ctl_backend_block.c revision 278672
1/*- 2 * Copyright (c) 2003 Silicon Graphics International Corp. 3 * Copyright (c) 2009-2011 Spectra Logic Corporation 4 * Copyright (c) 2012 The FreeBSD Foundation 5 * All rights reserved. 6 * 7 * Portions of this software were developed by Edward Tomasz Napierala 8 * under sponsorship from the FreeBSD Foundation. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions, and the following disclaimer, 15 * without modification. 16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17 * substantially similar to the "NO WARRANTY" disclaimer below 18 * ("Disclaimer") and any redistribution must be conditioned upon 19 * including a substantially similar Disclaimer requirement for further 20 * binary redistribution. 21 * 22 * NO WARRANTY 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGES. 34 * 35 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36 */ 37/* 38 * CAM Target Layer driver backend for block devices. 39 * 40 * Author: Ken Merry <ken@FreeBSD.org> 41 */ 42#include <sys/cdefs.h> 43__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl_backend_block.c 278672 2015-02-13 13:26:23Z mav $"); 44 45#include <sys/param.h> 46#include <sys/systm.h> 47#include <sys/kernel.h> 48#include <sys/types.h> 49#include <sys/kthread.h> 50#include <sys/bio.h> 51#include <sys/fcntl.h> 52#include <sys/limits.h> 53#include <sys/lock.h> 54#include <sys/mutex.h> 55#include <sys/condvar.h> 56#include <sys/malloc.h> 57#include <sys/conf.h> 58#include <sys/ioccom.h> 59#include <sys/queue.h> 60#include <sys/sbuf.h> 61#include <sys/endian.h> 62#include <sys/uio.h> 63#include <sys/buf.h> 64#include <sys/taskqueue.h> 65#include <sys/vnode.h> 66#include <sys/namei.h> 67#include <sys/mount.h> 68#include <sys/disk.h> 69#include <sys/fcntl.h> 70#include <sys/filedesc.h> 71#include <sys/filio.h> 72#include <sys/proc.h> 73#include <sys/pcpu.h> 74#include <sys/module.h> 75#include <sys/sdt.h> 76#include <sys/devicestat.h> 77#include <sys/sysctl.h> 78 79#include <geom/geom.h> 80 81#include <cam/cam.h> 82#include <cam/scsi/scsi_all.h> 83#include <cam/scsi/scsi_da.h> 84#include <cam/ctl/ctl_io.h> 85#include <cam/ctl/ctl.h> 86#include <cam/ctl/ctl_backend.h> 87#include <cam/ctl/ctl_frontend_internal.h> 88#include <cam/ctl/ctl_ioctl.h> 89#include <cam/ctl/ctl_scsi_all.h> 90#include <cam/ctl/ctl_error.h> 91 92/* 93 * The idea here is that we'll allocate enough S/G space to hold a 1MB 94 * I/O. If we get an I/O larger than that, we'll split it. 95 */ 96#define CTLBLK_HALF_IO_SIZE (512 * 1024) 97#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 98#define CTLBLK_MAX_SEG MAXPHYS 99#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 100#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 101 102#ifdef CTLBLK_DEBUG 103#define DPRINTF(fmt, args...) \ 104 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 105#else 106#define DPRINTF(fmt, args...) do {} while(0) 107#endif 108 109#define PRIV(io) \ 110 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 111#define ARGS(io) \ 112 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 113 114SDT_PROVIDER_DEFINE(cbb); 115 116typedef enum { 117 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 118 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 119 CTL_BE_BLOCK_LUN_WAITING = 0x04, 120 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 121} ctl_be_block_lun_flags; 122 123typedef enum { 124 CTL_BE_BLOCK_NONE, 125 CTL_BE_BLOCK_DEV, 126 CTL_BE_BLOCK_FILE 127} ctl_be_block_type; 128 129struct ctl_be_block_devdata { 130 struct cdev *cdev; 131 struct cdevsw *csw; 132 int dev_ref; 133}; 134 135struct ctl_be_block_filedata { 136 struct ucred *cred; 137}; 138 139union ctl_be_block_bedata { 140 struct ctl_be_block_devdata dev; 141 struct ctl_be_block_filedata file; 142}; 143 144struct ctl_be_block_io; 145struct ctl_be_block_lun; 146 147typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 148 struct ctl_be_block_io *beio); 149typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 150 const char *attrname); 151 152/* 153 * Backend LUN structure. There is a 1:1 mapping between a block device 154 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 155 */ 156struct ctl_be_block_lun { 157 struct ctl_lun_create_params params; 158 struct ctl_block_disk *disk; 159 char lunname[32]; 160 char *dev_path; 161 ctl_be_block_type dev_type; 162 struct vnode *vn; 163 union ctl_be_block_bedata backend; 164 cbb_dispatch_t dispatch; 165 cbb_dispatch_t lun_flush; 166 cbb_dispatch_t unmap; 167 cbb_dispatch_t get_lba_status; 168 cbb_getattr_t getattr; 169 uma_zone_t lun_zone; 170 uint64_t size_blocks; 171 uint64_t size_bytes; 172 uint32_t blocksize; 173 int blocksize_shift; 174 uint16_t pblockexp; 175 uint16_t pblockoff; 176 uint16_t ublockexp; 177 uint16_t ublockoff; 178 uint32_t atomicblock; 179 uint32_t opttxferlen; 180 struct ctl_be_block_softc *softc; 181 struct devstat *disk_stats; 182 ctl_be_block_lun_flags flags; 183 STAILQ_ENTRY(ctl_be_block_lun) links; 184 struct ctl_be_lun ctl_be_lun; 185 struct taskqueue *io_taskqueue; 186 struct task io_task; 187 int num_threads; 188 STAILQ_HEAD(, ctl_io_hdr) input_queue; 189 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 190 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 191 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 192 struct mtx_padalign io_lock; 193 struct mtx_padalign queue_lock; 194}; 195 196/* 197 * Overall softc structure for the block backend module. 198 */ 199struct ctl_be_block_softc { 200 struct mtx lock; 201 int num_disks; 202 STAILQ_HEAD(, ctl_block_disk) disk_list; 203 int num_luns; 204 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 205}; 206 207static struct ctl_be_block_softc backend_block_softc; 208 209/* 210 * Per-I/O information. 211 */ 212struct ctl_be_block_io { 213 union ctl_io *io; 214 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 215 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 216 int bio_cmd; 217 int num_segs; 218 int num_bios_sent; 219 int num_bios_done; 220 int send_complete; 221 int num_errors; 222 struct bintime ds_t0; 223 devstat_tag_type ds_tag_type; 224 devstat_trans_flags ds_trans_type; 225 uint64_t io_len; 226 uint64_t io_offset; 227 struct ctl_be_block_softc *softc; 228 struct ctl_be_block_lun *lun; 229 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 230}; 231 232static int cbb_num_threads = 14; 233SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 234 "CAM Target Layer Block Backend"); 235SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 236 &cbb_num_threads, 0, "Number of threads per backing file"); 237 238static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 239static void ctl_free_beio(struct ctl_be_block_io *beio); 240static void ctl_complete_beio(struct ctl_be_block_io *beio); 241static int ctl_be_block_move_done(union ctl_io *io); 242static void ctl_be_block_biodone(struct bio *bio); 243static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 244 struct ctl_be_block_io *beio); 245static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 246 struct ctl_be_block_io *beio); 247static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 248 struct ctl_be_block_io *beio); 249static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 250 const char *attrname); 251static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 252 struct ctl_be_block_io *beio); 253static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 254 struct ctl_be_block_io *beio); 255static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 256 struct ctl_be_block_io *beio); 257static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 258 const char *attrname); 259static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 260 union ctl_io *io); 261static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 262 union ctl_io *io); 263static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 264 union ctl_io *io); 265static void ctl_be_block_worker(void *context, int pending); 266static int ctl_be_block_submit(union ctl_io *io); 267static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 268 int flag, struct thread *td); 269static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 270 struct ctl_lun_req *req); 271static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 272 struct ctl_lun_req *req); 273static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 274static int ctl_be_block_open(struct ctl_be_block_softc *softc, 275 struct ctl_be_block_lun *be_lun, 276 struct ctl_lun_req *req); 277static int ctl_be_block_create(struct ctl_be_block_softc *softc, 278 struct ctl_lun_req *req); 279static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 280 struct ctl_lun_req *req); 281static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 282 struct ctl_lun_req *req); 283static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 284 struct ctl_lun_req *req); 285static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 286 struct ctl_lun_req *req); 287static void ctl_be_block_lun_shutdown(void *be_lun); 288static void ctl_be_block_lun_config_status(void *be_lun, 289 ctl_lun_config_status status); 290static int ctl_be_block_config_write(union ctl_io *io); 291static int ctl_be_block_config_read(union ctl_io *io); 292static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 293static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 294int ctl_be_block_init(void); 295 296static struct ctl_backend_driver ctl_be_block_driver = 297{ 298 .name = "block", 299 .flags = CTL_BE_FLAG_HAS_CONFIG, 300 .init = ctl_be_block_init, 301 .data_submit = ctl_be_block_submit, 302 .data_move_done = ctl_be_block_move_done, 303 .config_read = ctl_be_block_config_read, 304 .config_write = ctl_be_block_config_write, 305 .ioctl = ctl_be_block_ioctl, 306 .lun_info = ctl_be_block_lun_info, 307 .lun_attr = ctl_be_block_lun_attr 308}; 309 310MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 311CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 312 313static uma_zone_t beio_zone; 314 315static struct ctl_be_block_io * 316ctl_alloc_beio(struct ctl_be_block_softc *softc) 317{ 318 struct ctl_be_block_io *beio; 319 320 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 321 beio->softc = softc; 322 return (beio); 323} 324 325static void 326ctl_free_beio(struct ctl_be_block_io *beio) 327{ 328 int duplicate_free; 329 int i; 330 331 duplicate_free = 0; 332 333 for (i = 0; i < beio->num_segs; i++) { 334 if (beio->sg_segs[i].addr == NULL) 335 duplicate_free++; 336 337 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 338 beio->sg_segs[i].addr = NULL; 339 340 /* For compare we had two equal S/G lists. */ 341 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 342 uma_zfree(beio->lun->lun_zone, 343 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 344 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 345 } 346 } 347 348 if (duplicate_free > 0) { 349 printf("%s: %d duplicate frees out of %d segments\n", __func__, 350 duplicate_free, beio->num_segs); 351 } 352 353 uma_zfree(beio_zone, beio); 354} 355 356static void 357ctl_complete_beio(struct ctl_be_block_io *beio) 358{ 359 union ctl_io *io = beio->io; 360 361 if (beio->beio_cont != NULL) { 362 beio->beio_cont(beio); 363 } else { 364 ctl_free_beio(beio); 365 ctl_data_submit_done(io); 366 } 367} 368 369static int 370ctl_be_block_move_done(union ctl_io *io) 371{ 372 struct ctl_be_block_io *beio; 373 struct ctl_be_block_lun *be_lun; 374 struct ctl_lba_len_flags *lbalen; 375#ifdef CTL_TIME_IO 376 struct bintime cur_bt; 377#endif 378 int i; 379 380 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 381 be_lun = beio->lun; 382 383 DPRINTF("entered\n"); 384 385#ifdef CTL_TIME_IO 386 getbintime(&cur_bt); 387 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 388 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 389 io->io_hdr.num_dmas++; 390#endif 391 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 392 393 /* 394 * We set status at this point for read commands, and write 395 * commands with errors. 