ctl_backend_block.c revision 279005
1/*- 2 * Copyright (c) 2003 Silicon Graphics International Corp. 3 * Copyright (c) 2009-2011 Spectra Logic Corporation 4 * Copyright (c) 2012 The FreeBSD Foundation 5 * All rights reserved. 6 * 7 * Portions of this software were developed by Edward Tomasz Napierala 8 * under sponsorship from the FreeBSD Foundation. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions, and the following disclaimer, 15 * without modification. 16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17 * substantially similar to the "NO WARRANTY" disclaimer below 18 * ("Disclaimer") and any redistribution must be conditioned upon 19 * including a substantially similar Disclaimer requirement for further 20 * binary redistribution. 21 * 22 * NO WARRANTY 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGES. 34 * 35 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36 */ 37/* 38 * CAM Target Layer driver backend for block devices. 39 * 40 * Author: Ken Merry <ken@FreeBSD.org> 41 */ 42#include <sys/cdefs.h> 43__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl_backend_block.c 279005 2015-02-19 14:40:50Z mav $"); 44 45#include <opt_kdtrace.h> 46 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/kernel.h> 50#include <sys/types.h> 51#include <sys/kthread.h> 52#include <sys/bio.h> 53#include <sys/fcntl.h> 54#include <sys/limits.h> 55#include <sys/lock.h> 56#include <sys/mutex.h> 57#include <sys/condvar.h> 58#include <sys/malloc.h> 59#include <sys/conf.h> 60#include <sys/ioccom.h> 61#include <sys/queue.h> 62#include <sys/sbuf.h> 63#include <sys/endian.h> 64#include <sys/uio.h> 65#include <sys/buf.h> 66#include <sys/taskqueue.h> 67#include <sys/vnode.h> 68#include <sys/namei.h> 69#include <sys/mount.h> 70#include <sys/disk.h> 71#include <sys/fcntl.h> 72#include <sys/filedesc.h> 73#include <sys/filio.h> 74#include <sys/proc.h> 75#include <sys/pcpu.h> 76#include <sys/module.h> 77#include <sys/sdt.h> 78#include <sys/devicestat.h> 79#include <sys/sysctl.h> 80 81#include <geom/geom.h> 82 83#include <cam/cam.h> 84#include <cam/scsi/scsi_all.h> 85#include <cam/scsi/scsi_da.h> 86#include <cam/ctl/ctl_io.h> 87#include <cam/ctl/ctl.h> 88#include <cam/ctl/ctl_backend.h> 89#include <cam/ctl/ctl_frontend_internal.h> 90#include <cam/ctl/ctl_ioctl.h> 91#include <cam/ctl/ctl_scsi_all.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_MULTI_THREAD = 0x08 123} ctl_be_block_lun_flags; 124 125typedef enum { 126 CTL_BE_BLOCK_NONE, 127 CTL_BE_BLOCK_DEV, 128 CTL_BE_BLOCK_FILE 129} ctl_be_block_type; 130 131struct ctl_be_block_devdata { 132 struct cdev *cdev; 133 struct cdevsw *csw; 134 int dev_ref; 135}; 136 137struct ctl_be_block_filedata { 138 struct ucred *cred; 139}; 140 141union ctl_be_block_bedata { 142 struct ctl_be_block_devdata dev; 143 struct ctl_be_block_filedata file; 144}; 145 146struct ctl_be_block_io; 147struct ctl_be_block_lun; 148 149typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 150 struct ctl_be_block_io *beio); 151typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 152 const char *attrname); 153 154/* 155 * Backend LUN structure. There is a 1:1 mapping between a block device 156 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 157 */ 158struct ctl_be_block_lun { 159 struct ctl_lun_create_params params; 160 struct ctl_block_disk *disk; 161 char lunname[32]; 162 char *dev_path; 163 ctl_be_block_type dev_type; 164 struct vnode *vn; 165 union ctl_be_block_bedata backend; 166 cbb_dispatch_t dispatch; 167 cbb_dispatch_t lun_flush; 168 cbb_dispatch_t unmap; 169 cbb_dispatch_t get_lba_status; 170 cbb_getattr_t getattr; 171 uma_zone_t lun_zone; 172 uint64_t size_blocks; 173 uint64_t size_bytes; 174 uint32_t blocksize; 175 int blocksize_shift; 176 uint16_t pblockexp; 177 uint16_t pblockoff; 178 uint16_t ublockexp; 179 uint16_t ublockoff; 180 uint32_t atomicblock; 181 uint32_t opttxferlen; 182 struct ctl_be_block_softc *softc; 183 struct devstat *disk_stats; 184 ctl_be_block_lun_flags flags; 185 STAILQ_ENTRY(ctl_be_block_lun) links; 186 struct ctl_be_lun ctl_be_lun; 187 struct taskqueue *io_taskqueue; 188 struct task io_task; 189 int num_threads; 190 STAILQ_HEAD(, ctl_io_hdr) input_queue; 191 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 192 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 193 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 194 struct mtx_padalign io_lock; 195 struct mtx_padalign queue_lock; 196}; 197 198/* 199 * Overall softc structure for the block backend module. 200 */ 201struct ctl_be_block_softc { 202 struct mtx lock; 203 int num_disks; 204 STAILQ_HEAD(, ctl_block_disk) disk_list; 205 int num_luns; 206 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 207}; 208 209static struct ctl_be_block_softc backend_block_softc; 210 211/* 212 * Per-I/O information. 213 */ 214struct ctl_be_block_io { 215 union ctl_io *io; 216 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 217 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 218 int bio_cmd; 219 int num_segs; 220 int num_bios_sent; 221 int num_bios_done; 222 int send_complete; 223 int num_errors; 224 struct bintime ds_t0; 225 devstat_tag_type ds_tag_type; 226 devstat_trans_flags ds_trans_type; 227 uint64_t io_len; 228 uint64_t io_offset; 229 struct ctl_be_block_softc *softc; 230 struct ctl_be_block_lun *lun; 231 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 232}; 233 234static int cbb_num_threads = 14; 235TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 236SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 237 "CAM Target Layer Block Backend"); 238SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 239 &cbb_num_threads, 0, "Number of threads per backing file"); 240 241static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 242static void ctl_free_beio(struct ctl_be_block_io *beio); 243static void ctl_complete_beio(struct ctl_be_block_io *beio); 244static int ctl_be_block_move_done(union ctl_io *io); 245static void ctl_be_block_biodone(struct bio *bio); 246static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 247 struct ctl_be_block_io *beio); 248static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 249 struct ctl_be_block_io *beio); 250static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 251 struct ctl_be_block_io *beio); 252static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 253 const char *attrname); 254static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 255 struct ctl_be_block_io *beio); 256static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 257 struct ctl_be_block_io *beio); 258static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 259 struct ctl_be_block_io *beio); 260static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 261 const char *attrname); 262static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 263 union ctl_io *io); 264static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 265 union ctl_io *io); 266static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 267 union ctl_io *io); 268static void ctl_be_block_worker(void *context, int pending); 269static int ctl_be_block_submit(union ctl_io *io); 270static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 271 int flag, struct thread *td); 272static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 273 struct ctl_lun_req *req); 274static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 275 struct ctl_lun_req *req); 276static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 277static int ctl_be_block_open(struct ctl_be_block_softc *softc, 278 struct ctl_be_block_lun *be_lun, 279 struct ctl_lun_req *req); 280static int ctl_be_block_create(struct ctl_be_block_softc *softc, 281 struct ctl_lun_req *req); 282static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 283 struct ctl_lun_req *req); 284static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 285 struct ctl_lun_req *req); 286static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 287 struct ctl_lun_req *req); 288static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 289 struct ctl_lun_req *req); 290static void ctl_be_block_lun_shutdown(void *be_lun); 291static void ctl_be_block_lun_config_status(void *be_lun, 292 ctl_lun_config_status status); 293static int ctl_be_block_config_write(union ctl_io *io); 294static int ctl_be_block_config_read(union ctl_io *io); 295static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 296static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 297int ctl_be_block_init(void); 298 299static struct ctl_backend_driver ctl_be_block_driver = 300{ 301 .name = "block", 302 .flags = CTL_BE_FLAG_HAS_CONFIG, 303 .init = ctl_be_block_init, 304 .data_submit = ctl_be_block_submit, 305 .data_move_done = ctl_be_block_move_done, 306 .config_read = ctl_be_block_config_read, 307 .config_write = ctl_be_block_config_write, 308 .ioctl = ctl_be_block_ioctl, 309 .lun_info = ctl_be_block_lun_info, 310 .lun_attr = ctl_be_block_lun_attr 311}; 312 313MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 314CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 315 316static uma_zone_t beio_zone; 317 318static struct ctl_be_block_io * 319ctl_alloc_beio(struct ctl_be_block_softc *softc) 320{ 321 struct ctl_be_block_io *beio; 322 323 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 324 beio->softc = softc; 325 return (beio); 326} 327 328static void 329ctl_free_beio(struct ctl_be_block_io *beio) 330{ 331 int duplicate_free; 332 int i; 333 334 duplicate_free = 0; 335 336 for (i = 0; i < beio->num_segs; i++) { 337 if (beio->sg_segs[i].addr == NULL) 338 duplicate_free++; 339 340 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 341 beio->sg_segs[i].addr = NULL; 342 343 /* For compare we had two equal S/G lists. */ 344 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 345 uma_zfree(beio->lun->lun_zone, 346 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 347 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 348 } 349 } 350 351 if (duplicate_free > 0) { 352 printf("%s: %d duplicate frees out of %d segments\n", __func__, 353 duplicate_free, beio->num_segs); 354 } 355 356 uma_zfree(beio_zone, beio); 357} 358 359static void 360ctl_complete_beio(struct ctl_be_block_io *beio) 361{ 362 union ctl_io *io = beio->io; 363 364 if (beio->beio_cont != NULL) { 365 beio->beio_cont(beio); 366 } else { 367 ctl_free_beio(beio); 368 ctl_data_submit_done(io); 369 } 370} 371 372static int 373ctl_be_block_move_done(union ctl_io *io) 374{ 375 struct ctl_be_block_io *beio; 376 struct ctl_be_block_lun *be_lun; 377 struct ctl_lba_len_flags *lbalen; 378#ifdef CTL_TIME_IO 379 struct bintime cur_bt; 380#endif 381 int i; 382 383 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 384 be_lun = beio->lun; 385 386 DPRINTF("entered\n"); 387 388#ifdef CTL_TIME_IO 389 getbintime(&cur_bt); 390 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 391 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 392 io->io_hdr.num_dmas++; 393#endif 394 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 395 396 /* 397 * We set status at this point for read commands, and write 398 * commands with errors. 