ctl_backend_block.c revision 361256
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 361256 2020-05-19 14:42:09Z 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#include <sys/sx.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_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 *dev_path; 153 ctl_be_block_type dev_type; 154 struct vnode *vn; 155 union ctl_be_block_bedata backend; 156 cbb_dispatch_t dispatch; 157 cbb_dispatch_t lun_flush; 158 cbb_dispatch_t unmap; 159 cbb_dispatch_t get_lba_status; 160 cbb_getattr_t getattr; 161 uma_zone_t lun_zone; 162 uint64_t size_blocks; 163 uint64_t size_bytes; 164 struct ctl_be_block_softc *softc; 165 struct devstat *disk_stats; 166 ctl_be_block_lun_flags flags; 167 SLIST_ENTRY(ctl_be_block_lun) links; 168 struct ctl_be_lun cbe_lun; 169 struct taskqueue *io_taskqueue; 170 struct task io_task; 171 int num_threads; 172 STAILQ_HEAD(, ctl_io_hdr) input_queue; 173 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 174 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 175 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 176 struct mtx_padalign io_lock; 177 struct mtx_padalign queue_lock; 178}; 179 180/* 181 * Overall softc structure for the block backend module. 182 */ 183struct ctl_be_block_softc { 184 struct sx modify_lock; 185 struct mtx lock; 186 uma_zone_t beio_zone; 187 int num_luns; 188 SLIST_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 first_error; 206 uint64_t first_error_offset; 207 struct bintime ds_t0; 208 devstat_tag_type ds_tag_type; 209 devstat_trans_flags ds_trans_type; 210 uint64_t io_len; 211 uint64_t io_offset; 212 int io_arg; 213 struct ctl_be_block_softc *softc; 214 struct ctl_be_block_lun *lun; 215 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 216}; 217 218extern struct ctl_softc *control_softc; 219 220static int cbb_num_threads = 14; 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_RWTUN, 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_lun *be_lun, 263 struct ctl_lun_req *req); 264static int ctl_be_block_create(struct ctl_be_block_softc *softc, 265 struct ctl_lun_req *req); 266static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 267 struct ctl_lun_req *req); 268static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 269 struct ctl_lun_req *req); 270static void ctl_be_block_lun_shutdown(void *be_lun); 271static int ctl_be_block_config_write(union ctl_io *io); 272static int ctl_be_block_config_read(union ctl_io *io); 273static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 274static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 275static int ctl_be_block_init(void); 276static int ctl_be_block_shutdown(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 .shutdown = ctl_be_block_shutdown, 284 .data_submit = ctl_be_block_submit, 285 .data_move_done = ctl_be_block_move_done, 286 .config_read = ctl_be_block_config_read, 287 .config_write = ctl_be_block_config_write, 288 .ioctl = ctl_be_block_ioctl, 289 .lun_info = ctl_be_block_lun_info, 290 .lun_attr = ctl_be_block_lun_attr 291}; 292 293MALLOC_DEFINE(M_CTLBLK, "ctlblock", "Memory used for CTL block backend"); 294CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 295 296static struct ctl_be_block_io * 297ctl_alloc_beio(struct ctl_be_block_softc *softc) 298{ 299 struct ctl_be_block_io *beio; 300 301 beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO); 302 beio->softc = softc; 303 return (beio); 304} 305 306static void 307ctl_free_beio(struct ctl_be_block_io *beio) 308{ 309 int duplicate_free; 310 int i; 311 312 duplicate_free = 0; 313 314 for (i = 0; i < beio->num_segs; i++) { 315 if (beio->sg_segs[i].addr == NULL) 316 duplicate_free++; 317 318 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 319 beio->sg_segs[i].addr = NULL; 320 321 /* For compare we had two equal S/G lists. */ 322 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 323 uma_zfree(beio->lun->lun_zone, 324 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 325 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 326 } 327 } 328 329 if (duplicate_free > 0) { 330 printf("%s: %d duplicate frees out of %d segments\n", __func__, 331 duplicate_free, beio->num_segs); 332 } 333 334 uma_zfree(beio->softc->beio_zone, beio); 335} 336 337static void 338ctl_complete_beio(struct ctl_be_block_io *beio) 339{ 340 union ctl_io *io = beio->io; 341 342 if (beio->beio_cont != NULL) { 343 beio->beio_cont(beio); 344 } else { 345 ctl_free_beio(beio); 346 ctl_data_submit_done(io); 347 } 348} 349 350static size_t 351cmp(uint8_t *a, uint8_t *b, size_t size) 352{ 353 size_t i; 354 355 for (i = 0; i < size; i++) { 356 if (a[i] != b[i]) 357 break; 358 } 359 return (i); 360} 361 362static void 363ctl_be_block_compare(union ctl_io *io) 364{ 365 struct ctl_be_block_io *beio; 366 uint64_t off, res; 367 int i; 368 uint8_t info[8]; 369 370 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 371 off = 0; 372 for (i = 0; i < beio->num_segs; i++) { 373 res = cmp(beio->sg_segs[i].addr, 374 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 375 beio->sg_segs[i].len); 376 off += res; 377 if (res < beio->sg_segs[i].len) 378 break; 379 } 380 if (i < beio->num_segs) { 381 scsi_u64to8b(off, info); 382 ctl_set_sense(&io->scsiio, /*current_error*/ 1, 383 /*sense_key*/ SSD_KEY_MISCOMPARE, 384 /*asc*/ 0x1D, /*ascq*/ 0x00, 385 /*type*/ SSD_ELEM_INFO, 386 /*size*/ sizeof(info), /*data*/ &info, 387 /*type*/ SSD_ELEM_NONE); 388 } else 389 ctl_set_success(&io->scsiio); 390} 391 392static int 393ctl_be_block_move_done(union ctl_io *io) 394{ 395 struct ctl_be_block_io *beio; 396 struct ctl_be_block_lun *be_lun; 397 struct ctl_lba_len_flags *lbalen; 398#ifdef CTL_TIME_IO 399 struct bintime cur_bt; 400#endif 401 402 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 403 be_lun = beio->lun; 404 405 DPRINTF("entered\n"); 406 407#ifdef CTL_TIME_IO 408 getbinuptime(&cur_bt); 409 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 410 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 411#endif 412 io->io_hdr.num_dmas++; 413 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 414 415 /* 416 * We set status at this point for read commands, and write 417 * commands with errors. 418 */ 419 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 420 ; 421 } else if ((io->io_hdr.port_status != 0) && 422 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 423 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 424 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, 425 /*retry_count*/ io->io_hdr.port_status); 426 } else if (io->scsiio.kern_data_resid != 0 && 427 (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT && 428 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 429 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 430 ctl_set_invalid_field_ciu(&io->scsiio); 431 } else if ((io->io_hdr.port_status == 0) && 432 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 433 lbalen = ARGS(beio->io); 434 if (lbalen->flags & CTL_LLF_READ) { 435 ctl_set_success(&io->scsiio); 436 } else if (lbalen->flags & CTL_LLF_COMPARE) { 437 /* We have two data blocks ready for comparison. */ 438 ctl_be_block_compare(io); 439 } 440 } 441 442 /* 443 * If this is a read, or a write with errors, it is done. 444 */ 445 if ((beio->bio_cmd == BIO_READ) 446 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 447 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 448 ctl_complete_beio(beio); 449 return (0); 450 } 451 452 /* 453 * At this point, we have a write and the DMA completed 454 * successfully. We now have to queue it to the task queue to 455 * execute the backend I/O. That is because we do blocking 456 * memory allocations, and in the file backing case, blocking I/O. 457 * This move done routine is generally called in the SIM's 458 * interrupt context, and therefore we cannot block. 459 */ 460 mtx_lock(&be_lun->queue_lock); 461 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 462 mtx_unlock(&be_lun->queue_lock); 463 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 464 465 return (0); 466} 467 468static void 469ctl_be_block_biodone(struct bio *bio) 470{ 471 struct ctl_be_block_io *beio; 472 struct ctl_be_block_lun *be_lun; 473 union ctl_io *io; 474 int error; 475 476 beio = bio->bio_caller1; 477 be_lun = beio->lun; 478 io = beio->io; 479 480 DPRINTF("entered\n"); 481 482 error = bio->bio_error; 483 mtx_lock(&be_lun->io_lock); 484 if (error != 0 && 485 (beio->first_error == 0 || 486 bio->bio_offset < beio->first_error_offset)) { 487 beio->first_error = error; 488 beio->first_error_offset = bio->bio_offset; 489 } 490 491 beio->num_bios_done++; 492 493 /* 494 * XXX KDM will this cause WITNESS to complain? Holding a lock 495 * during the free might cause it to complain. 496 */ 497 g_destroy_bio(bio); 498 499 /* 500 * If the send complete bit isn't set, or we aren't the last I/O to 501 * complete, then we're done. 502 */ 503 if ((beio->send_complete == 0) 504 || (beio->num_bios_done < beio->num_bios_sent)) { 505 mtx_unlock(&be_lun->io_lock); 506 return; 507 } 508 509 /* 510 * At this point, we've verified that we are the last I/O to 511 * complete, so it's safe to drop the lock. 512 */ 513 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 514 beio->ds_tag_type, beio->ds_trans_type, 515 /*now*/ NULL, /*then*/&beio->ds_t0); 516 mtx_unlock(&be_lun->io_lock); 517 518 /* 519 * If there are any errors from the backing device, we fail the 520 * entire I/O with a medium error. 