 851 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 852 if (strcmp(attrname, "blocksused") == 0) { 853 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 854 if (error == 0) 855 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 856 } 857 if (strcmp(attrname, "blocksavail") == 0 && 858 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 859 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 860 if (error == 0) 861 val = statfs.f_bavail * statfs.f_bsize / 862 be_lun->cbe_lun.blocksize; 863 } 864 VOP_UNLOCK(be_lun->vn, 0); 865 return (val); 866} 867 868static void 869ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 870 struct ctl_be_block_io *beio) 871{ 872 union ctl_io *io; 873 struct cdevsw *csw; 874 struct cdev *dev; 875 struct uio xuio; 876 struct iovec *xiovec; 877 int error, flags, i, ref; 878 879 DPRINTF("entered\n"); 880 881 io = beio->io; 882 flags = 0; 883 if (ARGS(io)->flags & CTL_LLF_DPO) 884 flags |= IO_DIRECT; 885 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 886 flags |= IO_SYNC; 887 888 bzero(&xuio, sizeof(xuio)); 889 if (beio->bio_cmd == BIO_READ) { 890 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 891 xuio.uio_rw = UIO_READ; 892 } else { 893 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 894 xuio.uio_rw = UIO_WRITE; 895 } 896 xuio.uio_offset = beio->io_offset; 897 xuio.uio_resid = beio->io_len; 898 xuio.uio_segflg = UIO_SYSSPACE; 899 xuio.uio_iov = beio->xiovecs; 900 xuio.uio_iovcnt = beio->num_segs; 901 xuio.uio_td = curthread; 902 903 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 904 xiovec->iov_base = beio->sg_segs[i].addr; 905 xiovec->iov_len = beio->sg_segs[i].len; 906 } 907 908 binuptime(&beio->ds_t0); 909 mtx_lock(&be_lun->io_lock); 910 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 911 mtx_unlock(&be_lun->io_lock); 912 913 csw = devvn_refthread(be_lun->vn, &dev, &ref); 914 if (csw) { 915 if (beio->bio_cmd == BIO_READ) 916 error = csw->d_read(dev, &xuio, flags); 917 else 918 error = csw->d_write(dev, &xuio, flags); 919 dev_relthread(dev, ref); 920 } else 921 error = ENXIO; 922 923 if (beio->bio_cmd == BIO_READ) 924 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 925 else 926 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 927 928 mtx_lock(&be_lun->io_lock); 929 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 930 beio->ds_tag_type, beio->ds_trans_type, 931 /*now*/ NULL, /*then*/&beio->ds_t0); 932 mtx_unlock(&be_lun->io_lock); 933 934 /* 935 * If we got an error, set the sense data to "MEDIUM ERROR" and 936 * return the I/O to the user. 937 */ 938 if (error != 0) { 939 if (error == ENOSPC || error == EDQUOT) { 940 ctl_set_space_alloc_fail(&io->scsiio); 941 } else if (error == EROFS || error == EACCES) { 942 ctl_set_hw_write_protected(&io->scsiio); 943 } else { 944 ctl_set_medium_error(&io->scsiio, 945 beio->bio_cmd == BIO_READ); 946 } 947 ctl_complete_beio(beio); 948 return; 949 } 950 951 /* 952 * If this is a write or a verify, we're all done. 953 * If this is a read, we can now send the data to the user. 954 */ 955 if ((beio->bio_cmd == BIO_WRITE) || 956 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 957 ctl_set_success(&io->scsiio); 958 ctl_complete_beio(beio); 959 } else { 960 if ((ARGS(io)->flags & CTL_LLF_READ) && 961 beio->beio_cont == NULL) { 962 ctl_set_success(&io->scsiio); 963 ctl_serseq_done(io); 964 } 965#ifdef CTL_TIME_IO 966 getbintime(&io->io_hdr.dma_start_bt); 967#endif 968 ctl_datamove(io); 969 } 970} 971 972static void 973ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 974 struct ctl_be_block_io *beio) 975{ 976 union ctl_io *io = beio->io; 977 struct cdevsw *csw; 978 struct cdev *dev; 979 struct ctl_lba_len_flags *lbalen = ARGS(io); 980 struct scsi_get_lba_status_data *data; 981 off_t roff, off; 982 int error, ref, status; 983 984 DPRINTF("entered\n"); 985 986 csw = devvn_refthread(be_lun->vn, &dev, &ref); 987 if (csw == NULL) { 988 status = 0; /* unknown up to the end */ 989 off = be_lun->size_bytes; 990 goto done; 991 } 992 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 993 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, 994 curthread); 995 if (error == 0 && off > roff) 996 status = 0; /* mapped up to off */ 997 else { 998 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, 999 curthread); 1000 if (error == 0 && off > roff) 1001 status = 1; /* deallocated up to off */ 1002 else { 1003 status = 0; /* unknown up to the end */ 1004 off = be_lun->size_bytes; 1005 } 1006 } 1007 dev_relthread(dev, ref); 1008 1009done: 1010 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 1011 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 1012 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 1013 lbalen->lba), data->descr[0].length); 1014 data->descr[0].status = status; 1015 1016 ctl_complete_beio(beio); 1017} 1018 1019static void 1020ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 1021 struct ctl_be_block_io *beio) 1022{ 1023 struct bio *bio; 1024 union ctl_io *io; 1025 struct cdevsw *csw; 1026 struct cdev *dev; 1027 int ref; 1028 1029 io = beio->io; 1030 1031 DPRINTF("entered\n"); 1032 1033 /* This can't fail, it's a blocking allocation. */ 1034 bio = g_alloc_bio(); 1035 1036 bio->bio_cmd = BIO_FLUSH; 1037 bio->bio_offset = 0; 1038 bio->bio_data = 0; 1039 bio->bio_done = ctl_be_block_biodone; 1040 bio->bio_caller1 = beio; 1041 bio->bio_pblkno = 0; 1042 1043 /* 1044 * We don't need to acquire the LUN lock here, because we are only 1045 * sending one bio, and so there is no other context to synchronize 1046 * with. 1047 */ 1048 beio->num_bios_sent = 1; 1049 beio->send_complete = 1; 1050 1051 binuptime(&beio->ds_t0); 1052 mtx_lock(&be_lun->io_lock); 1053 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1054 mtx_unlock(&be_lun->io_lock); 1055 1056 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1057 if (csw) { 1058 bio->bio_dev = dev; 1059 csw->d_strategy(bio); 1060 dev_relthread(dev, ref); 1061 } else { 1062 bio->bio_error = ENXIO; 1063 ctl_be_block_biodone(bio); 1064 } 1065} 1066 1067static void 1068ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1069 struct ctl_be_block_io *beio, 1070 uint64_t off, uint64_t len, int last) 1071{ 1072 struct bio *bio; 1073 uint64_t maxlen; 1074 struct cdevsw *csw; 1075 struct cdev *dev; 1076 int ref; 1077 1078 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1079 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1080 while (len > 0) { 1081 bio = g_alloc_bio(); 1082 bio->bio_cmd = BIO_DELETE; 1083 bio->bio_dev = dev; 1084 bio->bio_offset = off; 1085 bio->bio_length = MIN(len, maxlen); 1086 bio->bio_data = 0; 1087 bio->bio_done = ctl_be_block_biodone; 1088 bio->bio_caller1 = beio; 1089 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1090 1091 off += bio->bio_length; 1092 len -= bio->bio_length; 1093 1094 mtx_lock(&be_lun->io_lock); 1095 beio->num_bios_sent++; 1096 if (last && len == 0) 1097 beio->send_complete = 1; 1098 mtx_unlock(&be_lun->io_lock); 1099 1100 if (csw) { 1101 csw->d_strategy(bio); 1102 } else { 1103 bio->bio_error = ENXIO; 1104 ctl_be_block_biodone(bio); 1105 } 1106 } 1107 if (csw) 1108 dev_relthread(dev, ref); 1109} 1110 1111static void 1112ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1113 struct ctl_be_block_io *beio) 1114{ 1115 union ctl_io *io; 1116 struct ctl_ptr_len_flags *ptrlen; 1117 struct scsi_unmap_desc *buf, *end; 1118 uint64_t len; 1119 1120 io = beio->io; 1121 1122 DPRINTF("entered\n"); 1123 1124 binuptime(&beio->ds_t0); 1125 mtx_lock(&be_lun->io_lock); 1126 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1127 mtx_unlock(&be_lun->io_lock); 1128 1129 if (beio->io_offset == -1) { 1130 beio->io_len = 0; 1131 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1132 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1133 end = buf + ptrlen->len / sizeof(*buf); 1134 for (; buf < end; buf++) { 1135 len = (uint64_t)scsi_4btoul(buf->length) * 1136 be_lun->cbe_lun.blocksize; 1137 beio->io_len += len; 1138 ctl_be_block_unmap_dev_range(be_lun, beio, 1139 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1140 len, (end - buf < 2) ? TRUE : FALSE); 1141 } 1142 } else 1143 ctl_be_block_unmap_dev_range(be_lun, beio, 1144 beio->io_offset, beio->io_len, TRUE); 1145} 1146 1147static void 1148ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1149 struct ctl_be_block_io *beio) 1150{ 1151 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1152 struct bio *bio; 1153 struct cdevsw *csw; 1154 struct cdev *dev; 1155 off_t cur_offset; 1156 int i, max_iosize, ref; 1157 1158 DPRINTF("entered\n"); 1159 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1160 1161 /* 1162 * We have to limit our I/O size to the maximum supported by the 1163 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1164 * set it properly, use DFLTPHYS. 