ctl_backend_block.c revision 274004
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 274004 2014-11-03 03:45:41Z 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/proc.h> 74#include <sys/pcpu.h> 75#include <sys/module.h> 76#include <sys/sdt.h> 77#include <sys/devicestat.h> 78#include <sys/sysctl.h> 79 80#include <geom/geom.h> 81 82#include <cam/cam.h> 83#include <cam/scsi/scsi_all.h> 84#include <cam/scsi/scsi_da.h> 85#include <cam/ctl/ctl_io.h> 86#include <cam/ctl/ctl.h> 87#include <cam/ctl/ctl_backend.h> 88#include <cam/ctl/ctl_frontend_internal.h> 89#include <cam/ctl/ctl_ioctl.h> 90#include <cam/ctl/ctl_scsi_all.h> 91#include <cam/ctl/ctl_error.h> 92 93/* 94 * The idea here is that we'll allocate enough S/G space to hold a 1MB 95 * I/O. If we get an I/O larger than that, we'll split it. 96 */ 97#define CTLBLK_HALF_IO_SIZE (512 * 1024) 98#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 99#define CTLBLK_MAX_SEG MAXPHYS 100#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 101#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 102 103#ifdef CTLBLK_DEBUG 104#define DPRINTF(fmt, args...) \ 105 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 106#else 107#define DPRINTF(fmt, args...) do {} while(0) 108#endif 109 110#define PRIV(io) \ 111 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 112#define ARGS(io) \ 113 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 114 115SDT_PROVIDER_DEFINE(cbb); 116 117typedef enum { 118 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 119 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 120 CTL_BE_BLOCK_LUN_WAITING = 0x04, 121 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 122} ctl_be_block_lun_flags; 123 124typedef enum { 125 CTL_BE_BLOCK_NONE, 126 CTL_BE_BLOCK_DEV, 127 CTL_BE_BLOCK_FILE 128} ctl_be_block_type; 129 130struct ctl_be_block_devdata { 131 struct cdev *cdev; 132 struct cdevsw *csw; 133 int dev_ref; 134}; 135 136struct ctl_be_block_filedata { 137 struct ucred *cred; 138}; 139 140union ctl_be_block_bedata { 141 struct ctl_be_block_devdata dev; 142 struct ctl_be_block_filedata file; 143}; 144 145struct ctl_be_block_io; 146struct ctl_be_block_lun; 147 148typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 149 struct ctl_be_block_io *beio); 150 151/* 152 * Backend LUN structure. There is a 1:1 mapping between a block device 153 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 154 */ 155struct ctl_be_block_lun { 156 struct ctl_lun_create_params params; 157 struct ctl_block_disk *disk; 158 char lunname[32]; 159 char *dev_path; 160 ctl_be_block_type dev_type; 161 struct vnode *vn; 162 union ctl_be_block_bedata backend; 163 cbb_dispatch_t dispatch; 164 cbb_dispatch_t lun_flush; 165 cbb_dispatch_t unmap; 166 uma_zone_t lun_zone; 167 uint64_t size_blocks; 168 uint64_t size_bytes; 169 uint32_t blocksize; 170 int blocksize_shift; 171 uint16_t pblockexp; 172 uint16_t pblockoff; 173 struct ctl_be_block_softc *softc; 174 struct devstat *disk_stats; 175 ctl_be_block_lun_flags flags; 176 STAILQ_ENTRY(ctl_be_block_lun) links; 177 struct ctl_be_lun ctl_be_lun; 178 struct taskqueue *io_taskqueue; 179 struct task io_task; 180 int num_threads; 181 STAILQ_HEAD(, ctl_io_hdr) input_queue; 182 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 183 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 184 struct mtx_padalign io_lock; 185 struct mtx_padalign queue_lock; 186}; 187 188/* 189 * Overall softc structure for the block backend module. 190 */ 191struct ctl_be_block_softc { 192 struct mtx lock; 193 int num_disks; 194 STAILQ_HEAD(, ctl_block_disk) disk_list; 195 int num_luns; 196 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 197}; 198 199static struct ctl_be_block_softc backend_block_softc; 200 201/* 202 * Per-I/O information. 203 */ 204struct ctl_be_block_io { 205 union ctl_io *io; 206 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 207 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 208 int bio_cmd; 209 int num_segs; 210 int num_bios_sent; 211 int num_bios_done; 212 int send_complete; 213 int num_errors; 214 struct bintime ds_t0; 215 devstat_tag_type ds_tag_type; 216 devstat_trans_flags ds_trans_type; 217 uint64_t io_len; 218 uint64_t io_offset; 219 struct ctl_be_block_softc *softc; 220 struct ctl_be_block_lun *lun; 221 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 222}; 223 224static int cbb_num_threads = 14; 225TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 226SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 227 "CAM Target Layer Block Backend"); 228SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 229 &cbb_num_threads, 0, "Number of threads per backing file"); 230 231static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 232static void ctl_free_beio(struct ctl_be_block_io *beio); 233static void ctl_complete_beio(struct ctl_be_block_io *beio); 234static int ctl_be_block_move_done(union ctl_io *io); 235static void ctl_be_block_biodone(struct bio *bio); 236static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 237 struct ctl_be_block_io *beio); 238static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 239 struct ctl_be_block_io *beio); 240static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 241 struct ctl_be_block_io *beio); 242static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 243 struct ctl_be_block_io *beio); 244static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 245 struct ctl_be_block_io *beio); 246static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 247 union ctl_io *io); 248static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 249 union ctl_io *io); 250static void ctl_be_block_worker(void *context, int pending); 251static int ctl_be_block_submit(union ctl_io *io); 252static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 253 int flag, struct thread *td); 254static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 255 struct ctl_lun_req *req); 256static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 257 struct ctl_lun_req *req); 258static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 259static int ctl_be_block_open(struct ctl_be_block_softc *softc, 260 struct ctl_be_block_lun *be_lun, 261 struct ctl_lun_req *req); 262static int ctl_be_block_create(struct ctl_be_block_softc *softc, 263 struct ctl_lun_req *req); 264static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 265 struct ctl_lun_req *req); 266static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 267 struct ctl_lun_req *req); 268static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 269 struct ctl_lun_req *req); 270static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 271 struct ctl_lun_req *req); 272static void ctl_be_block_lun_shutdown(void *be_lun); 273static void ctl_be_block_lun_config_status(void *be_lun, 274 ctl_lun_config_status status); 275static int ctl_be_block_config_write(union ctl_io *io); 276static int ctl_be_block_config_read(union ctl_io *io); 277static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 278int ctl_be_block_init(void); 279 280static struct ctl_backend_driver ctl_be_block_driver = 281{ 282 .name = "block", 283 .flags = CTL_BE_FLAG_HAS_CONFIG, 284 .init = ctl_be_block_init, 285 .data_submit = ctl_be_block_submit, 286 .data_move_done = ctl_be_block_move_done, 287 .config_read = ctl_be_block_config_read, 288 .config_write = ctl_be_block_config_write, 289 .ioctl = ctl_be_block_ioctl, 290 .lun_info = ctl_be_block_lun_info 291}; 292 293MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 294CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 295 296static uma_zone_t beio_zone; 297 298static struct ctl_be_block_io * 299ctl_alloc_beio(struct ctl_be_block_softc *softc) 300{ 301 struct ctl_be_block_io *beio; 302 303 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 304 beio->softc = softc; 305 return (beio); 306} 307 308static void 309ctl_free_beio(struct ctl_be_block_io *beio) 310{ 311 int duplicate_free; 312 int i; 313 314 duplicate_free = 0; 315 316 for (i = 0; i < beio->num_segs; i++) { 317 if (beio->sg_segs[i].addr == NULL) 318 duplicate_free++; 319 320 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 321 beio->sg_segs[i].addr = NULL; 322 323 /* For compare we had two equal S/G lists. */ 324 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 325 uma_zfree(beio->lun->lun_zone, 326 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 327 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 328 } 329 } 330 331 if (duplicate_free > 0) { 332 printf("%s: %d duplicate frees out of %d segments\n", __func__, 333 duplicate_free, beio->num_segs); 334 } 335 336 uma_zfree(beio_zone, beio); 337} 338 339static void 340ctl_complete_beio(struct ctl_be_block_io *beio) 341{ 342 union ctl_io *io = beio->io; 343 344 if (beio->beio_cont != NULL) { 345 beio->beio_cont(beio); 346 } else { 347 ctl_free_beio(beio); 348 ctl_data_submit_done(io); 349 } 350} 351 352static int 353ctl_be_block_move_done(union ctl_io *io) 354{ 355 struct ctl_be_block_io *beio; 356 struct ctl_be_block_lun *be_lun; 357 struct ctl_lba_len_flags *lbalen; 358#ifdef CTL_TIME_IO 359 struct bintime cur_bt; 360#endif 361 int i; 362 363 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 364 be_lun = beio->lun; 365 366 DPRINTF("entered\n"); 367 368#ifdef CTL_TIME_IO 369 getbintime(&cur_bt); 370 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 371 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 372 io->io_hdr.num_dmas++; 373#endif 374 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 375 376 /* 377 * We set status at this point for read commands, and write 378 * commands with errors. 