ctl_backend_block.c revision 274732
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 274732 2014-11-20 01:55:12Z 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); 150typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 151 const char *attrname); 152 153/* 154 * Backend LUN structure. There is a 1:1 mapping between a block device 155 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 156 */ 157struct ctl_be_block_lun { 158 struct ctl_lun_create_params params; 159 struct ctl_block_disk *disk; 160 char lunname[32]; 161 char *dev_path; 162 ctl_be_block_type dev_type; 163 struct vnode *vn; 164 union ctl_be_block_bedata backend; 165 cbb_dispatch_t dispatch; 166 cbb_dispatch_t lun_flush; 167 cbb_dispatch_t unmap; 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 struct ctl_be_block_softc *softc; 177 struct devstat *disk_stats; 178 ctl_be_block_lun_flags flags; 179 STAILQ_ENTRY(ctl_be_block_lun) links; 180 struct ctl_be_lun ctl_be_lun; 181 struct taskqueue *io_taskqueue; 182 struct task io_task; 183 int num_threads; 184 STAILQ_HEAD(, ctl_io_hdr) input_queue; 185 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 186 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 187 struct mtx_padalign io_lock; 188 struct mtx_padalign queue_lock; 189}; 190 191/* 192 * Overall softc structure for the block backend module. 193 */ 194struct ctl_be_block_softc { 195 struct mtx lock; 196 int num_disks; 197 STAILQ_HEAD(, ctl_block_disk) disk_list; 198 int num_luns; 199 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 200}; 201 202static struct ctl_be_block_softc backend_block_softc; 203 204/* 205 * Per-I/O information. 206 */ 207struct ctl_be_block_io { 208 union ctl_io *io; 209 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 210 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 211 int bio_cmd; 212 int num_segs; 213 int num_bios_sent; 214 int num_bios_done; 215 int send_complete; 216 int num_errors; 217 struct bintime ds_t0; 218 devstat_tag_type ds_tag_type; 219 devstat_trans_flags ds_trans_type; 220 uint64_t io_len; 221 uint64_t io_offset; 222 struct ctl_be_block_softc *softc; 223 struct ctl_be_block_lun *lun; 224 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 225}; 226 227static int cbb_num_threads = 14; 228TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 229SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 230 "CAM Target Layer Block Backend"); 231SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 232 &cbb_num_threads, 0, "Number of threads per backing file"); 233 234static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 235static void ctl_free_beio(struct ctl_be_block_io *beio); 236static void ctl_complete_beio(struct ctl_be_block_io *beio); 237static int ctl_be_block_move_done(union ctl_io *io); 238static void ctl_be_block_biodone(struct bio *bio); 239static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 240 struct ctl_be_block_io *beio); 241static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 242 struct ctl_be_block_io *beio); 243static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 244 struct ctl_be_block_io *beio); 245static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 246 struct ctl_be_block_io *beio); 247static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 248 struct ctl_be_block_io *beio); 249static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 250 const char *attrname); 251static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 252 union ctl_io *io); 253static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 254 union ctl_io *io); 255static void ctl_be_block_worker(void *context, int pending); 256static int ctl_be_block_submit(union ctl_io *io); 257static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 258 int flag, struct thread *td); 259static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 260 struct ctl_lun_req *req); 261static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 262 struct ctl_lun_req *req); 263static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 264static int ctl_be_block_open(struct ctl_be_block_softc *softc, 265 struct ctl_be_block_lun *be_lun, 266 struct ctl_lun_req *req); 267static int ctl_be_block_create(struct ctl_be_block_softc *softc, 268 struct ctl_lun_req *req); 269static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 270 struct ctl_lun_req *req); 271static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 272 struct ctl_lun_req *req); 273static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 274 struct ctl_lun_req *req); 275static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 276 struct ctl_lun_req *req); 277static void ctl_be_block_lun_shutdown(void *be_lun); 278static void ctl_be_block_lun_config_status(void *be_lun, 279 ctl_lun_config_status status); 280static int ctl_be_block_config_write(union ctl_io *io); 281static int ctl_be_block_config_read(union ctl_io *io); 282static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 283static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 284int ctl_be_block_init(void); 285 286static struct ctl_backend_driver ctl_be_block_driver = 287{ 288 .name = "block", 289 .flags = CTL_BE_FLAG_HAS_CONFIG, 290 .init = ctl_be_block_init, 291 .data_submit = ctl_be_block_submit, 292 .data_move_done = ctl_be_block_move_done, 293 .config_read = ctl_be_block_config_read, 294 .config_write = ctl_be_block_config_write, 295 .ioctl = ctl_be_block_ioctl, 296 .lun_info = ctl_be_block_lun_info, 297 .lun_attr = ctl_be_block_lun_attr 298}; 299 300MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 301CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 302 303static uma_zone_t beio_zone; 304 305static struct ctl_be_block_io * 306ctl_alloc_beio(struct ctl_be_block_softc *softc) 307{ 308 struct ctl_be_block_io *beio; 309 310 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 311 beio->softc = softc; 312 return (beio); 313} 314 315static void 316ctl_free_beio(struct ctl_be_block_io *beio) 317{ 318 int duplicate_free; 319 int i; 320 321 duplicate_free = 0; 322 323 for (i = 0; i < beio->num_segs; i++) { 324 if (beio->sg_segs[i].addr == NULL) 325 duplicate_free++; 326 327 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 328 beio->sg_segs[i].addr = NULL; 329 330 /* For compare we had two equal S/G lists. */ 331 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 332 uma_zfree(beio->lun->lun_zone, 333 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 334 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 335 } 336 } 337 338 if (duplicate_free > 0) { 339 printf("%s: %d duplicate frees out of %d segments\n", __func__, 340 duplicate_free, beio->num_segs); 341 } 342 343 uma_zfree(beio_zone, beio); 344} 345 346static void 347ctl_complete_beio(struct ctl_be_block_io *beio) 348{ 349 union ctl_io *io = beio->io; 350 351 if (beio->beio_cont != NULL) { 352 beio->beio_cont(beio); 353 } else { 354 ctl_free_beio(beio); 355 ctl_data_submit_done(io); 356 } 357} 358 359static int 360ctl_be_block_move_done(union ctl_io *io) 361{ 362 struct ctl_be_block_io *beio; 363 struct ctl_be_block_lun *be_lun; 364 struct ctl_lba_len_flags *lbalen; 365#ifdef CTL_TIME_IO 366 struct bintime cur_bt; 367#endif 368 int i; 369 370 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 371 be_lun = beio->lun; 372 373 DPRINTF("entered\n"); 374 375#ifdef CTL_TIME_IO 376 getbintime(&cur_bt); 377 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 378 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 379 io->io_hdr.num_dmas++; 380#endif 381 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 382 383 /* 384 * We set status at this point for read commands, and write 385 * commands with errors. 386 */ 387 if ((io->io_hdr.port_status == 0) && 388 ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 389 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 390 lbalen = ARGS(beio->io); 391 if (lbalen->flags & CTL_LLF_READ) { 392 ctl_set_success(&io->scsiio); 393 } else if (lbalen->flags & CTL_LLF_COMPARE) { 394 /* We have two data blocks ready for comparison. */ 395 for (i = 0; i < beio->num_segs; i++) { 396 if (memcmp(beio->sg_segs[i].addr, 397 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 398 beio->sg_segs[i].len) != 0) 399 break; 400 } 401 if (i < beio->num_segs) 402 ctl_set_sense(&io->scsiio, 403 /*current_error*/ 1, 404 /*sense_key*/ SSD_KEY_MISCOMPARE, 405 /*asc*/ 0x1D, 406 /*ascq*/ 0x00, 407 SSD_ELEM_NONE); 408 else 409 ctl_set_success(&io->scsiio); 410 } 411 } 412 else if ((io->io_hdr.port_status != 0) 413 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 414 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 415 /* 416 * For hardware error sense keys, the sense key 417 * specific value is defined to be a retry count, 418 * but we use it to pass back an internal FETD 419 * error code. XXX KDM Hopefully the FETD is only 420 * using 16 bits for an error code, since that's 421 * all the space we have in the sks field. 