ctl_backend_block.c revision 312835
1/*- 2 * Copyright (c) 2003 Silicon Graphics International Corp. 3 * Copyright (c) 2009-2011 Spectra Logic Corporation 4 * Copyright (c) 2012 The FreeBSD Foundation 5 * Copyright (c) 2014-2015 Alexander Motin <mav@FreeBSD.org> 6 * All rights reserved. 7 * 8 * Portions of this software were developed by Edward Tomasz Napierala 9 * under sponsorship from the FreeBSD Foundation. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions, and the following disclaimer, 16 * without modification. 17 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 18 * substantially similar to the "NO WARRANTY" disclaimer below 19 * ("Disclaimer") and any redistribution must be conditioned upon 20 * including a substantially similar Disclaimer requirement for further 21 * binary redistribution. 22 * 23 * NO WARRANTY 24 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 25 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 26 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 27 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 28 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 32 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 33 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 34 * POSSIBILITY OF SUCH DAMAGES. 35 * 36 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 37 */ 38/* 39 * CAM Target Layer driver backend for block devices. 40 * 41 * Author: Ken Merry <ken@FreeBSD.org> 42 */ 43#include <sys/cdefs.h> 44__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl_backend_block.c 312835 2017-01-26 20:50:01Z mav $"); 45 46#include <opt_kdtrace.h> 47 48#include <sys/param.h> 49#include <sys/systm.h> 50#include <sys/kernel.h> 51#include <sys/types.h> 52#include <sys/kthread.h> 53#include <sys/bio.h> 54#include <sys/fcntl.h> 55#include <sys/limits.h> 56#include <sys/lock.h> 57#include <sys/mutex.h> 58#include <sys/condvar.h> 59#include <sys/malloc.h> 60#include <sys/conf.h> 61#include <sys/ioccom.h> 62#include <sys/queue.h> 63#include <sys/sbuf.h> 64#include <sys/endian.h> 65#include <sys/uio.h> 66#include <sys/buf.h> 67#include <sys/taskqueue.h> 68#include <sys/vnode.h> 69#include <sys/namei.h> 70#include <sys/mount.h> 71#include <sys/disk.h> 72#include <sys/fcntl.h> 73#include <sys/filedesc.h> 74#include <sys/filio.h> 75#include <sys/proc.h> 76#include <sys/pcpu.h> 77#include <sys/module.h> 78#include <sys/sdt.h> 79#include <sys/devicestat.h> 80#include <sys/sysctl.h> 81 82#include <geom/geom.h> 83 84#include <cam/cam.h> 85#include <cam/scsi/scsi_all.h> 86#include <cam/scsi/scsi_da.h> 87#include <cam/ctl/ctl_io.h> 88#include <cam/ctl/ctl.h> 89#include <cam/ctl/ctl_backend.h> 90#include <cam/ctl/ctl_ioctl.h> 91#include <cam/ctl/ctl_ha.h> 92#include <cam/ctl/ctl_scsi_all.h> 93#include <cam/ctl/ctl_private.h> 94#include <cam/ctl/ctl_error.h> 95 96/* 97 * The idea here is that we'll allocate enough S/G space to hold a 1MB 98 * I/O. If we get an I/O larger than that, we'll split it. 99 */ 100#define CTLBLK_HALF_IO_SIZE (512 * 1024) 101#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 102#define CTLBLK_MAX_SEG MAXPHYS 103#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 104#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 105 106#ifdef CTLBLK_DEBUG 107#define DPRINTF(fmt, args...) \ 108 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 109#else 110#define DPRINTF(fmt, args...) do {} while(0) 111#endif 112 113#define PRIV(io) \ 114 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 115#define ARGS(io) \ 116 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 117 118SDT_PROVIDER_DEFINE(cbb); 119 120typedef enum { 121 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 122 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 123 CTL_BE_BLOCK_LUN_WAITING = 0x04, 124} ctl_be_block_lun_flags; 125 126typedef enum { 127 CTL_BE_BLOCK_NONE, 128 CTL_BE_BLOCK_DEV, 129 CTL_BE_BLOCK_FILE 130} ctl_be_block_type; 131 132struct ctl_be_block_filedata { 133 struct ucred *cred; 134}; 135 136union ctl_be_block_bedata { 137 struct ctl_be_block_filedata file; 138}; 139 140struct ctl_be_block_io; 141struct ctl_be_block_lun; 142 143typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 144 struct ctl_be_block_io *beio); 145typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 146 const char *attrname); 147 148/* 149 * Backend LUN structure. There is a 1:1 mapping between a block device 150 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 151 */ 152struct ctl_be_block_lun { 153 struct ctl_lun_create_params params; 154 char lunname[32]; 155 char *dev_path; 156 ctl_be_block_type dev_type; 157 struct vnode *vn; 158 union ctl_be_block_bedata backend; 159 cbb_dispatch_t dispatch; 160 cbb_dispatch_t lun_flush; 161 cbb_dispatch_t unmap; 162 cbb_dispatch_t get_lba_status; 163 cbb_getattr_t getattr; 164 uma_zone_t lun_zone; 165 uint64_t size_blocks; 166 uint64_t size_bytes; 167 struct ctl_be_block_softc *softc; 168 struct devstat *disk_stats; 169 ctl_be_block_lun_flags flags; 170 STAILQ_ENTRY(ctl_be_block_lun) links; 171 struct ctl_be_lun cbe_lun; 172 struct taskqueue *io_taskqueue; 173 struct task io_task; 174 int num_threads; 175 STAILQ_HEAD(, ctl_io_hdr) input_queue; 176 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 177 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 178 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 179 struct mtx_padalign io_lock; 180 struct mtx_padalign queue_lock; 181}; 182 183/* 184 * Overall softc structure for the block backend module. 185 */ 186struct ctl_be_block_softc { 187 struct mtx lock; 188 int num_luns; 189 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 190}; 191 192static struct ctl_be_block_softc backend_block_softc; 193 194/* 195 * Per-I/O information. 196 */ 197struct ctl_be_block_io { 198 union ctl_io *io; 199 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 200 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 201 int bio_cmd; 202 int num_segs; 203 int num_bios_sent; 204 int num_bios_done; 205 int send_complete; 206 int first_error; 207 uint64_t first_error_offset; 208 struct bintime ds_t0; 209 devstat_tag_type ds_tag_type; 210 devstat_trans_flags ds_trans_type; 211 uint64_t io_len; 212 uint64_t io_offset; 213 int io_arg; 214 struct ctl_be_block_softc *softc; 215 struct ctl_be_block_lun *lun; 216 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 217}; 218 219extern struct ctl_softc *control_softc; 220 221static int cbb_num_threads = 14; 222TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 223SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 224 "CAM Target Layer Block Backend"); 225SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 226 &cbb_num_threads, 0, "Number of threads per backing file"); 227 228static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 229static void ctl_free_beio(struct ctl_be_block_io *beio); 230static void ctl_complete_beio(struct ctl_be_block_io *beio); 231static int ctl_be_block_move_done(union ctl_io *io); 232static void ctl_be_block_biodone(struct bio *bio); 233static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 234 struct ctl_be_block_io *beio); 235static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 236 struct ctl_be_block_io *beio); 237static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 238 struct ctl_be_block_io *beio); 239static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 240 const char *attrname); 241static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 242 struct ctl_be_block_io *beio); 243static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 244 struct ctl_be_block_io *beio); 245static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 246 struct ctl_be_block_io *beio); 247static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 248 const char *attrname); 249static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 250 union ctl_io *io); 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_lun *be_lun, 265 struct ctl_lun_req *req); 266static int ctl_be_block_create(struct ctl_be_block_softc *softc, 267 struct ctl_lun_req *req); 268static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 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); 278static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 279int ctl_be_block_init(void); 280 281static struct ctl_backend_driver ctl_be_block_driver = 282{ 283 .name = "block", 284 .flags = CTL_BE_FLAG_HAS_CONFIG, 285 .init = ctl_be_block_init, 286 .data_submit = ctl_be_block_submit, 287 .data_move_done = ctl_be_block_move_done, 288 .config_read = ctl_be_block_config_read, 289 .config_write = ctl_be_block_config_write, 290 .ioctl = ctl_be_block_ioctl, 291 .lun_info = ctl_be_block_lun_info, 292 .lun_attr = ctl_be_block_lun_attr 293}; 294 295MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 296CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 297 298static uma_zone_t beio_zone; 299 300static struct ctl_be_block_io * 301ctl_alloc_beio(struct ctl_be_block_softc *softc) 302{ 303 struct ctl_be_block_io *beio; 304 305 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 306 beio->softc = softc; 307 return (beio); 308} 309 310static void 311ctl_free_beio(struct ctl_be_block_io *beio) 312{ 313 int duplicate_free; 314 int i; 315 316 duplicate_free = 0; 317 318 for (i = 0; i < beio->num_segs; i++) { 319 if (beio->sg_segs[i].addr == NULL) 320 duplicate_free++; 321 322 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 323 beio->sg_segs[i].addr = NULL; 324 325 /* For compare we had two equal S/G lists. */ 326 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 327 uma_zfree(beio->lun->lun_zone, 328 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 329 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 330 } 331 } 332 333 if (duplicate_free > 0) { 334 printf("%s: %d duplicate frees out of %d segments\n", __func__, 335 duplicate_free, beio->num_segs); 336 } 337 338 uma_zfree(beio_zone, beio); 339} 340 341static void 342ctl_complete_beio(struct ctl_be_block_io *beio) 343{ 344 union ctl_io *io = beio->io; 345 346 if (beio->beio_cont != NULL) { 347 beio->beio_cont(beio); 348 } else { 349 ctl_free_beio(beio); 350 ctl_data_submit_done(io); 351 } 352} 353 354static size_t 355cmp(uint8_t *a, uint8_t *b, size_t size) 356{ 357 size_t i; 358 359 for (i = 0; i < size; i++) { 360 if (a[i] != b[i]) 361 break; 362 } 363 return (i); 364} 365 366static void 367ctl_be_block_compare(union ctl_io *io) 368{ 369 struct ctl_be_block_io *beio; 370 uint64_t off, res; 371 int i; 372 uint8_t info[8]; 373 374 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 375 off = 0; 376 for (i = 0; i < beio->num_segs; i++) { 377 res = cmp(beio->sg_segs[i].addr, 378 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 379 beio->sg_segs[i].len); 380 off += res; 381 if (res < beio->sg_segs[i].len) 382 break; 383 } 384 if (i < beio->num_segs) { 385 scsi_u64to8b(off, info); 386 ctl_set_sense(&io->scsiio, /*current_error*/ 1, 387 /*sense_key*/ SSD_KEY_MISCOMPARE, 388 /*asc*/ 0x1D, /*ascq*/ 0x00, 389 /*type*/ SSD_ELEM_INFO, 390 /*size*/ sizeof(info), /*data*/ &info, 391 /*type*/ SSD_ELEM_NONE); 392 } else 393 ctl_set_success(&io->scsiio); 394} 395 396static int 397ctl_be_block_move_done(union ctl_io *io) 398{ 399 struct ctl_be_block_io *beio; 400 struct ctl_be_block_lun *be_lun; 401 struct ctl_lba_len_flags *lbalen; 402#ifdef CTL_TIME_IO 403 struct bintime cur_bt; 404#endif 405 406 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 407 be_lun = beio->lun; 408 409 DPRINTF("entered\n"); 410 411#ifdef CTL_TIME_IO 412 getbinuptime(&cur_bt); 413 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 414 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 415#endif 416 io->io_hdr.num_dmas++; 417 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 418 419 /* 420 * We set status at this point for read commands, and write 421 * commands with errors. 