379 */ 380 if ((io->io_hdr.port_status == 0) && 381 ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 382 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 383 lbalen = ARGS(beio->io); 384 if (lbalen->flags & CTL_LLF_READ) { 385 ctl_set_success(&io->scsiio); 386 } else if (lbalen->flags & CTL_LLF_COMPARE) { 387 /* We have two data blocks ready for comparison. */ 388 for (i = 0; i < beio->num_segs; i++) { 389 if (memcmp(beio->sg_segs[i].addr, 390 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 391 beio->sg_segs[i].len) != 0) 392 break; 393 } 394 if (i < beio->num_segs) 395 ctl_set_sense(&io->scsiio, 396 /*current_error*/ 1, 397 /*sense_key*/ SSD_KEY_MISCOMPARE, 398 /*asc*/ 0x1D, 399 /*ascq*/ 0x00, 400 SSD_ELEM_NONE); 401 else 402 ctl_set_success(&io->scsiio); 403 } 404 } 405 else if ((io->io_hdr.port_status != 0) 406 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 407 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 408 /* 409 * For hardware error sense keys, the sense key 410 * specific value is defined to be a retry count, 411 * but we use it to pass back an internal FETD 412 * error code. XXX KDM Hopefully the FETD is only 413 * using 16 bits for an error code, since that's 414 * all the space we have in the sks field. 415 */ 416 ctl_set_internal_failure(&io->scsiio, 417 /*sks_valid*/ 1, 418 /*retry_count*/ 419 io->io_hdr.port_status); 420 } 421 422 /* 423 * If this is a read, or a write with errors, it is done. 424 */ 425 if ((beio->bio_cmd == BIO_READ) 426 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 427 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 428 ctl_complete_beio(beio); 429 return (0); 430 } 431 432 /* 433 * At this point, we have a write and the DMA completed 434 * successfully. We now have to queue it to the task queue to 435 * execute the backend I/O. That is because we do blocking 436 * memory allocations, and in the file backing case, blocking I/O. 437 * This move done routine is generally called in the SIM's 438 * interrupt context, and therefore we cannot block. 439 */ 440 mtx_lock(&be_lun->queue_lock); 441 /* 442 * XXX KDM make sure that links is okay to use at this point. 443 * Otherwise, we either need to add another field to ctl_io_hdr, 444 * or deal with resource allocation here. 445 */ 446 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 447 mtx_unlock(&be_lun->queue_lock); 448 449 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 450 451 return (0); 452} 453 454static void 455ctl_be_block_biodone(struct bio *bio) 456{ 457 struct ctl_be_block_io *beio; 458 struct ctl_be_block_lun *be_lun; 459 union ctl_io *io; 460 int error; 461 462 beio = bio->bio_caller1; 463 be_lun = beio->lun; 464 io = beio->io; 465 466 DPRINTF("entered\n"); 467 468 error = bio->bio_error; 469 mtx_lock(&be_lun->io_lock); 470 if (error != 0) 471 beio->num_errors++; 472 473 beio->num_bios_done++; 474 475 /* 476 * XXX KDM will this cause WITNESS to complain? Holding a lock 477 * during the free might cause it to complain. 478 */ 479 g_destroy_bio(bio); 480 481 /* 482 * If the send complete bit isn't set, or we aren't the last I/O to 483 * complete, then we're done. 484 */ 485 if ((beio->send_complete == 0) 486 || (beio->num_bios_done < beio->num_bios_sent)) { 487 mtx_unlock(&be_lun->io_lock); 488 return; 489 } 490 491 /* 492 * At this point, we've verified that we are the last I/O to 493 * complete, so it's safe to drop the lock. 494 */ 495 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 496 beio->ds_tag_type, beio->ds_trans_type, 497 /*now*/ NULL, /*then*/&beio->ds_t0); 498 mtx_unlock(&be_lun->io_lock); 499 500 /* 501 * If there are any errors from the backing device, we fail the 502 * entire I/O with a medium error. 503 */ 504 if (beio->num_errors > 0) { 505 if (error == EOPNOTSUPP) { 506 ctl_set_invalid_opcode(&io->scsiio); 507 } else if (error == ENOSPC) { 508 ctl_set_space_alloc_fail(&io->scsiio); 509 } else if (beio->bio_cmd == BIO_FLUSH) { 510 /* XXX KDM is there is a better error here? */ 511 ctl_set_internal_failure(&io->scsiio, 512 /*sks_valid*/ 1, 513 /*retry_count*/ 0xbad2); 514 } else 515 ctl_set_medium_error(&io->scsiio); 516 ctl_complete_beio(beio); 517 return; 518 } 519 520 /* 521 * If this is a write, a flush, a delete or verify, we're all done. 522 * If this is a read, we can now send the data to the user. 523 */ 524 if ((beio->bio_cmd == BIO_WRITE) 525 || (beio->bio_cmd == BIO_FLUSH) 526 || (beio->bio_cmd == BIO_DELETE) 527 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 528 ctl_set_success(&io->scsiio); 529 ctl_complete_beio(beio); 530 } else { 531#ifdef CTL_TIME_IO 532 getbintime(&io->io_hdr.dma_start_bt); 533#endif 534 ctl_datamove(io); 535 } 536} 537 538static void 539ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 540 struct ctl_be_block_io *beio) 541{ 542 union ctl_io *io = beio->io; 543 struct mount *mountpoint; 544 int error, lock_flags; 545 546 DPRINTF("entered\n"); 547 548 binuptime(&beio->ds_t0); 549 mtx_lock(&be_lun->io_lock); 550 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 551 mtx_unlock(&be_lun->io_lock); 552 553 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 554 555 if (MNT_SHARED_WRITES(mountpoint) 556 || ((mountpoint == NULL) 557 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 558 lock_flags = LK_SHARED; 559 else 560 lock_flags = LK_EXCLUSIVE; 561 562 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 563 564 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 565 VOP_UNLOCK(be_lun->vn, 0); 566 567 vn_finished_write(mountpoint); 568 569 mtx_lock(&be_lun->io_lock); 570 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 571 beio->ds_tag_type, beio->ds_trans_type, 572 /*now*/ NULL, /*then*/&beio->ds_t0); 573 mtx_unlock(&be_lun->io_lock); 574 575 if (error == 0) 576 ctl_set_success(&io->scsiio); 577 else { 578 /* XXX KDM is there is a better error here? */ 579 ctl_set_internal_failure(&io->scsiio, 580 /*sks_valid*/ 1, 581 /*retry_count*/ 0xbad1); 582 } 583 584 ctl_complete_beio(beio); 585} 586 587SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 588SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 589SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 590SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 591 592static void 593ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 594 struct ctl_be_block_io *beio) 595{ 596 struct ctl_be_block_filedata *file_data; 597 union ctl_io *io; 598 struct uio xuio; 599 struct iovec *xiovec; 600 int flags; 601 int error, i; 602 603 DPRINTF("entered\n"); 604 605 file_data = &be_lun->backend.file; 606 io = beio->io; 607 flags = 0; 608 if (ARGS(io)->flags & CTL_LLF_DPO) 609 flags |= IO_DIRECT; 610 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 611 flags |= IO_SYNC; 612 613 bzero(&xuio, sizeof(xuio)); 614 if (beio->bio_cmd == BIO_READ) { 615 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 616 xuio.uio_rw = UIO_READ; 617 } else { 618 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 619 xuio.uio_rw = UIO_WRITE; 620 } 621 xuio.uio_offset = beio->io_offset; 622 xuio.uio_resid = beio->io_len; 623 xuio.uio_segflg = UIO_SYSSPACE; 624 xuio.uio_iov = beio->xiovecs; 625 xuio.uio_iovcnt = beio->num_segs; 626 xuio.uio_td = curthread; 627 628 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 629 xiovec->iov_base = beio->sg_segs[i].addr; 630 xiovec->iov_len = beio->sg_segs[i].len; 631 } 632 633 binuptime(&beio->ds_t0); 634 mtx_lock(&be_lun->io_lock); 635 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 636 mtx_unlock(&be_lun->io_lock); 637 638 if (beio->bio_cmd == BIO_READ) { 639 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 640 641 /* 642 * UFS pays attention to IO_DIRECT for reads. If the 643 * DIRECTIO option is configured into the kernel, it calls 644 * ffs_rawread(). But that only works for single-segment 645 * uios with user space addresses. In our case, with a 646 * kernel uio, it still reads into the buffer cache, but it 647 * will just try to release the buffer from the cache later 648 * on in ffs_read(). 649 * 650 * ZFS does not pay attention to IO_DIRECT for reads. 651 * 652 * UFS does not pay attention to IO_SYNC for reads. 653 * 654 * ZFS pays attention to IO_SYNC (which translates into the 655 * Solaris define FRSYNC for zfs_read()) for reads. It 656 * attempts to sync the file before reading. 657 * 658 * So, to attempt to provide some barrier semantics in the 659 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 660 */ 661 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 662 663 VOP_UNLOCK(be_lun->vn, 0); 664 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 665 } else { 666 struct mount *mountpoint; 667 int lock_flags; 668 669 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 670 671 if (MNT_SHARED_WRITES(mountpoint) 672 || ((mountpoint == NULL) 673 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 674 lock_flags = LK_SHARED; 675 else 676 lock_flags = LK_EXCLUSIVE; 677 678 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 679 680 /* 681 * UFS pays attention to IO_DIRECT for writes. The write 682 * is done asynchronously. (Normally the write would just 683 * get put into cache. 