ctl_backend_block.c revision 313364
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 * Copyright (c) 2014-2015 Alexander Motin <mav@FreeBSD.org> 6 * All rights reserved. 7 * 8 * Portions of this software were developed by Edward Tomasz Napierala 9 * under sponsorship from the FreeBSD Foundation. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions, and the following disclaimer, 16 * without modification. 17 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 18 * substantially similar to the "NO WARRANTY" disclaimer below 19 * ("Disclaimer") and any redistribution must be conditioned upon 20 * including a substantially similar Disclaimer requirement for further 21 * binary redistribution. 22 * 23 * NO WARRANTY 24 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 25 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 26 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 27 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 28 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 32 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 33 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 34 * POSSIBILITY OF SUCH DAMAGES. 35 * 36 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 37 */ 38/* 39 * CAM Target Layer driver backend for block devices. 40 * 41 * Author: Ken Merry <ken@FreeBSD.org> 42 */ 43#include <sys/cdefs.h> 44__FBSDID("$FreeBSD: stable/11/sys/cam/ctl/ctl_backend_block.c 313364 2017-02-07 01:42:13Z mav $"); 45 46#include <sys/param.h> 47#include <sys/systm.h> 48#include <sys/kernel.h> 49#include <sys/types.h> 50#include <sys/kthread.h> 51#include <sys/bio.h> 52#include <sys/fcntl.h> 53#include <sys/limits.h> 54#include <sys/lock.h> 55#include <sys/mutex.h> 56#include <sys/condvar.h> 57#include <sys/malloc.h> 58#include <sys/conf.h> 59#include <sys/ioccom.h> 60#include <sys/queue.h> 61#include <sys/sbuf.h> 62#include <sys/endian.h> 63#include <sys/uio.h> 64#include <sys/buf.h> 65#include <sys/taskqueue.h> 66#include <sys/vnode.h> 67#include <sys/namei.h> 68#include <sys/mount.h> 69#include <sys/disk.h> 70#include <sys/fcntl.h> 71#include <sys/filedesc.h> 72#include <sys/filio.h> 73#include <sys/proc.h> 74#include <sys/pcpu.h> 75#include <sys/module.h> 76#include <sys/sdt.h> 77#include <sys/devicestat.h> 78#include <sys/sysctl.h> 79 80#include <geom/geom.h> 81 82#include <cam/cam.h> 83#include <cam/scsi/scsi_all.h> 84#include <cam/scsi/scsi_da.h> 85#include <cam/ctl/ctl_io.h> 86#include <cam/ctl/ctl.h> 87#include <cam/ctl/ctl_backend.h> 88#include <cam/ctl/ctl_ioctl.h> 89#include <cam/ctl/ctl_ha.h> 90#include <cam/ctl/ctl_scsi_all.h> 91#include <cam/ctl/ctl_private.h> 92#include <cam/ctl/ctl_error.h> 93 94/* 95 * The idea here is that we'll allocate enough S/G space to hold a 1MB 96 * I/O. If we get an I/O larger than that, we'll split it. 97 */ 98#define CTLBLK_HALF_IO_SIZE (512 * 1024) 99#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 100#define CTLBLK_MAX_SEG MAXPHYS 101#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 102#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 103 104#ifdef CTLBLK_DEBUG 105#define DPRINTF(fmt, args...) \ 106 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 107#else 108#define DPRINTF(fmt, args...) do {} while(0) 109#endif 110 111#define PRIV(io) \ 112 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 113#define ARGS(io) \ 114 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 115 116SDT_PROVIDER_DEFINE(cbb); 117 118typedef enum { 119 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 120 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 121 CTL_BE_BLOCK_LUN_WAITING = 0x04, 122} ctl_be_block_lun_flags; 123 124typedef enum { 125 CTL_BE_BLOCK_NONE, 126 CTL_BE_BLOCK_DEV, 127 CTL_BE_BLOCK_FILE 128} ctl_be_block_type; 129 130struct ctl_be_block_filedata { 131 struct ucred *cred; 132}; 133 134union ctl_be_block_bedata { 135 struct ctl_be_block_filedata file; 136}; 137 138struct ctl_be_block_io; 139struct ctl_be_block_lun; 140 141typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 142 struct ctl_be_block_io *beio); 143typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 144 const char *attrname); 145 146/* 147 * Backend LUN structure. There is a 1:1 mapping between a block device 148 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 149 */ 150struct ctl_be_block_lun { 151 struct ctl_lun_create_params params; 152 char lunname[32]; 153 char *dev_path; 154 ctl_be_block_type dev_type; 155 struct vnode *vn; 156 union ctl_be_block_bedata backend; 157 cbb_dispatch_t dispatch; 158 cbb_dispatch_t lun_flush; 159 cbb_dispatch_t unmap; 160 cbb_dispatch_t get_lba_status; 161 cbb_getattr_t getattr; 162 uma_zone_t lun_zone; 163 uint64_t size_blocks; 164 uint64_t size_bytes; 165 struct ctl_be_block_softc *softc; 166 struct devstat *disk_stats; 167 ctl_be_block_lun_flags flags; 168 STAILQ_ENTRY(ctl_be_block_lun) links; 169 struct ctl_be_lun cbe_lun; 170 struct taskqueue *io_taskqueue; 171 struct task io_task; 172 int num_threads; 173 STAILQ_HEAD(, ctl_io_hdr) input_queue; 174 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 175 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 176 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 177 struct mtx_padalign io_lock; 178 struct mtx_padalign queue_lock; 179}; 180 181/* 182 * Overall softc structure for the block backend module. 183 */ 184struct ctl_be_block_softc { 185 struct mtx lock; 186 int num_luns; 187 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 188}; 189 190static struct ctl_be_block_softc backend_block_softc; 191 192/* 193 * Per-I/O information. 194 */ 195struct ctl_be_block_io { 196 union ctl_io *io; 197 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 198 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 199 int bio_cmd; 200 int num_segs; 201 int num_bios_sent; 202 int num_bios_done; 203 int send_complete; 204 int first_error; 205 uint64_t first_error_offset; 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; 220SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 221 "CAM Target Layer Block Backend"); 222SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 223 &cbb_num_threads, 0, "Number of threads per backing file"); 224 225static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 226static void ctl_free_beio(struct ctl_be_block_io *beio); 227static void ctl_complete_beio(struct ctl_be_block_io *beio); 228static int ctl_be_block_move_done(union ctl_io *io); 229static void ctl_be_block_biodone(struct bio *bio); 230static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 231 struct ctl_be_block_io *beio); 232static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 233 struct ctl_be_block_io *beio); 234static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 235 struct ctl_be_block_io *beio); 236static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 237 const char *attrname); 238static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 239 struct ctl_be_block_io *beio); 240static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 241 struct ctl_be_block_io *beio); 242static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 243 struct ctl_be_block_io *beio); 244static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 245 const char *attrname); 246static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 247 union ctl_io *io); 248static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 249 union ctl_io *io); 250static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 251 union ctl_io *io); 252static void ctl_be_block_worker(void *context, int pending); 253static int ctl_be_block_submit(union ctl_io *io); 254static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 255 int flag, struct thread *td); 256static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 257 struct ctl_lun_req *req); 258static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 259 struct ctl_lun_req *req); 260static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 261static int ctl_be_block_open(struct ctl_be_block_lun *be_lun, 262 struct ctl_lun_req *req); 263static int ctl_be_block_create(struct ctl_be_block_softc *softc, 264 struct ctl_lun_req *req); 265static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 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); 275static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 276int ctl_be_block_init(void); 277 278static struct ctl_backend_driver ctl_be_block_driver = 279{ 280 .name = "block", 281 .flags = CTL_BE_FLAG_HAS_CONFIG, 282 .init = ctl_be_block_init, 283 .data_submit = ctl_be_block_submit, 284 .data_move_done = ctl_be_block_move_done, 285 .config_read = ctl_be_block_config_read, 286 .config_write = ctl_be_block_config_write, 287 .ioctl = ctl_be_block_ioctl, 288 .lun_info = ctl_be_block_lun_info, 289 .lun_attr = ctl_be_block_lun_attr 290}; 291 292MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 293CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 294 295static uma_zone_t beio_zone; 296 297static struct ctl_be_block_io * 298ctl_alloc_beio(struct ctl_be_block_softc *softc) 299{ 300 struct ctl_be_block_io *beio; 301 302 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 303 beio->softc = softc; 304 return (beio); 305} 306 307static void 308ctl_free_beio(struct ctl_be_block_io *beio) 309{ 310 int duplicate_free; 311 int i; 312 313 duplicate_free = 0; 314 315 for (i = 0; i < beio->num_segs; i++) { 316 if (beio->sg_segs[i].addr == NULL) 317 duplicate_free++; 318 319 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 320 beio->sg_segs[i].addr = NULL; 321 322 /* For compare we had two equal S/G lists. */ 323 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 324 uma_zfree(beio->lun->lun_zone, 325 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 326 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 327 } 328 } 329 330 if (duplicate_free > 0) { 331 printf("%s: %d duplicate frees out of %d segments\n", __func__, 332 duplicate_free, beio->num_segs); 333 } 334 335 uma_zfree(beio_zone, beio); 336} 337 338static void 339ctl_complete_beio(struct ctl_be_block_io *beio) 340{ 341 union ctl_io *io = beio->io; 342 343 if (beio->beio_cont != NULL) { 344 beio->beio_cont(beio); 345 } else { 346 ctl_free_beio(beio); 347 ctl_data_submit_done(io); 348 } 349} 350 351static size_t 352cmp(uint8_t *a, uint8_t *b, size_t size) 353{ 354 size_t i; 355 356 for (i = 0; i < size; i++) { 357 if (a[i] != b[i]) 358 break; 359 } 360 return (i); 361} 362 363static void 364ctl_be_block_compare(union ctl_io *io) 365{ 366 struct ctl_be_block_io *beio; 367 uint64_t off, res; 368 int i; 369 uint8_t info[8]; 370 371 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 372 off = 0; 373 for (i = 0; i < beio->num_segs; i++) { 374 res = cmp(beio->sg_segs[i].addr, 375 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 376 beio->sg_segs[i].len); 377 off += res; 378 if (res < beio->sg_segs[i].len) 379 break; 380 } 381 if (i < beio->num_segs) { 382 scsi_u64to8b(off, info); 383 ctl_set_sense(&io->scsiio, /*current_error*/ 1, 384 /*sense_key*/ SSD_KEY_MISCOMPARE, 385 /*asc*/ 0x1D, /*ascq*/ 0x00, 386 /*type*/ SSD_ELEM_INFO, 387 /*size*/ sizeof(info), /*data*/ &info, 388 /*type*/ SSD_ELEM_NONE); 389 } else 390 ctl_set_success(&io->scsiio); 391} 392 393static int 394ctl_be_block_move_done(union ctl_io *io) 395{ 396 struct ctl_be_block_io *beio; 397 struct ctl_be_block_lun *be_lun; 398 struct ctl_lba_len_flags *lbalen; 399#ifdef CTL_TIME_IO 400 struct bintime cur_bt; 401#endif 402 403 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 404 be_lun = beio->lun; 405 406 DPRINTF("entered\n"); 407 408#ifdef CTL_TIME_IO 409 getbinuptime(&cur_bt); 410 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 411 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 412#endif 413 io->io_hdr.num_dmas++; 414 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 415 416 /* 417 * We set status at this point for read commands, and write 418 * commands with errors. 