1165 */ 1166 if (csw) { 1167 max_iosize = dev->si_iosize_max; 1168 if (max_iosize < PAGE_SIZE) 1169 max_iosize = DFLTPHYS; 1170 } else 1171 max_iosize = DFLTPHYS; 1172 1173 cur_offset = beio->io_offset; 1174 for (i = 0; i < beio->num_segs; i++) { 1175 size_t cur_size; 1176 uint8_t *cur_ptr; 1177 1178 cur_size = beio->sg_segs[i].len; 1179 cur_ptr = beio->sg_segs[i].addr; 1180 1181 while (cur_size > 0) { 1182 /* This can't fail, it's a blocking allocation. */ 1183 bio = g_alloc_bio(); 1184 1185 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1186 1187 bio->bio_cmd = beio->bio_cmd; 1188 bio->bio_dev = dev; 1189 bio->bio_caller1 = beio; 1190 bio->bio_length = min(cur_size, max_iosize); 1191 bio->bio_offset = cur_offset; 1192 bio->bio_data = cur_ptr; 1193 bio->bio_done = ctl_be_block_biodone; 1194 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1195 1196 cur_offset += bio->bio_length; 1197 cur_ptr += bio->bio_length; 1198 cur_size -= bio->bio_length; 1199 1200 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1201 beio->num_bios_sent++; 1202 } 1203 } 1204 binuptime(&beio->ds_t0); 1205 mtx_lock(&be_lun->io_lock); 1206 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1207 beio->send_complete = 1; 1208 mtx_unlock(&be_lun->io_lock); 1209 1210 /* 1211 * Fire off all allocated requests! 1212 */ 1213 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1214 TAILQ_REMOVE(&queue, bio, bio_queue); 1215 if (csw) 1216 csw->d_strategy(bio); 1217 else { 1218 bio->bio_error = ENXIO; 1219 ctl_be_block_biodone(bio); 1220 } 1221 } 1222 if (csw) 1223 dev_relthread(dev, ref); 1224} 1225 1226static uint64_t 1227ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1228{ 1229 struct diocgattr_arg arg; 1230 struct cdevsw *csw; 1231 struct cdev *dev; 1232 int error, ref; 1233 1234 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1235 if (csw == NULL) 1236 return (UINT64_MAX); 1237 strlcpy(arg.name, attrname, sizeof(arg.name)); 1238 arg.len = sizeof(arg.value.off); 1239 if (csw->d_ioctl) { 1240 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1241 curthread); 1242 } else 1243 error = ENODEV; 1244 dev_relthread(dev, ref); 1245 if (error != 0) 1246 return (UINT64_MAX); 1247 return (arg.value.off); 1248} 1249 1250static void 1251ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1252 union ctl_io *io) 1253{ 1254 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1255 struct ctl_be_block_io *beio; 1256 struct ctl_lba_len_flags *lbalen; 1257 1258 DPRINTF("entered\n"); 1259 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1260 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1261 1262 beio->io_len = lbalen->len * cbe_lun->blocksize; 1263 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1264 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1265 beio->bio_cmd = BIO_FLUSH; 1266 beio->ds_trans_type = DEVSTAT_NO_DATA; 1267 DPRINTF("SYNC\n"); 1268 be_lun->lun_flush(be_lun, beio); 1269} 1270 1271static void 1272ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1273{ 1274 union ctl_io *io; 1275 1276 io = beio->io; 1277 ctl_free_beio(beio); 1278 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1279 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1280 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1281 ctl_config_write_done(io); 1282 return; 1283 } 1284 1285 ctl_be_block_config_write(io); 1286} 1287 1288static void 1289ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1290 union ctl_io *io) 1291{ 1292 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1293 struct ctl_be_block_io *beio; 1294 struct ctl_lba_len_flags *lbalen; 1295 uint64_t len_left, lba; 1296 uint32_t pb, pbo, adj; 1297 int i, seglen; 1298 uint8_t *buf, *end; 1299 1300 DPRINTF("entered\n"); 1301 1302 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1303 lbalen = ARGS(beio->io); 1304 1305 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1306 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1307 ctl_free_beio(beio); 1308 ctl_set_invalid_field(&io->scsiio, 1309 /*sks_valid*/ 1, 1310 /*command*/ 1, 1311 /*field*/ 1, 1312 /*bit_valid*/ 0, 1313 /*bit*/ 0); 1314 ctl_config_write_done(io); 1315 return; 1316 } 1317 1318 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1319 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1320 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1321 beio->bio_cmd = BIO_DELETE; 1322 beio->ds_trans_type = DEVSTAT_FREE; 1323 1324 be_lun->unmap(be_lun, beio); 1325 return; 1326 } 1327 1328 beio->bio_cmd = BIO_WRITE; 1329 beio->ds_trans_type = DEVSTAT_WRITE; 1330 1331 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1332 (uintmax_t)lbalen->lba, lbalen->len); 1333 1334 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1335 if (be_lun->cbe_lun.pblockoff > 0) 1336 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1337 else 1338 pbo = 0; 1339 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1340 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1341 1342 /* 1343 * Setup the S/G entry for this chunk. 1344 */ 1345 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1346 if (pb > cbe_lun->blocksize) { 1347 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1348 seglen - pbo) % pb; 1349 if (seglen > adj) 1350 seglen -= adj; 1351 else 1352 seglen -= seglen % cbe_lun->blocksize; 1353 } else 1354 seglen -= seglen % cbe_lun->blocksize; 1355 beio->sg_segs[i].len = seglen; 1356 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1357 1358 DPRINTF("segment %d addr %p len %zd\n", i, 1359 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1360 1361 beio->num_segs++; 1362 len_left -= seglen; 1363 1364 buf = beio->sg_segs[i].addr; 1365 end = buf + seglen; 1366 for (; buf < end; buf += cbe_lun->blocksize) { 1367 memcpy(buf, io->scsiio.kern_data_ptr, cbe_lun->blocksize); 1368 if (lbalen->flags & SWS_LBDATA) 1369 scsi_ulto4b(lbalen->lba + lba, buf); 1370 lba++; 1371 } 1372 } 1373 1374 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1375 beio->io_len = lba * cbe_lun->blocksize; 1376 1377 /* We can not do all in one run. Correct and schedule rerun. */ 1378 if (len_left > 0) { 1379 lbalen->lba += lba; 1380 lbalen->len -= lba; 1381 beio->beio_cont = ctl_be_block_cw_done_ws; 1382 } 1383 1384 be_lun->dispatch(be_lun, beio); 1385} 1386 1387static void 1388ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1389 union ctl_io *io) 1390{ 1391 struct ctl_be_block_io *beio; 1392 struct ctl_ptr_len_flags *ptrlen; 1393 1394 DPRINTF("entered\n"); 1395 1396 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1397 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1398 1399 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1400 ctl_free_beio(beio); 1401 ctl_set_invalid_field(&io->scsiio, 1402 /*sks_valid*/ 0, 1403 /*command*/ 1, 1404 /*field*/ 0, 1405 /*bit_valid*/ 0, 1406 /*bit*/ 0); 1407 ctl_config_write_done(io); 1408 return; 1409 } 1410 1411 beio->io_len = 0; 1412 beio->io_offset = -1; 1413 beio->bio_cmd = BIO_DELETE; 1414 beio->ds_trans_type = DEVSTAT_FREE; 1415 DPRINTF("UNMAP\n"); 1416 be_lun->unmap(be_lun, beio); 1417} 1418 1419static void 1420ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1421{ 1422 union ctl_io *io; 1423 1424 io = beio->io; 1425 ctl_free_beio(beio); 1426 