ctl_backend_block.c revision 288732
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 288732 2015-10-05 08:57:16Z mav $"); 44 45#include <opt_kdtrace.h> 46 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/kernel.h> 50#include <sys/types.h> 51#include <sys/kthread.h> 52#include <sys/bio.h> 53#include <sys/fcntl.h> 54#include <sys/limits.h> 55#include <sys/lock.h> 56#include <sys/mutex.h> 57#include <sys/condvar.h> 58#include <sys/malloc.h> 59#include <sys/conf.h> 60#include <sys/ioccom.h> 61#include <sys/queue.h> 62#include <sys/sbuf.h> 63#include <sys/endian.h> 64#include <sys/uio.h> 65#include <sys/buf.h> 66#include <sys/taskqueue.h> 67#include <sys/vnode.h> 68#include <sys/namei.h> 69#include <sys/mount.h> 70#include <sys/disk.h> 71#include <sys/fcntl.h> 72#include <sys/filedesc.h> 73#include <sys/filio.h> 74#include <sys/proc.h> 75#include <sys/pcpu.h> 76#include <sys/module.h> 77#include <sys/sdt.h> 78#include <sys/devicestat.h> 79#include <sys/sysctl.h> 80 81#include <geom/geom.h> 82 83#include <cam/cam.h> 84#include <cam/scsi/scsi_all.h> 85#include <cam/scsi/scsi_da.h> 86#include <cam/ctl/ctl_io.h> 87#include <cam/ctl/ctl.h> 88#include <cam/ctl/ctl_backend.h> 89#include <cam/ctl/ctl_ioctl.h> 90#include <cam/ctl/ctl_ha.h> 91#include <cam/ctl/ctl_scsi_all.h> 92#include <cam/ctl/ctl_private.h> 93#include <cam/ctl/ctl_error.h> 94 95/* 96 * The idea here is that we'll allocate enough S/G space to hold a 1MB 97 * I/O. If we get an I/O larger than that, we'll split it. 98 */ 99#define CTLBLK_HALF_IO_SIZE (512 * 1024) 100#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 101#define CTLBLK_MAX_SEG MAXPHYS 102#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 103#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 104 105#ifdef CTLBLK_DEBUG 106#define DPRINTF(fmt, args...) \ 107 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 108#else 109#define DPRINTF(fmt, args...) do {} while(0) 110#endif 111 112#define PRIV(io) \ 113 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 114#define ARGS(io) \ 115 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 116 117SDT_PROVIDER_DEFINE(cbb); 118 119typedef enum { 120 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 121 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 122 CTL_BE_BLOCK_LUN_WAITING = 0x04, 123} ctl_be_block_lun_flags; 124 125typedef enum { 126 CTL_BE_BLOCK_NONE, 127 CTL_BE_BLOCK_DEV, 128 CTL_BE_BLOCK_FILE 129} ctl_be_block_type; 130 131struct ctl_be_block_devdata { 132 struct cdev *cdev; 133 struct cdevsw *csw; 134 int dev_ref; 135}; 136 137struct ctl_be_block_filedata { 138 struct ucred *cred; 139}; 140 141union ctl_be_block_bedata { 142 struct ctl_be_block_devdata dev; 143 struct ctl_be_block_filedata file; 144}; 145 146struct ctl_be_block_io; 147struct ctl_be_block_lun; 148 149typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 150 struct ctl_be_block_io *beio); 151typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 152 const char *attrname); 153 154/* 155 * Backend LUN structure. There is a 1:1 mapping between a block device 156 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 157 */ 158struct ctl_be_block_lun { 159 struct ctl_lun_create_params params; 160 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_dispatch_t get_lba_status; 169 cbb_getattr_t getattr; 170 uma_zone_t lun_zone; 171 uint64_t size_blocks; 172 uint64_t size_bytes; 173 struct ctl_be_block_softc *softc; 174 struct devstat *disk_stats; 175 ctl_be_block_lun_flags flags; 176 STAILQ_ENTRY(ctl_be_block_lun) links; 177 struct ctl_be_lun cbe_lun; 178 struct taskqueue *io_taskqueue; 179 struct task io_task; 180 int num_threads; 181 STAILQ_HEAD(, ctl_io_hdr) input_queue; 182 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 183 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 184 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 185 struct mtx_padalign io_lock; 186 struct mtx_padalign queue_lock; 187}; 188 189/* 190 * Overall softc structure for the block backend module. 191 */ 192struct ctl_be_block_softc { 193 struct mtx lock; 194 int num_luns; 195 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 196}; 197 198static struct ctl_be_block_softc backend_block_softc; 199 200/* 201 * Per-I/O information. 202 */ 203struct ctl_be_block_io { 204 union ctl_io *io; 205 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 206 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 207 int bio_cmd; 208 int num_segs; 209 int num_bios_sent; 210 int num_bios_done; 211 int send_complete; 212 int num_errors; 213 struct bintime ds_t0; 214 devstat_tag_type ds_tag_type; 215 devstat_trans_flags ds_trans_type; 216 uint64_t io_len; 217 uint64_t io_offset; 218 int io_arg; 219 struct ctl_be_block_softc *softc; 220 struct ctl_be_block_lun *lun; 221 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 222}; 223 224extern struct ctl_softc *control_softc; 225 226static int cbb_num_threads = 14; 227TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 228SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 229 "CAM Target Layer Block Backend"); 230SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 231 &cbb_num_threads, 0, "Number of threads per backing file"); 232 233static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 234static void ctl_free_beio(struct ctl_be_block_io *beio); 235static void ctl_complete_beio(struct ctl_be_block_io *beio); 236static int ctl_be_block_move_done(union ctl_io *io); 237static void ctl_be_block_biodone(struct bio *bio); 238static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 239 struct ctl_be_block_io *beio); 240static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 241 struct ctl_be_block_io *beio); 242static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 243 struct ctl_be_block_io *beio); 244static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 245 const char *attrname); 246static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 247 struct ctl_be_block_io *beio); 248static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 249 struct ctl_be_block_io *beio); 250static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 251 struct ctl_be_block_io *beio); 252static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 253 const char *attrname); 254static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 255 union ctl_io *io); 256static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 257 union ctl_io *io); 258static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 259 union ctl_io *io); 260static void ctl_be_block_worker(void *context, int pending); 261static int ctl_be_block_submit(union ctl_io *io); 262static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 263 int flag, struct thread *td); 264static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 265 struct ctl_lun_req *req); 266static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 267 struct ctl_lun_req *req); 268static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 269static int ctl_be_block_open(struct ctl_be_block_softc *softc, 270 struct ctl_be_block_lun *be_lun, 271 struct ctl_lun_req *req); 272static int ctl_be_block_create(struct ctl_be_block_softc *softc, 273 struct ctl_lun_req *req); 274static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 275 struct ctl_lun_req *req); 276static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 277 struct ctl_lun_req *req); 278static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 279 struct ctl_lun_req *req); 280static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 281 struct ctl_lun_req *req); 282static void ctl_be_block_lun_shutdown(void *be_lun); 283static void ctl_be_block_lun_config_status(void *be_lun, 284 ctl_lun_config_status status); 285static int ctl_be_block_config_write(union ctl_io *io); 286static int ctl_be_block_config_read(union ctl_io *io); 287static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 288static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 289int ctl_be_block_init(void); 290 291static struct ctl_backend_driver ctl_be_block_driver = 292{ 293 .name = "block", 294 .flags = CTL_BE_FLAG_HAS_CONFIG, 295 .init = ctl_be_block_init, 296 .data_submit = ctl_be_block_submit, 297 .data_move_done = ctl_be_block_move_done, 298 .config_read = ctl_be_block_config_read, 299 .config_write = ctl_be_block_config_write, 300 .ioctl = ctl_be_block_ioctl, 301 .lun_info = ctl_be_block_lun_info, 302 .lun_attr = ctl_be_block_lun_attr 303}; 304 305MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 306CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 307 308static uma_zone_t beio_zone; 309 310static struct ctl_be_block_io * 311ctl_alloc_beio(struct ctl_be_block_softc *softc) 312{ 313 struct ctl_be_block_io *beio; 314 315 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 316 beio->softc = softc; 317 return (beio); 318} 319 320static void 321ctl_free_beio(struct ctl_be_block_io *beio) 322{ 323 int duplicate_free; 324 int i; 325 326 duplicate_free = 0; 327 328 for (i = 0; i < beio->num_segs; i++) { 329 if (beio->sg_segs[i].addr == NULL) 330 duplicate_free++; 331 332 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 333 beio->sg_segs[i].addr = NULL; 334 335 /* For compare we had two equal S/G lists. */ 336 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 337 uma_zfree(beio->lun->lun_zone, 338 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 339 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 340 } 341 } 342 343 if (duplicate_free > 0) { 344 printf("%s: %d duplicate frees out of %d segments\n", __func__, 345 duplicate_free, beio->num_segs); 346 } 347 348 uma_zfree(beio_zone, beio); 349} 350 351static void 352ctl_complete_beio(struct ctl_be_block_io *beio) 353{ 354 union ctl_io *io = beio->io; 355 356 if (beio->beio_cont != NULL) { 357 beio->beio_cont(beio); 358 } else { 359 ctl_free_beio(beio); 360 ctl_data_submit_done(io); 361 } 362} 363 364static int 365ctl_be_block_move_done(union ctl_io *io) 366{ 367 struct ctl_be_block_io *beio; 368 struct ctl_be_block_lun *be_lun; 369 struct ctl_lba_len_flags *lbalen; 370#ifdef CTL_TIME_IO 371 struct bintime cur_bt; 372#endif 373 int i; 374 375 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 376 be_lun = beio->lun; 377 378 DPRINTF("entered\n"); 379 380#ifdef CTL_TIME_IO 381 getbintime(&cur_bt); 382 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 383 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 384 io->io_hdr.num_dmas++; 385#endif 386 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 387 388 /* 389 * We set status at this point for read commands, and write 390 * commands with errors. 