521 */ 522 error = beio->first_error; 523 if (error != 0) { 524 if (error == EOPNOTSUPP) { 525 ctl_set_invalid_opcode(&io->scsiio); 526 } else if (error == ENOSPC || error == EDQUOT) { 527 ctl_set_space_alloc_fail(&io->scsiio); 528 } else if (error == EROFS || error == EACCES) { 529 ctl_set_hw_write_protected(&io->scsiio); 530 } else if (beio->bio_cmd == BIO_FLUSH) { 531 /* XXX KDM is there is a better error here? */ 532 ctl_set_internal_failure(&io->scsiio, 533 /*sks_valid*/ 1, 534 /*retry_count*/ 0xbad2); 535 } else { 536 ctl_set_medium_error(&io->scsiio, 537 beio->bio_cmd == BIO_READ); 538 } 539 ctl_complete_beio(beio); 540 return; 541 } 542 543 /* 544 * If this is a write, a flush, a delete or verify, we're all done. 545 * If this is a read, we can now send the data to the user. 546 */ 547 if ((beio->bio_cmd == BIO_WRITE) 548 || (beio->bio_cmd == BIO_FLUSH) 549 || (beio->bio_cmd == BIO_DELETE) 550 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 551 ctl_set_success(&io->scsiio); 552 ctl_complete_beio(beio); 553 } else { 554 if ((ARGS(io)->flags & CTL_LLF_READ) && 555 beio->beio_cont == NULL) { 556 ctl_set_success(&io->scsiio); 557 ctl_serseq_done(io); 558 } 559#ifdef CTL_TIME_IO 560 getbinuptime(&io->io_hdr.dma_start_bt); 561#endif 562 ctl_datamove(io); 563 } 564} 565 566static void 567ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 568 struct ctl_be_block_io *beio) 569{ 570 union ctl_io *io = beio->io; 571 struct mount *mountpoint; 572 int error, lock_flags; 573 574 DPRINTF("entered\n"); 575 576 binuptime(&beio->ds_t0); 577 mtx_lock(&be_lun->io_lock); 578 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 579 mtx_unlock(&be_lun->io_lock); 580 581 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 582 583 if (MNT_SHARED_WRITES(mountpoint) || 584 ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 585 lock_flags = LK_SHARED; 586 else 587 lock_flags = LK_EXCLUSIVE; 588 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 589 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 590 curthread); 591 VOP_UNLOCK(be_lun->vn, 0); 592 593 vn_finished_write(mountpoint); 594 595 mtx_lock(&be_lun->io_lock); 596 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 597 beio->ds_tag_type, beio->ds_trans_type, 598 /*now*/ NULL, /*then*/&beio->ds_t0); 599 mtx_unlock(&be_lun->io_lock); 600 601 if (error == 0) 602 ctl_set_success(&io->scsiio); 603 else { 604 /* XXX KDM is there is a better error here? */ 605 ctl_set_internal_failure(&io->scsiio, 606 /*sks_valid*/ 1, 607 /*retry_count*/ 0xbad1); 608 } 609 610 ctl_complete_beio(beio); 611} 612 613SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t"); 614SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t"); 615SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t"); 616SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t"); 617 618static void 619ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 620 struct ctl_be_block_io *beio) 621{ 622 struct ctl_be_block_filedata *file_data; 623 union ctl_io *io; 624 struct uio xuio; 625 struct iovec *xiovec; 626 size_t s; 627 int error, flags, i; 628 629 DPRINTF("entered\n"); 630 631 file_data = &be_lun->backend.file; 632 io = beio->io; 633 flags = 0; 634 if (ARGS(io)->flags & CTL_LLF_DPO) 635 flags |= IO_DIRECT; 636 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 637 flags |= IO_SYNC; 638 639 bzero(&xuio, sizeof(xuio)); 640 if (beio->bio_cmd == BIO_READ) { 641 SDT_PROBE0(cbb, , read, file_start); 642 xuio.uio_rw = UIO_READ; 643 } else { 644 SDT_PROBE0(cbb, , write, file_start); 645 xuio.uio_rw = UIO_WRITE; 646 } 647 xuio.uio_offset = beio->io_offset; 648 xuio.uio_resid = beio->io_len; 649 xuio.uio_segflg = UIO_SYSSPACE; 650 xuio.uio_iov = beio->xiovecs; 651 xuio.uio_iovcnt = beio->num_segs; 652 xuio.uio_td = curthread; 653 654 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 655 xiovec->iov_base = beio->sg_segs[i].addr; 656 xiovec->iov_len = beio->sg_segs[i].len; 657 } 658 659 binuptime(&beio->ds_t0); 660 mtx_lock(&be_lun->io_lock); 661 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 662 mtx_unlock(&be_lun->io_lock); 663 664 if (beio->bio_cmd == BIO_READ) { 665 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 666 667 /* 668 * UFS pays attention to IO_DIRECT for reads. If the 669 * DIRECTIO option is configured into the kernel, it calls 670 * ffs_rawread(). But that only works for single-segment 671 * uios with user space addresses. In our case, with a 672 * kernel uio, it still reads into the buffer cache, but it 673 * will just try to release the buffer from the cache later 674 * on in ffs_read(). 675 * 676 * ZFS does not pay attention to IO_DIRECT for reads. 677 * 678 * UFS does not pay attention to IO_SYNC for reads. 679 * 680 * ZFS pays attention to IO_SYNC (which translates into the 681 * Solaris define FRSYNC for zfs_read()) for reads. It 682 * attempts to sync the file before reading. 683 */ 684 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 685 686 VOP_UNLOCK(be_lun->vn, 0); 687 SDT_PROBE0(cbb, , read, file_done); 688 if (error == 0 && xuio.uio_resid > 0) { 689 /* 690 * If we red less then requested (EOF), then 691 * we should clean the rest of the buffer. 692 */ 693 s = beio->io_len - xuio.uio_resid; 694 for (i = 0; i < beio->num_segs; i++) { 695 if (s >= beio->sg_segs[i].len) { 696 s -= beio->sg_segs[i].len; 697 continue; 698 } 699 bzero((uint8_t *)beio->sg_segs[i].addr + s, 700 beio->sg_segs[i].len - s); 701 s = 0; 702 } 703 } 704 } else { 705 struct mount *mountpoint; 706 int lock_flags; 707 708 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 709 710 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL) 711 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 712 lock_flags = LK_SHARED; 713 else 714 lock_flags = LK_EXCLUSIVE; 715 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 716 717 /* 718 * UFS pays attention to IO_DIRECT for writes. The write 719 * is done asynchronously. (Normally the write would just 720 * get put into cache. 721 * 722 * UFS pays attention to IO_SYNC for writes. It will 723 * attempt to write the buffer out synchronously if that 724 * flag is set. 725 * 726 * ZFS does not pay attention to IO_DIRECT for writes. 727 * 728 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 729 * for writes. It will flush the transaction from the 730 * cache before returning. 731 */ 732 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 733 VOP_UNLOCK(be_lun->vn, 0); 734 735 vn_finished_write(mountpoint); 736 SDT_PROBE0(cbb, , write, file_done); 737 } 738 739 mtx_lock(&be_lun->io_lock); 740 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 741 beio->ds_tag_type, beio->ds_trans_type, 742 /*now*/ NULL, /*then*/&beio->ds_t0); 743 mtx_unlock(&be_lun->io_lock); 744 745 /* 746 * If we got an error, set the sense data to "MEDIUM ERROR" and 747 * return the I/O to the user. 748 */ 749 if (error != 0) { 750 if (error == ENOSPC || error == EDQUOT) { 751 ctl_set_space_alloc_fail(&io->scsiio); 752 } else if (error == EROFS || error == EACCES) { 753 ctl_set_hw_write_protected(&io->scsiio); 754 } else { 755 ctl_set_medium_error(&io->scsiio, 756 beio->bio_cmd == BIO_READ); 757 } 758 ctl_complete_beio(beio); 759 return; 760 } 761 762 /* 763 * If this is a write or a verify, we're all done. 764 * If this is a read, we can now send the data to the user. 765 */ 766 if ((beio->bio_cmd == BIO_WRITE) || 767 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 768 ctl_set_success(&io->scsiio); 769 ctl_complete_beio(beio); 770 } else { 771 if ((ARGS(io)->flags & CTL_LLF_READ) && 772 beio->beio_cont == NULL) { 773 ctl_set_success(&io->scsiio); 774 ctl_serseq_done(io); 775 } 776#ifdef CTL_TIME_IO 777 getbinuptime(&io->io_hdr.dma_start_bt); 778#endif 779 ctl_datamove(io); 780 } 781} 782 783static void 784ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 785 struct ctl_be_block_io *beio) 786{ 787 union ctl_io *io = beio->io; 788 struct ctl_lba_len_flags *lbalen = ARGS(io); 789 struct scsi_get_lba_status_data *data; 790 off_t roff, off; 791 int error, status; 792 793 DPRINTF("entered\n"); 794 795 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 796 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 797 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 798 0, curthread->td_ucred, curthread); 799 if (error == 0 && off > roff) 800 status = 0; /* mapped up to off */ 801 else { 802 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 803 0, curthread->td_ucred, curthread); 804 if (error == 0 && off > roff) 805 status = 1; /* deallocated up to off */ 806 else { 807 status = 0; /* unknown up to the end */ 808 off = be_lun->size_bytes; 809 } 810 } 811 VOP_UNLOCK(be_lun->vn, 0); 812 813 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 814 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 815 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 816 lbalen->lba), data->descr[0].length); 817 data->descr[0].status = status; 818 819 ctl_complete_beio(beio); 820} 821 822static uint64_t 823ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 824{ 825 struct vattr vattr; 826 struct statfs statfs; 827 uint64_t val; 828 int error; 829 830 val = UINT64_MAX; 831 if (be_lun->vn == NULL) 832 return (val); 833 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 834 if (strcmp(attrname, "blocksused") == 0) { 835 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 836 if (error == 0) 837 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 838 } 839 if (strcmp(attrname, "blocksavail") == 0 && 840 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 841 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 842 if (error == 0) 843 val = statfs.f_bavail * statfs.f_bsize / 844 be_lun->cbe_lun.blocksize; 845 } 846 VOP_UNLOCK(be_lun->vn, 0); 847 return (val); 848} 849 850static void 851ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 852 struct ctl_be_block_io *beio) 853{ 854 union ctl_io *io; 855 struct cdevsw *csw; 856 struct cdev *dev; 857 struct uio xuio; 858 struct iovec *xiovec; 859 int error, flags, i, ref; 860 861 DPRINTF("entered\n"); 862 