ctl_config_read_done(io); 1427} 1428 1429static void 1430ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1431 union ctl_io *io) 1432{ 1433 struct ctl_be_block_io *beio; 1434 struct ctl_be_block_softc *softc; 1435 1436 DPRINTF("entered\n"); 1437 1438 softc = be_lun->softc; 1439 beio = ctl_alloc_beio(softc); 1440 beio->io = io; 1441 beio->lun = be_lun; 1442 beio->beio_cont = ctl_be_block_cr_done; 1443 PRIV(io)->ptr = (void *)beio; 1444 1445 switch (io->scsiio.cdb[0]) { 1446 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1447 beio->bio_cmd = -1; 1448 beio->ds_trans_type = DEVSTAT_NO_DATA; 1449 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1450 beio->io_len = 0; 1451 if (be_lun->get_lba_status) 1452 be_lun->get_lba_status(be_lun, beio); 1453 else 1454 ctl_be_block_cr_done(beio); 1455 break; 1456 default: 1457 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1458 break; 1459 } 1460} 1461 1462static void 1463ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1464{ 1465 union ctl_io *io; 1466 1467 io = beio->io; 1468 ctl_free_beio(beio); 1469 ctl_config_write_done(io); 1470} 1471 1472static void 1473ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1474 union ctl_io *io) 1475{ 1476 struct ctl_be_block_io *beio; 1477 struct ctl_be_block_softc *softc; 1478 1479 DPRINTF("entered\n"); 1480 1481 softc = be_lun->softc; 1482 beio = ctl_alloc_beio(softc); 1483 beio->io = io; 1484 beio->lun = be_lun; 1485 beio->beio_cont = ctl_be_block_cw_done; 1486 switch (io->scsiio.tag_type) { 1487 case CTL_TAG_ORDERED: 1488 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1489 break; 1490 case CTL_TAG_HEAD_OF_QUEUE: 1491 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1492 break; 1493 case CTL_TAG_UNTAGGED: 1494 case CTL_TAG_SIMPLE: 1495 case CTL_TAG_ACA: 1496 default: 1497 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1498 break; 1499 } 1500 PRIV(io)->ptr = (void *)beio; 1501 1502 switch (io->scsiio.cdb[0]) { 1503 case SYNCHRONIZE_CACHE: 1504 case SYNCHRONIZE_CACHE_16: 1505 ctl_be_block_cw_dispatch_sync(be_lun, io); 1506 break; 1507 case WRITE_SAME_10: 1508 case WRITE_SAME_16: 1509 ctl_be_block_cw_dispatch_ws(be_lun, io); 1510 break; 1511 case UNMAP: 1512 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1513 break; 1514 default: 1515 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1516 break; 1517 } 1518} 1519 1520SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1521SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1522SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1523SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1524 1525static void 1526ctl_be_block_next(struct ctl_be_block_io *beio) 1527{ 1528 struct ctl_be_block_lun *be_lun; 1529 union ctl_io *io; 1530 1531 io = beio->io; 1532 be_lun = beio->lun; 1533 ctl_free_beio(beio); 1534 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1535 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1536 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1537 ctl_data_submit_done(io); 1538 return; 1539 } 1540 1541 io->io_hdr.status &= ~CTL_STATUS_MASK; 1542 io->io_hdr.status |= CTL_STATUS_NONE; 1543 1544 mtx_lock(&be_lun->queue_lock); 1545 /* 1546 * XXX KDM make sure that links is okay to use at this point. 1547 * Otherwise, we either need to add another field to ctl_io_hdr, 1548 * or deal with resource allocation here. 1549 */ 1550 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1551 mtx_unlock(&be_lun->queue_lock); 1552 1553 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1554} 1555 1556static void 1557ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1558 union ctl_io *io) 1559{ 1560 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1561 struct ctl_be_block_io *beio; 1562 struct ctl_be_block_softc *softc; 1563 struct ctl_lba_len_flags *lbalen; 1564 struct ctl_ptr_len_flags *bptrlen; 1565 uint64_t len_left, lbas; 1566 int i; 1567 1568 softc = be_lun->softc; 1569 1570 DPRINTF("entered\n"); 1571 1572 lbalen = ARGS(io); 1573 if (lbalen->flags & CTL_LLF_WRITE) { 1574 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1575 } else { 1576 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1577 } 1578 1579 beio = ctl_alloc_beio(softc); 1580 beio->io = io; 1581 beio->lun = be_lun; 1582 bptrlen = PRIV(io); 1583 bptrlen->ptr = (void *)beio; 1584 1585 switch (io->scsiio.tag_type) { 1586 case CTL_TAG_ORDERED: 1587 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1588 break; 1589 case CTL_TAG_HEAD_OF_QUEUE: 1590 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1591 break; 1592 case CTL_TAG_UNTAGGED: 1593 case CTL_TAG_SIMPLE: 1594 case CTL_TAG_ACA: 1595 default: 1596 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1597 break; 1598 } 1599 1600 if (lbalen->flags & CTL_LLF_WRITE) { 1601 beio->bio_cmd = BIO_WRITE; 1602 beio->ds_trans_type = DEVSTAT_WRITE; 1603 } else { 1604 beio->bio_cmd = BIO_READ; 1605 beio->ds_trans_type = DEVSTAT_READ; 1606 } 1607 1608 DPRINTF("%s at LBA %jx len %u @%ju\n", 1609 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1610 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1611 if (lbalen->flags & CTL_LLF_COMPARE) 1612 lbas = CTLBLK_HALF_IO_SIZE; 1613 else 1614 lbas = CTLBLK_MAX_IO_SIZE; 1615 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1616 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1617 beio->io_len = lbas * cbe_lun->blocksize; 1618 bptrlen->len += lbas; 1619 1620 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1621 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1622 i, CTLBLK_MAX_SEGS)); 1623 1624 /* 1625 * Setup the S/G entry for this chunk. 1626 */ 1627 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1628 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1629 1630 DPRINTF("segment %d addr %p len %zd\n", i, 1631 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1632 1633 /* Set up second segment for compare operation. */ 1634 if (lbalen->flags & CTL_LLF_COMPARE) { 1635 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1636 beio->sg_segs[i].len; 1637 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1638 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1639 } 1640 1641 beio->num_segs++; 1642 len_left -= beio->sg_segs[i].len; 1643 } 1644 if (bptrlen->len < lbalen->len) 1645 beio->beio_cont = ctl_be_block_next; 1646 io->scsiio.be_move_done = ctl_be_block_move_done; 1647 /* For compare we have separate S/G lists for read and datamove. */ 1648 if (lbalen->flags & CTL_LLF_COMPARE) 1649 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1650 else 1651 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1652 io->scsiio.kern_data_len = beio->io_len; 1653 io->scsiio.kern_data_resid = 0; 1654 io->scsiio.kern_sg_entries = beio->num_segs; 1655 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1656 1657 /* 1658 * For the read case, we need to read the data into our buffers and 1659 * then we can send it back to the user. For the write case, we 1660 * need to get the data from the user first. 1661 */ 1662 if (beio->bio_cmd == BIO_READ) { 1663 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1664 be_lun->dispatch(be_lun, beio); 1665 } else { 1666 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1667#ifdef CTL_TIME_IO 1668 getbintime(&io->io_hdr.dma_start_bt); 1669#endif 1670 ctl_datamove(io); 1671 } 1672} 1673 1674static void 1675ctl_be_block_worker(void *context, int pending) 1676{ 1677 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1678 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1679 union ctl_io *io; 1680 struct ctl_be_block_io *beio; 1681 1682 DPRINTF("entered\n"); 1683 /* 1684 * Fetch and process I/Os from all queues. If we detect LUN 1685 * CTL_LUN_FLAG_OFFLINE status here -- it is result of a race, 1686 * so make response maximally opaque to not confuse initiator. 