422 */ 423 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 424 ; 425 } else if ((io->io_hdr.port_status == 0) && 426 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 427 lbalen = ARGS(beio->io); 428 if (lbalen->flags & CTL_LLF_READ) { 429 ctl_set_success(&io->scsiio); 430 } else if (lbalen->flags & CTL_LLF_COMPARE) { 431 /* We have two data blocks ready for comparison. */ 432 ctl_be_block_compare(io); 433 } 434 } else if ((io->io_hdr.port_status != 0) && 435 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 436 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 437 /* 438 * For hardware error sense keys, the sense key 439 * specific value is defined to be a retry count, 440 * but we use it to pass back an internal FETD 441 * error code. XXX KDM Hopefully the FETD is only 442 * using 16 bits for an error code, since that's 443 * all the space we have in the sks field. 444 */ 445 ctl_set_internal_failure(&io->scsiio, 446 /*sks_valid*/ 1, 447 /*retry_count*/ 448 io->io_hdr.port_status); 449 } 450 451 /* 452 * If this is a read, or a write with errors, it is done. 453 */ 454 if ((beio->bio_cmd == BIO_READ) 455 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 456 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 457 ctl_complete_beio(beio); 458 return (0); 459 } 460 461 /* 462 * At this point, we have a write and the DMA completed 463 * successfully. We now have to queue it to the task queue to 464 * execute the backend I/O. That is because we do blocking 465 * memory allocations, and in the file backing case, blocking I/O. 466 * This move done routine is generally called in the SIM's 467 * interrupt context, and therefore we cannot block. 468 */ 469 mtx_lock(&be_lun->queue_lock); 470 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 471 mtx_unlock(&be_lun->queue_lock); 472 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 473 474 return (0); 475} 476 477static void 478ctl_be_block_biodone(struct bio *bio) 479{ 480 struct ctl_be_block_io *beio; 481 struct ctl_be_block_lun *be_lun; 482 union ctl_io *io; 483 int error; 484 485 beio = bio->bio_caller1; 486 be_lun = beio->lun; 487 io = beio->io; 488 489 DPRINTF("entered\n"); 490 491 error = bio->bio_error; 492 mtx_lock(&be_lun->io_lock); 493 if (error != 0 && 494 (beio->first_error == 0 || 495 bio->bio_offset < beio->first_error_offset)) { 496 beio->first_error = error; 497 beio->first_error_offset = bio->bio_offset; 498 } 499 500 beio->num_bios_done++; 501 502 /* 503 * XXX KDM will this cause WITNESS to complain? Holding a lock 504 * during the free might cause it to complain. 505 */ 506 g_destroy_bio(bio); 507 508 /* 509 * If the send complete bit isn't set, or we aren't the last I/O to 510 * complete, then we're done. 511 */ 512 if ((beio->send_complete == 0) 513 || (beio->num_bios_done < beio->num_bios_sent)) { 514 mtx_unlock(&be_lun->io_lock); 515 return; 516 } 517 518 /* 519 * At this point, we've verified that we are the last I/O to 520 * complete, so it's safe to drop the lock. 521 */ 522 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 523 beio->ds_tag_type, beio->ds_trans_type, 524 /*now*/ NULL, /*then*/&beio->ds_t0); 525 mtx_unlock(&be_lun->io_lock); 526 527 /* 528 * If there are any errors from the backing device, we fail the 529 * entire I/O with a medium error. 530 */ 531 error = beio->first_error; 532 if (error != 0) { 533 if (error == EOPNOTSUPP) { 534 ctl_set_invalid_opcode(&io->scsiio); 535 } else if (error == ENOSPC || error == EDQUOT) { 536 ctl_set_space_alloc_fail(&io->scsiio); 537 } else if (error == EROFS || error == EACCES) { 538 ctl_set_hw_write_protected(&io->scsiio); 539 } else if (beio->bio_cmd == BIO_FLUSH) { 540 /* XXX KDM is there is a better error here? */ 541 ctl_set_internal_failure(&io->scsiio, 542 /*sks_valid*/ 1, 543 /*retry_count*/ 0xbad2); 544 } else { 545 ctl_set_medium_error(&io->scsiio, 546 beio->bio_cmd == BIO_READ); 547 } 548 ctl_complete_beio(beio); 549 return; 550 } 551 552 /* 553 * If this is a write, a flush, a delete or verify, we're all done. 554 * If this is a read, we can now send the data to the user. 555 */ 556 if ((beio->bio_cmd == BIO_WRITE) 557 || (beio->bio_cmd == BIO_FLUSH) 558 || (beio->bio_cmd == BIO_DELETE) 559 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 560 ctl_set_success(&io->scsiio); 561 ctl_complete_beio(beio); 562 } else { 563 if ((ARGS(io)->flags & CTL_LLF_READ) && 564 beio->beio_cont == NULL) { 565 ctl_set_success(&io->scsiio); 566 ctl_serseq_done(io); 567 } 568#ifdef CTL_TIME_IO 569 getbinuptime(&io->io_hdr.dma_start_bt); 570#endif 571 ctl_datamove(io); 572 } 573} 574 575static void 576ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 577 struct ctl_be_block_io *beio) 578{ 579 union ctl_io *io = beio->io; 580 struct mount *mountpoint; 581 int error, lock_flags; 582 583 DPRINTF("entered\n"); 584 585 binuptime(&beio->ds_t0); 586 mtx_lock(&be_lun->io_lock); 587 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 588 mtx_unlock(&be_lun->io_lock); 589 590 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 591 592 if (MNT_SHARED_WRITES(mountpoint) || 593 ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 594 lock_flags = LK_SHARED; 595 else 596 lock_flags = LK_EXCLUSIVE; 597 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 598 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 599 curthread); 600 VOP_UNLOCK(be_lun->vn, 0); 601 602 vn_finished_write(mountpoint); 603 604 mtx_lock(&be_lun->io_lock); 605 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 606 beio->ds_tag_type, beio->ds_trans_type, 607 /*now*/ NULL, /*then*/&beio->ds_t0); 608 mtx_unlock(&be_lun->io_lock); 609 610 if (error == 0) 611 ctl_set_success(&io->scsiio); 612 else { 613 /* XXX KDM is there is a better error here? */ 614 ctl_set_internal_failure(&io->scsiio, 615 /*sks_valid*/ 1, 616 /*retry_count*/ 0xbad1); 617 } 618 619 ctl_complete_beio(beio); 620} 621 622SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t"); 623SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t"); 624SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t"); 625SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t"); 626 627static void 628ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 629 struct ctl_be_block_io *beio) 630{ 631 struct ctl_be_block_filedata *file_data; 632 union ctl_io *io; 633 struct uio xuio; 634 struct iovec *xiovec; 635 size_t s; 636 int error, flags, i; 637 638 DPRINTF("entered\n"); 639 640 file_data = &be_lun->backend.file; 641 io = beio->io; 642 flags = 0; 643 if (ARGS(io)->flags & CTL_LLF_DPO) 644 flags |= IO_DIRECT; 645 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 646 flags |= IO_SYNC; 647 648 bzero(&xuio, sizeof(xuio)); 649 if (beio->bio_cmd == BIO_READ) { 650 SDT_PROBE0(cbb, , read, file_start); 651 xuio.uio_rw = UIO_READ; 652 } else { 653 SDT_PROBE0(cbb, , write, file_start); 654 xuio.uio_rw = UIO_WRITE; 655 } 656 xuio.uio_offset = beio->io_offset; 657 xuio.uio_resid = beio->io_len; 658 xuio.uio_segflg = UIO_SYSSPACE; 659 xuio.uio_iov = beio->xiovecs; 660 xuio.uio_iovcnt = beio->num_segs; 661 xuio.uio_td = curthread; 662 663 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 664 xiovec->iov_base = beio->sg_segs[i].addr; 665 xiovec->iov_len = beio->sg_segs[i].len; 666 } 667 668 binuptime(&beio->ds_t0); 669 mtx_lock(&be_lun->io_lock); 670 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 671 mtx_unlock(&be_lun->io_lock); 672 673 if (beio->bio_cmd == BIO_READ) { 674 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 675 676 /* 677 * UFS pays attention to IO_DIRECT for reads. If the 678 * DIRECTIO option is configured into the kernel, it calls 679 * ffs_rawread(). But that only works for single-segment 680 * uios with user space addresses. In our case, with a 681 * kernel uio, it still reads into the buffer cache, but it 682 * will just try to release the buffer from the cache later 683 * on in ffs_read(). 684 * 685 * ZFS does not pay attention to IO_DIRECT for reads. 686 * 687 * UFS does not pay attention to IO_SYNC for reads. 688 * 689 * ZFS pays attention to IO_SYNC (which translates into the 690 * Solaris define FRSYNC for zfs_read()) for reads. It 691 * attempts to sync the file before reading. 692 */ 693 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 694 695 VOP_UNLOCK(be_lun->vn, 0); 696 SDT_PROBE0(cbb, , read, file_done); 697 if (error == 0 && xuio.uio_resid > 0) { 698 /* 699 * If we red less then requested (EOF), then 700 * we should clean the rest of the buffer. 701 */ 702 s = beio->io_len - xuio.uio_resid; 703 for (i = 0; i < beio->num_segs; i++) { 704 if (s >= beio->sg_segs[i].len) { 705 s -= beio->sg_segs[i].len; 706 continue; 707 } 708 bzero((uint8_t *)beio->sg_segs[i].addr + s, 709 beio->sg_segs[i].len - s); 710 s = 0; 711 } 712 } 713 } else { 714 struct mount *mountpoint; 715 int lock_flags; 716 717 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 718 719 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL) 720 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 721 lock_flags = LK_SHARED; 722 else 723 lock_flags = LK_EXCLUSIVE; 724 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 725 726 /* 727 * UFS pays attention to IO_DIRECT for writes. The write 728 * is done asynchronously. (Normally the write would just 729 * get put into cache. 730 * 731 * UFS pays attention to IO_SYNC for writes. It will 732 * attempt to write the buffer out synchronously if that 733 * flag is set. 734 * 735 * ZFS does not pay attention to IO_DIRECT for writes. 