684 * 685 * UFS pays attention to IO_SYNC for writes. It will 686 * attempt to write the buffer out synchronously if that 687 * flag is set. 688 * 689 * ZFS does not pay attention to IO_DIRECT for writes. 690 * 691 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 692 * for writes. It will flush the transaction from the 693 * cache before returning. 694 * 695 * So if we've got the BIO_ORDERED flag set, we want 696 * IO_SYNC in either the UFS or ZFS case. 697 */ 698 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 699 VOP_UNLOCK(be_lun->vn, 0); 700 701 vn_finished_write(mountpoint); 702 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 703 } 704 705 mtx_lock(&be_lun->io_lock); 706 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 707 beio->ds_tag_type, beio->ds_trans_type, 708 /*now*/ NULL, /*then*/&beio->ds_t0); 709 mtx_unlock(&be_lun->io_lock); 710 711 /* 712 * If we got an error, set the sense data to "MEDIUM ERROR" and 713 * return the I/O to the user. 714 */ 715 if (error != 0) { 716 char path_str[32]; 717 718 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 719 printf("%s%s command returned errno %d\n", path_str, 720 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 721 if (error == ENOSPC) { 722 ctl_set_space_alloc_fail(&io->scsiio); 723 } else 724 ctl_set_medium_error(&io->scsiio); 725 ctl_complete_beio(beio); 726 return; 727 } 728 729 /* 730 * If this is a write or a verify, we're all done. 731 * If this is a read, we can now send the data to the user. 732 */ 733 if ((beio->bio_cmd == BIO_WRITE) || 734 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 735 ctl_set_success(&io->scsiio); 736 ctl_complete_beio(beio); 737 } else { 738#ifdef CTL_TIME_IO 739 getbintime(&io->io_hdr.dma_start_bt); 740#endif 741 ctl_datamove(io); 742 } 743} 744 745static void 746ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 747 struct ctl_be_block_io *beio) 748{ 749 struct ctl_be_block_devdata *dev_data; 750 union ctl_io *io; 751 struct uio xuio; 752 struct iovec *xiovec; 753 int flags; 754 int error, i; 755 756 DPRINTF("entered\n"); 757 758 dev_data = &be_lun->backend.dev; 759 io = beio->io; 760 flags = 0; 761 if (ARGS(io)->flags & CTL_LLF_DPO) 762 flags |= IO_DIRECT; 763 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 764 flags |= IO_SYNC; 765 766 bzero(&xuio, sizeof(xuio)); 767 if (beio->bio_cmd == BIO_READ) { 768 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 769 xuio.uio_rw = UIO_READ; 770 } else { 771 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 772 xuio.uio_rw = UIO_WRITE; 773 } 774 xuio.uio_offset = beio->io_offset; 775 xuio.uio_resid = beio->io_len; 776 xuio.uio_segflg = UIO_SYSSPACE; 777 xuio.uio_iov = beio->xiovecs; 778 xuio.uio_iovcnt = beio->num_segs; 779 xuio.uio_td = curthread; 780 781 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 782 xiovec->iov_base = beio->sg_segs[i].addr; 783 xiovec->iov_len = beio->sg_segs[i].len; 784 } 785 786 binuptime(&beio->ds_t0); 787 mtx_lock(&be_lun->io_lock); 788 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 789 mtx_unlock(&be_lun->io_lock); 790 791 if (beio->bio_cmd == BIO_READ) { 792 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 793 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 794 } else { 795 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 796 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 797 } 798 799 mtx_lock(&be_lun->io_lock); 800 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 801 beio->ds_tag_type, beio->ds_trans_type, 802 /*now*/ NULL, /*then*/&beio->ds_t0); 803 mtx_unlock(&be_lun->io_lock); 804 805 /* 806 * If we got an error, set the sense data to "MEDIUM ERROR" and 807 * return the I/O to the user. 808 */ 809 if (error != 0) { 810 if (error == ENOSPC) { 811 ctl_set_space_alloc_fail(&io->scsiio); 812 } else 813 ctl_set_medium_error(&io->scsiio); 814 ctl_complete_beio(beio); 815 return; 816 } 817 818 /* 819 * If this is a write or a verify, we're all done. 820 * If this is a read, we can now send the data to the user. 821 */ 822 if ((beio->bio_cmd == BIO_WRITE) || 823 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 824 ctl_set_success(&io->scsiio); 825 ctl_complete_beio(beio); 826 } else { 827#ifdef CTL_TIME_IO 828 getbintime(&io->io_hdr.dma_start_bt); 829#endif 830 ctl_datamove(io); 831 } 832} 833 834static void 835ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 836 struct ctl_be_block_io *beio) 837{ 838 struct bio *bio; 839 union ctl_io *io; 840 struct ctl_be_block_devdata *dev_data; 841 842 dev_data = &be_lun->backend.dev; 843 io = beio->io; 844 845 DPRINTF("entered\n"); 846 847 /* This can't fail, it's a blocking allocation. */ 848 bio = g_alloc_bio(); 849 850 bio->bio_cmd = BIO_FLUSH; 851 bio->bio_flags |= BIO_ORDERED; 852 bio->bio_dev = dev_data->cdev; 853 bio->bio_offset = 0; 854 bio->bio_data = 0; 855 bio->bio_done = ctl_be_block_biodone; 856 bio->bio_caller1 = beio; 857 bio->bio_pblkno = 0; 858 859 /* 860 * We don't need to acquire the LUN lock here, because we are only 861 * sending one bio, and so there is no other context to synchronize 862 * with. 863 */ 864 beio->num_bios_sent = 1; 865 beio->send_complete = 1; 866 867 binuptime(&beio->ds_t0); 868 mtx_lock(&be_lun->io_lock); 869 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 870 mtx_unlock(&be_lun->io_lock); 871 872 (*dev_data->csw->d_strategy)(bio); 873} 874 875static void 876ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 877 struct ctl_be_block_io *beio, 878 uint64_t off, uint64_t len, int last) 879{ 880 struct bio *bio; 881 struct ctl_be_block_devdata *dev_data; 882 uint64_t maxlen; 883 884 dev_data = &be_lun->backend.dev; 885 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 886 while (len > 0) { 887 bio = g_alloc_bio(); 888 bio->bio_cmd = BIO_DELETE; 889 bio->bio_dev = dev_data->cdev; 890 bio->bio_offset = off; 891 bio->bio_length = MIN(len, maxlen); 892 bio->bio_data = 0; 893 bio->bio_done = ctl_be_block_biodone; 894 bio->bio_caller1 = beio; 895 bio->bio_pblkno = off / be_lun->blocksize; 896 897 off += bio->bio_length; 898 len -= bio->bio_length; 899 900 mtx_lock(&be_lun->io_lock); 901 beio->num_bios_sent++; 902 if (last && len == 0) 903 beio->send_complete = 1; 904 mtx_unlock(&be_lun->io_lock); 905 906 (*dev_data->csw->d_strategy)(bio); 907 } 908} 909 910static void 911ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 912 struct ctl_be_block_io *beio) 913{ 914 union ctl_io *io; 915 struct ctl_be_block_devdata *dev_data; 916 struct ctl_ptr_len_flags *ptrlen; 917 struct scsi_unmap_desc *buf, *end; 918 uint64_t len; 919 920 dev_data = &be_lun->backend.dev; 921 io = beio->io; 922 923 DPRINTF("entered\n"); 924 925 binuptime(&beio->ds_t0); 926 mtx_lock(&be_lun->io_lock); 927 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 928 mtx_unlock(&be_lun->io_lock); 929 930 if (beio->io_offset == -1) { 931 beio->io_len = 0; 932 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 933 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 934 end = buf + ptrlen->len / sizeof(*buf); 935 for (; buf < end; buf++) { 936 len = (uint64_t)scsi_4btoul(buf->length) * 937 be_lun->blocksize; 938 beio->io_len += len; 939 ctl_be_block_unmap_dev_range(be_lun, beio, 940 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 941 (end - buf < 2) ? TRUE : FALSE); 942 } 943 } else 944 ctl_be_block_unmap_dev_range(be_lun, beio, 945 beio->io_offset, beio->io_len, TRUE); 946} 947 948static void 949ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 950 struct ctl_be_block_io *beio) 951{ 952 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 953 int i; 954 struct bio *bio; 955 struct ctl_be_block_devdata *dev_data; 956 off_t cur_offset; 957 int max_iosize; 958 959 DPRINTF("entered\n"); 960 961 dev_data = &be_lun->backend.dev; 962 963 /* 964 * We have to limit our I/O size to the maximum supported by the 965 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 966 * set it properly, use DFLTPHYS. 967 */ 968 max_iosize = dev_data->cdev->si_iosize_max; 969 if (max_iosize < PAGE_SIZE) 970 max_iosize = DFLTPHYS; 971 972 cur_offset = beio->io_offset; 973 for (i = 0; i < beio->num_segs; i++) { 974 size_t cur_size; 975 uint8_t *cur_ptr; 976 977 cur_size = beio->sg_segs[i].len; 978 cur_ptr = beio->sg_segs[i].addr; 979 980 while (cur_size > 0) { 981 /* This can't fail, it's a blocking allocation. */ 982 bio = g_alloc_bio(); 983 984 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 985 986 bio->bio_cmd = beio->bio_cmd; 987 bio->bio_dev = dev_data->cdev; 988 bio->bio_caller1 = beio; 989 bio->bio_length = min(cur_size, max_iosize); 990 bio->bio_offset = cur_offset; 991 bio->bio_data = cur_ptr; 992 bio->bio_done = ctl_be_block_biodone; 993 bio->bio_pblkno = cur_offset / be_lun->blocksize; 994 995 cur_offset += bio->bio_length; 996 cur_ptr += bio->bio_length; 997 cur_size -= bio->bio_length; 998 999 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1000 beio->num_bios_sent++; 1001 } 1002 } 1003 binuptime(&beio->ds_t0); 1004 mtx_lock(&be_lun->io_lock); 1005 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1006 beio->send_complete = 1; 1007 mtx_unlock(&be_lun->io_lock); 1008 1009 /* 1010 * Fire off all allocated requests! 