396 */ 397 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 398 ; 399 } else if ((io->io_hdr.port_status == 0) && 400 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 401 lbalen = ARGS(beio->io); 402 if (lbalen->flags & CTL_LLF_READ) { 403 ctl_set_success(&io->scsiio); 404 } else if (lbalen->flags & CTL_LLF_COMPARE) { 405 /* We have two data blocks ready for comparison. */ 406 for (i = 0; i < beio->num_segs; i++) { 407 if (memcmp(beio->sg_segs[i].addr, 408 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 409 beio->sg_segs[i].len) != 0) 410 break; 411 } 412 if (i < beio->num_segs) 413 ctl_set_sense(&io->scsiio, 414 /*current_error*/ 1, 415 /*sense_key*/ SSD_KEY_MISCOMPARE, 416 /*asc*/ 0x1D, 417 /*ascq*/ 0x00, 418 SSD_ELEM_NONE); 419 else 420 ctl_set_success(&io->scsiio); 421 } 422 } else if ((io->io_hdr.port_status != 0) && 423 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 424 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 425 /* 426 * For hardware error sense keys, the sense key 427 * specific value is defined to be a retry count, 428 * but we use it to pass back an internal FETD 429 * error code. XXX KDM Hopefully the FETD is only 430 * using 16 bits for an error code, since that's 431 * all the space we have in the sks field. 432 */ 433 ctl_set_internal_failure(&io->scsiio, 434 /*sks_valid*/ 1, 435 /*retry_count*/ 436 io->io_hdr.port_status); 437 } 438 439 /* 440 * If this is a read, or a write with errors, it is done. 441 */ 442 if ((beio->bio_cmd == BIO_READ) 443 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 444 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 445 ctl_complete_beio(beio); 446 return (0); 447 } 448 449 /* 450 * At this point, we have a write and the DMA completed 451 * successfully. We now have to queue it to the task queue to 452 * execute the backend I/O. That is because we do blocking 453 * memory allocations, and in the file backing case, blocking I/O. 454 * This move done routine is generally called in the SIM's 455 * interrupt context, and therefore we cannot block. 456 */ 457 mtx_lock(&be_lun->queue_lock); 458 /* 459 * XXX KDM make sure that links is okay to use at this point. 460 * Otherwise, we either need to add another field to ctl_io_hdr, 461 * or deal with resource allocation here. 462 */ 463 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 464 mtx_unlock(&be_lun->queue_lock); 465 466 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 467 468 return (0); 469} 470 471static void 472ctl_be_block_biodone(struct bio *bio) 473{ 474 struct ctl_be_block_io *beio; 475 struct ctl_be_block_lun *be_lun; 476 union ctl_io *io; 477 int error; 478 479 beio = bio->bio_caller1; 480 be_lun = beio->lun; 481 io = beio->io; 482 483 DPRINTF("entered\n"); 484 485 error = bio->bio_error; 486 mtx_lock(&be_lun->io_lock); 487 if (error != 0) 488 beio->num_errors++; 489 490 beio->num_bios_done++; 491 492 /* 493 * XXX KDM will this cause WITNESS to complain? Holding a lock 494 * during the free might cause it to complain. 495 */ 496 g_destroy_bio(bio); 497 498 /* 499 * If the send complete bit isn't set, or we aren't the last I/O to 500 * complete, then we're done. 501 */ 502 if ((beio->send_complete == 0) 503 || (beio->num_bios_done < beio->num_bios_sent)) { 504 mtx_unlock(&be_lun->io_lock); 505 return; 506 } 507 508 /* 509 * At this point, we've verified that we are the last I/O to 510 * complete, so it's safe to drop the lock. 511 */ 512 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 513 beio->ds_tag_type, beio->ds_trans_type, 514 /*now*/ NULL, /*then*/&beio->ds_t0); 515 mtx_unlock(&be_lun->io_lock); 516 517 /* 518 * If there are any errors from the backing device, we fail the 519 * entire I/O with a medium error. 520 */ 521 if (beio->num_errors > 0) { 522 if (error == EOPNOTSUPP) { 523 ctl_set_invalid_opcode(&io->scsiio); 524 } else if (error == ENOSPC) { 525 ctl_set_space_alloc_fail(&io->scsiio); 526 } else if (beio->bio_cmd == BIO_FLUSH) { 527 /* XXX KDM is there is a better error here? */ 528 ctl_set_internal_failure(&io->scsiio, 529 /*sks_valid*/ 1, 530 /*retry_count*/ 0xbad2); 531 } else 532 ctl_set_medium_error(&io->scsiio); 533 ctl_complete_beio(beio); 534 return; 535 } 536 537 /* 538 * If this is a write, a flush, a delete or verify, we're all done. 539 * If this is a read, we can now send the data to the user. 540 */ 541 if ((beio->bio_cmd == BIO_WRITE) 542 || (beio->bio_cmd == BIO_FLUSH) 543 || (beio->bio_cmd == BIO_DELETE) 544 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 545 ctl_set_success(&io->scsiio); 546 ctl_complete_beio(beio); 547 } else { 548 if ((ARGS(io)->flags & CTL_LLF_READ) && 549 beio->beio_cont == NULL) 550 ctl_set_success(&io->scsiio); 551#ifdef CTL_TIME_IO 552 getbintime(&io->io_hdr.dma_start_bt); 553#endif 554 ctl_datamove(io); 555 } 556} 557 558static void 559ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 560 struct ctl_be_block_io *beio) 561{ 562 union ctl_io *io = beio->io; 563 struct mount *mountpoint; 564 int error, lock_flags; 565 566 DPRINTF("entered\n"); 567 568 binuptime(&beio->ds_t0); 569 mtx_lock(&be_lun->io_lock); 570 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 571 mtx_unlock(&be_lun->io_lock); 572 573 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 574 575 if (MNT_SHARED_WRITES(mountpoint) 576 || ((mountpoint == NULL) 577 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 578 lock_flags = LK_SHARED; 579 else 580 lock_flags = LK_EXCLUSIVE; 581 582 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 583 584 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 585 VOP_UNLOCK(be_lun->vn, 0); 586 587 vn_finished_write(mountpoint); 588 589 mtx_lock(&be_lun->io_lock); 590 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 591 beio->ds_tag_type, beio->ds_trans_type, 592 /*now*/ NULL, /*then*/&beio->ds_t0); 593 mtx_unlock(&be_lun->io_lock); 594 595 if (error == 0) 596 ctl_set_success(&io->scsiio); 597 else { 598 /* XXX KDM is there is a better error here? */ 599 ctl_set_internal_failure(&io->scsiio, 600 /*sks_valid*/ 1, 601 /*retry_count*/ 0xbad1); 602 } 603 604 ctl_complete_beio(beio); 605} 606 607SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 608SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 609SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 610SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 611 612static void 613ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 614 struct ctl_be_block_io *beio) 615{ 616 struct ctl_be_block_filedata *file_data; 617 union ctl_io *io; 618 struct uio xuio; 619 struct iovec *xiovec; 620 int flags; 621 int error, i; 622 623 DPRINTF("entered\n"); 624 625 file_data = &be_lun->backend.file; 626 io = beio->io; 627 flags = 0; 628 if (ARGS(io)->flags & CTL_LLF_DPO) 629 flags |= IO_DIRECT; 630 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 631 flags |= IO_SYNC; 632 633 bzero(&xuio, sizeof(xuio)); 634 if (beio->bio_cmd == BIO_READ) { 635 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 636 xuio.uio_rw = UIO_READ; 637 } else { 638 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 639 xuio.uio_rw = UIO_WRITE; 640 } 641 xuio.uio_offset = beio->io_offset; 642 xuio.uio_resid = beio->io_len; 643 xuio.uio_segflg = UIO_SYSSPACE; 644 xuio.uio_iov = beio->xiovecs; 645 xuio.uio_iovcnt = beio->num_segs; 646 xuio.uio_td = curthread; 647 648 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 649 xiovec->iov_base = beio->sg_segs[i].addr; 650 xiovec->iov_len = beio->sg_segs[i].len; 651 } 652 653 binuptime(&beio->ds_t0); 654 mtx_lock(&be_lun->io_lock); 655 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 656 mtx_unlock(&be_lun->io_lock); 657 658 if (beio->bio_cmd == BIO_READ) { 659 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 660 661 /* 662 * UFS pays attention to IO_DIRECT for reads. If the 663 * DIRECTIO option is configured into the kernel, it calls 664 * ffs_rawread(). But that only works for single-segment 665 * uios with user space addresses. In our case, with a 666 * kernel uio, it still reads into the buffer cache, but it 667 * will just try to release the buffer from the cache later 668 * on in ffs_read(). 669 * 670 * ZFS does not pay attention to IO_DIRECT for reads. 671 * 672 * UFS does not pay attention to IO_SYNC for reads. 673 * 674 * ZFS pays attention to IO_SYNC (which translates into the 675 * Solaris define FRSYNC for zfs_read()) for reads. It 676 * attempts to sync the file before reading. 677 * 678 * So, to attempt to provide some barrier semantics in the 679 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 680 */ 681 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 682 683 VOP_UNLOCK(be_lun->vn, 0); 684 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 685 } else { 686 struct mount *mountpoint; 687 int lock_flags; 688 689 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 690 691 if (MNT_SHARED_WRITES(mountpoint) 692 || ((mountpoint == NULL) 693 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 694 lock_flags = LK_SHARED; 695 else 696 lock_flags = LK_EXCLUSIVE; 697 698 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 699 700 /* 701 * UFS pays attention to IO_DIRECT for writes. The write 702 * is done asynchronously. (Normally the write would just 703 * get put into cache. 704 * 705 * UFS pays attention to IO_SYNC for writes. It will 706 * attempt to write the buffer out synchronously if that 707 * flag is set. 708 * 709 * ZFS does not pay attention to IO_DIRECT for writes. 710 * 711 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 712 * for writes. It will flush the transaction from the 713 * cache before returning. 