419 */ 420 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 421 ; 422 } else if ((io->io_hdr.port_status != 0) && 423 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 424 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 425 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, 426 /*retry_count*/ io->io_hdr.port_status); 427 } else if (io->scsiio.kern_data_resid != 0 && 428 (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT && 429 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 430 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 431 ctl_set_invalid_field_ciu(&io->scsiio); 432 } else if ((io->io_hdr.port_status == 0) && 433 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 434 lbalen = ARGS(beio->io); 435 if (lbalen->flags & CTL_LLF_READ) { 436 ctl_set_success(&io->scsiio); 437 } else if (lbalen->flags & CTL_LLF_COMPARE) { 438 /* We have two data blocks ready for comparison. */ 439 ctl_be_block_compare(io); 440 } 441 } 442 443 /* 444 * If this is a read, or a write with errors, it is done. 445 */ 446 if ((beio->bio_cmd == BIO_READ) 447 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 448 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 449 ctl_complete_beio(beio); 450 return (0); 451 } 452 453 /* 454 * At this point, we have a write and the DMA completed 455 * successfully. We now have to queue it to the task queue to 456 * execute the backend I/O. That is because we do blocking 457 * memory allocations, and in the file backing case, blocking I/O. 458 * This move done routine is generally called in the SIM's 459 * interrupt context, and therefore we cannot block. 460 */ 461 mtx_lock(&be_lun->queue_lock); 462 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 463 mtx_unlock(&be_lun->queue_lock); 464 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 465 466 return (0); 467} 468 469static void 470ctl_be_block_biodone(struct bio *bio) 471{ 472 struct ctl_be_block_io *beio; 473 struct ctl_be_block_lun *be_lun; 474 union ctl_io *io; 475 int error; 476 477 beio = bio->bio_caller1; 478 be_lun = beio->lun; 479 io = beio->io; 480 481 DPRINTF("entered\n"); 482 483 error = bio->bio_error; 484 mtx_lock(&be_lun->io_lock); 485 if (error != 0 && 486 (beio->first_error == 0 || 487 bio->bio_offset < beio->first_error_offset)) { 488 beio->first_error = error; 489 beio->first_error_offset = bio->bio_offset; 490 } 491 492 beio->num_bios_done++; 493 494 /* 495 * XXX KDM will this cause WITNESS to complain? Holding a lock 496 * during the free might cause it to complain. 497 */ 498 g_destroy_bio(bio); 499 500 /* 501 * If the send complete bit isn't set, or we aren't the last I/O to 502 * complete, then we're done. 503 */ 504 if ((beio->send_complete == 0) 505 || (beio->num_bios_done < beio->num_bios_sent)) { 506 mtx_unlock(&be_lun->io_lock); 507 return; 508 } 509 510 /* 511 * At this point, we've verified that we are the last I/O to 512 * complete, so it's safe to drop the lock. 513 */ 514 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 515 beio->ds_tag_type, beio->ds_trans_type, 516 /*now*/ NULL, /*then*/&beio->ds_t0); 517 mtx_unlock(&be_lun->io_lock); 518 519 /* 520 * If there are any errors from the backing device, we fail the 521 * entire I/O with a medium error. 522 */ 523 error = beio->first_error; 524 if (error != 0) { 525 if (error == EOPNOTSUPP) { 526 ctl_set_invalid_opcode(&io->scsiio); 527 } else if (error == ENOSPC || error == EDQUOT) { 528 ctl_set_space_alloc_fail(&io->scsiio); 529 } else if (error == EROFS || error == EACCES) { 530 ctl_set_hw_write_protected(&io->scsiio); 531 } else if (beio->bio_cmd == BIO_FLUSH) { 532 /* XXX KDM is there is a better error here? */ 533 ctl_set_internal_failure(&io->scsiio, 534 /*sks_valid*/ 1, 535 /*retry_count*/ 0xbad2); 536 } else { 537 ctl_set_medium_error(&io->scsiio, 538 beio->bio_cmd == BIO_READ); 539 } 540 ctl_complete_beio(beio); 541 return; 542 } 543 544 /* 545 * If this is a write, a flush, a delete or verify, we're all done. 546 * If this is a read, we can now send the data to the user. 547 */ 548 if ((beio->bio_cmd == BIO_WRITE) 549 || (beio->bio_cmd == BIO_FLUSH) 550 || (beio->bio_cmd == BIO_DELETE) 551 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 552 ctl_set_success(&io->scsiio); 553 ctl_complete_beio(beio); 554 } else { 555 if ((ARGS(io)->flags & CTL_LLF_READ) && 556 beio->beio_cont == NULL) { 557 ctl_set_success(&io->scsiio); 558 ctl_serseq_done(io); 559 } 560#ifdef CTL_TIME_IO 561 getbinuptime(&io->io_hdr.dma_start_bt); 562#endif 563 ctl_datamove(io); 564 } 565} 566 567static void 568ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 569 struct ctl_be_block_io *beio) 570{ 571 union ctl_io *io = beio->io; 572 struct mount *mountpoint; 573 int error, lock_flags; 574 575 DPRINTF("entered\n"); 576 577 binuptime(&beio->ds_t0); 578 mtx_lock(&be_lun->io_lock); 579 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 580 mtx_unlock(&be_lun->io_lock); 581 582 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 583 584 if (MNT_SHARED_WRITES(mountpoint) || 585 ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 586 lock_flags = LK_SHARED; 587 else 588 lock_flags = LK_EXCLUSIVE; 589 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 590 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 591 curthread); 592 VOP_UNLOCK(be_lun->vn, 0); 593 594 vn_finished_write(mountpoint); 595 596 mtx_lock(&be_lun->io_lock); 597 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 598 beio->ds_tag_type, beio->ds_trans_type, 599 /*now*/ NULL, /*then*/&beio->ds_t0); 600 mtx_unlock(&be_lun->io_lock); 601 602 if (error == 0) 603 ctl_set_success(&io->scsiio); 604 else { 605 /* XXX KDM is there is a better error here? */ 606 ctl_set_internal_failure(&io->scsiio, 607 /*sks_valid*/ 1, 608 /*retry_count*/ 0xbad1); 609 } 610 611 ctl_complete_beio(beio); 612} 613 614SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t"); 615SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t"); 616SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t"); 617SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t"); 618 619static void 620ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 621 struct ctl_be_block_io *beio) 622{ 623 struct ctl_be_block_filedata *file_data; 624 union ctl_io *io; 625 struct uio xuio; 626 struct iovec *xiovec; 627 size_t s; 628 int error, flags, i; 629 630 DPRINTF("entered\n"); 631 632 file_data = &be_lun->backend.file; 633 io = beio->io; 634 flags = 0; 635 if (ARGS(io)->flags & CTL_LLF_DPO) 636 flags |= IO_DIRECT; 637 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 638 flags |= IO_SYNC; 639 640 bzero(&xuio, sizeof(xuio)); 641 if (beio->bio_cmd == BIO_READ) { 642 SDT_PROBE0(cbb, , read, file_start); 643 xuio.uio_rw = UIO_READ; 644 } else { 645 SDT_PROBE0(cbb, , write, file_start); 646 xuio.uio_rw = UIO_WRITE; 647 } 648 xuio.uio_offset = beio->io_offset; 649 xuio.uio_resid = beio->io_len; 650 xuio.uio_segflg = UIO_SYSSPACE; 651 xuio.uio_iov = beio->xiovecs; 652 xuio.uio_iovcnt = beio->num_segs; 653 xuio.uio_td = curthread; 654 655 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 656 xiovec->iov_base = beio->sg_segs[i].addr; 657 xiovec->iov_len = beio->sg_segs[i].len; 658 } 659 660 binuptime(&beio->ds_t0); 661 mtx_lock(&be_lun->io_lock); 662 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 663 mtx_unlock(&be_lun->io_lock); 664 665 if (beio->bio_cmd == BIO_READ) { 666 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 667 668 /* 669 * UFS pays attention to IO_DIRECT for reads. If the 670 * DIRECTIO option is configured into the kernel, it calls 671 * ffs_rawread(). But that only works for single-segment 672 * uios with user space addresses. In our case, with a 673 * kernel uio, it still reads into the buffer cache, but it 674 * will just try to release the buffer from the cache later 675 * on in ffs_read(). 676 * 677 * ZFS does not pay attention to IO_DIRECT for reads. 678 * 679 * UFS does not pay attention to IO_SYNC for reads. 680 * 681 * ZFS pays attention to IO_SYNC (which translates into the 682 * Solaris define FRSYNC for zfs_read()) for reads. It 683 * attempts to sync the file before reading. 684 */ 685 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 686 687 VOP_UNLOCK(be_lun->vn, 0); 688 SDT_PROBE0(cbb, , read, file_done); 689 if (error == 0 && xuio.uio_resid > 0) { 690 /* 691 * If we red less then requested (EOF), then 692 * we should clean the rest of the buffer. 693 */ 694 s = beio->io_len - xuio.uio_resid; 695 for (i = 0; i < beio->num_segs; i++) { 696 if (s >= beio->sg_segs[i].len) { 697 s -= beio->sg_segs[i].len; 698 continue; 699 } 700 bzero((uint8_t *)beio->sg_segs[i].addr + s, 701 beio->sg_segs[i].len - s); 702 s = 0; 703 } 704 } 705 } else { 706 struct mount *mountpoint; 707 int lock_flags; 708 709 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 710 711 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL) 712 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 713 lock_flags = LK_SHARED; 714 else 715 lock_flags = LK_EXCLUSIVE; 716 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 717 718 /* 719 * UFS pays attention to IO_DIRECT for writes. The write 720 * is done asynchronously. (Normally the write would just 721 * get put into cache. 