1011 */ 1012 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1013 TAILQ_REMOVE(&queue, bio, bio_queue); 1014 (*dev_data->csw->d_strategy)(bio); 1015 } 1016} 1017 1018static void 1019ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1020{ 1021 union ctl_io *io; 1022 1023 io = beio->io; 1024 ctl_free_beio(beio); 1025 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1026 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1027 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1028 ctl_config_write_done(io); 1029 return; 1030 } 1031 1032 ctl_be_block_config_write(io); 1033} 1034 1035static void 1036ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1037 union ctl_io *io) 1038{ 1039 struct ctl_be_block_io *beio; 1040 struct ctl_be_block_softc *softc; 1041 struct ctl_lba_len_flags *lbalen; 1042 uint64_t len_left, lba; 1043 int i, seglen; 1044 uint8_t *buf, *end; 1045 1046 DPRINTF("entered\n"); 1047 1048 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1049 softc = be_lun->softc; 1050 lbalen = ARGS(beio->io); 1051 1052 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1053 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1054 ctl_free_beio(beio); 1055 ctl_set_invalid_field(&io->scsiio, 1056 /*sks_valid*/ 1, 1057 /*command*/ 1, 1058 /*field*/ 1, 1059 /*bit_valid*/ 0, 1060 /*bit*/ 0); 1061 ctl_config_write_done(io); 1062 return; 1063 } 1064 1065 switch (io->scsiio.tag_type) { 1066 case CTL_TAG_ORDERED: 1067 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1068 break; 1069 case CTL_TAG_HEAD_OF_QUEUE: 1070 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1071 break; 1072 case CTL_TAG_UNTAGGED: 1073 case CTL_TAG_SIMPLE: 1074 case CTL_TAG_ACA: 1075 default: 1076 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1077 break; 1078 } 1079 1080 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1081 beio->io_offset = lbalen->lba * be_lun->blocksize; 1082 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1083 beio->bio_cmd = BIO_DELETE; 1084 beio->ds_trans_type = DEVSTAT_FREE; 1085 1086 be_lun->unmap(be_lun, beio); 1087 return; 1088 } 1089 1090 beio->bio_cmd = BIO_WRITE; 1091 beio->ds_trans_type = DEVSTAT_WRITE; 1092 1093 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1094 (uintmax_t)lbalen->lba, lbalen->len); 1095 1096 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1097 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1098 1099 /* 1100 * Setup the S/G entry for this chunk. 1101 */ 1102 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1103 seglen -= seglen % be_lun->blocksize; 1104 beio->sg_segs[i].len = seglen; 1105 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1106 1107 DPRINTF("segment %d addr %p len %zd\n", i, 1108 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1109 1110 beio->num_segs++; 1111 len_left -= seglen; 1112 1113 buf = beio->sg_segs[i].addr; 1114 end = buf + seglen; 1115 for (; buf < end; buf += be_lun->blocksize) { 1116 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1117 if (lbalen->flags & SWS_LBDATA) 1118 scsi_ulto4b(lbalen->lba + lba, buf); 1119 lba++; 1120 } 1121 } 1122 1123 beio->io_offset = lbalen->lba * be_lun->blocksize; 1124 beio->io_len = lba * be_lun->blocksize; 1125 1126 /* We can not do all in one run. Correct and schedule rerun. */ 1127 if (len_left > 0) { 1128 lbalen->lba += lba; 1129 lbalen->len -= lba; 1130 beio->beio_cont = ctl_be_block_cw_done_ws; 1131 } 1132 1133 be_lun->dispatch(be_lun, beio); 1134} 1135 1136static void 1137ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1138 union ctl_io *io) 1139{ 1140 struct ctl_be_block_io *beio; 1141 struct ctl_be_block_softc *softc; 1142 struct ctl_ptr_len_flags *ptrlen; 1143 1144 DPRINTF("entered\n"); 1145 1146 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1147 softc = be_lun->softc; 1148 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1149 1150 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1151 ctl_free_beio(beio); 1152 ctl_set_invalid_field(&io->scsiio, 1153 /*sks_valid*/ 0, 1154 /*command*/ 1, 1155 /*field*/ 0, 1156 /*bit_valid*/ 0, 1157 /*bit*/ 0); 1158 ctl_config_write_done(io); 1159 return; 1160 } 1161 1162 switch (io->scsiio.tag_type) { 1163 case CTL_TAG_ORDERED: 1164 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1165 break; 1166 case CTL_TAG_HEAD_OF_QUEUE: 1167 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1168 break; 1169 case CTL_TAG_UNTAGGED: 1170 case CTL_TAG_SIMPLE: 1171 case CTL_TAG_ACA: 1172 default: 1173 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1174 break; 1175 } 1176 1177 beio->io_len = 0; 1178 beio->io_offset = -1; 1179 1180 beio->bio_cmd = BIO_DELETE; 1181 beio->ds_trans_type = DEVSTAT_FREE; 1182 1183 DPRINTF("UNMAP\n"); 1184 1185 be_lun->unmap(be_lun, beio); 1186} 1187 1188static void 1189ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1190{ 1191 union ctl_io *io; 1192 1193 io = beio->io; 1194 ctl_free_beio(beio); 1195 ctl_config_write_done(io); 1196} 1197 1198static void 1199ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1200 union ctl_io *io) 1201{ 1202 struct ctl_be_block_io *beio; 1203 struct ctl_be_block_softc *softc; 1204 1205 DPRINTF("entered\n"); 1206 1207 softc = be_lun->softc; 1208 beio = ctl_alloc_beio(softc); 1209 beio->io = io; 1210 beio->lun = be_lun; 1211 beio->beio_cont = ctl_be_block_cw_done; 1212 PRIV(io)->ptr = (void *)beio; 1213 1214 switch (io->scsiio.cdb[0]) { 1215 case SYNCHRONIZE_CACHE: 1216 case SYNCHRONIZE_CACHE_16: 1217 beio->bio_cmd = BIO_FLUSH; 1218 beio->ds_trans_type = DEVSTAT_NO_DATA; 1219 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1220 beio->io_len = 0; 1221 be_lun->lun_flush(be_lun, beio); 1222 break; 1223 case WRITE_SAME_10: 1224 case WRITE_SAME_16: 1225 ctl_be_block_cw_dispatch_ws(be_lun, io); 1226 break; 1227 case UNMAP: 1228 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1229 break; 1230 default: 1231 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1232 break; 1233 } 1234} 1235 1236SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1237SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1238SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1239SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1240 1241static void 1242ctl_be_block_next(struct ctl_be_block_io *beio) 1243{ 1244 struct ctl_be_block_lun *be_lun; 1245 union ctl_io *io; 1246 1247 io = beio->io; 1248 be_lun = beio->lun; 1249 ctl_free_beio(beio); 1250 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1251 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1252 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1253 ctl_data_submit_done(io); 1254 return; 1255 } 1256 1257 io->io_hdr.status &= ~CTL_STATUS_MASK; 1258 io->io_hdr.status |= CTL_STATUS_NONE; 1259 1260 mtx_lock(&be_lun->queue_lock); 1261 /* 1262 * XXX KDM make sure that links is okay to use at this point. 1263 * Otherwise, we either need to add another field to ctl_io_hdr, 1264 * or deal with resource allocation here. 1265 */ 1266 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1267 mtx_unlock(&be_lun->queue_lock); 1268 1269 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1270} 1271 1272static void 1273ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1274 union ctl_io *io) 1275{ 1276 struct ctl_be_block_io *beio; 1277 struct ctl_be_block_softc *softc; 1278 struct ctl_lba_len_flags *lbalen; 1279 struct ctl_ptr_len_flags *bptrlen; 1280 uint64_t len_left, lbas; 1281 int i; 1282 1283 softc = be_lun->softc; 1284 1285 DPRINTF("entered\n"); 1286 1287 lbalen = ARGS(io); 1288 if (lbalen->flags & CTL_LLF_WRITE) { 1289 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1290 } else { 1291 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1292 } 1293 1294 beio = ctl_alloc_beio(softc); 1295 beio->io = io; 1296 beio->lun = be_lun; 1297 bptrlen = PRIV(io); 1298 bptrlen->ptr = (void *)beio; 1299 1300 switch (io->scsiio.tag_type) { 1301 case CTL_TAG_ORDERED: 1302 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1303 break; 1304 case CTL_TAG_HEAD_OF_QUEUE: 1305 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1306 break; 1307 case CTL_TAG_UNTAGGED: 1308 case CTL_TAG_SIMPLE: 1309 case CTL_TAG_ACA: 1310 default: 1311 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1312 break; 1313 } 1314 1315 if (lbalen->flags & CTL_LLF_WRITE) { 1316 beio->bio_cmd = BIO_WRITE; 1317 beio->ds_trans_type = DEVSTAT_WRITE; 1318 } else { 1319 beio->bio_cmd = BIO_READ; 1320 beio->ds_trans_type = DEVSTAT_READ; 1321 } 1322 1323 DPRINTF("%s at LBA %jx len %u @%ju\n", 1324 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1325 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1326 if (lbalen->flags & CTL_LLF_COMPARE) 1327 lbas = CTLBLK_HALF_IO_SIZE; 1328 else 1329 lbas = CTLBLK_MAX_IO_SIZE; 1330 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1331 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1332 beio->io_len = lbas * be_lun->blocksize; 1333 bptrlen->len += lbas; 1334 1335 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1336 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1337 i, CTLBLK_MAX_SEGS)); 1338 1339 /* 1340 * Setup the S/G entry for this chunk. 