399 */ 400 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 401 ; 402 } else if ((io->io_hdr.port_status == 0) && 403 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 404 lbalen = ARGS(beio->io); 405 if (lbalen->flags & CTL_LLF_READ) { 406 ctl_set_success(&io->scsiio); 407 } else if (lbalen->flags & CTL_LLF_COMPARE) { 408 /* We have two data blocks ready for comparison. */ 409 for (i = 0; i < beio->num_segs; i++) { 410 if (memcmp(beio->sg_segs[i].addr, 411 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 412 beio->sg_segs[i].len) != 0) 413 break; 414 } 415 if (i < beio->num_segs) 416 ctl_set_sense(&io->scsiio, 417 /*current_error*/ 1, 418 /*sense_key*/ SSD_KEY_MISCOMPARE, 419 /*asc*/ 0x1D, 420 /*ascq*/ 0x00, 421 SSD_ELEM_NONE); 422 else 423 ctl_set_success(&io->scsiio); 424 } 425 } else if ((io->io_hdr.port_status != 0) && 426 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 427 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 428 /* 429 * For hardware error sense keys, the sense key 430 * specific value is defined to be a retry count, 431 * but we use it to pass back an internal FETD 432 * error code. XXX KDM Hopefully the FETD is only 433 * using 16 bits for an error code, since that's 434 * all the space we have in the sks field. 435 */ 436 ctl_set_internal_failure(&io->scsiio, 437 /*sks_valid*/ 1, 438 /*retry_count*/ 439 io->io_hdr.port_status); 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 /* 462 * XXX KDM make sure that links is okay to use at this point. 463 * Otherwise, we either need to add another field to ctl_io_hdr, 464 * or deal with resource allocation here. 465 */ 466 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 467 mtx_unlock(&be_lun->queue_lock); 468 469 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 470 471 return (0); 472} 473 474static void 475ctl_be_block_biodone(struct bio *bio) 476{ 477 struct ctl_be_block_io *beio; 478 struct ctl_be_block_lun *be_lun; 479 union ctl_io *io; 480 int error; 481 482 beio = bio->bio_caller1; 483 be_lun = beio->lun; 484 io = beio->io; 485 486 DPRINTF("entered\n"); 487 488 error = bio->bio_error; 489 mtx_lock(&be_lun->io_lock); 490 if (error != 0) 491 beio->num_errors++; 492 493 beio->num_bios_done++; 494 495 /* 496 * XXX KDM will this cause WITNESS to complain? Holding a lock 497 * during the free might cause it to complain. 498 */ 499 g_destroy_bio(bio); 500 501 /* 502 * If the send complete bit isn't set, or we aren't the last I/O to 503 * complete, then we're done. 504 */ 505 if ((beio->send_complete == 0) 506 || (beio->num_bios_done < beio->num_bios_sent)) { 507 mtx_unlock(&be_lun->io_lock); 508 return; 509 } 510 511 /* 512 * At this point, we've verified that we are the last I/O to 513 * complete, so it's safe to drop the lock. 514 */ 515 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 516 beio->ds_tag_type, beio->ds_trans_type, 517 /*now*/ NULL, /*then*/&beio->ds_t0); 518 mtx_unlock(&be_lun->io_lock); 519 520 /* 521 * If there are any errors from the backing device, we fail the 522 * entire I/O with a medium error. 523 */ 524 if (beio->num_errors > 0) { 525 if (error == EOPNOTSUPP) { 526 ctl_set_invalid_opcode(&io->scsiio); 527 } else if (error == ENOSPC) { 528 ctl_set_space_alloc_fail(&io->scsiio); 529 } else if (beio->bio_cmd == BIO_FLUSH) { 530 /* XXX KDM is there is a better error here? */ 531 ctl_set_internal_failure(&io->scsiio, 532 /*sks_valid*/ 1, 533 /*retry_count*/ 0xbad2); 534 } else 535 ctl_set_medium_error(&io->scsiio); 536 ctl_complete_beio(beio); 537 return; 538 } 539 540 /* 541 * If this is a write, a flush, a delete or verify, we're all done. 542 * If this is a read, we can now send the data to the user. 543 */ 544 if ((beio->bio_cmd == BIO_WRITE) 545 || (beio->bio_cmd == BIO_FLUSH) 546 || (beio->bio_cmd == BIO_DELETE) 547 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 548 ctl_set_success(&io->scsiio); 549 ctl_complete_beio(beio); 550 } else { 551 if ((ARGS(io)->flags & CTL_LLF_READ) && 552 beio->beio_cont == NULL) 553 ctl_set_success(&io->scsiio); 554#ifdef CTL_TIME_IO 555 getbintime(&io->io_hdr.dma_start_bt); 556#endif 557 ctl_datamove(io); 558 } 559} 560 561static void 562ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 563 struct ctl_be_block_io *beio) 564{ 565 union ctl_io *io = beio->io; 566 struct mount *mountpoint; 567 int error, lock_flags; 568 569 DPRINTF("entered\n"); 570 571 binuptime(&beio->ds_t0); 572 mtx_lock(&be_lun->io_lock); 573 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 574 mtx_unlock(&be_lun->io_lock); 575 576 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 577 578 if (MNT_SHARED_WRITES(mountpoint) 579 || ((mountpoint == NULL) 580 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 581 lock_flags = LK_SHARED; 582 else 583 lock_flags = LK_EXCLUSIVE; 584 585 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 586 587 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 588 VOP_UNLOCK(be_lun->vn, 0); 589 590 vn_finished_write(mountpoint); 591 592 mtx_lock(&be_lun->io_lock); 593 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 594 beio->ds_tag_type, beio->ds_trans_type, 595 /*now*/ NULL, /*then*/&beio->ds_t0); 596 mtx_unlock(&be_lun->io_lock); 597 598 if (error == 0) 599 ctl_set_success(&io->scsiio); 600 else { 601 /* XXX KDM is there is a better error here? */ 602 ctl_set_internal_failure(&io->scsiio, 603 /*sks_valid*/ 1, 604 /*retry_count*/ 0xbad1); 605 } 606 607 ctl_complete_beio(beio); 608} 609 610SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 611SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 612SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 613SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 614 615static void 616ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 617 struct ctl_be_block_io *beio) 618{ 619 struct ctl_be_block_filedata *file_data; 620 union ctl_io *io; 621 struct uio xuio; 622 struct iovec *xiovec; 623 int flags; 624 int error, i; 625 626 DPRINTF("entered\n"); 627 628 file_data = &be_lun->backend.file; 629 io = beio->io; 630 flags = 0; 631 if (ARGS(io)->flags & CTL_LLF_DPO) 632 flags |= IO_DIRECT; 633 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 634 flags |= IO_SYNC; 635 636 bzero(&xuio, sizeof(xuio)); 637 if (beio->bio_cmd == BIO_READ) { 638 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 639 xuio.uio_rw = UIO_READ; 640 } else { 641 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 642 xuio.uio_rw = UIO_WRITE; 643 } 644 xuio.uio_offset = beio->io_offset; 645 xuio.uio_resid = beio->io_len; 646 xuio.uio_segflg = UIO_SYSSPACE; 647 xuio.uio_iov = beio->xiovecs; 648 xuio.uio_iovcnt = beio->num_segs; 649 xuio.uio_td = curthread; 650 651 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 652 xiovec->iov_base = beio->sg_segs[i].addr; 653 xiovec->iov_len = beio->sg_segs[i].len; 654 } 655 656 binuptime(&beio->ds_t0); 657 mtx_lock(&be_lun->io_lock); 658 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 659 mtx_unlock(&be_lun->io_lock); 660 661 if (beio->bio_cmd == BIO_READ) { 662 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 663 664 /* 665 * UFS pays attention to IO_DIRECT for reads. If the 666 * DIRECTIO option is configured into the kernel, it calls 667 * ffs_rawread(). But that only works for single-segment 668 * uios with user space addresses. In our case, with a 669 * kernel uio, it still reads into the buffer cache, but it 670 * will just try to release the buffer from the cache later 671 * on in ffs_read(). 672 * 673 * ZFS does not pay attention to IO_DIRECT for reads. 674 * 675 * UFS does not pay attention to IO_SYNC for reads. 676 * 677 * ZFS pays attention to IO_SYNC (which translates into the 678 * Solaris define FRSYNC for zfs_read()) for reads. It 679 * attempts to sync the file before reading. 680 * 681 * So, to attempt to provide some barrier semantics in the 682 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 683 */ 684 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 685 686 VOP_UNLOCK(be_lun->vn, 0); 687 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 688 } else { 689 struct mount *mountpoint; 690 int lock_flags; 691 692 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 693 694 if (MNT_SHARED_WRITES(mountpoint) 695 || ((mountpoint == NULL) 696 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 697 lock_flags = LK_SHARED; 698 else 699 lock_flags = LK_EXCLUSIVE; 700 701 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 702 703 /* 704 * UFS pays attention to IO_DIRECT for writes. The write 705 * is done asynchronously. (Normally the write would just 706 * get put into cache. 707 * 708 * UFS pays attention to IO_SYNC for writes. It will 709 * attempt to write the buffer out synchronously if that 710 * flag is set. 711 * 712 * ZFS does not pay attention to IO_DIRECT for writes. 713 * 714 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 715 * for writes. It will flush the transaction from the 716 * cache before returning. 