391 */ 392 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 393 ; 394 } else if ((io->io_hdr.port_status == 0) && 395 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 396 lbalen = ARGS(beio->io); 397 if (lbalen->flags & CTL_LLF_READ) { 398 ctl_set_success(&io->scsiio); 399 } else if (lbalen->flags & CTL_LLF_COMPARE) { 400 /* We have two data blocks ready for comparison. */ 401 for (i = 0; i < beio->num_segs; i++) { 402 if (memcmp(beio->sg_segs[i].addr, 403 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 404 beio->sg_segs[i].len) != 0) 405 break; 406 } 407 if (i < beio->num_segs) 408 ctl_set_sense(&io->scsiio, 409 /*current_error*/ 1, 410 /*sense_key*/ SSD_KEY_MISCOMPARE, 411 /*asc*/ 0x1D, 412 /*ascq*/ 0x00, 413 SSD_ELEM_NONE); 414 else 415 ctl_set_success(&io->scsiio); 416 } 417 } else if ((io->io_hdr.port_status != 0) && 418 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 419 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 420 /* 421 * For hardware error sense keys, the sense key 422 * specific value is defined to be a retry count, 423 * but we use it to pass back an internal FETD 424 * error code. XXX KDM Hopefully the FETD is only 425 * using 16 bits for an error code, since that's 426 * all the space we have in the sks field. 427 */ 428 ctl_set_internal_failure(&io->scsiio, 429 /*sks_valid*/ 1, 430 /*retry_count*/ 431 io->io_hdr.port_status); 432 } 433 434 /* 435 * If this is a read, or a write with errors, it is done. 436 */ 437 if ((beio->bio_cmd == BIO_READ) 438 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 439 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 440 ctl_complete_beio(beio); 441 return (0); 442 } 443 444 /* 445 * At this point, we have a write and the DMA completed 446 * successfully. We now have to queue it to the task queue to 447 * execute the backend I/O. That is because we do blocking 448 * memory allocations, and in the file backing case, blocking I/O. 449 * This move done routine is generally called in the SIM's 450 * interrupt context, and therefore we cannot block. 451 */ 452 mtx_lock(&be_lun->queue_lock); 453 /* 454 * XXX KDM make sure that links is okay to use at this point. 455 * Otherwise, we either need to add another field to ctl_io_hdr, 456 * or deal with resource allocation here. 457 */ 458 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 459 mtx_unlock(&be_lun->queue_lock); 460 461 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 462 463 return (0); 464} 465 466static void 467ctl_be_block_biodone(struct bio *bio) 468{ 469 struct ctl_be_block_io *beio; 470 struct ctl_be_block_lun *be_lun; 471 union ctl_io *io; 472 int error; 473 474 beio = bio->bio_caller1; 475 be_lun = beio->lun; 476 io = beio->io; 477 478 DPRINTF("entered\n"); 479 480 error = bio->bio_error; 481 mtx_lock(&be_lun->io_lock); 482 if (error != 0) 483 beio->num_errors++; 484 485 beio->num_bios_done++; 486 487 /* 488 * XXX KDM will this cause WITNESS to complain? Holding a lock 489 * during the free might cause it to complain. 490 */ 491 g_destroy_bio(bio); 492 493 /* 494 * If the send complete bit isn't set, or we aren't the last I/O to 495 * complete, then we're done. 496 */ 497 if ((beio->send_complete == 0) 498 || (beio->num_bios_done < beio->num_bios_sent)) { 499 mtx_unlock(&be_lun->io_lock); 500 return; 501 } 502 503 /* 504 * At this point, we've verified that we are the last I/O to 505 * complete, so it's safe to drop the lock. 506 */ 507 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 508 beio->ds_tag_type, beio->ds_trans_type, 509 /*now*/ NULL, /*then*/&beio->ds_t0); 510 mtx_unlock(&be_lun->io_lock); 511 512 /* 513 * If there are any errors from the backing device, we fail the 514 * entire I/O with a medium error. 515 */ 516 if (beio->num_errors > 0) { 517 if (error == EOPNOTSUPP) { 518 ctl_set_invalid_opcode(&io->scsiio); 519 } else if (error == ENOSPC || error == EDQUOT) { 520 ctl_set_space_alloc_fail(&io->scsiio); 521 } else if (beio->bio_cmd == BIO_FLUSH) { 522 /* XXX KDM is there is a better error here? */ 523 ctl_set_internal_failure(&io->scsiio, 524 /*sks_valid*/ 1, 525 /*retry_count*/ 0xbad2); 526 } else 527 ctl_set_medium_error(&io->scsiio); 528 ctl_complete_beio(beio); 529 return; 530 } 531 532 /* 533 * If this is a write, a flush, a delete or verify, we're all done. 534 * If this is a read, we can now send the data to the user. 535 */ 536 if ((beio->bio_cmd == BIO_WRITE) 537 || (beio->bio_cmd == BIO_FLUSH) 538 || (beio->bio_cmd == BIO_DELETE) 539 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 540 ctl_set_success(&io->scsiio); 541 ctl_complete_beio(beio); 542 } else { 543 if ((ARGS(io)->flags & CTL_LLF_READ) && 544 beio->beio_cont == NULL) 545 ctl_set_success(&io->scsiio); 546#ifdef CTL_TIME_IO 547 getbintime(&io->io_hdr.dma_start_bt); 548#endif 549 ctl_datamove(io); 550 } 551} 552 553static void 554ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 555 struct ctl_be_block_io *beio) 556{ 557 union ctl_io *io = beio->io; 558 struct mount *mountpoint; 559 int error, lock_flags; 560 561 DPRINTF("entered\n"); 562 563 binuptime(&beio->ds_t0); 564 mtx_lock(&be_lun->io_lock); 565 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 566 mtx_unlock(&be_lun->io_lock); 567 568 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 569 570 if (MNT_SHARED_WRITES(mountpoint) 571 || ((mountpoint == NULL) 572 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 573 lock_flags = LK_SHARED; 574 else 575 lock_flags = LK_EXCLUSIVE; 576 577 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 578 579 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 580 curthread); 581 VOP_UNLOCK(be_lun->vn, 0); 582 583 vn_finished_write(mountpoint); 584 585 mtx_lock(&be_lun->io_lock); 586 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 587 beio->ds_tag_type, beio->ds_trans_type, 588 /*now*/ NULL, /*then*/&beio->ds_t0); 589 mtx_unlock(&be_lun->io_lock); 590 591 if (error == 0) 592 ctl_set_success(&io->scsiio); 593 else { 594 /* XXX KDM is there is a better error here? */ 595 ctl_set_internal_failure(&io->scsiio, 596 /*sks_valid*/ 1, 597 /*retry_count*/ 0xbad1); 598 } 599 600 ctl_complete_beio(beio); 601} 602 603SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 604SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 605SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 606SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 607 608static void 609ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 610 struct ctl_be_block_io *beio) 611{ 612 struct ctl_be_block_filedata *file_data; 613 union ctl_io *io; 614 struct uio xuio; 615 struct iovec *xiovec; 616 int flags; 617 int error, i; 618 619 DPRINTF("entered\n"); 620 621 file_data = &be_lun->backend.file; 622 io = beio->io; 623 flags = 0; 624 if (ARGS(io)->flags & CTL_LLF_DPO) 625 flags |= IO_DIRECT; 626 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 627 flags |= IO_SYNC; 628 629 bzero(&xuio, sizeof(xuio)); 630 if (beio->bio_cmd == BIO_READ) { 631 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 632 xuio.uio_rw = UIO_READ; 633 } else { 634 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 635 xuio.uio_rw = UIO_WRITE; 636 } 637 xuio.uio_offset = beio->io_offset; 638 xuio.uio_resid = beio->io_len; 639 xuio.uio_segflg = UIO_SYSSPACE; 640 xuio.uio_iov = beio->xiovecs; 641 xuio.uio_iovcnt = beio->num_segs; 642 xuio.uio_td = curthread; 643 644 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 645 xiovec->iov_base = beio->sg_segs[i].addr; 646 xiovec->iov_len = beio->sg_segs[i].len; 647 } 648 649 binuptime(&beio->ds_t0); 650 mtx_lock(&be_lun->io_lock); 651 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 652 mtx_unlock(&be_lun->io_lock); 653 654 if (beio->bio_cmd == BIO_READ) { 655 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 656 657 /* 658 * UFS pays attention to IO_DIRECT for reads. If the 659 * DIRECTIO option is configured into the kernel, it calls 660 * ffs_rawread(). But that only works for single-segment 661 * uios with user space addresses. In our case, with a 662 * kernel uio, it still reads into the buffer cache, but it 663 * will just try to release the buffer from the cache later 664 * on in ffs_read(). 665 * 666 * ZFS does not pay attention to IO_DIRECT for reads. 667 * 668 * UFS does not pay attention to IO_SYNC for reads. 669 * 670 * ZFS pays attention to IO_SYNC (which translates into the 671 * Solaris define FRSYNC for zfs_read()) for reads. It 672 * attempts to sync the file before reading. 673 */ 674 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 675 676 VOP_UNLOCK(be_lun->vn, 0); 677 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 678 } else { 679 struct mount *mountpoint; 680 int lock_flags; 681 682 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 683 684 if (MNT_SHARED_WRITES(mountpoint) 685 || ((mountpoint == NULL) 686 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 687 lock_flags = LK_SHARED; 688 else 689 lock_flags = LK_EXCLUSIVE; 690 691 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 692 693 /* 694 * UFS pays attention to IO_DIRECT for writes. The write 695 * is done asynchronously. (Normally the write would just 696 * get put into cache. 697 * 698 * UFS pays attention to IO_SYNC for writes. It will 699 * attempt to write the buffer out synchronously if that 700 * flag is set. 701 * 702 * ZFS does not pay attention to IO_DIRECT for writes. 703 * 704 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 705 * for writes. It will flush the transaction from the 706 * cache before returning. 