714 * 715 * So if we've got the BIO_ORDERED flag set, we want 716 * IO_SYNC in either the UFS or ZFS case. 717 */ 718 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 719 VOP_UNLOCK(be_lun->vn, 0); 720 721 vn_finished_write(mountpoint); 722 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 723 } 724 725 mtx_lock(&be_lun->io_lock); 726 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 727 beio->ds_tag_type, beio->ds_trans_type, 728 /*now*/ NULL, /*then*/&beio->ds_t0); 729 mtx_unlock(&be_lun->io_lock); 730 731 /* 732 * If we got an error, set the sense data to "MEDIUM ERROR" and 733 * return the I/O to the user. 734 */ 735 if (error != 0) { 736 char path_str[32]; 737 738 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 739 printf("%s%s command returned errno %d\n", path_str, 740 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 741 if (error == ENOSPC) { 742 ctl_set_space_alloc_fail(&io->scsiio); 743 } else 744 ctl_set_medium_error(&io->scsiio); 745 ctl_complete_beio(beio); 746 return; 747 } 748 749 /* 750 * If this is a write or a verify, we're all done. 751 * If this is a read, we can now send the data to the user. 752 */ 753 if ((beio->bio_cmd == BIO_WRITE) || 754 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 755 ctl_set_success(&io->scsiio); 756 ctl_complete_beio(beio); 757 } else { 758 if ((ARGS(io)->flags & CTL_LLF_READ) && 759 beio->beio_cont == NULL) 760 ctl_set_success(&io->scsiio); 761#ifdef CTL_TIME_IO 762 getbintime(&io->io_hdr.dma_start_bt); 763#endif 764 ctl_datamove(io); 765 } 766} 767 768static void 769ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 770 struct ctl_be_block_io *beio) 771{ 772 union ctl_io *io = beio->io; 773 struct ctl_lba_len_flags *lbalen = ARGS(io); 774 struct scsi_get_lba_status_data *data; 775 off_t roff, off; 776 int error, status; 777 778 DPRINTF("entered\n"); 779 780 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 781 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 782 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 783 0, curthread->td_ucred, curthread); 784 if (error == 0 && off > roff) 785 status = 0; /* mapped up to off */ 786 else { 787 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 788 0, curthread->td_ucred, curthread); 789 if (error == 0 && off > roff) 790 status = 1; /* deallocated up to off */ 791 else { 792 status = 0; /* unknown up to the end */ 793 off = be_lun->size_bytes; 794 } 795 } 796 VOP_UNLOCK(be_lun->vn, 0); 797 798 off >>= be_lun->blocksize_shift; 799 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 800 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 801 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 802 data->descr[0].length); 803 data->descr[0].status = status; 804 805 ctl_complete_beio(beio); 806} 807 808static uint64_t 809ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 810{ 811 struct vattr vattr; 812 struct statfs statfs; 813 int error; 814 815 if (be_lun->vn == NULL) 816 return (UINT64_MAX); 817 if (strcmp(attrname, "blocksused") == 0) { 818 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 819 if (error != 0) 820 return (UINT64_MAX); 821 return (vattr.va_bytes >> be_lun->blocksize_shift); 822 } 823 if (strcmp(attrname, "blocksavail") == 0) { 824 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 825 if (error != 0) 826 return (UINT64_MAX); 827 return ((statfs.f_bavail * statfs.f_bsize) >> 828 be_lun->blocksize_shift); 829 } 830 return (UINT64_MAX); 831} 832 833static void 834ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 835 struct ctl_be_block_io *beio) 836{ 837 struct ctl_be_block_devdata *dev_data; 838 union ctl_io *io; 839 struct uio xuio; 840 struct iovec *xiovec; 841 int flags; 842 int error, i; 843 844 DPRINTF("entered\n"); 845 846 dev_data = &be_lun->backend.dev; 847 io = beio->io; 848 flags = 0; 849 if (ARGS(io)->flags & CTL_LLF_DPO) 850 flags |= IO_DIRECT; 851 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 852 flags |= IO_SYNC; 853 854 bzero(&xuio, sizeof(xuio)); 855 if (beio->bio_cmd == BIO_READ) { 856 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 857 xuio.uio_rw = UIO_READ; 858 } else { 859 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 860 xuio.uio_rw = UIO_WRITE; 861 } 862 xuio.uio_offset = beio->io_offset; 863 xuio.uio_resid = beio->io_len; 864 xuio.uio_segflg = UIO_SYSSPACE; 865 xuio.uio_iov = beio->xiovecs; 866 xuio.uio_iovcnt = beio->num_segs; 867 xuio.uio_td = curthread; 868 869 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 870 xiovec->iov_base = beio->sg_segs[i].addr; 871 xiovec->iov_len = beio->sg_segs[i].len; 872 } 873 874 binuptime(&beio->ds_t0); 875 mtx_lock(&be_lun->io_lock); 876 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 877 mtx_unlock(&be_lun->io_lock); 878 879 if (beio->bio_cmd == BIO_READ) { 880 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 881 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 882 } else { 883 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 884 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 885 } 886 887 mtx_lock(&be_lun->io_lock); 888 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 889 beio->ds_tag_type, beio->ds_trans_type, 890 /*now*/ NULL, /*then*/&beio->ds_t0); 891 mtx_unlock(&be_lun->io_lock); 892 893 /* 894 * If we got an error, set the sense data to "MEDIUM ERROR" and 895 * return the I/O to the user. 896 */ 897 if (error != 0) { 898 if (error == ENOSPC) { 899 ctl_set_space_alloc_fail(&io->scsiio); 900 } else 901 ctl_set_medium_error(&io->scsiio); 902 ctl_complete_beio(beio); 903 return; 904 } 905 906 /* 907 * If this is a write or a verify, we're all done. 908 * If this is a read, we can now send the data to the user. 909 */ 910 if ((beio->bio_cmd == BIO_WRITE) || 911 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 912 ctl_set_success(&io->scsiio); 913 ctl_complete_beio(beio); 914 } else { 915 if ((ARGS(io)->flags & CTL_LLF_READ) && 916 beio->beio_cont == NULL) 917 ctl_set_success(&io->scsiio); 918#ifdef CTL_TIME_IO 919 getbintime(&io->io_hdr.dma_start_bt); 920#endif 921 ctl_datamove(io); 922 } 923} 924 925static void 926ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 927 struct ctl_be_block_io *beio) 928{ 929 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 930 union ctl_io *io = beio->io; 931 struct ctl_lba_len_flags *lbalen = ARGS(io); 932 struct scsi_get_lba_status_data *data; 933 off_t roff, off; 934 int error, status; 935 936 DPRINTF("entered\n"); 937 938 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 939 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE, 940 (caddr_t)&off, FREAD, curthread); 941 if (error == 0 && off > roff) 942 status = 0; /* mapped up to off */ 943 else { 944 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA, 945 (caddr_t)&off, FREAD, curthread); 946 if (error == 0 && off > roff) 947 status = 1; /* deallocated up to off */ 948 else { 949 status = 0; /* unknown up to the end */ 950 off = be_lun->size_bytes; 951 } 952 } 953 954 off >>= be_lun->blocksize_shift; 955 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 956 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 957 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 958 data->descr[0].length); 959 data->descr[0].status = status; 960 961 ctl_complete_beio(beio); 962} 963 964static void 965ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 966 struct ctl_be_block_io *beio) 967{ 968 struct bio *bio; 969 union ctl_io *io; 970 struct ctl_be_block_devdata *dev_data; 971 972 dev_data = &be_lun->backend.dev; 973 io = beio->io; 974 975 DPRINTF("entered\n"); 976 977 /* This can't fail, it's a blocking allocation. */ 978 bio = g_alloc_bio(); 979 980 bio->bio_cmd = BIO_FLUSH; 981 bio->bio_flags |= BIO_ORDERED; 982 bio->bio_dev = dev_data->cdev; 983 bio->bio_offset = 0; 984 bio->bio_data = 0; 985 bio->bio_done = ctl_be_block_biodone; 986 bio->bio_caller1 = beio; 987 bio->bio_pblkno = 0; 988 989 /* 990 * We don't need to acquire the LUN lock here, because we are only 991 * sending one bio, and so there is no other context to synchronize 992 * with. 993 */ 994 beio->num_bios_sent = 1; 995 beio->send_complete = 1; 996 997 binuptime(&beio->ds_t0); 998 mtx_lock(&be_lun->io_lock); 999 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1000 mtx_unlock(&be_lun->io_lock); 1001 1002 (*dev_data->csw->d_strategy)(bio); 1003} 1004 1005static void 1006ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1007 struct ctl_be_block_io *beio, 1008 uint64_t off, uint64_t len, int last) 1009{ 1010 struct bio *bio; 1011 struct ctl_be_block_devdata *dev_data; 1012 uint64_t maxlen; 1013 1014 dev_data = &be_lun->backend.dev; 1015 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 1016 while (len > 0) { 1017 bio = g_alloc_bio(); 1018 bio->bio_cmd = BIO_DELETE; 1019 bio->bio_dev = dev_data->cdev; 1020 bio->bio_offset = off; 1021 bio->bio_length = MIN(len, maxlen); 1022 bio->bio_data = 0; 1023 bio->bio_done = ctl_be_block_biodone; 1024 bio->bio_caller1 = beio; 1025 bio->bio_pblkno = off / be_lun->blocksize; 1026 1027 off += bio->bio_length; 1028 len -= bio->bio_length; 1029 1030 mtx_lock(&be_lun->io_lock); 1031 beio->num_bios_sent++; 1032 if (last && len == 0) 1033 beio->send_complete = 1; 1034 mtx_unlock(&be_lun->io_lock); 1035 1036 (*dev_data->csw->d_strategy)(bio); 1037 } 1038} 1039 1040static void 1041ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1042 struct ctl_be_block_io *beio) 1043{ 1044 union ctl_io *io; 1045 struct ctl_be_block_devdata *dev_data; 1046 struct ctl_ptr_len_flags *ptrlen; 1047 struct scsi_unmap_desc *buf, *end; 1048 uint64_t len; 1049 1050 dev_data = &be_lun->backend.dev; 1051 io = beio->io; 1052 1053 DPRINTF("entered\n"); 1054 1055 binuptime(&beio->ds_t0); 1056 mtx_lock(&be_lun->io_lock); 1057 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1058 mtx_unlock(&be_lun->io_lock); 1059 1060 if (beio->io_offset == -1) { 1061 beio->io_len = 0; 1062 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1063 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1064 end = buf + ptrlen->len / sizeof(*buf); 1065 for (; buf < end; buf++) { 1066 len = (uint64_t)scsi_4btoul(buf->length) * 1067 be_lun->blocksize; 1068 beio->io_len += len; 1069 ctl_be_block_unmap_dev_range(be_lun, beio, 1070 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 1071 (end - buf < 2) ? TRUE : FALSE); 1072 } 1073 } else 1074 ctl_be_block_unmap_dev_range(be_lun, beio, 1075 beio->io_offset, beio->io_len, TRUE); 1076} 1077 1078static void 1079ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1080 struct ctl_be_block_io *beio) 1081{ 1082 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1083 int i; 1084 struct bio *bio; 1085 struct ctl_be_block_devdata *dev_data; 1086 off_t cur_offset; 1087 int max_iosize; 1088 1089 DPRINTF("entered\n"); 1090 1091 dev_data = &be_lun->backend.dev; 1092 1093 /* 1094 * We have to limit our I/O size to the maximum supported by the 1095 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1096 * set it properly, use DFLTPHYS. 