1687 */ 1688 for (;;) { 1689 mtx_lock(&be_lun->queue_lock); 1690 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1691 if (io != NULL) { 1692 DPRINTF("datamove queue\n"); 1693 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1694 ctl_io_hdr, links); 1695 mtx_unlock(&be_lun->queue_lock); 1696 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1697 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1698 ctl_set_busy(&io->scsiio); 1699 ctl_complete_beio(beio); 1700 return; 1701 } 1702 be_lun->dispatch(be_lun, beio); 1703 continue; 1704 } 1705 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1706 if (io != NULL) { 1707 DPRINTF("config write queue\n"); 1708 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1709 ctl_io_hdr, links); 1710 mtx_unlock(&be_lun->queue_lock); 1711 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1712 ctl_set_busy(&io->scsiio); 1713 ctl_config_write_done(io); 1714 return; 1715 } 1716 ctl_be_block_cw_dispatch(be_lun, io); 1717 continue; 1718 } 1719 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1720 if (io != NULL) { 1721 DPRINTF("config read queue\n"); 1722 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1723 ctl_io_hdr, links); 1724 mtx_unlock(&be_lun->queue_lock); 1725 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1726 ctl_set_busy(&io->scsiio); 1727 ctl_config_read_done(io); 1728 return; 1729 } 1730 ctl_be_block_cr_dispatch(be_lun, io); 1731 continue; 1732 } 1733 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1734 if (io != NULL) { 1735 DPRINTF("input queue\n"); 1736 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1737 ctl_io_hdr, links); 1738 mtx_unlock(&be_lun->queue_lock); 1739 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1740 ctl_set_busy(&io->scsiio); 1741 ctl_data_submit_done(io); 1742 return; 1743 } 1744 ctl_be_block_dispatch(be_lun, io); 1745 continue; 1746 } 1747 1748 /* 1749 * If we get here, there is no work left in the queues, so 1750 * just break out and let the task queue go to sleep. 1751 */ 1752 mtx_unlock(&be_lun->queue_lock); 1753 break; 1754 } 1755} 1756 1757/* 1758 * Entry point from CTL to the backend for I/O. We queue everything to a 1759 * work thread, so this just puts the I/O on a queue and wakes up the 1760 * thread. 1761 */ 1762static int 1763ctl_be_block_submit(union ctl_io *io) 1764{ 1765 struct ctl_be_block_lun *be_lun; 1766 struct ctl_be_lun *cbe_lun; 1767 1768 DPRINTF("entered\n"); 1769 1770 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1771 CTL_PRIV_BACKEND_LUN].ptr; 1772 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1773 1774 /* 1775 * Make sure we only get SCSI I/O. 1776 */ 1777 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1778 "%#x) encountered", io->io_hdr.io_type)); 1779 1780 PRIV(io)->len = 0; 1781 1782 mtx_lock(&be_lun->queue_lock); 1783 /* 1784 * XXX KDM make sure that links is okay to use at this point. 1785 * Otherwise, we either need to add another field to ctl_io_hdr, 1786 * or deal with resource allocation here. 1787 */ 1788 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1789 mtx_unlock(&be_lun->queue_lock); 1790 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1791 1792 return (CTL_RETVAL_COMPLETE); 1793} 1794 1795static int 1796ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1797 int flag, struct thread *td) 1798{ 1799 struct ctl_be_block_softc *softc; 1800 int error; 1801 1802 softc = &backend_block_softc; 1803 1804 error = 0; 1805 1806 switch (cmd) { 1807 case CTL_LUN_REQ: { 1808 struct ctl_lun_req *lun_req; 1809 1810 lun_req = (struct ctl_lun_req *)addr; 1811 1812 switch (lun_req->reqtype) { 1813 case CTL_LUNREQ_CREATE: 1814 error = ctl_be_block_create(softc, lun_req); 1815 break; 1816 case CTL_LUNREQ_RM: 1817 error = ctl_be_block_rm(softc, lun_req); 1818 break; 1819 case CTL_LUNREQ_MODIFY: 1820 error = ctl_be_block_modify(softc, lun_req); 1821 break; 1822 default: 1823 lun_req->status = CTL_LUN_ERROR; 1824 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1825 "invalid LUN request type %d", 1826 lun_req->reqtype); 1827 break; 1828 } 1829 break; 1830 } 1831 default: 1832 error = ENOTTY; 1833 break; 1834 } 1835 1836 return (error); 1837} 1838 1839static int 1840ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1841{ 1842 struct ctl_be_lun *cbe_lun; 1843 struct ctl_be_block_filedata *file_data; 1844 struct ctl_lun_create_params *params; 1845 char *value; 1846 struct vattr vattr; 1847 off_t ps, pss, po, pos, us, uss, uo, uos; 1848 int error; 1849 1850 error = 0; 1851 cbe_lun = &be_lun->cbe_lun; 1852 file_data = &be_lun->backend.file; 1853 params = &be_lun->params; 1854 1855 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1856 be_lun->dispatch = ctl_be_block_dispatch_file; 1857 be_lun->lun_flush = ctl_be_block_flush_file; 1858 be_lun->get_lba_status = ctl_be_block_gls_file; 1859 be_lun->getattr = ctl_be_block_getattr_file; 1860 be_lun->unmap = NULL; 1861 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1862 1863 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1864 if (error != 0) { 1865 snprintf(req->error_str, sizeof(req->error_str), 1866 "error calling VOP_GETATTR() for file %s", 1867 be_lun->dev_path); 1868 return (error); 1869 } 1870 1871 /* 1872 * Verify that we have the ability to upgrade to exclusive 1873 * access on this file so we can trap errors at open instead 1874 * of reporting them during first access. 1875 */ 1876 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1877 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1878 if (be_lun->vn->v_iflag & VI_DOOMED) { 1879 error = EBADF; 1880 snprintf(req->error_str, sizeof(req->error_str), 1881 "error locking file %s", be_lun->dev_path); 1882 return (error); 1883 } 1884 } 1885 1886 file_data->cred = crhold(curthread->td_ucred); 1887 if (params->lun_size_bytes != 0) 1888 be_lun->size_bytes = params->lun_size_bytes; 1889 else 1890 be_lun->size_bytes = vattr.va_size; 1891 1892 /* 1893 * For files we can use any logical block size. Prefer 512 bytes 1894 * for compatibility reasons. If file's vattr.va_blocksize 1895 * (preferred I/O block size) is bigger and multiple to chosen 1896 * logical block size -- report it as physical block size. 1897 */ 1898 if (params->blocksize_bytes != 0) 1899 cbe_lun->blocksize = params->blocksize_bytes; 1900 else 1901 cbe_lun->blocksize = 512; 1902 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1903 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1904 0 : (be_lun->size_blocks - 1); 1905 1906 us = ps = vattr.va_blocksize; 1907 uo = po = 0; 1908 1909 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1910 if (value != NULL) 1911 ctl_expand_number(value, &ps); 1912 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1913 if (value != NULL) 1914 ctl_expand_number(value, &po); 1915 pss = ps / cbe_lun->blocksize; 1916 pos = po / cbe_lun->blocksize; 1917 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1918 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1919 cbe_lun->pblockexp = fls(pss) - 1; 1920 cbe_lun->pblockoff = (pss - pos) % pss; 1921 } 1922 1923 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1924 if (value != NULL) 1925 ctl_expand_number(value, &us); 1926 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 1927 if (value != NULL) 1928 ctl_expand_number(value, &uo); 1929 uss = us / cbe_lun->blocksize; 1930 uos = uo / cbe_lun->blocksize; 1931 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1932 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1933 cbe_lun->ublockexp = fls(uss) - 1; 1934 cbe_lun->ublockoff = (uss - uos) % uss; 1935 } 1936 1937 /* 1938 * Sanity check. The media size has to be at least one 1939 * sector long. 1940 */ 1941 if (be_lun->size_bytes < cbe_lun->blocksize) { 1942 error = EINVAL; 1943 snprintf(req->error_str, sizeof(req->error_str), 1944 "file %s size %ju < block size %u", be_lun->dev_path, 1945 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1946 } 1947 1948 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1949 return (error); 1950} 1951 1952static int 1953ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1954{ 1955 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1956 struct ctl_lun_create_params *params; 1957 struct cdevsw *csw; 