422 */ 423 ctl_set_internal_failure(&io->scsiio, 424 /*sks_valid*/ 1, 425 /*retry_count*/ 426 io->io_hdr.port_status); 427 } 428 429 /* 430 * If this is a read, or a write with errors, it is done. 431 */ 432 if ((beio->bio_cmd == BIO_READ) 433 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 434 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 435 ctl_complete_beio(beio); 436 return (0); 437 } 438 439 /* 440 * At this point, we have a write and the DMA completed 441 * successfully. We now have to queue it to the task queue to 442 * execute the backend I/O. That is because we do blocking 443 * memory allocations, and in the file backing case, blocking I/O. 444 * This move done routine is generally called in the SIM's 445 * interrupt context, and therefore we cannot block. 446 */ 447 mtx_lock(&be_lun->queue_lock); 448 /* 449 * XXX KDM make sure that links is okay to use at this point. 450 * Otherwise, we either need to add another field to ctl_io_hdr, 451 * or deal with resource allocation here. 452 */ 453 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 454 mtx_unlock(&be_lun->queue_lock); 455 456 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 457 458 return (0); 459} 460 461static void 462ctl_be_block_biodone(struct bio *bio) 463{ 464 struct ctl_be_block_io *beio; 465 struct ctl_be_block_lun *be_lun; 466 union ctl_io *io; 467 int error; 468 469 beio = bio->bio_caller1; 470 be_lun = beio->lun; 471 io = beio->io; 472 473 DPRINTF("entered\n"); 474 475 error = bio->bio_error; 476 mtx_lock(&be_lun->io_lock); 477 if (error != 0) 478 beio->num_errors++; 479 480 beio->num_bios_done++; 481 482 /* 483 * XXX KDM will this cause WITNESS to complain? Holding a lock 484 * during the free might cause it to complain. 485 */ 486 g_destroy_bio(bio); 487 488 /* 489 * If the send complete bit isn't set, or we aren't the last I/O to 490 * complete, then we're done. 491 */ 492 if ((beio->send_complete == 0) 493 || (beio->num_bios_done < beio->num_bios_sent)) { 494 mtx_unlock(&be_lun->io_lock); 495 return; 496 } 497 498 /* 499 * At this point, we've verified that we are the last I/O to 500 * complete, so it's safe to drop the lock. 501 */ 502 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 503 beio->ds_tag_type, beio->ds_trans_type, 504 /*now*/ NULL, /*then*/&beio->ds_t0); 505 mtx_unlock(&be_lun->io_lock); 506 507 /* 508 * If there are any errors from the backing device, we fail the 509 * entire I/O with a medium error. 510 */ 511 if (beio->num_errors > 0) { 512 if (error == EOPNOTSUPP) { 513 ctl_set_invalid_opcode(&io->scsiio); 514 } else if (error == ENOSPC) { 515 ctl_set_space_alloc_fail(&io->scsiio); 516 } else if (beio->bio_cmd == BIO_FLUSH) { 517 /* XXX KDM is there is a better error here? */ 518 ctl_set_internal_failure(&io->scsiio, 519 /*sks_valid*/ 1, 520 /*retry_count*/ 0xbad2); 521 } else 522 ctl_set_medium_error(&io->scsiio); 523 ctl_complete_beio(beio); 524 return; 525 } 526 527 /* 528 * If this is a write, a flush, a delete or verify, we're all done. 529 * If this is a read, we can now send the data to the user. 530 */ 531 if ((beio->bio_cmd == BIO_WRITE) 532 || (beio->bio_cmd == BIO_FLUSH) 533 || (beio->bio_cmd == BIO_DELETE) 534 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 535 ctl_set_success(&io->scsiio); 536 ctl_complete_beio(beio); 537 } else { 538#ifdef CTL_TIME_IO 539 getbintime(&io->io_hdr.dma_start_bt); 540#endif 541 ctl_datamove(io); 542 } 543} 544 545static void 546ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 547 struct ctl_be_block_io *beio) 548{ 549 union ctl_io *io = beio->io; 550 struct mount *mountpoint; 551 int error, lock_flags; 552 553 DPRINTF("entered\n"); 554 555 binuptime(&beio->ds_t0); 556 mtx_lock(&be_lun->io_lock); 557 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 558 mtx_unlock(&be_lun->io_lock); 559 560 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 561 562 if (MNT_SHARED_WRITES(mountpoint) 563 || ((mountpoint == NULL) 564 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 565 lock_flags = LK_SHARED; 566 else 567 lock_flags = LK_EXCLUSIVE; 568 569 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 570 571 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 572 VOP_UNLOCK(be_lun->vn, 0); 573 574 vn_finished_write(mountpoint); 575 576 mtx_lock(&be_lun->io_lock); 577 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 578 beio->ds_tag_type, beio->ds_trans_type, 579 /*now*/ NULL, /*then*/&beio->ds_t0); 580 mtx_unlock(&be_lun->io_lock); 581 582 if (error == 0) 583 ctl_set_success(&io->scsiio); 584 else { 585 /* XXX KDM is there is a better error here? */ 586 ctl_set_internal_failure(&io->scsiio, 587 /*sks_valid*/ 1, 588 /*retry_count*/ 0xbad1); 589 } 590 591 ctl_complete_beio(beio); 592} 593 594SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 595SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 596SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 597SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 598 599static void 600ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 601 struct ctl_be_block_io *beio) 602{ 603 struct ctl_be_block_filedata *file_data; 604 union ctl_io *io; 605 struct uio xuio; 606 struct iovec *xiovec; 607 int flags; 608 int error, i; 609 610 DPRINTF("entered\n"); 611 612 file_data = &be_lun->backend.file; 613 io = beio->io; 614 flags = 0; 615 if (ARGS(io)->flags & CTL_LLF_DPO) 616 flags |= IO_DIRECT; 617 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 618 flags |= IO_SYNC; 619 620 bzero(&xuio, sizeof(xuio)); 621 if (beio->bio_cmd == BIO_READ) { 622 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 623 xuio.uio_rw = UIO_READ; 624 } else { 625 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 626 xuio.uio_rw = UIO_WRITE; 627 } 628 xuio.uio_offset = beio->io_offset; 629 xuio.uio_resid = beio->io_len; 630 xuio.uio_segflg = UIO_SYSSPACE; 631 xuio.uio_iov = beio->xiovecs; 632 xuio.uio_iovcnt = beio->num_segs; 633 xuio.uio_td = curthread; 634 635 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 636 xiovec->iov_base = beio->sg_segs[i].addr; 637 xiovec->iov_len = beio->sg_segs[i].len; 638 } 639 640 binuptime(&beio->ds_t0); 641 mtx_lock(&be_lun->io_lock); 642 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 643 mtx_unlock(&be_lun->io_lock); 644 645 if (beio->bio_cmd == BIO_READ) { 646 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 647 648 /* 649 * UFS pays attention to IO_DIRECT for reads. If the 650 * DIRECTIO option is configured into the kernel, it calls 651 * ffs_rawread(). But that only works for single-segment 652 * uios with user space addresses. In our case, with a 653 * kernel uio, it still reads into the buffer cache, but it 654 * will just try to release the buffer from the cache later 655 * on in ffs_read(). 656 * 657 * ZFS does not pay attention to IO_DIRECT for reads. 658 * 659 * UFS does not pay attention to IO_SYNC for reads. 660 * 661 * ZFS pays attention to IO_SYNC (which translates into the 662 * Solaris define FRSYNC for zfs_read()) for reads. It 663 * attempts to sync the file before reading. 664 * 665 * So, to attempt to provide some barrier semantics in the 666 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 667 */ 668 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 669 670 VOP_UNLOCK(be_lun->vn, 0); 671 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 672 } else { 673 struct mount *mountpoint; 674 int lock_flags; 675 676 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 677 678 if (MNT_SHARED_WRITES(mountpoint) 679 || ((mountpoint == NULL) 680 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 681 lock_flags = LK_SHARED; 682 else 683 lock_flags = LK_EXCLUSIVE; 684 685 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 686 687 /* 688 * UFS pays attention to IO_DIRECT for writes. The write 689 * is done asynchronously. (Normally the write would just 690 * get put into cache. 691 * 692 * UFS pays attention to IO_SYNC for writes. It will 693 * attempt to write the buffer out synchronously if that 694 * flag is set. 695 * 696 * ZFS does not pay attention to IO_DIRECT for writes. 697 * 698 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 699 * for writes. It will flush the transaction from the 700 * cache before returning. 