1097 */ 1098 max_iosize = dev_data->cdev->si_iosize_max; 1099 if (max_iosize < PAGE_SIZE) 1100 max_iosize = DFLTPHYS; 1101 1102 cur_offset = beio->io_offset; 1103 for (i = 0; i < beio->num_segs; i++) { 1104 size_t cur_size; 1105 uint8_t *cur_ptr; 1106 1107 cur_size = beio->sg_segs[i].len; 1108 cur_ptr = beio->sg_segs[i].addr; 1109 1110 while (cur_size > 0) { 1111 /* This can't fail, it's a blocking allocation. */ 1112 bio = g_alloc_bio(); 1113 1114 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1115 1116 bio->bio_cmd = beio->bio_cmd; 1117 bio->bio_dev = dev_data->cdev; 1118 bio->bio_caller1 = beio; 1119 bio->bio_length = min(cur_size, max_iosize); 1120 bio->bio_offset = cur_offset; 1121 bio->bio_data = cur_ptr; 1122 bio->bio_done = ctl_be_block_biodone; 1123 bio->bio_pblkno = cur_offset / be_lun->blocksize; 1124 1125 cur_offset += bio->bio_length; 1126 cur_ptr += bio->bio_length; 1127 cur_size -= bio->bio_length; 1128 1129 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1130 beio->num_bios_sent++; 1131 } 1132 } 1133 binuptime(&beio->ds_t0); 1134 mtx_lock(&be_lun->io_lock); 1135 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1136 beio->send_complete = 1; 1137 mtx_unlock(&be_lun->io_lock); 1138 1139 /* 1140 * Fire off all allocated requests! 1141 */ 1142 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1143 TAILQ_REMOVE(&queue, bio, bio_queue); 1144 (*dev_data->csw->d_strategy)(bio); 1145 } 1146} 1147 1148static uint64_t 1149ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1150{ 1151 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1152 struct diocgattr_arg arg; 1153 int error; 1154 1155 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1156 return (UINT64_MAX); 1157 strlcpy(arg.name, attrname, sizeof(arg.name)); 1158 arg.len = sizeof(arg.value.off); 1159 error = dev_data->csw->d_ioctl(dev_data->cdev, 1160 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1161 if (error != 0) 1162 return (UINT64_MAX); 1163 return (arg.value.off); 1164} 1165 1166static void 1167ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1168{ 1169 union ctl_io *io; 1170 1171 io = beio->io; 1172 ctl_free_beio(beio); 1173 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1174 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1175 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1176 ctl_config_write_done(io); 1177 return; 1178 } 1179 1180 ctl_be_block_config_write(io); 1181} 1182 1183static void 1184ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1185 union ctl_io *io) 1186{ 1187 struct ctl_be_block_io *beio; 1188 struct ctl_be_block_softc *softc; 1189 struct ctl_lba_len_flags *lbalen; 1190 uint64_t len_left, lba; 1191 uint32_t pb, pbo, adj; 1192 int i, seglen; 1193 uint8_t *buf, *end; 1194 1195 DPRINTF("entered\n"); 1196 1197 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1198 softc = be_lun->softc; 1199 lbalen = ARGS(beio->io); 1200 1201 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1202 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1203 ctl_free_beio(beio); 1204 ctl_set_invalid_field(&io->scsiio, 1205 /*sks_valid*/ 1, 1206 /*command*/ 1, 1207 /*field*/ 1, 1208 /*bit_valid*/ 0, 1209 /*bit*/ 0); 1210 ctl_config_write_done(io); 1211 return; 1212 } 1213 1214 switch (io->scsiio.tag_type) { 1215 case CTL_TAG_ORDERED: 1216 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1217 break; 1218 case CTL_TAG_HEAD_OF_QUEUE: 1219 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1220 break; 1221 case CTL_TAG_UNTAGGED: 1222 case CTL_TAG_SIMPLE: 1223 case CTL_TAG_ACA: 1224 default: 1225 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1226 break; 1227 } 1228 1229 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1230 beio->io_offset = lbalen->lba * be_lun->blocksize; 1231 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1232 beio->bio_cmd = BIO_DELETE; 1233 beio->ds_trans_type = DEVSTAT_FREE; 1234 1235 be_lun->unmap(be_lun, beio); 1236 return; 1237 } 1238 1239 beio->bio_cmd = BIO_WRITE; 1240 beio->ds_trans_type = DEVSTAT_WRITE; 1241 1242 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1243 (uintmax_t)lbalen->lba, lbalen->len); 1244 1245 pb = be_lun->blocksize << be_lun->pblockexp; 1246 if (be_lun->pblockoff > 0) 1247 pbo = pb - be_lun->blocksize * be_lun->pblockoff; 1248 else 1249 pbo = 0; 1250 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1251 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1252 1253 /* 1254 * Setup the S/G entry for this chunk. 1255 */ 1256 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1257 if (pb > be_lun->blocksize) { 1258 adj = ((lbalen->lba + lba) * be_lun->blocksize + 1259 seglen - pbo) % pb; 1260 if (seglen > adj) 1261 seglen -= adj; 1262 else 1263 seglen -= seglen % be_lun->blocksize; 1264 } else 1265 seglen -= seglen % be_lun->blocksize; 1266 beio->sg_segs[i].len = seglen; 1267 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1268 1269 DPRINTF("segment %d addr %p len %zd\n", i, 1270 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1271 1272 beio->num_segs++; 1273 len_left -= seglen; 1274 1275 buf = beio->sg_segs[i].addr; 1276 end = buf + seglen; 1277 for (; buf < end; buf += be_lun->blocksize) { 1278 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1279 if (lbalen->flags & SWS_LBDATA) 1280 scsi_ulto4b(lbalen->lba + lba, buf); 1281 lba++; 1282 } 1283 } 1284 1285 beio->io_offset = lbalen->lba * be_lun->blocksize; 1286 beio->io_len = lba * be_lun->blocksize; 1287 1288 /* We can not do all in one run. Correct and schedule rerun. */ 1289 if (len_left > 0) { 1290 lbalen->lba += lba; 1291 lbalen->len -= lba; 1292 beio->beio_cont = ctl_be_block_cw_done_ws; 1293 } 1294 1295 be_lun->dispatch(be_lun, beio); 1296} 1297 1298static void 1299ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1300 union ctl_io *io) 1301{ 1302 struct ctl_be_block_io *beio; 1303 struct ctl_be_block_softc *softc; 1304 struct ctl_ptr_len_flags *ptrlen; 1305 1306 DPRINTF("entered\n"); 1307 1308 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1309 softc = be_lun->softc; 1310 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1311 1312 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1313 ctl_free_beio(beio); 1314 ctl_set_invalid_field(&io->scsiio, 1315 /*sks_valid*/ 0, 1316 /*command*/ 1, 1317 /*field*/ 0, 1318 /*bit_valid*/ 0, 1319 /*bit*/ 0); 1320 ctl_config_write_done(io); 1321 return; 1322 } 1323 1324 switch (io->scsiio.tag_type) { 1325 case CTL_TAG_ORDERED: 1326 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1327 break; 1328 case CTL_TAG_HEAD_OF_QUEUE: 1329 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1330 break; 1331 case CTL_TAG_UNTAGGED: 1332 case CTL_TAG_SIMPLE: 1333 case CTL_TAG_ACA: 1334 default: 1335 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1336 break; 1337 } 1338 1339 beio->io_len = 0; 1340 beio->io_offset = -1; 1341 1342 beio->bio_cmd = BIO_DELETE; 1343 beio->ds_trans_type = DEVSTAT_FREE; 1344 1345 DPRINTF("UNMAP\n"); 1346 1347 be_lun->unmap(be_lun, beio); 1348} 1349 1350static void 1351ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1352{ 1353 union ctl_io *io; 1354 1355 io = beio->io; 1356 ctl_free_beio(beio); 1357 ctl_config_read_done(io); 1358} 1359 1360static void 1361ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1362 union ctl_io *io) 1363{ 1364 struct ctl_be_block_io *beio; 1365 struct ctl_be_block_softc *softc; 1366 1367 DPRINTF("entered\n"); 1368 1369 softc = be_lun->softc; 1370 beio = ctl_alloc_beio(softc); 1371 beio->io = io; 1372 beio->lun = be_lun; 1373 beio->beio_cont = ctl_be_block_cr_done; 1374 PRIV(io)->ptr = (void *)beio; 1375 1376 switch (io->scsiio.cdb[0]) { 1377 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1378 beio->bio_cmd = -1; 1379 beio->ds_trans_type = DEVSTAT_NO_DATA; 1380 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1381 beio->io_len = 0; 1382 if (be_lun->get_lba_status) 1383 be_lun->get_lba_status(be_lun, beio); 1384 else 1385 ctl_be_block_cr_done(beio); 1386 break; 1387 default: 1388 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1389 break; 1390 } 1391} 1392 1393static void 1394ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1395{ 1396 union ctl_io *io; 1397 1398 io = beio->io; 1399 ctl_free_beio(beio); 1400 ctl_config_write_done(io); 1401} 1402 1403static void 1404ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1405 union ctl_io *io) 1406{ 1407 struct ctl_be_block_io *beio; 1408 struct ctl_be_block_softc *softc; 1409 1410 DPRINTF("entered\n"); 1411 1412 softc = be_lun->softc; 1413 beio = ctl_alloc_beio(softc); 1414 beio->io = io; 1415 beio->lun = be_lun; 1416 beio->beio_cont = ctl_be_block_cw_done; 1417 PRIV(io)->ptr = (void *)beio; 1418 1419 switch (io->scsiio.cdb[0]) { 1420 case SYNCHRONIZE_CACHE: 1421 case SYNCHRONIZE_CACHE_16: 1422 beio->bio_cmd = BIO_FLUSH; 1423 beio->ds_trans_type = DEVSTAT_NO_DATA; 1424 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1425 beio->io_len = 0; 1426 be_lun->lun_flush(be_lun, beio); 1427 break; 1428 case WRITE_SAME_10: 1429 case WRITE_SAME_16: 1430 ctl_be_block_cw_dispatch_ws(be_lun, io); 1431 break; 1432 case UNMAP: 1433 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1434 break; 1435 default: 1436 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1437 break; 1438 } 1439} 1440 1441SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1442SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1443SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1444SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1445 1446static void 1447ctl_be_block_next(struct ctl_be_block_io *beio) 1448{ 1449 struct ctl_be_block_lun *be_lun; 1450 union ctl_io *io; 1451 1452 io = beio->io; 1453 be_lun = beio->lun; 1454 ctl_free_beio(beio); 1455 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1456 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1457 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1458 ctl_data_submit_done(io); 1459 return; 1460 } 1461 1462 io->io_hdr.status &= ~CTL_STATUS_MASK; 1463 io->io_hdr.status |= CTL_STATUS_NONE; 1464 1465 mtx_lock(&be_lun->queue_lock); 1466 /* 1467 * XXX KDM make sure that links is okay to use at this point. 