1341 */ 1342 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1343 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1344 1345 DPRINTF("segment %d addr %p len %zd\n", i, 1346 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1347 1348 /* Set up second segment for compare operation. */ 1349 if (lbalen->flags & CTL_LLF_COMPARE) { 1350 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1351 beio->sg_segs[i].len; 1352 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1353 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1354 } 1355 1356 beio->num_segs++; 1357 len_left -= beio->sg_segs[i].len; 1358 } 1359 if (bptrlen->len < lbalen->len) 1360 beio->beio_cont = ctl_be_block_next; 1361 io->scsiio.be_move_done = ctl_be_block_move_done; 1362 /* For compare we have separate S/G lists for read and datamove. */ 1363 if (lbalen->flags & CTL_LLF_COMPARE) 1364 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1365 else 1366 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1367 io->scsiio.kern_data_len = beio->io_len; 1368 io->scsiio.kern_data_resid = 0; 1369 io->scsiio.kern_sg_entries = beio->num_segs; 1370 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1371 1372 /* 1373 * For the read case, we need to read the data into our buffers and 1374 * then we can send it back to the user. For the write case, we 1375 * need to get the data from the user first. 1376 */ 1377 if (beio->bio_cmd == BIO_READ) { 1378 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1379 be_lun->dispatch(be_lun, beio); 1380 } else { 1381 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1382#ifdef CTL_TIME_IO 1383 getbintime(&io->io_hdr.dma_start_bt); 1384#endif 1385 ctl_datamove(io); 1386 } 1387} 1388 1389static void 1390ctl_be_block_worker(void *context, int pending) 1391{ 1392 struct ctl_be_block_lun *be_lun; 1393 struct ctl_be_block_softc *softc; 1394 union ctl_io *io; 1395 1396 be_lun = (struct ctl_be_block_lun *)context; 1397 softc = be_lun->softc; 1398 1399 DPRINTF("entered\n"); 1400 1401 mtx_lock(&be_lun->queue_lock); 1402 for (;;) { 1403 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1404 if (io != NULL) { 1405 struct ctl_be_block_io *beio; 1406 1407 DPRINTF("datamove queue\n"); 1408 1409 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1410 ctl_io_hdr, links); 1411 1412 mtx_unlock(&be_lun->queue_lock); 1413 1414 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1415 1416 be_lun->dispatch(be_lun, beio); 1417 1418 mtx_lock(&be_lun->queue_lock); 1419 continue; 1420 } 1421 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1422 if (io != NULL) { 1423 1424 DPRINTF("config write queue\n"); 1425 1426 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1427 ctl_io_hdr, links); 1428 1429 mtx_unlock(&be_lun->queue_lock); 1430 1431 ctl_be_block_cw_dispatch(be_lun, io); 1432 1433 mtx_lock(&be_lun->queue_lock); 1434 continue; 1435 } 1436 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1437 if (io != NULL) { 1438 DPRINTF("input queue\n"); 1439 1440 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1441 ctl_io_hdr, links); 1442 mtx_unlock(&be_lun->queue_lock); 1443 1444 /* 1445 * We must drop the lock, since this routine and 1446 * its children may sleep. 1447 */ 1448 ctl_be_block_dispatch(be_lun, io); 1449 1450 mtx_lock(&be_lun->queue_lock); 1451 continue; 1452 } 1453 1454 /* 1455 * If we get here, there is no work left in the queues, so 1456 * just break out and let the task queue go to sleep. 1457 */ 1458 break; 1459 } 1460 mtx_unlock(&be_lun->queue_lock); 1461} 1462 1463/* 1464 * Entry point from CTL to the backend for I/O. We queue everything to a 1465 * work thread, so this just puts the I/O on a queue and wakes up the 1466 * thread. 1467 */ 1468static int 1469ctl_be_block_submit(union ctl_io *io) 1470{ 1471 struct ctl_be_block_lun *be_lun; 1472 struct ctl_be_lun *ctl_be_lun; 1473 1474 DPRINTF("entered\n"); 1475 1476 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1477 CTL_PRIV_BACKEND_LUN].ptr; 1478 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1479 1480 /* 1481 * Make sure we only get SCSI I/O. 1482 */ 1483 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1484 "%#x) encountered", io->io_hdr.io_type)); 1485 1486 PRIV(io)->len = 0; 1487 1488 mtx_lock(&be_lun->queue_lock); 1489 /* 1490 * XXX KDM make sure that links is okay to use at this point. 1491 * Otherwise, we either need to add another field to ctl_io_hdr, 1492 * or deal with resource allocation here. 1493 */ 1494 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1495 mtx_unlock(&be_lun->queue_lock); 1496 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1497 1498 return (CTL_RETVAL_COMPLETE); 1499} 1500 1501static int 1502ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1503 int flag, struct thread *td) 1504{ 1505 struct ctl_be_block_softc *softc; 1506 int error; 1507 1508 softc = &backend_block_softc; 1509 1510 error = 0; 1511 1512 switch (cmd) { 1513 case CTL_LUN_REQ: { 1514 struct ctl_lun_req *lun_req; 1515 1516 lun_req = (struct ctl_lun_req *)addr; 1517 1518 switch (lun_req->reqtype) { 1519 case CTL_LUNREQ_CREATE: 1520 error = ctl_be_block_create(softc, lun_req); 1521 break; 1522 case CTL_LUNREQ_RM: 1523 error = ctl_be_block_rm(softc, lun_req); 1524 break; 1525 case CTL_LUNREQ_MODIFY: 1526 error = ctl_be_block_modify(softc, lun_req); 1527 break; 1528 default: 1529 lun_req->status = CTL_LUN_ERROR; 1530 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1531 "invalid LUN request type %d", 1532 lun_req->reqtype); 1533 break; 1534 } 1535 break; 1536 } 1537 default: 1538 error = ENOTTY; 1539 break; 1540 } 1541 1542 return (error); 1543} 1544 1545static int 1546ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1547{ 1548 struct ctl_be_block_filedata *file_data; 1549 struct ctl_lun_create_params *params; 1550 struct vattr vattr; 1551 off_t pss; 1552 int error; 1553 1554 error = 0; 1555 file_data = &be_lun->backend.file; 1556 params = &be_lun->params; 1557 1558 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1559 be_lun->dispatch = ctl_be_block_dispatch_file; 1560 be_lun->lun_flush = ctl_be_block_flush_file; 1561 1562 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1563 if (error != 0) { 1564 snprintf(req->error_str, sizeof(req->error_str), 1565 "error calling VOP_GETATTR() for file %s", 1566 be_lun->dev_path); 1567 return (error); 1568 } 1569 1570 /* 1571 * Verify that we have the ability to upgrade to exclusive 1572 * access on this file so we can trap errors at open instead 1573 * of reporting them during first access. 1574 */ 1575 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1576 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1577 if (be_lun->vn->v_iflag & VI_DOOMED) { 1578 error = EBADF; 1579 snprintf(req->error_str, sizeof(req->error_str), 1580 "error locking file %s", be_lun->dev_path); 1581 return (error); 1582 } 1583 } 1584 1585 1586 file_data->cred = crhold(curthread->td_ucred); 1587 if (params->lun_size_bytes != 0) 1588 be_lun->size_bytes = params->lun_size_bytes; 1589 else 1590 be_lun->size_bytes = vattr.va_size; 1591 /* 1592 * We set the multi thread flag for file operations because all 1593 * filesystems (in theory) are capable of allowing multiple readers 1594 * of a file at once. So we want to get the maximum possible 1595 * concurrency. 1596 */ 1597 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1598 1599 /* 1600 * For files we can use any logical block size. Prefer 512 bytes 1601 * for compatibility reasons. If file's vattr.va_blocksize 1602 * (preferred I/O block size) is bigger and multiple to chosen 1603 * logical block size -- report it as physical block size. 1604 */ 1605 if (params->blocksize_bytes != 0) 1606 be_lun->blocksize = params->blocksize_bytes; 1607 else 1608 be_lun->blocksize = 512; 1609 pss = vattr.va_blocksize / be_lun->blocksize; 1610 if ((pss > 0) && (pss * be_lun->blocksize == vattr.va_blocksize) && 1611 ((pss & (pss - 1)) == 0)) { 1612 be_lun->pblockexp = fls(pss) - 1; 1613 be_lun->pblockoff = 0; 1614 } 1615 1616 /* 1617 * Sanity check. The media size has to be at least one 1618 * sector long. 1619 */ 1620 if (be_lun->size_bytes < be_lun->blocksize) { 1621 error = EINVAL; 1622 snprintf(req->error_str, sizeof(req->error_str), 1623 "file %s size %ju < block size %u", be_lun->dev_path, 1624 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1625 } 1626 return (error); 1627} 1628 1629static int 1630ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1631{ 1632 struct ctl_lun_create_params *params; 1633 struct vattr vattr; 1634 struct cdev *dev; 1635 struct cdevsw *devsw; 1636 int error; 1637 off_t ps, pss, po, pos; 1638 1639 params = &be_lun->params; 1640 1641 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1642 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1643 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1644 &be_lun->backend.dev.dev_ref); 1645 if (be_lun->backend.dev.csw == NULL) 1646 panic("Unable to retrieve device switch"); 1647 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) 1648 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1649 else 1650 be_lun->dispatch = ctl_be_block_dispatch_dev; 1651 be_lun->lun_flush = ctl_be_block_flush_dev; 1652 be_lun->unmap = ctl_be_block_unmap_dev; 1653 1654 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1655 if (error) { 1656 snprintf(req->error_str, sizeof(req->error_str), 1657 "error getting vnode attributes for device %s", 1658 be_lun->dev_path); 1659 return (error); 1660 } 1661 1662 dev = be_lun->vn->v_rdev; 1663 devsw = dev->si_devsw; 1664 if (!devsw->d_ioctl) { 1665 snprintf(req->error_str, sizeof(req->error_str), 1666 "no d_ioctl for device %s!", 1667 be_lun->dev_path); 1668 return (ENODEV); 1669 } 1670 1671 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1672 (caddr_t)&be_lun->blocksize, FREAD, 1673 curthread); 1674 if (error) { 1675 snprintf(req->error_str, sizeof(req->error_str), 1676 "error %d returned for DIOCGSECTORSIZE ioctl " 1677 "on %s!", error, be_lun->dev_path); 1678 return (error); 1679 } 1680 1681 /* 1682 * If the user has asked for a blocksize that is greater than the 1683 * backing device's blocksize, we can do it only if the blocksize 1684 * the user is asking for is an even multiple of the underlying 1685 * device's blocksize. 