701 * 702 * So if we've got the BIO_ORDERED flag set, we want 703 * IO_SYNC in either the UFS or ZFS case. 704 */ 705 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 706 VOP_UNLOCK(be_lun->vn, 0); 707 708 vn_finished_write(mountpoint); 709 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 710 } 711 712 mtx_lock(&be_lun->io_lock); 713 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 714 beio->ds_tag_type, beio->ds_trans_type, 715 /*now*/ NULL, /*then*/&beio->ds_t0); 716 mtx_unlock(&be_lun->io_lock); 717 718 /* 719 * If we got an error, set the sense data to "MEDIUM ERROR" and 720 * return the I/O to the user. 721 */ 722 if (error != 0) { 723 char path_str[32]; 724 725 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 726 printf("%s%s command returned errno %d\n", path_str, 727 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 728 if (error == ENOSPC) { 729 ctl_set_space_alloc_fail(&io->scsiio); 730 } else 731 ctl_set_medium_error(&io->scsiio); 732 ctl_complete_beio(beio); 733 return; 734 } 735 736 /* 737 * If this is a write or a verify, we're all done. 738 * If this is a read, we can now send the data to the user. 739 */ 740 if ((beio->bio_cmd == BIO_WRITE) || 741 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 742 ctl_set_success(&io->scsiio); 743 ctl_complete_beio(beio); 744 } else { 745#ifdef CTL_TIME_IO 746 getbintime(&io->io_hdr.dma_start_bt); 747#endif 748 ctl_datamove(io); 749 } 750} 751 752static void 753ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 754 struct ctl_be_block_io *beio) 755{ 756 struct ctl_be_block_devdata *dev_data; 757 union ctl_io *io; 758 struct uio xuio; 759 struct iovec *xiovec; 760 int flags; 761 int error, i; 762 763 DPRINTF("entered\n"); 764 765 dev_data = &be_lun->backend.dev; 766 io = beio->io; 767 flags = 0; 768 if (ARGS(io)->flags & CTL_LLF_DPO) 769 flags |= IO_DIRECT; 770 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 771 flags |= IO_SYNC; 772 773 bzero(&xuio, sizeof(xuio)); 774 if (beio->bio_cmd == BIO_READ) { 775 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 776 xuio.uio_rw = UIO_READ; 777 } else { 778 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 779 xuio.uio_rw = UIO_WRITE; 780 } 781 xuio.uio_offset = beio->io_offset; 782 xuio.uio_resid = beio->io_len; 783 xuio.uio_segflg = UIO_SYSSPACE; 784 xuio.uio_iov = beio->xiovecs; 785 xuio.uio_iovcnt = beio->num_segs; 786 xuio.uio_td = curthread; 787 788 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 789 xiovec->iov_base = beio->sg_segs[i].addr; 790 xiovec->iov_len = beio->sg_segs[i].len; 791 } 792 793 binuptime(&beio->ds_t0); 794 mtx_lock(&be_lun->io_lock); 795 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 796 mtx_unlock(&be_lun->io_lock); 797 798 if (beio->bio_cmd == BIO_READ) { 799 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 800 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 801 } else { 802 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 803 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 804 } 805 806 mtx_lock(&be_lun->io_lock); 807 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 808 beio->ds_tag_type, beio->ds_trans_type, 809 /*now*/ NULL, /*then*/&beio->ds_t0); 810 mtx_unlock(&be_lun->io_lock); 811 812 /* 813 * If we got an error, set the sense data to "MEDIUM ERROR" and 814 * return the I/O to the user. 815 */ 816 if (error != 0) { 817 if (error == ENOSPC) { 818 ctl_set_space_alloc_fail(&io->scsiio); 819 } else 820 ctl_set_medium_error(&io->scsiio); 821 ctl_complete_beio(beio); 822 return; 823 } 824 825 /* 826 * If this is a write or a verify, we're all done. 827 * If this is a read, we can now send the data to the user. 828 */ 829 if ((beio->bio_cmd == BIO_WRITE) || 830 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 831 ctl_set_success(&io->scsiio); 832 ctl_complete_beio(beio); 833 } else { 834#ifdef CTL_TIME_IO 835 getbintime(&io->io_hdr.dma_start_bt); 836#endif 837 ctl_datamove(io); 838 } 839} 840 841static void 842ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 843 struct ctl_be_block_io *beio) 844{ 845 struct bio *bio; 846 union ctl_io *io; 847 struct ctl_be_block_devdata *dev_data; 848 849 dev_data = &be_lun->backend.dev; 850 io = beio->io; 851 852 DPRINTF("entered\n"); 853 854 /* This can't fail, it's a blocking allocation. */ 855 bio = g_alloc_bio(); 856 857 bio->bio_cmd = BIO_FLUSH; 858 bio->bio_flags |= BIO_ORDERED; 859 bio->bio_dev = dev_data->cdev; 860 bio->bio_offset = 0; 861 bio->bio_data = 0; 862 bio->bio_done = ctl_be_block_biodone; 863 bio->bio_caller1 = beio; 864 bio->bio_pblkno = 0; 865 866 /* 867 * We don't need to acquire the LUN lock here, because we are only 868 * sending one bio, and so there is no other context to synchronize 869 * with. 870 */ 871 beio->num_bios_sent = 1; 872 beio->send_complete = 1; 873 874 binuptime(&beio->ds_t0); 875 mtx_lock(&be_lun->io_lock); 876 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 877 mtx_unlock(&be_lun->io_lock); 878 879 (*dev_data->csw->d_strategy)(bio); 880} 881 882static void 883ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 884 struct ctl_be_block_io *beio, 885 uint64_t off, uint64_t len, int last) 886{ 887 struct bio *bio; 888 struct ctl_be_block_devdata *dev_data; 889 uint64_t maxlen; 890 891 dev_data = &be_lun->backend.dev; 892 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 893 while (len > 0) { 894 bio = g_alloc_bio(); 895 bio->bio_cmd = BIO_DELETE; 896 bio->bio_dev = dev_data->cdev; 897 bio->bio_offset = off; 898 bio->bio_length = MIN(len, maxlen); 899 bio->bio_data = 0; 900 bio->bio_done = ctl_be_block_biodone; 901 bio->bio_caller1 = beio; 902 bio->bio_pblkno = off / be_lun->blocksize; 903 904 off += bio->bio_length; 905 len -= bio->bio_length; 906 907 mtx_lock(&be_lun->io_lock); 908 beio->num_bios_sent++; 909 if (last && len == 0) 910 beio->send_complete = 1; 911 mtx_unlock(&be_lun->io_lock); 912 913 (*dev_data->csw->d_strategy)(bio); 914 } 915} 916 917static void 918ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 919 struct ctl_be_block_io *beio) 920{ 921 union ctl_io *io; 922 struct ctl_be_block_devdata *dev_data; 923 struct ctl_ptr_len_flags *ptrlen; 924 struct scsi_unmap_desc *buf, *end; 925 uint64_t len; 926 927 dev_data = &be_lun->backend.dev; 928 io = beio->io; 929 930 DPRINTF("entered\n"); 931 932 binuptime(&beio->ds_t0); 933 mtx_lock(&be_lun->io_lock); 934 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 935 mtx_unlock(&be_lun->io_lock); 936 937 if (beio->io_offset == -1) { 938 beio->io_len = 0; 939 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 940 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 941 end = buf + ptrlen->len / sizeof(*buf); 942 for (; buf < end; buf++) { 943 len = (uint64_t)scsi_4btoul(buf->length) * 944 be_lun->blocksize; 945 beio->io_len += len; 946 ctl_be_block_unmap_dev_range(be_lun, beio, 947 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 948 (end - buf < 2) ? TRUE : FALSE); 949 } 950 } else 951 ctl_be_block_unmap_dev_range(be_lun, beio, 952 beio->io_offset, beio->io_len, TRUE); 953} 954 955static void 956ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 957 struct ctl_be_block_io *beio) 958{ 959 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 960 int i; 961 struct bio *bio; 962 struct ctl_be_block_devdata *dev_data; 963 off_t cur_offset; 964 int max_iosize; 965 966 DPRINTF("entered\n"); 967 968 dev_data = &be_lun->backend.dev; 969 970 /* 971 * We have to limit our I/O size to the maximum supported by the 972 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 973 * set it properly, use DFLTPHYS. 974 */ 975 max_iosize = dev_data->cdev->si_iosize_max; 976 if (max_iosize < PAGE_SIZE) 977 max_iosize = DFLTPHYS; 978 979 cur_offset = beio->io_offset; 980 for (i = 0; i < beio->num_segs; i++) { 981 size_t cur_size; 982 uint8_t *cur_ptr; 983 984 cur_size = beio->sg_segs[i].len; 985 cur_ptr = beio->sg_segs[i].addr; 986 987 while (cur_size > 0) { 988 /* This can't fail, it's a blocking allocation. */ 989 bio = g_alloc_bio(); 990 991 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 992 993 bio->bio_cmd = beio->bio_cmd; 994 bio->bio_dev = dev_data->cdev; 995 bio->bio_caller1 = beio; 996 bio->bio_length = min(cur_size, max_iosize); 997 bio->bio_offset = cur_offset; 998 bio->bio_data = cur_ptr; 999 bio->bio_done = ctl_be_block_biodone; 1000 bio->bio_pblkno = cur_offset / be_lun->blocksize; 1001 1002 cur_offset += bio->bio_length; 1003 cur_ptr += bio->bio_length; 1004 cur_size -= bio->bio_length; 1005 1006 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1007 beio->num_bios_sent++; 1008 } 1009 } 1010 binuptime(&beio->ds_t0); 1011 mtx_lock(&be_lun->io_lock); 1012 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1013 beio->send_complete = 1; 1014 mtx_unlock(&be_lun->io_lock); 1015 1016 /* 1017 * Fire off all allocated requests! 