1468 * Otherwise, we either need to add another field to ctl_io_hdr, 1469 * or deal with resource allocation here. 1470 */ 1471 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1472 mtx_unlock(&be_lun->queue_lock); 1473 1474 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1475} 1476 1477static void 1478ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1479 union ctl_io *io) 1480{ 1481 struct ctl_be_block_io *beio; 1482 struct ctl_be_block_softc *softc; 1483 struct ctl_lba_len_flags *lbalen; 1484 struct ctl_ptr_len_flags *bptrlen; 1485 uint64_t len_left, lbas; 1486 int i; 1487 1488 softc = be_lun->softc; 1489 1490 DPRINTF("entered\n"); 1491 1492 lbalen = ARGS(io); 1493 if (lbalen->flags & CTL_LLF_WRITE) { 1494 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1495 } else { 1496 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1497 } 1498 1499 beio = ctl_alloc_beio(softc); 1500 beio->io = io; 1501 beio->lun = be_lun; 1502 bptrlen = PRIV(io); 1503 bptrlen->ptr = (void *)beio; 1504 1505 switch (io->scsiio.tag_type) { 1506 case CTL_TAG_ORDERED: 1507 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1508 break; 1509 case CTL_TAG_HEAD_OF_QUEUE: 1510 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1511 break; 1512 case CTL_TAG_UNTAGGED: 1513 case CTL_TAG_SIMPLE: 1514 case CTL_TAG_ACA: 1515 default: 1516 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1517 break; 1518 } 1519 1520 if (lbalen->flags & CTL_LLF_WRITE) { 1521 beio->bio_cmd = BIO_WRITE; 1522 beio->ds_trans_type = DEVSTAT_WRITE; 1523 } else { 1524 beio->bio_cmd = BIO_READ; 1525 beio->ds_trans_type = DEVSTAT_READ; 1526 } 1527 1528 DPRINTF("%s at LBA %jx len %u @%ju\n", 1529 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1530 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1531 if (lbalen->flags & CTL_LLF_COMPARE) 1532 lbas = CTLBLK_HALF_IO_SIZE; 1533 else 1534 lbas = CTLBLK_MAX_IO_SIZE; 1535 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1536 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1537 beio->io_len = lbas * be_lun->blocksize; 1538 bptrlen->len += lbas; 1539 1540 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1541 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1542 i, CTLBLK_MAX_SEGS)); 1543 1544 /* 1545 * Setup the S/G entry for this chunk. 1546 */ 1547 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1548 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1549 1550 DPRINTF("segment %d addr %p len %zd\n", i, 1551 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1552 1553 /* Set up second segment for compare operation. */ 1554 if (lbalen->flags & CTL_LLF_COMPARE) { 1555 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1556 beio->sg_segs[i].len; 1557 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1558 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1559 } 1560 1561 beio->num_segs++; 1562 len_left -= beio->sg_segs[i].len; 1563 } 1564 if (bptrlen->len < lbalen->len) 1565 beio->beio_cont = ctl_be_block_next; 1566 io->scsiio.be_move_done = ctl_be_block_move_done; 1567 /* For compare we have separate S/G lists for read and datamove. */ 1568 if (lbalen->flags & CTL_LLF_COMPARE) 1569 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1570 else 1571 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1572 io->scsiio.kern_data_len = beio->io_len; 1573 io->scsiio.kern_data_resid = 0; 1574 io->scsiio.kern_sg_entries = beio->num_segs; 1575 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1576 1577 /* 1578 * For the read case, we need to read the data into our buffers and 1579 * then we can send it back to the user. For the write case, we 1580 * need to get the data from the user first. 1581 */ 1582 if (beio->bio_cmd == BIO_READ) { 1583 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1584 be_lun->dispatch(be_lun, beio); 1585 } else { 1586 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1587#ifdef CTL_TIME_IO 1588 getbintime(&io->io_hdr.dma_start_bt); 1589#endif 1590 ctl_datamove(io); 1591 } 1592} 1593 1594static void 1595ctl_be_block_worker(void *context, int pending) 1596{ 1597 struct ctl_be_block_lun *be_lun; 1598 struct ctl_be_block_softc *softc; 1599 union ctl_io *io; 1600 1601 be_lun = (struct ctl_be_block_lun *)context; 1602 softc = be_lun->softc; 1603 1604 DPRINTF("entered\n"); 1605 1606 mtx_lock(&be_lun->queue_lock); 1607 for (;;) { 1608 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1609 if (io != NULL) { 1610 struct ctl_be_block_io *beio; 1611 1612 DPRINTF("datamove queue\n"); 1613 1614 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1615 ctl_io_hdr, links); 1616 1617 mtx_unlock(&be_lun->queue_lock); 1618 1619 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1620 1621 be_lun->dispatch(be_lun, beio); 1622 1623 mtx_lock(&be_lun->queue_lock); 1624 continue; 1625 } 1626 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1627 if (io != NULL) { 1628 DPRINTF("config write queue\n"); 1629 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1630 ctl_io_hdr, links); 1631 mtx_unlock(&be_lun->queue_lock); 1632 ctl_be_block_cw_dispatch(be_lun, io); 1633 mtx_lock(&be_lun->queue_lock); 1634 continue; 1635 } 1636 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1637 if (io != NULL) { 1638 DPRINTF("config read queue\n"); 1639 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1640 ctl_io_hdr, links); 1641 mtx_unlock(&be_lun->queue_lock); 1642 ctl_be_block_cr_dispatch(be_lun, io); 1643 mtx_lock(&be_lun->queue_lock); 1644 continue; 1645 } 1646 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1647 if (io != NULL) { 1648 DPRINTF("input queue\n"); 1649 1650 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1651 ctl_io_hdr, links); 1652 mtx_unlock(&be_lun->queue_lock); 1653 1654 /* 1655 * We must drop the lock, since this routine and 1656 * its children may sleep. 1657 */ 1658 ctl_be_block_dispatch(be_lun, io); 1659 1660 mtx_lock(&be_lun->queue_lock); 1661 continue; 1662 } 1663 1664 /* 1665 * If we get here, there is no work left in the queues, so 1666 * just break out and let the task queue go to sleep. 1667 */ 1668 break; 1669 } 1670 mtx_unlock(&be_lun->queue_lock); 1671} 1672 1673/* 1674 * Entry point from CTL to the backend for I/O. We queue everything to a 1675 * work thread, so this just puts the I/O on a queue and wakes up the 1676 * thread. 1677 */ 1678static int 1679ctl_be_block_submit(union ctl_io *io) 1680{ 1681 struct ctl_be_block_lun *be_lun; 1682 struct ctl_be_lun *ctl_be_lun; 1683 1684 DPRINTF("entered\n"); 1685 1686 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1687 CTL_PRIV_BACKEND_LUN].ptr; 1688 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1689 1690 /* 1691 * Make sure we only get SCSI I/O. 1692 */ 1693 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1694 "%#x) encountered", io->io_hdr.io_type)); 1695 1696 PRIV(io)->len = 0; 1697 1698 mtx_lock(&be_lun->queue_lock); 1699 /* 1700 * XXX KDM make sure that links is okay to use at this point. 1701 * Otherwise, we either need to add another field to ctl_io_hdr, 1702 * or deal with resource allocation here. 1703 */ 1704 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1705 mtx_unlock(&be_lun->queue_lock); 1706 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1707 1708 return (CTL_RETVAL_COMPLETE); 1709} 1710 1711static int 1712ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1713 int flag, struct thread *td) 1714{ 1715 struct ctl_be_block_softc *softc; 1716 int error; 1717 1718 softc = &backend_block_softc; 1719 1720 error = 0; 1721 1722 switch (cmd) { 1723 case CTL_LUN_REQ: { 1724 struct ctl_lun_req *lun_req; 1725 1726 lun_req = (struct ctl_lun_req *)addr; 1727 1728 switch (lun_req->reqtype) { 1729 case CTL_LUNREQ_CREATE: 1730 error = ctl_be_block_create(softc, lun_req); 1731 break; 1732 case CTL_LUNREQ_RM: 1733 error = ctl_be_block_rm(softc, lun_req); 1734 break; 1735 case CTL_LUNREQ_MODIFY: 1736 error = ctl_be_block_modify(softc, lun_req); 1737 break; 1738 default: 1739 lun_req->status = CTL_LUN_ERROR; 1740 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1741 "invalid LUN request type %d", 1742 lun_req->reqtype); 1743 break; 1744 } 1745 break; 1746 } 1747 default: 1748 error = ENOTTY; 1749 break; 1750 } 1751 1752 return (error); 1753} 1754 1755static int 1756ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1757{ 1758 struct ctl_be_block_filedata *file_data; 1759 struct ctl_lun_create_params *params; 1760 char *value; 1761 struct vattr vattr; 1762 off_t ps, pss, po, pos, us, uss, uo, uos; 1763 int error; 1764 1765 error = 0; 1766 file_data = &be_lun->backend.file; 1767 params = &be_lun->params; 1768 1769 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1770 be_lun->dispatch = ctl_be_block_dispatch_file; 1771 be_lun->lun_flush = ctl_be_block_flush_file; 1772 be_lun->get_lba_status = ctl_be_block_gls_file; 1773 be_lun->getattr = ctl_be_block_getattr_file; 1774 1775 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1776 if (error != 0) { 1777 snprintf(req->error_str, sizeof(req->error_str), 1778 "error calling VOP_GETATTR() for file %s", 1779 be_lun->dev_path); 1780 return (error); 1781 } 1782 1783 /* 1784 * Verify that we have the ability to upgrade to exclusive 1785 * access on this file so we can trap errors at open instead 1786 * of reporting them during first access. 1787 */ 1788 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1789 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1790 if (be_lun->vn->v_iflag & VI_DOOMED) { 1791 error = EBADF; 1792 snprintf(req->error_str, sizeof(req->error_str), 1793 "error locking file %s", be_lun->dev_path); 1794 return (error); 1795 } 1796 } 1797 1798 1799 file_data->cred = crhold(curthread->td_ucred); 1800 if (params->lun_size_bytes != 0) 1801 be_lun->size_bytes = params->lun_size_bytes; 1802 else 1803 be_lun->size_bytes = vattr.va_size; 1804 /* 1805 * We set the multi thread flag for file operations because all 1806 * filesystems (in theory) are capable of allowing multiple readers 1807 * of a file at once. So we want to get the maximum possible 1808 * concurrency. 