863 io = beio->io; 864 flags = 0; 865 if (ARGS(io)->flags & CTL_LLF_DPO) 866 flags |= IO_DIRECT; 867 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 868 flags |= IO_SYNC; 869 870 bzero(&xuio, sizeof(xuio)); 871 if (beio->bio_cmd == BIO_READ) { 872 SDT_PROBE0(cbb, , read, file_start); 873 xuio.uio_rw = UIO_READ; 874 } else { 875 SDT_PROBE0(cbb, , write, file_start); 876 xuio.uio_rw = UIO_WRITE; 877 } 878 xuio.uio_offset = beio->io_offset; 879 xuio.uio_resid = beio->io_len; 880 xuio.uio_segflg = UIO_SYSSPACE; 881 xuio.uio_iov = beio->xiovecs; 882 xuio.uio_iovcnt = beio->num_segs; 883 xuio.uio_td = curthread; 884 885 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 886 xiovec->iov_base = beio->sg_segs[i].addr; 887 xiovec->iov_len = beio->sg_segs[i].len; 888 } 889 890 binuptime(&beio->ds_t0); 891 mtx_lock(&be_lun->io_lock); 892 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 893 mtx_unlock(&be_lun->io_lock); 894 895 csw = devvn_refthread(be_lun->vn, &dev, &ref); 896 if (csw) { 897 if (beio->bio_cmd == BIO_READ) 898 error = csw->d_read(dev, &xuio, flags); 899 else 900 error = csw->d_write(dev, &xuio, flags); 901 dev_relthread(dev, ref); 902 } else 903 error = ENXIO; 904 905 if (beio->bio_cmd == BIO_READ) 906 SDT_PROBE0(cbb, , read, file_done); 907 else 908 SDT_PROBE0(cbb, , write, file_done); 909 910 mtx_lock(&be_lun->io_lock); 911 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 912 beio->ds_tag_type, beio->ds_trans_type, 913 /*now*/ NULL, /*then*/&beio->ds_t0); 914 mtx_unlock(&be_lun->io_lock); 915 916 /* 917 * If we got an error, set the sense data to "MEDIUM ERROR" and 918 * return the I/O to the user. 919 */ 920 if (error != 0) { 921 if (error == ENOSPC || error == EDQUOT) { 922 ctl_set_space_alloc_fail(&io->scsiio); 923 } else if (error == EROFS || error == EACCES) { 924 ctl_set_hw_write_protected(&io->scsiio); 925 } else { 926 ctl_set_medium_error(&io->scsiio, 927 beio->bio_cmd == BIO_READ); 928 } 929 ctl_complete_beio(beio); 930 return; 931 } 932 933 /* 934 * If this is a write or a verify, we're all done. 935 * If this is a read, we can now send the data to the user. 936 */ 937 if ((beio->bio_cmd == BIO_WRITE) || 938 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 939 ctl_set_success(&io->scsiio); 940 ctl_complete_beio(beio); 941 } else { 942 if ((ARGS(io)->flags & CTL_LLF_READ) && 943 beio->beio_cont == NULL) { 944 ctl_set_success(&io->scsiio); 945 ctl_serseq_done(io); 946 } 947#ifdef CTL_TIME_IO 948 getbinuptime(&io->io_hdr.dma_start_bt); 949#endif 950 ctl_datamove(io); 951 } 952} 953 954static void 955ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 956 struct ctl_be_block_io *beio) 957{ 958 union ctl_io *io = beio->io; 959 struct cdevsw *csw; 960 struct cdev *dev; 961 struct ctl_lba_len_flags *lbalen = ARGS(io); 962 struct scsi_get_lba_status_data *data; 963 off_t roff, off; 964 int error, ref, status; 965 966 DPRINTF("entered\n"); 967 968 csw = devvn_refthread(be_lun->vn, &dev, &ref); 969 if (csw == NULL) { 970 status = 0; /* unknown up to the end */ 971 off = be_lun->size_bytes; 972 goto done; 973 } 974 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 975 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, 976 curthread); 977 if (error == 0 && off > roff) 978 status = 0; /* mapped up to off */ 979 else { 980 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, 981 curthread); 982 if (error == 0 && off > roff) 983 status = 1; /* deallocated up to off */ 984 else { 985 status = 0; /* unknown up to the end */ 986 off = be_lun->size_bytes; 987 } 988 } 989 dev_relthread(dev, ref); 990 991done: 992 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 993 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 994 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 995 lbalen->lba), data->descr[0].length); 996 data->descr[0].status = status; 997 998 ctl_complete_beio(beio); 999} 1000 1001static void 1002ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 1003 struct ctl_be_block_io *beio) 1004{ 1005 struct bio *bio; 1006 struct cdevsw *csw; 1007 struct cdev *dev; 1008 int ref; 1009 1010 DPRINTF("entered\n"); 1011 1012 /* This can't fail, it's a blocking allocation. */ 1013 bio = g_alloc_bio(); 1014 1015 bio->bio_cmd = BIO_FLUSH; 1016 bio->bio_offset = 0; 1017 bio->bio_data = 0; 1018 bio->bio_done = ctl_be_block_biodone; 1019 bio->bio_caller1 = beio; 1020 bio->bio_pblkno = 0; 1021 1022 /* 1023 * We don't need to acquire the LUN lock here, because we are only 1024 * sending one bio, and so there is no other context to synchronize 1025 * with. 1026 */ 1027 beio->num_bios_sent = 1; 1028 beio->send_complete = 1; 1029 1030 binuptime(&beio->ds_t0); 1031 mtx_lock(&be_lun->io_lock); 1032 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1033 mtx_unlock(&be_lun->io_lock); 1034 1035 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1036 if (csw) { 1037 bio->bio_dev = dev; 1038 csw->d_strategy(bio); 1039 dev_relthread(dev, ref); 1040 } else { 1041 bio->bio_error = ENXIO; 1042 ctl_be_block_biodone(bio); 1043 } 1044} 1045 1046static void 1047ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1048 struct ctl_be_block_io *beio, 1049 uint64_t off, uint64_t len, int last) 1050{ 1051 struct bio *bio; 1052 uint64_t maxlen; 1053 struct cdevsw *csw; 1054 struct cdev *dev; 1055 int ref; 1056 1057 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1058 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1059 while (len > 0) { 1060 bio = g_alloc_bio(); 1061 bio->bio_cmd = BIO_DELETE; 1062 bio->bio_dev = dev; 1063 bio->bio_offset = off; 1064 bio->bio_length = MIN(len, maxlen); 1065 bio->bio_data = 0; 1066 bio->bio_done = ctl_be_block_biodone; 1067 bio->bio_caller1 = beio; 1068 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1069 1070 off += bio->bio_length; 1071 len -= bio->bio_length; 1072 1073 mtx_lock(&be_lun->io_lock); 1074 beio->num_bios_sent++; 1075 if (last && len == 0) 1076 beio->send_complete = 1; 1077 mtx_unlock(&be_lun->io_lock); 1078 1079 if (csw) { 1080 csw->d_strategy(bio); 1081 } else { 1082 bio->bio_error = ENXIO; 1083 ctl_be_block_biodone(bio); 1084 } 1085 } 1086 if (csw) 1087 dev_relthread(dev, ref); 1088} 1089 1090static void 1091ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1092 struct ctl_be_block_io *beio) 1093{ 1094 union ctl_io *io; 1095 struct ctl_ptr_len_flags *ptrlen; 1096 struct scsi_unmap_desc *buf, *end; 1097 uint64_t len; 1098 1099 io = beio->io; 1100 1101 DPRINTF("entered\n"); 1102 1103 binuptime(&beio->ds_t0); 1104 mtx_lock(&be_lun->io_lock); 1105 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1106 mtx_unlock(&be_lun->io_lock); 1107 1108 if (beio->io_offset == -1) { 1109 beio->io_len = 0; 1110 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1111 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1112 end = buf + ptrlen->len / sizeof(*buf); 1113 for (; buf < end; buf++) { 1114 len = (uint64_t)scsi_4btoul(buf->length) * 1115 be_lun->cbe_lun.blocksize; 1116 beio->io_len += len; 1117 ctl_be_block_unmap_dev_range(be_lun, beio, 1118 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1119 len, (end - buf < 2) ? TRUE : FALSE); 1120 } 1121 } else 1122 ctl_be_block_unmap_dev_range(be_lun, beio, 1123 beio->io_offset, beio->io_len, TRUE); 1124} 1125 1126static void 1127ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1128 struct ctl_be_block_io *beio) 1129{ 1130 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1131 struct bio *bio; 1132 struct cdevsw *csw; 1133 struct cdev *dev; 1134 off_t cur_offset; 1135 int i, max_iosize, ref; 1136 1137 DPRINTF("entered\n"); 1138 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1139 1140 /* 1141 * We have to limit our I/O size to the maximum supported by the 1142 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1143 * set it properly, use DFLTPHYS. 1144 */ 1145 if (csw) { 1146 max_iosize = dev->si_iosize_max; 1147 if (max_iosize < PAGE_SIZE) 1148 max_iosize = DFLTPHYS; 1149 } else 1150 max_iosize = DFLTPHYS; 1151 1152 cur_offset = beio->io_offset; 1153 for (i = 0; i < beio->num_segs; i++) { 1154 size_t cur_size; 1155 uint8_t *cur_ptr; 1156 1157 cur_size = beio->sg_segs[i].len; 1158 cur_ptr = beio->sg_segs[i].addr; 1159 1160 while (cur_size > 0) { 1161 /* This can't fail, it's a blocking allocation. */ 1162 bio = g_alloc_bio(); 1163 1164 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1165 1166 bio->bio_cmd = beio->bio_cmd; 1167 bio->bio_dev = dev; 1168 bio->bio_caller1 = beio; 1169 bio->bio_length = min(cur_size, max_iosize); 1170 bio->bio_offset = cur_offset; 1171 bio->bio_data = cur_ptr; 1172 bio->bio_done = ctl_be_block_biodone; 1173 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1174 1175 cur_offset += bio->bio_length; 1176 cur_ptr += bio->bio_length; 1177 cur_size -= bio->bio_length; 1178 1179 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1180 beio->num_bios_sent++; 1181 } 1182 } 1183 binuptime(&beio->ds_t0); 1184 mtx_lock(&be_lun->io_lock); 1185 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1186 beio->send_complete = 1; 1187 mtx_unlock(&be_lun->io_lock); 1188 1189 /* 1190 * Fire off all allocated requests! 