707 */ 708 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 709 VOP_UNLOCK(be_lun->vn, 0); 710 711 vn_finished_write(mountpoint); 712 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 713 } 714 715 mtx_lock(&be_lun->io_lock); 716 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 717 beio->ds_tag_type, beio->ds_trans_type, 718 /*now*/ NULL, /*then*/&beio->ds_t0); 719 mtx_unlock(&be_lun->io_lock); 720 721 /* 722 * If we got an error, set the sense data to "MEDIUM ERROR" and 723 * return the I/O to the user. 724 */ 725 if (error != 0) { 726 char path_str[32]; 727 728 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 729 printf("%s%s command returned errno %d\n", path_str, 730 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 731 if (error == ENOSPC || error == EDQUOT) { 732 ctl_set_space_alloc_fail(&io->scsiio); 733 } else 734 ctl_set_medium_error(&io->scsiio); 735 ctl_complete_beio(beio); 736 return; 737 } 738 739 /* 740 * If this is a write or a verify, we're all done. 741 * If this is a read, we can now send the data to the user. 742 */ 743 if ((beio->bio_cmd == BIO_WRITE) || 744 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 745 ctl_set_success(&io->scsiio); 746 ctl_complete_beio(beio); 747 } else { 748 if ((ARGS(io)->flags & CTL_LLF_READ) && 749 beio->beio_cont == NULL) 750 ctl_set_success(&io->scsiio); 751#ifdef CTL_TIME_IO 752 getbintime(&io->io_hdr.dma_start_bt); 753#endif 754 ctl_datamove(io); 755 } 756} 757 758static void 759ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 760 struct ctl_be_block_io *beio) 761{ 762 union ctl_io *io = beio->io; 763 struct ctl_lba_len_flags *lbalen = ARGS(io); 764 struct scsi_get_lba_status_data *data; 765 off_t roff, off; 766 int error, status; 767 768 DPRINTF("entered\n"); 769 770 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 771 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 772 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 773 0, curthread->td_ucred, curthread); 774 if (error == 0 && off > roff) 775 status = 0; /* mapped up to off */ 776 else { 777 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 778 0, curthread->td_ucred, curthread); 779 if (error == 0 && off > roff) 780 status = 1; /* deallocated up to off */ 781 else { 782 status = 0; /* unknown up to the end */ 783 off = be_lun->size_bytes; 784 } 785 } 786 VOP_UNLOCK(be_lun->vn, 0); 787 788 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 789 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 790 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 791 lbalen->lba), data->descr[0].length); 792 data->descr[0].status = status; 793 794 ctl_complete_beio(beio); 795} 796 797static uint64_t 798ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 799{ 800 struct vattr vattr; 801 struct statfs statfs; 802 uint64_t val; 803 int error; 804 805 val = UINT64_MAX; 806 if (be_lun->vn == NULL) 807 return (val); 808 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 809 if (strcmp(attrname, "blocksused") == 0) { 810 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 811 if (error == 0) 812 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 813 } 814 if (strcmp(attrname, "blocksavail") == 0 && 815 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 816 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 817 if (error == 0) 818 val = statfs.f_bavail * statfs.f_bsize / 819 be_lun->cbe_lun.blocksize; 820 } 821 VOP_UNLOCK(be_lun->vn, 0); 822 return (val); 823} 824 825static void 826ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 827 struct ctl_be_block_io *beio) 828{ 829 struct ctl_be_block_devdata *dev_data; 830 union ctl_io *io; 831 struct uio xuio; 832 struct iovec *xiovec; 833 int flags; 834 int error, i; 835 836 DPRINTF("entered\n"); 837 838 dev_data = &be_lun->backend.dev; 839 io = beio->io; 840 flags = 0; 841 if (ARGS(io)->flags & CTL_LLF_DPO) 842 flags |= IO_DIRECT; 843 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 844 flags |= IO_SYNC; 845 846 bzero(&xuio, sizeof(xuio)); 847 if (beio->bio_cmd == BIO_READ) { 848 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 849 xuio.uio_rw = UIO_READ; 850 } else { 851 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 852 xuio.uio_rw = UIO_WRITE; 853 } 854 xuio.uio_offset = beio->io_offset; 855 xuio.uio_resid = beio->io_len; 856 xuio.uio_segflg = UIO_SYSSPACE; 857 xuio.uio_iov = beio->xiovecs; 858 xuio.uio_iovcnt = beio->num_segs; 859 xuio.uio_td = curthread; 860 861 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 862 xiovec->iov_base = beio->sg_segs[i].addr; 863 xiovec->iov_len = beio->sg_segs[i].len; 864 } 865 866 binuptime(&beio->ds_t0); 867 mtx_lock(&be_lun->io_lock); 868 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 869 mtx_unlock(&be_lun->io_lock); 870 871 if (beio->bio_cmd == BIO_READ) { 872 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 873 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 874 } else { 875 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 876 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 877 } 878 879 mtx_lock(&be_lun->io_lock); 880 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 881 beio->ds_tag_type, beio->ds_trans_type, 882 /*now*/ NULL, /*then*/&beio->ds_t0); 883 mtx_unlock(&be_lun->io_lock); 884 885 /* 886 * If we got an error, set the sense data to "MEDIUM ERROR" and 887 * return the I/O to the user. 888 */ 889 if (error != 0) { 890 if (error == ENOSPC || error == EDQUOT) { 891 ctl_set_space_alloc_fail(&io->scsiio); 892 } else 893 ctl_set_medium_error(&io->scsiio); 894 ctl_complete_beio(beio); 895 return; 896 } 897 898 /* 899 * If this is a write or a verify, we're all done. 900 * If this is a read, we can now send the data to the user. 901 */ 902 if ((beio->bio_cmd == BIO_WRITE) || 903 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 904 ctl_set_success(&io->scsiio); 905 ctl_complete_beio(beio); 906 } else { 907 if ((ARGS(io)->flags & CTL_LLF_READ) && 908 beio->beio_cont == NULL) 909 ctl_set_success(&io->scsiio); 910#ifdef CTL_TIME_IO 911 getbintime(&io->io_hdr.dma_start_bt); 912#endif 913 ctl_datamove(io); 914 } 915} 916 917static void 918ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 919 struct ctl_be_block_io *beio) 920{ 921 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 922 union ctl_io *io = beio->io; 923 struct ctl_lba_len_flags *lbalen = ARGS(io); 924 struct scsi_get_lba_status_data *data; 925 off_t roff, off; 926 int error, status; 927 928 DPRINTF("entered\n"); 929 930 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 931 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE, 932 (caddr_t)&off, FREAD, curthread); 933 if (error == 0 && off > roff) 934 status = 0; /* mapped up to off */ 935 else { 936 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA, 937 (caddr_t)&off, FREAD, curthread); 938 if (error == 0 && off > roff) 939 status = 1; /* deallocated up to off */ 940 else { 941 status = 0; /* unknown up to the end */ 942 off = be_lun->size_bytes; 943 } 944 } 945 946 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 947 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 948 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 949 lbalen->lba), data->descr[0].length); 950 data->descr[0].status = status; 951 952 ctl_complete_beio(beio); 953} 954 955static void 956ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 957 struct ctl_be_block_io *beio) 958{ 959 struct bio *bio; 960 union ctl_io *io; 961 struct ctl_be_block_devdata *dev_data; 962 963 dev_data = &be_lun->backend.dev; 964 io = beio->io; 965 966 DPRINTF("entered\n"); 967 968 /* This can't fail, it's a blocking allocation. */ 969 bio = g_alloc_bio(); 970 971 bio->bio_cmd = BIO_FLUSH; 972 bio->bio_dev = dev_data->cdev; 973 bio->bio_offset = 0; 974 bio->bio_data = 0; 975 bio->bio_done = ctl_be_block_biodone; 976 bio->bio_caller1 = beio; 977 bio->bio_pblkno = 0; 978 979 /* 980 * We don't need to acquire the LUN lock here, because we are only 981 * sending one bio, and so there is no other context to synchronize 982 * with. 983 */ 984 beio->num_bios_sent = 1; 985 beio->send_complete = 1; 986 987 binuptime(&beio->ds_t0); 988 mtx_lock(&be_lun->io_lock); 989 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 990 mtx_unlock(&be_lun->io_lock); 991 992 (*dev_data->csw->d_strategy)(bio); 993} 994 995static void 996ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 997 struct ctl_be_block_io *beio, 998 uint64_t off, uint64_t len, int last) 999{ 1000 struct bio *bio; 1001 struct ctl_be_block_devdata *dev_data; 1002 uint64_t maxlen; 1003 1004 dev_data = &be_lun->backend.dev; 1005 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1006 while (len > 0) { 1007 bio = g_alloc_bio(); 1008 bio->bio_cmd = BIO_DELETE; 1009 bio->bio_dev = dev_data->cdev; 1010 bio->bio_offset = off; 1011 bio->bio_length = MIN(len, maxlen); 1012 bio->bio_data = 0; 1013 bio->bio_done = ctl_be_block_biodone; 1014 bio->bio_caller1 = beio; 1015 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1016 1017 off += bio->bio_length; 1018 len -= bio->bio_length; 1019 1020 mtx_lock(&be_lun->io_lock); 1021 beio->num_bios_sent++; 1022 if (last && len == 0) 1023 beio->send_complete = 1; 1024 mtx_unlock(&be_lun->io_lock); 1025 1026 (*dev_data->csw->d_strategy)(bio); 1027 } 1028} 1029 1030static void 1031ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1032 struct ctl_be_block_io *beio) 1033{ 1034 union ctl_io *io; 1035 struct ctl_be_block_devdata *dev_data; 1036 struct ctl_ptr_len_flags *ptrlen; 1037 struct scsi_unmap_desc *buf, *end; 1038 uint64_t len; 1039 1040 dev_data = &be_lun->backend.dev; 1041 io = beio->io; 1042 1043 DPRINTF("entered\n"); 1044 1045 binuptime(&beio->ds_t0); 1046 mtx_lock(&be_lun->io_lock); 1047 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1048 mtx_unlock(&be_lun->io_lock); 1049 1050 if (beio->io_offset == -1) { 1051 beio->io_len = 0; 1052 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1053 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1054 end = buf + ptrlen->len / sizeof(*buf); 1055 for (; buf < end; buf++) { 1056 len = (uint64_t)scsi_4btoul(buf->length) * 1057 be_lun->cbe_lun.blocksize; 1058 beio->io_len += len; 1059 ctl_be_block_unmap_dev_range(be_lun, beio, 1060 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1061 len, (end - buf < 2) ? TRUE : FALSE); 1062 } 1063 } else 1064 ctl_be_block_unmap_dev_range(be_lun, beio, 1065 beio->io_offset, beio->io_len, TRUE); 1066} 1067 1068static void 1069ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1070 struct ctl_be_block_io *beio) 1071{ 1072 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1073 int i; 1074 struct bio *bio; 1075 struct ctl_be_block_devdata *dev_data; 1076 off_t cur_offset; 1077 int max_iosize; 1078 1079 DPRINTF("entered\n"); 1080 1081 dev_data = &be_lun->backend.dev; 1082 1083 /* 1084 * We have to limit our I/O size to the maximum supported by the 1085 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1086 * set it properly, use DFLTPHYS. 1087 */ 1088 max_iosize = dev_data->cdev->si_iosize_max; 1089 if (max_iosize < PAGE_SIZE) 1090 max_iosize = DFLTPHYS; 1091 1092 cur_offset = beio->io_offset; 1093 for (i = 0; i < beio->num_segs; i++) { 1094 size_t cur_size; 1095 uint8_t *cur_ptr; 1096 1097 cur_size = beio->sg_segs[i].len; 1098 cur_ptr = beio->sg_segs[i].addr; 1099 1100 while (cur_size > 0) { 1101 /* This can't fail, it's a blocking allocation. */ 1102 bio = g_alloc_bio(); 1103 1104 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1105 1106 bio->bio_cmd = beio->bio_cmd; 1107 bio->bio_dev = dev_data->cdev; 1108 bio->bio_caller1 = beio; 1109 bio->bio_length = min(cur_size, max_iosize); 1110 bio->bio_offset = cur_offset; 1111 bio->bio_data = cur_ptr; 1112 bio->bio_done = ctl_be_block_biodone; 1113 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1114 1115 cur_offset += bio->bio_length; 1116 cur_ptr += bio->bio_length; 1117 cur_size -= bio->bio_length; 1118 1119 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1120 beio->num_bios_sent++; 1121 } 1122 } 1123 binuptime(&beio->ds_t0); 1124 mtx_lock(&be_lun->io_lock); 1125 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1126 beio->send_complete = 1; 1127 mtx_unlock(&be_lun->io_lock); 1128 1129 /* 1130 * Fire off all allocated requests! 