1958 struct cdev *dev; 1959 char *value; 1960 int error, atomic, maxio, ref, unmap, tmp; 1961 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1962 1963 params = &be_lun->params; 1964 1965 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1966 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1967 if (csw == NULL) 1968 return (ENXIO); 1969 if (strcmp(csw->d_name, "zvol") == 0) { 1970 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1971 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1972 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1973 } else { 1974 be_lun->dispatch = ctl_be_block_dispatch_dev; 1975 be_lun->get_lba_status = NULL; 1976 atomic = 0; 1977 maxio = dev->si_iosize_max; 1978 if (maxio <= 0) 1979 maxio = DFLTPHYS; 1980 if (maxio > CTLBLK_MAX_IO_SIZE) 1981 maxio = CTLBLK_MAX_IO_SIZE; 1982 } 1983 be_lun->lun_flush = ctl_be_block_flush_dev; 1984 be_lun->getattr = ctl_be_block_getattr_dev; 1985 be_lun->unmap = ctl_be_block_unmap_dev; 1986 1987 if (!csw->d_ioctl) { 1988 dev_relthread(dev, ref); 1989 snprintf(req->error_str, sizeof(req->error_str), 1990 "no d_ioctl for device %s!", be_lun->dev_path); 1991 return (ENODEV); 1992 } 1993 1994 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1995 curthread); 1996 if (error) { 1997 dev_relthread(dev, ref); 1998 snprintf(req->error_str, sizeof(req->error_str), 1999 "error %d returned for DIOCGSECTORSIZE ioctl " 2000 "on %s!", error, be_lun->dev_path); 2001 return (error); 2002 } 2003 2004 /* 2005 * If the user has asked for a blocksize that is greater than the 2006 * backing device's blocksize, we can do it only if the blocksize 2007 * the user is asking for is an even multiple of the underlying 2008 * device's blocksize. 2009 */ 2010 if ((params->blocksize_bytes != 0) && 2011 (params->blocksize_bytes >= tmp)) { 2012 if (params->blocksize_bytes % tmp == 0) { 2013 cbe_lun->blocksize = params->blocksize_bytes; 2014 } else { 2015 dev_relthread(dev, ref); 2016 snprintf(req->error_str, sizeof(req->error_str), 2017 "requested blocksize %u is not an even " 2018 "multiple of backing device blocksize %u", 2019 params->blocksize_bytes, tmp); 2020 return (EINVAL); 2021 } 2022 } else if (params->blocksize_bytes != 0) { 2023 dev_relthread(dev, ref); 2024 snprintf(req->error_str, sizeof(req->error_str), 2025 "requested blocksize %u < backing device " 2026 "blocksize %u", params->blocksize_bytes, tmp); 2027 return (EINVAL); 2028 } else 2029 cbe_lun->blocksize = tmp; 2030 2031 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 2032 curthread); 2033 if (error) { 2034 dev_relthread(dev, ref); 2035 snprintf(req->error_str, sizeof(req->error_str), 2036 "error %d returned for DIOCGMEDIASIZE " 2037 " ioctl on %s!", error, 2038 be_lun->dev_path); 2039 return (error); 2040 } 2041 2042 if (params->lun_size_bytes != 0) { 2043 if (params->lun_size_bytes > otmp) { 2044 dev_relthread(dev, ref); 2045 snprintf(req->error_str, sizeof(req->error_str), 2046 "requested LUN size %ju > backing device " 2047 "size %ju", 2048 (uintmax_t)params->lun_size_bytes, 2049 (uintmax_t)otmp); 2050 return (EINVAL); 2051 } 2052 2053 be_lun->size_bytes = params->lun_size_bytes; 2054 } else 2055 be_lun->size_bytes = otmp; 2056 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2057 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2058 0 : (be_lun->size_blocks - 1); 2059 2060 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2061 curthread); 2062 if (error) 2063 ps = po = 0; 2064 else { 2065 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2066 FREAD, curthread); 2067 if (error) 2068 po = 0; 2069 } 2070 us = ps; 2071 uo = po; 2072 2073 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 2074 if (value != NULL) 2075 ctl_expand_number(value, &ps); 2076 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 2077 if (value != NULL) 2078 ctl_expand_number(value, &po); 2079 pss = ps / cbe_lun->blocksize; 2080 pos = po / cbe_lun->blocksize; 2081 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2082 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2083 cbe_lun->pblockexp = fls(pss) - 1; 2084 cbe_lun->pblockoff = (pss - pos) % pss; 2085 } 2086 2087 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 2088 if (value != NULL) 2089 ctl_expand_number(value, &us); 2090 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 2091 if (value != NULL) 2092 ctl_expand_number(value, &uo); 2093 uss = us / cbe_lun->blocksize; 2094 uos = uo / cbe_lun->blocksize; 2095 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2096 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2097 cbe_lun->ublockexp = fls(uss) - 1; 2098 cbe_lun->ublockoff = (uss - uos) % uss; 2099 } 2100 2101 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2102 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2103 2104 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2105 unmap = 1; 2106 } else { 2107 struct diocgattr_arg arg; 2108 2109 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2110 arg.len = sizeof(arg.value.i); 2111 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2112 curthread); 2113 unmap = (error == 0) ? arg.value.i : 0; 2114 } 2115 value = ctl_get_opt(&cbe_lun->options, "unmap"); 2116 if (value != NULL) 2117 unmap = (strcmp(value, "on") == 0); 2118 if (unmap) 2119 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2120 else 2121 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2122 2123 dev_relthread(dev, ref); 2124 return (0); 2125} 2126 2127static int 2128ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2129{ 2130 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2131 int flags; 2132 2133 if (be_lun->vn) { 2134 flags = FREAD; 2135 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2136 flags |= FWRITE; 2137 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2138 be_lun->vn = NULL; 2139 2140 switch (be_lun->dev_type) { 2141 case CTL_BE_BLOCK_DEV: 2142 break; 2143 case CTL_BE_BLOCK_FILE: 2144 if (be_lun->backend.file.cred != NULL) { 2145 crfree(be_lun->backend.file.cred); 2146 be_lun->backend.file.cred = NULL; 2147 } 2148 break; 2149 case CTL_BE_BLOCK_NONE: 2150 break; 2151 default: 2152 panic("Unexpected backend type."); 2153 break; 2154 } 2155 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2156 } 2157 return (0); 2158} 2159 2160static int 2161ctl_be_block_open(struct ctl_be_block_softc *softc, 2162 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2163{ 2164 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2165 struct nameidata nd; 2166 char *value; 2167 int error, flags; 2168 2169 error = 0; 2170 if (rootvnode == NULL) { 2171 snprintf(req->error_str, sizeof(req->error_str), 2172 "Root filesystem is not mounted"); 2173 return (1); 2174 } 2175 if (!curthread->td_proc->p_fd->fd_cdir) { 2176 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2177 VREF(rootvnode); 2178 } 2179 if (!curthread->td_proc->p_fd->fd_rdir) { 2180 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2181 VREF(rootvnode); 2182 } 2183 if (!curthread->td_proc->p_fd->fd_jdir) { 2184 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2185 VREF(rootvnode); 2186 } 2187 2188 value = ctl_get_opt(&cbe_lun->options, "file"); 2189 if (value == NULL) { 2190 snprintf(req->error_str, sizeof(req->error_str), 2191 "no file argument specified"); 2192 return (1); 2193 } 2194 free(be_lun->dev_path, M_CTLBLK); 2195 be_lun->dev_path = strdup(value, M_CTLBLK); 2196 2197 flags = FREAD; 2198 value = ctl_get_opt(&cbe_lun->options, "readonly"); 2199 if (value == NULL || strcmp(value, "on") != 0) 2200 flags |= FWRITE; 2201 2202again: 2203 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2204 error = vn_open(&nd, &flags, 0, NULL); 2205 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2206 flags &= ~FWRITE; 2207 goto again; 2208 } 2209 if (error) { 2210 /* 2211 * This is the only reasonable guess we can make as far as 2212 * path if the user doesn't give us a fully qualified path. 2213 * If they want to specify a file, they need to specify the 2214 * full path. 2215 */ 