1191 */ 1192 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1193 TAILQ_REMOVE(&queue, bio, bio_queue); 1194 if (csw) 1195 csw->d_strategy(bio); 1196 else { 1197 bio->bio_error = ENXIO; 1198 ctl_be_block_biodone(bio); 1199 } 1200 } 1201 if (csw) 1202 dev_relthread(dev, ref); 1203} 1204 1205static uint64_t 1206ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1207{ 1208 struct diocgattr_arg arg; 1209 struct cdevsw *csw; 1210 struct cdev *dev; 1211 int error, ref; 1212 1213 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1214 if (csw == NULL) 1215 return (UINT64_MAX); 1216 strlcpy(arg.name, attrname, sizeof(arg.name)); 1217 arg.len = sizeof(arg.value.off); 1218 if (csw->d_ioctl) { 1219 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1220 curthread); 1221 } else 1222 error = ENODEV; 1223 dev_relthread(dev, ref); 1224 if (error != 0) 1225 return (UINT64_MAX); 1226 return (arg.value.off); 1227} 1228 1229static void 1230ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1231 union ctl_io *io) 1232{ 1233 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1234 struct ctl_be_block_io *beio; 1235 struct ctl_lba_len_flags *lbalen; 1236 1237 DPRINTF("entered\n"); 1238 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1239 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1240 1241 beio->io_len = lbalen->len * cbe_lun->blocksize; 1242 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1243 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1244 beio->bio_cmd = BIO_FLUSH; 1245 beio->ds_trans_type = DEVSTAT_NO_DATA; 1246 DPRINTF("SYNC\n"); 1247 be_lun->lun_flush(be_lun, beio); 1248} 1249 1250static void 1251ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1252{ 1253 union ctl_io *io; 1254 1255 io = beio->io; 1256 ctl_free_beio(beio); 1257 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1258 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1259 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1260 ctl_config_write_done(io); 1261 return; 1262 } 1263 1264 ctl_be_block_config_write(io); 1265} 1266 1267static void 1268ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1269 union ctl_io *io) 1270{ 1271 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1272 struct ctl_be_block_io *beio; 1273 struct ctl_lba_len_flags *lbalen; 1274 uint64_t len_left, lba; 1275 uint32_t pb, pbo, adj; 1276 int i, seglen; 1277 uint8_t *buf, *end; 1278 1279 DPRINTF("entered\n"); 1280 1281 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1282 lbalen = ARGS(beio->io); 1283 1284 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1285 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1286 ctl_free_beio(beio); 1287 ctl_set_invalid_field(&io->scsiio, 1288 /*sks_valid*/ 1, 1289 /*command*/ 1, 1290 /*field*/ 1, 1291 /*bit_valid*/ 0, 1292 /*bit*/ 0); 1293 ctl_config_write_done(io); 1294 return; 1295 } 1296 1297 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1298 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1299 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1300 beio->bio_cmd = BIO_DELETE; 1301 beio->ds_trans_type = DEVSTAT_FREE; 1302 1303 be_lun->unmap(be_lun, beio); 1304 return; 1305 } 1306 1307 beio->bio_cmd = BIO_WRITE; 1308 beio->ds_trans_type = DEVSTAT_WRITE; 1309 1310 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1311 (uintmax_t)lbalen->lba, lbalen->len); 1312 1313 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1314 if (be_lun->cbe_lun.pblockoff > 0) 1315 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1316 else 1317 pbo = 0; 1318 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1319 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1320 1321 /* 1322 * Setup the S/G entry for this chunk. 1323 */ 1324 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1325 if (pb > cbe_lun->blocksize) { 1326 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1327 seglen - pbo) % pb; 1328 if (seglen > adj) 1329 seglen -= adj; 1330 else 1331 seglen -= seglen % cbe_lun->blocksize; 1332 } else 1333 seglen -= seglen % cbe_lun->blocksize; 1334 beio->sg_segs[i].len = seglen; 1335 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1336 1337 DPRINTF("segment %d addr %p len %zd\n", i, 1338 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1339 1340 beio->num_segs++; 1341 len_left -= seglen; 1342 1343 buf = beio->sg_segs[i].addr; 1344 end = buf + seglen; 1345 for (; buf < end; buf += cbe_lun->blocksize) { 1346 if (lbalen->flags & SWS_NDOB) { 1347 memset(buf, 0, cbe_lun->blocksize); 1348 } else { 1349 memcpy(buf, io->scsiio.kern_data_ptr, 1350 cbe_lun->blocksize); 1351 } 1352 if (lbalen->flags & SWS_LBDATA) 1353 scsi_ulto4b(lbalen->lba + lba, buf); 1354 lba++; 1355 } 1356 } 1357 1358 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1359 beio->io_len = lba * cbe_lun->blocksize; 1360 1361 /* We can not do all in one run. Correct and schedule rerun. */ 1362 if (len_left > 0) { 1363 lbalen->lba += lba; 1364 lbalen->len -= lba; 1365 beio->beio_cont = ctl_be_block_cw_done_ws; 1366 } 1367 1368 be_lun->dispatch(be_lun, beio); 1369} 1370 1371static void 1372ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1373 union ctl_io *io) 1374{ 1375 struct ctl_be_block_io *beio; 1376 struct ctl_ptr_len_flags *ptrlen; 1377 1378 DPRINTF("entered\n"); 1379 1380 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1381 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1382 1383 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1384 ctl_free_beio(beio); 1385 ctl_set_invalid_field(&io->scsiio, 1386 /*sks_valid*/ 0, 1387 /*command*/ 1, 1388 /*field*/ 0, 1389 /*bit_valid*/ 0, 1390 /*bit*/ 0); 1391 ctl_config_write_done(io); 1392 return; 1393 } 1394 1395 beio->io_len = 0; 1396 beio->io_offset = -1; 1397 beio->bio_cmd = BIO_DELETE; 1398 beio->ds_trans_type = DEVSTAT_FREE; 1399 DPRINTF("UNMAP\n"); 1400 be_lun->unmap(be_lun, beio); 1401} 1402 1403static void 1404ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1405{ 1406 union ctl_io *io; 1407 1408 io = beio->io; 1409 ctl_free_beio(beio); 1410 ctl_config_read_done(io); 1411} 1412 1413static void 1414ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1415 union ctl_io *io) 1416{ 1417 struct ctl_be_block_io *beio; 1418 struct ctl_be_block_softc *softc; 1419 1420 DPRINTF("entered\n"); 1421 1422 softc = be_lun->softc; 1423 beio = ctl_alloc_beio(softc); 1424 beio->io = io; 1425 beio->lun = be_lun; 1426 beio->beio_cont = ctl_be_block_cr_done; 1427 PRIV(io)->ptr = (void *)beio; 1428 1429 switch (io->scsiio.cdb[0]) { 1430 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1431 beio->bio_cmd = -1; 1432 beio->ds_trans_type = DEVSTAT_NO_DATA; 1433 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1434 beio->io_len = 0; 1435 if (be_lun->get_lba_status) 1436 be_lun->get_lba_status(be_lun, beio); 1437 else 1438 ctl_be_block_cr_done(beio); 1439 break; 1440 default: 1441 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1442 break; 1443 } 1444} 1445 1446static void 1447ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1448{ 1449 union ctl_io *io; 1450 1451 io = beio->io; 1452 ctl_free_beio(beio); 1453 ctl_config_write_done(io); 1454} 1455 1456static void 1457ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1458 union ctl_io *io) 1459{ 1460 struct ctl_be_block_io *beio; 1461 struct ctl_be_block_softc *softc; 1462 1463 DPRINTF("entered\n"); 1464 1465 softc = be_lun->softc; 1466 beio = ctl_alloc_beio(softc); 1467 beio->io = io; 1468 beio->lun = be_lun; 1469 beio->beio_cont = ctl_be_block_cw_done; 1470 switch (io->scsiio.tag_type) { 1471 case CTL_TAG_ORDERED: 1472 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1473 break; 1474 case CTL_TAG_HEAD_OF_QUEUE: 1475 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1476 break; 1477 case CTL_TAG_UNTAGGED: 1478 case CTL_TAG_SIMPLE: 1479 case CTL_TAG_ACA: 1480 default: 1481 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1482 break; 1483 } 1484 PRIV(io)->ptr = (void *)beio; 1485 1486 switch (io->scsiio.cdb[0]) { 1487 case SYNCHRONIZE_CACHE: 1488 case SYNCHRONIZE_CACHE_16: 1489 ctl_be_block_cw_dispatch_sync(be_lun, io); 1490 break; 1491 case WRITE_SAME_10: 1492 case WRITE_SAME_16: 1493 ctl_be_block_cw_dispatch_ws(be_lun, io); 1494 break; 1495 case UNMAP: 1496 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1497 break; 1498 default: 1499 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1500 break; 1501 } 1502} 1503 1504SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t"); 1505SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t"); 1506SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t"); 1507SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t"); 1508 1509static void 1510ctl_be_block_next(struct ctl_be_block_io *beio) 1511{ 1512 struct ctl_be_block_lun *be_lun; 1513 union ctl_io *io; 1514 1515 io = beio->io; 1516 be_lun = beio->lun; 1517 ctl_free_beio(beio); 1518 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1519 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1520 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1521 ctl_data_submit_done(io); 1522 return; 1523 } 1524 1525 io->io_hdr.status &= ~CTL_STATUS_MASK; 1526 io->io_hdr.status |= CTL_STATUS_NONE; 1527 1528 mtx_lock(&be_lun->queue_lock); 1529 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1530 mtx_unlock(&be_lun->queue_lock); 1531 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1532} 1533 1534static void 1535ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1536 union ctl_io *io) 1537{ 1538 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1539 struct ctl_be_block_io *beio; 1540 struct ctl_be_block_softc *softc; 1541 struct ctl_lba_len_flags *lbalen; 1542 struct ctl_ptr_len_flags *bptrlen; 1543 uint64_t len_left, lbas; 1544 int i; 1545 1546 softc = be_lun->softc; 1547 1548 DPRINTF("entered\n"); 1549 1550 lbalen = ARGS(io); 1551 if (lbalen->flags & CTL_LLF_WRITE) { 1552 SDT_PROBE0(cbb, , write, start); 1553 } else { 1554 SDT_PROBE0(cbb, , read, start); 1555 } 1556 1557 beio = ctl_alloc_beio(softc); 1558 beio->io = io; 1559 beio->lun = be_lun; 1560 bptrlen = PRIV(io); 1561 bptrlen->ptr = (void *)beio; 1562 1563 switch (io->scsiio.tag_type) { 1564 case CTL_TAG_ORDERED: 1565 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1566 break; 1567 case CTL_TAG_HEAD_OF_QUEUE: 1568 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1569 break; 1570 case CTL_TAG_UNTAGGED: 1571 case CTL_TAG_SIMPLE: 1572 case CTL_TAG_ACA: 1573 default: 1574 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1575 break; 1576 } 1577 1578 if (lbalen->flags & CTL_LLF_WRITE) { 1579 beio->bio_cmd = BIO_WRITE; 1580 beio->ds_trans_type = DEVSTAT_WRITE; 1581 } else { 1582 beio->bio_cmd = BIO_READ; 1583 beio->ds_trans_type = DEVSTAT_READ; 1584 } 1585 1586 DPRINTF("%s at LBA %jx len %u @%ju\n", 1587 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1588 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1589 if (lbalen->flags & CTL_LLF_COMPARE) 1590 lbas = CTLBLK_HALF_IO_SIZE; 1591 else 1592 lbas = CTLBLK_MAX_IO_SIZE; 1593 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1594 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1595 beio->io_len = lbas * cbe_lun->blocksize; 1596 bptrlen->len += lbas; 1597 1598 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1599 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1600 i, CTLBLK_MAX_SEGS)); 1601 1602 /* 1603 * Setup the S/G entry for this chunk. 