722 * 723 * UFS pays attention to IO_SYNC for writes. It will 724 * attempt to write the buffer out synchronously if that 725 * flag is set. 726 * 727 * ZFS does not pay attention to IO_DIRECT for writes. 728 * 729 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 730 * for writes. It will flush the transaction from the 731 * cache before returning. 732 */ 733 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 734 VOP_UNLOCK(be_lun->vn, 0); 735 736 vn_finished_write(mountpoint); 737 SDT_PROBE0(cbb, , write, file_done); 738 } 739 740 mtx_lock(&be_lun->io_lock); 741 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 742 beio->ds_tag_type, beio->ds_trans_type, 743 /*now*/ NULL, /*then*/&beio->ds_t0); 744 mtx_unlock(&be_lun->io_lock); 745 746 /* 747 * If we got an error, set the sense data to "MEDIUM ERROR" and 748 * return the I/O to the user. 749 */ 750 if (error != 0) { 751 if (error == ENOSPC || error == EDQUOT) { 752 ctl_set_space_alloc_fail(&io->scsiio); 753 } else if (error == EROFS || error == EACCES) { 754 ctl_set_hw_write_protected(&io->scsiio); 755 } else { 756 ctl_set_medium_error(&io->scsiio, 757 beio->bio_cmd == BIO_READ); 758 } 759 ctl_complete_beio(beio); 760 return; 761 } 762 763 /* 764 * If this is a write or a verify, we're all done. 765 * If this is a read, we can now send the data to the user. 766 */ 767 if ((beio->bio_cmd == BIO_WRITE) || 768 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 769 ctl_set_success(&io->scsiio); 770 ctl_complete_beio(beio); 771 } else { 772 if ((ARGS(io)->flags & CTL_LLF_READ) && 773 beio->beio_cont == NULL) { 774 ctl_set_success(&io->scsiio); 775 ctl_serseq_done(io); 776 } 777#ifdef CTL_TIME_IO 778 getbinuptime(&io->io_hdr.dma_start_bt); 779#endif 780 ctl_datamove(io); 781 } 782} 783 784static void 785ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 786 struct ctl_be_block_io *beio) 787{ 788 union ctl_io *io = beio->io; 789 struct ctl_lba_len_flags *lbalen = ARGS(io); 790 struct scsi_get_lba_status_data *data; 791 off_t roff, off; 792 int error, status; 793 794 DPRINTF("entered\n"); 795 796 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 797 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 798 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 799 0, curthread->td_ucred, curthread); 800 if (error == 0 && off > roff) 801 status = 0; /* mapped up to off */ 802 else { 803 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 804 0, curthread->td_ucred, curthread); 805 if (error == 0 && off > roff) 806 status = 1; /* deallocated up to off */ 807 else { 808 status = 0; /* unknown up to the end */ 809 off = be_lun->size_bytes; 810 } 811 } 812 VOP_UNLOCK(be_lun->vn, 0); 813 814 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 815 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 816 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 817 lbalen->lba), data->descr[0].length); 818 data->descr[0].status = status; 819 820 ctl_complete_beio(beio); 821} 822 823static uint64_t 824ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 825{ 826 struct vattr vattr; 827 struct statfs statfs; 828 uint64_t val; 829 int error; 830 831 val = UINT64_MAX; 832 if (be_lun->vn == NULL) 833 return (val); 834 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 835 if (strcmp(attrname, "blocksused") == 0) { 836 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 837 if (error == 0) 838 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 839 } 840 if (strcmp(attrname, "blocksavail") == 0 && 841 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 842 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 843 if (error == 0) 844 val = statfs.f_bavail * statfs.f_bsize / 845 be_lun->cbe_lun.blocksize; 846 } 847 VOP_UNLOCK(be_lun->vn, 0); 848 return (val); 849} 850 851static void 852ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 853 struct ctl_be_block_io *beio) 854{ 855 union ctl_io *io; 856 struct cdevsw *csw; 857 struct cdev *dev; 858 struct uio xuio; 859 struct iovec *xiovec; 860 int error, flags, i, ref; 861 862 DPRINTF("entered\n"); 863 864 io = beio->io; 865 flags = 0; 866 if (ARGS(io)->flags & CTL_LLF_DPO) 867 flags |= IO_DIRECT; 868 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 869 flags |= IO_SYNC; 870 871 bzero(&xuio, sizeof(xuio)); 872 if (beio->bio_cmd == BIO_READ) { 873 SDT_PROBE0(cbb, , read, file_start); 874 xuio.uio_rw = UIO_READ; 875 } else { 876 SDT_PROBE0(cbb, , write, file_start); 877 xuio.uio_rw = UIO_WRITE; 878 } 879 xuio.uio_offset = beio->io_offset; 880 xuio.uio_resid = beio->io_len; 881 xuio.uio_segflg = UIO_SYSSPACE; 882 xuio.uio_iov = beio->xiovecs; 883 xuio.uio_iovcnt = beio->num_segs; 884 xuio.uio_td = curthread; 885 886 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 887 xiovec->iov_base = beio->sg_segs[i].addr; 888 xiovec->iov_len = beio->sg_segs[i].len; 889 } 890 891 binuptime(&beio->ds_t0); 892 mtx_lock(&be_lun->io_lock); 893 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 894 mtx_unlock(&be_lun->io_lock); 895 896 csw = devvn_refthread(be_lun->vn, &dev, &ref); 897 if (csw) { 898 if (beio->bio_cmd == BIO_READ) 899 error = csw->d_read(dev, &xuio, flags); 900 else 901 error = csw->d_write(dev, &xuio, flags); 902 dev_relthread(dev, ref); 903 } else 904 error = ENXIO; 905 906 if (beio->bio_cmd == BIO_READ) 907 SDT_PROBE0(cbb, , read, file_done); 908 else 909 SDT_PROBE0(cbb, , write, file_done); 910 911 mtx_lock(&be_lun->io_lock); 912 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 913 beio->ds_tag_type, beio->ds_trans_type, 914 /*now*/ NULL, /*then*/&beio->ds_t0); 915 mtx_unlock(&be_lun->io_lock); 916 917 /* 918 * If we got an error, set the sense data to "MEDIUM ERROR" and 919 * return the I/O to the user. 920 */ 921 if (error != 0) { 922 if (error == ENOSPC || error == EDQUOT) { 923 ctl_set_space_alloc_fail(&io->scsiio); 924 } else if (error == EROFS || error == EACCES) { 925 ctl_set_hw_write_protected(&io->scsiio); 926 } else { 927 ctl_set_medium_error(&io->scsiio, 928 beio->bio_cmd == BIO_READ); 929 } 930 ctl_complete_beio(beio); 931 return; 932 } 933 934 /* 935 * If this is a write or a verify, we're all done. 936 * If this is a read, we can now send the data to the user. 937 */ 938 if ((beio->bio_cmd == BIO_WRITE) || 939 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 940 ctl_set_success(&io->scsiio); 941 ctl_complete_beio(beio); 942 } else { 943 if ((ARGS(io)->flags & CTL_LLF_READ) && 944 beio->beio_cont == NULL) { 945 ctl_set_success(&io->scsiio); 946 ctl_serseq_done(io); 947 } 948#ifdef CTL_TIME_IO 949 getbinuptime(&io->io_hdr.dma_start_bt); 950#endif 951 ctl_datamove(io); 952 } 953} 954 955static void 956ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 957 struct ctl_be_block_io *beio) 958{ 959 union ctl_io *io = beio->io; 960 struct cdevsw *csw; 961 struct cdev *dev; 962 struct ctl_lba_len_flags *lbalen = ARGS(io); 963 struct scsi_get_lba_status_data *data; 964 off_t roff, off; 965 int error, ref, status; 966 967 DPRINTF("entered\n"); 968 969 csw = devvn_refthread(be_lun->vn, &dev, &ref); 970 if (csw == NULL) { 971 status = 0; /* unknown up to the end */ 972 off = be_lun->size_bytes; 973 goto done; 974 } 975 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 976 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, 977 curthread); 978 if (error == 0 && off > roff) 979 status = 0; /* mapped up to off */ 980 else { 981 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, 982 curthread); 983 if (error == 0 && off > roff) 984 status = 1; /* deallocated up to off */ 985 else { 986 status = 0; /* unknown up to the end */ 987 off = be_lun->size_bytes; 988 } 989 } 990 dev_relthread(dev, ref); 991 992done: 993 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 994 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 995 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 996 lbalen->lba), data->descr[0].length); 997 data->descr[0].status = status; 998 999 ctl_complete_beio(beio); 1000} 1001 1002static void 1003ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 1004 struct ctl_be_block_io *beio) 1005{ 1006 struct bio *bio; 1007 struct cdevsw *csw; 1008 struct cdev *dev; 1009 int ref; 1010 1011 DPRINTF("entered\n"); 1012 1013 /* This can't fail, it's a blocking allocation. */ 1014 bio = g_alloc_bio(); 1015 1016 bio->bio_cmd = BIO_FLUSH; 1017 bio->bio_offset = 0; 1018 bio->bio_data = 0; 1019 bio->bio_done = ctl_be_block_biodone; 1020 bio->bio_caller1 = beio; 1021 bio->bio_pblkno = 0; 1022 1023 /* 1024 * We don't need to acquire the LUN lock here, because we are only 1025 * sending one bio, and so there is no other context to synchronize 1026 * with. 1027 */ 1028 beio->num_bios_sent = 1; 1029 beio->send_complete = 1; 1030 1031 binuptime(&beio->ds_t0); 1032 mtx_lock(&be_lun->io_lock); 1033 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1034 mtx_unlock(&be_lun->io_lock); 1035 1036 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1037 if (csw) { 1038 bio->bio_dev = dev; 1039 csw->d_strategy(bio); 1040 dev_relthread(dev, ref); 1041 } else { 1042 bio->bio_error = ENXIO; 1043 ctl_be_block_biodone(bio); 1044 } 1045} 1046 1047static void 1048ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1049 struct ctl_be_block_io *beio, 1050 uint64_t off, uint64_t len, int last) 1051{ 1052 struct bio *bio; 1053 uint64_t maxlen; 1054 struct cdevsw *csw; 1055 struct cdev *dev; 1056 int ref; 1057 1058 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1059 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1060 while (len > 0) { 1061 bio = g_alloc_bio(); 1062 bio->bio_cmd = BIO_DELETE; 1063 bio->bio_dev = dev; 1064 bio->bio_offset = off; 1065 bio->bio_length = MIN(len, maxlen); 1066 bio->bio_data = 0; 1067 bio->bio_done = ctl_be_block_biodone; 1068 bio->bio_caller1 = beio; 1069 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1070 1071 off += bio->bio_length; 1072 len -= bio->bio_length; 1073 1074 mtx_lock(&be_lun->io_lock); 1075 beio->num_bios_sent++; 1076 if (last && len == 0) 1077 beio->send_complete = 1; 1078 mtx_unlock(&be_lun->io_lock); 1079 1080 if (csw) { 1081 csw->d_strategy(bio); 1082 } else { 1083 bio->bio_error = ENXIO; 1084 ctl_be_block_biodone(bio); 1085 } 1086 } 1087 if (csw) 1088 dev_relthread(dev, ref); 1089} 1090 1091static void 1092ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1093 struct ctl_be_block_io *beio) 1094{ 1095 union ctl_io *io; 1096 struct ctl_ptr_len_flags *ptrlen; 1097 struct scsi_unmap_desc *buf, *end; 1098 uint64_t len; 1099 1100 io = beio->io; 1101 1102 DPRINTF("entered\n"); 1103 1104 binuptime(&beio->ds_t0); 1105 mtx_lock(&be_lun->io_lock); 1106 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1107 mtx_unlock(&be_lun->io_lock); 1108 1109 if (beio->io_offset == -1) { 1110 beio->io_len = 0; 1111 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1112 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1113 end = buf + ptrlen->len / sizeof(*buf); 1114 for (; buf < end; buf++) { 1115 len = (uint64_t)scsi_4btoul(buf->length) * 1116 be_lun->cbe_lun.blocksize; 1117 beio->io_len += len; 1118 ctl_be_block_unmap_dev_range(be_lun, beio, 1119 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1120 len, (end - buf < 2) ? TRUE : FALSE); 1121 } 1122 } else 1123 ctl_be_block_unmap_dev_range(be_lun, beio, 1124 beio->io_offset, beio->io_len, TRUE); 1125} 1126 1127static void 1128ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1129 struct ctl_be_block_io *beio) 1130{ 1131 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1132 struct bio *bio; 1133 struct cdevsw *csw; 1134 struct cdev *dev; 1135 off_t cur_offset; 1136 int i, max_iosize, ref; 1137 1138 DPRINTF("entered\n"); 1139 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1140 1141 /* 1142 * We have to limit our I/O size to the maximum supported by the 1143 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1144 * set it properly, use DFLTPHYS. 