1809 */ 1810 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1811 1812 /* 1813 * For files we can use any logical block size. Prefer 512 bytes 1814 * for compatibility reasons. If file's vattr.va_blocksize 1815 * (preferred I/O block size) is bigger and multiple to chosen 1816 * logical block size -- report it as physical block size. 1817 */ 1818 if (params->blocksize_bytes != 0) 1819 be_lun->blocksize = params->blocksize_bytes; 1820 else 1821 be_lun->blocksize = 512; 1822 1823 us = ps = vattr.va_blocksize; 1824 uo = po = 0; 1825 1826 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 1827 if (value != NULL) 1828 ctl_expand_number(value, &ps); 1829 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 1830 if (value != NULL) 1831 ctl_expand_number(value, &po); 1832 pss = ps / be_lun->blocksize; 1833 pos = po / be_lun->blocksize; 1834 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1835 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1836 be_lun->pblockexp = fls(pss) - 1; 1837 be_lun->pblockoff = (pss - pos) % pss; 1838 } 1839 1840 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 1841 if (value != NULL) 1842 ctl_expand_number(value, &us); 1843 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 1844 if (value != NULL) 1845 ctl_expand_number(value, &uo); 1846 uss = us / be_lun->blocksize; 1847 uos = uo / be_lun->blocksize; 1848 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 1849 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 1850 be_lun->ublockexp = fls(uss) - 1; 1851 be_lun->ublockoff = (uss - uos) % uss; 1852 } 1853 1854 /* 1855 * Sanity check. The media size has to be at least one 1856 * sector long. 1857 */ 1858 if (be_lun->size_bytes < be_lun->blocksize) { 1859 error = EINVAL; 1860 snprintf(req->error_str, sizeof(req->error_str), 1861 "file %s size %ju < block size %u", be_lun->dev_path, 1862 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1863 } 1864 1865 be_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / be_lun->blocksize; 1866 return (error); 1867} 1868 1869static int 1870ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1871{ 1872 struct ctl_lun_create_params *params; 1873 struct vattr vattr; 1874 struct cdev *dev; 1875 struct cdevsw *devsw; 1876 char *value; 1877 int error, atomic, maxio, unmap; 1878 off_t ps, pss, po, pos, us, uss, uo, uos; 1879 1880 params = &be_lun->params; 1881 1882 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1883 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1884 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1885 &be_lun->backend.dev.dev_ref); 1886 if (be_lun->backend.dev.csw == NULL) 1887 panic("Unable to retrieve device switch"); 1888 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) { 1889 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1890 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1891 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1892 } else { 1893 be_lun->dispatch = ctl_be_block_dispatch_dev; 1894 atomic = 0; 1895 maxio = be_lun->backend.dev.cdev->si_iosize_max; 1896 if (maxio <= 0) 1897 maxio = DFLTPHYS; 1898 if (maxio > CTLBLK_MAX_IO_SIZE) 1899 maxio = CTLBLK_MAX_IO_SIZE; 1900 } 1901 be_lun->lun_flush = ctl_be_block_flush_dev; 1902 be_lun->getattr = ctl_be_block_getattr_dev; 1903 1904 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1905 if (error) { 1906 snprintf(req->error_str, sizeof(req->error_str), 1907 "error getting vnode attributes for device %s", 1908 be_lun->dev_path); 1909 return (error); 1910 } 1911 1912 dev = be_lun->vn->v_rdev; 1913 devsw = dev->si_devsw; 1914 if (!devsw->d_ioctl) { 1915 snprintf(req->error_str, sizeof(req->error_str), 1916 "no d_ioctl for device %s!", 1917 be_lun->dev_path); 1918 return (ENODEV); 1919 } 1920 1921 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1922 (caddr_t)&be_lun->blocksize, FREAD, 1923 curthread); 1924 if (error) { 1925 snprintf(req->error_str, sizeof(req->error_str), 1926 "error %d returned for DIOCGSECTORSIZE ioctl " 1927 "on %s!", error, be_lun->dev_path); 1928 return (error); 1929 } 1930 1931 /* 1932 * If the user has asked for a blocksize that is greater than the 1933 * backing device's blocksize, we can do it only if the blocksize 1934 * the user is asking for is an even multiple of the underlying 1935 * device's blocksize. 1936 */ 1937 if ((params->blocksize_bytes != 0) 1938 && (params->blocksize_bytes > be_lun->blocksize)) { 1939 uint32_t bs_multiple, tmp_blocksize; 1940 1941 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1942 1943 tmp_blocksize = bs_multiple * be_lun->blocksize; 1944 1945 if (tmp_blocksize == params->blocksize_bytes) { 1946 be_lun->blocksize = params->blocksize_bytes; 1947 } else { 1948 snprintf(req->error_str, sizeof(req->error_str), 1949 "requested blocksize %u is not an even " 1950 "multiple of backing device blocksize %u", 1951 params->blocksize_bytes, 1952 be_lun->blocksize); 1953 return (EINVAL); 1954 1955 } 1956 } else if ((params->blocksize_bytes != 0) 1957 && (params->blocksize_bytes != be_lun->blocksize)) { 1958 snprintf(req->error_str, sizeof(req->error_str), 1959 "requested blocksize %u < backing device " 1960 "blocksize %u", params->blocksize_bytes, 1961 be_lun->blocksize); 1962 return (EINVAL); 1963 } 1964 1965 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1966 (caddr_t)&be_lun->size_bytes, FREAD, 1967 curthread); 1968 if (error) { 1969 snprintf(req->error_str, sizeof(req->error_str), 1970 "error %d returned for DIOCGMEDIASIZE " 1971 " ioctl on %s!", error, 1972 be_lun->dev_path); 1973 return (error); 1974 } 1975 1976 if (params->lun_size_bytes != 0) { 1977 if (params->lun_size_bytes > be_lun->size_bytes) { 1978 snprintf(req->error_str, sizeof(req->error_str), 1979 "requested LUN size %ju > backing device " 1980 "size %ju", 1981 (uintmax_t)params->lun_size_bytes, 1982 (uintmax_t)be_lun->size_bytes); 1983 return (EINVAL); 1984 } 1985 1986 be_lun->size_bytes = params->lun_size_bytes; 1987 } 1988 1989 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1990 (caddr_t)&ps, FREAD, curthread); 1991 if (error) 1992 ps = po = 0; 1993 else { 1994 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1995 (caddr_t)&po, FREAD, curthread); 1996 if (error) 1997 po = 0; 1998 } 1999 us = ps; 2000 uo = po; 2001 2002 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 2003 if (value != NULL) 2004 ctl_expand_number(value, &ps); 2005 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 2006 if (value != NULL) 2007 ctl_expand_number(value, &po); 2008 pss = ps / be_lun->blocksize; 2009 pos = po / be_lun->blocksize; 2010 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 2011 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 2012 be_lun->pblockexp = fls(pss) - 1; 2013 be_lun->pblockoff = (pss - pos) % pss; 2014 } 2015 2016 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 2017 if (value != NULL) 2018 ctl_expand_number(value, &us); 2019 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 2020 if (value != NULL) 2021 ctl_expand_number(value, &uo); 2022 uss = us / be_lun->blocksize; 2023 uos = uo / be_lun->blocksize; 2024 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 2025 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 2026 be_lun->ublockexp = fls(uss) - 1; 2027 be_lun->ublockoff = (uss - uos) % uss; 2028 } 2029 2030 be_lun->atomicblock = atomic / be_lun->blocksize; 2031 be_lun->opttxferlen = maxio / be_lun->blocksize; 2032 2033 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2034 unmap = 1; 2035 } else { 2036 struct diocgattr_arg arg; 2037 2038 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2039 arg.len = sizeof(arg.value.i); 2040 error = devsw->d_ioctl(dev, DIOCGATTR, 2041 (caddr_t)&arg, FREAD, curthread); 2042 unmap = (error == 0) ? arg.value.i : 0; 2043 } 2044 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2045 if (value != NULL) 2046 unmap = (strcmp(value, "on") == 0); 2047 if (unmap) 2048 be_lun->unmap = ctl_be_block_unmap_dev; 2049 2050 return (0); 2051} 2052 2053static int 2054ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2055{ 2056 DROP_GIANT(); 2057 if (be_lun->vn) { 2058 int flags = FREAD | FWRITE; 2059 2060 switch (be_lun->dev_type) { 2061 case CTL_BE_BLOCK_DEV: 2062 if (be_lun->backend.dev.csw) { 2063 dev_relthread(be_lun->backend.dev.cdev, 2064 be_lun->backend.dev.dev_ref); 2065 be_lun->backend.dev.csw = NULL; 2066 be_lun->backend.dev.cdev = NULL; 2067 } 2068 break; 2069 case CTL_BE_BLOCK_FILE: 2070 break; 2071 case CTL_BE_BLOCK_NONE: 2072 break; 2073 default: 2074 panic("Unexpected backend type."); 2075 break; 2076 } 2077 2078 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2079 be_lun->vn = NULL; 2080 2081 switch (be_lun->dev_type) { 2082 case CTL_BE_BLOCK_DEV: 2083 break; 2084 case CTL_BE_BLOCK_FILE: 2085 if (be_lun->backend.file.cred != NULL) { 2086 crfree(be_lun->backend.file.cred); 2087 be_lun->backend.file.cred = NULL; 2088 } 2089 break; 2090 case CTL_BE_BLOCK_NONE: 2091 break; 2092 default: 2093 panic("Unexpected backend type."); 2094 break; 2095 } 2096 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2097 } 2098 PICKUP_GIANT(); 2099 2100 return (0); 2101} 2102 2103static int 2104ctl_be_block_open(struct ctl_be_block_softc *softc, 2105 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2106{ 2107 struct nameidata nd; 2108 int flags; 2109 int error; 2110 2111 /* 2112 * XXX KDM allow a read-only option? 2113 */ 2114 flags = FREAD | FWRITE; 2115 error = 0; 2116 2117 if (rootvnode == NULL) { 2118 snprintf(req->error_str, sizeof(req->error_str), 2119 "Root filesystem is not mounted"); 2120 return (1); 2121 } 2122 2123 if (!curthread->td_proc->p_fd->fd_cdir) { 2124 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2125 VREF(rootvnode); 2126 } 2127 if (!curthread->td_proc->p_fd->fd_rdir) { 2128 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2129 VREF(rootvnode); 2130 } 2131 if (!curthread->td_proc->p_fd->fd_jdir) { 2132 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2133 VREF(rootvnode); 2134 } 2135 2136 again: 2137 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2138 error = vn_open(&nd, &flags, 0, NULL); 2139 if (error) { 2140 /* 2141 * This is the only reasonable guess we can make as far as 2142 * path if the user doesn't give us a fully qualified path. 2143 * If they want to specify a file, they need to specify the 2144 * full path. 2145 */ 2146 if (be_lun->dev_path[0] != '/') { 2147 char *dev_path = "/dev/"; 2148 char *dev_name; 2149 2150 /* Try adding device path at beginning of name */ 2151 dev_name = malloc(strlen(be_lun->dev_path) 2152 + strlen(dev_path) + 1, 2153 M_CTLBLK, M_WAITOK); 2154 if (dev_name) { 2155 sprintf(dev_name, "%s%s", dev_path, 2156 be_lun->dev_path); 2157 free(be_lun->dev_path, M_CTLBLK); 2158 be_lun->dev_path = dev_name; 2159 goto again; 2160 } 2161 } 2162 snprintf(req->error_str, sizeof(req->error_str), 2163 "error opening %s: %d", be_lun->dev_path, error); 2164 return (error); 2165 } 2166 2167 NDFREE(&nd, NDF_ONLY_PNBUF); 2168 2169 be_lun->vn = nd.ni_vp; 2170 2171 /* We only support disks and files. */ 2172 if (vn_isdisk(be_lun->vn, &error)) { 2173 error = ctl_be_block_open_dev(be_lun, req); 2174 } else if (be_lun->vn->v_type == VREG) { 2175 error = ctl_be_block_open_file(be_lun, req); 2176 } else { 2177 error = EINVAL; 2178 snprintf(req->error_str, sizeof(req->error_str), 2179 "%s is not a disk or plain file", be_lun->dev_path); 2180 } 2181 VOP_UNLOCK(be_lun->vn, 0); 2182 2183 if (error != 0) { 2184 ctl_be_block_close(be_lun); 2185 return (error); 2186 } 2187 2188 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2189 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2190 2191 return (0); 2192} 2193 2194static int 2195ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2196{ 2197 struct ctl_be_block_lun *be_lun; 2198 struct ctl_lun_create_params *params; 2199 char num_thread_str[16]; 2200 char tmpstr[32]; 2201 char *value; 2202 int retval, num_threads; 2203 int tmp_num_threads; 2204 2205 params = &req->reqdata.create; 2206 retval = 0; 2207 req->status = CTL_LUN_OK; 2208 2209 num_threads = cbb_num_threads; 2210 2211 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2212 2213 be_lun->params = req->reqdata.create; 2214 be_lun->softc = softc; 2215 STAILQ_INIT(&be_lun->input_queue); 