1686 */ 1687 if ((params->blocksize_bytes != 0) 1688 && (params->blocksize_bytes > be_lun->blocksize)) { 1689 uint32_t bs_multiple, tmp_blocksize; 1690 1691 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1692 1693 tmp_blocksize = bs_multiple * be_lun->blocksize; 1694 1695 if (tmp_blocksize == params->blocksize_bytes) { 1696 be_lun->blocksize = params->blocksize_bytes; 1697 } else { 1698 snprintf(req->error_str, sizeof(req->error_str), 1699 "requested blocksize %u is not an even " 1700 "multiple of backing device blocksize %u", 1701 params->blocksize_bytes, 1702 be_lun->blocksize); 1703 return (EINVAL); 1704 1705 } 1706 } else if ((params->blocksize_bytes != 0) 1707 && (params->blocksize_bytes != be_lun->blocksize)) { 1708 snprintf(req->error_str, sizeof(req->error_str), 1709 "requested blocksize %u < backing device " 1710 "blocksize %u", params->blocksize_bytes, 1711 be_lun->blocksize); 1712 return (EINVAL); 1713 } 1714 1715 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1716 (caddr_t)&be_lun->size_bytes, FREAD, 1717 curthread); 1718 if (error) { 1719 snprintf(req->error_str, sizeof(req->error_str), 1720 "error %d returned for DIOCGMEDIASIZE " 1721 " ioctl on %s!", error, 1722 be_lun->dev_path); 1723 return (error); 1724 } 1725 1726 if (params->lun_size_bytes != 0) { 1727 if (params->lun_size_bytes > be_lun->size_bytes) { 1728 snprintf(req->error_str, sizeof(req->error_str), 1729 "requested LUN size %ju > backing device " 1730 "size %ju", 1731 (uintmax_t)params->lun_size_bytes, 1732 (uintmax_t)be_lun->size_bytes); 1733 return (EINVAL); 1734 } 1735 1736 be_lun->size_bytes = params->lun_size_bytes; 1737 } 1738 1739 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1740 (caddr_t)&ps, FREAD, curthread); 1741 if (error) 1742 ps = po = 0; 1743 else { 1744 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1745 (caddr_t)&po, FREAD, curthread); 1746 if (error) 1747 po = 0; 1748 } 1749 pss = ps / be_lun->blocksize; 1750 pos = po / be_lun->blocksize; 1751 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1752 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1753 be_lun->pblockexp = fls(pss) - 1; 1754 be_lun->pblockoff = (pss - pos) % pss; 1755 } 1756 1757 return (0); 1758} 1759 1760static int 1761ctl_be_block_close(struct ctl_be_block_lun *be_lun) 1762{ 1763 DROP_GIANT(); 1764 if (be_lun->vn) { 1765 int flags = FREAD | FWRITE; 1766 1767 switch (be_lun->dev_type) { 1768 case CTL_BE_BLOCK_DEV: 1769 if (be_lun->backend.dev.csw) { 1770 dev_relthread(be_lun->backend.dev.cdev, 1771 be_lun->backend.dev.dev_ref); 1772 be_lun->backend.dev.csw = NULL; 1773 be_lun->backend.dev.cdev = NULL; 1774 } 1775 break; 1776 case CTL_BE_BLOCK_FILE: 1777 break; 1778 case CTL_BE_BLOCK_NONE: 1779 break; 1780 default: 1781 panic("Unexpected backend type."); 1782 break; 1783 } 1784 1785 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1786 be_lun->vn = NULL; 1787 1788 switch (be_lun->dev_type) { 1789 case CTL_BE_BLOCK_DEV: 1790 break; 1791 case CTL_BE_BLOCK_FILE: 1792 if (be_lun->backend.file.cred != NULL) { 1793 crfree(be_lun->backend.file.cred); 1794 be_lun->backend.file.cred = NULL; 1795 } 1796 break; 1797 case CTL_BE_BLOCK_NONE: 1798 break; 1799 default: 1800 panic("Unexpected backend type."); 1801 break; 1802 } 1803 be_lun->dev_type = CTL_BE_BLOCK_NONE; 1804 } 1805 PICKUP_GIANT(); 1806 1807 return (0); 1808} 1809 1810static int 1811ctl_be_block_open(struct ctl_be_block_softc *softc, 1812 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1813{ 1814 struct nameidata nd; 1815 int flags; 1816 int error; 1817 1818 /* 1819 * XXX KDM allow a read-only option? 1820 */ 1821 flags = FREAD | FWRITE; 1822 error = 0; 1823 1824 if (rootvnode == NULL) { 1825 snprintf(req->error_str, sizeof(req->error_str), 1826 "Root filesystem is not mounted"); 1827 return (1); 1828 } 1829 1830 if (!curthread->td_proc->p_fd->fd_cdir) { 1831 curthread->td_proc->p_fd->fd_cdir = rootvnode; 1832 VREF(rootvnode); 1833 } 1834 if (!curthread->td_proc->p_fd->fd_rdir) { 1835 curthread->td_proc->p_fd->fd_rdir = rootvnode; 1836 VREF(rootvnode); 1837 } 1838 if (!curthread->td_proc->p_fd->fd_jdir) { 1839 curthread->td_proc->p_fd->fd_jdir = rootvnode; 1840 VREF(rootvnode); 1841 } 1842 1843 again: 1844 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1845 error = vn_open(&nd, &flags, 0, NULL); 1846 if (error) { 1847 /* 1848 * This is the only reasonable guess we can make as far as 1849 * path if the user doesn't give us a fully qualified path. 1850 * If they want to specify a file, they need to specify the 1851 * full path. 1852 */ 1853 if (be_lun->dev_path[0] != '/') { 1854 char *dev_path = "/dev/"; 1855 char *dev_name; 1856 1857 /* Try adding device path at beginning of name */ 1858 dev_name = malloc(strlen(be_lun->dev_path) 1859 + strlen(dev_path) + 1, 1860 M_CTLBLK, M_WAITOK); 1861 if (dev_name) { 1862 sprintf(dev_name, "%s%s", dev_path, 1863 be_lun->dev_path); 1864 free(be_lun->dev_path, M_CTLBLK); 1865 be_lun->dev_path = dev_name; 1866 goto again; 1867 } 1868 } 1869 snprintf(req->error_str, sizeof(req->error_str), 1870 "error opening %s: %d", be_lun->dev_path, error); 1871 return (error); 1872 } 1873 1874 NDFREE(&nd, NDF_ONLY_PNBUF); 1875 1876 be_lun->vn = nd.ni_vp; 1877 1878 /* We only support disks and files. */ 1879 if (vn_isdisk(be_lun->vn, &error)) { 1880 error = ctl_be_block_open_dev(be_lun, req); 1881 } else if (be_lun->vn->v_type == VREG) { 1882 error = ctl_be_block_open_file(be_lun, req); 1883 } else { 1884 error = EINVAL; 1885 snprintf(req->error_str, sizeof(req->error_str), 1886 "%s is not a disk or plain file", be_lun->dev_path); 1887 } 1888 VOP_UNLOCK(be_lun->vn, 0); 1889 1890 if (error != 0) { 1891 ctl_be_block_close(be_lun); 1892 return (error); 1893 } 1894 1895 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1896 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1897 1898 return (0); 1899} 1900 1901static int 1902ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1903{ 1904 struct ctl_be_block_lun *be_lun; 1905 struct ctl_lun_create_params *params; 1906 char num_thread_str[16]; 1907 char tmpstr[32]; 1908 char *value; 1909 int retval, num_threads, unmap; 1910 int tmp_num_threads; 1911 1912 params = &req->reqdata.create; 1913 retval = 0; 1914 req->status = CTL_LUN_OK; 1915 1916 num_threads = cbb_num_threads; 1917 1918 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1919 1920 be_lun->params = req->reqdata.create; 1921 be_lun->softc = softc; 1922 STAILQ_INIT(&be_lun->input_queue); 1923 STAILQ_INIT(&be_lun->config_write_queue); 1924 STAILQ_INIT(&be_lun->datamove_queue); 1925 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1926 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1927 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1928 ctl_init_opts(&be_lun->ctl_be_lun.options, 1929 req->num_be_args, req->kern_be_args); 1930 1931 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1932 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1933 1934 if (be_lun->lun_zone == NULL) { 1935 snprintf(req->error_str, sizeof(req->error_str), 1936 "error allocating UMA zone"); 1937 goto bailout_error; 1938 } 1939 1940 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1941 be_lun->ctl_be_lun.lun_type = params->device_type; 1942 else 1943 be_lun->ctl_be_lun.lun_type = T_DIRECT; 1944 1945 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1946 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1947 if (value == NULL) { 1948 snprintf(req->error_str, sizeof(req->error_str), 1949 "no file argument specified"); 1950 goto bailout_error; 1951 } 1952 be_lun->dev_path = strdup(value, M_CTLBLK); 1953 be_lun->blocksize = 512; 1954 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1955 1956 retval = ctl_be_block_open(softc, be_lun, req); 1957 if (retval != 0) { 1958 retval = 0; 1959 req->status = CTL_LUN_WARNING; 1960 } 1961 } else { 1962 /* 1963 * For processor devices, we don't have any size. 1964 */ 1965 be_lun->blocksize = 0; 1966 be_lun->pblockexp = 0; 1967 be_lun->pblockoff = 0; 1968 be_lun->size_blocks = 0; 1969 be_lun->size_bytes = 0; 1970 be_lun->ctl_be_lun.maxlba = 0; 1971 1972 /* 1973 * Default to just 1 thread for processor devices. 