1145 */ 1146 if (csw) { 1147 max_iosize = dev->si_iosize_max; 1148 if (max_iosize < PAGE_SIZE) 1149 max_iosize = DFLTPHYS; 1150 } else 1151 max_iosize = DFLTPHYS; 1152 1153 cur_offset = beio->io_offset; 1154 for (i = 0; i < beio->num_segs; i++) { 1155 size_t cur_size; 1156 uint8_t *cur_ptr; 1157 1158 cur_size = beio->sg_segs[i].len; 1159 cur_ptr = beio->sg_segs[i].addr; 1160 1161 while (cur_size > 0) { 1162 /* This can't fail, it's a blocking allocation. */ 1163 bio = g_alloc_bio(); 1164 1165 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1166 1167 bio->bio_cmd = beio->bio_cmd; 1168 bio->bio_dev = dev; 1169 bio->bio_caller1 = beio; 1170 bio->bio_length = min(cur_size, max_iosize); 1171 bio->bio_offset = cur_offset; 1172 bio->bio_data = cur_ptr; 1173 bio->bio_done = ctl_be_block_biodone; 1174 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1175 1176 cur_offset += bio->bio_length; 1177 cur_ptr += bio->bio_length; 1178 cur_size -= bio->bio_length; 1179 1180 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1181 beio->num_bios_sent++; 1182 } 1183 } 1184 binuptime(&beio->ds_t0); 1185 mtx_lock(&be_lun->io_lock); 1186 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1187 beio->send_complete = 1; 1188 mtx_unlock(&be_lun->io_lock); 1189 1190 /* 1191 * Fire off all allocated requests! 1192 */ 1193 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1194 TAILQ_REMOVE(&queue, bio, bio_queue); 1195 if (csw) 1196 csw->d_strategy(bio); 1197 else { 1198 bio->bio_error = ENXIO; 1199 ctl_be_block_biodone(bio); 1200 } 1201 } 1202 if (csw) 1203 dev_relthread(dev, ref); 1204} 1205 1206static uint64_t 1207ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1208{ 1209 struct diocgattr_arg arg; 1210 struct cdevsw *csw; 1211 struct cdev *dev; 1212 int error, ref; 1213 1214 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1215 if (csw == NULL) 1216 return (UINT64_MAX); 1217 strlcpy(arg.name, attrname, sizeof(arg.name)); 1218 arg.len = sizeof(arg.value.off); 1219 if (csw->d_ioctl) { 1220 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1221 curthread); 1222 } else 1223 error = ENODEV; 1224 dev_relthread(dev, ref); 1225 if (error != 0) 1226 return (UINT64_MAX); 1227 return (arg.value.off); 1228} 1229 1230static void 1231ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1232 union ctl_io *io) 1233{ 1234 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1235 struct ctl_be_block_io *beio; 1236 struct ctl_lba_len_flags *lbalen; 1237 1238 DPRINTF("entered\n"); 1239 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1240 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1241 1242 beio->io_len = lbalen->len * cbe_lun->blocksize; 1243 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1244 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1245 beio->bio_cmd = BIO_FLUSH; 1246 beio->ds_trans_type = DEVSTAT_NO_DATA; 1247 DPRINTF("SYNC\n"); 1248 be_lun->lun_flush(be_lun, beio); 1249} 1250 1251static void 1252ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1253{ 1254 union ctl_io *io; 1255 1256 io = beio->io; 1257 ctl_free_beio(beio); 1258 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1259 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1260 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1261 ctl_config_write_done(io); 1262 return; 1263 } 1264 1265 ctl_be_block_config_write(io); 1266} 1267 1268static void 1269ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1270 union ctl_io *io) 1271{ 1272 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1273 struct ctl_be_block_io *beio; 1274 struct ctl_lba_len_flags *lbalen; 1275 uint64_t len_left, lba; 1276 uint32_t pb, pbo, adj; 1277 int i, seglen; 1278 uint8_t *buf, *end; 1279 1280 DPRINTF("entered\n"); 1281 1282 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1283 lbalen = ARGS(beio->io); 1284 1285 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1286 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1287 ctl_free_beio(beio); 1288 ctl_set_invalid_field(&io->scsiio, 1289 /*sks_valid*/ 1, 1290 /*command*/ 1, 1291 /*field*/ 1, 1292 /*bit_valid*/ 0, 1293 /*bit*/ 0); 1294 ctl_config_write_done(io); 1295 return; 1296 } 1297 1298 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1299 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1300 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1301 beio->bio_cmd = BIO_DELETE; 1302 beio->ds_trans_type = DEVSTAT_FREE; 1303 1304 be_lun->unmap(be_lun, beio); 1305 return; 1306 } 1307 1308 beio->bio_cmd = BIO_WRITE; 1309 beio->ds_trans_type = DEVSTAT_WRITE; 1310 1311 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1312 (uintmax_t)lbalen->lba, lbalen->len); 1313 1314 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1315 if (be_lun->cbe_lun.pblockoff > 0) 1316 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1317 else 1318 pbo = 0; 1319 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1320 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1321 1322 /* 1323 * Setup the S/G entry for this chunk. 1324 */ 1325 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1326 if (pb > cbe_lun->blocksize) { 1327 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1328 seglen - pbo) % pb; 1329 if (seglen > adj) 1330 seglen -= adj; 1331 else 1332 seglen -= seglen % cbe_lun->blocksize; 1333 } else 1334 seglen -= seglen % cbe_lun->blocksize; 1335 beio->sg_segs[i].len = seglen; 1336 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1337 1338 DPRINTF("segment %d addr %p len %zd\n", i, 1339 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1340 1341 beio->num_segs++; 1342 len_left -= seglen; 1343 1344 buf = beio->sg_segs[i].addr; 1345 end = buf + seglen; 1346 for (; buf < end; buf += cbe_lun->blocksize) { 1347 if (lbalen->flags & SWS_NDOB) { 1348 memset(buf, 0, cbe_lun->blocksize); 1349 } else { 1350 memcpy(buf, io->scsiio.kern_data_ptr, 1351 cbe_lun->blocksize); 1352 } 1353 if (lbalen->flags & SWS_LBDATA) 1354 scsi_ulto4b(lbalen->lba + lba, buf); 1355 lba++; 1356 } 1357 } 1358 1359 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1360 beio->io_len = lba * cbe_lun->blocksize; 1361 1362 /* We can not do all in one run. Correct and schedule rerun. */ 1363 if (len_left > 0) { 1364 lbalen->lba += lba; 1365 lbalen->len -= lba; 1366 beio->beio_cont = ctl_be_block_cw_done_ws; 1367 } 1368 1369 be_lun->dispatch(be_lun, beio); 1370} 1371 1372static void 1373ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1374 union ctl_io *io) 1375{ 1376 struct ctl_be_block_io *beio; 1377 struct ctl_ptr_len_flags *ptrlen; 1378 1379 DPRINTF("entered\n"); 1380 1381 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1382 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1383 1384 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1385 ctl_free_beio(beio); 1386 ctl_set_invalid_field(&io->scsiio, 1387 /*sks_valid*/ 0, 1388 /*command*/ 1, 1389 /*field*/ 0, 1390 /*bit_valid*/ 0, 1391 /*bit*/ 0); 1392 ctl_config_write_done(io); 1393 return; 1394 } 1395 1396 beio->io_len = 0; 1397 beio->io_offset = -1; 1398 beio->bio_cmd = BIO_DELETE; 1399 beio->ds_trans_type = DEVSTAT_FREE; 1400 DPRINTF("UNMAP\n"); 1401 be_lun->unmap(be_lun, beio); 1402} 1403 1404static void 1405ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1406{ 1407 union ctl_io *io; 1408 1409 io = beio->io; 1410 ctl_free_beio(beio); 1411 ctl_config_read_done(io); 1412} 1413 1414static void 1415ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1416 union ctl_io *io) 1417{ 1418 struct ctl_be_block_io *beio; 1419 struct ctl_be_block_softc *softc; 1420 1421 DPRINTF("entered\n"); 1422 1423 softc = be_lun->softc; 1424 beio = ctl_alloc_beio(softc); 1425 beio->io = io; 1426 beio->lun = be_lun; 1427 beio->beio_cont = ctl_be_block_cr_done; 1428 PRIV(io)->ptr = (void *)beio; 1429 1430 switch (io->scsiio.cdb[0]) { 1431 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1432 beio->bio_cmd = -1; 1433 beio->ds_trans_type = DEVSTAT_NO_DATA; 1434 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1435 beio->io_len = 0; 1436 if (be_lun->get_lba_status) 1437 be_lun->get_lba_status(be_lun, beio); 1438 else 1439 ctl_be_block_cr_done(beio); 1440 break; 1441 default: 1442 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1443 break; 1444 } 1445} 1446 1447static void 1448ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1449{ 1450 union ctl_io *io; 1451 1452 io = beio->io; 1453 ctl_free_beio(beio); 1454 ctl_config_write_done(io); 1455} 1456 1457static void 1458ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1459 union ctl_io *io) 1460{ 1461 struct ctl_be_block_io *beio; 1462 struct ctl_be_block_softc *softc; 1463 1464 DPRINTF("entered\n"); 1465 1466 softc = be_lun->softc; 1467 beio = ctl_alloc_beio(softc); 1468 beio->io = io; 1469 beio->lun = be_lun; 1470 beio->beio_cont = ctl_be_block_cw_done; 1471 switch (io->scsiio.tag_type) { 1472 case CTL_TAG_ORDERED: 1473 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1474 break; 1475 case CTL_TAG_HEAD_OF_QUEUE: 1476 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1477 break; 1478 case CTL_TAG_UNTAGGED: 1479 case CTL_TAG_SIMPLE: 1480 case CTL_TAG_ACA: 1481 default: 1482 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1483 break; 1484 } 1485 PRIV(io)->ptr = (void *)beio; 1486 1487 switch (io->scsiio.cdb[0]) { 1488 case SYNCHRONIZE_CACHE: 1489 case SYNCHRONIZE_CACHE_16: 1490 ctl_be_block_cw_dispatch_sync(be_lun, io); 1491 break; 1492 case WRITE_SAME_10: 1493 case WRITE_SAME_16: 1494 ctl_be_block_cw_dispatch_ws(be_lun, io); 