2216 STAILQ_INIT(&be_lun->config_read_queue); 2217 STAILQ_INIT(&be_lun->config_write_queue); 2218 STAILQ_INIT(&be_lun->datamove_queue); 2219 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2220 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2221 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2222 ctl_init_opts(&be_lun->ctl_be_lun.options, 2223 req->num_be_args, req->kern_be_args); 2224 2225 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2226 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2227 2228 if (be_lun->lun_zone == NULL) { 2229 snprintf(req->error_str, sizeof(req->error_str), 2230 "error allocating UMA zone"); 2231 goto bailout_error; 2232 } 2233 2234 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2235 be_lun->ctl_be_lun.lun_type = params->device_type; 2236 else 2237 be_lun->ctl_be_lun.lun_type = T_DIRECT; 2238 2239 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 2240 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 2241 if (value == NULL) { 2242 snprintf(req->error_str, sizeof(req->error_str), 2243 "no file argument specified"); 2244 goto bailout_error; 2245 } 2246 be_lun->dev_path = strdup(value, M_CTLBLK); 2247 be_lun->blocksize = 512; 2248 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2249 2250 retval = ctl_be_block_open(softc, be_lun, req); 2251 if (retval != 0) { 2252 retval = 0; 2253 req->status = CTL_LUN_WARNING; 2254 } 2255 } else { 2256 /* 2257 * For processor devices, we don't have any size. 2258 */ 2259 be_lun->blocksize = 0; 2260 be_lun->pblockexp = 0; 2261 be_lun->pblockoff = 0; 2262 be_lun->ublockexp = 0; 2263 be_lun->ublockoff = 0; 2264 be_lun->size_blocks = 0; 2265 be_lun->size_bytes = 0; 2266 be_lun->ctl_be_lun.maxlba = 0; 2267 2268 /* 2269 * Default to just 1 thread for processor devices. 2270 */ 2271 num_threads = 1; 2272 } 2273 2274 /* 2275 * XXX This searching loop might be refactored to be combined with 2276 * the loop above, 2277 */ 2278 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2279 if (value != NULL) { 2280 tmp_num_threads = strtol(value, NULL, 0); 2281 2282 /* 2283 * We don't let the user specify less than one 2284 * thread, but hope he's clueful enough not to 2285 * specify 1000 threads. 2286 */ 2287 if (tmp_num_threads < 1) { 2288 snprintf(req->error_str, sizeof(req->error_str), 2289 "invalid number of threads %s", 2290 num_thread_str); 2291 goto bailout_error; 2292 } 2293 num_threads = tmp_num_threads; 2294 } 2295 2296 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2297 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2298 if (be_lun->vn == NULL) 2299 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2300 if (be_lun->unmap != NULL) 2301 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2302 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2303 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_SERSEQ_READ; 2304 be_lun->ctl_be_lun.be_lun = be_lun; 2305 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2306 0 : (be_lun->size_blocks - 1); 2307 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2308 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2309 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2310 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2311 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2312 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2313 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2314 /* Tell the user the blocksize we ended up using */ 2315 params->lun_size_bytes = be_lun->size_bytes; 2316 params->blocksize_bytes = be_lun->blocksize; 2317 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2318 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2319 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2320 } else 2321 be_lun->ctl_be_lun.req_lun_id = 0; 2322 2323 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2324 be_lun->ctl_be_lun.lun_config_status = 2325 ctl_be_block_lun_config_status; 2326 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2327 2328 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2329 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2330 softc->num_luns); 2331 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2332 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2333 sizeof(tmpstr))); 2334 2335 /* Tell the user what we used for a serial number */ 2336 strncpy((char *)params->serial_num, tmpstr, 2337 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2338 } else { 2339 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2340 params->serial_num, 2341 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2342 sizeof(params->serial_num))); 2343 } 2344 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2345 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2346 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2347 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2348 sizeof(tmpstr))); 2349 2350 /* Tell the user what we used for a device ID */ 2351 strncpy((char *)params->device_id, tmpstr, 2352 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2353 } else { 2354 strncpy((char *)be_lun->ctl_be_lun.device_id, 2355 params->device_id, 2356 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2357 sizeof(params->device_id))); 2358 } 2359 2360 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2361 2362 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2363 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2364 2365 if (be_lun->io_taskqueue == NULL) { 2366 snprintf(req->error_str, sizeof(req->error_str), 2367 "unable to create taskqueue"); 2368 goto bailout_error; 2369 } 2370 2371 /* 2372 * Note that we start the same number of threads by default for 2373 * both the file case and the block device case. For the file 2374 * case, we need multiple threads to allow concurrency, because the 2375 * vnode interface is designed to be a blocking interface. For the 2376 * block device case, ZFS zvols at least will block the caller's 2377 * context in many instances, and so we need multiple threads to 2378 * overcome that problem. Other block devices don't need as many 2379 * threads, but they shouldn't cause too many problems. 2380 * 2381 * If the user wants to just have a single thread for a block 2382 * device, he can specify that when the LUN is created, or change 2383 * the tunable/sysctl to alter the default number of threads. 2384 */ 2385 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2386 /*num threads*/num_threads, 2387 /*priority*/PWAIT, 2388 /*thread name*/ 2389 "%s taskq", be_lun->lunname); 2390 2391 if (retval != 0) 2392 goto bailout_error; 2393 2394 be_lun->num_threads = num_threads; 2395 2396 mtx_lock(&softc->lock); 2397 softc->num_luns++; 2398 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2399 2400 mtx_unlock(&softc->lock); 2401 2402 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2403 if (retval != 0) { 2404 mtx_lock(&softc->lock); 2405 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2406 links); 2407 softc->num_luns--; 2408 mtx_unlock(&softc->lock); 2409 snprintf(req->error_str, sizeof(req->error_str), 2410 "ctl_add_lun() returned error %d, see dmesg for " 2411 "details", retval); 2412 retval = 0; 2413 goto bailout_error; 2414 } 2415 2416 mtx_lock(&softc->lock); 2417 2418 /* 2419 * Tell the config_status routine that we're waiting so it won't 2420 * clean up the LUN in the event of an error. 2421 */ 2422 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2423 2424 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2425 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2426 if (retval == EINTR) 2427 break; 2428 } 2429 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2430 2431 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2432 snprintf(req->error_str, sizeof(req->error_str), 2433 "LUN configuration error, see dmesg for details"); 2434 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2435 links); 2436 softc->num_luns--; 2437 mtx_unlock(&softc->lock); 2438 goto bailout_error; 2439 } else { 2440 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2441 } 2442 2443 mtx_unlock(&softc->lock); 2444 2445 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2446 be_lun->blocksize, 2447 DEVSTAT_ALL_SUPPORTED, 2448 be_lun->ctl_be_lun.lun_type 2449 | DEVSTAT_TYPE_IF_OTHER, 2450 DEVSTAT_PRIORITY_OTHER); 2451 2452 return (retval); 2453 2454bailout_error: 2455 req->status = CTL_LUN_ERROR; 2456 2457 if (be_lun->io_taskqueue != NULL) 2458 taskqueue_free(be_lun->io_taskqueue); 2459 ctl_be_block_close(be_lun); 2460 if (be_lun->dev_path != NULL) 2461 free(be_lun->dev_path, M_CTLBLK); 2462 if (be_lun->lun_zone != NULL) 2463 uma_zdestroy(be_lun->lun_zone); 2464 ctl_free_opts(&be_lun->ctl_be_lun.options); 2465 mtx_destroy(&be_lun->queue_lock); 2466 mtx_destroy(&be_lun->io_lock); 2467 free(be_lun, M_CTLBLK); 2468 2469 return (retval); 2470} 2471 2472static int 2473ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2474{ 2475 struct ctl_lun_rm_params *params; 2476 struct ctl_be_block_lun *be_lun; 2477 int retval; 2478 2479 params = &req->reqdata.rm; 2480 2481 mtx_lock(&softc->lock); 2482 2483 be_lun = NULL; 2484 2485 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2486 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2487 break; 2488 } 2489 mtx_unlock(&softc->lock); 2490 2491 if (be_lun == NULL) { 2492 snprintf(req->error_str, sizeof(req->error_str), 2493 "LUN %u is not managed by the block backend", 2494 params->lun_id); 2495 goto bailout_error; 2496 } 2497 2498 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2499 2500 if (retval != 0) { 2501 snprintf(req->error_str, sizeof(req->error_str), 2502 "error %d returned from ctl_disable_lun() for " 2503 "LUN %d", retval, params->lun_id); 2504 goto bailout_error; 2505 2506 } 2507 2508 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2509 if (retval != 0) { 2510 snprintf(req->error_str, sizeof(req->error_str), 2511 "error %d returned from ctl_invalidate_lun() for " 2512 "LUN %d", retval, params->lun_id); 2513 goto bailout_error; 2514 } 2515 2516 mtx_lock(&softc->lock); 2517 2518 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2519 2520 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2521 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2522 if (retval == EINTR) 2523 break; 2524 } 2525 2526 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2527 2528 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2529 snprintf(req->error_str, sizeof(req->error_str), 2530 "interrupted waiting for LUN to be freed"); 2531 mtx_unlock(&softc->lock); 2532 goto bailout_error; 2533 } 2534 2535 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2536 2537 softc->num_luns--; 2538 mtx_unlock(&softc->lock); 2539 2540 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2541 2542 taskqueue_free(be_lun->io_taskqueue); 2543 2544 ctl_be_block_close(be_lun); 