717 * 718 * So if we've got the BIO_ORDERED flag set, we want 719 * IO_SYNC in either the UFS or ZFS case. 720 */ 721 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 722 VOP_UNLOCK(be_lun->vn, 0); 723 724 vn_finished_write(mountpoint); 725 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 726 } 727 728 mtx_lock(&be_lun->io_lock); 729 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 730 beio->ds_tag_type, beio->ds_trans_type, 731 /*now*/ NULL, /*then*/&beio->ds_t0); 732 mtx_unlock(&be_lun->io_lock); 733 734 /* 735 * If we got an error, set the sense data to "MEDIUM ERROR" and 736 * return the I/O to the user. 737 */ 738 if (error != 0) { 739 char path_str[32]; 740 741 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 742 printf("%s%s command returned errno %d\n", path_str, 743 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 744 if (error == ENOSPC) { 745 ctl_set_space_alloc_fail(&io->scsiio); 746 } else 747 ctl_set_medium_error(&io->scsiio); 748 ctl_complete_beio(beio); 749 return; 750 } 751 752 /* 753 * If this is a write or a verify, we're all done. 754 * If this is a read, we can now send the data to the user. 755 */ 756 if ((beio->bio_cmd == BIO_WRITE) || 757 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 758 ctl_set_success(&io->scsiio); 759 ctl_complete_beio(beio); 760 } else { 761 if ((ARGS(io)->flags & CTL_LLF_READ) && 762 beio->beio_cont == NULL) 763 ctl_set_success(&io->scsiio); 764#ifdef CTL_TIME_IO 765 getbintime(&io->io_hdr.dma_start_bt); 766#endif 767 ctl_datamove(io); 768 } 769} 770 771static void 772ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 773 struct ctl_be_block_io *beio) 774{ 775 union ctl_io *io = beio->io; 776 struct ctl_lba_len_flags *lbalen = ARGS(io); 777 struct scsi_get_lba_status_data *data; 778 off_t roff, off; 779 int error, status; 780 781 DPRINTF("entered\n"); 782 783 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 784 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 785 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 786 0, curthread->td_ucred, curthread); 787 if (error == 0 && off > roff) 788 status = 0; /* mapped up to off */ 789 else { 790 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 791 0, curthread->td_ucred, curthread); 792 if (error == 0 && off > roff) 793 status = 1; /* deallocated up to off */ 794 else { 795 status = 0; /* unknown up to the end */ 796 off = be_lun->size_bytes; 797 } 798 } 799 VOP_UNLOCK(be_lun->vn, 0); 800 801 off >>= be_lun->blocksize_shift; 802 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 803 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 804 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 805 data->descr[0].length); 806 data->descr[0].status = status; 807 808 ctl_complete_beio(beio); 809} 810 811static uint64_t 812ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 813{ 814 struct vattr vattr; 815 struct statfs statfs; 816 int error; 817 818 if (be_lun->vn == NULL) 819 return (UINT64_MAX); 820 if (strcmp(attrname, "blocksused") == 0) { 821 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 822 if (error != 0) 823 return (UINT64_MAX); 824 return (vattr.va_bytes >> be_lun->blocksize_shift); 825 } 826 if (strcmp(attrname, "blocksavail") == 0) { 827 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 828 if (error != 0) 829 return (UINT64_MAX); 830 return ((statfs.f_bavail * statfs.f_bsize) >> 831 be_lun->blocksize_shift); 832 } 833 return (UINT64_MAX); 834} 835 836static void 837ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 838 struct ctl_be_block_io *beio) 839{ 840 struct ctl_be_block_devdata *dev_data; 841 union ctl_io *io; 842 struct uio xuio; 843 struct iovec *xiovec; 844 int flags; 845 int error, i; 846 847 DPRINTF("entered\n"); 848 849 dev_data = &be_lun->backend.dev; 850 io = beio->io; 851 flags = 0; 852 if (ARGS(io)->flags & CTL_LLF_DPO) 853 flags |= IO_DIRECT; 854 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 855 flags |= IO_SYNC; 856 857 bzero(&xuio, sizeof(xuio)); 858 if (beio->bio_cmd == BIO_READ) { 859 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 860 xuio.uio_rw = UIO_READ; 861 } else { 862 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 863 xuio.uio_rw = UIO_WRITE; 864 } 865 xuio.uio_offset = beio->io_offset; 866 xuio.uio_resid = beio->io_len; 867 xuio.uio_segflg = UIO_SYSSPACE; 868 xuio.uio_iov = beio->xiovecs; 869 xuio.uio_iovcnt = beio->num_segs; 870 xuio.uio_td = curthread; 871 872 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 873 xiovec->iov_base = beio->sg_segs[i].addr; 874 xiovec->iov_len = beio->sg_segs[i].len; 875 } 876 877 binuptime(&beio->ds_t0); 878 mtx_lock(&be_lun->io_lock); 879 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 880 mtx_unlock(&be_lun->io_lock); 881 882 if (beio->bio_cmd == BIO_READ) { 883 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 884 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 885 } else { 886 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 887 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 888 } 889 890 mtx_lock(&be_lun->io_lock); 891 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 892 beio->ds_tag_type, beio->ds_trans_type, 893 /*now*/ NULL, /*then*/&beio->ds_t0); 894 mtx_unlock(&be_lun->io_lock); 895 896 /* 897 * If we got an error, set the sense data to "MEDIUM ERROR" and 898 * return the I/O to the user. 899 */ 900 if (error != 0) { 901 if (error == ENOSPC) { 902 ctl_set_space_alloc_fail(&io->scsiio); 903 } else 904 ctl_set_medium_error(&io->scsiio); 905 ctl_complete_beio(beio); 906 return; 907 } 908 909 /* 910 * If this is a write or a verify, we're all done. 911 * If this is a read, we can now send the data to the user. 912 */ 913 if ((beio->bio_cmd == BIO_WRITE) || 914 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 915 ctl_set_success(&io->scsiio); 916 ctl_complete_beio(beio); 917 } else { 918 if ((ARGS(io)->flags & CTL_LLF_READ) && 919 beio->beio_cont == NULL) 920 ctl_set_success(&io->scsiio); 921#ifdef CTL_TIME_IO 922 getbintime(&io->io_hdr.dma_start_bt); 923#endif 924 ctl_datamove(io); 925 } 926} 927 928static void 929ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 930 struct ctl_be_block_io *beio) 931{ 932 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 933 union ctl_io *io = beio->io; 934 struct ctl_lba_len_flags *lbalen = ARGS(io); 935 struct scsi_get_lba_status_data *data; 936 off_t roff, off; 937 int error, status; 938 939 DPRINTF("entered\n"); 940 941 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 942 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE, 943 (caddr_t)&off, FREAD, curthread); 944 if (error == 0 && off > roff) 945 status = 0; /* mapped up to off */ 946 else { 947 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA, 948 (caddr_t)&off, FREAD, curthread); 949 if (error == 0 && off > roff) 950 status = 1; /* deallocated up to off */ 951 else { 952 status = 0; /* unknown up to the end */ 953 off = be_lun->size_bytes; 954 } 955 } 956 957 off >>= be_lun->blocksize_shift; 958 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 959 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 960 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 961 data->descr[0].length); 962 data->descr[0].status = status; 963 964 ctl_complete_beio(beio); 965} 966 967static void 968ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 969 struct ctl_be_block_io *beio) 970{ 971 struct bio *bio; 972 union ctl_io *io; 973 struct ctl_be_block_devdata *dev_data; 974 975 dev_data = &be_lun->backend.dev; 976 io = beio->io; 977 978 DPRINTF("entered\n"); 979 980 /* This can't fail, it's a blocking allocation. */ 981 bio = g_alloc_bio(); 982 983 bio->bio_cmd = BIO_FLUSH; 984 bio->bio_flags |= BIO_ORDERED; 985 bio->bio_dev = dev_data->cdev; 986 bio->bio_offset = 0; 987 bio->bio_data = 0; 988 bio->bio_done = ctl_be_block_biodone; 989 bio->bio_caller1 = beio; 990 bio->bio_pblkno = 0; 991 992 /* 993 * We don't need to acquire the LUN lock here, because we are only 994 * sending one bio, and so there is no other context to synchronize 995 * with. 996 */ 997 beio->num_bios_sent = 1; 998 beio->send_complete = 1; 999 1000 binuptime(&beio->ds_t0); 1001 mtx_lock(&be_lun->io_lock); 1002 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1003 mtx_unlock(&be_lun->io_lock); 1004 1005 (*dev_data->csw->d_strategy)(bio); 1006} 1007 1008static void 1009ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1010 struct ctl_be_block_io *beio, 1011 uint64_t off, uint64_t len, int last) 1012{ 1013 struct bio *bio; 1014 struct ctl_be_block_devdata *dev_data; 1015 uint64_t maxlen; 1016 1017 dev_data = &be_lun->backend.dev; 1018 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 1019 while (len > 0) { 1020 bio = g_alloc_bio(); 1021 bio->bio_cmd = BIO_DELETE; 1022 bio->bio_dev = dev_data->cdev; 1023 bio->bio_offset = off; 1024 bio->bio_length = MIN(len, maxlen); 1025 bio->bio_data = 0; 1026 bio->bio_done = ctl_be_block_biodone; 1027 bio->bio_caller1 = beio; 1028 bio->bio_pblkno = off / be_lun->blocksize; 1029 1030 off += bio->bio_length; 1031 len -= bio->bio_length; 1032 1033 mtx_lock(&be_lun->io_lock); 1034 beio->num_bios_sent++; 1035 if (last && len == 0) 1036 beio->send_complete = 1; 1037 mtx_unlock(&be_lun->io_lock); 1038 1039 (*dev_data->csw->d_strategy)(bio); 1040 } 1041} 1042 1043static void 1044ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1045 struct ctl_be_block_io *beio) 1046{ 1047 union ctl_io *io; 1048 struct ctl_be_block_devdata *dev_data; 1049 struct ctl_ptr_len_flags *ptrlen; 1050 struct scsi_unmap_desc *buf, *end; 1051 uint64_t len; 1052 1053 dev_data = &be_lun->backend.dev; 1054 io = beio->io; 1055 1056 DPRINTF("entered\n"); 1057 1058 binuptime(&beio->ds_t0); 1059 mtx_lock(&be_lun->io_lock); 1060 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1061 mtx_unlock(&be_lun->io_lock); 1062 1063 if (beio->io_offset == -1) { 1064 beio->io_len = 0; 1065 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1066 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1067 end = buf + ptrlen->len / sizeof(*buf); 1068 for (; buf < end; buf++) { 1069 len = (uint64_t)scsi_4btoul(buf->length) * 1070 be_lun->blocksize; 1071 beio->io_len += len; 1072 ctl_be_block_unmap_dev_range(be_lun, beio, 1073 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 1074 (end - buf < 2) ? TRUE : FALSE); 1075 } 1076 } else 1077 ctl_be_block_unmap_dev_range(be_lun, beio, 1078 beio->io_offset, beio->io_len, TRUE); 1079} 1080 1081static void 1082ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1083 struct ctl_be_block_io *beio) 1084{ 1085 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1086 int i; 1087 struct bio *bio; 1088 struct ctl_be_block_devdata *dev_data; 1089 off_t cur_offset; 1090 int max_iosize; 1091 1092 DPRINTF("entered\n"); 1093 1094 dev_data = &be_lun->backend.dev; 1095 1096 /* 1097 * We have to limit our I/O size to the maximum supported by the 1098 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1099 * set it properly, use DFLTPHYS. 1100 */ 1101 max_iosize = dev_data->cdev->si_iosize_max; 1102 if (max_iosize < PAGE_SIZE) 1103 max_iosize = DFLTPHYS; 1104 1105 cur_offset = beio->io_offset; 1106 for (i = 0; i < beio->num_segs; i++) { 1107 size_t cur_size; 1108 uint8_t *cur_ptr; 1109 1110 cur_size = beio->sg_segs[i].len; 1111 cur_ptr = beio->sg_segs[i].addr; 1112 1113 while (cur_size > 0) { 1114 /* This can't fail, it's a blocking allocation. */ 1115 bio = g_alloc_bio(); 1116 1117 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1118 1119 bio->bio_cmd = beio->bio_cmd; 1120 bio->bio_dev = dev_data->cdev; 1121 bio->bio_caller1 = beio; 1122 bio->bio_length = min(cur_size, max_iosize); 1123 bio->bio_offset = cur_offset; 1124 bio->bio_data = cur_ptr; 1125 bio->bio_done = ctl_be_block_biodone; 1126 bio->bio_pblkno = cur_offset / be_lun->blocksize; 1127 1128 cur_offset += bio->bio_length; 1129 cur_ptr += bio->bio_length; 1130 cur_size -= bio->bio_length; 1131 1132 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1133 beio->num_bios_sent++; 1134 } 1135 } 1136 binuptime(&beio->ds_t0); 1137 mtx_lock(&be_lun->io_lock); 1138 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1139 beio->send_complete = 1; 1140 mtx_unlock(&be_lun->io_lock); 1141 1142 /* 1143 * Fire off all allocated requests! 