1604 */ 1605 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1606 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1607 1608 DPRINTF("segment %d addr %p len %zd\n", i, 1609 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1610 1611 /* Set up second segment for compare operation. */ 1612 if (lbalen->flags & CTL_LLF_COMPARE) { 1613 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1614 beio->sg_segs[i].len; 1615 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1616 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1617 } 1618 1619 beio->num_segs++; 1620 len_left -= beio->sg_segs[i].len; 1621 } 1622 if (bptrlen->len < lbalen->len) 1623 beio->beio_cont = ctl_be_block_next; 1624 io->scsiio.be_move_done = ctl_be_block_move_done; 1625 /* For compare we have separate S/G lists for read and datamove. */ 1626 if (lbalen->flags & CTL_LLF_COMPARE) 1627 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1628 else 1629 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1630 io->scsiio.kern_data_len = beio->io_len; 1631 io->scsiio.kern_sg_entries = beio->num_segs; 1632 io->io_hdr.flags |= CTL_FLAG_ALLOCATED; 1633 1634 /* 1635 * For the read case, we need to read the data into our buffers and 1636 * then we can send it back to the user. For the write case, we 1637 * need to get the data from the user first. 1638 */ 1639 if (beio->bio_cmd == BIO_READ) { 1640 SDT_PROBE0(cbb, , read, alloc_done); 1641 be_lun->dispatch(be_lun, beio); 1642 } else { 1643 SDT_PROBE0(cbb, , write, alloc_done); 1644#ifdef CTL_TIME_IO 1645 getbinuptime(&io->io_hdr.dma_start_bt); 1646#endif 1647 ctl_datamove(io); 1648 } 1649} 1650 1651static void 1652ctl_be_block_worker(void *context, int pending) 1653{ 1654 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1655 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1656 union ctl_io *io; 1657 struct ctl_be_block_io *beio; 1658 1659 DPRINTF("entered\n"); 1660 /* 1661 * Fetch and process I/Os from all queues. If we detect LUN 1662 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race, 1663 * so make response maximally opaque to not confuse initiator. 1664 */ 1665 for (;;) { 1666 mtx_lock(&be_lun->queue_lock); 1667 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1668 if (io != NULL) { 1669 DPRINTF("datamove queue\n"); 1670 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1671 ctl_io_hdr, links); 1672 mtx_unlock(&be_lun->queue_lock); 1673 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1674 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1675 ctl_set_busy(&io->scsiio); 1676 ctl_complete_beio(beio); 1677 return; 1678 } 1679 be_lun->dispatch(be_lun, beio); 1680 continue; 1681 } 1682 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1683 if (io != NULL) { 1684 DPRINTF("config write queue\n"); 1685 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1686 ctl_io_hdr, links); 1687 mtx_unlock(&be_lun->queue_lock); 1688 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1689 ctl_set_busy(&io->scsiio); 1690 ctl_config_write_done(io); 1691 return; 1692 } 1693 ctl_be_block_cw_dispatch(be_lun, io); 1694 continue; 1695 } 1696 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1697 if (io != NULL) { 1698 DPRINTF("config read queue\n"); 1699 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1700 ctl_io_hdr, links); 1701 mtx_unlock(&be_lun->queue_lock); 1702 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1703 ctl_set_busy(&io->scsiio); 1704 ctl_config_read_done(io); 1705 return; 1706 } 1707 ctl_be_block_cr_dispatch(be_lun, io); 1708 continue; 1709 } 1710 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1711 if (io != NULL) { 1712 DPRINTF("input queue\n"); 1713 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1714 ctl_io_hdr, links); 1715 mtx_unlock(&be_lun->queue_lock); 1716 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1717 ctl_set_busy(&io->scsiio); 1718 ctl_data_submit_done(io); 1719 return; 1720 } 1721 ctl_be_block_dispatch(be_lun, io); 1722 continue; 1723 } 1724 1725 /* 1726 * If we get here, there is no work left in the queues, so 1727 * just break out and let the task queue go to sleep. 1728 */ 1729 mtx_unlock(&be_lun->queue_lock); 1730 break; 1731 } 1732} 1733 1734/* 1735 * Entry point from CTL to the backend for I/O. We queue everything to a 1736 * work thread, so this just puts the I/O on a queue and wakes up the 1737 * thread. 1738 */ 1739static int 1740ctl_be_block_submit(union ctl_io *io) 1741{ 1742 struct ctl_be_block_lun *be_lun; 1743 struct ctl_be_lun *cbe_lun; 1744 1745 DPRINTF("entered\n"); 1746 1747 cbe_lun = CTL_BACKEND_LUN(io); 1748 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1749 1750 /* 1751 * Make sure we only get SCSI I/O. 1752 */ 1753 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1754 "%#x) encountered", io->io_hdr.io_type)); 1755 1756 PRIV(io)->len = 0; 1757 1758 mtx_lock(&be_lun->queue_lock); 1759 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1760 mtx_unlock(&be_lun->queue_lock); 1761 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1762 1763 return (CTL_RETVAL_COMPLETE); 1764} 1765 1766static int 1767ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1768 int flag, struct thread *td) 1769{ 1770 struct ctl_be_block_softc *softc = &backend_block_softc; 1771 int error; 1772 1773 error = 0; 1774 switch (cmd) { 1775 case CTL_LUN_REQ: { 1776 struct ctl_lun_req *lun_req; 1777 1778 lun_req = (struct ctl_lun_req *)addr; 1779 1780 switch (lun_req->reqtype) { 1781 case CTL_LUNREQ_CREATE: 1782 error = ctl_be_block_create(softc, lun_req); 1783 break; 1784 case CTL_LUNREQ_RM: 1785 error = ctl_be_block_rm(softc, lun_req); 1786 break; 1787 case CTL_LUNREQ_MODIFY: 1788 error = ctl_be_block_modify(softc, lun_req); 1789 break; 1790 default: 1791 lun_req->status = CTL_LUN_ERROR; 1792 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1793 "invalid LUN request type %d", 1794 lun_req->reqtype); 1795 break; 1796 } 1797 break; 1798 } 1799 default: 1800 error = ENOTTY; 1801 break; 1802 } 1803 1804 return (error); 1805} 1806 1807static int 1808ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1809{ 1810 struct ctl_be_lun *cbe_lun; 1811 struct ctl_be_block_filedata *file_data; 1812 struct ctl_lun_create_params *params; 1813 char *value; 1814 struct vattr vattr; 1815 off_t ps, pss, po, pos, us, uss, uo, uos; 1816 int error; 1817 1818 cbe_lun = &be_lun->cbe_lun; 1819 file_data = &be_lun->backend.file; 1820 params = &be_lun->params; 1821 1822 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1823 be_lun->dispatch = ctl_be_block_dispatch_file; 1824 be_lun->lun_flush = ctl_be_block_flush_file; 1825 be_lun->get_lba_status = ctl_be_block_gls_file; 1826 be_lun->getattr = ctl_be_block_getattr_file; 1827 be_lun->unmap = NULL; 1828 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1829 1830 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1831 if (error != 0) { 1832 snprintf(req->error_str, sizeof(req->error_str), 1833 "error calling VOP_GETATTR() for file %s", 1834 be_lun->dev_path); 1835 return (error); 1836 } 1837 1838 file_data->cred = crhold(curthread->td_ucred); 1839 if (params->lun_size_bytes != 0) 1840 be_lun->size_bytes = params->lun_size_bytes; 1841 else 1842 be_lun->size_bytes = vattr.va_size; 1843 1844 /* 1845 * For files we can use any logical block size. Prefer 512 bytes 1846 * for compatibility reasons. If file's vattr.va_blocksize 1847 * (preferred I/O block size) is bigger and multiple to chosen 1848 * logical block size -- report it as physical block size. 1849 */ 1850 if (params->blocksize_bytes != 0) 1851 cbe_lun->blocksize = params->blocksize_bytes; 1852 else if (cbe_lun->lun_type == T_CDROM) 1853 cbe_lun->blocksize = 2048; 1854 else 1855 cbe_lun->blocksize = 512; 1856 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1857 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1858 0 : (be_lun->size_blocks - 1); 1859 1860 us = ps = vattr.va_blocksize; 1861 uo = po = 0; 1862 1863 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1864 if (value != NULL) 1865 ctl_expand_number(value, &ps); 1866 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1867 if (value != NULL) 1868 ctl_expand_number(value, &po); 1869 pss = ps / cbe_lun->blocksize; 1870 pos = po / cbe_lun->blocksize; 1871 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1872 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1873 cbe_lun->pblockexp = fls(pss) - 1; 1874 cbe_lun->pblockoff = (pss - pos) % pss; 1875 } 1876 1877 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1878 if (value != NULL) 1879 ctl_expand_number(value, &us); 1880 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 1881 if (value != NULL) 1882 ctl_expand_number(value, &uo); 1883 uss = us / cbe_lun->blocksize; 1884 uos = uo / cbe_lun->blocksize; 