1131 */ 1132 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1133 TAILQ_REMOVE(&queue, bio, bio_queue); 1134 (*dev_data->csw->d_strategy)(bio); 1135 } 1136} 1137 1138static uint64_t 1139ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1140{ 1141 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1142 struct diocgattr_arg arg; 1143 int error; 1144 1145 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1146 return (UINT64_MAX); 1147 strlcpy(arg.name, attrname, sizeof(arg.name)); 1148 arg.len = sizeof(arg.value.off); 1149 error = dev_data->csw->d_ioctl(dev_data->cdev, 1150 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1151 if (error != 0) 1152 return (UINT64_MAX); 1153 return (arg.value.off); 1154} 1155 1156static void 1157ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1158 union ctl_io *io) 1159{ 1160 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1161 struct ctl_be_block_io *beio; 1162 struct ctl_lba_len_flags *lbalen; 1163 1164 DPRINTF("entered\n"); 1165 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1166 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1167 1168 beio->io_len = lbalen->len * cbe_lun->blocksize; 1169 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1170 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1171 beio->bio_cmd = BIO_FLUSH; 1172 beio->ds_trans_type = DEVSTAT_NO_DATA; 1173 DPRINTF("SYNC\n"); 1174 be_lun->lun_flush(be_lun, beio); 1175} 1176 1177static void 1178ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1179{ 1180 union ctl_io *io; 1181 1182 io = beio->io; 1183 ctl_free_beio(beio); 1184 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1185 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1186 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1187 ctl_config_write_done(io); 1188 return; 1189 } 1190 1191 ctl_be_block_config_write(io); 1192} 1193 1194static void 1195ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1196 union ctl_io *io) 1197{ 1198 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1199 struct ctl_be_block_io *beio; 1200 struct ctl_lba_len_flags *lbalen; 1201 uint64_t len_left, lba; 1202 uint32_t pb, pbo, adj; 1203 int i, seglen; 1204 uint8_t *buf, *end; 1205 1206 DPRINTF("entered\n"); 1207 1208 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1209 lbalen = ARGS(beio->io); 1210 1211 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1212 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1213 ctl_free_beio(beio); 1214 ctl_set_invalid_field(&io->scsiio, 1215 /*sks_valid*/ 1, 1216 /*command*/ 1, 1217 /*field*/ 1, 1218 /*bit_valid*/ 0, 1219 /*bit*/ 0); 1220 ctl_config_write_done(io); 1221 return; 1222 } 1223 1224 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1225 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1226 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1227 beio->bio_cmd = BIO_DELETE; 1228 beio->ds_trans_type = DEVSTAT_FREE; 1229 1230 be_lun->unmap(be_lun, beio); 1231 return; 1232 } 1233 1234 beio->bio_cmd = BIO_WRITE; 1235 beio->ds_trans_type = DEVSTAT_WRITE; 1236 1237 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1238 (uintmax_t)lbalen->lba, lbalen->len); 1239 1240 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1241 if (be_lun->cbe_lun.pblockoff > 0) 1242 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1243 else 1244 pbo = 0; 1245 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1246 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1247 1248 /* 1249 * Setup the S/G entry for this chunk. 1250 */ 1251 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1252 if (pb > cbe_lun->blocksize) { 1253 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1254 seglen - pbo) % pb; 1255 if (seglen > adj) 1256 seglen -= adj; 1257 else 1258 seglen -= seglen % cbe_lun->blocksize; 1259 } else 1260 seglen -= seglen % cbe_lun->blocksize; 1261 beio->sg_segs[i].len = seglen; 1262 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1263 1264 DPRINTF("segment %d addr %p len %zd\n", i, 1265 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1266 1267 beio->num_segs++; 1268 len_left -= seglen; 1269 1270 buf = beio->sg_segs[i].addr; 1271 end = buf + seglen; 1272 for (; buf < end; buf += cbe_lun->blocksize) { 1273 memcpy(buf, io->scsiio.kern_data_ptr, cbe_lun->blocksize); 1274 if (lbalen->flags & SWS_LBDATA) 1275 scsi_ulto4b(lbalen->lba + lba, buf); 1276 lba++; 1277 } 1278 } 1279 1280 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1281 beio->io_len = lba * cbe_lun->blocksize; 1282 1283 /* We can not do all in one run. Correct and schedule rerun. */ 1284 if (len_left > 0) { 1285 lbalen->lba += lba; 1286 lbalen->len -= lba; 1287 beio->beio_cont = ctl_be_block_cw_done_ws; 1288 } 1289 1290 be_lun->dispatch(be_lun, beio); 1291} 1292 1293static void 1294ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1295 union ctl_io *io) 1296{ 1297 struct ctl_be_block_io *beio; 1298 struct ctl_ptr_len_flags *ptrlen; 1299 1300 DPRINTF("entered\n"); 1301 1302 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1303 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1304 1305 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1306 ctl_free_beio(beio); 1307 ctl_set_invalid_field(&io->scsiio, 1308 /*sks_valid*/ 0, 1309 /*command*/ 1, 1310 /*field*/ 0, 1311 /*bit_valid*/ 0, 1312 /*bit*/ 0); 1313 ctl_config_write_done(io); 1314 return; 1315 } 1316 1317 beio->io_len = 0; 1318 beio->io_offset = -1; 1319 beio->bio_cmd = BIO_DELETE; 1320 beio->ds_trans_type = DEVSTAT_FREE; 1321 DPRINTF("UNMAP\n"); 1322 be_lun->unmap(be_lun, beio); 1323} 1324 1325static void 1326ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1327{ 1328 union ctl_io *io; 1329 1330 io = beio->io; 1331 ctl_free_beio(beio); 1332 ctl_config_read_done(io); 1333} 1334 1335static void 1336ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1337 union ctl_io *io) 1338{ 1339 struct ctl_be_block_io *beio; 1340 struct ctl_be_block_softc *softc; 1341 1342 DPRINTF("entered\n"); 1343 1344 softc = be_lun->softc; 1345 beio = ctl_alloc_beio(softc); 1346 beio->io = io; 1347 beio->lun = be_lun; 1348 beio->beio_cont = ctl_be_block_cr_done; 1349 PRIV(io)->ptr = (void *)beio; 1350 1351 switch (io->scsiio.cdb[0]) { 1352 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1353 beio->bio_cmd = -1; 1354 beio->ds_trans_type = DEVSTAT_NO_DATA; 1355 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1356 beio->io_len = 0; 1357 if (be_lun->get_lba_status) 1358 be_lun->get_lba_status(be_lun, beio); 1359 else 1360 ctl_be_block_cr_done(beio); 1361 break; 1362 default: 1363 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1364 break; 1365 } 1366} 1367 1368static void 1369ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1370{ 1371 union ctl_io *io; 1372 1373 io = beio->io; 1374 ctl_free_beio(beio); 1375 ctl_config_write_done(io); 1376} 1377 1378static void 1379ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1380 union ctl_io *io) 1381{ 1382 struct ctl_be_block_io *beio; 1383 struct ctl_be_block_softc *softc; 1384 1385 DPRINTF("entered\n"); 1386 1387 softc = be_lun->softc; 1388 beio = ctl_alloc_beio(softc); 1389 beio->io = io; 1390 beio->lun = be_lun; 1391 beio->beio_cont = ctl_be_block_cw_done; 1392 switch (io->scsiio.tag_type) { 1393 case CTL_TAG_ORDERED: 1394 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1395 break; 1396 case CTL_TAG_HEAD_OF_QUEUE: 1397 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1398 break; 1399 case CTL_TAG_UNTAGGED: 1400 case CTL_TAG_SIMPLE: 1401 case CTL_TAG_ACA: 1402 default: 1403 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1404 break; 1405 } 1406 PRIV(io)->ptr = (void *)beio; 1407 1408 switch (io->scsiio.cdb[0]) { 1409 case SYNCHRONIZE_CACHE: 1410 case SYNCHRONIZE_CACHE_16: 1411 ctl_be_block_cw_dispatch_sync(be_lun, io); 1412 break; 1413 case WRITE_SAME_10: 1414 case WRITE_SAME_16: 1415 ctl_be_block_cw_dispatch_ws(be_lun, io); 1416 break; 1417 case UNMAP: 1418 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1419 break; 1420 default: 1421 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1422 break; 1423 } 1424} 1425 1426SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1427SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1428SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1429SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1430 1431static void 1432ctl_be_block_next(struct ctl_be_block_io *beio) 1433{ 1434 struct ctl_be_block_lun *be_lun; 1435 union ctl_io *io; 1436 1437 io = beio->io; 1438 be_lun = beio->lun; 1439 ctl_free_beio(beio); 1440 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1441 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1442 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1443 ctl_data_submit_done(io); 1444 return; 1445 } 1446 1447 io->io_hdr.status &= ~CTL_STATUS_MASK; 1448 io->io_hdr.status |= CTL_STATUS_NONE; 1449 1450 mtx_lock(&be_lun->queue_lock); 1451 /* 1452 * XXX KDM make sure that links is okay to use at this point. 1453 * Otherwise, we either need to add another field to ctl_io_hdr, 1454 * or deal with resource allocation here. 1455 */ 1456 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1457 mtx_unlock(&be_lun->queue_lock); 1458 1459 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1460} 1461 1462static void 1463ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1464 union ctl_io *io) 1465{ 1466 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1467 struct ctl_be_block_io *beio; 1468 struct ctl_be_block_softc *softc; 1469 struct ctl_lba_len_flags *lbalen; 1470 struct ctl_ptr_len_flags *bptrlen; 1471 uint64_t len_left, lbas; 1472 int i; 1473 1474 softc = be_lun->softc; 1475 1476 DPRINTF("entered\n"); 1477 1478 lbalen = ARGS(io); 1479 if (lbalen->flags & CTL_LLF_WRITE) { 1480 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1481 } else { 1482 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1483 } 1484 1485 beio = ctl_alloc_beio(softc); 1486 beio->io = io; 1487 beio->lun = be_lun; 1488 bptrlen = PRIV(io); 1489 bptrlen->ptr = (void *)beio; 1490 1491 switch (io->scsiio.tag_type) { 1492 case CTL_TAG_ORDERED: 1493 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1494 break; 1495 case CTL_TAG_HEAD_OF_QUEUE: 1496 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1497 break; 1498 case CTL_TAG_UNTAGGED: 1499 case CTL_TAG_SIMPLE: 1500 case CTL_TAG_ACA: 1501 default: 1502 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1503 break; 1504 } 1505 1506 if (lbalen->flags & CTL_LLF_WRITE) { 1507 beio->bio_cmd = BIO_WRITE; 1508 beio->ds_trans_type = DEVSTAT_WRITE; 1509 } else { 1510 beio->bio_cmd = BIO_READ; 1511 beio->ds_trans_type = DEVSTAT_READ; 1512 } 1513 1514 DPRINTF("%s at LBA %jx len %u @%ju\n", 1515 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1516 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1517 if (lbalen->flags & CTL_LLF_COMPARE) 1518 lbas = CTLBLK_HALF_IO_SIZE; 1519 else 1520 lbas = CTLBLK_MAX_IO_SIZE; 1521 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1522 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1523 beio->io_len = lbas * cbe_lun->blocksize; 1524 bptrlen->len += lbas; 1525 1526 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1527 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1528 i, CTLBLK_MAX_SEGS)); 1529 1530 /* 1531 * Setup the S/G entry for this chunk. 