1974 */ 1975 num_threads = 1; 1976 } 1977 1978 /* 1979 * XXX This searching loop might be refactored to be combined with 1980 * the loop above, 1981 */ 1982 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 1983 if (value != NULL) { 1984 tmp_num_threads = strtol(value, NULL, 0); 1985 1986 /* 1987 * We don't let the user specify less than one 1988 * thread, but hope he's clueful enough not to 1989 * specify 1000 threads. 1990 */ 1991 if (tmp_num_threads < 1) { 1992 snprintf(req->error_str, sizeof(req->error_str), 1993 "invalid number of threads %s", 1994 num_thread_str); 1995 goto bailout_error; 1996 } 1997 num_threads = tmp_num_threads; 1998 } 1999 unmap = 0; 2000 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2001 if (value != NULL && strcmp(value, "on") == 0) 2002 unmap = 1; 2003 2004 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2005 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2006 if (be_lun->vn == NULL) 2007 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2008 if (unmap) 2009 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2010 be_lun->ctl_be_lun.be_lun = be_lun; 2011 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2012 0 : (be_lun->size_blocks - 1); 2013 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2014 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2015 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2016 if (be_lun->dispatch == ctl_be_block_dispatch_zvol && 2017 be_lun->blocksize != 0) 2018 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / 2019 be_lun->blocksize; 2020 /* Tell the user the blocksize we ended up using */ 2021 params->lun_size_bytes = be_lun->size_bytes; 2022 params->blocksize_bytes = be_lun->blocksize; 2023 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2024 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2025 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2026 } else 2027 be_lun->ctl_be_lun.req_lun_id = 0; 2028 2029 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2030 be_lun->ctl_be_lun.lun_config_status = 2031 ctl_be_block_lun_config_status; 2032 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2033 2034 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2035 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2036 softc->num_luns); 2037 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2038 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2039 sizeof(tmpstr))); 2040 2041 /* Tell the user what we used for a serial number */ 2042 strncpy((char *)params->serial_num, tmpstr, 2043 ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 2044 } else { 2045 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2046 params->serial_num, 2047 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2048 sizeof(params->serial_num))); 2049 } 2050 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2051 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2052 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2053 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2054 sizeof(tmpstr))); 2055 2056 /* Tell the user what we used for a device ID */ 2057 strncpy((char *)params->device_id, tmpstr, 2058 ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 2059 } else { 2060 strncpy((char *)be_lun->ctl_be_lun.device_id, 2061 params->device_id, 2062 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2063 sizeof(params->device_id))); 2064 } 2065 2066 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2067 2068 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2069 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2070 2071 if (be_lun->io_taskqueue == NULL) { 2072 snprintf(req->error_str, sizeof(req->error_str), 2073 "unable to create taskqueue"); 2074 goto bailout_error; 2075 } 2076 2077 /* 2078 * Note that we start the same number of threads by default for 2079 * both the file case and the block device case. For the file 2080 * case, we need multiple threads to allow concurrency, because the 2081 * vnode interface is designed to be a blocking interface. For the 2082 * block device case, ZFS zvols at least will block the caller's 2083 * context in many instances, and so we need multiple threads to 2084 * overcome that problem. Other block devices don't need as many 2085 * threads, but they shouldn't cause too many problems. 2086 * 2087 * If the user wants to just have a single thread for a block 2088 * device, he can specify that when the LUN is created, or change 2089 * the tunable/sysctl to alter the default number of threads. 2090 */ 2091 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2092 /*num threads*/num_threads, 2093 /*priority*/PWAIT, 2094 /*thread name*/ 2095 "%s taskq", be_lun->lunname); 2096 2097 if (retval != 0) 2098 goto bailout_error; 2099 2100 be_lun->num_threads = num_threads; 2101 2102 mtx_lock(&softc->lock); 2103 softc->num_luns++; 2104 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2105 2106 mtx_unlock(&softc->lock); 2107 2108 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2109 if (retval != 0) { 2110 mtx_lock(&softc->lock); 2111 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2112 links); 2113 softc->num_luns--; 2114 mtx_unlock(&softc->lock); 2115 snprintf(req->error_str, sizeof(req->error_str), 2116 "ctl_add_lun() returned error %d, see dmesg for " 2117 "details", retval); 2118 retval = 0; 2119 goto bailout_error; 2120 } 2121 2122 mtx_lock(&softc->lock); 2123 2124 /* 2125 * Tell the config_status routine that we're waiting so it won't 2126 * clean up the LUN in the event of an error. 2127 */ 2128 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2129 2130 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2131 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2132 if (retval == EINTR) 2133 break; 2134 } 2135 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2136 2137 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2138 snprintf(req->error_str, sizeof(req->error_str), 2139 "LUN configuration error, see dmesg for details"); 2140 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2141 links); 2142 softc->num_luns--; 2143 mtx_unlock(&softc->lock); 2144 goto bailout_error; 2145 } else { 2146 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2147 } 2148 2149 mtx_unlock(&softc->lock); 2150 2151 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2152 be_lun->blocksize, 2153 DEVSTAT_ALL_SUPPORTED, 2154 be_lun->ctl_be_lun.lun_type 2155 | DEVSTAT_TYPE_IF_OTHER, 2156 DEVSTAT_PRIORITY_OTHER); 2157 2158 return (retval); 2159 2160bailout_error: 2161 req->status = CTL_LUN_ERROR; 2162 2163 if (be_lun->io_taskqueue != NULL) 2164 taskqueue_free(be_lun->io_taskqueue); 2165 ctl_be_block_close(be_lun); 2166 if (be_lun->dev_path != NULL) 2167 free(be_lun->dev_path, M_CTLBLK); 2168 if (be_lun->lun_zone != NULL) 2169 uma_zdestroy(be_lun->lun_zone); 2170 ctl_free_opts(&be_lun->ctl_be_lun.options); 2171 mtx_destroy(&be_lun->queue_lock); 2172 mtx_destroy(&be_lun->io_lock); 2173 free(be_lun, M_CTLBLK); 2174 2175 return (retval); 2176} 2177 2178static int 2179ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2180{ 2181 struct ctl_lun_rm_params *params; 2182 struct ctl_be_block_lun *be_lun; 2183 int retval; 2184 2185 params = &req->reqdata.rm; 2186 2187 mtx_lock(&softc->lock); 2188 2189 be_lun = NULL; 2190 2191 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2192 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2193 break; 2194 } 2195 mtx_unlock(&softc->lock); 2196 2197 if (be_lun == NULL) { 2198 snprintf(req->error_str, sizeof(req->error_str), 2199 "LUN %u is not managed by the block backend", 2200 params->lun_id); 2201 goto bailout_error; 2202 } 2203 2204 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2205 2206 if (retval != 0) { 2207 snprintf(req->error_str, sizeof(req->error_str), 2208 "error %d returned from ctl_disable_lun() for " 2209 "LUN %d", retval, params->lun_id); 2210 goto bailout_error; 2211 2212 } 2213 2214 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2215 if (retval != 0) { 2216 snprintf(req->error_str, sizeof(req->error_str), 2217 "error %d returned from ctl_invalidate_lun() for " 2218 "LUN %d", retval, params->lun_id); 2219 goto bailout_error; 2220 } 2221 2222 mtx_lock(&softc->lock); 2223 2224 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2225 2226 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2227 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2228 if (retval == EINTR) 2229 break; 2230 } 2231 2232 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2233 2234 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2235 snprintf(req->error_str, sizeof(req->error_str), 2236 "interrupted waiting for LUN to be freed"); 2237 mtx_unlock(&softc->lock); 2238 goto bailout_error; 2239 } 2240 2241 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2242 2243 softc->num_luns--; 2244 mtx_unlock(&softc->lock); 2245 2246 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2247 2248 taskqueue_free(be_lun->io_taskqueue); 2249 2250 ctl_be_block_close(be_lun); 2251 2252 if (be_lun->disk_stats != NULL) 2253 devstat_remove_entry(be_lun->disk_stats); 2254 2255 uma_zdestroy(be_lun->lun_zone); 2256 2257 