1018 */ 1019 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1020 TAILQ_REMOVE(&queue, bio, bio_queue); 1021 (*dev_data->csw->d_strategy)(bio); 1022 } 1023} 1024 1025static uint64_t 1026ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1027{ 1028 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1029 struct diocgattr_arg arg; 1030 int error; 1031 1032 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1033 return (UINT64_MAX); 1034 strlcpy(arg.name, attrname, sizeof(arg.name)); 1035 arg.len = sizeof(arg.value.off); 1036 error = dev_data->csw->d_ioctl(dev_data->cdev, 1037 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1038 if (error != 0) 1039 return (UINT64_MAX); 1040 return (arg.value.off); 1041} 1042 1043static void 1044ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1045{ 1046 union ctl_io *io; 1047 1048 io = beio->io; 1049 ctl_free_beio(beio); 1050 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1051 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1052 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1053 ctl_config_write_done(io); 1054 return; 1055 } 1056 1057 ctl_be_block_config_write(io); 1058} 1059 1060static void 1061ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1062 union ctl_io *io) 1063{ 1064 struct ctl_be_block_io *beio; 1065 struct ctl_be_block_softc *softc; 1066 struct ctl_lba_len_flags *lbalen; 1067 uint64_t len_left, lba; 1068 int i, seglen; 1069 uint8_t *buf, *end; 1070 1071 DPRINTF("entered\n"); 1072 1073 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1074 softc = be_lun->softc; 1075 lbalen = ARGS(beio->io); 1076 1077 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1078 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1079 ctl_free_beio(beio); 1080 ctl_set_invalid_field(&io->scsiio, 1081 /*sks_valid*/ 1, 1082 /*command*/ 1, 1083 /*field*/ 1, 1084 /*bit_valid*/ 0, 1085 /*bit*/ 0); 1086 ctl_config_write_done(io); 1087 return; 1088 } 1089 1090 switch (io->scsiio.tag_type) { 1091 case CTL_TAG_ORDERED: 1092 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1093 break; 1094 case CTL_TAG_HEAD_OF_QUEUE: 1095 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1096 break; 1097 case CTL_TAG_UNTAGGED: 1098 case CTL_TAG_SIMPLE: 1099 case CTL_TAG_ACA: 1100 default: 1101 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1102 break; 1103 } 1104 1105 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1106 beio->io_offset = lbalen->lba * be_lun->blocksize; 1107 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1108 beio->bio_cmd = BIO_DELETE; 1109 beio->ds_trans_type = DEVSTAT_FREE; 1110 1111 be_lun->unmap(be_lun, beio); 1112 return; 1113 } 1114 1115 beio->bio_cmd = BIO_WRITE; 1116 beio->ds_trans_type = DEVSTAT_WRITE; 1117 1118 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1119 (uintmax_t)lbalen->lba, lbalen->len); 1120 1121 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1122 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1123 1124 /* 1125 * Setup the S/G entry for this chunk. 1126 */ 1127 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1128 seglen -= seglen % be_lun->blocksize; 1129 beio->sg_segs[i].len = seglen; 1130 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1131 1132 DPRINTF("segment %d addr %p len %zd\n", i, 1133 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1134 1135 beio->num_segs++; 1136 len_left -= seglen; 1137 1138 buf = beio->sg_segs[i].addr; 1139 end = buf + seglen; 1140 for (; buf < end; buf += be_lun->blocksize) { 1141 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1142 if (lbalen->flags & SWS_LBDATA) 1143 scsi_ulto4b(lbalen->lba + lba, buf); 1144 lba++; 1145 } 1146 } 1147 1148 beio->io_offset = lbalen->lba * be_lun->blocksize; 1149 beio->io_len = lba * be_lun->blocksize; 1150 1151 /* We can not do all in one run. Correct and schedule rerun. */ 1152 if (len_left > 0) { 1153 lbalen->lba += lba; 1154 lbalen->len -= lba; 1155 beio->beio_cont = ctl_be_block_cw_done_ws; 1156 } 1157 1158 be_lun->dispatch(be_lun, beio); 1159} 1160 1161static void 1162ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1163 union ctl_io *io) 1164{ 1165 struct ctl_be_block_io *beio; 1166 struct ctl_be_block_softc *softc; 1167 struct ctl_ptr_len_flags *ptrlen; 1168 1169 DPRINTF("entered\n"); 1170 1171 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1172 softc = be_lun->softc; 1173 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1174 1175 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1176 ctl_free_beio(beio); 1177 ctl_set_invalid_field(&io->scsiio, 1178 /*sks_valid*/ 0, 1179 /*command*/ 1, 1180 /*field*/ 0, 1181 /*bit_valid*/ 0, 1182 /*bit*/ 0); 1183 ctl_config_write_done(io); 1184 return; 1185 } 1186 1187 switch (io->scsiio.tag_type) { 1188 case CTL_TAG_ORDERED: 1189 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1190 break; 1191 case CTL_TAG_HEAD_OF_QUEUE: 1192 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1193 break; 1194 case CTL_TAG_UNTAGGED: 1195 case CTL_TAG_SIMPLE: 1196 case CTL_TAG_ACA: 1197 default: 1198 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1199 break; 1200 } 1201 1202 beio->io_len = 0; 1203 beio->io_offset = -1; 1204 1205 beio->bio_cmd = BIO_DELETE; 1206 beio->ds_trans_type = DEVSTAT_FREE; 1207 1208 DPRINTF("UNMAP\n"); 1209 1210 be_lun->unmap(be_lun, beio); 1211} 1212 1213static void 1214ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1215{ 1216 union ctl_io *io; 1217 1218 io = beio->io; 1219 ctl_free_beio(beio); 1220 ctl_config_write_done(io); 1221} 1222 1223static void 1224ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1225 union ctl_io *io) 1226{ 1227 struct ctl_be_block_io *beio; 1228 struct ctl_be_block_softc *softc; 1229 1230 DPRINTF("entered\n"); 1231 1232 softc = be_lun->softc; 1233 beio = ctl_alloc_beio(softc); 1234 beio->io = io; 1235 beio->lun = be_lun; 1236 beio->beio_cont = ctl_be_block_cw_done; 1237 PRIV(io)->ptr = (void *)beio; 1238 1239 switch (io->scsiio.cdb[0]) { 1240 case SYNCHRONIZE_CACHE: 1241 case SYNCHRONIZE_CACHE_16: 1242 beio->bio_cmd = BIO_FLUSH; 1243 beio->ds_trans_type = DEVSTAT_NO_DATA; 1244 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1245 beio->io_len = 0; 1246 be_lun->lun_flush(be_lun, beio); 1247 break; 1248 case WRITE_SAME_10: 1249 case WRITE_SAME_16: 1250 ctl_be_block_cw_dispatch_ws(be_lun, io); 1251 break; 1252 case UNMAP: 1253 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1254 break; 1255 default: 1256 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1257 break; 1258 } 1259} 1260 1261SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1262SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1263SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1264SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1265 1266static void 1267ctl_be_block_next(struct ctl_be_block_io *beio) 1268{ 1269 struct ctl_be_block_lun *be_lun; 1270 union ctl_io *io; 1271 1272 io = beio->io; 1273 be_lun = beio->lun; 1274 ctl_free_beio(beio); 1275 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1276 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1277 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1278 ctl_data_submit_done(io); 1279 return; 1280 } 1281 1282 io->io_hdr.status &= ~CTL_STATUS_MASK; 1283 io->io_hdr.status |= CTL_STATUS_NONE; 1284 1285 mtx_lock(&be_lun->queue_lock); 1286 /* 1287 * XXX KDM make sure that links is okay to use at this point. 1288 * Otherwise, we either need to add another field to ctl_io_hdr, 1289 * or deal with resource allocation here. 1290 */ 1291 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1292 mtx_unlock(&be_lun->queue_lock); 1293 1294 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1295} 1296 1297static void 1298ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1299 union ctl_io *io) 1300{ 1301 struct ctl_be_block_io *beio; 1302 struct ctl_be_block_softc *softc; 1303 struct ctl_lba_len_flags *lbalen; 1304 struct ctl_ptr_len_flags *bptrlen; 1305 uint64_t len_left, lbas; 1306 int i; 1307 1308 softc = be_lun->softc; 1309 1310 DPRINTF("entered\n"); 1311 1312 lbalen = ARGS(io); 1313 if (lbalen->flags & CTL_LLF_WRITE) { 1314 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1315 } else { 1316 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1317 } 1318 1319 beio = ctl_alloc_beio(softc); 1320 beio->io = io; 1321 beio->lun = be_lun; 1322 bptrlen = PRIV(io); 1323 bptrlen->ptr = (void *)beio; 1324 1325 switch (io->scsiio.tag_type) { 1326 case CTL_TAG_ORDERED: 1327 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1328 break; 1329 case CTL_TAG_HEAD_OF_QUEUE: 1330 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1331 break; 1332 case CTL_TAG_UNTAGGED: 1333 case CTL_TAG_SIMPLE: 1334 case CTL_TAG_ACA: 1335 default: 1336 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1337 break; 1338 } 1339 1340 if (lbalen->flags & CTL_LLF_WRITE) { 1341 beio->bio_cmd = BIO_WRITE; 1342 beio->ds_trans_type = DEVSTAT_WRITE; 1343 } else { 1344 beio->bio_cmd = BIO_READ; 1345 beio->ds_trans_type = DEVSTAT_READ; 1346 } 1347 1348 DPRINTF("%s at LBA %jx len %u @%ju\n", 1349 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1350 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1351 if (lbalen->flags & CTL_LLF_COMPARE) 1352 lbas = CTLBLK_HALF_IO_SIZE; 1353 else 1354 lbas = CTLBLK_MAX_IO_SIZE; 1355 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1356 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1357 beio->io_len = lbas * be_lun->blocksize; 1358 bptrlen->len += lbas; 1359 1360 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1361 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1362 i, CTLBLK_MAX_SEGS)); 1363 1364 /* 1365 * Setup the S/G entry for this chunk. 1366 */ 1367 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1368 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1369 1370 DPRINTF("segment %d addr %p len %zd\n", i, 1371 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1372 1373 /* Set up second segment for compare operation. */ 1374 if (lbalen->flags & CTL_LLF_COMPARE) { 1375 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1376 beio->sg_segs[i].len; 1377 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1378 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1379 } 1380 1381 beio->num_segs++; 1382 len_left -= beio->sg_segs[i].len; 1383 } 1384 if (bptrlen->len < lbalen->len) 1385 beio->beio_cont = ctl_be_block_next; 1386 io->scsiio.be_move_done = ctl_be_block_move_done; 1387 /* For compare we have separate S/G lists for read and datamove. */ 1388 if (lbalen->flags & CTL_LLF_COMPARE) 1389 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1390 else 1391 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1392 io->scsiio.kern_data_len = beio->io_len; 1393 io->scsiio.kern_data_resid = 0; 1394 io->scsiio.kern_sg_entries = beio->num_segs; 1395 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1396 1397 /* 1398 * For the read case, we need to read the data into our buffers and 1399 * then we can send it back to the user. For the write case, we 1400 * need to get the data from the user first. 