ctl_backend_block.c revision 286931
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 286931 2015-08-19 17:47:47Z 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_frontend_internal.h> 90#include <cam/ctl/ctl_ioctl.h> 91#include <cam/ctl/ctl_scsi_all.h> 92#include <cam/ctl/ctl_error.h> 93 94/* 95 * The idea here is that we'll allocate enough S/G space to hold a 1MB 96 * I/O. If we get an I/O larger than that, we'll split it. 97 */ 98#define CTLBLK_HALF_IO_SIZE (512 * 1024) 99#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 100#define CTLBLK_MAX_SEG MAXPHYS 101#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 102#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 103 104#ifdef CTLBLK_DEBUG 105#define DPRINTF(fmt, args...) \ 106 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 107#else 108#define DPRINTF(fmt, args...) do {} while(0) 109#endif 110 111#define PRIV(io) \ 112 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 113#define ARGS(io) \ 114 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 115 116SDT_PROVIDER_DEFINE(cbb); 117 118typedef enum { 119 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 120 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 121 CTL_BE_BLOCK_LUN_WAITING = 0x04, 122 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 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 struct ctl_block_disk *disk; 161 char lunname[32]; 162 char *dev_path; 163 ctl_be_block_type dev_type; 164 struct vnode *vn; 165 union ctl_be_block_bedata backend; 166 cbb_dispatch_t dispatch; 167 cbb_dispatch_t lun_flush; 168 cbb_dispatch_t unmap; 169 cbb_dispatch_t get_lba_status; 170 cbb_getattr_t getattr; 171 uma_zone_t lun_zone; 172 uint64_t size_blocks; 173 uint64_t size_bytes; 174 uint32_t blocksize; 175 int blocksize_shift; 176 uint16_t pblockexp; 177 uint16_t pblockoff; 178 uint16_t ublockexp; 179 uint16_t ublockoff; 180 uint32_t atomicblock; 181 uint32_t opttxferlen; 182 struct ctl_be_block_softc *softc; 183 struct devstat *disk_stats; 184 ctl_be_block_lun_flags flags; 185 STAILQ_ENTRY(ctl_be_block_lun) links; 186 struct ctl_be_lun ctl_be_lun; 187 struct taskqueue *io_taskqueue; 188 struct task io_task; 189 int num_threads; 190 STAILQ_HEAD(, ctl_io_hdr) input_queue; 191 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 192 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 193 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 194 struct mtx_padalign io_lock; 195 struct mtx_padalign queue_lock; 196}; 197 198/* 199 * Overall softc structure for the block backend module. 200 */ 201struct ctl_be_block_softc { 202 struct mtx lock; 203 int num_disks; 204 STAILQ_HEAD(, ctl_block_disk) disk_list; 205 int num_luns; 206 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 207}; 208 209static struct ctl_be_block_softc backend_block_softc; 210 211/* 212 * Per-I/O information. 213 */ 214struct ctl_be_block_io { 215 union ctl_io *io; 216 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 217 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 218 int bio_cmd; 219 int num_segs; 220 int num_bios_sent; 221 int num_bios_done; 222 int send_complete; 223 int num_errors; 224 struct bintime ds_t0; 225 devstat_tag_type ds_tag_type; 226 devstat_trans_flags ds_trans_type; 227 uint64_t io_len; 228 uint64_t io_offset; 229 int io_arg; 230 struct ctl_be_block_softc *softc; 231 struct ctl_be_block_lun *lun; 232 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 233}; 234 235static int cbb_num_threads = 14; 236TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 237SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 238 "CAM Target Layer Block Backend"); 239SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 240 &cbb_num_threads, 0, "Number of threads per backing file"); 241 242static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 243static void ctl_free_beio(struct ctl_be_block_io *beio); 244static void ctl_complete_beio(struct ctl_be_block_io *beio); 245static int ctl_be_block_move_done(union ctl_io *io); 246static void ctl_be_block_biodone(struct bio *bio); 247static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 248 struct ctl_be_block_io *beio); 249static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 250 struct ctl_be_block_io *beio); 251static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 252 struct ctl_be_block_io *beio); 253static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 254 const char *attrname); 255static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 256 struct ctl_be_block_io *beio); 257static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 258 struct ctl_be_block_io *beio); 259static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 260 struct ctl_be_block_io *beio); 261static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 262 const char *attrname); 263static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 264 union ctl_io *io); 265static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 266 union ctl_io *io); 267static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 268 union ctl_io *io); 269static void ctl_be_block_worker(void *context, int pending); 270static int ctl_be_block_submit(union ctl_io *io); 