2545 2546 if (be_lun->disk_stats != NULL) 2547 devstat_remove_entry(be_lun->disk_stats); 2548 2549 uma_zdestroy(be_lun->lun_zone); 2550 2551 ctl_free_opts(&be_lun->ctl_be_lun.options); 2552 free(be_lun->dev_path, M_CTLBLK); 2553 mtx_destroy(&be_lun->queue_lock); 2554 mtx_destroy(&be_lun->io_lock); 2555 free(be_lun, M_CTLBLK); 2556 2557 req->status = CTL_LUN_OK; 2558 2559 return (0); 2560 2561bailout_error: 2562 2563 req->status = CTL_LUN_ERROR; 2564 2565 return (0); 2566} 2567 2568static int 2569ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2570 struct ctl_lun_req *req) 2571{ 2572 struct vattr vattr; 2573 int error; 2574 struct ctl_lun_create_params *params = &be_lun->params; 2575 2576 if (params->lun_size_bytes != 0) { 2577 be_lun->size_bytes = params->lun_size_bytes; 2578 } else { 2579 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2580 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2581 VOP_UNLOCK(be_lun->vn, 0); 2582 if (error != 0) { 2583 snprintf(req->error_str, sizeof(req->error_str), 2584 "error calling VOP_GETATTR() for file %s", 2585 be_lun->dev_path); 2586 return (error); 2587 } 2588 2589 be_lun->size_bytes = vattr.va_size; 2590 } 2591 2592 return (0); 2593} 2594 2595static int 2596ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2597 struct ctl_lun_req *req) 2598{ 2599 struct ctl_be_block_devdata *dev_data; 2600 int error; 2601 struct ctl_lun_create_params *params = &be_lun->params; 2602 uint64_t size_bytes; 2603 2604 dev_data = &be_lun->backend.dev; 2605 if (!dev_data->csw->d_ioctl) { 2606 snprintf(req->error_str, sizeof(req->error_str), 2607 "no d_ioctl for device %s!", be_lun->dev_path); 2608 return (ENODEV); 2609 } 2610 2611 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2612 (caddr_t)&size_bytes, FREAD, 2613 curthread); 2614 if (error) { 2615 snprintf(req->error_str, sizeof(req->error_str), 2616 "error %d returned for DIOCGMEDIASIZE ioctl " 2617 "on %s!", error, be_lun->dev_path); 2618 return (error); 2619 } 2620 2621 if (params->lun_size_bytes != 0) { 2622 if (params->lun_size_bytes > size_bytes) { 2623 snprintf(req->error_str, sizeof(req->error_str), 2624 "requested LUN size %ju > backing device " 2625 "size %ju", 2626 (uintmax_t)params->lun_size_bytes, 2627 (uintmax_t)size_bytes); 2628 return (EINVAL); 2629 } 2630 2631 be_lun->size_bytes = params->lun_size_bytes; 2632 } else { 2633 be_lun->size_bytes = size_bytes; 2634 } 2635 2636 return (0); 2637} 2638 2639static int 2640ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2641{ 2642 struct ctl_lun_modify_params *params; 2643 struct ctl_be_block_lun *be_lun; 2644 uint64_t oldsize; 2645 int error; 2646 2647 params = &req->reqdata.modify; 2648 2649 mtx_lock(&softc->lock); 2650 be_lun = NULL; 2651 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2652 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2653 break; 2654 } 2655 mtx_unlock(&softc->lock); 2656 2657 if (be_lun == NULL) { 2658 snprintf(req->error_str, sizeof(req->error_str), 2659 "LUN %u is not managed by the block backend", 2660 params->lun_id); 2661 goto bailout_error; 2662 } 2663 2664 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2665 2666 oldsize = be_lun->size_bytes; 2667 if (be_lun->vn == NULL) 2668 error = ctl_be_block_open(softc, be_lun, req); 2669 else if (be_lun->vn->v_type == VREG) 2670 error = ctl_be_block_modify_file(be_lun, req); 2671 else 2672 error = ctl_be_block_modify_dev(be_lun, req); 2673 2674 if (error == 0 && be_lun->size_bytes != oldsize) { 2675 be_lun->size_blocks = be_lun->size_bytes >> 2676 be_lun->blocksize_shift; 2677 2678 /* 2679 * The maximum LBA is the size - 1. 2680 * 2681 * XXX: Note that this field is being updated without locking, 2682 * which might cause problems on 32-bit architectures. 2683 */ 2684 if (be_lun->unmap != NULL) 2685 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2686 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2687 0 : (be_lun->size_blocks - 1); 2688 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2689 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2690 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2691 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2692 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2693 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2694 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2695 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2696 if (oldsize == 0 && be_lun->size_blocks != 0) 2697 ctl_lun_online(&be_lun->ctl_be_lun); 2698 } 2699 2700 /* Tell the user the exact size we ended up using */ 2701 params->lun_size_bytes = be_lun->size_bytes; 2702 2703 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2704 2705 return (0); 2706 2707bailout_error: 2708 req->status = CTL_LUN_ERROR; 2709 2710 return (0); 2711} 2712 2713static void 2714ctl_be_block_lun_shutdown(void *be_lun) 2715{ 2716 struct ctl_be_block_lun *lun; 2717 struct ctl_be_block_softc *softc; 2718 2719 lun = (struct ctl_be_block_lun *)be_lun; 2720 2721 softc = lun->softc; 2722 2723 mtx_lock(&softc->lock); 2724 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2725 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2726 wakeup(lun); 2727 mtx_unlock(&softc->lock); 2728 2729} 2730 2731static void 2732ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2733{ 2734 struct ctl_be_block_lun *lun; 2735 struct ctl_be_block_softc *softc; 2736 2737 lun = (struct ctl_be_block_lun *)be_lun; 2738 softc = lun->softc; 2739 2740 if (status == CTL_LUN_CONFIG_OK) { 2741 mtx_lock(&softc->lock); 2742 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2743 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2744 wakeup(lun); 2745 mtx_unlock(&softc->lock); 2746 2747 /* 2748 * We successfully added the LUN, attempt to enable it. 2749 */ 2750 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2751 printf("%s: ctl_enable_lun() failed!\n", __func__); 2752 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2753 printf("%s: ctl_invalidate_lun() failed!\n", 2754 __func__); 2755 } 2756 } 2757 2758 return; 2759 } 2760 2761 2762 mtx_lock(&softc->lock); 2763 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2764 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2765 wakeup(lun); 2766 mtx_unlock(&softc->lock); 2767} 2768 2769 2770static int 2771ctl_be_block_config_write(union ctl_io *io) 2772{ 2773 struct ctl_be_block_lun *be_lun; 2774 struct ctl_be_lun *ctl_be_lun; 2775 int retval; 2776 2777 retval = 0; 2778 2779 DPRINTF("entered\n"); 2780 2781 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2782 CTL_PRIV_BACKEND_LUN].ptr; 2783 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2784 2785 switch (io->scsiio.cdb[0]) { 2786 case SYNCHRONIZE_CACHE: 2787 case SYNCHRONIZE_CACHE_16: 2788 case WRITE_SAME_10: 2789 case WRITE_SAME_16: 2790 case UNMAP: 2791 /* 2792 * The upper level CTL code will filter out any CDBs with 2793 * the immediate bit set and return the proper error. 2794 * 2795 * We don't really need to worry about what LBA range the 2796 * user asked to be synced out. When they issue a sync 2797 * cache command, we'll sync out the whole thing. 2798 */ 2799 mtx_lock(&be_lun->queue_lock); 2800 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2801 links); 2802 mtx_unlock(&be_lun->queue_lock); 2803 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2804 break; 2805 case START_STOP_UNIT: { 2806 struct scsi_start_stop_unit *cdb; 2807 2808 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2809 2810 if (cdb->how & SSS_START) 2811 retval = ctl_start_lun(ctl_be_lun); 2812 else { 2813 retval = ctl_stop_lun(ctl_be_lun); 2814 /* 2815 * XXX KDM Copan-specific offline behavior. 2816 * Figure out a reasonable way to port this? 2817 */ 2818#ifdef NEEDTOPORT 2819 if ((retval == 0) 2820 && (cdb->byte2 & SSS_ONOFFLINE)) 2821 retval = ctl_lun_offline(ctl_be_lun); 2822#endif 2823 } 2824 2825 /* 2826 * In general, the above routines should not fail. They 2827 * just set state for the LUN. So we've got something 2828 * pretty wrong here if we can't start or stop the LUN. 2829 */ 2830 if (retval != 0) { 2831 ctl_set_internal_failure(&io->scsiio, 2832 /*sks_valid*/ 1, 2833 /*retry_count*/ 0xf051); 2834 retval = CTL_RETVAL_COMPLETE; 2835 } else { 2836 ctl_set_success(&io->scsiio); 2837 } 2838 ctl_config_write_done(io); 2839 break; 2840 } 2841 default: 2842 ctl_set_invalid_opcode(&io->scsiio); 2843 ctl_config_write_done(io); 2844 retval = CTL_RETVAL_COMPLETE; 2845 break; 2846 } 2847 2848 return (retval); 2849} 2850 2851static int 2852ctl_be_block_config_read(union ctl_io *io) 2853{ 2854 struct ctl_be_block_lun *be_lun; 2855 struct ctl_be_lun *ctl_be_lun; 2856 int retval = 0; 2857 2858 DPRINTF("entered\n"); 2859 2860 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2861 CTL_PRIV_BACKEND_LUN].ptr; 2862 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2863 2864 switch (io->scsiio.cdb[0]) { 2865 case SERVICE_ACTION_IN: 2866 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2867 mtx_lock(&be_lun->queue_lock); 2868 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2869 &io->io_hdr, links); 2870 mtx_unlock(&be_lun->queue_lock); 2871 taskqueue_enqueue(be_lun->io_taskqueue, 2872 &be_lun->io_task); 2873 retval = CTL_RETVAL_QUEUED; 2874 break; 2875 } 2876 ctl_set_invalid_field(&io->scsiio, 2877 /*sks_valid*/ 1, 2878 /*command*/ 1, 2879 /*field*/ 1, 2880 /*bit_valid*/ 1, 2881 /*bit*/ 4); 2882 ctl_config_read_done(io); 2883 retval = CTL_RETVAL_COMPLETE; 2884 break; 2885 default: 2886 ctl_set_invalid_opcode(&io->scsiio); 2887 ctl_config_read_done(io); 2888 retval = CTL_RETVAL_COMPLETE; 2889 break; 2890 } 2891 2892 return (retval); 2893} 2894 2895static int 2896ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2897{ 2898 struct ctl_be_block_lun *lun; 2899 int retval; 2900 2901 lun = (struct ctl_be_block_lun *)be_lun; 2902 retval = 0; 2903 2904 retval = sbuf_printf(sb, "\t<num_threads>"); 2905 2906 if (retval != 0) 2907 goto bailout; 2908 2909 retval = sbuf_printf(sb, "%d", lun->num_threads); 2910 2911 if (retval != 0) 2912 goto bailout; 2913 2914 retval = sbuf_printf(sb, "</num_threads>\n"); 2915 2916bailout: 2917 2918 return (retval); 2919} 2920 2921static uint64_t 2922ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2923{ 2924 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2925 2926 if (lun->getattr == NULL) 2927 return (UINT64_MAX); 2928 return (lun->getattr(lun, attrname)); 2929} 2930 2931int 2932ctl_be_block_init(void) 2933{ 2934 struct ctl_be_block_softc *softc; 2935 int retval; 2936 2937 softc = &backend_block_softc; 2938 retval = 0; 2939 2940 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2941 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2942 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2943 STAILQ_INIT(&softc->disk_list); 2944 STAILQ_INIT(&softc->lun_list); 2945 2946 return (retval); 2947} 2948