1401 */ 1402 if (beio->bio_cmd == BIO_READ) { 1403 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1404 be_lun->dispatch(be_lun, beio); 1405 } else { 1406 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1407#ifdef CTL_TIME_IO 1408 getbintime(&io->io_hdr.dma_start_bt); 1409#endif 1410 ctl_datamove(io); 1411 } 1412} 1413 1414static void 1415ctl_be_block_worker(void *context, int pending) 1416{ 1417 struct ctl_be_block_lun *be_lun; 1418 struct ctl_be_block_softc *softc; 1419 union ctl_io *io; 1420 1421 be_lun = (struct ctl_be_block_lun *)context; 1422 softc = be_lun->softc; 1423 1424 DPRINTF("entered\n"); 1425 1426 mtx_lock(&be_lun->queue_lock); 1427 for (;;) { 1428 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1429 if (io != NULL) { 1430 struct ctl_be_block_io *beio; 1431 1432 DPRINTF("datamove queue\n"); 1433 1434 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1435 ctl_io_hdr, links); 1436 1437 mtx_unlock(&be_lun->queue_lock); 1438 1439 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1440 1441 be_lun->dispatch(be_lun, beio); 1442 1443 mtx_lock(&be_lun->queue_lock); 1444 continue; 1445 } 1446 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1447 if (io != NULL) { 1448 1449 DPRINTF("config write queue\n"); 1450 1451 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1452 ctl_io_hdr, links); 1453 1454 mtx_unlock(&be_lun->queue_lock); 1455 1456 ctl_be_block_cw_dispatch(be_lun, io); 1457 1458 mtx_lock(&be_lun->queue_lock); 1459 continue; 1460 } 1461 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1462 if (io != NULL) { 1463 DPRINTF("input queue\n"); 1464 1465 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1466 ctl_io_hdr, links); 1467 mtx_unlock(&be_lun->queue_lock); 1468 1469 /* 1470 * We must drop the lock, since this routine and 1471 * its children may sleep. 1472 */ 1473 ctl_be_block_dispatch(be_lun, io); 1474 1475 mtx_lock(&be_lun->queue_lock); 1476 continue; 1477 } 1478 1479 /* 1480 * If we get here, there is no work left in the queues, so 1481 * just break out and let the task queue go to sleep. 1482 */ 1483 break; 1484 } 1485 mtx_unlock(&be_lun->queue_lock); 1486} 1487 1488/* 1489 * Entry point from CTL to the backend for I/O. We queue everything to a 1490 * work thread, so this just puts the I/O on a queue and wakes up the 1491 * thread. 1492 */ 1493static int 1494ctl_be_block_submit(union ctl_io *io) 1495{ 1496 struct ctl_be_block_lun *be_lun; 1497 struct ctl_be_lun *ctl_be_lun; 1498 1499 DPRINTF("entered\n"); 1500 1501 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1502 CTL_PRIV_BACKEND_LUN].ptr; 1503 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1504 1505 /* 1506 * Make sure we only get SCSI I/O. 1507 */ 1508 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1509 "%#x) encountered", io->io_hdr.io_type)); 1510 1511 PRIV(io)->len = 0; 1512 1513 mtx_lock(&be_lun->queue_lock); 1514 /* 1515 * XXX KDM make sure that links is okay to use at this point. 1516 * Otherwise, we either need to add another field to ctl_io_hdr, 1517 * or deal with resource allocation here. 1518 */ 1519 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1520 mtx_unlock(&be_lun->queue_lock); 1521 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1522 1523 return (CTL_RETVAL_COMPLETE); 1524} 1525 1526static int 1527ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1528 int flag, struct thread *td) 1529{ 1530 struct ctl_be_block_softc *softc; 1531 int error; 1532 1533 softc = &backend_block_softc; 1534 1535 error = 0; 1536 1537 switch (cmd) { 1538 case CTL_LUN_REQ: { 1539 struct ctl_lun_req *lun_req; 1540 1541 lun_req = (struct ctl_lun_req *)addr; 1542 1543 switch (lun_req->reqtype) { 1544 case CTL_LUNREQ_CREATE: 1545 error = ctl_be_block_create(softc, lun_req); 1546 break; 1547 case CTL_LUNREQ_RM: 1548 error = ctl_be_block_rm(softc, lun_req); 1549 break; 1550 case CTL_LUNREQ_MODIFY: 1551 error = ctl_be_block_modify(softc, lun_req); 1552 break; 1553 default: 1554 lun_req->status = CTL_LUN_ERROR; 1555 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1556 "invalid LUN request type %d", 1557 lun_req->reqtype); 1558 break; 1559 } 1560 break; 1561 } 1562 default: 1563 error = ENOTTY; 1564 break; 1565 } 1566 1567 return (error); 1568} 1569 1570static int 1571ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1572{ 1573 struct ctl_be_block_filedata *file_data; 1574 struct ctl_lun_create_params *params; 1575 struct vattr vattr; 1576 off_t pss; 1577 int error; 1578 1579 error = 0; 1580 file_data = &be_lun->backend.file; 1581 params = &be_lun->params; 1582 1583 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1584 be_lun->dispatch = ctl_be_block_dispatch_file; 1585 be_lun->lun_flush = ctl_be_block_flush_file; 1586 1587 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1588 if (error != 0) { 1589 snprintf(req->error_str, sizeof(req->error_str), 1590 "error calling VOP_GETATTR() for file %s", 1591 be_lun->dev_path); 1592 return (error); 1593 } 1594 1595 /* 1596 * Verify that we have the ability to upgrade to exclusive 1597 * access on this file so we can trap errors at open instead 1598 * of reporting them during first access. 1599 */ 1600 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1601 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1602 if (be_lun->vn->v_iflag & VI_DOOMED) { 1603 error = EBADF; 1604 snprintf(req->error_str, sizeof(req->error_str), 1605 "error locking file %s", be_lun->dev_path); 1606 return (error); 1607 } 1608 } 1609 1610 1611 file_data->cred = crhold(curthread->td_ucred); 1612 if (params->lun_size_bytes != 0) 1613 be_lun->size_bytes = params->lun_size_bytes; 1614 else 1615 be_lun->size_bytes = vattr.va_size; 1616 /* 1617 * We set the multi thread flag for file operations because all 1618 * filesystems (in theory) are capable of allowing multiple readers 1619 * of a file at once. So we want to get the maximum possible 1620 * concurrency. 1621 */ 1622 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1623 1624 /* 1625 * For files we can use any logical block size. Prefer 512 bytes 1626 * for compatibility reasons. If file's vattr.va_blocksize 1627 * (preferred I/O block size) is bigger and multiple to chosen 1628 * logical block size -- report it as physical block size. 1629 */ 1630 if (params->blocksize_bytes != 0) 1631 be_lun->blocksize = params->blocksize_bytes; 1632 else 1633 be_lun->blocksize = 512; 1634 pss = vattr.va_blocksize / be_lun->blocksize; 1635 if ((pss > 0) && (pss * be_lun->blocksize == vattr.va_blocksize) && 1636 ((pss & (pss - 1)) == 0)) { 1637 be_lun->pblockexp = fls(pss) - 1; 1638 be_lun->pblockoff = 0; 1639 } 1640 1641 /* 1642 * Sanity check. The media size has to be at least one 1643 * sector long. 1644 */ 1645 if (be_lun->size_bytes < be_lun->blocksize) { 1646 error = EINVAL; 1647 snprintf(req->error_str, sizeof(req->error_str), 1648 "file %s size %ju < block size %u", be_lun->dev_path, 1649 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1650 } 1651 return (error); 1652} 1653 1654static int 1655ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1656{ 1657 struct ctl_lun_create_params *params; 1658 struct vattr vattr; 1659 struct cdev *dev; 1660 struct cdevsw *devsw; 1661 int error; 1662 off_t ps, pss, po, pos; 1663 1664 params = &be_lun->params; 1665 1666 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1667 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1668 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1669 &be_lun->backend.dev.dev_ref); 1670 if (be_lun->backend.dev.csw == NULL) 1671 panic("Unable to retrieve device switch"); 1672 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) 1673 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1674 else 1675 be_lun->dispatch = ctl_be_block_dispatch_dev; 1676 be_lun->lun_flush = ctl_be_block_flush_dev; 1677 be_lun->unmap = ctl_be_block_unmap_dev; 1678 be_lun->getattr = ctl_be_block_getattr_dev; 1679 1680 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1681 if (error) { 1682 snprintf(req->error_str, sizeof(req->error_str), 1683 "error getting vnode attributes for device %s", 1684 be_lun->dev_path); 1685 return (error); 1686 } 1687 1688 dev = be_lun->vn->v_rdev; 1689 devsw = dev->si_devsw; 1690 if (!devsw->d_ioctl) { 1691 snprintf(req->error_str, sizeof(req->error_str), 1692 "no d_ioctl for device %s!", 1693 be_lun->dev_path); 1694 return (ENODEV); 1695 } 1696 1697 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1698 (caddr_t)&be_lun->blocksize, FREAD, 1699 curthread); 1700 if (error) { 1701 snprintf(req->error_str, sizeof(req->error_str), 1702 "error %d returned for DIOCGSECTORSIZE ioctl " 1703 "on %s!", error, be_lun->dev_path); 1704 return (error); 1705 } 1706 1707 /* 1708 * If the user has asked for a blocksize that is greater than the 1709 * backing device's blocksize, we can do it only if the blocksize 1710 * the user is asking for is an even multiple of the underlying 1711 * device's blocksize. 