1885 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1886 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1887 cbe_lun->ublockexp = fls(uss) - 1; 1888 cbe_lun->ublockoff = (uss - uos) % uss; 1889 } 1890 1891 /* 1892 * Sanity check. The media size has to be at least one 1893 * sector long. 1894 */ 1895 if (be_lun->size_bytes < cbe_lun->blocksize) { 1896 error = EINVAL; 1897 snprintf(req->error_str, sizeof(req->error_str), 1898 "file %s size %ju < block size %u", be_lun->dev_path, 1899 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1900 } 1901 1902 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1903 return (error); 1904} 1905 1906static int 1907ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1908{ 1909 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1910 struct ctl_lun_create_params *params; 1911 struct cdevsw *csw; 1912 struct cdev *dev; 1913 char *value; 1914 int error, atomic, maxio, ref, unmap, tmp; 1915 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1916 1917 params = &be_lun->params; 1918 1919 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1920 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1921 if (csw == NULL) 1922 return (ENXIO); 1923 if (strcmp(csw->d_name, "zvol") == 0) { 1924 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1925 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1926 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1927 } else { 1928 be_lun->dispatch = ctl_be_block_dispatch_dev; 1929 be_lun->get_lba_status = NULL; 1930 atomic = 0; 1931 maxio = dev->si_iosize_max; 1932 if (maxio <= 0) 1933 maxio = DFLTPHYS; 1934 if (maxio > CTLBLK_MAX_IO_SIZE) 1935 maxio = CTLBLK_MAX_IO_SIZE; 1936 } 1937 be_lun->lun_flush = ctl_be_block_flush_dev; 1938 be_lun->getattr = ctl_be_block_getattr_dev; 1939 be_lun->unmap = ctl_be_block_unmap_dev; 1940 1941 if (!csw->d_ioctl) { 1942 dev_relthread(dev, ref); 1943 snprintf(req->error_str, sizeof(req->error_str), 1944 "no d_ioctl for device %s!", be_lun->dev_path); 1945 return (ENODEV); 1946 } 1947 1948 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1949 curthread); 1950 if (error) { 1951 dev_relthread(dev, ref); 1952 snprintf(req->error_str, sizeof(req->error_str), 1953 "error %d returned for DIOCGSECTORSIZE ioctl " 1954 "on %s!", error, be_lun->dev_path); 1955 return (error); 1956 } 1957 1958 /* 1959 * If the user has asked for a blocksize that is greater than the 1960 * backing device's blocksize, we can do it only if the blocksize 1961 * the user is asking for is an even multiple of the underlying 1962 * device's blocksize. 1963 */ 1964 if ((params->blocksize_bytes != 0) && 1965 (params->blocksize_bytes >= tmp)) { 1966 if (params->blocksize_bytes % tmp == 0) { 1967 cbe_lun->blocksize = params->blocksize_bytes; 1968 } else { 1969 dev_relthread(dev, ref); 1970 snprintf(req->error_str, sizeof(req->error_str), 1971 "requested blocksize %u is not an even " 1972 "multiple of backing device blocksize %u", 1973 params->blocksize_bytes, tmp); 1974 return (EINVAL); 1975 } 1976 } else if (params->blocksize_bytes != 0) { 1977 dev_relthread(dev, ref); 1978 snprintf(req->error_str, sizeof(req->error_str), 1979 "requested blocksize %u < backing device " 1980 "blocksize %u", params->blocksize_bytes, tmp); 1981 return (EINVAL); 1982 } else if (cbe_lun->lun_type == T_CDROM) 1983 cbe_lun->blocksize = MAX(tmp, 2048); 1984 else 1985 cbe_lun->blocksize = tmp; 1986 1987 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 1988 curthread); 1989 if (error) { 1990 dev_relthread(dev, ref); 1991 snprintf(req->error_str, sizeof(req->error_str), 1992 "error %d returned for DIOCGMEDIASIZE " 1993 " ioctl on %s!", error, 1994 be_lun->dev_path); 1995 return (error); 1996 } 1997 1998 if (params->lun_size_bytes != 0) { 1999 if (params->lun_size_bytes > otmp) { 2000 dev_relthread(dev, ref); 2001 snprintf(req->error_str, sizeof(req->error_str), 2002 "requested LUN size %ju > backing device " 2003 "size %ju", 2004 (uintmax_t)params->lun_size_bytes, 2005 (uintmax_t)otmp); 2006 return (EINVAL); 2007 } 2008 2009 be_lun->size_bytes = params->lun_size_bytes; 2010 } else 2011 be_lun->size_bytes = otmp; 2012 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2013 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2014 0 : (be_lun->size_blocks - 1); 2015 2016 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2017 curthread); 2018 if (error) 2019 ps = po = 0; 2020 else { 2021 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2022 FREAD, curthread); 2023 if (error) 2024 po = 0; 2025 } 2026 us = ps; 2027 uo = po; 2028 2029 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 2030 if (value != NULL) 2031 ctl_expand_number(value, &ps); 2032 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 2033 if (value != NULL) 2034 ctl_expand_number(value, &po); 2035 pss = ps / cbe_lun->blocksize; 2036 pos = po / cbe_lun->blocksize; 2037 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2038 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2039 cbe_lun->pblockexp = fls(pss) - 1; 2040 cbe_lun->pblockoff = (pss - pos) % pss; 2041 } 2042 2043 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 2044 if (value != NULL) 2045 ctl_expand_number(value, &us); 2046 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 2047 if (value != NULL) 2048 ctl_expand_number(value, &uo); 2049 uss = us / cbe_lun->blocksize; 2050 uos = uo / cbe_lun->blocksize; 2051 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2052 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2053 cbe_lun->ublockexp = fls(uss) - 1; 2054 cbe_lun->ublockoff = (uss - uos) % uss; 2055 } 2056 2057 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2058 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2059 2060 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2061 unmap = 1; 2062 } else { 2063 struct diocgattr_arg arg; 2064 2065 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2066 arg.len = sizeof(arg.value.i); 2067 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2068 curthread); 2069 unmap = (error == 0) ? arg.value.i : 0; 2070 } 2071 value = ctl_get_opt(&cbe_lun->options, "unmap"); 2072 if (value != NULL) 2073 unmap = (strcmp(value, "on") == 0); 2074 if (unmap) 2075 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2076 else 2077 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2078 2079 dev_relthread(dev, ref); 2080 return (0); 2081} 2082 2083static int 2084ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2085{ 2086 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2087 int flags; 2088 2089 if (be_lun->vn) { 2090 flags = FREAD; 2091 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2092 flags |= FWRITE; 2093 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2094 be_lun->vn = NULL; 2095 2096 switch (be_lun->dev_type) { 2097 case CTL_BE_BLOCK_DEV: 2098 break; 2099 case CTL_BE_BLOCK_FILE: 2100 if (be_lun->backend.file.cred != NULL) { 2101 crfree(be_lun->backend.file.cred); 2102 be_lun->backend.file.cred = NULL; 2103 } 2104 break; 2105 case CTL_BE_BLOCK_NONE: 2106 break; 2107 default: 2108 panic("Unexpected backend type %d", be_lun->dev_type); 2109 break; 2110 } 2111 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2112 } 2113 return (0); 2114} 2115 2116static int 2117ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2118{ 2119 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2120 struct nameidata nd; 2121 char *value; 2122 int error, flags; 2123 2124 error = 0; 2125 if (rootvnode == NULL) { 2126 snprintf(req->error_str, sizeof(req->error_str), 2127 "Root filesystem is not mounted"); 2128 return (1); 2129 } 2130 pwd_ensure_dirs(); 2131 2132 value = ctl_get_opt(&cbe_lun->options, "file"); 2133 if (value == NULL) { 2134 snprintf(req->error_str, sizeof(req->error_str), 2135 "no file argument specified"); 2136 return (1); 2137 } 2138 free(be_lun->dev_path, M_CTLBLK); 2139 be_lun->dev_path = strdup(value, M_CTLBLK); 2140 2141 flags = FREAD; 2142 value = ctl_get_opt(&cbe_lun->options, "readonly"); 2143 if (value != NULL) { 2144 if (strcmp(value, "on") != 0) 2145 flags |= FWRITE; 2146 } else if (cbe_lun->lun_type == T_DIRECT) 2147 flags |= FWRITE; 2148 2149again: 2150 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2151 error = vn_open(&nd, &flags, 0, NULL); 2152 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2153 flags &= ~FWRITE; 2154 goto again; 2155 } 2156 if (error) { 2157 /* 2158 * This is the only reasonable guess we can make as far as 2159 * path if the user doesn't give us a fully qualified path. 2160 * If they want to specify a file, they need to specify the 2161 * full path. 2162 */ 2163 if (be_lun->dev_path[0] != '/') { 2164 char *dev_name; 2165 2166 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2167 be_lun->dev_path); 2168 free(be_lun->dev_path, M_CTLBLK); 2169 be_lun->dev_path = dev_name; 2170 goto again; 2171 } 2172 snprintf(req->error_str, sizeof(req->error_str), 2173 "error opening %s: %d", be_lun->dev_path, error); 2174 return (error); 2175 } 2176 if (flags & FWRITE) 2177 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2178 else 2179 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2180 2181 NDFREE(&nd, NDF_ONLY_PNBUF); 2182 be_lun->vn = nd.ni_vp; 2183 2184 /* We only support disks and files. */ 2185 if (vn_isdisk(be_lun->vn, &error)) { 2186 error = ctl_be_block_open_dev(be_lun, req); 2187 } else if (be_lun->vn->v_type == VREG) { 2188 error = ctl_be_block_open_file(be_lun, req); 2189 } else { 2190 error = EINVAL; 2191 snprintf(req->error_str, sizeof(req->error_str), 2192 "%s is not a disk or plain file", be_lun->dev_path); 2193 } 2194 VOP_UNLOCK(be_lun->vn, 0); 2195 2196 if (error != 0) 2197 ctl_be_block_close(be_lun); 2198 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2199 