2216 if (be_lun->dev_path[0] != '/') { 2217 char *dev_name; 2218 2219 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2220 be_lun->dev_path); 2221 free(be_lun->dev_path, M_CTLBLK); 2222 be_lun->dev_path = dev_name; 2223 goto again; 2224 } 2225 snprintf(req->error_str, sizeof(req->error_str), 2226 "error opening %s: %d", be_lun->dev_path, error); 2227 return (error); 2228 } 2229 if (flags & FWRITE) 2230 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2231 else 2232 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2233 2234 NDFREE(&nd, NDF_ONLY_PNBUF); 2235 be_lun->vn = nd.ni_vp; 2236 2237 /* We only support disks and files. */ 2238 if (vn_isdisk(be_lun->vn, &error)) { 2239 error = ctl_be_block_open_dev(be_lun, req); 2240 } else if (be_lun->vn->v_type == VREG) { 2241 error = ctl_be_block_open_file(be_lun, req); 2242 } else { 2243 error = EINVAL; 2244 snprintf(req->error_str, sizeof(req->error_str), 2245 "%s is not a disk or plain file", be_lun->dev_path); 2246 } 2247 VOP_UNLOCK(be_lun->vn, 0); 2248 2249 if (error != 0) 2250 ctl_be_block_close(be_lun); 2251 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2252 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2253 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2254 value = ctl_get_opt(&cbe_lun->options, "serseq"); 2255 if (value != NULL && strcmp(value, "on") == 0) 2256 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2257 else if (value != NULL && strcmp(value, "read") == 0) 2258 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2259 else if (value != NULL && strcmp(value, "off") == 0) 2260 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2261 return (0); 2262} 2263 2264static int 2265ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2266{ 2267 struct ctl_be_lun *cbe_lun; 2268 struct ctl_be_block_lun *be_lun; 2269 struct ctl_lun_create_params *params; 2270 char num_thread_str[16]; 2271 char tmpstr[32]; 2272 char *value; 2273 int retval, num_threads; 2274 int tmp_num_threads; 2275 2276 params = &req->reqdata.create; 2277 retval = 0; 2278 req->status = CTL_LUN_OK; 2279 2280 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2281 cbe_lun = &be_lun->cbe_lun; 2282 cbe_lun->be_lun = be_lun; 2283 be_lun->params = req->reqdata.create; 2284 be_lun->softc = softc; 2285 STAILQ_INIT(&be_lun->input_queue); 2286 STAILQ_INIT(&be_lun->config_read_queue); 2287 STAILQ_INIT(&be_lun->config_write_queue); 2288 STAILQ_INIT(&be_lun->datamove_queue); 2289 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2290 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2291 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2292 ctl_init_opts(&cbe_lun->options, 2293 req->num_be_args, req->kern_be_args); 2294 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2295 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2296 if (be_lun->lun_zone == NULL) { 2297 snprintf(req->error_str, sizeof(req->error_str), 2298 "error allocating UMA zone"); 2299 goto bailout_error; 2300 } 2301 2302 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2303 cbe_lun->lun_type = params->device_type; 2304 else 2305 cbe_lun->lun_type = T_DIRECT; 2306 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2307 cbe_lun->flags = 0; 2308 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2309 if (value != NULL) { 2310 if (strcmp(value, "primary") == 0) 2311 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2312 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2313 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2314 2315 if (cbe_lun->lun_type == T_DIRECT) { 2316 be_lun->size_bytes = params->lun_size_bytes; 2317 if (params->blocksize_bytes != 0) 2318 cbe_lun->blocksize = params->blocksize_bytes; 2319 else 2320 cbe_lun->blocksize = 512; 2321 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2322 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2323 0 : (be_lun->size_blocks - 1); 2324 2325 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2326 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2327 retval = ctl_be_block_open(softc, be_lun, req); 2328 if (retval != 0) { 2329 retval = 0; 2330 req->status = CTL_LUN_WARNING; 2331 } 2332 } 2333 num_threads = cbb_num_threads; 2334 } else { 2335 num_threads = 1; 2336 } 2337 2338 /* 2339 * XXX This searching loop might be refactored to be combined with 2340 * the loop above, 2341 */ 2342 value = ctl_get_opt(&cbe_lun->options, "num_threads"); 2343 if (value != NULL) { 2344 tmp_num_threads = strtol(value, NULL, 0); 2345 2346 /* 2347 * We don't let the user specify less than one 2348 * thread, but hope he's clueful enough not to 2349 * specify 1000 threads. 2350 */ 2351 if (tmp_num_threads < 1) { 2352 snprintf(req->error_str, sizeof(req->error_str), 2353 "invalid number of threads %s", 2354 num_thread_str); 2355 goto bailout_error; 2356 } 2357 num_threads = tmp_num_threads; 2358 } 2359 2360 if (be_lun->vn == NULL) 2361 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2362 /* Tell the user the blocksize we ended up using */ 2363 params->lun_size_bytes = be_lun->size_bytes; 2364 params->blocksize_bytes = cbe_lun->blocksize; 2365 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2366 cbe_lun->req_lun_id = params->req_lun_id; 2367 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2368 } else 2369 cbe_lun->req_lun_id = 0; 2370 2371 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2372 cbe_lun->lun_config_status = ctl_be_block_lun_config_status; 2373 cbe_lun->be = &ctl_be_block_driver; 2374 2375 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2376 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2377 softc->num_luns); 2378 strncpy((char *)cbe_lun->serial_num, tmpstr, 2379 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2380 2381 /* Tell the user what we used for a serial number */ 2382 strncpy((char *)params->serial_num, tmpstr, 2383 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2384 } else { 2385 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2386 MIN(sizeof(cbe_lun->serial_num), 2387 sizeof(params->serial_num))); 2388 } 2389 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2390 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2391 strncpy((char *)cbe_lun->device_id, tmpstr, 2392 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2393 2394 /* Tell the user what we used for a device ID */ 2395 strncpy((char *)params->device_id, tmpstr, 2396 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2397 } else { 2398 strncpy((char *)cbe_lun->device_id, params->device_id, 2399 MIN(sizeof(cbe_lun->device_id), 2400 sizeof(params->device_id))); 2401 } 2402 2403 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2404 2405 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2406 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2407 2408 if (be_lun->io_taskqueue == NULL) { 2409 snprintf(req->error_str, sizeof(req->error_str), 2410 "unable to create taskqueue"); 2411 goto bailout_error; 2412 } 2413 2414 /* 2415 * Note that we start the same number of threads by default for 2416 * both the file case and the block device case. For the file 2417 * case, we need multiple threads to allow concurrency, because the 2418 * vnode interface is designed to be a blocking interface. For the 2419 * block device case, ZFS zvols at least will block the caller's 2420 * context in many instances, and so we need multiple threads to 2421 * overcome that problem. Other block devices don't need as many 2422 * threads, but they shouldn't cause too many problems. 2423 * 2424 * If the user wants to just have a single thread for a block 2425 * device, he can specify that when the LUN is created, or change 2426 * the tunable/sysctl to alter the default number of threads. 