1712 */ 1713 if ((params->blocksize_bytes != 0) 1714 && (params->blocksize_bytes > be_lun->blocksize)) { 1715 uint32_t bs_multiple, tmp_blocksize; 1716 1717 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1718 1719 tmp_blocksize = bs_multiple * be_lun->blocksize; 1720 1721 if (tmp_blocksize == params->blocksize_bytes) { 1722 be_lun->blocksize = params->blocksize_bytes; 1723 } else { 1724 snprintf(req->error_str, sizeof(req->error_str), 1725 "requested blocksize %u is not an even " 1726 "multiple of backing device blocksize %u", 1727 params->blocksize_bytes, 1728 be_lun->blocksize); 1729 return (EINVAL); 1730 1731 } 1732 } else if ((params->blocksize_bytes != 0) 1733 && (params->blocksize_bytes != be_lun->blocksize)) { 1734 snprintf(req->error_str, sizeof(req->error_str), 1735 "requested blocksize %u < backing device " 1736 "blocksize %u", params->blocksize_bytes, 1737 be_lun->blocksize); 1738 return (EINVAL); 1739 } 1740 1741 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1742 (caddr_t)&be_lun->size_bytes, FREAD, 1743 curthread); 1744 if (error) { 1745 snprintf(req->error_str, sizeof(req->error_str), 1746 "error %d returned for DIOCGMEDIASIZE " 1747 " ioctl on %s!", error, 1748 be_lun->dev_path); 1749 return (error); 1750 } 1751 1752 if (params->lun_size_bytes != 0) { 1753 if (params->lun_size_bytes > be_lun->size_bytes) { 1754 snprintf(req->error_str, sizeof(req->error_str), 1755 "requested LUN size %ju > backing device " 1756 "size %ju", 1757 (uintmax_t)params->lun_size_bytes, 1758 (uintmax_t)be_lun->size_bytes); 1759 return (EINVAL); 1760 } 1761 1762 be_lun->size_bytes = params->lun_size_bytes; 1763 } 1764 1765 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1766 (caddr_t)&ps, FREAD, curthread); 1767 if (error) 1768 ps = po = 0; 1769 else { 1770 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1771 (caddr_t)&po, FREAD, curthread); 1772 if (error) 1773 po = 0; 1774 } 1775 pss = ps / be_lun->blocksize; 1776 pos = po / be_lun->blocksize; 1777 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1778 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1779 be_lun->pblockexp = fls(pss) - 1; 1780 be_lun->pblockoff = (pss - pos) % pss; 1781 } 1782 1783 return (0); 1784} 1785 1786static int 1787ctl_be_block_close(struct ctl_be_block_lun *be_lun) 1788{ 1789 DROP_GIANT(); 1790 if (be_lun->vn) { 1791 int flags = FREAD | FWRITE; 1792 1793 switch (be_lun->dev_type) { 1794 case CTL_BE_BLOCK_DEV: 1795 if (be_lun->backend.dev.csw) { 1796 dev_relthread(be_lun->backend.dev.cdev, 1797 be_lun->backend.dev.dev_ref); 1798 be_lun->backend.dev.csw = NULL; 1799 be_lun->backend.dev.cdev = NULL; 1800 } 1801 break; 1802 case CTL_BE_BLOCK_FILE: 1803 break; 1804 case CTL_BE_BLOCK_NONE: 1805 break; 1806 default: 1807 panic("Unexpected backend type."); 1808 break; 1809 } 1810 1811 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1812 be_lun->vn = NULL; 1813 1814 switch (be_lun->dev_type) { 1815 case CTL_BE_BLOCK_DEV: 1816 break; 1817 case CTL_BE_BLOCK_FILE: 1818 if (be_lun->backend.file.cred != NULL) { 1819 crfree(be_lun->backend.file.cred); 1820 be_lun->backend.file.cred = NULL; 1821 } 1822 break; 1823 case CTL_BE_BLOCK_NONE: 1824 break; 1825 default: 1826 panic("Unexpected backend type."); 1827 break; 1828 } 1829 be_lun->dev_type = CTL_BE_BLOCK_NONE; 1830 } 1831 PICKUP_GIANT(); 1832 1833 return (0); 1834} 1835 1836static int 1837ctl_be_block_open(struct ctl_be_block_softc *softc, 1838 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1839{ 1840 struct nameidata nd; 1841 int flags; 1842 int error; 1843 1844 /* 1845 * XXX KDM allow a read-only option? 1846 */ 1847 flags = FREAD | FWRITE; 1848 error = 0; 1849 1850 if (rootvnode == NULL) { 1851 snprintf(req->error_str, sizeof(req->error_str), 1852 "Root filesystem is not mounted"); 1853 return (1); 1854 } 1855 1856 if (!curthread->td_proc->p_fd->fd_cdir) { 1857 curthread->td_proc->p_fd->fd_cdir = rootvnode; 1858 VREF(rootvnode); 1859 } 1860 if (!curthread->td_proc->p_fd->fd_rdir) { 1861 curthread->td_proc->p_fd->fd_rdir = rootvnode; 1862 VREF(rootvnode); 1863 } 1864 if (!curthread->td_proc->p_fd->fd_jdir) { 1865 curthread->td_proc->p_fd->fd_jdir = rootvnode; 1866 VREF(rootvnode); 1867 } 1868 1869 again: 1870 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1871 error = vn_open(&nd, &flags, 0, NULL); 1872 if (error) { 1873 /* 1874 * This is the only reasonable guess we can make as far as 1875 * path if the user doesn't give us a fully qualified path. 1876 * If they want to specify a file, they need to specify the 1877 * full path. 1878 */ 1879 if (be_lun->dev_path[0] != '/') { 1880 char *dev_path = "/dev/"; 1881 char *dev_name; 1882 1883 /* Try adding device path at beginning of name */ 1884 dev_name = malloc(strlen(be_lun->dev_path) 1885 + strlen(dev_path) + 1, 1886 M_CTLBLK, M_WAITOK); 1887 if (dev_name) { 1888 sprintf(dev_name, "%s%s", dev_path, 1889 be_lun->dev_path); 1890 free(be_lun->dev_path, M_CTLBLK); 1891 be_lun->dev_path = dev_name; 1892 goto again; 1893 } 1894 } 1895 snprintf(req->error_str, sizeof(req->error_str), 1896 "error opening %s: %d", be_lun->dev_path, error); 1897 return (error); 1898 } 1899 1900 NDFREE(&nd, NDF_ONLY_PNBUF); 1901 1902 be_lun->vn = nd.ni_vp; 1903 1904 /* We only support disks and files. */ 1905 if (vn_isdisk(be_lun->vn, &error)) { 1906 error = ctl_be_block_open_dev(be_lun, req); 1907 } else if (be_lun->vn->v_type == VREG) { 1908 error = ctl_be_block_open_file(be_lun, req); 1909 } else { 1910 error = EINVAL; 1911 snprintf(req->error_str, sizeof(req->error_str), 1912 "%s is not a disk or plain file", be_lun->dev_path); 1913 } 1914 VOP_UNLOCK(be_lun->vn, 0); 1915 1916 if (error != 0) { 1917 ctl_be_block_close(be_lun); 1918 return (error); 1919 } 1920 1921 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1922 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1923 1924 return (0); 1925} 1926 1927static int 1928ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1929{ 1930 struct ctl_be_block_lun *be_lun; 1931 struct ctl_lun_create_params *params; 1932 char num_thread_str[16]; 1933 char tmpstr[32]; 1934 char *value; 1935 int retval, num_threads, unmap; 1936 int tmp_num_threads; 1937 1938 params = &req->reqdata.create; 1939 retval = 0; 1940 req->status = CTL_LUN_OK; 1941 1942 num_threads = cbb_num_threads; 1943 1944 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1945 1946 be_lun->params = req->reqdata.create; 1947 be_lun->softc = softc; 1948 STAILQ_INIT(&be_lun->input_queue); 1949 STAILQ_INIT(&be_lun->config_write_queue); 1950 STAILQ_INIT(&be_lun->datamove_queue); 1951 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1952 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1953 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1954 ctl_init_opts(&be_lun->ctl_be_lun.options, 1955 req->num_be_args, req->kern_be_args); 1956 1957 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1958 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1959 1960 if (be_lun->lun_zone == NULL) { 1961 snprintf(req->error_str, sizeof(req->error_str), 1962 "error allocating UMA zone"); 1963 goto bailout_error; 1964 } 1965 1966 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1967 be_lun->ctl_be_lun.lun_type = params->device_type; 1968 else 1969 be_lun->ctl_be_lun.lun_type = T_DIRECT; 1970 1971 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1972 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1973 if (value == NULL) { 1974 snprintf(req->error_str, sizeof(req->error_str), 1975 "no file argument specified"); 1976 goto bailout_error; 1977 } 1978 be_lun->dev_path = strdup(value, M_CTLBLK); 1979 be_lun->blocksize = 512; 1980 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1981 1982 retval = ctl_be_block_open(softc, be_lun, req); 1983 if (retval != 0) { 1984 retval = 0; 1985 req->status = CTL_LUN_WARNING; 1986 } 1987 } else { 1988 /* 1989 * For processor devices, we don't have any size. 1990 */ 1991 be_lun->blocksize = 0; 1992 be_lun->pblockexp = 0; 1993 be_lun->pblockoff = 0; 1994 be_lun->size_blocks = 0; 1995 be_lun->size_bytes = 0; 1996 be_lun->ctl_be_lun.maxlba = 0; 1997 1998 /* 1999 * Default to just 1 thread for processor devices. 2000 */ 2001 num_threads = 1; 2002 } 2003 2004 /* 2005 * XXX This searching loop might be refactored to be combined with 2006 * the loop above, 2007 */ 2008 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2009 if (value != NULL) { 2010 tmp_num_threads = strtol(value, NULL, 0); 2011 2012 /* 2013 * We don't let the user specify less than one 2014 * thread, but hope he's clueful enough not to 2015 * specify 1000 threads. 