736 * 737 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 738 * for writes. It will flush the transaction from the 739 * cache before returning. 740 */ 741 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 742 VOP_UNLOCK(be_lun->vn, 0); 743 744 vn_finished_write(mountpoint); 745 SDT_PROBE0(cbb, , write, file_done); 746 } 747 748 mtx_lock(&be_lun->io_lock); 749 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 750 beio->ds_tag_type, beio->ds_trans_type, 751 /*now*/ NULL, /*then*/&beio->ds_t0); 752 mtx_unlock(&be_lun->io_lock); 753 754 /* 755 * If we got an error, set the sense data to "MEDIUM ERROR" and 756 * return the I/O to the user. 757 */ 758 if (error != 0) { 759 if (error == ENOSPC || error == EDQUOT) { 760 ctl_set_space_alloc_fail(&io->scsiio); 761 } else if (error == EROFS || error == EACCES) { 762 ctl_set_hw_write_protected(&io->scsiio); 763 } else { 764 ctl_set_medium_error(&io->scsiio, 765 beio->bio_cmd == BIO_READ); 766 } 767 ctl_complete_beio(beio); 768 return; 769 } 770 771 /* 772 * If this is a write or a verify, we're all done. 773 * If this is a read, we can now send the data to the user. 774 */ 775 if ((beio->bio_cmd == BIO_WRITE) || 776 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 777 ctl_set_success(&io->scsiio); 778 ctl_complete_beio(beio); 779 } else { 780 if ((ARGS(io)->flags & CTL_LLF_READ) && 781 beio->beio_cont == NULL) { 782 ctl_set_success(&io->scsiio); 783 ctl_serseq_done(io); 784 } 785#ifdef CTL_TIME_IO 786 getbinuptime(&io->io_hdr.dma_start_bt); 787#endif 788 ctl_datamove(io); 789 } 790} 791 792static void 793ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 794 struct ctl_be_block_io *beio) 795{ 796 union ctl_io *io = beio->io; 797 struct ctl_lba_len_flags *lbalen = ARGS(io); 798 struct scsi_get_lba_status_data *data; 799 off_t roff, off; 800 int error, status; 801 802 DPRINTF("entered\n"); 803 804 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 805 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 806 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 807 0, curthread->td_ucred, curthread); 808 if (error == 0 && off > roff) 809 status = 0; /* mapped up to off */ 810 else { 811 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 812 0, curthread->td_ucred, curthread); 813 if (error == 0 && off > roff) 814 status = 1; /* deallocated up to off */ 815 else { 816 status = 0; /* unknown up to the end */ 817 off = be_lun->size_bytes; 818 } 819 } 820 VOP_UNLOCK(be_lun->vn, 0); 821 822 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 823 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 824 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 825 lbalen->lba), data->descr[0].length); 826 data->descr[0].status = status; 827 828 ctl_complete_beio(beio); 829} 830 831static uint64_t 832ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 833{ 834 struct vattr vattr; 835 struct statfs statfs; 836 uint64_t val; 837 int error; 838 839 val = UINT64_MAX; 840 if (be_lun->vn == NULL) 841 return (val); 842 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 843 if (strcmp(attrname, "blocksused") == 0) { 844 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 845 if (error == 0) 846 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 847 } 848 if (strcmp(attrname, "blocksavail") == 0 && 849 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 850 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 851 if (error == 0) 852 val = statfs.f_bavail * statfs.f_bsize / 853 be_lun->cbe_lun.blocksize; 854 } 855 VOP_UNLOCK(be_lun->vn, 0); 856 return (val); 857} 858 859static void 860ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 861 struct ctl_be_block_io *beio) 862{ 863 union ctl_io *io; 864 struct cdevsw *csw; 865 struct cdev *dev; 866 struct uio xuio; 867 struct iovec *xiovec; 868 int error, flags, i, ref; 869 870 DPRINTF("entered\n"); 871 872 io = beio->io; 873 flags = 0; 874 if (ARGS(io)->flags & CTL_LLF_DPO) 875 flags |= IO_DIRECT; 876 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 877 flags |= IO_SYNC; 878 879 bzero(&xuio, sizeof(xuio)); 880 if (beio->bio_cmd == BIO_READ) { 881 SDT_PROBE0(cbb, , read, file_start); 882 xuio.uio_rw = UIO_READ; 883 } else { 884 SDT_PROBE0(cbb, , write, file_start); 885 xuio.uio_rw = UIO_WRITE; 886 } 887 xuio.uio_offset = beio->io_offset; 888 xuio.uio_resid = beio->io_len; 889 xuio.uio_segflg = UIO_SYSSPACE; 890 xuio.uio_iov = beio->xiovecs; 891 xuio.uio_iovcnt = beio->num_segs; 892 xuio.uio_td = curthread; 893 894 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 895 xiovec->iov_base = beio->sg_segs[i].addr; 896 xiovec->iov_len = beio->sg_segs[i].len; 897 } 898 899 binuptime(&beio->ds_t0); 900 mtx_lock(&be_lun->io_lock); 901 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 902 mtx_unlock(&be_lun->io_lock); 903 904 csw = devvn_refthread(be_lun->vn, &dev, &ref); 905 if (csw) { 906 if (beio->bio_cmd == BIO_READ) 907 error = csw->d_read(dev, &xuio, flags); 908 else 909 error = csw->d_write(dev, &xuio, flags); 910 dev_relthread(dev, ref); 911 } else 912 error = ENXIO; 913 914 if (beio->bio_cmd == BIO_READ) 915 SDT_PROBE0(cbb, , read, file_done); 916 else 917 SDT_PROBE0(cbb, , write, file_done); 918 919 mtx_lock(&be_lun->io_lock); 920 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 921 beio->ds_tag_type, beio->ds_trans_type, 922 /*now*/ NULL, /*then*/&beio->ds_t0); 923 mtx_unlock(&be_lun->io_lock); 924 925 /* 926 * If we got an error, set the sense data to "MEDIUM ERROR" and 927 * return the I/O to the user. 928 */ 929 if (error != 0) { 930 if (error == ENOSPC || error == EDQUOT) { 931 ctl_set_space_alloc_fail(&io->scsiio); 932 } else if (error == EROFS || error == EACCES) { 933 ctl_set_hw_write_protected(&io->scsiio); 934 } else { 935 ctl_set_medium_error(&io->scsiio, 936 beio->bio_cmd == BIO_READ); 937 } 938 ctl_complete_beio(beio); 939 return; 940 } 941 942 /* 943 * If this is a write or a verify, we're all done. 944 * If this is a read, we can now send the data to the user. 945 */ 946 if ((beio->bio_cmd == BIO_WRITE) || 947 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 948 ctl_set_success(&io->scsiio); 949 ctl_complete_beio(beio); 950 } else { 951 if ((ARGS(io)->flags & CTL_LLF_READ) && 952 beio->beio_cont == NULL) { 953 ctl_set_success(&io->scsiio); 954 ctl_serseq_done(io); 955 } 956#ifdef CTL_TIME_IO 957 getbinuptime(&io->io_hdr.dma_start_bt); 958#endif 959 ctl_datamove(io); 960 } 961} 962 963static void 964ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 965 struct ctl_be_block_io *beio) 966{ 967 union ctl_io *io = beio->io; 968 struct cdevsw *csw; 969 struct cdev *dev; 970 struct ctl_lba_len_flags *lbalen = ARGS(io); 971 struct scsi_get_lba_status_data *data; 972 off_t roff, off; 973 int error, ref, status; 974 975 DPRINTF("entered\n"); 976 977 csw = devvn_refthread(be_lun->vn, &dev, &ref); 978 if (csw == NULL) { 979 status = 0; /* unknown up to the end */ 980 off = be_lun->size_bytes; 981 goto done; 982 } 983 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 984 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, 985 curthread); 986 if (error == 0 && off > roff) 987 status = 0; /* mapped up to off */ 988 else { 989 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, 990 curthread); 991 if (error == 0 && off > roff) 992 status = 1; /* deallocated up to off */ 993 else { 994 status = 0; /* unknown up to the end */ 995 off = be_lun->size_bytes; 996 } 997 } 998 dev_relthread(dev, ref); 999 1000done: 1001 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 1002 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 1003 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 1004 lbalen->lba), data->descr[0].length); 1005 data->descr[0].status = status; 1006 1007 ctl_complete_beio(beio); 1008} 1009 1010static void 1011ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 1012 struct ctl_be_block_io *beio) 1013{ 1014 struct bio *bio; 1015 struct cdevsw *csw; 1016 struct cdev *dev; 1017 int ref; 1018 1019 DPRINTF("entered\n"); 1020 1021 /* This can't fail, it's a blocking allocation. */ 1022 bio = g_alloc_bio(); 1023 1024 bio->bio_cmd = BIO_FLUSH; 1025 bio->bio_offset = 0; 1026 bio->bio_data = 0; 1027 bio->bio_done = ctl_be_block_biodone; 1028 bio->bio_caller1 = beio; 1029 bio->bio_pblkno = 0; 1030 1031 /* 1032 * We don't need to acquire the LUN lock here, because we are only 1033 * sending one bio, and so there is no other context to synchronize 1034 * with. 1035 */ 1036 beio->num_bios_sent = 1; 1037 beio->send_complete = 1; 1038 1039 binuptime(&beio->ds_t0); 1040 mtx_lock(&be_lun->io_lock); 1041 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1042 mtx_unlock(&be_lun->io_lock); 1043 1044 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1045 if (csw) { 1046 bio->bio_dev = dev; 1047 csw->d_strategy(bio); 1048 dev_relthread(dev, ref); 1049 } else { 1050 bio->bio_error = ENXIO; 1051 ctl_be_block_biodone(bio); 1052 } 1053} 1054 1055static void 1056ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1057 struct ctl_be_block_io *beio, 1058 uint64_t off, uint64_t len, int last) 1059{ 1060 struct bio *bio; 1061 uint64_t maxlen; 1062 struct cdevsw *csw; 1063 struct cdev *dev; 1064 int ref; 1065 1066 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1067 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1068 while (len > 0) { 1069 bio = g_alloc_bio(); 1070 bio->bio_cmd = BIO_DELETE; 1071 bio->bio_dev = dev; 1072 bio->bio_offset = off; 1073 bio->bio_length = MIN(len, maxlen); 1074 bio->bio_data = 0; 1075 bio->bio_done = ctl_be_block_biodone; 1076 bio->bio_caller1 = beio; 1077 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1078 1079 off += bio->bio_length; 1080 len -= bio->bio_length; 1081 1082 mtx_lock(&be_lun->io_lock); 1083 beio->num_bios_sent++; 1084 if (last && len == 0) 1085 beio->send_complete = 1; 1086 mtx_unlock(&be_lun->io_lock); 1087 1088 if (csw) { 1089 csw->d_strategy(bio); 1090 } else { 1091 bio->bio_error = ENXIO; 1092 ctl_be_block_biodone(bio); 1093 } 1094 } 1095 if (csw) 1096 dev_relthread(dev, ref); 1097} 1098 1099static void 1100ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1101 struct ctl_be_block_io *beio) 1102{ 1103 union ctl_io *io; 1104 struct ctl_ptr_len_flags *ptrlen; 1105 struct scsi_unmap_desc *buf, *end; 1106 uint64_t len; 1107 1108 io = beio->io; 1109 1110 DPRINTF("entered\n"); 1111 1112 binuptime(&beio->ds_t0); 1113 mtx_lock(&be_lun->io_lock); 1114 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1115 mtx_unlock(&be_lun->io_lock); 1116 1117 if (beio->io_offset == -1) { 1118 beio->io_len = 0; 1119 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1120 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1121 end = buf + ptrlen->len / sizeof(*buf); 1122 for (; buf < end; buf++) { 1123 len = (uint64_t)scsi_4btoul(buf->length) * 1124 be_lun->cbe_lun.blocksize; 1125 beio->io_len += len; 1126 ctl_be_block_unmap_dev_range(be_lun, beio, 1127 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1128 len, (end - buf < 2) ? TRUE : FALSE); 1129 } 1130 } else 1131 ctl_be_block_unmap_dev_range(be_lun, beio, 1132 beio->io_offset, beio->io_len, TRUE); 1133} 1134 1135static void 1136ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1137 struct ctl_be_block_io *beio) 1138{ 1139 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1140 struct bio *bio; 1141 struct cdevsw *csw; 1142 struct cdev *dev; 1143 off_t cur_offset; 1144 int i, max_iosize, ref; 1145 1146 DPRINTF("entered\n"); 1147 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1148 1149 /* 1150 * We have to limit our I/O size to the maximum supported by the 1151 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1152 * set it properly, use DFLTPHYS. 