1495 break; 1496 case UNMAP: 1497 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1498 break; 1499 default: 1500 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1501 break; 1502 } 1503} 1504 1505SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t"); 1506SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t"); 1507SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t"); 1508SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t"); 1509 1510static void 1511ctl_be_block_next(struct ctl_be_block_io *beio) 1512{ 1513 struct ctl_be_block_lun *be_lun; 1514 union ctl_io *io; 1515 1516 io = beio->io; 1517 be_lun = beio->lun; 1518 ctl_free_beio(beio); 1519 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1520 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1521 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1522 ctl_data_submit_done(io); 1523 return; 1524 } 1525 1526 io->io_hdr.status &= ~CTL_STATUS_MASK; 1527 io->io_hdr.status |= CTL_STATUS_NONE; 1528 1529 mtx_lock(&be_lun->queue_lock); 1530 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1531 mtx_unlock(&be_lun->queue_lock); 1532 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1533} 1534 1535static void 1536ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1537 union ctl_io *io) 1538{ 1539 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1540 struct ctl_be_block_io *beio; 1541 struct ctl_be_block_softc *softc; 1542 struct ctl_lba_len_flags *lbalen; 1543 struct ctl_ptr_len_flags *bptrlen; 1544 uint64_t len_left, lbas; 1545 int i; 1546 1547 softc = be_lun->softc; 1548 1549 DPRINTF("entered\n"); 1550 1551 lbalen = ARGS(io); 1552 if (lbalen->flags & CTL_LLF_WRITE) { 1553 SDT_PROBE0(cbb, , write, start); 1554 } else { 1555 SDT_PROBE0(cbb, , read, start); 1556 } 1557 1558 beio = ctl_alloc_beio(softc); 1559 beio->io = io; 1560 beio->lun = be_lun; 1561 bptrlen = PRIV(io); 1562 bptrlen->ptr = (void *)beio; 1563 1564 switch (io->scsiio.tag_type) { 1565 case CTL_TAG_ORDERED: 1566 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1567 break; 1568 case CTL_TAG_HEAD_OF_QUEUE: 1569 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1570 break; 1571 case CTL_TAG_UNTAGGED: 1572 case CTL_TAG_SIMPLE: 1573 case CTL_TAG_ACA: 1574 default: 1575 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1576 break; 1577 } 1578 1579 if (lbalen->flags & CTL_LLF_WRITE) { 1580 beio->bio_cmd = BIO_WRITE; 1581 beio->ds_trans_type = DEVSTAT_WRITE; 1582 } else { 1583 beio->bio_cmd = BIO_READ; 1584 beio->ds_trans_type = DEVSTAT_READ; 1585 } 1586 1587 DPRINTF("%s at LBA %jx len %u @%ju\n", 1588 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1589 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1590 if (lbalen->flags & CTL_LLF_COMPARE) 1591 lbas = CTLBLK_HALF_IO_SIZE; 1592 else 1593 lbas = CTLBLK_MAX_IO_SIZE; 1594 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1595 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1596 beio->io_len = lbas * cbe_lun->blocksize; 1597 bptrlen->len += lbas; 1598 1599 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1600 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1601 i, CTLBLK_MAX_SEGS)); 1602 1603 /* 1604 * Setup the S/G entry for this chunk. 1605 */ 1606 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1607 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1608 1609 DPRINTF("segment %d addr %p len %zd\n", i, 1610 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1611 1612 /* Set up second segment for compare operation. */ 1613 if (lbalen->flags & CTL_LLF_COMPARE) { 1614 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1615 beio->sg_segs[i].len; 1616 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1617 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1618 } 1619 1620 beio->num_segs++; 1621 len_left -= beio->sg_segs[i].len; 1622 } 1623 if (bptrlen->len < lbalen->len) 1624 beio->beio_cont = ctl_be_block_next; 1625 io->scsiio.be_move_done = ctl_be_block_move_done; 1626 /* For compare we have separate S/G lists for read and datamove. */ 1627 if (lbalen->flags & CTL_LLF_COMPARE) 1628 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1629 else 1630 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1631 io->scsiio.kern_data_len = beio->io_len; 1632 io->scsiio.kern_sg_entries = beio->num_segs; 1633 io->io_hdr.flags |= CTL_FLAG_ALLOCATED; 1634 1635 /* 1636 * For the read case, we need to read the data into our buffers and 1637 * then we can send it back to the user. For the write case, we 1638 * need to get the data from the user first. 1639 */ 1640 if (beio->bio_cmd == BIO_READ) { 1641 SDT_PROBE0(cbb, , read, alloc_done); 1642 be_lun->dispatch(be_lun, beio); 1643 } else { 1644 SDT_PROBE0(cbb, , write, alloc_done); 1645#ifdef CTL_TIME_IO 1646 getbinuptime(&io->io_hdr.dma_start_bt); 1647#endif 1648 ctl_datamove(io); 1649 } 1650} 1651 1652static void 1653ctl_be_block_worker(void *context, int pending) 1654{ 1655 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1656 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1657 union ctl_io *io; 1658 struct ctl_be_block_io *beio; 1659 1660 DPRINTF("entered\n"); 1661 /* 1662 * Fetch and process I/Os from all queues. If we detect LUN 1663 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race, 1664 * so make response maximally opaque to not confuse initiator. 1665 */ 1666 for (;;) { 1667 mtx_lock(&be_lun->queue_lock); 1668 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1669 if (io != NULL) { 1670 DPRINTF("datamove queue\n"); 1671 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1672 ctl_io_hdr, links); 1673 mtx_unlock(&be_lun->queue_lock); 1674 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1675 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1676 ctl_set_busy(&io->scsiio); 1677 ctl_complete_beio(beio); 1678 return; 1679 } 1680 be_lun->dispatch(be_lun, beio); 1681 continue; 1682 } 1683 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1684 if (io != NULL) { 1685 DPRINTF("config write queue\n"); 1686 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1687 ctl_io_hdr, links); 1688 mtx_unlock(&be_lun->queue_lock); 1689 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1690 ctl_set_busy(&io->scsiio); 1691 ctl_config_write_done(io); 1692 return; 1693 } 1694 ctl_be_block_cw_dispatch(be_lun, io); 1695 continue; 1696 } 1697 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1698 if (io != NULL) { 1699 DPRINTF("config read queue\n"); 1700 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1701 ctl_io_hdr, links); 1702 mtx_unlock(&be_lun->queue_lock); 1703 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1704 ctl_set_busy(&io->scsiio); 1705 ctl_config_read_done(io); 1706 return; 1707 } 1708 ctl_be_block_cr_dispatch(be_lun, io); 1709 continue; 1710 } 1711 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1712 if (io != NULL) { 1713 DPRINTF("input queue\n"); 1714 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1715 ctl_io_hdr, links); 1716 mtx_unlock(&be_lun->queue_lock); 1717 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1718 ctl_set_busy(&io->scsiio); 1719 ctl_data_submit_done(io); 1720 return; 1721 } 1722 ctl_be_block_dispatch(be_lun, io); 1723 continue; 1724 } 1725 1726 /* 1727 * If we get here, there is no work left in the queues, so 1728 * just break out and let the task queue go to sleep. 1729 */ 1730 mtx_unlock(&be_lun->queue_lock); 1731 break; 1732 } 1733} 1734 1735/* 1736 * Entry point from CTL to the backend for I/O. We queue everything to a 1737 * work thread, so this just puts the I/O on a queue and wakes up the 1738 * thread. 1739 */ 1740static int 1741ctl_be_block_submit(union ctl_io *io) 1742{ 1743 struct ctl_be_block_lun *be_lun; 1744 struct ctl_be_lun *cbe_lun; 1745 1746 DPRINTF("entered\n"); 1747 1748 cbe_lun = CTL_BACKEND_LUN(io); 1749 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1750 1751 /* 1752 * Make sure we only get SCSI I/O. 1753 */ 1754 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1755 "%#x) encountered", io->io_hdr.io_type)); 1756 1757 PRIV(io)->len = 0; 1758 1759 mtx_lock(&be_lun->queue_lock); 1760 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1761 mtx_unlock(&be_lun->queue_lock); 1762 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1763 1764 return (CTL_RETVAL_COMPLETE); 1765} 1766 1767static int 1768ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1769 int flag, struct thread *td) 1770{ 1771 struct ctl_be_block_softc *softc; 1772 int error; 1773 1774 softc = &backend_block_softc; 1775 1776 error = 0; 1777 1778 switch (cmd) { 1779 case CTL_LUN_REQ: { 1780 struct ctl_lun_req *lun_req; 1781 1782 lun_req = (struct ctl_lun_req *)addr; 1783 1784 switch (lun_req->reqtype) { 1785 case CTL_LUNREQ_CREATE: 1786 error = ctl_be_block_create(softc, lun_req); 1787 break; 1788 case CTL_LUNREQ_RM: 1789 error = ctl_be_block_rm(softc, lun_req); 1790 break; 1791 case CTL_LUNREQ_MODIFY: 1792 error = ctl_be_block_modify(softc, lun_req); 1793 break; 1794 default: 1795 lun_req->status = CTL_LUN_ERROR; 1796 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1797 "invalid LUN request type %d", 1798 lun_req->reqtype); 1799 break; 1800 } 1801 break; 1802 } 1803 default: 1804 error = ENOTTY; 1805 break; 1806 } 1807 1808 return (error); 1809} 1810 1811static int 1812ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1813{ 1814 struct ctl_be_lun *cbe_lun; 1815 struct ctl_be_block_filedata *file_data; 1816 struct ctl_lun_create_params *params; 1817 char *value; 1818 struct vattr vattr; 1819 off_t ps, pss, po, pos, us, uss, uo, uos; 1820 int error; 1821 1822 cbe_lun = &be_lun->cbe_lun; 1823 file_data = &be_lun->backend.file; 1824 params = &be_lun->params; 1825 1826 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1827 be_lun->dispatch = ctl_be_block_dispatch_file; 1828 be_lun->lun_flush = ctl_be_block_flush_file; 1829 be_lun->get_lba_status = ctl_be_block_gls_file; 1830 be_lun->getattr = ctl_be_block_getattr_file; 1831 be_lun->unmap = NULL; 1832 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1833 1834 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1835 if (error != 0) { 1836 snprintf(req->error_str, sizeof(req->error_str), 1837 "error calling VOP_GETATTR() for file %s", 1838 be_lun->dev_path); 1839 return (error); 1840 } 1841 1842 file_data->cred = crhold(curthread->td_ucred); 1843 if (params->lun_size_bytes != 0) 1844 be_lun->size_bytes = params->lun_size_bytes; 1845 else 1846 be_lun->size_bytes = vattr.va_size; 1847 1848 /* 1849 * For files we can use any logical block size. Prefer 512 bytes 1850 * for compatibility reasons. If file's vattr.va_blocksize 1851 * (preferred I/O block size) is