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2200 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2201 value = ctl_get_opt(&cbe_lun->options, "serseq"); 2202 if (value != NULL && strcmp(value, "on") == 0) 2203 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2204 else if (value != NULL && strcmp(value, "read") == 0) 2205 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2206 else if (value != NULL && strcmp(value, "off") == 0) 2207 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2208 return (0); 2209} 2210 2211static int 2212ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2213{ 2214 struct ctl_be_lun *cbe_lun; 2215 struct ctl_be_block_lun *be_lun; 2216 struct ctl_lun_create_params *params; 2217 char num_thread_str[16]; 2218 char tmpstr[32]; 2219 char *value; 2220 int retval, num_threads; 2221 int tmp_num_threads; 2222 2223 params = &req->reqdata.create; 2224 retval = 0; 2225 req->status = CTL_LUN_OK; 2226 2227 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2228 cbe_lun = &be_lun->cbe_lun; 2229 cbe_lun->be_lun = be_lun; 2230 be_lun->params = req->reqdata.create; 2231 be_lun->softc = softc; 2232 STAILQ_INIT(&be_lun->input_queue); 2233 STAILQ_INIT(&be_lun->config_read_queue); 2234 STAILQ_INIT(&be_lun->config_write_queue); 2235 STAILQ_INIT(&be_lun->datamove_queue); 2236 mtx_init(&be_lun->io_lock, "ctlblock io", NULL, MTX_DEF); 2237 mtx_init(&be_lun->queue_lock, "ctlblock queue", NULL, MTX_DEF); 2238 ctl_init_opts(&cbe_lun->options, 2239 req->num_be_args, req->kern_be_args); 2240 be_lun->lun_zone = uma_zcreate("ctlblock", CTLBLK_MAX_SEG, 2241 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2242 if (be_lun->lun_zone == NULL) { 2243 snprintf(req->error_str, sizeof(req->error_str), 2244 "error allocating UMA zone"); 2245 goto bailout_error; 2246 } 2247 2248 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2249 cbe_lun->lun_type = params->device_type; 2250 else 2251 cbe_lun->lun_type = T_DIRECT; 2252 be_lun->flags = 0; 2253 cbe_lun->flags = 0; 2254 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2255 if (value != NULL) { 2256 if (strcmp(value, "primary") == 0) 2257 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2258 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2259 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2260 2261 if (cbe_lun->lun_type == T_DIRECT || 2262 cbe_lun->lun_type == T_CDROM) { 2263 be_lun->size_bytes = params->lun_size_bytes; 2264 if (params->blocksize_bytes != 0) 2265 cbe_lun->blocksize = params->blocksize_bytes; 2266 else if (cbe_lun->lun_type == T_CDROM) 2267 cbe_lun->blocksize = 2048; 2268 else 2269 cbe_lun->blocksize = 512; 2270 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2271 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2272 0 : (be_lun->size_blocks - 1); 2273 2274 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2275 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2276 retval = ctl_be_block_open(be_lun, req); 2277 if (retval != 0) { 2278 retval = 0; 2279 req->status = CTL_LUN_WARNING; 2280 } 2281 } 2282 num_threads = cbb_num_threads; 2283 } else { 2284 num_threads = 1; 2285 } 2286 2287 value = ctl_get_opt(&cbe_lun->options, "num_threads"); 2288 if (value != NULL) { 2289 tmp_num_threads = strtol(value, NULL, 0); 2290 2291 /* 2292 * We don't let the user specify less than one 2293 * thread, but hope he's clueful enough not to 2294 * specify 1000 threads. 2295 */ 2296 if (tmp_num_threads < 1) { 2297 snprintf(req->error_str, sizeof(req->error_str), 2298 "invalid number of threads %s", 2299 num_thread_str); 2300 goto bailout_error; 2301 } 2302 num_threads = tmp_num_threads; 2303 } 2304 2305 if (be_lun->vn == NULL) 2306 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2307 /* Tell the user the blocksize we ended up using */ 2308 params->lun_size_bytes = be_lun->size_bytes; 2309 params->blocksize_bytes = cbe_lun->blocksize; 2310 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2311 cbe_lun->req_lun_id = params->req_lun_id; 2312 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2313 } else 2314 cbe_lun->req_lun_id = 0; 2315 2316 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2317 cbe_lun->be = &ctl_be_block_driver; 2318 2319 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2320 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2321 softc->num_luns); 2322 strncpy((char *)cbe_lun->serial_num, tmpstr, 2323 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2324 2325 /* Tell the user what we used for a serial number */ 2326 strncpy((char *)params->serial_num, tmpstr, 2327 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2328 } else { 2329 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2330 MIN(sizeof(cbe_lun->serial_num), 2331 sizeof(params->serial_num))); 2332 } 2333 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2334 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2335 strncpy((char *)cbe_lun->device_id, tmpstr, 2336 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2337 2338 /* Tell the user what we used for a device ID */ 2339 strncpy((char *)params->device_id, tmpstr, 2340 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2341 } else { 2342 strncpy((char *)cbe_lun->device_id, params->device_id, 2343 MIN(sizeof(cbe_lun->device_id), 2344 sizeof(params->device_id))); 2345 } 2346 2347 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2348 2349 be_lun->io_taskqueue = taskqueue_create("ctlblocktq", M_WAITOK, 2350 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2351 2352 if (be_lun->io_taskqueue == NULL) { 2353 snprintf(req->error_str, sizeof(req->error_str), 2354 "unable to create taskqueue"); 2355 goto bailout_error; 2356 } 2357 2358 /* 2359 * Note that we start the same number of threads by default for 2360 * both the file case and the block device case. For the file 2361 * case, we need multiple threads to allow concurrency, because the 2362 * vnode interface is designed to be a blocking interface. For the 2363 * block device case, ZFS zvols at least will block the caller's 2364 * context in many instances, and so we need multiple threads to 2365 * overcome that problem. Other block devices don't need as many 2366 * threads, but they shouldn't cause too many problems. 2367 * 2368 * If the user wants to just have a single thread for a block 2369 * device, he can specify that when the LUN is created, or change 2370 * the tunable/sysctl to alter the default number of threads. 2371 */ 2372 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2373 /*num threads*/num_threads, 2374 /*priority*/PUSER, 2375 /*thread name*/"block"); 2376 2377 if (retval != 0) 2378 goto bailout_error; 2379 2380 be_lun->num_threads = num_threads; 2381 2382 retval = ctl_add_lun(&be_lun->cbe_lun); 2383 if (retval != 0) { 2384 snprintf(req->error_str, sizeof(req->error_str), 2385 "ctl_add_lun() returned error %d, see dmesg for " 2386 "details", retval); 2387 retval = 0; 2388 goto bailout_error; 2389 } 2390 2391 be_lun->disk_stats = devstat_new_entry("cbb", cbe_lun->lun_id, 2392 cbe_lun->blocksize, 2393 DEVSTAT_ALL_SUPPORTED, 2394 cbe_lun->lun_type 2395 | DEVSTAT_TYPE_IF_OTHER, 2396 DEVSTAT_PRIORITY_OTHER); 2397 2398 mtx_lock(&softc->lock); 2399 softc->num_luns++; 2400 SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links); 2401 mtx_unlock(&softc->lock); 2402 2403 params->req_lun_id = cbe_lun->lun_id; 2404 2405 return (retval); 2406 2407bailout_error: 2408 req->status = CTL_LUN_ERROR; 2409 2410 if (be_lun->io_taskqueue != NULL) 2411 taskqueue_free(be_lun->io_taskqueue); 2412 ctl_be_block_close(be_lun); 2413 if (be_lun->dev_path != NULL) 2414 free(be_lun->dev_path, M_CTLBLK); 2415 if (be_lun->lun_zone != NULL) 2416 uma_zdestroy(be_lun->lun_zone); 2417 ctl_free_opts(&cbe_lun->options); 2418 mtx_destroy(&be_lun->queue_lock); 2419 mtx_destroy(&be_lun->io_lock); 2420 free(be_lun, M_CTLBLK); 2421 2422 return (retval); 2423} 2424 2425static int 2426ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2427{ 2428 struct ctl_lun_rm_params *params; 2429 struct ctl_be_block_lun *be_lun; 2430 struct ctl_be_lun *cbe_lun; 2431 int retval; 2432 2433 params = &req->reqdata.rm; 2434 2435 sx_xlock(&softc->modify_lock); 2436 mtx_lock(&softc->lock); 2437 SLIST_FOREACH(be_lun, &softc->lun_list, links) { 2438 if (be_lun->cbe_lun.lun_id == params->lun_id) { 2439 SLIST_REMOVE(&softc->lun_list, be_lun, 2440 ctl_be_block_lun, links); 2441 softc->num_luns--; 2442 break; 2443 } 2444 } 2445 mtx_unlock(&softc->lock); 2446 sx_xunlock(&softc->modify_lock); 2447 if (be_lun == NULL) { 2448 snprintf(req->error_str, sizeof(req->error_str), 2449 "LUN %u is not managed by the block backend", 2450 params->lun_id); 2451 goto bailout_error; 2452 } 2453 cbe_lun = &be_lun->cbe_lun; 2454 2455 if (be_lun->vn != NULL) { 2456 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2457 ctl_lun_no_media(cbe_lun); 2458 taskqueue_drain_all(be_lun->io_taskqueue); 2459 ctl_be_block_close(be_lun); 2460 } 2461 2462 mtx_lock(&softc->lock); 2463 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2464 mtx_unlock(&softc->lock); 2465 2466 retval = ctl_remove_lun(cbe_lun); 2467 if (retval != 0) { 2468 snprintf(req->error_str, sizeof(req->error_str), 2469 "error %d returned from ctl_remove_lun() for " 2470 "LUN %d", retval, params->lun_id); 2471 mtx_lock(&softc->lock); 2472 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2473 mtx_unlock(&softc->lock); 2474 goto bailout_error; 2475 } 2476 2477 mtx_lock(&softc->lock); 2478 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2479 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblockrm", 0); 2480 if (retval == EINTR) 2481 break; 2482 } 2483 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2484 if (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2485 