2016 */ 2017 if (tmp_num_threads < 1) { 2018 snprintf(req->error_str, sizeof(req->error_str), 2019 "invalid number of threads %s", 2020 num_thread_str); 2021 goto bailout_error; 2022 } 2023 num_threads = tmp_num_threads; 2024 } 2025 unmap = (be_lun->dispatch == ctl_be_block_dispatch_zvol); 2026 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2027 if (value != NULL) 2028 unmap = (strcmp(value, "on") == 0); 2029 2030 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2031 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2032 if (be_lun->vn == NULL) 2033 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2034 if (unmap) 2035 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2036 be_lun->ctl_be_lun.be_lun = be_lun; 2037 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2038 0 : (be_lun->size_blocks - 1); 2039 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2040 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2041 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2042 if (be_lun->dispatch == ctl_be_block_dispatch_zvol && 2043 be_lun->blocksize != 0) 2044 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / 2045 be_lun->blocksize; 2046 /* Tell the user the blocksize we ended up using */ 2047 params->lun_size_bytes = be_lun->size_bytes; 2048 params->blocksize_bytes = be_lun->blocksize; 2049 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2050 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2051 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2052 } else 2053 be_lun->ctl_be_lun.req_lun_id = 0; 2054 2055 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2056 be_lun->ctl_be_lun.lun_config_status = 2057 ctl_be_block_lun_config_status; 2058 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2059 2060 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2061 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2062 softc->num_luns); 2063 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2064 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2065 sizeof(tmpstr))); 2066 2067 /* Tell the user what we used for a serial number */ 2068 strncpy((char *)params->serial_num, tmpstr, 2069 ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 2070 } else { 2071 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2072 params->serial_num, 2073 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2074 sizeof(params->serial_num))); 2075 } 2076 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2077 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2078 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2079 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2080 sizeof(tmpstr))); 2081 2082 /* Tell the user what we used for a device ID */ 2083 strncpy((char *)params->device_id, tmpstr, 2084 ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 2085 } else { 2086 strncpy((char *)be_lun->ctl_be_lun.device_id, 2087 params->device_id, 2088 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2089 sizeof(params->device_id))); 2090 } 2091 2092 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2093 2094 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2095 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2096 2097 if (be_lun->io_taskqueue == NULL) { 2098 snprintf(req->error_str, sizeof(req->error_str), 2099 "unable to create taskqueue"); 2100 goto bailout_error; 2101 } 2102 2103 /* 2104 * Note that we start the same number of threads by default for 2105 * both the file case and the block device case. For the file 2106 * case, we need multiple threads to allow concurrency, because the 2107 * vnode interface is designed to be a blocking interface. For the 2108 * block device case, ZFS zvols at least will block the caller's 2109 * context in many instances, and so we need multiple threads to 2110 * overcome that problem. Other block devices don't need as many 2111 * threads, but they shouldn't cause too many problems. 2112 * 2113 * If the user wants to just have a single thread for a block 2114 * device, he can specify that when the LUN is created, or change 2115 * the tunable/sysctl to alter the default number of threads. 2116 */ 2117 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2118 /*num threads*/num_threads, 2119 /*priority*/PWAIT, 2120 /*thread name*/ 2121 "%s taskq", be_lun->lunname); 2122 2123 if (retval != 0) 2124 goto bailout_error; 2125 2126 be_lun->num_threads = num_threads; 2127 2128 mtx_lock(&softc->lock); 2129 softc->num_luns++; 2130 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2131 2132 mtx_unlock(&softc->lock); 2133 2134 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2135 if (retval != 0) { 2136 mtx_lock(&softc->lock); 2137 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2138 links); 2139 softc->num_luns--; 2140 mtx_unlock(&softc->lock); 2141 snprintf(req->error_str, sizeof(req->error_str), 2142 "ctl_add_lun() returned error %d, see dmesg for " 2143 "details", retval); 2144 retval = 0; 2145 goto bailout_error; 2146 } 2147 2148 mtx_lock(&softc->lock); 2149 2150 /* 2151 * Tell the config_status routine that we're waiting so it won't 2152 * clean up the LUN in the event of an error. 2153 */ 2154 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2155 2156 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2157 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2158 if (retval == EINTR) 2159 break; 2160 } 2161 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2162 2163 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2164 snprintf(req->error_str, sizeof(req->error_str), 2165 "LUN configuration error, see dmesg for details"); 2166 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2167 links); 2168 softc->num_luns--; 2169 mtx_unlock(&softc->lock); 2170 goto bailout_error; 2171 } else { 2172 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2173 } 2174 2175 mtx_unlock(&softc->lock); 2176 2177 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2178 be_lun->blocksize, 2179 DEVSTAT_ALL_SUPPORTED, 2180 be_lun->ctl_be_lun.lun_type 2181 | DEVSTAT_TYPE_IF_OTHER, 2182 DEVSTAT_PRIORITY_OTHER); 2183 2184 return (retval); 2185 2186bailout_error: 2187 req->status = CTL_LUN_ERROR; 2188 2189 if (be_lun->io_taskqueue != NULL) 2190 taskqueue_free(be_lun->io_taskqueue); 2191 ctl_be_block_close(be_lun); 2192 if (be_lun->dev_path != NULL) 2193 free(be_lun->dev_path, M_CTLBLK); 2194 if (be_lun->lun_zone != NULL) 2195 uma_zdestroy(be_lun->lun_zone); 2196 ctl_free_opts(&be_lun->ctl_be_lun.options); 2197 mtx_destroy(&be_lun->queue_lock); 2198 mtx_destroy(&be_lun->io_lock); 2199 free(be_lun, M_CTLBLK); 2200 2201 return (retval); 2202} 2203 2204static int 2205ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2206{ 2207 struct ctl_lun_rm_params *params; 2208 struct ctl_be_block_lun *be_lun; 2209 int retval; 2210 2211 params = &req->reqdata.rm; 2212 2213 mtx_lock(&softc->lock); 2214 2215 be_lun = NULL; 2216 2217 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2218 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2219 break; 2220 } 2221 mtx_unlock(&softc->lock); 2222 2223 if (be_lun == NULL) { 2224 snprintf(req->error_str, sizeof(req->error_str), 2225 "LUN %u is not managed by the block backend", 2226 params->lun_id); 2227 goto bailout_error; 2228 } 2229 2230 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2231 2232 if (retval != 0) { 2233 snprintf(req->error_str, sizeof(req->error_str), 2234 "error %d returned from ctl_disable_lun() for " 2235 "LUN %d", retval, params->lun_id); 2236 goto bailout_error; 2237 2238 } 2239 2240 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2241 if (retval != 0) { 2242 snprintf(req->error_str, sizeof(req->error_str), 2243 "error %d returned from ctl_invalidate_lun() for " 2244 "LUN %d", retval, params->lun_id); 2245 goto bailout_error; 2246 } 2247 2248 mtx_lock(&softc->lock); 2249 2250 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2251 2252 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2253 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2254 if (retval == EINTR) 2255 break; 2256 } 2257 2258 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2259 2260 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2261 snprintf(req->error_str, sizeof(req->error_str), 2262 "interrupted waiting for LUN to be freed"); 2263 mtx_unlock(&softc->lock); 2264 goto bailout_error; 2265 } 2266 2267 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2268 2269 softc->num_luns--; 2270 mtx_unlock(&softc->lock); 2271 2272 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2273 2274 taskqueue_free(be_lun->io_taskqueue); 2275 2276 ctl_be_block_close(be_lun); 2277 2278 if (be_lun->disk_stats != NULL) 2279 devstat_remove_entry(be_lun->disk_stats); 2280 2281 uma_zdestroy(be_lun->lun_zone); 2282 2283 ctl_free_opts(&be_lun->ctl_be_lun.options); 2284 free(be_lun->dev_path, M_CTLBLK); 2285 mtx_destroy(&be_lun->queue_lock); 2286 mtx_destroy(&be_lun->io_lock); 2287 free(be_lun, M_CTLBLK); 2288 2289 req->status = CTL_LUN_OK; 2290 2291 return (0); 2292 2293bailout_error: 2294 2295 req->status = CTL_LUN_ERROR; 2296 2297 return (0); 2298} 2299 2300static int 2301ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2302 