bigger and multiple to chosen 1852 * logical block size -- report it as physical block size. 1853 */ 1854 if (params->blocksize_bytes != 0) 1855 cbe_lun->blocksize = params->blocksize_bytes; 1856 else if (cbe_lun->lun_type == T_CDROM) 1857 cbe_lun->blocksize = 2048; 1858 else 1859 cbe_lun->blocksize = 512; 1860 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1861 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1862 0 : (be_lun->size_blocks - 1); 1863 1864 us = ps = vattr.va_blocksize; 1865 uo = po = 0; 1866 1867 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1868 if (value != NULL) 1869 ctl_expand_number(value, &ps); 1870 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1871 if (value != NULL) 1872 ctl_expand_number(value, &po); 1873 pss = ps / cbe_lun->blocksize; 1874 pos = po / cbe_lun->blocksize; 1875 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1876 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1877 cbe_lun->pblockexp = fls(pss) - 1; 1878 cbe_lun->pblockoff = (pss - pos) % pss; 1879 } 1880 1881 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1882 if (value != NULL) 1883 ctl_expand_number(value, &us); 1884 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 1885 if (value != NULL) 1886 ctl_expand_number(value, &uo); 1887 uss = us / cbe_lun->blocksize; 1888 uos = uo / cbe_lun->blocksize; 1889 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1890 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1891 cbe_lun->ublockexp = fls(uss) - 1; 1892 cbe_lun->ublockoff = (uss - uos) % uss; 1893 } 1894 1895 /* 1896 * Sanity check. The media size has to be at least one 1897 * sector long. 1898 */ 1899 if (be_lun->size_bytes < cbe_lun->blocksize) { 1900 error = EINVAL; 1901 snprintf(req->error_str, sizeof(req->error_str), 1902 "file %s size %ju < block size %u", be_lun->dev_path, 1903 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1904 } 1905 1906 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1907 return (error); 1908} 1909 1910static int 1911ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1912{ 1913 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1914 struct ctl_lun_create_params *params; 1915 struct cdevsw *csw; 1916 struct cdev *dev; 1917 char *value; 1918 int error, atomic, maxio, ref, unmap, tmp; 1919 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1920 1921 params = &be_lun->params; 1922 1923 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1924 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1925 if (csw == NULL) 1926 return (ENXIO); 1927 if (strcmp(csw->d_name, "zvol") == 0) { 1928 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1929 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1930 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1931 } else { 1932 be_lun->dispatch = ctl_be_block_dispatch_dev; 1933 be_lun->get_lba_status = NULL; 1934 atomic = 0; 1935 maxio = dev->si_iosize_max; 1936 if (maxio <= 0) 1937 maxio = DFLTPHYS; 1938 if (maxio > CTLBLK_MAX_IO_SIZE) 1939 maxio = CTLBLK_MAX_IO_SIZE; 1940 } 1941 be_lun->lun_flush = ctl_be_block_flush_dev; 1942 be_lun->getattr = ctl_be_block_getattr_dev; 1943 be_lun->unmap = ctl_be_block_unmap_dev; 1944 1945 if (!csw->d_ioctl) { 1946 dev_relthread(dev, ref); 1947 snprintf(req->error_str, sizeof(req->error_str), 1948 "no d_ioctl for device %s!", be_lun->dev_path); 1949 return (ENODEV); 1950 } 1951 1952 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1953 curthread); 1954 if (error) { 1955 dev_relthread(dev, ref); 1956 snprintf(req->error_str, sizeof(req->error_str), 1957 "error %d returned for DIOCGSECTORSIZE ioctl " 1958 "on %s!", error, be_lun->dev_path); 1959 return (error); 1960 } 1961 1962 /* 1963 * If the user has asked for a blocksize that is greater than the 1964 * backing device's blocksize, we can do it only if the blocksize 1965 * the user is asking for is an even multiple of the underlying 1966 * device's blocksize. 1967 */ 1968 if ((params->blocksize_bytes != 0) && 1969 (params->blocksize_bytes >= tmp)) { 1970 if (params->blocksize_bytes % tmp == 0) { 1971 cbe_lun->blocksize = params->blocksize_bytes; 1972 } else { 1973 dev_relthread(dev, ref); 1974 snprintf(req->error_str, sizeof(req->error_str), 1975 "requested blocksize %u is not an even " 1976 "multiple of backing device blocksize %u", 1977 params->blocksize_bytes, tmp); 1978 return (EINVAL); 1979 } 1980 } else if (params->blocksize_bytes != 0) { 1981 dev_relthread(dev, ref); 1982 snprintf(req->error_str, sizeof(req->error_str), 1983 "requested blocksize %u < backing device " 1984 "blocksize %u", params->blocksize_bytes, tmp); 1985 return (EINVAL); 1986 } else if (cbe_lun->lun_type == T_CDROM) 1987 cbe_lun->blocksize = MAX(tmp, 2048); 1988 else 1989 cbe_lun->blocksize = tmp; 1990 1991 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 1992 curthread); 1993 if (error) { 1994 dev_relthread(dev, ref); 1995 snprintf(req->error_str, sizeof(req->error_str), 1996 "error %d returned for DIOCGMEDIASIZE " 1997 " ioctl on %s!", error, 1998 be_lun->dev_path); 1999 return (error); 2000 } 2001 2002 if (params->lun_size_bytes != 0) { 2003 if (params->lun_size_bytes > otmp) { 2004 dev_relthread(dev, ref); 2005 snprintf(req->error_str, sizeof(req->error_str), 2006 "requested LUN size %ju > backing device " 2007 "size %ju", 2008 (uintmax_t)params->lun_size_bytes, 2009 (uintmax_t)otmp); 2010 return (EINVAL); 2011 } 2012 2013 be_lun->size_bytes = params->lun_size_bytes; 2014 } else 2015 be_lun->size_bytes = otmp; 2016 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2017 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2018 0 : (be_lun->size_blocks - 1); 2019 2020 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2021 curthread); 2022 if (error) 2023 ps = po = 0; 2024 else { 2025 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2026 FREAD, curthread); 2027 if (error) 2028 po = 0; 2029 } 2030 us = ps; 2031 uo = po; 2032 2033 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 2034 if (value != NULL) 2035 ctl_expand_number(value, &ps); 2036 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 2037 if (value != NULL) 2038 ctl_expand_number(value, &po); 2039 pss = ps / cbe_lun->blocksize; 2040 pos = po / cbe_lun->blocksize; 2041 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2042 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2043 cbe_lun->pblockexp = fls(pss) - 1; 2044 cbe_lun->pblockoff = (pss - pos) % pss; 2045 } 2046 2047 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 2048 if (value != NULL) 2049 ctl_expand_number(value, &us); 2050 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 2051 if (value != NULL) 2052 ctl_expand_number(value, &uo); 2053 uss = us / cbe_lun->blocksize; 2054 uos = uo / cbe_lun->blocksize; 2055 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2056 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2057 cbe_lun->ublockexp = fls(uss) - 1; 2058 cbe_lun->ublockoff = (uss - uos) % uss; 2059 } 2060 2061 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2062 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2063 2064 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2065 unmap = 1; 2066 } else { 2067 struct diocgattr_arg arg; 2068 2069 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2070 arg.len = sizeof(arg.value.i); 2071 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2072 curthread); 2073 unmap = (error == 0) ? arg.value.i : 0; 2074 } 2075 value = ctl_get_opt(&cbe_lun->options, "unmap"); 2076 if (value != NULL) 2077 unmap = (strcmp(value, "on") == 0); 2078 if (unmap) 2079 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2080 else 2081 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2082 2083 dev_relthread(dev, ref); 2084 return (0); 2085} 2086 2087static int 2088ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2089{ 2090 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2091 int flags; 2092 2093 if (be_lun->vn) { 2094 flags = FREAD; 2095 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2096 flags |= FWRITE; 2097 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2098 be_lun->vn = NULL; 2099 2100 switch (be_lun->dev_type) { 2101 case CTL_BE_BLOCK_DEV: 2102 break; 2103 case CTL_BE_BLOCK_FILE: 2104 if (be_lun->backend.file.cred != NULL) { 2105 crfree(be_lun->backend.file.cred); 2106 be_lun->backend.file.cred = NULL; 2107 } 2108 break; 2109 case CTL_BE_BLOCK_NONE: 2110 break; 2111 default: 2112 panic("Unexpected backend type %d", be_lun->dev_type); 2113 break; 2114 } 2115 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2116 } 2117 return (0); 2118} 2119 2120static int 2121ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2122{ 2123 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2124 struct nameidata nd; 2125 char *value; 2126 int error, flags; 2127 2128 error = 0; 2129 if (rootvnode == NULL) { 2130 snprintf(req->error_str, sizeof(req->error_str), 2131 "Root filesystem is not mounted"); 2132 return (1); 2133 } 2134 pwd_ensure_dirs(); 2135 2136 value = ctl_get_opt(&cbe_lun->options, "file"); 2137 if (value == NULL) { 2138 snprintf(req->error_str, sizeof(req->error_str), 2139 "no file argument specified"); 2140 return (1); 2141 } 2142 free(be_lun->dev_path, M_CTLBLK); 2143 be_lun->dev_path = strdup(value, M_CTLBLK); 2144 2145 flags = FREAD; 2146 value = ctl_get_opt(&cbe_lun->options, "readonly"); 2147 if (value != NULL) { 2148 if (strcmp(value, "on") != 0) 2149 flags |= FWRITE; 2150 } else if (cbe_lun->lun_type == T_DIRECT) 2151 flags |= FWRITE; 2152 2153again: 2154 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2155 error = vn_open(&nd, &flags, 0, NULL); 2156 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2157 flags &= ~FWRITE; 2158 goto again; 2159 } 2160 if (error) { 2161 /* 2162 * This is the only reasonable guess we can make as far as 2163 * path if the user doesn't give us a fully qualified path. 2164 * If they want to specify a file, they need to specify the 2165 * full path. 2166 */ 2167 if (be_lun->dev_path[0] != '/') { 2168 char *dev_name; 2169 2170 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2171 be_lun->dev_path); 2172 free(be_lun->dev_path, M_CTLBLK); 2173 be_lun->dev_path = dev_name; 2174 goto again; 2175 } 2176 snprintf(req->error_str, sizeof(req->error_str), 2177 "error opening %s: %d", be_lun->dev_path, error); 2178 return (error); 2179 } 2180 if (flags & FWRITE) 2181 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2182 else 2183 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2184 2185 NDFREE(&nd, NDF_ONLY_PNBUF); 2186 be_lun->vn = nd.ni_vp; 2187 2188 /* We only support disks and files. */ 2189 if (vn_isdisk(be_lun->vn, &error)) { 