ctl_free_opts(&be_lun->ctl_be_lun.options); 2258 free(be_lun->dev_path, M_CTLBLK); 2259 mtx_destroy(&be_lun->queue_lock); 2260 mtx_destroy(&be_lun->io_lock); 2261 free(be_lun, M_CTLBLK); 2262 2263 req->status = CTL_LUN_OK; 2264 2265 return (0); 2266 2267bailout_error: 2268 2269 req->status = CTL_LUN_ERROR; 2270 2271 return (0); 2272} 2273 2274static int 2275ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2276 struct ctl_lun_req *req) 2277{ 2278 struct vattr vattr; 2279 int error; 2280 struct ctl_lun_create_params *params = &be_lun->params; 2281 2282 if (params->lun_size_bytes != 0) { 2283 be_lun->size_bytes = params->lun_size_bytes; 2284 } else { 2285 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2286 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2287 VOP_UNLOCK(be_lun->vn, 0); 2288 if (error != 0) { 2289 snprintf(req->error_str, sizeof(req->error_str), 2290 "error calling VOP_GETATTR() for file %s", 2291 be_lun->dev_path); 2292 return (error); 2293 } 2294 2295 be_lun->size_bytes = vattr.va_size; 2296 } 2297 2298 return (0); 2299} 2300 2301static int 2302ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2303 struct ctl_lun_req *req) 2304{ 2305 struct ctl_be_block_devdata *dev_data; 2306 int error; 2307 struct ctl_lun_create_params *params = &be_lun->params; 2308 uint64_t size_bytes; 2309 2310 dev_data = &be_lun->backend.dev; 2311 if (!dev_data->csw->d_ioctl) { 2312 snprintf(req->error_str, sizeof(req->error_str), 2313 "no d_ioctl for device %s!", be_lun->dev_path); 2314 return (ENODEV); 2315 } 2316 2317 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2318 (caddr_t)&size_bytes, FREAD, 2319 curthread); 2320 if (error) { 2321 snprintf(req->error_str, sizeof(req->error_str), 2322 "error %d returned for DIOCGMEDIASIZE ioctl " 2323 "on %s!", error, be_lun->dev_path); 2324 return (error); 2325 } 2326 2327 if (params->lun_size_bytes != 0) { 2328 if (params->lun_size_bytes > size_bytes) { 2329 snprintf(req->error_str, sizeof(req->error_str), 2330 "requested LUN size %ju > backing device " 2331 "size %ju", 2332 (uintmax_t)params->lun_size_bytes, 2333 (uintmax_t)size_bytes); 2334 return (EINVAL); 2335 } 2336 2337 be_lun->size_bytes = params->lun_size_bytes; 2338 } else { 2339 be_lun->size_bytes = size_bytes; 2340 } 2341 2342 return (0); 2343} 2344 2345static int 2346ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2347{ 2348 struct ctl_lun_modify_params *params; 2349 struct ctl_be_block_lun *be_lun; 2350 uint64_t oldsize; 2351 int error; 2352 2353 params = &req->reqdata.modify; 2354 2355 mtx_lock(&softc->lock); 2356 be_lun = NULL; 2357 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2358 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2359 break; 2360 } 2361 mtx_unlock(&softc->lock); 2362 2363 if (be_lun == NULL) { 2364 snprintf(req->error_str, sizeof(req->error_str), 2365 "LUN %u is not managed by the block backend", 2366 params->lun_id); 2367 goto bailout_error; 2368 } 2369 2370 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2371 2372 oldsize = be_lun->size_blocks; 2373 if (be_lun->vn == NULL) 2374 error = ctl_be_block_open(softc, be_lun, req); 2375 else if (be_lun->vn->v_type == VREG) 2376 error = ctl_be_block_modify_file(be_lun, req); 2377 else 2378 error = ctl_be_block_modify_dev(be_lun, req); 2379 2380 if (error == 0 && be_lun->size_blocks != oldsize) { 2381 be_lun->size_blocks = be_lun->size_bytes >> 2382 be_lun->blocksize_shift; 2383 2384 /* 2385 * The maximum LBA is the size - 1. 2386 * 2387 * XXX: Note that this field is being updated without locking, 2388 * which might cause problems on 32-bit architectures. 2389 */ 2390 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2391 0 : (be_lun->size_blocks - 1); 2392 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2393 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2394 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2395 if (be_lun->dispatch == ctl_be_block_dispatch_zvol && 2396 be_lun->blocksize != 0) 2397 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / 2398 be_lun->blocksize; 2399 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2400 if (oldsize == 0 && be_lun->size_blocks != 0) 2401 ctl_lun_online(&be_lun->ctl_be_lun); 2402 } 2403 2404 /* Tell the user the exact size we ended up using */ 2405 params->lun_size_bytes = be_lun->size_bytes; 2406 2407 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2408 2409 return (0); 2410 2411bailout_error: 2412 req->status = CTL_LUN_ERROR; 2413 2414 return (0); 2415} 2416 2417static void 2418ctl_be_block_lun_shutdown(void *be_lun) 2419{ 2420 struct ctl_be_block_lun *lun; 2421 struct ctl_be_block_softc *softc; 2422 2423 lun = (struct ctl_be_block_lun *)be_lun; 2424 2425 softc = lun->softc; 2426 2427 mtx_lock(&softc->lock); 2428 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2429 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2430 wakeup(lun); 2431 mtx_unlock(&softc->lock); 2432 2433} 2434 2435static void 2436ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2437{ 2438 struct ctl_be_block_lun *lun; 2439 struct ctl_be_block_softc *softc; 2440 2441 lun = (struct ctl_be_block_lun *)be_lun; 2442 softc = lun->softc; 2443 2444 if (status == CTL_LUN_CONFIG_OK) { 2445 mtx_lock(&softc->lock); 2446 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2447 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2448 wakeup(lun); 2449 mtx_unlock(&softc->lock); 2450 2451 /* 2452 * We successfully added the LUN, attempt to enable it. 2453 */ 2454 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2455 printf("%s: ctl_enable_lun() failed!\n", __func__); 2456 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2457 printf("%s: ctl_invalidate_lun() failed!\n", 2458 __func__); 2459 } 2460 } 2461 2462 return; 2463 } 2464 2465 2466 mtx_lock(&softc->lock); 2467 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2468 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2469 wakeup(lun); 2470 mtx_unlock(&softc->lock); 2471} 2472 2473 2474static int 2475ctl_be_block_config_write(union ctl_io *io) 2476{ 2477 struct ctl_be_block_lun *be_lun; 2478 struct ctl_be_lun *ctl_be_lun; 2479 int retval; 2480 2481 retval = 0; 2482 2483 DPRINTF("entered\n"); 2484 2485 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2486 CTL_PRIV_BACKEND_LUN].ptr; 2487 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2488 2489 switch (io->scsiio.cdb[0]) { 2490 case SYNCHRONIZE_CACHE: 2491 case SYNCHRONIZE_CACHE_16: 2492 case WRITE_SAME_10: 2493 case WRITE_SAME_16: 2494 case UNMAP: 2495 /* 2496 * The upper level CTL code will filter out any CDBs with 2497 * the immediate bit set and return the proper error. 2498 * 2499 * We don't really need to worry about what LBA range the 2500 * user asked to be synced out. When they issue a sync 2501 * cache command, we'll sync out the whole thing. 2502 */ 2503 mtx_lock(&be_lun->queue_lock); 2504 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2505 links); 2506 mtx_unlock(&be_lun->queue_lock); 2507 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2508 break; 2509 case START_STOP_UNIT: { 2510 struct scsi_start_stop_unit *cdb; 2511 2512 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2513 2514 if (cdb->how & SSS_START) 2515 retval = ctl_start_lun(ctl_be_lun); 2516 else { 2517 retval = ctl_stop_lun(ctl_be_lun); 2518 /* 2519 * XXX KDM Copan-specific offline behavior. 2520 * Figure out a reasonable way to port this? 2521 */ 2522#ifdef NEEDTOPORT 2523 if ((retval == 0) 2524 && (cdb->byte2 & SSS_ONOFFLINE)) 2525 retval = ctl_lun_offline(ctl_be_lun); 2526#endif 2527 } 2528 2529 /* 2530 * In general, the above routines should not fail. They 2531 * just set state for the LUN. So we've got something 2532 * pretty wrong here if we can't start or stop the LUN. 2533 */ 2534 if (retval != 0) { 2535 ctl_set_internal_failure(&io->scsiio, 2536 /*sks_valid*/ 1, 2537 /*retry_count*/ 0xf051); 2538 retval = CTL_RETVAL_COMPLETE; 2539 } else { 2540 ctl_set_success(&io->scsiio); 2541 } 2542 ctl_config_write_done(io); 2543 break; 2544 } 2545 default: 2546 ctl_set_invalid_opcode(&io->scsiio); 2547 ctl_config_write_done(io); 2548 retval = CTL_RETVAL_COMPLETE; 2549 break; 2550 } 2551 2552 return (retval); 2553 2554} 2555 2556static int 2557ctl_be_block_config_read(union ctl_io *io) 2558{ 2559 return (0); 2560} 2561 2562static int 2563ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2564{ 2565 struct ctl_be_block_lun *lun; 2566 int retval; 2567 2568 lun = (struct ctl_be_block_lun *)be_lun; 2569 retval = 0; 2570 2571 retval = sbuf_printf(sb, "\t<num_threads>"); 2572 2573 if (retval != 0) 2574 goto bailout; 2575 2576 retval = sbuf_printf(sb, "%d", lun->num_threads); 2577 2578 if (retval != 0) 2579 goto bailout; 2580 2581 retval = sbuf_printf(sb, "</num_threads>\n"); 2582 2583bailout: 2584 2585 return (retval); 2586} 2587 2588int 2589ctl_be_block_init(void) 2590{ 2591 struct ctl_be_block_softc *softc; 2592 int retval; 2593 2594 softc = &backend_block_softc; 2595 retval = 0; 2596 2597 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2598 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2599 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2600 STAILQ_INIT(&softc->disk_list); 2601 STAILQ_INIT(&softc->lun_list); 2602 2603 return (retval); 2604} 2605