2427 */ 2428 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2429 /*num threads*/num_threads, 2430 /*priority*/PWAIT, 2431 /*thread name*/ 2432 "%s taskq", be_lun->lunname); 2433 2434 if (retval != 0) 2435 goto bailout_error; 2436 2437 be_lun->num_threads = num_threads; 2438 2439 mtx_lock(&softc->lock); 2440 softc->num_luns++; 2441 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2442 2443 mtx_unlock(&softc->lock); 2444 2445 retval = ctl_add_lun(&be_lun->cbe_lun); 2446 if (retval != 0) { 2447 mtx_lock(&softc->lock); 2448 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2449 links); 2450 softc->num_luns--; 2451 mtx_unlock(&softc->lock); 2452 snprintf(req->error_str, sizeof(req->error_str), 2453 "ctl_add_lun() returned error %d, see dmesg for " 2454 "details", retval); 2455 retval = 0; 2456 goto bailout_error; 2457 } 2458 2459 mtx_lock(&softc->lock); 2460 2461 /* 2462 * Tell the config_status routine that we're waiting so it won't 2463 * clean up the LUN in the event of an error. 2464 */ 2465 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2466 2467 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2468 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2469 if (retval == EINTR) 2470 break; 2471 } 2472 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2473 2474 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2475 snprintf(req->error_str, sizeof(req->error_str), 2476 "LUN configuration error, see dmesg for details"); 2477 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2478 links); 2479 softc->num_luns--; 2480 mtx_unlock(&softc->lock); 2481 goto bailout_error; 2482 } else { 2483 params->req_lun_id = cbe_lun->lun_id; 2484 } 2485 2486 mtx_unlock(&softc->lock); 2487 2488 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2489 cbe_lun->blocksize, 2490 DEVSTAT_ALL_SUPPORTED, 2491 cbe_lun->lun_type 2492 | DEVSTAT_TYPE_IF_OTHER, 2493 DEVSTAT_PRIORITY_OTHER); 2494 2495 return (retval); 2496 2497bailout_error: 2498 req->status = CTL_LUN_ERROR; 2499 2500 if (be_lun->io_taskqueue != NULL) 2501 taskqueue_free(be_lun->io_taskqueue); 2502 ctl_be_block_close(be_lun); 2503 if (be_lun->dev_path != NULL) 2504 free(be_lun->dev_path, M_CTLBLK); 2505 if (be_lun->lun_zone != NULL) 2506 uma_zdestroy(be_lun->lun_zone); 2507 ctl_free_opts(&cbe_lun->options); 2508 mtx_destroy(&be_lun->queue_lock); 2509 mtx_destroy(&be_lun->io_lock); 2510 free(be_lun, M_CTLBLK); 2511 2512 return (retval); 2513} 2514 2515static int 2516ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2517{ 2518 struct ctl_lun_rm_params *params; 2519 struct ctl_be_block_lun *be_lun; 2520 struct ctl_be_lun *cbe_lun; 2521 int retval; 2522 2523 params = &req->reqdata.rm; 2524 2525 mtx_lock(&softc->lock); 2526 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2527 if (be_lun->cbe_lun.lun_id == params->lun_id) 2528 break; 2529 } 2530 mtx_unlock(&softc->lock); 2531 2532 if (be_lun == NULL) { 2533 snprintf(req->error_str, sizeof(req->error_str), 2534 "LUN %u is not managed by the block backend", 2535 params->lun_id); 2536 goto bailout_error; 2537 } 2538 cbe_lun = &be_lun->cbe_lun; 2539 2540 retval = ctl_disable_lun(cbe_lun); 2541 if (retval != 0) { 2542 snprintf(req->error_str, sizeof(req->error_str), 2543 "error %d returned from ctl_disable_lun() for " 2544 "LUN %d", retval, params->lun_id); 2545 goto bailout_error; 2546 } 2547 2548 if (be_lun->vn != NULL) { 2549 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2550 ctl_lun_offline(cbe_lun); 2551 taskqueue_drain_all(be_lun->io_taskqueue); 2552 ctl_be_block_close(be_lun); 2553 } 2554 2555 retval = ctl_invalidate_lun(cbe_lun); 2556 if (retval != 0) { 2557 snprintf(req->error_str, sizeof(req->error_str), 2558 "error %d returned from ctl_invalidate_lun() for " 2559 "LUN %d", retval, params->lun_id); 2560 goto bailout_error; 2561 } 2562 2563 mtx_lock(&softc->lock); 2564 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2565 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2566 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2567 if (retval == EINTR) 2568 break; 2569 } 2570 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2571 2572 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2573 snprintf(req->error_str, sizeof(req->error_str), 2574 "interrupted waiting for LUN to be freed"); 2575 mtx_unlock(&softc->lock); 2576 goto bailout_error; 2577 } 2578 2579 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2580 2581 softc->num_luns--; 2582 mtx_unlock(&softc->lock); 2583 2584 taskqueue_drain_all(be_lun->io_taskqueue); 2585 taskqueue_free(be_lun->io_taskqueue); 2586 2587 if (be_lun->disk_stats != NULL) 2588 devstat_remove_entry(be_lun->disk_stats); 2589 2590 uma_zdestroy(be_lun->lun_zone); 2591 2592 ctl_free_opts(&cbe_lun->options); 2593 free(be_lun->dev_path, M_CTLBLK); 2594 mtx_destroy(&be_lun->queue_lock); 2595 mtx_destroy(&be_lun->io_lock); 2596 free(be_lun, M_CTLBLK); 2597 2598 req->status = CTL_LUN_OK; 2599 2600 return (0); 2601 2602bailout_error: 2603 2604 req->status = CTL_LUN_ERROR; 2605 2606 return (0); 2607} 2608 2609static int 2610ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2611 struct ctl_lun_req *req) 2612{ 2613 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2614 struct vattr vattr; 2615 int error; 2616 struct ctl_lun_create_params *params = &be_lun->params; 2617 2618 if (params->lun_size_bytes != 0) { 2619 be_lun->size_bytes = params->lun_size_bytes; 2620 } else { 2621 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2622 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2623 VOP_UNLOCK(be_lun->vn, 0); 2624 if (error != 0) { 2625 snprintf(req->error_str, sizeof(req->error_str), 2626 "error calling VOP_GETATTR() for file %s", 2627 be_lun->dev_path); 2628 return (error); 2629 } 2630 be_lun->size_bytes = vattr.va_size; 2631 } 2632 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2633 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2634 0 : (be_lun->size_blocks - 1); 2635 return (0); 2636} 2637 2638static int 2639ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2640 struct ctl_lun_req *req) 2641{ 2642 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2643 struct ctl_lun_create_params *params = &be_lun->params; 2644 struct cdevsw *csw; 2645 struct cdev *dev; 2646 uint64_t size_bytes; 2647 int error, ref; 2648 2649 csw = devvn_refthread(be_lun->vn, &dev, &ref); 2650 if (csw == NULL) 2651 return (ENXIO); 2652 if (csw->d_ioctl == NULL) { 2653 dev_relthread(dev, ref); 2654 snprintf(req->error_str, sizeof(req->error_str), 2655 "no d_ioctl for device %s!", be_lun->dev_path); 2656 return (ENODEV); 2657 } 2658 2659 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&size_bytes, FREAD, 2660 curthread); 2661 dev_relthread(dev, ref); 2662 if (error) { 2663 snprintf(req->error_str, sizeof(req->error_str), 2664 "error %d returned for DIOCGMEDIASIZE ioctl " 2665 "on %s!", error, be_lun->dev_path); 2666 return (error); 2667 } 2668 2669 if (params->lun_size_bytes != 0) { 2670 if (params->lun_size_bytes > size_bytes) { 2671 snprintf(req->error_str, sizeof(req->error_str), 2672 "requested LUN size %ju > backing device " 2673 "size %ju", 2674 (uintmax_t)params->lun_size_bytes, 2675 (uintmax_t)size_bytes); 2676 return (EINVAL); 2677 } 2678 be_lun->size_bytes = params->lun_size_bytes; 2679 } else { 2680 be_lun->size_bytes = size_bytes; 2681 } 2682 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2683 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2684 0 : (be_lun->size_blocks - 1); 2685 return (0); 2686} 2687 2688static int 2689ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2690{ 2691 struct ctl_lun_modify_params *params; 2692 struct ctl_be_block_lun *be_lun; 2693 struct ctl_be_lun *cbe_lun; 2694 char *value; 2695 uint64_t oldsize; 2696 int error, wasprim; 2697 2698 params = &req->reqdata.modify; 2699 2700 mtx_lock(&softc->lock); 2701 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2702 if (be_lun->cbe_lun.lun_id == params->lun_id) 2703 break; 2704 } 2705 mtx_unlock(&softc->lock); 2706 2707 if (be_lun == NULL) { 2708 snprintf(req->error_str, sizeof(req->error_str), 2709 "LUN %u is not managed by the block backend", 2710 params->lun_id); 2711 goto bailout_error; 2712 } 2713 cbe_lun = &be_lun->cbe_lun; 2714 2715 if (params->lun_size_bytes != 0) 2716 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2717 