1532 */ 1533 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1534 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1535 1536 DPRINTF("segment %d addr %p len %zd\n", i, 1537 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1538 1539 /* Set up second segment for compare operation. */ 1540 if (lbalen->flags & CTL_LLF_COMPARE) { 1541 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1542 beio->sg_segs[i].len; 1543 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1544 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1545 } 1546 1547 beio->num_segs++; 1548 len_left -= beio->sg_segs[i].len; 1549 } 1550 if (bptrlen->len < lbalen->len) 1551 beio->beio_cont = ctl_be_block_next; 1552 io->scsiio.be_move_done = ctl_be_block_move_done; 1553 /* For compare we have separate S/G lists for read and datamove. */ 1554 if (lbalen->flags & CTL_LLF_COMPARE) 1555 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1556 else 1557 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1558 io->scsiio.kern_data_len = beio->io_len; 1559 io->scsiio.kern_data_resid = 0; 1560 io->scsiio.kern_sg_entries = beio->num_segs; 1561 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1562 1563 /* 1564 * For the read case, we need to read the data into our buffers and 1565 * then we can send it back to the user. For the write case, we 1566 * need to get the data from the user first. 1567 */ 1568 if (beio->bio_cmd == BIO_READ) { 1569 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1570 be_lun->dispatch(be_lun, beio); 1571 } else { 1572 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1573#ifdef CTL_TIME_IO 1574 getbintime(&io->io_hdr.dma_start_bt); 1575#endif 1576 ctl_datamove(io); 1577 } 1578} 1579 1580static void 1581ctl_be_block_worker(void *context, int pending) 1582{ 1583 struct ctl_be_block_lun *be_lun; 1584 struct ctl_be_block_softc *softc; 1585 union ctl_io *io; 1586 1587 be_lun = (struct ctl_be_block_lun *)context; 1588 softc = be_lun->softc; 1589 1590 DPRINTF("entered\n"); 1591 1592 mtx_lock(&be_lun->queue_lock); 1593 for (;;) { 1594 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1595 if (io != NULL) { 1596 struct ctl_be_block_io *beio; 1597 1598 DPRINTF("datamove queue\n"); 1599 1600 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1601 ctl_io_hdr, links); 1602 1603 mtx_unlock(&be_lun->queue_lock); 1604 1605 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1606 1607 be_lun->dispatch(be_lun, beio); 1608 1609 mtx_lock(&be_lun->queue_lock); 1610 continue; 1611 } 1612 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1613 if (io != NULL) { 1614 DPRINTF("config write queue\n"); 1615 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1616 ctl_io_hdr, links); 1617 mtx_unlock(&be_lun->queue_lock); 1618 ctl_be_block_cw_dispatch(be_lun, io); 1619 mtx_lock(&be_lun->queue_lock); 1620 continue; 1621 } 1622 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1623 if (io != NULL) { 1624 DPRINTF("config read queue\n"); 1625 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1626 ctl_io_hdr, links); 1627 mtx_unlock(&be_lun->queue_lock); 1628 ctl_be_block_cr_dispatch(be_lun, io); 1629 mtx_lock(&be_lun->queue_lock); 1630 continue; 1631 } 1632 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1633 if (io != NULL) { 1634 DPRINTF("input queue\n"); 1635 1636 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1637 ctl_io_hdr, links); 1638 mtx_unlock(&be_lun->queue_lock); 1639 1640 /* 1641 * We must drop the lock, since this routine and 1642 * its children may sleep. 1643 */ 1644 ctl_be_block_dispatch(be_lun, io); 1645 1646 mtx_lock(&be_lun->queue_lock); 1647 continue; 1648 } 1649 1650 /* 1651 * If we get here, there is no work left in the queues, so 1652 * just break out and let the task queue go to sleep. 1653 */ 1654 break; 1655 } 1656 mtx_unlock(&be_lun->queue_lock); 1657} 1658 1659/* 1660 * Entry point from CTL to the backend for I/O. We queue everything to a 1661 * work thread, so this just puts the I/O on a queue and wakes up the 1662 * thread. 1663 */ 1664static int 1665ctl_be_block_submit(union ctl_io *io) 1666{ 1667 struct ctl_be_block_lun *be_lun; 1668 struct ctl_be_lun *cbe_lun; 1669 1670 DPRINTF("entered\n"); 1671 1672 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1673 CTL_PRIV_BACKEND_LUN].ptr; 1674 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1675 1676 /* 1677 * Make sure we only get SCSI I/O. 1678 */ 1679 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1680 "%#x) encountered", io->io_hdr.io_type)); 1681 1682 PRIV(io)->len = 0; 1683 1684 mtx_lock(&be_lun->queue_lock); 1685 /* 1686 * XXX KDM make sure that links is okay to use at this point. 1687 * Otherwise, we either need to add another field to ctl_io_hdr, 1688 * or deal with resource allocation here. 1689 */ 1690 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1691 mtx_unlock(&be_lun->queue_lock); 1692 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1693 1694 return (CTL_RETVAL_COMPLETE); 1695} 1696 1697static int 1698ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1699 int flag, struct thread *td) 1700{ 1701 struct ctl_be_block_softc *softc; 1702 int error; 1703 1704 softc = &backend_block_softc; 1705 1706 error = 0; 1707 1708 switch (cmd) { 1709 case CTL_LUN_REQ: { 1710 struct ctl_lun_req *lun_req; 1711 1712 lun_req = (struct ctl_lun_req *)addr; 1713 1714 switch (lun_req->reqtype) { 1715 case CTL_LUNREQ_CREATE: 1716 error = ctl_be_block_create(softc, lun_req); 1717 break; 1718 case CTL_LUNREQ_RM: 1719 error = ctl_be_block_rm(softc, lun_req); 1720 break; 1721 case CTL_LUNREQ_MODIFY: 1722 error = ctl_be_block_modify(softc, lun_req); 1723 break; 1724 default: 1725 lun_req->status = CTL_LUN_ERROR; 1726 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1727 "invalid LUN request type %d", 1728 lun_req->reqtype); 1729 break; 1730 } 1731 break; 1732 } 1733 default: 1734 error = ENOTTY; 1735 break; 1736 } 1737 1738 return (error); 1739} 1740 1741static int 1742ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1743{ 1744 struct ctl_be_lun *cbe_lun; 1745 struct ctl_be_block_filedata *file_data; 1746 struct ctl_lun_create_params *params; 1747 char *value; 1748 struct vattr vattr; 1749 off_t ps, pss, po, pos, us, uss, uo, uos; 1750 int error; 1751 1752 error = 0; 1753 cbe_lun = &be_lun->cbe_lun; 1754 file_data = &be_lun->backend.file; 1755 params = &be_lun->params; 1756 1757 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1758 be_lun->dispatch = ctl_be_block_dispatch_file; 1759 be_lun->lun_flush = ctl_be_block_flush_file; 1760 be_lun->get_lba_status = ctl_be_block_gls_file; 1761 be_lun->getattr = ctl_be_block_getattr_file; 1762 be_lun->unmap = NULL; 1763 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1764 1765 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1766 if (error != 0) { 1767 snprintf(req->error_str, sizeof(req->error_str), 1768 "error calling VOP_GETATTR() for file %s", 1769 be_lun->dev_path); 1770 return (error); 1771 } 1772 1773 /* 1774 * Verify that we have the ability to upgrade to exclusive 1775 * access on this file so we can trap errors at open instead 1776 * of reporting them during first access. 1777 */ 1778 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1779 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1780 if (be_lun->vn->v_iflag & VI_DOOMED) { 1781 error = EBADF; 1782 snprintf(req->error_str, sizeof(req->error_str), 1783 "error locking file %s", be_lun->dev_path); 1784 return (error); 1785 } 1786 } 1787 1788 file_data->cred = crhold(curthread->td_ucred); 1789 if (params->lun_size_bytes != 0) 1790 be_lun->size_bytes = params->lun_size_bytes; 1791 else 1792 be_lun->size_bytes = vattr.va_size; 1793 1794 /* 1795 * For files we can use any logical block size. Prefer 512 bytes 1796 * for compatibility reasons. If file's vattr.va_blocksize 1797 * (preferred I/O block size) is bigger and multiple to chosen 1798 * logical block size -- report it as physical block size. 1799 */ 1800 if (params->blocksize_bytes != 0) 1801 cbe_lun->blocksize = params->blocksize_bytes; 1802 else 1803 cbe_lun->blocksize = 512; 1804 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1805 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1806 0 : (be_lun->size_blocks - 1); 1807 1808 us = ps = vattr.va_blocksize; 1809 uo = po = 0; 1810 1811 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1812 if (value != NULL) 1813 ctl_expand_number(value, &ps); 1814 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1815 if (value != NULL) 1816 ctl_expand_number(value, &po); 1817 pss = ps / cbe_lun->blocksize; 1818 pos = po / cbe_lun->blocksize; 1819 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1820 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1821 cbe_lun->pblockexp = fls(pss) - 1; 1822 cbe_lun->pblockoff = (pss - pos) % pss; 1823 } 1824 1825 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1826 if (value != NULL) 1827 ctl_expand_number(value, &us); 1828 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 1829 if (value != NULL) 1830 ctl_expand_number(value, &uo); 1831 uss = us / cbe_lun->blocksize; 1832 uos = uo / cbe_lun->blocksize; 1833 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1834 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1835 cbe_lun->ublockexp = fls(uss) - 1; 1836 cbe_lun->ublockoff = (uss - uos) % uss; 1837 } 1838 1839 /* 1840 * Sanity check. The media size has to be at least one 1841 * sector long. 1842 */ 1843 if (be_lun->size_bytes < cbe_lun->blocksize) { 1844 error = EINVAL; 1845 snprintf(req->error_str, sizeof(req->error_str), 1846 "file %s size %ju < block size %u", be_lun->dev_path, 1847 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1848 } 1849 1850 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1851 return (error); 1852} 1853 1854static int 1855ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1856{ 1857 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1858 struct ctl_lun_create_params *params; 1859 struct vattr vattr; 1860 struct cdev *dev; 1861 struct cdevsw *devsw; 1862 char *value; 1863 int error, atomic, maxio, unmap, tmp; 1864 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1865 