2190 error = ctl_be_block_open_dev(be_lun, req); 2191 } else if (be_lun->vn->v_type == VREG) { 2192 error = ctl_be_block_open_file(be_lun, req); 2193 } else { 2194 error = EINVAL; 2195 snprintf(req->error_str, sizeof(req->error_str), 2196 "%s is not a disk or plain file", be_lun->dev_path); 2197 } 2198 VOP_UNLOCK(be_lun->vn, 0); 2199 2200 if (error != 0) 2201 ctl_be_block_close(be_lun); 2202 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2203 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2204 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2205 value = ctl_get_opt(&cbe_lun->options, "serseq"); 2206 if (value != NULL && strcmp(value, "on") == 0) 2207 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2208 else if (value != NULL && strcmp(value, "read") == 0) 2209 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2210 else if (value != NULL && strcmp(value, "off") == 0) 2211 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2212 return (0); 2213} 2214 2215static int 2216ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2217{ 2218 struct ctl_be_lun *cbe_lun; 2219 struct ctl_be_block_lun *be_lun; 2220 struct ctl_lun_create_params *params; 2221 char num_thread_str[16]; 2222 char tmpstr[32]; 2223 char *value; 2224 int retval, num_threads; 2225 int tmp_num_threads; 2226 2227 params = &req->reqdata.create; 2228 retval = 0; 2229 req->status = CTL_LUN_OK; 2230 2231 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2232 cbe_lun = &be_lun->cbe_lun; 2233 cbe_lun->be_lun = be_lun; 2234 be_lun->params = req->reqdata.create; 2235 be_lun->softc = softc; 2236 STAILQ_INIT(&be_lun->input_queue); 2237 STAILQ_INIT(&be_lun->config_read_queue); 2238 STAILQ_INIT(&be_lun->config_write_queue); 2239 STAILQ_INIT(&be_lun->datamove_queue); 2240 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2241 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2242 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2243 ctl_init_opts(&cbe_lun->options, 2244 req->num_be_args, req->kern_be_args); 2245 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2246 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2247 if (be_lun->lun_zone == NULL) { 2248 snprintf(req->error_str, sizeof(req->error_str), 2249 "error allocating UMA zone"); 2250 goto bailout_error; 2251 } 2252 2253 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2254 cbe_lun->lun_type = params->device_type; 2255 else 2256 cbe_lun->lun_type = T_DIRECT; 2257 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2258 cbe_lun->flags = 0; 2259 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2260 if (value != NULL) { 2261 if (strcmp(value, "primary") == 0) 2262 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2263 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2264 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2265 2266 if (cbe_lun->lun_type == T_DIRECT || 2267 cbe_lun->lun_type == T_CDROM) { 2268 be_lun->size_bytes = params->lun_size_bytes; 2269 if (params->blocksize_bytes != 0) 2270 cbe_lun->blocksize = params->blocksize_bytes; 2271 else if (cbe_lun->lun_type == T_CDROM) 2272 cbe_lun->blocksize = 2048; 2273 else 2274 cbe_lun->blocksize = 512; 2275 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2276 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2277 0 : (be_lun->size_blocks - 1); 2278 2279 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2280 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2281 retval = ctl_be_block_open(be_lun, req); 2282 if (retval != 0) { 2283 retval = 0; 2284 req->status = CTL_LUN_WARNING; 2285 } 2286 } 2287 num_threads = cbb_num_threads; 2288 } else { 2289 num_threads = 1; 2290 } 2291 2292 value = ctl_get_opt(&cbe_lun->options, "num_threads"); 2293 if (value != NULL) { 2294 tmp_num_threads = strtol(value, NULL, 0); 2295 2296 /* 2297 * We don't let the user specify less than one 2298 * thread, but hope he's clueful enough not to 2299 * specify 1000 threads. 2300 */ 2301 if (tmp_num_threads < 1) { 2302 snprintf(req->error_str, sizeof(req->error_str), 2303 "invalid number of threads %s", 2304 num_thread_str); 2305 goto bailout_error; 2306 } 2307 num_threads = tmp_num_threads; 2308 } 2309 2310 if (be_lun->vn == NULL) 2311 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2312 /* Tell the user the blocksize we ended up using */ 2313 params->lun_size_bytes = be_lun->size_bytes; 2314 params->blocksize_bytes = cbe_lun->blocksize; 2315 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2316 cbe_lun->req_lun_id = params->req_lun_id; 2317 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2318 } else 2319 cbe_lun->req_lun_id = 0; 2320 2321 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2322 cbe_lun->lun_config_status = ctl_be_block_lun_config_status; 2323 cbe_lun->be = &ctl_be_block_driver; 2324 2325 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2326 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2327 softc->num_luns); 2328 strncpy((char *)cbe_lun->serial_num, tmpstr, 2329 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2330 2331 /* Tell the user what we used for a serial number */ 2332 strncpy((char *)params->serial_num, tmpstr, 2333 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2334 } else { 2335 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2336 MIN(sizeof(cbe_lun->serial_num), 2337 sizeof(params->serial_num))); 2338 } 2339 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2340 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2341 strncpy((char *)cbe_lun->device_id, tmpstr, 2342 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2343 2344 /* Tell the user what we used for a device ID */ 2345 strncpy((char *)params->device_id, tmpstr, 2346 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2347 } else { 2348 strncpy((char *)cbe_lun->device_id, params->device_id, 2349 MIN(sizeof(cbe_lun->device_id), 2350 sizeof(params->device_id))); 2351 } 2352 2353 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2354 2355 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2356 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2357 2358 if (be_lun->io_taskqueue == NULL) { 2359 snprintf(req->error_str, sizeof(req->error_str), 2360 "unable to create taskqueue"); 2361 goto bailout_error; 2362 } 2363 2364 /* 2365 * Note that we start the same number of threads by default for 2366 * both the file case and the block device case. For the file 2367 * case, we need multiple threads to allow concurrency, because the 2368 * vnode interface is designed to be a blocking interface. For the 2369 * block device case, ZFS zvols at least will block the caller's 2370 * context in many instances, and so we need multiple threads to 2371 * overcome that problem. Other block devices don't need as many 2372 * threads, but they shouldn't cause too many problems. 2373 * 2374 * If the user wants to just have a single thread for a block 2375 * device, he can specify that when the LUN is created, or change 2376 * the tunable/sysctl to alter the default number of threads. 2377 */ 2378 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2379 /*num threads*/num_threads, 2380 /*priority*/PWAIT, 2381 /*thread name*/ 2382 "%s taskq", be_lun->lunname); 2383 2384 if (retval != 0) 2385 goto bailout_error; 2386 2387 be_lun->num_threads = num_threads; 2388 2389 mtx_lock(&softc->lock); 2390 softc->num_luns++; 2391 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2392 2393 mtx_unlock(&softc->lock); 2394 2395 retval = ctl_add_lun(&be_lun->cbe_lun); 2396 if (retval != 0) { 2397 mtx_lock(&softc->lock); 2398 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2399 links); 2400 softc->num_luns--; 2401 mtx_unlock(&softc->lock); 2402 snprintf(req->error_str, sizeof(req->error_str), 2403 "ctl_add_lun() returned error %d, see dmesg for " 2404 "details", retval); 2405 retval = 0; 2406 goto bailout_error; 2407 } 2408 2409 mtx_lock(&softc->lock); 2410 2411 /* 2412 * Tell the config_status routine that we're waiting so it won't 2413 * clean up the LUN in the event of an error. 2414 */ 2415 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2416 2417 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2418 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2419 if (retval == EINTR) 2420 break; 2421 } 2422 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2423 2424 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2425 snprintf(req->error_str, sizeof(req->error_str), 2426 "LUN configuration error, see dmesg for details"); 2427 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2428 links); 2429 softc->num_luns--; 2430 mtx_unlock(&softc->lock); 2431 goto bailout_error; 2432 } else { 2433 params->req_lun_id = cbe_lun->lun_id; 2434 } 2435 2436 mtx_unlock(&softc->lock); 2437 2438 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2439 cbe_lun->blocksize, 2440 DEVSTAT_ALL_SUPPORTED, 2441 cbe_lun->lun_type 2442 | DEVSTAT_TYPE_IF_OTHER, 2443 DEVSTAT_PRIORITY_OTHER); 2444 2445 return (retval); 2446 2447bailout_error: 2448 req->status = CTL_LUN_ERROR; 2449 2450 if (be_lun->io_taskqueue != NULL) 2451 taskqueue_free(be_lun->io_taskqueue); 2452 ctl_be_block_close(be_lun); 2453 if (be_lun->dev_path != NULL) 2454 free(be_lun->dev_path, M_CTLBLK); 2455 if (be_lun->lun_zone != NULL) 2456 uma_zdestroy(be_lun->lun_zone); 2457 ctl_free_opts(&cbe_lun->options); 2458 mtx_destroy(&be_lun->queue_lock); 2459 mtx_destroy(&be_lun->io_lock); 2460 free(be_lun, M_CTLBLK); 2461 2462 return (retval); 2463} 2464 2465static int 2466ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2467{ 2468 struct ctl_lun_rm_params *params; 2469 struct ctl_be_block_lun *be_lun; 2470 struct ctl_be_lun *cbe_lun; 2471 int retval; 2472 2473 params = &req->reqdata.rm; 2474 2475 mtx_lock(&softc->lock); 2476 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2477 if (be_lun->cbe_lun.lun_id == params->lun_id) 2478 break; 2479 } 2480 mtx_unlock(&softc->lock); 2481 if (be_lun == NULL) { 2482 snprintf(req->error_str, sizeof(req->error_str), 2483 "LUN %u is not managed by the block backend", 2484 params->lun_id); 2485 goto bailout_error; 2486 } 2487 cbe_lun = &be_lun->cbe_lun; 2488 2489 retval = ctl_disable_lun(cbe_lun); 2490 if (retval != 0) { 2491 snprintf(req->error_str, sizeof(req->error_str), 2492 "error %d returned from ctl_disable_lun() for " 2493 "LUN %d", retval, params->lun_id); 2494 goto bailout_error; 2495 } 2496 2497 if (be_lun->vn != NULL) { 2498 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2499 ctl_lun_no_media(cbe_lun); 2500 taskqueue_drain_all(be_lun->io_taskqueue); 2501 ctl_be_block_close(be_lun); 2502 } 2503 2504 retval = ctl_invalidate_lun(cbe_lun); 2505 if (retval != 0) { 