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); 2718 2719 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2720 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2721 if (value != NULL) { 2722 if (strcmp(value, "primary") == 0) 2723 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2724 else 2725 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2726 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2727 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2728 else 2729 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2730 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2731 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2732 ctl_lun_primary(cbe_lun); 2733 else 2734 ctl_lun_secondary(cbe_lun); 2735 } 2736 2737 oldsize = be_lun->size_blocks; 2738 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2739 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2740 if (be_lun->vn == NULL) 2741 error = ctl_be_block_open(softc, be_lun, req); 2742 else if (vn_isdisk(be_lun->vn, &error)) 2743 error = ctl_be_block_modify_dev(be_lun, req); 2744 else if (be_lun->vn->v_type == VREG) 2745 error = ctl_be_block_modify_file(be_lun, req); 2746 else 2747 error = EINVAL; 2748 if ((cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) && 2749 be_lun->vn != NULL) { 2750 cbe_lun->flags &= ~CTL_LUN_FLAG_OFFLINE; 2751 ctl_lun_online(cbe_lun); 2752 } 2753 } else { 2754 if (be_lun->vn != NULL) { 2755 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2756 ctl_lun_offline(cbe_lun); 2757 taskqueue_drain_all(be_lun->io_taskqueue); 2758 error = ctl_be_block_close(be_lun); 2759 } else 2760 error = 0; 2761 } 2762 if (be_lun->size_blocks != oldsize) 2763 ctl_lun_capacity_changed(cbe_lun); 2764 2765 /* Tell the user the exact size we ended up using */ 2766 params->lun_size_bytes = be_lun->size_bytes; 2767 2768 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2769 return (0); 2770 2771bailout_error: 2772 req->status = CTL_LUN_ERROR; 2773 return (0); 2774} 2775 2776static void 2777ctl_be_block_lun_shutdown(void *be_lun) 2778{ 2779 struct ctl_be_block_lun *lun; 2780 struct ctl_be_block_softc *softc; 2781 2782 lun = (struct ctl_be_block_lun *)be_lun; 2783 2784 softc = lun->softc; 2785 2786 mtx_lock(&softc->lock); 2787 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2788 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2789 wakeup(lun); 2790 mtx_unlock(&softc->lock); 2791 2792} 2793 2794static void 2795ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2796{ 2797 struct ctl_be_block_lun *lun; 2798 struct ctl_be_block_softc *softc; 2799 2800 lun = (struct ctl_be_block_lun *)be_lun; 2801 softc = lun->softc; 2802 2803 if (status == CTL_LUN_CONFIG_OK) { 2804 mtx_lock(&softc->lock); 2805 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2806 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2807 wakeup(lun); 2808 mtx_unlock(&softc->lock); 2809 2810 /* 2811 * We successfully added the LUN, attempt to enable it. 2812 */ 2813 if (ctl_enable_lun(&lun->cbe_lun) != 0) { 2814 printf("%s: ctl_enable_lun() failed!\n", __func__); 2815 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { 2816 printf("%s: ctl_invalidate_lun() failed!\n", 2817 __func__); 2818 } 2819 } 2820 2821 return; 2822 } 2823 2824 2825 mtx_lock(&softc->lock); 2826 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2827 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2828 wakeup(lun); 2829 mtx_unlock(&softc->lock); 2830} 2831 2832 2833static int 2834ctl_be_block_config_write(union ctl_io *io) 2835{ 2836 struct ctl_be_block_lun *be_lun; 2837 struct ctl_be_lun *cbe_lun; 2838 int retval; 2839 2840 retval = 0; 2841 2842 DPRINTF("entered\n"); 2843 2844 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2845 CTL_PRIV_BACKEND_LUN].ptr; 2846 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2847 2848 switch (io->scsiio.cdb[0]) { 2849 case SYNCHRONIZE_CACHE: 2850 case SYNCHRONIZE_CACHE_16: 2851 case WRITE_SAME_10: 2852 case WRITE_SAME_16: 2853 case UNMAP: 2854 /* 2855 * The upper level CTL code will filter out any CDBs with 2856 * the immediate bit set and return the proper error. 2857 * 2858 * We don't really need to worry about what LBA range the 2859 * user asked to be synced out. When they issue a sync 2860 * cache command, we'll sync out the whole thing. 2861 */ 2862 mtx_lock(&be_lun->queue_lock); 2863 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2864 links); 2865 mtx_unlock(&be_lun->queue_lock); 2866 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2867 break; 2868 case START_STOP_UNIT: { 2869 struct scsi_start_stop_unit *cdb; 2870 2871 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2872 2873 if (cdb->how & SSS_START) 2874 retval = ctl_start_lun(cbe_lun); 2875 else { 2876 retval = ctl_stop_lun(cbe_lun); 2877 /* 2878 * XXX KDM Copan-specific offline behavior. 2879 * Figure out a reasonable way to port this? 2880 */ 2881#ifdef NEEDTOPORT 2882 if ((retval == 0) 2883 && (cdb->byte2 & SSS_ONOFFLINE)) 2884 retval = ctl_lun_offline(cbe_lun); 2885#endif 2886 } 2887 2888 /* 2889 * In general, the above routines should not fail. They 2890 * just set state for the LUN. So we've got something 2891 * pretty wrong here if we can't start or stop the LUN. 2892 */ 2893 if (retval != 0) { 2894 ctl_set_internal_failure(&io->scsiio, 2895 /*sks_valid*/ 1, 2896 /*retry_count*/ 0xf051); 2897 retval = CTL_RETVAL_COMPLETE; 2898 } else { 2899 ctl_set_success(&io->scsiio); 2900 } 2901 ctl_config_write_done(io); 2902 break; 2903 } 2904 default: 2905 ctl_set_invalid_opcode(&io->scsiio); 2906 ctl_config_write_done(io); 2907 retval = CTL_RETVAL_COMPLETE; 2908 break; 2909 } 2910 2911 return (retval); 2912} 2913 2914static int 2915ctl_be_block_config_read(union ctl_io *io) 2916{ 2917 struct ctl_be_block_lun *be_lun; 2918 struct ctl_be_lun *cbe_lun; 2919 int retval = 0; 2920 2921 DPRINTF("entered\n"); 2922 2923 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2924 CTL_PRIV_BACKEND_LUN].ptr; 2925 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2926 2927 switch (io->scsiio.cdb[0]) { 2928 case SERVICE_ACTION_IN: 2929 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2930 mtx_lock(&be_lun->queue_lock); 2931 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2932 &io->io_hdr, links); 2933 mtx_unlock(&be_lun->queue_lock); 2934 taskqueue_enqueue(be_lun->io_taskqueue, 2935 &be_lun->io_task); 2936 retval = CTL_RETVAL_QUEUED; 2937 break; 2938 } 2939 ctl_set_invalid_field(&io->scsiio, 2940 /*sks_valid*/ 1, 2941 /*command*/ 1, 2942 /*field*/ 1, 2943 /*bit_valid*/ 1, 2944 /*bit*/ 4); 2945 ctl_config_read_done(io); 2946 retval = CTL_RETVAL_COMPLETE; 2947 break; 2948 default: 2949 ctl_set_invalid_opcode(&io->scsiio); 2950 ctl_config_read_done(io); 2951 retval = CTL_RETVAL_COMPLETE; 2952 break; 2953 } 2954 2955 return (retval); 2956} 2957 2958static int 2959ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2960{ 2961 struct ctl_be_block_lun *lun; 2962 int retval; 2963 2964 lun = (struct ctl_be_block_lun *)be_lun; 2965 retval = 0; 2966 2967 retval = sbuf_printf(sb, "\t<num_threads>"); 2968 2969 if (retval != 0) 2970 goto bailout; 2971 2972 retval = sbuf_printf(sb, "%d", lun->num_threads); 2973 2974 if (retval != 0) 2975 goto bailout; 2976 2977 retval = sbuf_printf(sb, "</num_threads>\n"); 2978 2979bailout: 2980 2981 return (retval); 2982} 2983 2984static uint64_t 2985ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2986{ 2987 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2988 2989 if (lun->getattr == NULL) 2990 return (UINT64_MAX); 2991 return (lun->getattr(lun, attrname)); 2992} 2993 2994int 2995ctl_be_block_init(void) 2996{ 2997 struct ctl_be_block_softc *softc; 2998 int retval; 2999 3000 softc = &backend_block_softc; 3001 retval = 0; 3002 3003 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 3004 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 3005 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 3006 STAILQ_INIT(&softc->lun_list); 3007 3008 return (retval); 3009} 3010