1153 */ 1154 if (csw) { 1155 max_iosize = dev->si_iosize_max; 1156 if (max_iosize < PAGE_SIZE) 1157 max_iosize = DFLTPHYS; 1158 } else 1159 max_iosize = DFLTPHYS; 1160 1161 cur_offset = beio->io_offset; 1162 for (i = 0; i < beio->num_segs; i++) { 1163 size_t cur_size; 1164 uint8_t *cur_ptr; 1165 1166 cur_size = beio->sg_segs[i].len; 1167 cur_ptr = beio->sg_segs[i].addr; 1168 1169 while (cur_size > 0) { 1170 /* This can't fail, it's a blocking allocation. */ 1171 bio = g_alloc_bio(); 1172 1173 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1174 1175 bio->bio_cmd = beio->bio_cmd; 1176 bio->bio_dev = dev; 1177 bio->bio_caller1 = beio; 1178 bio->bio_length = min(cur_size, max_iosize); 1179 bio->bio_offset = cur_offset; 1180 bio->bio_data = cur_ptr; 1181 bio->bio_done = ctl_be_block_biodone; 1182 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1183 1184 cur_offset += bio->bio_length; 1185 cur_ptr += bio->bio_length; 1186 cur_size -= bio->bio_length; 1187 1188 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1189 beio->num_bios_sent++; 1190 } 1191 } 1192 binuptime(&beio->ds_t0); 1193 mtx_lock(&be_lun->io_lock); 1194 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1195 beio->send_complete = 1; 1196 mtx_unlock(&be_lun->io_lock); 1197 1198 /* 1199 * Fire off all allocated requests! 1200 */ 1201 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1202 TAILQ_REMOVE(&queue, bio, bio_queue); 1203 if (csw) 1204 csw->d_strategy(bio); 1205 else { 1206 bio->bio_error = ENXIO; 1207 ctl_be_block_biodone(bio); 1208 } 1209 } 1210 if (csw) 1211 dev_relthread(dev, ref); 1212} 1213 1214static uint64_t 1215ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1216{ 1217 struct diocgattr_arg arg; 1218 struct cdevsw *csw; 1219 struct cdev *dev; 1220 int error, ref; 1221 1222 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1223 if (csw == NULL) 1224 return (UINT64_MAX); 1225 strlcpy(arg.name, attrname, sizeof(arg.name)); 1226 arg.len = sizeof(arg.value.off); 1227 if (csw->d_ioctl) { 1228 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1229 curthread); 1230 } else 1231 error = ENODEV; 1232 dev_relthread(dev, ref); 1233 if (error != 0) 1234 return (UINT64_MAX); 1235 return (arg.value.off); 1236} 1237 1238static void 1239ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1240 union ctl_io *io) 1241{ 1242 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1243 struct ctl_be_block_io *beio; 1244 struct ctl_lba_len_flags *lbalen; 1245 1246 DPRINTF("entered\n"); 1247 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1248 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1249 1250 beio->io_len = lbalen->len * cbe_lun->blocksize; 1251 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1252 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1253 beio->bio_cmd = BIO_FLUSH; 1254 beio->ds_trans_type = DEVSTAT_NO_DATA; 1255 DPRINTF("SYNC\n"); 1256 be_lun->lun_flush(be_lun, beio); 1257} 1258 1259static void 1260ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1261{ 1262 union ctl_io *io; 1263 1264 io = beio->io; 1265 ctl_free_beio(beio); 1266 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1267 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1268 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1269 ctl_config_write_done(io); 1270 return; 1271 } 1272 1273 ctl_be_block_config_write(io); 1274} 1275 1276static void 1277ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1278 union ctl_io *io) 1279{ 1280 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1281 struct ctl_be_block_io *beio; 1282 struct ctl_lba_len_flags *lbalen; 1283 uint64_t len_left, lba; 1284 uint32_t pb, pbo, adj; 1285 int i, seglen; 1286 uint8_t *buf, *end; 1287 1288 DPRINTF("entered\n"); 1289 1290 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1291 lbalen = ARGS(beio->io); 1292 1293 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1294 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1295 ctl_free_beio(beio); 1296 ctl_set_invalid_field(&io->scsiio, 1297 /*sks_valid*/ 1, 1298 /*command*/ 1, 1299 /*field*/ 1, 1300 /*bit_valid*/ 0, 1301 /*bit*/ 0); 1302 ctl_config_write_done(io); 1303 return; 1304 } 1305 1306 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1307 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1308 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1309 beio->bio_cmd = BIO_DELETE; 1310 beio->ds_trans_type = DEVSTAT_FREE; 1311 1312 be_lun->unmap(be_lun, beio); 1313 return; 1314 } 1315 1316 beio->bio_cmd = BIO_WRITE; 1317 beio->ds_trans_type = DEVSTAT_WRITE; 1318 1319 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1320 (uintmax_t)lbalen->lba, lbalen->len); 1321 1322 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1323 if (be_lun->cbe_lun.pblockoff > 0) 1324 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1325 else 1326 pbo = 0; 1327 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1328 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1329 1330 /* 1331 * Setup the S/G entry for this chunk. 1332 */ 1333 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1334 if (pb > cbe_lun->blocksize) { 1335 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1336 seglen - pbo) % pb; 1337 if (seglen > adj) 1338 seglen -= adj; 1339 else 1340 seglen -= seglen % cbe_lun->blocksize; 1341 } else 1342 seglen -= seglen % cbe_lun->blocksize; 1343 beio->sg_segs[i].len = seglen; 1344 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1345 1346 DPRINTF("segment %d addr %p len %zd\n", i, 1347 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1348 1349 beio->num_segs++; 1350 len_left -= seglen; 1351 1352 buf = beio->sg_segs[i].addr; 1353 end = buf + seglen; 1354 for (; buf < end; buf += cbe_lun->blocksize) { 1355 if (lbalen->flags & SWS_NDOB) { 1356 memset(buf, 0, cbe_lun->blocksize); 1357 } else { 1358 memcpy(buf, io->scsiio.kern_data_ptr, 1359 cbe_lun->blocksize); 1360 } 1361 if (lbalen->flags & SWS_LBDATA) 1362 scsi_ulto4b(lbalen->lba + lba, buf); 1363 lba++; 1364 } 1365 } 1366 1367 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1368 beio->io_len = lba * cbe_lun->blocksize; 1369 1370 /* We can not do all in one run. Correct and schedule rerun. */ 1371 if (len_left > 0) { 1372 lbalen->lba += lba; 1373 lbalen->len -= lba; 1374 beio->beio_cont = ctl_be_block_cw_done_ws; 1375 } 1376 1377 be_lun->dispatch(be_lun, beio); 1378} 1379 1380static void 1381ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1382 union ctl_io *io) 1383{ 1384 struct ctl_be_block_io *beio; 1385 struct ctl_ptr_len_flags *ptrlen; 1386 1387 DPRINTF("entered\n"); 1388 1389 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1390 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1391 1392 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1393 ctl_free_beio(beio); 1394 ctl_set_invalid_field(&io->scsiio, 1395 /*sks_valid*/ 0, 1396 /*command*/ 1, 1397 /*field*/ 0, 1398 /*bit_valid*/ 0, 1399 /*bit*/ 0); 1400 ctl_config_write_done(io); 1401 return; 1402 } 1403 1404 beio->io_len = 0; 1405 beio->io_offset = -1; 1406 beio->bio_cmd = BIO_DELETE; 1407 beio->ds_trans_type = DEVSTAT_FREE; 1408 DPRINTF("UNMAP\n"); 1409 be_lun->unmap(be_lun, beio); 1410} 1411 1412static void 1413ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1414{ 1415 union ctl_io *io; 1416 1417 io = beio->io; 1418 ctl_free_beio(beio); 1419 ctl_config_read_done(io); 1420} 1421 1422static void 1423ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1424 union ctl_io *io) 1425{ 1426 struct ctl_be_block_io *beio; 1427 struct ctl_be_block_softc *softc; 1428 1429 DPRINTF("entered\n"); 1430 1431 softc = be_lun->softc; 1432 beio = ctl_alloc_beio(softc); 1433 beio->io = io; 1434 beio->lun = be_lun; 1435 beio->beio_cont = ctl_be_block_cr_done; 1436 PRIV(io)->ptr = (void *)beio; 1437 1438 switch (io->scsiio.cdb[0]) { 1439 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1440 beio->bio_cmd = -1; 1441 beio->ds_trans_type = DEVSTAT_NO_DATA; 1442 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1443 beio->io_len = 0; 1444 if (be_lun->get_lba_status) 1445 be_lun->get_lba_status(be_lun, beio); 1446 else 1447 ctl_be_block_cr_done(beio); 1448 break; 1449 default: 1450 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1451 break; 1452 } 1453} 1454 1455static void 1456ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1457{ 1458 union ctl_io *io; 1459 1460 io = beio->io; 1461 ctl_free_beio(beio); 1462 ctl_config_write_done(io); 1463} 1464 1465static void 1466ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1467 union ctl_io *io) 1468{ 1469 struct ctl_be_block_io *beio; 1470 struct ctl_be_block_softc *softc; 1471 1472 DPRINTF("entered\n"); 1473 1474 softc = be_lun->softc; 1475 beio = ctl_alloc_beio(softc); 1476 beio->io = io; 1477 beio->lun = be_lun; 1478 beio->beio_cont = ctl_be_block_cw_done; 1479 switch (io->scsiio.tag_type) { 1480 case CTL_TAG_ORDERED: 1481 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1482 break; 1483 case CTL_TAG_HEAD_OF_QUEUE: 1484 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1485 break; 1486 case CTL_TAG_UNTAGGED: 1487 case CTL_TAG_SIMPLE: 1488 case CTL_TAG_ACA: 1489 default: 1490 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1491 break; 1492 } 1493 PRIV(io)->ptr = (void *)beio; 1494 1495 switch (io->scsiio.cdb[0]) { 1496 case SYNCHRONIZE_CACHE: 1497 case SYNCHRONIZE_CACHE_16: 1498 ctl_be_block_cw_dispatch_sync(be_lun, io); 1499 break; 1500 case WRITE_SAME_10: 1501 case WRITE_SAME_16: 1502 ctl_be_block_cw_dispatch_ws(be_lun, io); 1503 break; 1504 case UNMAP: 1505 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1506 break; 1507 default: 1508 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1509 break; 1510 } 1511} 1512 