1866 params = &be_lun->params; 1867 1868 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1869 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1870 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1871 &be_lun->backend.dev.dev_ref); 1872 if (be_lun->backend.dev.csw == NULL) 1873 panic("Unable to retrieve device switch"); 1874 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) { 1875 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1876 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1877 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1878 } else { 1879 be_lun->dispatch = ctl_be_block_dispatch_dev; 1880 be_lun->get_lba_status = NULL; 1881 atomic = 0; 1882 maxio = be_lun->backend.dev.cdev->si_iosize_max; 1883 if (maxio <= 0) 1884 maxio = DFLTPHYS; 1885 if (maxio > CTLBLK_MAX_IO_SIZE) 1886 maxio = CTLBLK_MAX_IO_SIZE; 1887 } 1888 be_lun->lun_flush = ctl_be_block_flush_dev; 1889 be_lun->getattr = ctl_be_block_getattr_dev; 1890 be_lun->unmap = ctl_be_block_unmap_dev; 1891 1892 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1893 if (error) { 1894 snprintf(req->error_str, sizeof(req->error_str), 1895 "error getting vnode attributes for device %s", 1896 be_lun->dev_path); 1897 return (error); 1898 } 1899 1900 dev = be_lun->vn->v_rdev; 1901 devsw = dev->si_devsw; 1902 if (!devsw->d_ioctl) { 1903 snprintf(req->error_str, sizeof(req->error_str), 1904 "no d_ioctl for device %s!", 1905 be_lun->dev_path); 1906 return (ENODEV); 1907 } 1908 1909 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1910 curthread); 1911 if (error) { 1912 snprintf(req->error_str, sizeof(req->error_str), 1913 "error %d returned for DIOCGSECTORSIZE ioctl " 1914 "on %s!", error, be_lun->dev_path); 1915 return (error); 1916 } 1917 1918 /* 1919 * If the user has asked for a blocksize that is greater than the 1920 * backing device's blocksize, we can do it only if the blocksize 1921 * the user is asking for is an even multiple of the underlying 1922 * device's blocksize. 1923 */ 1924 if ((params->blocksize_bytes != 0) && 1925 (params->blocksize_bytes >= tmp)) { 1926 if (params->blocksize_bytes % tmp == 0) { 1927 cbe_lun->blocksize = params->blocksize_bytes; 1928 } else { 1929 snprintf(req->error_str, sizeof(req->error_str), 1930 "requested blocksize %u is not an even " 1931 "multiple of backing device blocksize %u", 1932 params->blocksize_bytes, tmp); 1933 return (EINVAL); 1934 1935 } 1936 } else if (params->blocksize_bytes != 0) { 1937 snprintf(req->error_str, sizeof(req->error_str), 1938 "requested blocksize %u < backing device " 1939 "blocksize %u", params->blocksize_bytes, tmp); 1940 return (EINVAL); 1941 } else 1942 cbe_lun->blocksize = tmp; 1943 1944 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 1945 curthread); 1946 if (error) { 1947 snprintf(req->error_str, sizeof(req->error_str), 1948 "error %d returned for DIOCGMEDIASIZE " 1949 " ioctl on %s!", error, 1950 be_lun->dev_path); 1951 return (error); 1952 } 1953 1954 if (params->lun_size_bytes != 0) { 1955 if (params->lun_size_bytes > otmp) { 1956 snprintf(req->error_str, sizeof(req->error_str), 1957 "requested LUN size %ju > backing device " 1958 "size %ju", 1959 (uintmax_t)params->lun_size_bytes, 1960 (uintmax_t)otmp); 1961 return (EINVAL); 1962 } 1963 1964 be_lun->size_bytes = params->lun_size_bytes; 1965 } else 1966 be_lun->size_bytes = otmp; 1967 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1968 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1969 0 : (be_lun->size_blocks - 1); 1970 1971 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1972 (caddr_t)&ps, FREAD, curthread); 1973 if (error) 1974 ps = po = 0; 1975 else { 1976 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1977 (caddr_t)&po, FREAD, curthread); 1978 if (error) 1979 po = 0; 1980 } 1981 us = ps; 1982 uo = po; 1983 1984 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1985 if (value != NULL) 1986 ctl_expand_number(value, &ps); 1987 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1988 if (value != NULL) 1989 ctl_expand_number(value, &po); 1990 pss = ps / cbe_lun->blocksize; 1991 pos = po / cbe_lun->blocksize; 1992 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1993 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1994 cbe_lun->pblockexp = fls(pss) - 1; 1995 cbe_lun->pblockoff = (pss - pos) % pss; 1996 } 1997 1998 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1999 if (value != NULL) 2000 ctl_expand_number(value, &us); 2001 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 2002 if (value != NULL) 2003 ctl_expand_number(value, &uo); 2004 uss = us / cbe_lun->blocksize; 2005 uos = uo / cbe_lun->blocksize; 2006 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2007 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2008 cbe_lun->ublockexp = fls(uss) - 1; 2009 cbe_lun->ublockoff = (uss - uos) % uss; 2010 } 2011 2012 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2013 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2014 2015 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2016 unmap = 1; 2017 } else { 2018 struct diocgattr_arg arg; 2019 2020 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2021 arg.len = sizeof(arg.value.i); 2022 error = devsw->d_ioctl(dev, DIOCGATTR, 2023 (caddr_t)&arg, FREAD, curthread); 2024 unmap = (error == 0) ? arg.value.i : 0; 2025 } 2026 value = ctl_get_opt(&cbe_lun->options, "unmap"); 2027 if (value != NULL) 2028 unmap = (strcmp(value, "on") == 0); 2029 if (unmap) 2030 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2031 else 2032 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2033 2034 return (0); 2035} 2036 2037static int 2038ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2039{ 2040 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2041 int flags; 2042 2043 if (be_lun->vn) { 2044 switch (be_lun->dev_type) { 2045 case CTL_BE_BLOCK_DEV: 2046 if (be_lun->backend.dev.csw) { 2047 dev_relthread(be_lun->backend.dev.cdev, 2048 be_lun->backend.dev.dev_ref); 2049 be_lun->backend.dev.csw = NULL; 2050 be_lun->backend.dev.cdev = NULL; 2051 } 2052 break; 2053 case CTL_BE_BLOCK_FILE: 2054 break; 2055 case CTL_BE_BLOCK_NONE: 2056 break; 2057 default: 2058 panic("Unexpected backend type."); 2059 break; 2060 } 2061 2062 flags = FREAD; 2063 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2064 flags |= FWRITE; 2065 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2066 be_lun->vn = NULL; 2067 2068 switch (be_lun->dev_type) { 2069 case CTL_BE_BLOCK_DEV: 2070 break; 2071 case CTL_BE_BLOCK_FILE: 2072 if (be_lun->backend.file.cred != NULL) { 2073 crfree(be_lun->backend.file.cred); 2074 be_lun->backend.file.cred = NULL; 2075 } 2076 break; 2077 case CTL_BE_BLOCK_NONE: 2078 break; 2079 default: 2080 panic("Unexpected backend type."); 2081 break; 2082 } 2083 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2084 } 2085 return (0); 2086} 2087 2088static int 2089ctl_be_block_open(struct ctl_be_block_softc *softc, 2090 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2091{ 2092 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2093 struct nameidata nd; 2094 char *value; 2095 int error, flags; 2096 2097 error = 0; 2098 if (rootvnode == NULL) { 2099 snprintf(req->error_str, sizeof(req->error_str), 2100 "Root filesystem is not mounted"); 2101 return (1); 2102 } 2103 if (!curthread->td_proc->p_fd->fd_cdir) { 2104 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2105 VREF(rootvnode); 2106 } 2107 if (!curthread->td_proc->p_fd->fd_rdir) { 2108 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2109 VREF(rootvnode); 2110 } 2111 if (!curthread->td_proc->p_fd->fd_jdir) { 2112 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2113 VREF(rootvnode); 2114 } 2115 2116 value = ctl_get_opt(&cbe_lun->options, "file"); 2117 if (value == NULL) { 2118 snprintf(req->error_str, sizeof(req->error_str), 2119 "no file argument specified"); 2120 return (1); 2121 } 2122 free(be_lun->dev_path, M_CTLBLK); 2123 be_lun->dev_path = strdup(value, M_CTLBLK); 2124 2125 flags = FREAD; 2126 value = ctl_get_opt(&cbe_lun->options, "readonly"); 2127 if (value == NULL || strcmp(value, "on") != 0) 2128 flags |= FWRITE; 2129 2130again: 2131 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2132 error = vn_open(&nd, &flags, 0, NULL); 2133 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2134 flags &= ~FWRITE; 2135 goto again; 2136 } 2137 if (error) { 2138 /* 2139 * This is the only reasonable guess we can make as far as 2140 * path if the user doesn't give us a fully qualified path. 2141 * If they want to specify a file, they need to specify the 2142 * full path. 2143 */ 2144 if (be_lun->dev_path[0] != '/') { 2145 char *dev_name; 2146 2147 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2148 be_lun->dev_path); 2149 free(be_lun->dev_path, M_CTLBLK); 2150 be_lun->dev_path = dev_name; 2151 goto again; 2152 } 2153 snprintf(req->error_str, sizeof(req->error_str), 2154 "error opening %s: %d", be_lun->dev_path, error); 2155 return (error); 2156 } 2157 if (flags & FWRITE) 2158 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2159 else 2160 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2161 2162 NDFREE(&nd, NDF_ONLY_PNBUF); 2163 be_lun->vn = nd.ni_vp; 2164 2165 /* We only support disks and files. */ 2166 if (vn_isdisk(be_lun->vn, &error)) { 2167 error = ctl_be_block_open_dev(be_lun, req); 2168 } else if (be_lun->vn->v_type == VREG) { 2169 error = ctl_be_block_open_file(be_lun, req); 2170 } else { 2171 error = EINVAL; 2172 snprintf(req->error_str, sizeof(req->error_str), 2173 "%s is not a disk or plain file", be_lun->dev_path); 2174 } 2175 VOP_UNLOCK(be_lun->vn, 0); 2176 2177 if (error != 0) 2178 ctl_be_block_close(be_lun); 2179 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2180 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2181 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2182 value = ctl_get_opt(&cbe_lun->options, "serseq"); 2183 if (value != NULL && strcmp(value, "on") == 0) 2184 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2185 else if (value != NULL && strcmp(value, "read") == 0) 2186 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2187 else if (value != NULL && strcmp(value, "off") == 0) 2188 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2189 return (0); 2190} 2191 2192static int 2193ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2194{ 2195 struct ctl_be_lun *cbe_lun; 2196 struct ctl_be_block_lun *be_lun; 2197 struct ctl_lun_create_params *params; 2198 char num_thread_str[16]; 2199 char tmpstr[32]; 2200 char *value; 2201 int retval, num_threads; 2202 int tmp_num_threads; 2203 2204 params = &req->reqdata.create; 2205 retval = 0; 2206 req->status = CTL_LUN_OK; 2207 2208 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2209 cbe_lun = &be_lun->cbe_lun; 