1144 */ 1145 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1146 TAILQ_REMOVE(&queue, bio, bio_queue); 1147 (*dev_data->csw->d_strategy)(bio); 1148 } 1149} 1150 1151static uint64_t 1152ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1153{ 1154 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1155 struct diocgattr_arg arg; 1156 int error; 1157 1158 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1159 return (UINT64_MAX); 1160 strlcpy(arg.name, attrname, sizeof(arg.name)); 1161 arg.len = sizeof(arg.value.off); 1162 error = dev_data->csw->d_ioctl(dev_data->cdev, 1163 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1164 if (error != 0) 1165 return (UINT64_MAX); 1166 return (arg.value.off); 1167} 1168 1169static void 1170ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1171{ 1172 union ctl_io *io; 1173 1174 io = beio->io; 1175 ctl_free_beio(beio); 1176 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1177 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1178 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1179 ctl_config_write_done(io); 1180 return; 1181 } 1182 1183 ctl_be_block_config_write(io); 1184} 1185 1186static void 1187ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1188 union ctl_io *io) 1189{ 1190 struct ctl_be_block_io *beio; 1191 struct ctl_be_block_softc *softc; 1192 struct ctl_lba_len_flags *lbalen; 1193 uint64_t len_left, lba; 1194 uint32_t pb, pbo, adj; 1195 int i, seglen; 1196 uint8_t *buf, *end; 1197 1198 DPRINTF("entered\n"); 1199 1200 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1201 softc = be_lun->softc; 1202 lbalen = ARGS(beio->io); 1203 1204 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1205 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1206 ctl_free_beio(beio); 1207 ctl_set_invalid_field(&io->scsiio, 1208 /*sks_valid*/ 1, 1209 /*command*/ 1, 1210 /*field*/ 1, 1211 /*bit_valid*/ 0, 1212 /*bit*/ 0); 1213 ctl_config_write_done(io); 1214 return; 1215 } 1216 1217 switch (io->scsiio.tag_type) { 1218 case CTL_TAG_ORDERED: 1219 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1220 break; 1221 case CTL_TAG_HEAD_OF_QUEUE: 1222 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1223 break; 1224 case CTL_TAG_UNTAGGED: 1225 case CTL_TAG_SIMPLE: 1226 case CTL_TAG_ACA: 1227 default: 1228 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1229 break; 1230 } 1231 1232 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1233 beio->io_offset = lbalen->lba * be_lun->blocksize; 1234 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1235 beio->bio_cmd = BIO_DELETE; 1236 beio->ds_trans_type = DEVSTAT_FREE; 1237 1238 be_lun->unmap(be_lun, beio); 1239 return; 1240 } 1241 1242 beio->bio_cmd = BIO_WRITE; 1243 beio->ds_trans_type = DEVSTAT_WRITE; 1244 1245 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1246 (uintmax_t)lbalen->lba, lbalen->len); 1247 1248 pb = be_lun->blocksize << be_lun->pblockexp; 1249 if (be_lun->pblockoff > 0) 1250 pbo = pb - be_lun->blocksize * be_lun->pblockoff; 1251 else 1252 pbo = 0; 1253 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1254 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1255 1256 /* 1257 * Setup the S/G entry for this chunk. 1258 */ 1259 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1260 if (pb > be_lun->blocksize) { 1261 adj = ((lbalen->lba + lba) * be_lun->blocksize + 1262 seglen - pbo) % pb; 1263 if (seglen > adj) 1264 seglen -= adj; 1265 else 1266 seglen -= seglen % be_lun->blocksize; 1267 } else 1268 seglen -= seglen % be_lun->blocksize; 1269 beio->sg_segs[i].len = seglen; 1270 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1271 1272 DPRINTF("segment %d addr %p len %zd\n", i, 1273 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1274 1275 beio->num_segs++; 1276 len_left -= seglen; 1277 1278 buf = beio->sg_segs[i].addr; 1279 end = buf + seglen; 1280 for (; buf < end; buf += be_lun->blocksize) { 1281 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1282 if (lbalen->flags & SWS_LBDATA) 1283 scsi_ulto4b(lbalen->lba + lba, buf); 1284 lba++; 1285 } 1286 } 1287 1288 beio->io_offset = lbalen->lba * be_lun->blocksize; 1289 beio->io_len = lba * be_lun->blocksize; 1290 1291 /* We can not do all in one run. Correct and schedule rerun. */ 1292 if (len_left > 0) { 1293 lbalen->lba += lba; 1294 lbalen->len -= lba; 1295 beio->beio_cont = ctl_be_block_cw_done_ws; 1296 } 1297 1298 be_lun->dispatch(be_lun, beio); 1299} 1300 1301static void 1302ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1303 union ctl_io *io) 1304{ 1305 struct ctl_be_block_io *beio; 1306 struct ctl_be_block_softc *softc; 1307 struct ctl_ptr_len_flags *ptrlen; 1308 1309 DPRINTF("entered\n"); 1310 1311 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1312 softc = be_lun->softc; 1313 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1314 1315 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1316 ctl_free_beio(beio); 1317 ctl_set_invalid_field(&io->scsiio, 1318 /*sks_valid*/ 0, 1319 /*command*/ 1, 1320 /*field*/ 0, 1321 /*bit_valid*/ 0, 1322 /*bit*/ 0); 1323 ctl_config_write_done(io); 1324 return; 1325 } 1326 1327 switch (io->scsiio.tag_type) { 1328 case CTL_TAG_ORDERED: 1329 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1330 break; 1331 case CTL_TAG_HEAD_OF_QUEUE: 1332 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1333 break; 1334 case CTL_TAG_UNTAGGED: 1335 case CTL_TAG_SIMPLE: 1336 case CTL_TAG_ACA: 1337 default: 1338 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1339 break; 1340 } 1341 1342 beio->io_len = 0; 1343 beio->io_offset = -1; 1344 1345 beio->bio_cmd = BIO_DELETE; 1346 beio->ds_trans_type = DEVSTAT_FREE; 1347 1348 DPRINTF("UNMAP\n"); 1349 1350 be_lun->unmap(be_lun, beio); 1351} 1352 1353static void 1354ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1355{ 1356 union ctl_io *io; 1357 1358 io = beio->io; 1359 ctl_free_beio(beio); 1360 ctl_config_read_done(io); 1361} 1362 1363static void 1364ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1365 union ctl_io *io) 1366{ 1367 struct ctl_be_block_io *beio; 1368 struct ctl_be_block_softc *softc; 1369 1370 DPRINTF("entered\n"); 1371 1372 softc = be_lun->softc; 1373 beio = ctl_alloc_beio(softc); 1374 beio->io = io; 1375 beio->lun = be_lun; 1376 beio->beio_cont = ctl_be_block_cr_done; 1377 PRIV(io)->ptr = (void *)beio; 1378 1379 switch (io->scsiio.cdb[0]) { 1380 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1381 beio->bio_cmd = -1; 1382 beio->ds_trans_type = DEVSTAT_NO_DATA; 1383 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1384 beio->io_len = 0; 1385 if (be_lun->get_lba_status) 1386 be_lun->get_lba_status(be_lun, beio); 1387 else 1388 ctl_be_block_cr_done(beio); 1389 break; 1390 default: 1391 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1392 break; 1393 } 1394} 1395 1396static void 1397ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1398{ 1399 union ctl_io *io; 1400 1401 io = beio->io; 1402 ctl_free_beio(beio); 1403 ctl_config_write_done(io); 1404} 1405 1406static void 1407ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1408 union ctl_io *io) 1409{ 1410 struct ctl_be_block_io *beio; 1411 struct ctl_be_block_softc *softc; 1412 1413 DPRINTF("entered\n"); 1414 1415 softc = be_lun->softc; 1416 beio = ctl_alloc_beio(softc); 1417 beio->io = io; 1418 beio->lun = be_lun; 1419 beio->beio_cont = ctl_be_block_cw_done; 1420 PRIV(io)->ptr = (void *)beio; 1421 1422 switch (io->scsiio.cdb[0]) { 1423 case SYNCHRONIZE_CACHE: 1424 case SYNCHRONIZE_CACHE_16: 1425 beio->bio_cmd = BIO_FLUSH; 1426 beio->ds_trans_type = DEVSTAT_NO_DATA; 1427 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1428 beio->io_len = 0; 1429 be_lun->lun_flush(be_lun, beio); 1430 break; 1431 case WRITE_SAME_10: 1432 case WRITE_SAME_16: 1433 ctl_be_block_cw_dispatch_ws(be_lun, io); 1434 break; 1435 case UNMAP: 1436 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1437 break; 1438 default: 1439 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1440 break; 1441 } 1442} 1443 1444SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1445SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1446SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1447SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1448 1449static void 1450ctl_be_block_next(struct ctl_be_block_io *beio) 1451{ 1452 struct ctl_be_block_lun *be_lun; 1453 union ctl_io *io; 1454 1455 io = beio->io; 1456 be_lun = beio->lun; 1457 ctl_free_beio(beio); 1458 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1459 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1460 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1461 ctl_data_submit_done(io); 1462 return; 1463 } 1464 1465 io->io_hdr.status &= ~CTL_STATUS_MASK; 1466 io->io_hdr.status |= CTL_STATUS_NONE; 1467 1468 mtx_lock(&be_lun->queue_lock); 1469 /* 1470 * XXX KDM make sure that links is okay to use at this point. 1471 * Otherwise, we either need to add another field to ctl_io_hdr, 1472 * or deal with resource allocation here. 1473 */ 1474 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1475 mtx_unlock(&be_lun->queue_lock); 1476 1477 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1478} 1479 1480static void 1481ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1482 union ctl_io *io) 1483{ 1484 struct ctl_be_block_io *beio; 1485 struct ctl_be_block_softc *softc; 1486 struct ctl_lba_len_flags *lbalen; 1487 struct ctl_ptr_len_flags *bptrlen; 1488 uint64_t len_left, lbas; 1489 int i; 1490 1491 softc = be_lun->softc; 1492 1493 DPRINTF("entered\n"); 1494 1495 lbalen = ARGS(io); 1496 if (lbalen->flags & CTL_LLF_WRITE) { 1497 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1498 } else { 1499 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1500 } 1501 1502 beio = ctl_alloc_beio(softc); 1503 beio->io = io; 1504 beio->lun = be_lun; 1505 bptrlen = PRIV(io); 1506 bptrlen->ptr = (void *)beio; 1507 1508 switch (io->scsiio.tag_type) { 1509 case CTL_TAG_ORDERED: 1510 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1511 break; 1512 case CTL_TAG_HEAD_OF_QUEUE: 1513 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1514 break; 1515 case CTL_TAG_UNTAGGED: 1516 case CTL_TAG_SIMPLE: 1517 case CTL_TAG_ACA: 1518 default: 1519 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1520 break; 1521 } 1522 1523 if (lbalen->flags & CTL_LLF_WRITE) { 1524 beio->bio_cmd = BIO_WRITE; 1525 beio->ds_trans_type = DEVSTAT_WRITE; 1526 } else { 1527 beio->bio_cmd = BIO_READ; 1528 beio->ds_trans_type = DEVSTAT_READ; 1529 } 1530 1531 DPRINTF("%s at LBA %jx len %u @%ju\n", 1532 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1533 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1534 if (lbalen->flags & CTL_LLF_COMPARE) 1535 lbas = CTLBLK_HALF_IO_SIZE; 1536 else 1537 lbas = CTLBLK_MAX_IO_SIZE; 1538 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1539 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1540 beio->io_len = lbas * be_lun->blocksize; 1541 bptrlen->len += lbas; 1542 1543 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1544 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1545 i, CTLBLK_MAX_SEGS)); 1546 1547 /* 1548 * Setup the S/G entry for this chunk. 