1513SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t"); 1514SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t"); 1515SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t"); 1516SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t"); 1517 1518static void 1519ctl_be_block_next(struct ctl_be_block_io *beio) 1520{ 1521 struct ctl_be_block_lun *be_lun; 1522 union ctl_io *io; 1523 1524 io = beio->io; 1525 be_lun = beio->lun; 1526 ctl_free_beio(beio); 1527 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1528 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1529 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1530 ctl_data_submit_done(io); 1531 return; 1532 } 1533 1534 io->io_hdr.status &= ~CTL_STATUS_MASK; 1535 io->io_hdr.status |= CTL_STATUS_NONE; 1536 1537 mtx_lock(&be_lun->queue_lock); 1538 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1539 mtx_unlock(&be_lun->queue_lock); 1540 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1541} 1542 1543static void 1544ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1545 union ctl_io *io) 1546{ 1547 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1548 struct ctl_be_block_io *beio; 1549 struct ctl_be_block_softc *softc; 1550 struct ctl_lba_len_flags *lbalen; 1551 struct ctl_ptr_len_flags *bptrlen; 1552 uint64_t len_left, lbas; 1553 int i; 1554 1555 softc = be_lun->softc; 1556 1557 DPRINTF("entered\n"); 1558 1559 lbalen = ARGS(io); 1560 if (lbalen->flags & CTL_LLF_WRITE) { 1561 SDT_PROBE0(cbb, , write, start); 1562 } else { 1563 SDT_PROBE0(cbb, , read, start); 1564 } 1565 1566 beio = ctl_alloc_beio(softc); 1567 beio->io = io; 1568 beio->lun = be_lun; 1569 bptrlen = PRIV(io); 1570 bptrlen->ptr = (void *)beio; 1571 1572 switch (io->scsiio.tag_type) { 1573 case CTL_TAG_ORDERED: 1574 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1575 break; 1576 case CTL_TAG_HEAD_OF_QUEUE: 1577 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1578 break; 1579 case CTL_TAG_UNTAGGED: 1580 case CTL_TAG_SIMPLE: 1581 case CTL_TAG_ACA: 1582 default: 1583 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1584 break; 1585 } 1586 1587 if (lbalen->flags & CTL_LLF_WRITE) { 1588 beio->bio_cmd = BIO_WRITE; 1589 beio->ds_trans_type = DEVSTAT_WRITE; 1590 } else { 1591 beio->bio_cmd = BIO_READ; 1592 beio->ds_trans_type = DEVSTAT_READ; 1593 } 1594 1595 DPRINTF("%s at LBA %jx len %u @%ju\n", 1596 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1597 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1598 if (lbalen->flags & CTL_LLF_COMPARE) 1599 lbas = CTLBLK_HALF_IO_SIZE; 1600 else 1601 lbas = CTLBLK_MAX_IO_SIZE; 1602 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1603 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1604 beio->io_len = lbas * cbe_lun->blocksize; 1605 bptrlen->len += lbas; 1606 1607 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1608 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1609 i, CTLBLK_MAX_SEGS)); 1610 1611 /* 1612 * Setup the S/G entry for this chunk. 1613 */ 1614 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1615 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1616 1617 DPRINTF("segment %d addr %p len %zd\n", i, 1618 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1619 1620 /* Set up second segment for compare operation. */ 1621 if (lbalen->flags & CTL_LLF_COMPARE) { 1622 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1623 beio->sg_segs[i].len; 1624 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1625 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1626 } 1627 1628 beio->num_segs++; 1629 len_left -= beio->sg_segs[i].len; 1630 } 1631 if (bptrlen->len < lbalen->len) 1632 beio->beio_cont = ctl_be_block_next; 1633 io->scsiio.be_move_done = ctl_be_block_move_done; 1634 /* For compare we have separate S/G lists for read and datamove. */ 1635 if (lbalen->flags & CTL_LLF_COMPARE) 1636 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1637 else 1638 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1639 io->scsiio.kern_data_len = beio->io_len; 1640 io->scsiio.kern_data_resid = 0; 1641 io->scsiio.kern_sg_entries = beio->num_segs; 1642 io->io_hdr.flags |= CTL_FLAG_ALLOCATED; 1643 1644 /* 1645 * For the read case, we need to read the data into our buffers and 1646 * then we can send it back to the user. For the write case, we 1647 * need to get the data from the user first. 1648 */ 1649 if (beio->bio_cmd == BIO_READ) { 1650 SDT_PROBE0(cbb, , read, alloc_done); 1651 be_lun->dispatch(be_lun, beio); 1652 } else { 1653 SDT_PROBE0(cbb, , write, alloc_done); 1654#ifdef CTL_TIME_IO 1655 getbinuptime(&io->io_hdr.dma_start_bt); 1656#endif 1657 ctl_datamove(io); 1658 } 1659} 1660 1661static void 1662ctl_be_block_worker(void *context, int pending) 1663{ 1664 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1665 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1666 union ctl_io *io; 1667 struct ctl_be_block_io *beio; 1668 1669 DPRINTF("entered\n"); 1670 /* 1671 * Fetch and process I/Os from all queues. If we detect LUN 1672 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race, 1673 * so make response maximally opaque to not confuse initiator. 1674 */ 1675 for (;;) { 1676 mtx_lock(&be_lun->queue_lock); 1677 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1678 if (io != NULL) { 1679 DPRINTF("datamove queue\n"); 1680 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1681 ctl_io_hdr, links); 1682 mtx_unlock(&be_lun->queue_lock); 1683 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1684 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1685 ctl_set_busy(&io->scsiio); 1686 ctl_complete_beio(beio); 1687 return; 1688 } 1689 be_lun->dispatch(be_lun, beio); 1690 continue; 1691 } 1692 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1693 if (io != NULL) { 1694 DPRINTF("config write queue\n"); 1695 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1696 ctl_io_hdr, links); 1697 mtx_unlock(&be_lun->queue_lock); 1698 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1699 ctl_set_busy(&io->scsiio); 1700 ctl_config_write_done(io); 1701 return; 1702 } 1703 ctl_be_block_cw_dispatch(be_lun, io); 1704 continue; 1705 } 1706 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1707 if (io != NULL) { 1708 DPRINTF("config read queue\n"); 1709 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1710 ctl_io_hdr, links); 1711 mtx_unlock(&be_lun->queue_lock); 1712 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1713 ctl_set_busy(&io->scsiio); 1714 ctl_config_read_done(io); 1715 return; 1716 } 1717 ctl_be_block_cr_dispatch(be_lun, io); 1718 continue; 1719 } 1720 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1721 if (io != NULL) { 1722 DPRINTF("input queue\n"); 1723 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1724 ctl_io_hdr, links); 1725 mtx_unlock(&be_lun->queue_lock); 1726 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1727 ctl_set_busy(&io->scsiio); 1728 ctl_data_submit_done(io); 1729 return; 1730 } 1731 ctl_be_block_dispatch(be_lun, io); 1732 continue; 1733 } 1734 1735 /* 1736 * If we get here, there is no work left in the queues, so 1737 * just break out and let the task queue go to sleep. 1738 */ 1739 mtx_unlock(&be_lun->queue_lock); 1740 break; 1741 } 1742} 1743 1744/* 1745 * Entry point from CTL to the backend for I/O. We queue everything to a 1746 * work thread, so this just puts the I/O on a queue and wakes up the 1747 * thread. 1748 */ 1749static int 1750ctl_be_block_submit(union ctl_io *io) 1751{ 1752 struct ctl_be_block_lun *be_lun; 1753 struct ctl_be_lun *cbe_lun; 1754 1755 DPRINTF("entered\n"); 1756 1757 cbe_lun = CTL_BACKEND_LUN(io); 1758 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1759 1760 /* 1761 * Make sure we only get SCSI I/O. 1762 */ 1763 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1764 "%#x) encountered", io->io_hdr.io_type)); 1765 1766 PRIV(io)->len = 0; 1767 1768 mtx_lock(&be_lun->queue_lock); 1769 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1770 mtx_unlock(&be_lun->queue_lock); 1771 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1772 1773 return (CTL_RETVAL_COMPLETE); 1774} 1775 1776static int 1777ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1778 int flag, struct thread *td) 1779{ 1780 struct ctl_be_block_softc *softc; 1781 int error; 1782 1783 softc = &backend_block_softc; 1784 1785 error = 0; 1786 1787 switch (cmd) { 1788 case CTL_LUN_REQ: { 1789 struct ctl_lun_req *lun_req; 1790 1791 lun_req = (struct ctl_lun_req *)addr; 1792 1793 switch (lun_req->reqtype) { 1794 case CTL_LUNREQ_CREATE: 1795 error = ctl_be_block_create(softc, lun_req); 1796 break; 1797 case CTL_LUNREQ_RM: 1798 error = ctl_be_block_rm(softc, lun_req); 1799 break; 1800 case CTL_LUNREQ_MODIFY: 1801 error = ctl_be_block_modify(softc, lun_req); 1802 break; 1803 default: 1804 lun_req->status = CTL_LUN_ERROR; 1805 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1806 "invalid LUN request type %d", 1807 lun_req->reqtype); 1808 break; 1809 } 1810 break; 1811 } 1812 default: 1813 error = ENOTTY; 1814 break; 1815 } 1816 1817 return (error); 1818} 1819 1820static int 1821ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1822{ 1823 struct ctl_be_lun *cbe_lun; 1824 struct ctl_be_block_filedata *file_data; 1825 struct ctl_lun_create_params *params; 1826 char *value; 1827 struct vattr vattr; 1828 off_t ps, pss, po, pos, us, uss, uo, uos; 1829 int error; 1830 1831 cbe_lun = &be_lun->cbe_lun; 1832 file_data = &be_lun->backend.file; 1833 params = &be_lun->params; 1834 1835 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1836 be_lun->dispatch = ctl_be_block_dispatch_file; 1837 be_lun->lun_flush = ctl_be_block_flush_file; 1838 be_lun->get_lba_status = ctl_be_block_gls_file; 1839 be_lun->getattr = ctl_be_block_getattr_file; 1840 be_lun->unmap = NULL; 1841 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1842 1843 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1844 if (error != 0) { 1845 snprintf(req->error_str, sizeof(req->error_str), 1846 "error calling VOP_GETATTR() for file %s", 1847 be_lun->dev_path); 1848 return (error); 1849 } 1850 1851 file_data->cred = crhold(curthread->td_ucred); 1852 if (params->lun_size_bytes != 0) 1853 be_lun->size_bytes = params->lun_size_bytes; 1854 else 1855 be_lun->size_bytes = vattr.va_size; 1856 1857 /* 1858 * For files we can use any logical block size. Prefer 512 bytes 1859 * for compatibility reasons. If file's vattr.va_blocksize 1860 * (preferred I/O block size) is bigger and multiple to chosen 1861 * logical block size -- report it as physical block size. 1862 */ 1863 if (params->blocksize_bytes != 0) 1864 cbe_lun->blocksize = params->blocksize_bytes; 1865 else if (cbe_lun->lun_type == T_CDROM) 