2210 cbe_lun->be_lun = be_lun; 2211 be_lun->params = req->reqdata.create; 2212 be_lun->softc = softc; 2213 STAILQ_INIT(&be_lun->input_queue); 2214 STAILQ_INIT(&be_lun->config_read_queue); 2215 STAILQ_INIT(&be_lun->config_write_queue); 2216 STAILQ_INIT(&be_lun->datamove_queue); 2217 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2218 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2219 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2220 ctl_init_opts(&cbe_lun->options, 2221 req->num_be_args, req->kern_be_args); 2222 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2223 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2224 if (be_lun->lun_zone == NULL) { 2225 snprintf(req->error_str, sizeof(req->error_str), 2226 "error allocating UMA zone"); 2227 goto bailout_error; 2228 } 2229 2230 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2231 cbe_lun->lun_type = params->device_type; 2232 else 2233 cbe_lun->lun_type = T_DIRECT; 2234 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2235 cbe_lun->flags = 0; 2236 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2237 if (value != NULL) { 2238 if (strcmp(value, "primary") == 0) 2239 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2240 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2241 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2242 2243 if (cbe_lun->lun_type == T_DIRECT) { 2244 be_lun->size_bytes = params->lun_size_bytes; 2245 if (params->blocksize_bytes != 0) 2246 cbe_lun->blocksize = params->blocksize_bytes; 2247 else 2248 cbe_lun->blocksize = 512; 2249 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2250 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2251 0 : (be_lun->size_blocks - 1); 2252 2253 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2254 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2255 retval = ctl_be_block_open(softc, be_lun, req); 2256 if (retval != 0) { 2257 retval = 0; 2258 req->status = CTL_LUN_WARNING; 2259 } 2260 } 2261 num_threads = cbb_num_threads; 2262 } else { 2263 num_threads = 1; 2264 } 2265 2266 /* 2267 * XXX This searching loop might be refactored to be combined with 2268 * the loop above, 2269 */ 2270 value = ctl_get_opt(&cbe_lun->options, "num_threads"); 2271 if (value != NULL) { 2272 tmp_num_threads = strtol(value, NULL, 0); 2273 2274 /* 2275 * We don't let the user specify less than one 2276 * thread, but hope he's clueful enough not to 2277 * specify 1000 threads. 2278 */ 2279 if (tmp_num_threads < 1) { 2280 snprintf(req->error_str, sizeof(req->error_str), 2281 "invalid number of threads %s", 2282 num_thread_str); 2283 goto bailout_error; 2284 } 2285 num_threads = tmp_num_threads; 2286 } 2287 2288 if (be_lun->vn == NULL) 2289 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2290 /* Tell the user the blocksize we ended up using */ 2291 params->lun_size_bytes = be_lun->size_bytes; 2292 params->blocksize_bytes = cbe_lun->blocksize; 2293 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2294 cbe_lun->req_lun_id = params->req_lun_id; 2295 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2296 } else 2297 cbe_lun->req_lun_id = 0; 2298 2299 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2300 cbe_lun->lun_config_status = ctl_be_block_lun_config_status; 2301 cbe_lun->be = &ctl_be_block_driver; 2302 2303 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2304 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2305 softc->num_luns); 2306 strncpy((char *)cbe_lun->serial_num, tmpstr, 2307 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2308 2309 /* Tell the user what we used for a serial number */ 2310 strncpy((char *)params->serial_num, tmpstr, 2311 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2312 } else { 2313 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2314 MIN(sizeof(cbe_lun->serial_num), 2315 sizeof(params->serial_num))); 2316 } 2317 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2318 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2319 strncpy((char *)cbe_lun->device_id, tmpstr, 2320 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2321 2322 /* Tell the user what we used for a device ID */ 2323 strncpy((char *)params->device_id, tmpstr, 2324 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2325 } else { 2326 strncpy((char *)cbe_lun->device_id, params->device_id, 2327 MIN(sizeof(cbe_lun->device_id), 2328 sizeof(params->device_id))); 2329 } 2330 2331 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2332 2333 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2334 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2335 2336 if (be_lun->io_taskqueue == NULL) { 2337 snprintf(req->error_str, sizeof(req->error_str), 2338 "unable to create taskqueue"); 2339 goto bailout_error; 2340 } 2341 2342 /* 2343 * Note that we start the same number of threads by default for 2344 * both the file case and the block device case. For the file 2345 * case, we need multiple threads to allow concurrency, because the 2346 * vnode interface is designed to be a blocking interface. For the 2347 * block device case, ZFS zvols at least will block the caller's 2348 * context in many instances, and so we need multiple threads to 2349 * overcome that problem. Other block devices don't need as many 2350 * threads, but they shouldn't cause too many problems. 2351 * 2352 * If the user wants to just have a single thread for a block 2353 * device, he can specify that when the LUN is created, or change 2354 * the tunable/sysctl to alter the default number of threads. 2355 */ 2356 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2357 /*num threads*/num_threads, 2358 /*priority*/PWAIT, 2359 /*thread name*/ 2360 "%s taskq", be_lun->lunname); 2361 2362 if (retval != 0) 2363 goto bailout_error; 2364 2365 be_lun->num_threads = num_threads; 2366 2367 mtx_lock(&softc->lock); 2368 softc->num_luns++; 2369 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2370 2371 mtx_unlock(&softc->lock); 2372 2373 retval = ctl_add_lun(&be_lun->cbe_lun); 2374 if (retval != 0) { 2375 mtx_lock(&softc->lock); 2376 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2377 links); 2378 softc->num_luns--; 2379 mtx_unlock(&softc->lock); 2380 snprintf(req->error_str, sizeof(req->error_str), 2381 "ctl_add_lun() returned error %d, see dmesg for " 2382 "details", retval); 2383 retval = 0; 2384 goto bailout_error; 2385 } 2386 2387 mtx_lock(&softc->lock); 2388 2389 /* 2390 * Tell the config_status routine that we're waiting so it won't 2391 * clean up the LUN in the event of an error. 2392 */ 2393 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2394 2395 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2396 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2397 if (retval == EINTR) 2398 break; 2399 } 2400 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2401 2402 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2403 snprintf(req->error_str, sizeof(req->error_str), 2404 "LUN configuration error, see dmesg for details"); 2405 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2406 links); 2407 softc->num_luns--; 2408 mtx_unlock(&softc->lock); 2409 goto bailout_error; 2410 } else { 2411 params->req_lun_id = cbe_lun->lun_id; 2412 } 2413 2414 mtx_unlock(&softc->lock); 2415 2416 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2417 cbe_lun->blocksize, 2418 DEVSTAT_ALL_SUPPORTED, 2419 cbe_lun->lun_type 2420 | DEVSTAT_TYPE_IF_OTHER, 2421 DEVSTAT_PRIORITY_OTHER); 2422 2423 return (retval); 2424 2425bailout_error: 2426 req->status = CTL_LUN_ERROR; 2427 2428 if (be_lun->io_taskqueue != NULL) 2429 taskqueue_free(be_lun->io_taskqueue); 2430 ctl_be_block_close(be_lun); 2431 if (be_lun->dev_path != NULL) 2432 free(be_lun->dev_path, M_CTLBLK); 2433 if (be_lun->lun_zone != NULL) 2434 uma_zdestroy(be_lun->lun_zone); 2435 ctl_free_opts(&cbe_lun->options); 2436 mtx_destroy(&be_lun->queue_lock); 2437 mtx_destroy(&be_lun->io_lock); 2438 free(be_lun, M_CTLBLK); 2439 2440 return (retval); 2441} 2442 2443static int 2444ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2445{ 2446 struct ctl_lun_rm_params *params; 2447 struct ctl_be_block_lun *be_lun; 2448 int retval; 2449 2450 params = &req->reqdata.rm; 2451 2452 mtx_lock(&softc->lock); 2453 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2454 if (be_lun->cbe_lun.lun_id == params->lun_id) 2455 break; 2456 } 2457 mtx_unlock(&softc->lock); 2458 2459 if (be_lun == NULL) { 2460 snprintf(req->error_str, sizeof(req->error_str), 2461 "LUN %u is not managed by the block backend", 2462 params->lun_id); 2463 goto bailout_error; 2464 } 2465 2466 retval = ctl_disable_lun(&be_lun->cbe_lun); 2467 2468 if (retval != 0) { 2469 snprintf(req->error_str, sizeof(req->error_str), 2470 "error %d returned from ctl_disable_lun() for " 2471 "LUN %d", retval, params->lun_id); 2472 goto bailout_error; 2473 2474 } 2475 2476 retval = ctl_invalidate_lun(&be_lun->cbe_lun); 2477 if (retval != 0) { 2478 snprintf(req->error_str, sizeof(req->error_str), 2479 "error %d returned from ctl_invalidate_lun() for " 2480 "LUN %d", retval, params->lun_id); 2481 goto bailout_error; 2482 } 2483 2484 mtx_lock(&softc->lock); 2485 2486 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2487 2488 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2489 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2490 if (retval == EINTR) 2491 break; 2492 } 2493 2494 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2495 2496 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2497 snprintf(req->error_str, sizeof(req->error_str), 2498 "interrupted waiting for LUN to be freed"); 2499 mtx_unlock(&softc->lock); 2500 goto bailout_error; 2501 } 2502 2503 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2504 2505 softc->num_luns--; 2506 mtx_unlock(&softc->lock); 2507 2508 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2509 2510 taskqueue_free(be_lun->io_taskqueue); 2511 2512 ctl_be_block_close(be_lun); 2513 2514 if (be_lun->disk_stats != NULL) 2515 devstat_remove_entry(be_lun->disk_stats); 2516 2517 uma_zdestroy(be_lun->lun_zone); 2518 2519 ctl_free_opts(&be_lun->cbe_lun.options); 2520 free(be_lun->dev_path, M_CTLBLK); 2521 mtx_destroy(&be_lun->queue_lock); 2522 mtx_destroy(&be_lun->io_lock); 2523 free(be_lun, M_CTLBLK); 2524 2525 req->status = CTL_LUN_OK; 2526 2527 return (0); 2528 2529bailout_error: 2530 2531 req->status = CTL_LUN_ERROR; 2532 2533 return (0); 2534} 2535 2536static int 2537ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2538 struct ctl_lun_req *req) 2539{ 2540 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2541 struct vattr vattr; 2542 int error; 2543 struct ctl_lun_create_params *params = &be_lun->params; 2544 2545 if (params->lun_size_bytes != 0) { 2546 be_lun->size_bytes = params->lun_size_bytes; 2547 } else { 2548 