1549 */ 1550 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1551 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1552 1553 DPRINTF("segment %d addr %p len %zd\n", i, 1554 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1555 1556 /* Set up second segment for compare operation. */ 1557 if (lbalen->flags & CTL_LLF_COMPARE) { 1558 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1559 beio->sg_segs[i].len; 1560 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1561 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1562 } 1563 1564 beio->num_segs++; 1565 len_left -= beio->sg_segs[i].len; 1566 } 1567 if (bptrlen->len < lbalen->len) 1568 beio->beio_cont = ctl_be_block_next; 1569 io->scsiio.be_move_done = ctl_be_block_move_done; 1570 /* For compare we have separate S/G lists for read and datamove. */ 1571 if (lbalen->flags & CTL_LLF_COMPARE) 1572 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1573 else 1574 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1575 io->scsiio.kern_data_len = beio->io_len; 1576 io->scsiio.kern_data_resid = 0; 1577 io->scsiio.kern_sg_entries = beio->num_segs; 1578 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1579 1580 /* 1581 * For the read case, we need to read the data into our buffers and 1582 * then we can send it back to the user. For the write case, we 1583 * need to get the data from the user first. 1584 */ 1585 if (beio->bio_cmd == BIO_READ) { 1586 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1587 be_lun->dispatch(be_lun, beio); 1588 } else { 1589 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1590#ifdef CTL_TIME_IO 1591 getbintime(&io->io_hdr.dma_start_bt); 1592#endif 1593 ctl_datamove(io); 1594 } 1595} 1596 1597static void 1598ctl_be_block_worker(void *context, int pending) 1599{ 1600 struct ctl_be_block_lun *be_lun; 1601 struct ctl_be_block_softc *softc; 1602 union ctl_io *io; 1603 1604 be_lun = (struct ctl_be_block_lun *)context; 1605 softc = be_lun->softc; 1606 1607 DPRINTF("entered\n"); 1608 1609 mtx_lock(&be_lun->queue_lock); 1610 for (;;) { 1611 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1612 if (io != NULL) { 1613 struct ctl_be_block_io *beio; 1614 1615 DPRINTF("datamove queue\n"); 1616 1617 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1618 ctl_io_hdr, links); 1619 1620 mtx_unlock(&be_lun->queue_lock); 1621 1622 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1623 1624 be_lun->dispatch(be_lun, beio); 1625 1626 mtx_lock(&be_lun->queue_lock); 1627 continue; 1628 } 1629 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1630 if (io != NULL) { 1631 DPRINTF("config write queue\n"); 1632 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1633 ctl_io_hdr, links); 1634 mtx_unlock(&be_lun->queue_lock); 1635 ctl_be_block_cw_dispatch(be_lun, io); 1636 mtx_lock(&be_lun->queue_lock); 1637 continue; 1638 } 1639 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1640 if (io != NULL) { 1641 DPRINTF("config read queue\n"); 1642 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1643 ctl_io_hdr, links); 1644 mtx_unlock(&be_lun->queue_lock); 1645 ctl_be_block_cr_dispatch(be_lun, io); 1646 mtx_lock(&be_lun->queue_lock); 1647 continue; 1648 } 1649 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1650 if (io != NULL) { 1651 DPRINTF("input queue\n"); 1652 1653 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1654 ctl_io_hdr, links); 1655 mtx_unlock(&be_lun->queue_lock); 1656 1657 /* 1658 * We must drop the lock, since this routine and 1659 * its children may sleep. 1660 */ 1661 ctl_be_block_dispatch(be_lun, io); 1662 1663 mtx_lock(&be_lun->queue_lock); 1664 continue; 1665 } 1666 1667 /* 1668 * If we get here, there is no work left in the queues, so 1669 * just break out and let the task queue go to sleep. 1670 */ 1671 break; 1672 } 1673 mtx_unlock(&be_lun->queue_lock); 1674} 1675 1676/* 1677 * Entry point from CTL to the backend for I/O. We queue everything to a 1678 * work thread, so this just puts the I/O on a queue and wakes up the 1679 * thread. 1680 */ 1681static int 1682ctl_be_block_submit(union ctl_io *io) 1683{ 1684 struct ctl_be_block_lun *be_lun; 1685 struct ctl_be_lun *ctl_be_lun; 1686 1687 DPRINTF("entered\n"); 1688 1689 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1690 CTL_PRIV_BACKEND_LUN].ptr; 1691 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1692 1693 /* 1694 * Make sure we only get SCSI I/O. 1695 */ 1696 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1697 "%#x) encountered", io->io_hdr.io_type)); 1698 1699 PRIV(io)->len = 0; 1700 1701 mtx_lock(&be_lun->queue_lock); 1702 /* 1703 * XXX KDM make sure that links is okay to use at this point. 1704 * Otherwise, we either need to add another field to ctl_io_hdr, 1705 * or deal with resource allocation here. 1706 */ 1707 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1708 mtx_unlock(&be_lun->queue_lock); 1709 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1710 1711 return (CTL_RETVAL_COMPLETE); 1712} 1713 1714static int 1715ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1716 int flag, struct thread *td) 1717{ 1718 struct ctl_be_block_softc *softc; 1719 int error; 1720 1721 softc = &backend_block_softc; 1722 1723 error = 0; 1724 1725 switch (cmd) { 1726 case CTL_LUN_REQ: { 1727 struct ctl_lun_req *lun_req; 1728 1729 lun_req = (struct ctl_lun_req *)addr; 1730 1731 switch (lun_req->reqtype) { 1732 case CTL_LUNREQ_CREATE: 1733 error = ctl_be_block_create(softc, lun_req); 1734 break; 1735 case CTL_LUNREQ_RM: 1736 error = ctl_be_block_rm(softc, lun_req); 1737 break; 1738 case CTL_LUNREQ_MODIFY: 1739 error = ctl_be_block_modify(softc, lun_req); 1740 break; 1741 default: 1742 lun_req->status = CTL_LUN_ERROR; 1743 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1744 "invalid LUN request type %d", 1745 lun_req->reqtype); 1746 break; 1747 } 1748 break; 1749 } 1750 default: 1751 error = ENOTTY; 1752 break; 1753 } 1754 1755 return (error); 1756} 1757 1758static int 1759ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1760{ 1761 struct ctl_be_block_filedata *file_data; 1762 struct ctl_lun_create_params *params; 1763 char *value; 1764 struct vattr vattr; 1765 off_t ps, pss, po, pos, us, uss, uo, uos; 1766 int error; 1767 1768 error = 0; 1769 file_data = &be_lun->backend.file; 1770 params = &be_lun->params; 1771 1772 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1773 be_lun->dispatch = ctl_be_block_dispatch_file; 1774 be_lun->lun_flush = ctl_be_block_flush_file; 1775 be_lun->get_lba_status = ctl_be_block_gls_file; 1776 be_lun->getattr = ctl_be_block_getattr_file; 1777 1778 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1779 if (error != 0) { 1780 snprintf(req->error_str, sizeof(req->error_str), 1781 "error calling VOP_GETATTR() for file %s", 1782 be_lun->dev_path); 1783 return (error); 1784 } 1785 1786 /* 1787 * Verify that we have the ability to upgrade to exclusive 1788 * access on this file so we can trap errors at open instead 1789 * of reporting them during first access. 1790 */ 1791 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1792 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1793 if (be_lun->vn->v_iflag & VI_DOOMED) { 1794 error = EBADF; 1795 snprintf(req->error_str, sizeof(req->error_str), 1796 "error locking file %s", be_lun->dev_path); 1797 return (error); 1798 } 1799 } 1800 1801 1802 file_data->cred = crhold(curthread->td_ucred); 1803 if (params->lun_size_bytes != 0) 1804 be_lun->size_bytes = params->lun_size_bytes; 1805 else 1806 be_lun->size_bytes = vattr.va_size; 1807 /* 1808 * We set the multi thread flag for file operations because all 1809 * filesystems (in theory) are capable of allowing multiple readers 1810 * of a file at once. So we want to get the maximum possible 1811 * concurrency. 1812 */ 1813 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1814 1815 /* 1816 * For files we can use any logical block size. Prefer 512 bytes 1817 * for compatibility reasons. If file's vattr.va_blocksize 1818 * (preferred I/O block size) is bigger and multiple to chosen 1819 * logical block size -- report it as physical block size. 1820 */ 1821 if (params->blocksize_bytes != 0) 1822 be_lun->blocksize = params->blocksize_bytes; 1823 else 1824 be_lun->blocksize = 512; 1825 1826 us = ps = vattr.va_blocksize; 1827 uo = po = 0; 1828 1829 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 1830 if (value != NULL) 1831 ctl_expand_number(value, &ps); 1832 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 1833 if (value != NULL) 1834 ctl_expand_number(value, &po); 1835 pss = ps / be_lun->blocksize; 1836 pos = po / be_lun->blocksize; 1837 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1838 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1839 be_lun->pblockexp = fls(pss) - 1; 1840 be_lun->pblockoff = (pss - pos) % pss; 1841 } 1842 1843 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 1844 if (value != NULL) 1845 ctl_expand_number(value, &us); 1846 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 1847 if (value != NULL) 1848 ctl_expand_number(value, &uo); 1849 uss = us / be_lun->blocksize; 1850 uos = uo / be_lun->blocksize; 1851 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 1852 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 1853 be_lun->ublockexp = fls(uss) - 1; 1854 be_lun->ublockoff = (uss - uos) % uss; 1855 } 1856 1857 /* 1858 * Sanity check. The media size has to be at least one 1859 * sector long. 1860 */ 1861 if (be_lun->size_bytes < be_lun->blocksize) { 1862 error = EINVAL; 1863 snprintf(req->error_str, sizeof(req->error_str), 1864 "file %s size %ju < block size %u", be_lun->dev_path, 1865 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1866 } 1867 1868 be_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / be_lun->blocksize; 1869 return (error); 1870} 1871 1872static int 1873ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1874{ 1875 struct ctl_lun_create_params *params; 1876 struct vattr vattr; 1877 struct cdev *dev; 1878 struct cdevsw *devsw; 1879 char *value; 1880 int error, atomic, maxio, unmap; 1881 off_t ps, pss, po, pos, us, uss, uo, uos; 1882 1883 params = &be_lun->params; 1884 1885 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1886 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1887 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1888 &be_lun->backend.dev.dev_ref); 1889 if (be_lun->backend.dev.csw == NULL) 1890 panic("Unable to retrieve device switch"); 1891 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) { 1892 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1893 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1894 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1895 } else { 1896 be_lun->dispatch = ctl_be_block_dispatch_dev; 1897 atomic = 0; 1898 maxio = be_lun->backend.dev.cdev->si_iosize_max; 1899 if (maxio <= 0) 1900 maxio = DFLTPHYS; 1901 if (maxio > CTLBLK_MAX_IO_SIZE) 1902 maxio = CTLBLK_MAX_IO_SIZE; 1903 } 1904 be_lun->lun_flush = ctl_be_block_flush_dev; 1905 be_lun->getattr = ctl_be_block_getattr_dev; 1906 1907 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1908 if (error) { 1909 snprintf(req->error_str, sizeof(req->error_str), 1910 "error getting vnode attributes for device %s", 1911 be_lun->dev_path); 1912 return (error); 1913 } 1914 1915 dev = be_lun->vn->v_rdev; 1916 devsw = dev->si_devsw; 1917 if (!devsw->d_ioctl) { 1918 snprintf(req->error_str, sizeof(req->error_str), 1919 "no d_ioctl for device %s!", 1920 be_lun->dev_path); 1921 return (ENODEV); 1922 } 1923 1924 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1925 (caddr_t)&be_lun->blocksize, FREAD, 1926 curthread); 1927 if (error) { 1928 snprintf(req->error_str, sizeof(req->error_str), 1929 "error %d returned for DIOCGSECTORSIZE ioctl " 1930 "on %s!", error, be_lun->dev_path); 1931 return (error); 1932 } 1933 1934 /* 1935 * If the user has asked for a blocksize that is greater than the 1936 * backing device's blocksize, we can do it only if the blocksize 1937 * the user is asking for is an even multiple of the underlying 1938 * device's blocksize. 1939 */ 1940 if ((params->blocksize_bytes != 0) 1941 && (params->blocksize_bytes > be_lun->blocksize)) { 1942 uint32_t bs_multiple, tmp_blocksize; 1943 1944 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1945 1946 tmp_blocksize = bs_multiple * be_lun->blocksize; 1947 1948 if (tmp_blocksize == params->blocksize_bytes) { 1949 be_lun->blocksize = params->blocksize_bytes; 1950 } else { 1951 snprintf(req->error_str, sizeof(req->error_str), 1952 "requested blocksize %u is not an even " 1953 "multiple of backing device blocksize %u", 1954 params->blocksize_bytes, 1955 be_lun->blocksize); 1956 return (EINVAL); 1957 1958 } 1959 } else if ((params->blocksize_bytes != 0) 1960 && (params->blocksize_bytes != be_lun->blocksize)) { 1961 snprintf(req->error_str, sizeof(req->error_str), 1962 "requested blocksize %u < backing device " 1963 "blocksize %u", params->blocksize_bytes, 1964 be_lun->blocksize); 1965 return (EINVAL); 1966 } 1967 1968 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1969 (caddr_t)&be_lun->size_bytes, FREAD, 1970 curthread); 1971 if (error) { 1972 snprintf(req->error_str, sizeof(req->error_str), 1973 "error %d returned for DIOCGMEDIASIZE " 1974 " ioctl on %s!", error, 1975 be_lun->dev_path); 1976 return (error); 1977 } 1978 1979 if (params->lun_size_bytes != 0) { 1980 if (params->lun_size_bytes > be_lun->size_bytes) { 1981 snprintf(req->error_str, sizeof(req->error_str), 1982 "requested LUN size %ju > backing device " 1983 "size %ju", 1984 (uintmax_t)params->lun_size_bytes, 1985 (uintmax_t)be_lun->size_bytes); 1986 return (EINVAL); 1987 } 1988 1989 be_lun->size_bytes = params->lun_size_bytes; 1990 } 1991 1992 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1993 (caddr_t)&ps, FREAD, curthread); 1994 if (error) 1995 ps = po = 0; 1996 else { 1997 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1998 (caddr_t)&po, FREAD, curthread); 1999 if (error) 2000 po = 0; 2001 } 2002 us = ps; 2003 uo = po; 2004 2005 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 2006 if (value != NULL) 2007 ctl_expand_number(value, &ps); 2008 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 2009 if (value != NULL) 2010 ctl_expand_number(value, &po); 2011 pss = ps / be_lun->blocksize; 2012 pos = po / be_lun->blocksize; 2013 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 2014 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 2015 be_lun->pblockexp = fls(pss) - 1; 2016 be_lun->pblockoff = (pss - pos) % pss; 2017 } 2018 2019 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 2020 if (value != NULL) 2021 ctl_expand_number(value, &us); 2022 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 2023 if (value != NULL) 2024 ctl_expand_number(value, &uo); 2025 uss = us / be_lun->blocksize; 2026 uos = uo / be_lun->blocksize; 2027 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 2028 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 2029 be_lun->ublockexp = fls(uss) - 1; 2030 be_lun->ublockoff = (uss - uos) % uss; 2031 } 2032 2033 be_lun->atomicblock = atomic / be_lun->blocksize; 2034 be_lun->opttxferlen = maxio / be_lun->blocksize; 2035 2036 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2037 unmap = 1; 2038 } else { 2039 struct diocgattr_arg arg; 2040 2041 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2042 arg.len = sizeof(arg.value.i); 2043 error = devsw->d_ioctl(dev, DIOCGATTR, 2044 (caddr_t)&arg, FREAD, curthread); 2045 unmap = (error == 0) ? arg.value.i : 0; 2046 } 2047 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2048 if (value != NULL) 2049 unmap = (strcmp(value, "on") == 0); 2050 if (unmap) 2051 be_lun->unmap = ctl_be_block_unmap_dev; 2052 2053 return (0); 2054} 2055 2056static int 2057ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2058{ 2059 DROP_GIANT(); 2060 if (be_lun->vn) { 2061 int flags = FREAD | FWRITE; 2062 2063 switch (be_lun->dev_type) { 2064 case CTL_BE_BLOCK_DEV: 2065 if (be_lun->backend.dev.csw) { 2066 dev_relthread(be_lun->backend.dev.cdev, 2067 be_lun->backend.dev.dev_ref); 2068 be_lun->backend.dev.csw = NULL; 2069 be_lun->backend.dev.cdev = NULL; 2070 } 2071 break; 2072 case CTL_BE_BLOCK_FILE: 2073 break; 2074 case CTL_BE_BLOCK_NONE: 2075 break; 2076 default: 2077 panic("Unexpected backend type."); 2078 break; 2079 } 2080 2081 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2082 be_lun->vn = NULL; 2083 2084 switch (be_lun->dev_type) { 2085 case CTL_BE_BLOCK_DEV: 2086 break; 2087 case CTL_BE_BLOCK_FILE: 2088 if (be_lun->backend.file.cred != NULL) { 2089 crfree(be_lun->backend.file.cred); 2090 be_lun->backend.file.cred = NULL; 2091 } 2092 break; 2093 case CTL_BE_BLOCK_NONE: 2094 break; 2095 default: 2096 panic("Unexpected backend type."); 2097 break; 2098 } 2099 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2100 } 2101 PICKUP_GIANT(); 2102 2103 return (0); 2104} 2105 2106static int 2107ctl_be_block_open(struct ctl_be_block_softc *softc, 2108 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2109{ 2110 struct nameidata nd; 2111 int flags; 2112 int error; 2113 2114 /* 2115 * XXX KDM allow a read-only option? 2116 */ 2117 flags = FREAD | FWRITE; 2118 error = 0; 2119 2120 if (rootvnode == NULL) { 2121 snprintf(req->error_str, sizeof(req->error_str), 2122 "Root filesystem is not mounted"); 2123 return (1); 2124 } 2125 2126 if (!curthread->td_proc->p_fd->fd_cdir) { 2127 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2128 VREF(rootvnode); 2129 } 2130 if (!curthread->td_proc->p_fd->fd_rdir) { 2131 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2132 VREF(rootvnode); 2133 } 2134 if (!curthread->td_proc->p_fd->fd_jdir) { 2135 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2136 VREF(rootvnode); 2137 } 2138 2139 again: 2140 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2141 error = vn_open(&nd, &flags, 0, NULL); 2142 if (error) { 2143 /* 2144 * This is the only reasonable guess we can make as far as 2145 * path if the user doesn't give us a fully qualified path. 2146 * If they want to specify a file, they need to specify the 2147 * full path. 2148 */ 2149 if (be_lun->dev_path[0] != '/') { 2150 char *dev_path = "/dev/"; 2151 char *dev_name; 2152 2153 /* Try adding device path at beginning of name */ 2154 dev_name = malloc(strlen(be_lun->dev_path) 2155 + strlen(dev_path) + 1, 2156 M_CTLBLK, M_WAITOK); 2157 if (dev_name) { 2158 sprintf(dev_name, "%s%s", dev_path, 2159 be_lun->dev_path); 2160 free(be_lun->dev_path, M_CTLBLK); 2161 be_lun->dev_path = dev_name; 2162 goto again; 2163 } 2164 } 2165 snprintf(req->error_str, sizeof(req->error_str), 2166 "error opening %s: %d", be_lun->dev_path, error); 2167 return (error); 2168 } 2169 2170 NDFREE(&nd, NDF_ONLY_PNBUF); 2171 2172 be_lun->vn = nd.ni_vp; 2173 2174 /* We only support disks and files. */ 2175 if (vn_isdisk(be_lun->vn, &error)) { 2176 error = ctl_be_block_open_dev(be_lun, req); 2177 } else if (be_lun->vn->v_type == VREG) { 2178 error = ctl_be_block_open_file(be_lun, req); 2179 } else { 2180 error = EINVAL; 2181 snprintf(req->error_str, sizeof(req->error_str), 2182 "%s is not a disk or plain file", be_lun->dev_path); 2183 } 2184 VOP_UNLOCK(be_lun->vn, 0); 2185 2186 if (error != 0) { 2187 ctl_be_block_close(be_lun); 2188 return (error); 2189 } 2190 2191 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2192 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2193 2194 return (0); 2195} 2196 2197static int 2198ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2199{ 2200 struct ctl_be_block_lun *be_lun; 2201 struct ctl_lun_create_params *params; 2202 char num_thread_str[16]; 2203 char tmpstr[32]; 2204 char *value; 2205 int retval, num_threads; 2206 int tmp_num_threads; 2207 2208 params = &req->reqdata.create; 2209 retval = 0; 2210 req->status = CTL_LUN_OK; 2211 2212 num_threads = cbb_num_threads; 2213 2214 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2215 2216 be_lun->params = req->reqdata.create; 2217 be_lun->softc = softc; 2218 STAILQ_INIT(&be_lun->input_queue); 2219 STAILQ_INIT(&be_lun->config_read_queue); 2220 STAILQ_INIT(&be_lun->config_write_queue); 2221 STAILQ_INIT(&be_lun->datamove_queue); 2222 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2223 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2224 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2225 ctl_init_opts(&be_lun->ctl_be_lun.options, 2226 req->num_be_args, req->kern_be_args); 2227 2228 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2229 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2230 2231 if (be_lun->lun_zone == NULL) { 2232 snprintf(req->error_str, sizeof(req->error_str), 2233 "error allocating UMA zone"); 2234 goto bailout_error; 2235 } 2236 2237 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2238 be_lun->ctl_be_lun.lun_type = params->device_type; 2239 else 2240 be_lun->ctl_be_lun.lun_type = T_DIRECT; 2241 2242 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 2243 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 2244 if (value == NULL) { 2245 snprintf(req->error_str, sizeof(req->error_str), 2246 "no file argument specified"); 2247 goto bailout_error; 2248 } 2249 be_lun->dev_path = strdup(value, M_CTLBLK); 2250 be_lun->blocksize = 512; 2251 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2252 2253 retval = ctl_be_block_open(softc, be_lun, req); 2254 if (retval != 0) { 2255 retval = 0; 2256 req->status = CTL_LUN_WARNING; 2257 } 2258 } else { 2259 /* 2260 * For processor devices, we don't have any size. 2261 */ 2262 be_lun->blocksize = 0; 2263 be_lun->pblockexp = 0; 2264 be_lun->pblockoff = 0; 2265 be_lun->ublockexp = 0; 2266 be_lun->ublockoff = 0; 2267 be_lun->size_blocks = 0; 2268 be_lun->size_bytes = 0; 2269 be_lun->ctl_be_lun.maxlba = 0; 2270 2271 /* 2272 * Default to just 1 thread for processor devices. 