2506 snprintf(req->error_str, sizeof(req->error_str), 2507 "error %d returned from ctl_invalidate_lun() for " 2508 "LUN %d", retval, params->lun_id); 2509 goto bailout_error; 2510 } 2511 2512 mtx_lock(&softc->lock); 2513 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2514 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2515 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2516 if (retval == EINTR) 2517 break; 2518 } 2519 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2520 2521 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2522 snprintf(req->error_str, sizeof(req->error_str), 2523 "interrupted waiting for LUN to be freed"); 2524 mtx_unlock(&softc->lock); 2525 goto bailout_error; 2526 } 2527 2528 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2529 2530 softc->num_luns--; 2531 mtx_unlock(&softc->lock); 2532 2533 taskqueue_drain_all(be_lun->io_taskqueue); 2534 taskqueue_free(be_lun->io_taskqueue); 2535 2536 if (be_lun->disk_stats != NULL) 2537 devstat_remove_entry(be_lun->disk_stats); 2538 2539 uma_zdestroy(be_lun->lun_zone); 2540 2541 ctl_free_opts(&cbe_lun->options); 2542 free(be_lun->dev_path, M_CTLBLK); 2543 mtx_destroy(&be_lun->queue_lock); 2544 mtx_destroy(&be_lun->io_lock); 2545 free(be_lun, M_CTLBLK); 2546 2547 req->status = CTL_LUN_OK; 2548 return (0); 2549 2550bailout_error: 2551 req->status = CTL_LUN_ERROR; 2552 return (0); 2553} 2554 2555static int 2556ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2557{ 2558 struct ctl_lun_modify_params *params; 2559 struct ctl_be_block_lun *be_lun; 2560 struct ctl_be_lun *cbe_lun; 2561 char *value; 2562 uint64_t oldsize; 2563 int error, wasprim; 2564 2565 params = &req->reqdata.modify; 2566 2567 mtx_lock(&softc->lock); 2568 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2569 if (be_lun->cbe_lun.lun_id == params->lun_id) 2570 break; 2571 } 2572 mtx_unlock(&softc->lock); 2573 if (be_lun == NULL) { 2574 snprintf(req->error_str, sizeof(req->error_str), 2575 "LUN %u is not managed by the block backend", 2576 params->lun_id); 2577 goto bailout_error; 2578 } 2579 cbe_lun = &be_lun->cbe_lun; 2580 2581 if (params->lun_size_bytes != 0) 2582 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2583 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); 2584 2585 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2586 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2587 if (value != NULL) { 2588 if (strcmp(value, "primary") == 0) 2589 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2590 else 2591 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2592 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2593 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2594 else 2595 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2596 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2597 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2598 ctl_lun_primary(cbe_lun); 2599 else 2600 ctl_lun_secondary(cbe_lun); 2601 } 2602 2603 oldsize = be_lun->size_blocks; 2604 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2605 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2606 if (be_lun->vn == NULL) 2607 error = ctl_be_block_open(be_lun, req); 2608 else if (vn_isdisk(be_lun->vn, &error)) 2609 error = ctl_be_block_open_dev(be_lun, req); 2610 else if (be_lun->vn->v_type == VREG) { 2611 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2612 error = ctl_be_block_open_file(be_lun, req); 2613 VOP_UNLOCK(be_lun->vn, 0); 2614 } else 2615 error = EINVAL; 2616 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) && 2617 be_lun->vn != NULL) { 2618 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2619 ctl_lun_has_media(cbe_lun); 2620 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 && 2621 be_lun->vn == NULL) { 2622 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2623 ctl_lun_no_media(cbe_lun); 2624 } 2625 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2626 } else { 2627 if (be_lun->vn != NULL) { 2628 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2629 ctl_lun_no_media(cbe_lun); 2630 taskqueue_drain_all(be_lun->io_taskqueue); 2631 error = ctl_be_block_close(be_lun); 2632 } else 2633 error = 0; 2634 } 2635 if (be_lun->size_blocks != oldsize) 2636 ctl_lun_capacity_changed(cbe_lun); 2637 2638 /* Tell the user the exact size we ended up using */ 2639 params->lun_size_bytes = be_lun->size_bytes; 2640 2641 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2642 return (0); 2643 2644bailout_error: 2645 req->status = CTL_LUN_ERROR; 2646 return (0); 2647} 2648 2649static void 2650ctl_be_block_lun_shutdown(void *be_lun) 2651{ 2652 struct ctl_be_block_lun *lun; 2653 struct ctl_be_block_softc *softc; 2654 2655 lun = (struct ctl_be_block_lun *)be_lun; 2656 softc = lun->softc; 2657 2658 mtx_lock(&softc->lock); 2659 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2660 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2661 wakeup(lun); 2662 mtx_unlock(&softc->lock); 2663} 2664 2665static void 2666ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2667{ 2668 struct ctl_be_block_lun *lun; 2669 struct ctl_be_block_softc *softc; 2670 2671 lun = (struct ctl_be_block_lun *)be_lun; 2672 softc = lun->softc; 2673 2674 if (status == CTL_LUN_CONFIG_OK) { 2675 mtx_lock(&softc->lock); 2676 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2677 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2678 wakeup(lun); 2679 mtx_unlock(&softc->lock); 2680 2681 /* 2682 * We successfully added the LUN, attempt to enable it. 2683 */ 2684 if (ctl_enable_lun(&lun->cbe_lun) != 0) { 2685 printf("%s: ctl_enable_lun() failed!\n", __func__); 2686 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { 2687 printf("%s: ctl_invalidate_lun() failed!\n", 2688 __func__); 2689 } 2690 } 2691 2692 return; 2693 } 2694 2695 2696 mtx_lock(&softc->lock); 2697 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2698 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2699 wakeup(lun); 2700 mtx_unlock(&softc->lock); 2701} 2702 2703 2704static int 2705ctl_be_block_config_write(union ctl_io *io) 2706{ 2707 struct ctl_be_block_lun *be_lun; 2708 struct ctl_be_lun *cbe_lun; 2709 int retval; 2710 2711 DPRINTF("entered\n"); 2712 2713 cbe_lun = CTL_BACKEND_LUN(io); 2714 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2715 2716 retval = 0; 2717 switch (io->scsiio.cdb[0]) { 2718 case SYNCHRONIZE_CACHE: 2719 case SYNCHRONIZE_CACHE_16: 2720 case WRITE_SAME_10: 2721 case WRITE_SAME_16: 2722 case UNMAP: 2723 /* 2724 * The upper level CTL code will filter out any CDBs with 2725 * the immediate bit set and return the proper error. 2726 * 2727 * We don't really need to worry about what LBA range the 2728 * user asked to be synced out. When they issue a sync 2729 * cache command, we'll sync out the whole thing. 2730 */ 2731 mtx_lock(&be_lun->queue_lock); 2732 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2733 links); 2734 mtx_unlock(&be_lun->queue_lock); 2735 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2736 break; 2737 case START_STOP_UNIT: { 2738 struct scsi_start_stop_unit *cdb; 2739 struct ctl_lun_req req; 2740 2741 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2742 if ((cdb->how & SSS_PC_MASK) != 0) { 2743 ctl_set_success(&io->scsiio); 2744 ctl_config_write_done(io); 2745 break; 2746 } 2747 if (cdb->how & SSS_START) { 2748 if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) { 2749 retval = ctl_be_block_open(be_lun, &req); 2750 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2751 if (retval == 0) { 2752 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2753 ctl_lun_has_media(cbe_lun); 2754 } else { 2755 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2756 ctl_lun_no_media(cbe_lun); 2757 } 2758 } 2759 ctl_start_lun(cbe_lun); 2760 } else { 2761 ctl_stop_lun(cbe_lun); 2762 if (cdb->how & SSS_LOEJ) { 2763 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2764 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED; 2765 ctl_lun_ejected(cbe_lun); 2766 if (be_lun->vn != NULL) 2767 ctl_be_block_close(be_lun); 2768 } 2769 } 2770 2771 ctl_set_success(&io->scsiio); 2772 ctl_config_write_done(io); 2773 break; 2774 } 2775 case PREVENT_ALLOW: 2776 ctl_set_success(&io->scsiio); 2777 ctl_config_write_done(io); 2778 break; 2779 default: 2780 ctl_set_invalid_opcode(&io->scsiio); 2781 ctl_config_write_done(io); 2782 retval = CTL_RETVAL_COMPLETE; 2783 break; 2784 } 2785 2786 return (retval); 2787} 2788 2789static int 2790ctl_be_block_config_read(union ctl_io *io) 2791{ 2792 struct ctl_be_block_lun *be_lun; 2793 struct ctl_be_lun *cbe_lun; 2794 int retval = 0; 2795 2796 DPRINTF("entered\n"); 2797 2798 cbe_lun = CTL_BACKEND_LUN(io); 2799 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2800 2801 switch (io->scsiio.cdb[0]) { 2802 case SERVICE_ACTION_IN: 2803 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2804 mtx_lock(&be_lun->queue_lock); 2805 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2806 &io->io_hdr, links); 2807 mtx_unlock(&be_lun->queue_lock); 2808 taskqueue_enqueue(be_lun->io_taskqueue, 2809 &be_lun->io_task); 2810 retval = CTL_RETVAL_QUEUED; 2811 break; 2812 } 2813 ctl_set_invalid_field(&io->scsiio, 2814 /*sks_valid*/ 1, 2815 /*command*/ 1, 2816 /*field*/ 1, 2817 /*bit_valid*/ 1, 2818 /*bit*/ 4); 2819 ctl_config_read_done(io); 2820 retval = CTL_RETVAL_COMPLETE; 2821 break; 2822 default: 2823 ctl_set_invalid_opcode(&io->scsiio); 2824 ctl_config_read_done(io); 2825 retval = CTL_RETVAL_COMPLETE; 2826 break; 2827 } 2828 2829 return (retval); 2830} 2831 2832static int 2833ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2834{ 2835 struct ctl_be_block_lun *lun; 2836 int retval; 2837 2838 lun = (struct ctl_be_block_lun *)be_lun; 2839 2840 retval = sbuf_printf(sb, "\t<num_threads>"); 2841 if (retval != 0) 2842 goto bailout; 2843 retval = sbuf_printf(sb, "%d", lun->num_threads); 2844 if (retval != 0) 2845 goto bailout; 2846 retval = sbuf_printf(sb, "</num_threads>\n"); 2847 2848bailout: 2849 return (retval); 2850} 2851 2852static uint64_t 2853ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2854{ 2855 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2856 2857 if (lun->getattr == NULL) 2858 return (UINT64_MAX); 2859 return (lun->getattr(lun, attrname)); 2860} 2861 2862int 2863ctl_be_block_init(void) 2864{ 2865 struct ctl_be_block_softc *softc; 2866 int retval; 2867 2868 softc = &backend_block_softc; 2869 retval = 0; 2870 2871 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2872 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2873 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2874 STAILQ_INIT(&softc->lun_list); 2875 2876 return (retval); 2877} 2878