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2549 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2550 VOP_UNLOCK(be_lun->vn, 0); 2551 if (error != 0) { 2552 snprintf(req->error_str, sizeof(req->error_str), 2553 "error calling VOP_GETATTR() for file %s", 2554 be_lun->dev_path); 2555 return (error); 2556 } 2557 be_lun->size_bytes = vattr.va_size; 2558 } 2559 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2560 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2561 0 : (be_lun->size_blocks - 1); 2562 return (0); 2563} 2564 2565static int 2566ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2567 struct ctl_lun_req *req) 2568{ 2569 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2570 struct ctl_be_block_devdata *dev_data; 2571 int error; 2572 struct ctl_lun_create_params *params = &be_lun->params; 2573 uint64_t size_bytes; 2574 2575 dev_data = &be_lun->backend.dev; 2576 if (!dev_data->csw->d_ioctl) { 2577 snprintf(req->error_str, sizeof(req->error_str), 2578 "no d_ioctl for device %s!", be_lun->dev_path); 2579 return (ENODEV); 2580 } 2581 2582 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2583 (caddr_t)&size_bytes, FREAD, 2584 curthread); 2585 if (error) { 2586 snprintf(req->error_str, sizeof(req->error_str), 2587 "error %d returned for DIOCGMEDIASIZE ioctl " 2588 "on %s!", error, be_lun->dev_path); 2589 return (error); 2590 } 2591 2592 if (params->lun_size_bytes != 0) { 2593 if (params->lun_size_bytes > size_bytes) { 2594 snprintf(req->error_str, sizeof(req->error_str), 2595 "requested LUN size %ju > backing device " 2596 "size %ju", 2597 (uintmax_t)params->lun_size_bytes, 2598 (uintmax_t)size_bytes); 2599 return (EINVAL); 2600 } 2601 be_lun->size_bytes = params->lun_size_bytes; 2602 } else { 2603 be_lun->size_bytes = size_bytes; 2604 } 2605 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2606 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2607 0 : (be_lun->size_blocks - 1); 2608 return (0); 2609} 2610 2611static int 2612ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2613{ 2614 struct ctl_lun_modify_params *params; 2615 struct ctl_be_block_lun *be_lun; 2616 struct ctl_be_lun *cbe_lun; 2617 char *value; 2618 uint64_t oldsize; 2619 int error, wasprim; 2620 2621 params = &req->reqdata.modify; 2622 2623 mtx_lock(&softc->lock); 2624 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2625 if (be_lun->cbe_lun.lun_id == params->lun_id) 2626 break; 2627 } 2628 mtx_unlock(&softc->lock); 2629 2630 if (be_lun == NULL) { 2631 snprintf(req->error_str, sizeof(req->error_str), 2632 "LUN %u is not managed by the block backend", 2633 params->lun_id); 2634 goto bailout_error; 2635 } 2636 cbe_lun = &be_lun->cbe_lun; 2637 2638 if (params->lun_size_bytes != 0) 2639 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2640 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); 2641 2642 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2643 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2644 if (value != NULL) { 2645 if (strcmp(value, "primary") == 0) 2646 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2647 else 2648 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2649 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2650 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2651 else 2652 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2653 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2654 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2655 ctl_lun_primary(cbe_lun); 2656 else 2657 ctl_lun_secondary(cbe_lun); 2658 } 2659 2660 oldsize = be_lun->size_blocks; 2661 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2662 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2663 if (be_lun->vn == NULL) 2664 error = ctl_be_block_open(softc, be_lun, req); 2665 else if (vn_isdisk(be_lun->vn, &error)) 2666 error = ctl_be_block_modify_dev(be_lun, req); 2667 else if (be_lun->vn->v_type == VREG) 2668 error = ctl_be_block_modify_file(be_lun, req); 2669 else 2670 error = EINVAL; 2671 if ((cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) && 2672 be_lun->vn != NULL) { 2673 cbe_lun->flags &= ~CTL_LUN_FLAG_OFFLINE; 2674 ctl_lun_online(cbe_lun); 2675 } 2676 } else { 2677 if (be_lun->vn != NULL) { 2678 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2679 ctl_lun_offline(cbe_lun); 2680 pause("CTL LUN offline", hz / 8); // XXX 2681 error = ctl_be_block_close(be_lun); 2682 } else 2683 error = 0; 2684 } 2685 if (be_lun->size_blocks != oldsize) 2686 ctl_lun_capacity_changed(cbe_lun); 2687 2688 /* Tell the user the exact size we ended up using */ 2689 params->lun_size_bytes = be_lun->size_bytes; 2690 2691 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2692 return (0); 2693 2694bailout_error: 2695 req->status = CTL_LUN_ERROR; 2696 return (0); 2697} 2698 2699static void 2700ctl_be_block_lun_shutdown(void *be_lun) 2701{ 2702 struct ctl_be_block_lun *lun; 2703 struct ctl_be_block_softc *softc; 2704 2705 lun = (struct ctl_be_block_lun *)be_lun; 2706 2707 softc = lun->softc; 2708 2709 mtx_lock(&softc->lock); 2710 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2711 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2712 wakeup(lun); 2713 mtx_unlock(&softc->lock); 2714 2715} 2716 2717static void 2718ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2719{ 2720 struct ctl_be_block_lun *lun; 2721 struct ctl_be_block_softc *softc; 2722 2723 lun = (struct ctl_be_block_lun *)be_lun; 2724 softc = lun->softc; 2725 2726 if (status == CTL_LUN_CONFIG_OK) { 2727 mtx_lock(&softc->lock); 2728 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2729 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2730 wakeup(lun); 2731 mtx_unlock(&softc->lock); 2732 2733 /* 2734 * We successfully added the LUN, attempt to enable it. 2735 */ 2736 if (ctl_enable_lun(&lun->cbe_lun) != 0) { 2737 printf("%s: ctl_enable_lun() failed!\n", __func__); 2738 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { 2739 printf("%s: ctl_invalidate_lun() failed!\n", 2740 __func__); 2741 } 2742 } 2743 2744 return; 2745 } 2746 2747 2748 mtx_lock(&softc->lock); 2749 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2750 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2751 wakeup(lun); 2752 mtx_unlock(&softc->lock); 2753} 2754 2755 2756static int 2757ctl_be_block_config_write(union ctl_io *io) 2758{ 2759 struct ctl_be_block_lun *be_lun; 2760 struct ctl_be_lun *cbe_lun; 2761 int retval; 2762 2763 retval = 0; 2764 2765 DPRINTF("entered\n"); 2766 2767 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2768 CTL_PRIV_BACKEND_LUN].ptr; 2769 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2770 2771 switch (io->scsiio.cdb[0]) { 2772 case SYNCHRONIZE_CACHE: 2773 case SYNCHRONIZE_CACHE_16: 2774 case WRITE_SAME_10: 2775 case WRITE_SAME_16: 2776 case UNMAP: 2777 /* 2778 * The upper level CTL code will filter out any CDBs with 2779 * the immediate bit set and return the proper error. 2780 * 2781 * We don't really need to worry about what LBA range the 2782 * user asked to be synced out. When they issue a sync 2783 * cache command, we'll sync out the whole thing. 2784 */ 2785 mtx_lock(&be_lun->queue_lock); 2786 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2787 links); 2788 mtx_unlock(&be_lun->queue_lock); 2789 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2790 break; 2791 case START_STOP_UNIT: { 2792 struct scsi_start_stop_unit *cdb; 2793 2794 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2795 2796 if (cdb->how & SSS_START) 2797 retval = ctl_start_lun(cbe_lun); 2798 else { 2799 retval = ctl_stop_lun(cbe_lun); 2800 /* 2801 * XXX KDM Copan-specific offline behavior. 2802 * Figure out a reasonable way to port this? 2803 */ 2804#ifdef NEEDTOPORT 2805 if ((retval == 0) 2806 && (cdb->byte2 & SSS_ONOFFLINE)) 2807 retval = ctl_lun_offline(cbe_lun); 2808#endif 2809 } 2810 2811 /* 2812 * In general, the above routines should not fail. They 2813 * just set state for the LUN. So we've got something 2814 * pretty wrong here if we can't start or stop the LUN. 2815 */ 2816 if (retval != 0) { 2817 ctl_set_internal_failure(&io->scsiio, 2818 /*sks_valid*/ 1, 2819 /*retry_count*/ 0xf051); 2820 retval = CTL_RETVAL_COMPLETE; 2821 } else { 2822 ctl_set_success(&io->scsiio); 2823 } 2824 ctl_config_write_done(io); 2825 break; 2826 } 2827 default: 2828 ctl_set_invalid_opcode(&io->scsiio); 2829 ctl_config_write_done(io); 2830 retval = CTL_RETVAL_COMPLETE; 2831 break; 2832 } 2833 2834 return (retval); 2835} 2836 2837static int 2838ctl_be_block_config_read(union ctl_io *io) 2839{ 2840 struct ctl_be_block_lun *be_lun; 2841 struct ctl_be_lun *cbe_lun; 2842 int retval = 0; 2843 2844 DPRINTF("entered\n"); 2845 2846 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2847 CTL_PRIV_BACKEND_LUN].ptr; 2848 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2849 2850 switch (io->scsiio.cdb[0]) { 2851 case SERVICE_ACTION_IN: 2852 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2853 mtx_lock(&be_lun->queue_lock); 2854 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2855 &io->io_hdr, links); 2856 mtx_unlock(&be_lun->queue_lock); 2857 taskqueue_enqueue(be_lun->io_taskqueue, 2858 &be_lun->io_task); 2859 retval = CTL_RETVAL_QUEUED; 2860 break; 2861 } 2862 ctl_set_invalid_field(&io->scsiio, 2863 /*sks_valid*/ 1, 2864 /*command*/ 1, 2865 /*field*/ 1, 2866 /*bit_valid*/ 1, 2867 /*bit*/ 4); 2868 ctl_config_read_done(io); 2869 retval = CTL_RETVAL_COMPLETE; 2870 break; 2871 default: 2872 ctl_set_invalid_opcode(&io->scsiio); 2873 ctl_config_read_done(io); 2874 retval = CTL_RETVAL_COMPLETE; 2875 break; 2876 } 2877 2878 return (retval); 2879} 2880 2881static int 2882ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2883{ 2884 struct ctl_be_block_lun *lun; 2885 int retval; 2886 2887 lun = (struct ctl_be_block_lun *)be_lun; 2888 retval = 0; 2889 2890 retval = sbuf_printf(sb, "\t<num_threads>"); 2891 2892 if (retval != 0) 2893 goto bailout; 2894 2895 retval = sbuf_printf(sb, "%d", lun->num_threads); 2896 2897 if (retval != 0) 2898 goto bailout; 2899 2900 retval = sbuf_printf(sb, "</num_threads>\n"); 2901 2902bailout: 2903 2904 return (retval); 2905} 2906 2907static uint64_t 2908ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2909{ 2910 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2911 2912 if (lun->getattr == NULL) 2913 return (UINT64_MAX); 2914 return (lun->getattr(lun, attrname)); 2915} 2916 2917int 2918ctl_be_block_init(void) 2919{ 2920 struct ctl_be_block_softc *softc; 2921 int retval; 2922 2923 softc = &backend_block_softc; 2924 retval = 0; 2925 2926 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2927 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2928 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2929 STAILQ_INIT(&softc->lun_list); 2930 2931 return (retval); 2932} 2933