2273 */ 2274 num_threads = 1; 2275 } 2276 2277 /* 2278 * XXX This searching loop might be refactored to be combined with 2279 * the loop above, 2280 */ 2281 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2282 if (value != NULL) { 2283 tmp_num_threads = strtol(value, NULL, 0); 2284 2285 /* 2286 * We don't let the user specify less than one 2287 * thread, but hope he's clueful enough not to 2288 * specify 1000 threads. 2289 */ 2290 if (tmp_num_threads < 1) { 2291 snprintf(req->error_str, sizeof(req->error_str), 2292 "invalid number of threads %s", 2293 num_thread_str); 2294 goto bailout_error; 2295 } 2296 num_threads = tmp_num_threads; 2297 } 2298 2299 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2300 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2301 if (be_lun->vn == NULL) 2302 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2303 if (be_lun->unmap != NULL) 2304 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2305 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2306 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_SERSEQ_READ; 2307 be_lun->ctl_be_lun.be_lun = be_lun; 2308 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2309 0 : (be_lun->size_blocks - 1); 2310 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2311 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2312 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2313 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2314 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2315 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2316 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2317 /* Tell the user the blocksize we ended up using */ 2318 params->lun_size_bytes = be_lun->size_bytes; 2319 params->blocksize_bytes = be_lun->blocksize; 2320 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2321 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2322 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2323 } else 2324 be_lun->ctl_be_lun.req_lun_id = 0; 2325 2326 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2327 be_lun->ctl_be_lun.lun_config_status = 2328 ctl_be_block_lun_config_status; 2329 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2330 2331 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2332 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2333 softc->num_luns); 2334 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2335 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2336 sizeof(tmpstr))); 2337 2338 /* Tell the user what we used for a serial number */ 2339 strncpy((char *)params->serial_num, tmpstr, 2340 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2341 } else { 2342 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2343 params->serial_num, 2344 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2345 sizeof(params->serial_num))); 2346 } 2347 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2348 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2349 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2350 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2351 sizeof(tmpstr))); 2352 2353 /* Tell the user what we used for a device ID */ 2354 strncpy((char *)params->device_id, tmpstr, 2355 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2356 } else { 2357 strncpy((char *)be_lun->ctl_be_lun.device_id, 2358 params->device_id, 2359 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2360 sizeof(params->device_id))); 2361 } 2362 2363 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2364 2365 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2366 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2367 2368 if (be_lun->io_taskqueue == NULL) { 2369 snprintf(req->error_str, sizeof(req->error_str), 2370 "unable to create taskqueue"); 2371 goto bailout_error; 2372 } 2373 2374 /* 2375 * Note that we start the same number of threads by default for 2376 * both the file case and the block device case. For the file 2377 * case, we need multiple threads to allow concurrency, because the 2378 * vnode interface is designed to be a blocking interface. For the 2379 * block device case, ZFS zvols at least will block the caller's 2380 * context in many instances, and so we need multiple threads to 2381 * overcome that problem. Other block devices don't need as many 2382 * threads, but they shouldn't cause too many problems. 2383 * 2384 * If the user wants to just have a single thread for a block 2385 * device, he can specify that when the LUN is created, or change 2386 * the tunable/sysctl to alter the default number of threads. 2387 */ 2388 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2389 /*num threads*/num_threads, 2390 /*priority*/PWAIT, 2391 /*thread name*/ 2392 "%s taskq", be_lun->lunname); 2393 2394 if (retval != 0) 2395 goto bailout_error; 2396 2397 be_lun->num_threads = num_threads; 2398 2399 mtx_lock(&softc->lock); 2400 softc->num_luns++; 2401 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2402 2403 mtx_unlock(&softc->lock); 2404 2405 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2406 if (retval != 0) { 2407 mtx_lock(&softc->lock); 2408 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2409 links); 2410 softc->num_luns--; 2411 mtx_unlock(&softc->lock); 2412 snprintf(req->error_str, sizeof(req->error_str), 2413 "ctl_add_lun() returned error %d, see dmesg for " 2414 "details", retval); 2415 retval = 0; 2416 goto bailout_error; 2417 } 2418 2419 mtx_lock(&softc->lock); 2420 2421 /* 2422 * Tell the config_status routine that we're waiting so it won't 2423 * clean up the LUN in the event of an error. 2424 */ 2425 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2426 2427 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2428 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2429 if (retval == EINTR) 2430 break; 2431 } 2432 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2433 2434 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2435 snprintf(req->error_str, sizeof(req->error_str), 2436 "LUN configuration error, see dmesg for details"); 2437 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2438 links); 2439 softc->num_luns--; 2440 mtx_unlock(&softc->lock); 2441 goto bailout_error; 2442 } else { 2443 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2444 } 2445 2446 mtx_unlock(&softc->lock); 2447 2448 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2449 be_lun->blocksize, 2450 DEVSTAT_ALL_SUPPORTED, 2451 be_lun->ctl_be_lun.lun_type 2452 | DEVSTAT_TYPE_IF_OTHER, 2453 DEVSTAT_PRIORITY_OTHER); 2454 2455 return (retval); 2456 2457bailout_error: 2458 req->status = CTL_LUN_ERROR; 2459 2460 if (be_lun->io_taskqueue != NULL) 2461 taskqueue_free(be_lun->io_taskqueue); 2462 ctl_be_block_close(be_lun); 2463 if (be_lun->dev_path != NULL) 2464 free(be_lun->dev_path, M_CTLBLK); 2465 if (be_lun->lun_zone != NULL) 2466 uma_zdestroy(be_lun->lun_zone); 2467 ctl_free_opts(&be_lun->ctl_be_lun.options); 2468 mtx_destroy(&be_lun->queue_lock); 2469 mtx_destroy(&be_lun->io_lock); 2470 free(be_lun, M_CTLBLK); 2471 2472 return (retval); 2473} 2474 2475static int 2476ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2477{ 2478 struct ctl_lun_rm_params *params; 2479 struct ctl_be_block_lun *be_lun; 2480 int retval; 2481 2482 params = &req->reqdata.rm; 2483 2484 mtx_lock(&softc->lock); 2485 2486 be_lun = NULL; 2487 2488 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2489 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2490 break; 2491 } 2492 mtx_unlock(&softc->lock); 2493 2494 if (be_lun == NULL) { 2495 snprintf(req->error_str, sizeof(req->error_str), 2496 "LUN %u is not managed by the block backend", 2497 params->lun_id); 2498 goto bailout_error; 2499 } 2500 2501 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2502 2503 if (retval != 0) { 2504 snprintf(req->error_str, sizeof(req->error_str), 2505 "error %d returned from ctl_disable_lun() for " 2506 "LUN %d", retval, params->lun_id); 2507 goto bailout_error; 2508 2509 } 2510 2511 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2512 if (retval != 0) { 2513 snprintf(req->error_str, sizeof(req->error_str), 2514 "error %d returned from ctl_invalidate_lun() for " 2515 "LUN %d", retval, params->lun_id); 2516 goto bailout_error; 2517 } 2518 2519 mtx_lock(&softc->lock); 2520 2521 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2522 2523 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2524 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2525 if (retval == EINTR) 2526 break; 2527 } 2528 2529 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2530 2531 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2532 snprintf(req->error_str, sizeof(req->error_str), 2533 "interrupted waiting for LUN to be freed"); 2534 mtx_unlock(&softc->lock); 2535 goto bailout_error; 2536 } 2537 2538 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2539 2540 softc->num_luns--; 2541 mtx_unlock(&softc->lock); 2542 2543 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2544 2545 taskqueue_free(be_lun->io_taskqueue); 2546 2547 ctl_be_block_close(be_lun); 2548 2549 if (be_lun->disk_stats != NULL) 2550 devstat_remove_entry(be_lun->disk_stats); 2551 2552 uma_zdestroy(be_lun->lun_zone); 2553 2554 ctl_free_opts(&be_lun->ctl_be_lun.options); 2555 