mtx_unlock(&softc->lock); 2486 free(be_lun, M_CTLBLK); 2487 } else { 2488 mtx_unlock(&softc->lock); 2489 return (EINTR); 2490 } 2491 2492 req->status = CTL_LUN_OK; 2493 return (0); 2494 2495bailout_error: 2496 req->status = CTL_LUN_ERROR; 2497 return (0); 2498} 2499 2500static int 2501ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2502{ 2503 struct ctl_lun_modify_params *params; 2504 struct ctl_be_block_lun *be_lun; 2505 struct ctl_be_lun *cbe_lun; 2506 char *value; 2507 uint64_t oldsize; 2508 int error, wasprim; 2509 2510 params = &req->reqdata.modify; 2511 2512 sx_xlock(&softc->modify_lock); 2513 mtx_lock(&softc->lock); 2514 SLIST_FOREACH(be_lun, &softc->lun_list, links) { 2515 if (be_lun->cbe_lun.lun_id == params->lun_id) 2516 break; 2517 } 2518 mtx_unlock(&softc->lock); 2519 if (be_lun == NULL) { 2520 snprintf(req->error_str, sizeof(req->error_str), 2521 "LUN %u is not managed by the block backend", 2522 params->lun_id); 2523 goto bailout_error; 2524 } 2525 cbe_lun = &be_lun->cbe_lun; 2526 2527 if (params->lun_size_bytes != 0) 2528 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2529 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); 2530 2531 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2532 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2533 if (value != NULL) { 2534 if (strcmp(value, "primary") == 0) 2535 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2536 else 2537 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2538 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2539 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2540 else 2541 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2542 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2543 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2544 ctl_lun_primary(cbe_lun); 2545 else 2546 ctl_lun_secondary(cbe_lun); 2547 } 2548 2549 oldsize = be_lun->size_blocks; 2550 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2551 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2552 if (be_lun->vn == NULL) 2553 error = ctl_be_block_open(be_lun, req); 2554 else if (vn_isdisk(be_lun->vn, &error)) 2555 error = ctl_be_block_open_dev(be_lun, req); 2556 else if (be_lun->vn->v_type == VREG) { 2557 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2558 error = ctl_be_block_open_file(be_lun, req); 2559 VOP_UNLOCK(be_lun->vn, 0); 2560 } else 2561 error = EINVAL; 2562 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) && 2563 be_lun->vn != NULL) { 2564 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2565 ctl_lun_has_media(cbe_lun); 2566 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 && 2567 be_lun->vn == NULL) { 2568 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2569 ctl_lun_no_media(cbe_lun); 2570 } 2571 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2572 } else { 2573 if (be_lun->vn != NULL) { 2574 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2575 ctl_lun_no_media(cbe_lun); 2576 taskqueue_drain_all(be_lun->io_taskqueue); 2577 error = ctl_be_block_close(be_lun); 2578 } else 2579 error = 0; 2580 } 2581 if (be_lun->size_blocks != oldsize) 2582 ctl_lun_capacity_changed(cbe_lun); 2583 2584 /* Tell the user the exact size we ended up using */ 2585 params->lun_size_bytes = be_lun->size_bytes; 2586 2587 sx_xunlock(&softc->modify_lock); 2588 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2589 return (0); 2590 2591bailout_error: 2592 sx_xunlock(&softc->modify_lock); 2593 req->status = CTL_LUN_ERROR; 2594 return (0); 2595} 2596 2597static void 2598ctl_be_block_lun_shutdown(void *lun) 2599{ 2600 struct ctl_be_block_lun *be_lun = lun; 2601 struct ctl_be_block_softc *softc = be_lun->softc; 2602 2603 taskqueue_drain_all(be_lun->io_taskqueue); 2604 taskqueue_free(be_lun->io_taskqueue); 2605 if (be_lun->disk_stats != NULL) 2606 devstat_remove_entry(be_lun->disk_stats); 2607 uma_zdestroy(be_lun->lun_zone); 2608 ctl_free_opts(&be_lun->cbe_lun.options); 2609 free(be_lun->dev_path, M_CTLBLK); 2610 mtx_destroy(&be_lun->queue_lock); 2611 mtx_destroy(&be_lun->io_lock); 2612 2613 mtx_lock(&softc->lock); 2614 be_lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2615 if (be_lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2616 wakeup(be_lun); 2617 else 2618 free(be_lun, M_CTLBLK); 2619 mtx_unlock(&softc->lock); 2620} 2621 2622static int 2623ctl_be_block_config_write(union ctl_io *io) 2624{ 2625 struct ctl_be_block_lun *be_lun; 2626 struct ctl_be_lun *cbe_lun; 2627 int retval; 2628 2629 DPRINTF("entered\n"); 2630 2631 cbe_lun = CTL_BACKEND_LUN(io); 2632 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2633 2634 retval = 0; 2635 switch (io->scsiio.cdb[0]) { 2636 case SYNCHRONIZE_CACHE: 2637 case SYNCHRONIZE_CACHE_16: 2638 case WRITE_SAME_10: 2639 case WRITE_SAME_16: 2640 case UNMAP: 2641 /* 2642 * The upper level CTL code will filter out any CDBs with 2643 * the immediate bit set and return the proper error. 2644 * 2645 * We don't really need to worry about what LBA range the 2646 * user asked to be synced out. When they issue a sync 2647 * cache command, we'll sync out the whole thing. 2648 */ 2649 mtx_lock(&be_lun->queue_lock); 2650 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2651 links); 2652 mtx_unlock(&be_lun->queue_lock); 2653 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2654 break; 2655 case START_STOP_UNIT: { 2656 struct scsi_start_stop_unit *cdb; 2657 struct ctl_lun_req req; 2658 2659 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2660 if ((cdb->how & SSS_PC_MASK) != 0) { 2661 ctl_set_success(&io->scsiio); 2662 ctl_config_write_done(io); 2663 break; 2664 } 2665 if (cdb->how & SSS_START) { 2666 if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) { 2667 retval = ctl_be_block_open(be_lun, &req); 2668 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2669 if (retval == 0) { 2670 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2671 ctl_lun_has_media(cbe_lun); 2672 } else { 2673 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2674 ctl_lun_no_media(cbe_lun); 2675 } 2676 } 2677 ctl_start_lun(cbe_lun); 2678 } else { 2679 ctl_stop_lun(cbe_lun); 2680 if (cdb->how & SSS_LOEJ) { 2681 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2682 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED; 2683 ctl_lun_ejected(cbe_lun); 2684 if (be_lun->vn != NULL) 2685 ctl_be_block_close(be_lun); 2686 } 2687 } 2688 2689 ctl_set_success(&io->scsiio); 2690 ctl_config_write_done(io); 2691 break; 2692 } 2693 case PREVENT_ALLOW: 2694 ctl_set_success(&io->scsiio); 2695 ctl_config_write_done(io); 2696 break; 2697 default: 2698 ctl_set_invalid_opcode(&io->scsiio); 2699 ctl_config_write_done(io); 2700 retval = CTL_RETVAL_COMPLETE; 2701 break; 2702 } 2703 2704 return (retval); 2705} 2706 2707static int 2708ctl_be_block_config_read(union ctl_io *io) 2709{ 2710 struct ctl_be_block_lun *be_lun; 2711 struct ctl_be_lun *cbe_lun; 2712 int retval = 0; 2713 2714 DPRINTF("entered\n"); 2715 2716 cbe_lun = CTL_BACKEND_LUN(io); 2717 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2718 2719 switch (io->scsiio.cdb[0]) { 2720 case SERVICE_ACTION_IN: 2721 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2722 mtx_lock(&be_lun->queue_lock); 2723 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2724 &io->io_hdr, links); 2725 mtx_unlock(&be_lun->queue_lock); 2726 taskqueue_enqueue(be_lun->io_taskqueue, 2727 &be_lun->io_task); 2728 retval = CTL_RETVAL_QUEUED; 2729 break; 2730 } 2731 ctl_set_invalid_field(&io->scsiio, 2732 /*sks_valid*/ 1, 2733 /*command*/ 1, 2734 /*field*/ 1, 2735 /*bit_valid*/ 1, 2736 /*bit*/ 4); 2737 ctl_config_read_done(io); 2738 retval = CTL_RETVAL_COMPLETE; 2739 break; 2740 default: 2741 ctl_set_invalid_opcode(&io->scsiio); 2742 ctl_config_read_done(io); 2743 retval = CTL_RETVAL_COMPLETE; 2744 break; 2745 } 2746 2747 return (retval); 2748} 2749 2750static int 2751ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2752{ 2753 struct ctl_be_block_lun *lun; 2754 int retval; 2755 2756 lun = (struct ctl_be_block_lun *)be_lun; 2757 2758 retval = sbuf_printf(sb, "\t<num_threads>"); 2759 if (retval != 0) 2760 goto bailout; 2761 retval = sbuf_printf(sb, "%d", lun->num_threads); 2762 if (retval != 0) 2763 goto bailout; 2764 retval = sbuf_printf(sb, "</num_threads>\n"); 2765 2766bailout: 2767 return (retval); 2768} 2769 2770static uint64_t 2771ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2772{ 2773 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2774 2775 if (lun->getattr == NULL) 2776 return (UINT64_MAX); 2777 return (lun->getattr(lun, attrname)); 2778} 2779 2780static int 2781ctl_be_block_init(void) 2782{ 2783 struct ctl_be_block_softc *softc = &backend_block_softc; 2784 2785 sx_init(&softc->modify_lock, "ctlblock modify"); 2786 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2787 softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2788 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2789 SLIST_INIT(&softc->lun_list); 2790 return (0); 2791} 2792 2793 2794static int 2795ctl_be_block_shutdown(void) 2796{ 2797 struct ctl_be_block_softc *softc = &backend_block_softc; 2798 struct ctl_be_block_lun *lun; 2799 2800 mtx_lock(&softc->lock); 2801 while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) { 2802 SLIST_REMOVE_HEAD(&softc->lun_list, links); 2803 softc->num_luns--; 2804 /* 2805 * Drop our lock here. Since ctl_remove_lun() can call 2806 * back into us, this could potentially lead to a recursive 2807 * lock of the same mutex, which would cause a hang. 2808 */ 2809 mtx_unlock(&softc->lock); 2810 ctl_remove_lun(&lun->cbe_lun); 2811 mtx_lock(&softc->lock); 2812 } 2813 mtx_unlock(&softc->lock); 2814 uma_zdestroy(softc->beio_zone); 2815 mtx_destroy(&softc->lock); 2816 sx_destroy(&softc->modify_lock); 2817 return (0); 2818} 2819