struct ctl_lun_req *req) 2303{ 2304 struct vattr vattr; 2305 int error; 2306 struct ctl_lun_create_params *params = &be_lun->params; 2307 2308 if (params->lun_size_bytes != 0) { 2309 be_lun->size_bytes = params->lun_size_bytes; 2310 } else { 2311 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2312 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2313 VOP_UNLOCK(be_lun->vn, 0); 2314 if (error != 0) { 2315 snprintf(req->error_str, sizeof(req->error_str), 2316 "error calling VOP_GETATTR() for file %s", 2317 be_lun->dev_path); 2318 return (error); 2319 } 2320 2321 be_lun->size_bytes = vattr.va_size; 2322 } 2323 2324 return (0); 2325} 2326 2327static int 2328ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2329 struct ctl_lun_req *req) 2330{ 2331 struct ctl_be_block_devdata *dev_data; 2332 int error; 2333 struct ctl_lun_create_params *params = &be_lun->params; 2334 uint64_t size_bytes; 2335 2336 dev_data = &be_lun->backend.dev; 2337 if (!dev_data->csw->d_ioctl) { 2338 snprintf(req->error_str, sizeof(req->error_str), 2339 "no d_ioctl for device %s!", be_lun->dev_path); 2340 return (ENODEV); 2341 } 2342 2343 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2344 (caddr_t)&size_bytes, FREAD, 2345 curthread); 2346 if (error) { 2347 snprintf(req->error_str, sizeof(req->error_str), 2348 "error %d returned for DIOCGMEDIASIZE ioctl " 2349 "on %s!", error, be_lun->dev_path); 2350 return (error); 2351 } 2352 2353 if (params->lun_size_bytes != 0) { 2354 if (params->lun_size_bytes > size_bytes) { 2355 snprintf(req->error_str, sizeof(req->error_str), 2356 "requested LUN size %ju > backing device " 2357 "size %ju", 2358 (uintmax_t)params->lun_size_bytes, 2359 (uintmax_t)size_bytes); 2360 return (EINVAL); 2361 } 2362 2363 be_lun->size_bytes = params->lun_size_bytes; 2364 } else { 2365 be_lun->size_bytes = size_bytes; 2366 } 2367 2368 return (0); 2369} 2370 2371static int 2372ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2373{ 2374 struct ctl_lun_modify_params *params; 2375 struct ctl_be_block_lun *be_lun; 2376 uint64_t oldsize; 2377 int error; 2378 2379 params = &req->reqdata.modify; 2380 2381 mtx_lock(&softc->lock); 2382 be_lun = NULL; 2383 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2384 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2385 break; 2386 } 2387 mtx_unlock(&softc->lock); 2388 2389 if (be_lun == NULL) { 2390 snprintf(req->error_str, sizeof(req->error_str), 2391 "LUN %u is not managed by the block backend", 2392 params->lun_id); 2393 goto bailout_error; 2394 } 2395 2396 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2397 2398 oldsize = be_lun->size_bytes; 2399 if (be_lun->vn == NULL) 2400 error = ctl_be_block_open(softc, be_lun, req); 2401 else if (be_lun->vn->v_type == VREG) 2402 error = ctl_be_block_modify_file(be_lun, req); 2403 else 2404 error = ctl_be_block_modify_dev(be_lun, req); 2405 2406 if (error == 0 && be_lun->size_bytes != oldsize) { 2407 be_lun->size_blocks = be_lun->size_bytes >> 2408 be_lun->blocksize_shift; 2409 2410 /* 2411 * The maximum LBA is the size - 1. 2412 * 2413 * XXX: Note that this field is being updated without locking, 2414 * which might cause problems on 32-bit architectures. 2415 */ 2416 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2417 0 : (be_lun->size_blocks - 1); 2418 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2419 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2420 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2421 if (be_lun->dispatch == ctl_be_block_dispatch_zvol && 2422 be_lun->blocksize != 0) 2423 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / 2424 be_lun->blocksize; 2425 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2426 if (oldsize == 0 && be_lun->size_blocks != 0) 2427 ctl_lun_online(&be_lun->ctl_be_lun); 2428 } 2429 2430 /* Tell the user the exact size we ended up using */ 2431 params->lun_size_bytes = be_lun->size_bytes; 2432 2433 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2434 2435 return (0); 2436 2437bailout_error: 2438 req->status = CTL_LUN_ERROR; 2439 2440 return (0); 2441} 2442 2443static void 2444ctl_be_block_lun_shutdown(void *be_lun) 2445{ 2446 struct ctl_be_block_lun *lun; 2447 struct ctl_be_block_softc *softc; 2448 2449 lun = (struct ctl_be_block_lun *)be_lun; 2450 2451 softc = lun->softc; 2452 2453 mtx_lock(&softc->lock); 2454 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2455 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2456 wakeup(lun); 2457 mtx_unlock(&softc->lock); 2458 2459} 2460 2461static void 2462ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2463{ 2464 struct ctl_be_block_lun *lun; 2465 struct ctl_be_block_softc *softc; 2466 2467 lun = (struct ctl_be_block_lun *)be_lun; 2468 softc = lun->softc; 2469 2470 if (status == CTL_LUN_CONFIG_OK) { 2471 mtx_lock(&softc->lock); 2472 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2473 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2474 wakeup(lun); 2475 mtx_unlock(&softc->lock); 2476 2477 /* 2478 * We successfully added the LUN, attempt to enable it. 2479 */ 2480 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2481 printf("%s: ctl_enable_lun() failed!\n", __func__); 2482 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2483 printf("%s: ctl_invalidate_lun() failed!\n", 2484 __func__); 2485 } 2486 } 2487 2488 return; 2489 } 2490 2491 2492 mtx_lock(&softc->lock); 2493 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2494 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2495 wakeup(lun); 2496 mtx_unlock(&softc->lock); 2497} 2498 2499 2500static int 2501ctl_be_block_config_write(union ctl_io *io) 2502{ 2503 struct ctl_be_block_lun *be_lun; 2504 struct ctl_be_lun *ctl_be_lun; 2505 int retval; 2506 2507 retval = 0; 2508 2509 DPRINTF("entered\n"); 2510 2511 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2512 CTL_PRIV_BACKEND_LUN].ptr; 2513 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2514 2515 switch (io->scsiio.cdb[0]) { 2516 case SYNCHRONIZE_CACHE: 2517 case SYNCHRONIZE_CACHE_16: 2518 case WRITE_SAME_10: 2519 case WRITE_SAME_16: 2520 case UNMAP: 2521 /* 2522 * The upper level CTL code will filter out any CDBs with 2523 * the immediate bit set and return the proper error. 2524 * 2525 * We don't really need to worry about what LBA range the 2526 * user asked to be synced out. When they issue a sync 2527 * cache command, we'll sync out the whole thing. 2528 */ 2529 mtx_lock(&be_lun->queue_lock); 2530 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2531 links); 2532 mtx_unlock(&be_lun->queue_lock); 2533 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2534 break; 2535 case START_STOP_UNIT: { 2536 struct scsi_start_stop_unit *cdb; 2537 2538 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2539 2540 if (cdb->how & SSS_START) 2541 retval = ctl_start_lun(ctl_be_lun); 2542 else { 2543 retval = ctl_stop_lun(ctl_be_lun); 2544 /* 2545 * XXX KDM Copan-specific offline behavior. 2546 * Figure out a reasonable way to port this? 2547 */ 2548#ifdef NEEDTOPORT 2549 if ((retval == 0) 2550 && (cdb->byte2 & SSS_ONOFFLINE)) 2551 retval = ctl_lun_offline(ctl_be_lun); 2552#endif 2553 } 2554 2555 /* 2556 * In general, the above routines should not fail. They 2557 * just set state for the LUN. So we've got something 2558 * pretty wrong here if we can't start or stop the LUN. 2559 */ 2560 if (retval != 0) { 2561 ctl_set_internal_failure(&io->scsiio, 2562 /*sks_valid*/ 1, 2563 /*retry_count*/ 0xf051); 2564 retval = CTL_RETVAL_COMPLETE; 2565 } else { 2566 ctl_set_success(&io->scsiio); 2567 } 2568 ctl_config_write_done(io); 2569 break; 2570 } 2571 default: 2572 ctl_set_invalid_opcode(&io->scsiio); 2573 ctl_config_write_done(io); 2574 retval = CTL_RETVAL_COMPLETE; 2575 break; 2576 } 2577 2578 return (retval); 2579 2580} 2581 2582static int 2583ctl_be_block_config_read(union ctl_io *io) 2584{ 2585 return (0); 2586} 2587 2588static int 2589ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2590{ 2591 struct ctl_be_block_lun *lun; 2592 int retval; 2593 2594 lun = (struct ctl_be_block_lun *)be_lun; 2595 retval = 0; 2596 2597 retval = sbuf_printf(sb, "\t<num_threads>"); 2598 2599 if (retval != 0) 2600 goto bailout; 2601 2602 retval = sbuf_printf(sb, "%d", lun->num_threads); 2603 2604 if (retval != 0) 2605 goto bailout; 2606 2607 retval = sbuf_printf(sb, "</num_threads>\n"); 2608 2609bailout: 2610 2611 return (retval); 2612} 2613 2614static uint64_t 2615ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2616{ 2617 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2618 2619 if (lun->getattr == NULL) 2620 return (UINT64_MAX); 2621 return (lun->getattr(lun, attrname)); 2622} 2623 2624int 2625ctl_be_block_init(void) 2626{ 2627 struct ctl_be_block_softc *softc; 2628 int retval; 2629 2630 softc = &backend_block_softc; 2631 retval = 0; 2632 2633 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2634 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2635 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2636 STAILQ_INIT(&softc->disk_list); 2637 STAILQ_INIT(&softc->lun_list); 2638 2639 return (retval); 2640} 2641