free(be_lun->dev_path, M_CTLBLK); 2556 mtx_destroy(&be_lun->queue_lock); 2557 mtx_destroy(&be_lun->io_lock); 2558 free(be_lun, M_CTLBLK); 2559 2560 req->status = CTL_LUN_OK; 2561 2562 return (0); 2563 2564bailout_error: 2565 2566 req->status = CTL_LUN_ERROR; 2567 2568 return (0); 2569} 2570 2571static int 2572ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2573 struct ctl_lun_req *req) 2574{ 2575 struct vattr vattr; 2576 int error; 2577 struct ctl_lun_create_params *params = &be_lun->params; 2578 2579 if (params->lun_size_bytes != 0) { 2580 be_lun->size_bytes = params->lun_size_bytes; 2581 } else { 2582 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2583 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2584 VOP_UNLOCK(be_lun->vn, 0); 2585 if (error != 0) { 2586 snprintf(req->error_str, sizeof(req->error_str), 2587 "error calling VOP_GETATTR() for file %s", 2588 be_lun->dev_path); 2589 return (error); 2590 } 2591 2592 be_lun->size_bytes = vattr.va_size; 2593 } 2594 2595 return (0); 2596} 2597 2598static int 2599ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2600 struct ctl_lun_req *req) 2601{ 2602 struct ctl_be_block_devdata *dev_data; 2603 int error; 2604 struct ctl_lun_create_params *params = &be_lun->params; 2605 uint64_t size_bytes; 2606 2607 dev_data = &be_lun->backend.dev; 2608 if (!dev_data->csw->d_ioctl) { 2609 snprintf(req->error_str, sizeof(req->error_str), 2610 "no d_ioctl for device %s!", be_lun->dev_path); 2611 return (ENODEV); 2612 } 2613 2614 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2615 (caddr_t)&size_bytes, FREAD, 2616 curthread); 2617 if (error) { 2618 snprintf(req->error_str, sizeof(req->error_str), 2619 "error %d returned for DIOCGMEDIASIZE ioctl " 2620 "on %s!", error, be_lun->dev_path); 2621 return (error); 2622 } 2623 2624 if (params->lun_size_bytes != 0) { 2625 if (params->lun_size_bytes > size_bytes) { 2626 snprintf(req->error_str, sizeof(req->error_str), 2627 "requested LUN size %ju > backing device " 2628 "size %ju", 2629 (uintmax_t)params->lun_size_bytes, 2630 (uintmax_t)size_bytes); 2631 return (EINVAL); 2632 } 2633 2634 be_lun->size_bytes = params->lun_size_bytes; 2635 } else { 2636 be_lun->size_bytes = size_bytes; 2637 } 2638 2639 return (0); 2640} 2641 2642static int 2643ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2644{ 2645 struct ctl_lun_modify_params *params; 2646 struct ctl_be_block_lun *be_lun; 2647 uint64_t oldsize; 2648 int error; 2649 2650 params = &req->reqdata.modify; 2651 2652 mtx_lock(&softc->lock); 2653 be_lun = NULL; 2654 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2655 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2656 break; 2657 } 2658 mtx_unlock(&softc->lock); 2659 2660 if (be_lun == NULL) { 2661 snprintf(req->error_str, sizeof(req->error_str), 2662 "LUN %u is not managed by the block backend", 2663 params->lun_id); 2664 goto bailout_error; 2665 } 2666 2667 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2668 2669 oldsize = be_lun->size_bytes; 2670 if (be_lun->vn == NULL) 2671 error = ctl_be_block_open(softc, be_lun, req); 2672 else if (be_lun->vn->v_type == VREG) 2673 error = ctl_be_block_modify_file(be_lun, req); 2674 else 2675 error = ctl_be_block_modify_dev(be_lun, req); 2676 2677 if (error == 0 && be_lun->size_bytes != oldsize) { 2678 be_lun->size_blocks = be_lun->size_bytes >> 2679 be_lun->blocksize_shift; 2680 2681 /* 2682 * The maximum LBA is the size - 1. 2683 * 2684 * XXX: Note that this field is being updated without locking, 2685 * which might cause problems on 32-bit architectures. 2686 */ 2687 if (be_lun->unmap != NULL) 2688 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2689 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2690 0 : (be_lun->size_blocks - 1); 2691 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2692 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2693 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2694 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2695 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2696 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2697 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2698 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2699 if (oldsize == 0 && be_lun->size_blocks != 0) 2700 ctl_lun_online(&be_lun->ctl_be_lun); 2701 } 2702 2703 /* Tell the user the exact size we ended up using */ 2704 params->lun_size_bytes = be_lun->size_bytes; 2705 2706 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2707 2708 return (0); 2709 2710bailout_error: 2711 req->status = CTL_LUN_ERROR; 2712 2713 return (0); 2714} 2715 2716static void 2717ctl_be_block_lun_shutdown(void *be_lun) 2718{ 2719 struct ctl_be_block_lun *lun; 2720 struct ctl_be_block_softc *softc; 2721 2722 lun = (struct ctl_be_block_lun *)be_lun; 2723 2724 softc = lun->softc; 2725 2726 mtx_lock(&softc->lock); 2727 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2728 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2729 wakeup(lun); 2730 mtx_unlock(&softc->lock); 2731 2732} 2733 2734static void 2735ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2736{ 2737 struct ctl_be_block_lun *lun; 2738 struct ctl_be_block_softc *softc; 2739 2740 lun = (struct ctl_be_block_lun *)be_lun; 2741 softc = lun->softc; 2742 2743 if (status == CTL_LUN_CONFIG_OK) { 2744 mtx_lock(&softc->lock); 2745 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2746 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2747 wakeup(lun); 2748 mtx_unlock(&softc->lock); 2749 2750 /* 2751 * We successfully added the LUN, attempt to enable it. 2752 */ 2753 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2754 printf("%s: ctl_enable_lun() failed!\n", __func__); 2755 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2756 printf("%s: ctl_invalidate_lun() failed!\n", 2757 __func__); 2758 } 2759 } 2760 2761 return; 2762 } 2763 2764 2765 mtx_lock(&softc->lock); 2766 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2767 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2768 wakeup(lun); 2769 mtx_unlock(&softc->lock); 2770} 2771 2772 2773static int 2774ctl_be_block_config_write(union ctl_io *io) 2775{ 2776 struct ctl_be_block_lun *be_lun; 2777 struct ctl_be_lun *ctl_be_lun; 2778 int retval; 2779 2780 retval = 0; 2781 2782 DPRINTF("entered\n"); 2783 2784 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2785 CTL_PRIV_BACKEND_LUN].ptr; 2786 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2787 2788 switch (io->scsiio.cdb[0]) { 2789 case SYNCHRONIZE_CACHE: 2790 case SYNCHRONIZE_CACHE_16: 2791 case WRITE_SAME_10: 2792 case WRITE_SAME_16: 2793 case UNMAP: 2794 /* 2795 * The upper level CTL code will filter out any CDBs with 2796 * the immediate bit set and return the proper error. 2797 * 2798 * We don't really need to worry about what LBA range the 2799 * user asked to be synced out. When they issue a sync 2800 * cache command, we'll sync out the whole thing. 2801 */ 2802 mtx_lock(&be_lun->queue_lock); 2803 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2804 links); 2805 mtx_unlock(&be_lun->queue_lock); 2806 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2807 break; 2808 case START_STOP_UNIT: { 2809 struct scsi_start_stop_unit *cdb; 2810 2811 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2812 2813 if (cdb->how & SSS_START) 2814 retval = ctl_start_lun(ctl_be_lun); 2815 else { 2816 retval = ctl_stop_lun(ctl_be_lun); 2817 /* 2818 * XXX KDM Copan-specific offline behavior. 2819 * Figure out a reasonable way to port this? 2820 */ 2821#ifdef NEEDTOPORT 2822 if ((retval == 0) 2823 && (cdb->byte2 & SSS_ONOFFLINE)) 2824 retval = ctl_lun_offline(ctl_be_lun); 2825#endif 2826 } 2827 2828 /* 2829 * In general, the above routines should not fail. They 2830 * just set state for the LUN. So we've got something 2831 * pretty wrong here if we can't start or stop the LUN. 2832 */ 2833 if (retval != 0) { 2834 ctl_set_internal_failure(&io->scsiio, 2835 /*sks_valid*/ 1, 2836 /*retry_count*/ 0xf051); 2837 retval = CTL_RETVAL_COMPLETE; 2838 } else { 2839 ctl_set_success(&io->scsiio); 2840 } 2841 ctl_config_write_done(io); 2842 break; 2843 } 2844 default: 2845 ctl_set_invalid_opcode(&io->scsiio); 2846 ctl_config_write_done(io); 2847 retval = CTL_RETVAL_COMPLETE; 2848 break; 2849 } 2850 2851 return (retval); 2852} 2853 2854static int 2855ctl_be_block_config_read(union ctl_io *io) 2856{ 2857 struct ctl_be_block_lun *be_lun; 2858 struct ctl_be_lun *ctl_be_lun; 2859 int retval = 0; 2860 2861 DPRINTF("entered\n"); 2862 2863 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2864 CTL_PRIV_BACKEND_LUN].ptr; 2865 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2866 2867 switch (io->scsiio.cdb[0]) { 2868 case SERVICE_ACTION_IN: 2869 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2870 mtx_lock(&be_lun->queue_lock); 2871 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2872 &io->io_hdr, links); 2873 mtx_unlock(&be_lun->queue_lock); 2874 taskqueue_enqueue(be_lun->io_taskqueue, 2875 &be_lun->io_task); 2876 retval = CTL_RETVAL_QUEUED; 2877 break; 2878 } 2879 ctl_set_invalid_field(&io->scsiio, 2880 /*sks_valid*/ 1, 2881 /*command*/ 1, 2882 /*field*/ 1, 2883 /*bit_valid*/ 1, 2884 /*bit*/ 4); 2885 ctl_config_read_done(io); 2886 retval = CTL_RETVAL_COMPLETE; 2887 break; 2888 default: 2889 ctl_set_invalid_opcode(&io->scsiio); 2890 ctl_config_read_done(io); 2891 retval = CTL_RETVAL_COMPLETE; 2892 break; 2893 } 2894 2895 return (retval); 2896} 2897 2898static int 2899ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2900{ 2901 struct ctl_be_block_lun *lun; 2902 int retval; 2903 2904 lun = (struct ctl_be_block_lun *)be_lun; 2905 retval = 0; 2906 2907 retval = sbuf_printf(sb, "\t<num_threads>"); 2908 2909 if (retval != 0) 2910 goto bailout; 2911 2912 retval = sbuf_printf(sb, "%d", lun->num_threads); 2913 2914 if (retval != 0) 2915 goto bailout; 2916 2917 retval = sbuf_printf(sb, "</num_threads>\n"); 2918 2919bailout: 2920 2921 return (retval); 2922} 2923 2924static uint64_t 2925ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2926{ 2927 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2928 2929 if (lun->getattr == NULL) 2930 return (UINT64_MAX); 2931 return (lun->getattr(lun, attrname)); 2932} 2933 2934int 2935ctl_be_block_init(void) 2936{ 2937 struct ctl_be_block_softc *softc; 2938 int retval; 2939 2940 softc = &backend_block_softc; 2941 retval = 0; 2942 2943 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2944 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2945 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2946 STAILQ_INIT(&softc->disk_list); 2947 STAILQ_INIT(&softc->lun_list); 2948 2949 return (retval); 2950} 2951