1866 cbe_lun->blocksize = 2048; 1867 else 1868 cbe_lun->blocksize = 512; 1869 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1870 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1871 0 : (be_lun->size_blocks - 1); 1872 1873 us = ps = vattr.va_blocksize; 1874 uo = po = 0; 1875 1876 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1877 if (value != NULL) 1878 ctl_expand_number(value, &ps); 1879 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1880 if (value != NULL) 1881 ctl_expand_number(value, &po); 1882 pss = ps / cbe_lun->blocksize; 1883 pos = po / cbe_lun->blocksize; 1884 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1885 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1886 cbe_lun->pblockexp = fls(pss) - 1; 1887 cbe_lun->pblockoff = (pss - pos) % pss; 1888 } 1889 1890 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1891 if (value != NULL) 1892 ctl_expand_number(value, &us); 1893 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 1894 if (value != NULL) 1895 ctl_expand_number(value, &uo); 1896 uss = us / cbe_lun->blocksize; 1897 uos = uo / cbe_lun->blocksize; 1898 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1899 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1900 cbe_lun->ublockexp = fls(uss) - 1; 1901 cbe_lun->ublockoff = (uss - uos) % uss; 1902 } 1903 1904 /* 1905 * Sanity check. The media size has to be at least one 1906 * sector long. 1907 */ 1908 if (be_lun->size_bytes < cbe_lun->blocksize) { 1909 error = EINVAL; 1910 snprintf(req->error_str, sizeof(req->error_str), 1911 "file %s size %ju < block size %u", be_lun->dev_path, 1912 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1913 } 1914 1915 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1916 return (error); 1917} 1918 1919static int 1920ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1921{ 1922 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1923 struct ctl_lun_create_params *params; 1924 struct cdevsw *csw; 1925 struct cdev *dev; 1926 char *value; 1927 int error, atomic, maxio, ref, unmap, tmp; 1928 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1929 1930 params = &be_lun->params; 1931 1932 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1933 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1934 if (csw == NULL) 1935 return (ENXIO); 1936 if (strcmp(csw->d_name, "zvol") == 0) { 1937 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1938 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1939 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1940 } else { 1941 be_lun->dispatch = ctl_be_block_dispatch_dev; 1942 be_lun->get_lba_status = NULL; 1943 atomic = 0; 1944 maxio = dev->si_iosize_max; 1945 if (maxio <= 0) 1946 maxio = DFLTPHYS; 1947 if (maxio > CTLBLK_MAX_IO_SIZE) 1948 maxio = CTLBLK_MAX_IO_SIZE; 1949 } 1950 be_lun->lun_flush = ctl_be_block_flush_dev; 1951 be_lun->getattr = ctl_be_block_getattr_dev; 1952 be_lun->unmap = ctl_be_block_unmap_dev; 1953 1954 if (!csw->d_ioctl) { 1955 dev_relthread(dev, ref); 1956 snprintf(req->error_str, sizeof(req->error_str), 1957 "no d_ioctl for device %s!", be_lun->dev_path); 1958 return (ENODEV); 1959 } 1960 1961 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1962 curthread); 1963 if (error) { 1964 dev_relthread(dev, ref); 1965 snprintf(req->error_str, sizeof(req->error_str), 1966 "error %d returned for DIOCGSECTORSIZE ioctl " 1967 "on %s!", error, be_lun->dev_path); 1968 return (error); 1969 } 1970 1971 /* 1972 * If the user has asked for a blocksize that is greater than the 1973 * backing device's blocksize, we can do it only if the blocksize 1974 * the user is asking for is an even multiple of the underlying 1975 * device's blocksize. 1976 */ 1977 if ((params->blocksize_bytes != 0) && 1978 (params->blocksize_bytes >= tmp)) { 1979 if (params->blocksize_bytes % tmp == 0) { 1980 cbe_lun->blocksize = params->blocksize_bytes; 1981 } else { 1982 dev_relthread(dev, ref); 1983 snprintf(req->error_str, sizeof(req->error_str), 1984 "requested blocksize %u is not an even " 1985 "multiple of backing device blocksize %u", 1986 params->blocksize_bytes, tmp); 1987 return (EINVAL); 1988 } 1989 } else if (params->blocksize_bytes != 0) { 1990 dev_relthread(dev, ref); 1991 snprintf(req->error_str, sizeof(req->error_str), 1992 "requested blocksize %u < backing device " 1993 "blocksize %u", params->blocksize_bytes, tmp); 1994 return (EINVAL); 1995 } else if (cbe_lun->lun_type == T_CDROM) 1996 cbe_lun->blocksize = MAX(tmp, 2048); 1997 else 1998 cbe_lun->blocksize = tmp; 1999 2000 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 2001 curthread); 2002 if (error) { 2003 dev_relthread(dev, ref); 2004 snprintf(req->error_str, sizeof(req->error_str), 2005 "error %d returned for DIOCGMEDIASIZE " 2006 " ioctl on %s!", error, 2007 be_lun->dev_path); 2008 return (error); 2009 } 2010 2011 if (params->lun_size_bytes != 0) { 2012 if (params->lun_size_bytes > otmp) { 2013 dev_relthread(dev, ref); 2014 snprintf(req->error_str, sizeof(req->error_str), 2015 "requested LUN size %ju > backing device " 2016 "size %ju", 2017 (uintmax_t)params->lun_size_bytes, 2018 (uintmax_t)otmp); 2019 return (EINVAL); 2020 } 2021 2022 be_lun->size_bytes = params->lun_size_bytes; 2023 } else 2024 be_lun->size_bytes = otmp; 2025 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2026 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2027 0 : (be_lun->size_blocks - 1); 2028 2029 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2030 curthread); 2031 if (error) 2032 ps = po = 0; 2033 else { 2034 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2035 FREAD, curthread); 2036 if (error) 2037 po = 0; 2038 } 2039 us = ps; 2040 uo = po; 2041 2042 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 2043 if (value != NULL) 2044 ctl_expand_number(value, &ps); 2045 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 2046 if (value != NULL) 2047 ctl_expand_number(value, &po); 2048 pss = ps / cbe_lun->blocksize; 2049 pos = po / cbe_lun->blocksize; 2050 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2051 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2052 cbe_lun->pblockexp = fls(pss) - 1; 2053 cbe_lun->pblockoff = (pss - pos) % pss; 2054 } 2055 2056 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 2057 if (value != NULL) 2058 ctl_expand_number(value, &us); 2059 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 2060 if (value != NULL) 2061 ctl_expand_number(value, &uo); 2062 uss = us / cbe_lun->blocksize; 2063 uos = uo / cbe_lun->blocksize; 2064 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2065 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2066 cbe_lun->ublockexp = fls(uss) - 1; 2067 cbe_lun->ublockoff = (uss - uos) % uss; 2068 } 2069 2070 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2071 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2072 2073 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2074 unmap = 1; 2075 } else { 2076 struct diocgattr_arg arg; 2077 2078 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2079 arg.len = sizeof(arg.value.i); 2080 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2081 curthread); 2082 unmap = (error == 0) ? arg.value.i : 0; 2083 } 2084 value = ctl_get_opt(&cbe_lun->options, "unmap"); 2085 if (value != NULL) 2086 unmap = (strcmp(value, "on") == 0); 2087 if (unmap) 2088 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2089 else 2090 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2091 2092 dev_relthread(dev, ref); 2093 return (0); 2094} 2095 2096static int 2097ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2098{ 2099 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2100 int flags; 2101 2102 if (be_lun->vn) { 2103 flags = FREAD; 2104 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2105 flags |= FWRITE; 2106 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2107 be_lun->vn = NULL; 2108 2109 switch (be_lun->dev_type) { 2110 case CTL_BE_BLOCK_DEV: 2111 break; 2112 case CTL_BE_BLOCK_FILE: 2113 if (be_lun->backend.file.cred != NULL) { 2114 crfree(be_lun->backend.file.cred); 2115 be_lun->backend.file.cred = NULL; 2116 } 2117 break; 2118 case CTL_BE_BLOCK_NONE: 2119 break; 2120 default: 2121 panic("Unexpected backend type %d", be_lun->dev_type); 2122 break; 2123 } 2124 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2125 } 2126 return (0); 2127} 2128 2129static int 2130ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2131{ 2132 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2133 struct nameidata nd; 2134 char *value; 2135 int error, flags; 2136 2137 error = 0; 2138 if (rootvnode == NULL) { 2139 snprintf(req->error_str, sizeof(req->error_str), 2140 "Root filesystem is not mounted"); 2141 return (1); 2142 } 2143 if (!curthread->td_proc->p_fd->fd_cdir) { 2144 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2145 VREF(rootvnode); 2146 } 2147 if (!curthread->td_proc->p_fd->fd_rdir) { 2148 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2149 VREF(rootvnode); 2150 } 2151 if (!curthread->td_proc->p_fd->fd_jdir) { 2152 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2153 VREF(rootvnode); 2154 } 2155 2156 value = ctl_get_opt(&cbe_lun->options, "file"); 2157 if (value == NULL) { 2158 snprintf(req->error_str, sizeof(req->error_str), 2159 "no file argument specified"); 2160 return (1); 2161 } 2162 free(be_lun->dev_path, M_CTLBLK); 2163 be_lun->dev_path = strdup(value, M_CTLBLK); 2164 2165 flags = FREAD; 2166 value = ctl_get_opt(&cbe_lun->options, "readonly"); 2167 if (value != NULL) { 2168 if (strcmp(value, "on") != 0) 2169 flags |= FWRITE; 2170 } else if (cbe_lun->lun_type == T_DIRECT) 2171 flags |= FWRITE; 2172 2173again: 2174 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2175 error = vn_open(&nd, &flags, 0, NULL); 2176 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2177 flags &= ~FWRITE; 2178 goto again; 2179 } 2180 if (error) { 2181 /* 2182 * This is the only reasonable guess we can make as far as 2183 * path if the user doesn't give us a fully qualified path. 2184 * If they want to specify a file, they need to specify the 2185 * full path. 2186 */ 2187 if (be_lun->dev_path[0] != '/') { 2188 char *dev_name; 2189 2190 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2191 be_lun->dev_path); 2192 free(be_lun->dev_path, M_CTLBLK); 2193 be_lun->dev_path = dev_name; 2194 goto again; 2195 } 2196 snprintf(req->error_str, sizeof(req->error_str), 2197 "error opening %s: %d", be_lun->dev_path, error); 2198 return (error); 2199 } 2200 if (flags & FWRITE) 2201 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2202 else 2203 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2204 2205 NDFREE(&nd, NDF_ONLY_PNBUF); 2206 be_lun->vn = nd.ni_vp; 2207 2208 /* We only support disks and files. */ 2209 if (vn_isdisk(be_lun->vn, &error)) { 2210 error = ctl_be_block_open_dev(be_lun, req); 2211 } else if (be_lun->vn->v_type == VREG) { 2212 error = ctl_be_block_open_file(be_lun, req); 2213 } else { 2214 error = EINVAL; 2215 snprintf(req->error_str, sizeof(req->error_str), 2216 "%s is not a disk or plain file", be_lun->dev_path); 2217 } 2218 VOP_UNLOCK(be_lun->vn, 0); 2219 2220 if (error != 0) 2221 ctl_be_block_close(be_lun); 2222 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2223 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2224 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2225 value = ctl_get_opt(&cbe_lun->options, "serseq"); 2226 if (value != NULL && strcmp(value, "on") == 0) 2227 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2228 else if (value != NULL && strcmp(value, "read") == 0) 2229 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2230 else if (value != NULL && strcmp(value, "off") == 0) 2231 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2232 return (0); 2233} 2234 2235static int 2236ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2237{ 2238 struct ctl_be_lun *cbe_lun; 2239 struct ctl_be_block_lun *be_lun; 2240 struct ctl_lun_create_params *params; 2241 char num_thread_str[16]; 2242 char tmpstr[32]; 2243 char *value; 2244 int retval, num_threads; 2245 int tmp_num_threads; 2246 2247 params = &req->reqdata.create; 2248 retval = 0; 2249 req->status = CTL_LUN_OK; 2250 2251 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2252 cbe_lun = &be_lun->cbe_lun; 2253 cbe_lun->be_lun = be_lun; 2254 be_lun->params = req->reqdata.create; 2255 be_lun->softc = softc; 2256 STAILQ_INIT(&be_lun->input_queue); 2257 STAILQ_INIT(&be_lun->config_read_queue); 2258 STAILQ_INIT(&be_lun->config_write_queue); 2259 STAILQ_INIT(&be_lun->datamove_queue); 2260 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2261 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2262 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2263 ctl_init_opts(&cbe_lun->options, 2264 req->num_be_args, req->kern_be_args); 2265 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2266 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2267 if (be_lun->lun_zone == NULL) { 2268 snprintf(req->error_str, sizeof(req->error_str), 2269 "error allocating UMA zone"); 2270 goto bailout_error; 2271 } 2272 2273 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2274 cbe_lun->lun_type = params->device_type; 2275 else 2276 cbe_lun->lun_type = T_DIRECT; 2277 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2278 cbe_lun->flags = 0; 2279 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2280 if (value != NULL) { 2281 if (strcmp(value, "primary") == 0) 2282 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2283 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2284 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2285 2286 if (cbe_lun->lun_type == T_DIRECT || 2287 cbe_lun->lun_type == T_CDROM) { 2288 be_lun->size_bytes = params->lun_size_bytes; 2289 if (params->blocksize_bytes != 0) 2290 cbe_lun->blocksize = params->blocksize_bytes; 2291 else if (cbe_lun->lun_type == T_CDROM) 2292 cbe_lun->blocksize = 2048; 2293 else 2294 cbe_lun->blocksize = 512; 2295 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2296 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2297 0 : (be_lun->size_blocks - 1); 2298 2299 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2300 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2301 retval = ctl_be_block_open(be_lun, req); 2302 if (retval != 0) { 2303 retval = 0; 2304 req->status = CTL_LUN_WARNING; 2305 } 2306 } 2307 num_threads = cbb_num_threads; 2308 } else { 2309 num_threads = 1; 2310 } 2311 2312 value = ctl_get_opt(&cbe_lun->options, "num_threads"); 2313 if (value != NULL) { 2314 tmp_num_threads = strtol(value, NULL, 0); 2315 2316 /* 2317 * We don't let the user specify less than one 2318 * thread, but hope he's clueful enough not to 2319 * specify 1000 threads. 2320 */ 2321 if (tmp_num_threads < 1) { 2322 snprintf(req->error_str, sizeof(req->error_str), 2323 "invalid number of threads %s", 2324 num_thread_str); 2325 goto bailout_error; 2326 } 2327 num_threads = tmp_num_threads; 2328 } 2329 2330 if (be_lun->vn == NULL) 2331 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2332 /* Tell the user the blocksize we ended up using */ 2333 params->lun_size_bytes = be_lun->size_bytes; 2334 params->blocksize_bytes = cbe_lun->blocksize; 2335 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2336 cbe_lun->req_lun_id = params->req_lun_id; 2337 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2338 } else 2339 cbe_lun->req_lun_id = 0; 2340 2341 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2342 cbe_lun->lun_config_status = ctl_be_block_lun_config_status; 2343 cbe_lun->be = &ctl_be_block_driver; 2344 2345 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2346 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2347 softc->num_luns); 2348 strncpy((char *)cbe_lun->serial_num, tmpstr, 2349 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2350 2351 /* Tell the user what we used for a serial number */ 2352 strncpy((char *)params->serial_num, tmpstr, 2353 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2354 } else { 2355 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2356 MIN(sizeof(cbe_lun->serial_num), 2357 sizeof(params->serial_num))); 2358 } 2359 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2360 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2361 strncpy((char *)cbe_lun->device_id, tmpstr, 2362 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2363 2364 /* Tell the user what we used for a device ID */ 2365 strncpy((char *)params->device_id, tmpstr, 2366 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2367 } else { 2368 strncpy((char *)cbe_lun->device_id, params->device_id, 2369 MIN(sizeof(cbe_lun->device_id), 2370 sizeof(params->device_id))); 2371 } 2372 2373 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2374 2375 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2376 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2377 2378 if (be_lun->io_taskqueue == NULL) { 2379 snprintf(req->error_str, sizeof(req->error_str), 2380 "unable to create taskqueue"); 2381 goto bailout_error; 2382 } 2383 2384 /* 2385 * Note that we start the same number of threads by default for 2386 * both the file case and the block device case. For the file 2387 * case, we need multiple threads to allow concurrency, because the 2388 * vnode interface is designed to be a blocking interface. For the 2389 * block device case, ZFS zvols at least will block the caller's 2390 * context in many instances, and so we need multiple threads to 2391 * overcome that problem. Other block devices don't need as many 2392 * threads, but they shouldn't cause too many problems. 2393 * 2394 * If the user wants to just have a single thread for a block 2395 * device, he can specify that when the LUN is created, or change 2396 * the tunable/sysctl to alter the default number of threads. 2397 */ 2398 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2399 /*num threads*/num_threads, 2400 /*priority*/PWAIT, 2401 /*thread name*/ 2402 "%s taskq", be_lun->lunname); 2403 2404 if (retval != 0) 2405 goto bailout_error; 2406 2407 be_lun->num_threads = num_threads; 2408 2409 mtx_lock(&softc->lock); 2410 softc->num_luns++; 2411 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2412 2413 mtx_unlock(&softc->lock); 2414 2415 retval = ctl_add_lun(&be_lun->cbe_lun); 2416 if (retval != 0) { 2417 mtx_lock(&softc->lock); 2418 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2419 links); 2420 softc->num_luns--; 2421 mtx_unlock(&softc->lock); 2422 snprintf(req->error_str, sizeof(req->error_str), 2423 "ctl_add_lun() returned error %d, see dmesg for " 2424 "details", retval); 2425 retval = 0; 2426 goto bailout_error; 2427 } 2428 2429 mtx_lock(&softc->lock); 2430 2431 /* 2432 * Tell the config_status routine that we're waiting so it won't 2433 * clean up the LUN in the event of an error. 2434 */ 2435 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2436 2437 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2438 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2439 if (retval == EINTR) 2440 break; 2441 } 2442 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2443 2444 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2445 snprintf(req->error_str, sizeof(req->error_str), 2446 "LUN configuration error, see dmesg for details"); 2447 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2448 links); 2449 softc->num_luns--; 2450 mtx_unlock(&softc->lock); 2451 goto bailout_error; 2452 } else { 2453 params->req_lun_id = cbe_lun->lun_id; 2454 } 2455 2456 mtx_unlock(&softc->lock); 2457 2458 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2459 cbe_lun->blocksize, 2460 DEVSTAT_ALL_SUPPORTED, 2461 cbe_lun->lun_type 2462 | DEVSTAT_TYPE_IF_OTHER, 2463 DEVSTAT_PRIORITY_OTHER); 2464 2465 return (retval); 2466 2467bailout_error: 2468 req->status = CTL_LUN_ERROR; 2469 2470 if (be_lun->io_taskqueue != NULL) 2471 taskqueue_free(be_lun->io_taskqueue); 2472 ctl_be_block_close(be_lun); 2473 if (be_lun->dev_path != NULL) 2474 free(be_lun->dev_path, M_CTLBLK); 2475 if (be_lun->lun_zone != NULL) 2476 uma_zdestroy(be_lun->lun_zone); 2477 ctl_free_opts(&cbe_lun->options); 2478 mtx_destroy(&be_lun->queue_lock); 2479 mtx_destroy(&be_lun->io_lock); 2480 free(be_lun, M_CTLBLK); 2481 2482 return (retval); 2483} 2484 2485static int 2486ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2487{ 2488 struct ctl_lun_rm_params *params; 2489 struct ctl_be_block_lun *be_lun; 2490 struct ctl_be_lun *cbe_lun; 2491 int retval; 2492 2493 params = &req->reqdata.rm; 2494 2495 mtx_lock(&softc->lock); 2496 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2497 if (be_lun->cbe_lun.lun_id == params->lun_id) 2498 break; 2499 } 2500 mtx_unlock(&softc->lock); 2501 if (be_lun == NULL) { 2502 snprintf(req->error_str, sizeof(req->error_str), 2503 "LUN %u is not managed by the block backend", 2504 params->lun_id); 2505 goto bailout_error; 2506 } 2507 cbe_lun = &be_lun->cbe_lun; 2508 2509 retval = ctl_disable_lun(cbe_lun); 2510 if (retval != 0) { 2511 snprintf(req->error_str, sizeof(req->error_str), 2512 "error %d returned from ctl_disable_lun() for " 2513 "LUN %d", retval, params->lun_id); 2514 goto bailout_error; 2515 } 2516 2517 if (be_lun->vn != NULL) { 2518 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2519 ctl_lun_no_media(cbe_lun); 2520 taskqueue_drain_all(be_lun->io_taskqueue); 2521 ctl_be_block_close(be_lun); 2522 } 2523 2524 retval = ctl_invalidate_lun(cbe_lun); 2525 if (retval != 0) { 2526 snprintf(req->error_str, sizeof(req->error_str), 2527 "error %d returned from ctl_invalidate_lun() for " 2528 "LUN %d", retval, params->lun_id); 2529 goto bailout_error; 2530 } 2531 2532 mtx_lock(&softc->lock); 2533 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2534 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2535 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2536 if (retval == EINTR) 2537 break; 2538 } 2539 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2540 2541 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2542 snprintf(req->error_str, sizeof(req->error_str), 2543 "interrupted waiting for LUN to be freed"); 2544 mtx_unlock(&softc->lock); 2545 goto bailout_error; 2546 } 2547 2548 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2549 2550 softc->num_luns--; 2551 mtx_unlock(&softc->lock); 2552 2553 taskqueue_drain_all(be_lun->io_taskqueue); 2554 taskqueue_free(be_lun->io_taskqueue); 2555 2556 if (be_lun->disk_stats != NULL) 2557 devstat_remove_entry(be_lun->disk_stats); 2558 2559 uma_zdestroy(be_lun->lun_zone); 2560 2561 ctl_free_opts(&cbe_lun->options); 2562 free(be_lun->dev_path, M_CTLBLK); 2563 mtx_destroy(&be_lun->queue_lock); 2564 mtx_destroy(&be_lun->io_lock); 2565 free(be_lun, M_CTLBLK); 2566 2567 req->status = CTL_LUN_OK; 2568 return (0); 2569 2570bailout_error: 2571 req->status = CTL_LUN_ERROR; 2572 return (0); 2573} 2574 2575static int 2576ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2577{ 2578 struct ctl_lun_modify_params *params; 2579 struct ctl_be_block_lun *be_lun; 2580 struct ctl_be_lun *cbe_lun; 2581 char *value; 2582 uint64_t oldsize; 2583 int error, wasprim; 2584 2585 params = &req->reqdata.modify; 2586 2587 mtx_lock(&softc->lock); 2588 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2589 if (be_lun->cbe_lun.lun_id == params->lun_id) 2590 break; 2591 } 2592 mtx_unlock(&softc->lock); 2593 if (be_lun == NULL) { 2594 snprintf(req->error_str, sizeof(req->error_str), 2595 "LUN %u is not managed by the block backend", 2596 params->lun_id); 2597 goto bailout_error; 2598 } 2599 cbe_lun = &be_lun->cbe_lun; 2600 2601 if (params->lun_size_bytes != 0) 2602 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2603 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); 2604 2605 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2606 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2607 if (value != NULL) { 2608 if (strcmp(value, "primary") == 0) 2609 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2610 else 2611 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2612 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2613 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2614 else 2615 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2616 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2617 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2618 ctl_lun_primary(cbe_lun); 2619 else 2620 ctl_lun_secondary(cbe_lun); 2621 } 2622 2623 oldsize = be_lun->size_blocks; 2624 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2625 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2626 if (be_lun->vn == NULL) 2627 error = ctl_be_block_open(be_lun, req); 2628 else if (vn_isdisk(be_lun->vn, &error)) 2629 error = ctl_be_block_open_dev(be_lun, req); 2630 else if (be_lun->vn->v_type == VREG) { 2631 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2632 error = ctl_be_block_open_file(be_lun, req); 2633 VOP_UNLOCK(be_lun->vn, 0); 2634 } else 2635 error = EINVAL; 2636 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) && 2637 be_lun->vn != NULL) { 2638 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2639 ctl_lun_has_media(cbe_lun); 2640 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 && 2641 be_lun->vn == NULL) { 2642 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2643 ctl_lun_no_media(cbe_lun); 2644 } 2645 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2646 } else { 2647 if (be_lun->vn != NULL) { 2648 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2649 ctl_lun_no_media(cbe_lun); 2650 taskqueue_drain_all(be_lun->io_taskqueue); 2651 error = ctl_be_block_close(be_lun); 2652 } else 2653 error = 0; 2654 } 2655 if (be_lun->size_blocks != oldsize) 2656 ctl_lun_capacity_changed(cbe_lun); 2657 2658 /* Tell the user the exact size we ended up using */ 2659 params->lun_size_bytes = be_lun->size_bytes; 2660 2661 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2662 return (0); 2663 2664bailout_error: 2665 req->status = CTL_LUN_ERROR; 2666 return (0); 2667} 2668 2669static void 2670ctl_be_block_lun_shutdown(void *be_lun) 2671{ 2672 struct ctl_be_block_lun *lun; 2673 struct ctl_be_block_softc *softc; 2674 2675 lun = (struct ctl_be_block_lun *)be_lun; 2676 softc = lun->softc; 2677 2678 mtx_lock(&softc->lock); 2679 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2680 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2681 wakeup(lun); 2682 mtx_unlock(&softc->lock); 2683} 2684 2685static void 2686ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2687{ 2688 struct ctl_be_block_lun *lun; 2689 struct ctl_be_block_softc *softc; 2690 2691 lun = (struct ctl_be_block_lun *)be_lun; 2692 softc = lun->softc; 2693 2694 if (status == CTL_LUN_CONFIG_OK) { 2695 mtx_lock(&softc->lock); 2696 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2697 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2698 wakeup(lun); 2699 mtx_unlock(&softc->lock); 2700 2701 /* 2702 * We successfully added the LUN, attempt to enable it. 2703 */ 2704 if (ctl_enable_lun(&lun->cbe_lun) != 0) { 2705 printf("%s: ctl_enable_lun() failed!\n", __func__); 2706 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { 2707 printf("%s: ctl_invalidate_lun() failed!\n", 2708 __func__); 2709 } 2710 } 2711 2712 return; 2713 } 2714 2715 2716 mtx_lock(&softc->lock); 2717 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2718 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2719 wakeup(lun); 2720 mtx_unlock(&softc->lock); 2721} 2722 2723 2724static int 2725ctl_be_block_config_write(union ctl_io *io) 2726{ 2727 struct ctl_be_block_lun *be_lun; 2728 struct ctl_be_lun *cbe_lun; 2729 int retval; 2730 2731 DPRINTF("entered\n"); 2732 2733 cbe_lun = CTL_BACKEND_LUN(io); 2734 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2735 2736 retval = 0; 2737 switch (io->scsiio.cdb[0]) { 2738 case SYNCHRONIZE_CACHE: 2739 case SYNCHRONIZE_CACHE_16: 2740 case WRITE_SAME_10: 2741 case WRITE_SAME_16: 2742 case UNMAP: 2743 /* 2744 * The upper level CTL code will filter out any CDBs with 2745 * the immediate bit set and return the proper error. 2746 * 2747 * We don't really need to worry about what LBA range the 2748 * user asked to be synced out. When they issue a sync 2749 * cache command, we'll sync out the whole thing. 2750 */ 2751 mtx_lock(&be_lun->queue_lock); 2752 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2753 links); 2754 mtx_unlock(&be_lun->queue_lock); 2755 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2756 break; 2757 case START_STOP_UNIT: { 2758 struct scsi_start_stop_unit *cdb; 2759 struct ctl_lun_req req; 2760 2761 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2762 if ((cdb->how & SSS_PC_MASK) != 0) { 2763 ctl_set_success(&io->scsiio); 2764 ctl_config_write_done(io); 2765 break; 2766 } 2767 if (cdb->how & SSS_START) { 2768 if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) { 2769 retval = ctl_be_block_open(be_lun, &req); 2770 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2771 if (retval == 0) { 2772 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2773 ctl_lun_has_media(cbe_lun); 2774 } else { 2775 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2776 ctl_lun_no_media(cbe_lun); 2777 } 2778 } 2779 ctl_start_lun(cbe_lun); 2780 } else { 2781 ctl_stop_lun(cbe_lun); 2782 if (cdb->how & SSS_LOEJ) { 2783 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2784 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED; 2785 ctl_lun_ejected(cbe_lun); 2786 if (be_lun->vn != NULL) 2787 ctl_be_block_close(be_lun); 2788 } 2789 } 2790 2791 ctl_set_success(&io->scsiio); 2792 ctl_config_write_done(io); 2793 break; 2794 } 2795 case PREVENT_ALLOW: 2796 ctl_set_success(&io->scsiio); 2797 ctl_config_write_done(io); 2798 break; 2799 default: 2800 ctl_set_invalid_opcode(&io->scsiio); 2801 ctl_config_write_done(io); 2802 retval = CTL_RETVAL_COMPLETE; 2803 break; 2804 } 2805 2806 return (retval); 2807} 2808 2809static int 2810ctl_be_block_config_read(union ctl_io *io) 2811{ 2812 struct ctl_be_block_lun *be_lun; 2813 struct ctl_be_lun *cbe_lun; 2814 int retval = 0; 2815 2816 DPRINTF("entered\n"); 2817 2818 cbe_lun = CTL_BACKEND_LUN(io); 2819 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2820 2821 switch (io->scsiio.cdb[0]) { 2822 case SERVICE_ACTION_IN: 2823 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2824 mtx_lock(&be_lun->queue_lock); 2825 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2826 &io->io_hdr, links); 2827 mtx_unlock(&be_lun->queue_lock); 2828 taskqueue_enqueue(be_lun->io_taskqueue, 2829 &be_lun->io_task); 2830 retval = CTL_RETVAL_QUEUED; 2831 break; 2832 } 2833 ctl_set_invalid_field(&io->scsiio, 2834 /*sks_valid*/ 1, 2835 /*command*/ 1, 2836 /*field*/ 1, 2837 /*bit_valid*/ 1, 2838 /*bit*/ 4); 2839 ctl_config_read_done(io); 2840 retval = CTL_RETVAL_COMPLETE; 2841 break; 2842 default: 2843 ctl_set_invalid_opcode(&io->scsiio); 2844 ctl_config_read_done(io); 2845 retval = CTL_RETVAL_COMPLETE; 2846 break; 2847 } 2848 2849 return (retval); 2850} 2851 2852static int 2853ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2854{ 2855 struct ctl_be_block_lun *lun; 2856 int retval; 2857 2858 lun = (struct ctl_be_block_lun *)be_lun; 2859 2860 retval = sbuf_printf(sb, "\t<num_threads>"); 2861 if (retval != 0) 2862 goto bailout; 2863 retval = sbuf_printf(sb, "%d", lun->num_threads); 2864 if (retval != 0) 2865 goto bailout; 2866 retval = sbuf_printf(sb, "</num_threads>\n"); 2867 2868bailout: 2869 return (retval); 2870} 2871 2872static uint64_t 2873ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2874{ 2875 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2876 2877 if (lun->getattr == NULL) 2878 return (UINT64_MAX); 2879 return (lun->getattr(lun, attrname)); 2880} 2881 2882int 2883ctl_be_block_init(void) 2884{ 2885 struct ctl_be_block_softc *softc; 2886 int retval; 2887 2888 softc = &backend_block_softc; 2889 retval = 0; 2890 2891 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2892 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2893 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2894 STAILQ_INIT(&softc->lun_list); 2895 2896 return (retval); 2897} 2898