271static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 272 int flag, struct thread *td); 273static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 274 struct ctl_lun_req *req); 275static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 276 struct ctl_lun_req *req); 277static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 278static int ctl_be_block_open(struct ctl_be_block_softc *softc, 279 struct ctl_be_block_lun *be_lun, 280 struct ctl_lun_req *req); 281static int ctl_be_block_create(struct ctl_be_block_softc *softc, 282 struct ctl_lun_req *req); 283static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 284 struct ctl_lun_req *req); 285static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 286 struct ctl_lun_req *req); 287static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 288 struct ctl_lun_req *req); 289static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 290 struct ctl_lun_req *req); 291static void ctl_be_block_lun_shutdown(void *be_lun); 292static void ctl_be_block_lun_config_status(void *be_lun, 293 ctl_lun_config_status status); 294static int ctl_be_block_config_write(union ctl_io *io); 295static int ctl_be_block_config_read(union ctl_io *io); 296static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 297static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 298int ctl_be_block_init(void); 299 300static struct ctl_backend_driver ctl_be_block_driver = 301{ 302 .name = "block", 303 .flags = CTL_BE_FLAG_HAS_CONFIG, 304 .init = ctl_be_block_init, 305 .data_submit = ctl_be_block_submit, 306 .data_move_done = ctl_be_block_move_done, 307 .config_read = ctl_be_block_config_read, 308 .config_write = ctl_be_block_config_write, 309 .ioctl = ctl_be_block_ioctl, 310 .lun_info = ctl_be_block_lun_info, 311 .lun_attr = ctl_be_block_lun_attr 312}; 313 314MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 315CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 316 317static uma_zone_t beio_zone; 318 319static struct ctl_be_block_io * 320ctl_alloc_beio(struct ctl_be_block_softc *softc) 321{ 322 struct ctl_be_block_io *beio; 323 324 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 325 beio->softc = softc; 326 return (beio); 327} 328 329static void 330ctl_free_beio(struct ctl_be_block_io *beio) 331{ 332 int duplicate_free; 333 int i; 334 335 duplicate_free = 0; 336 337 for (i = 0; i < beio->num_segs; i++) { 338 if (beio->sg_segs[i].addr == NULL) 339 duplicate_free++; 340 341 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 342 beio->sg_segs[i].addr = NULL; 343 344 /* For compare we had two equal S/G lists. */ 345 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 346 uma_zfree(beio->lun->lun_zone, 347 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 348 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 349 } 350 } 351 352 if (duplicate_free > 0) { 353 printf("%s: %d duplicate frees out of %d segments\n", __func__, 354 duplicate_free, beio->num_segs); 355 } 356 357 uma_zfree(beio_zone, beio); 358} 359 360static void 361ctl_complete_beio(struct ctl_be_block_io *beio) 362{ 363 union ctl_io *io = beio->io; 364 365 if (beio->beio_cont != NULL) { 366 beio->beio_cont(beio); 367 } else { 368 ctl_free_beio(beio); 369 ctl_data_submit_done(io); 370 } 371} 372 373static int 374ctl_be_block_move_done(union ctl_io *io) 375{ 376 struct ctl_be_block_io *beio; 377 struct ctl_be_block_lun *be_lun; 378 struct ctl_lba_len_flags *lbalen; 379#ifdef CTL_TIME_IO 380 struct bintime cur_bt; 381#endif 382 int i; 383 384 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 385 be_lun = beio->lun; 386 387 DPRINTF("entered\n"); 388 389#ifdef CTL_TIME_IO 390 getbintime(&cur_bt); 391 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 392 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 393 io->io_hdr.num_dmas++; 394#endif 395 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 396 397 /* 398 * We set status at this point for read commands, and write 399 * commands with errors. 400 */ 401 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 402 ; 403 } else if ((io->io_hdr.port_status == 0) && 404 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 405 lbalen = ARGS(beio->io); 406 if (lbalen->flags & CTL_LLF_READ) { 407 ctl_set_success(&io->scsiio); 408 } else if (lbalen->flags & CTL_LLF_COMPARE) { 409 /* We have two data blocks ready for comparison. */ 410 for (i = 0; i < beio->num_segs; i++) { 411 if (memcmp(beio->sg_segs[i].addr, 412 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 413 beio->sg_segs[i].len) != 0) 414 break; 415 } 416 if (i < beio->num_segs) 417 ctl_set_sense(&io->scsiio, 418 /*current_error*/ 1, 419 /*sense_key*/ SSD_KEY_MISCOMPARE, 420 /*asc*/ 0x1D, 421 /*ascq*/ 0x00, 422 SSD_ELEM_NONE); 423 else 424 ctl_set_success(&io->scsiio); 425 } 426 } else if ((io->io_hdr.port_status != 0) && 427 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 428 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 429 /* 430 * For hardware error sense keys, the sense key 431 * specific value is defined to be a retry count, 432 * but we use it to pass back an internal FETD 433 * error code. XXX KDM Hopefully the FETD is only 434 * using 16 bits for an error code, since that's 435 * all the space we have in the sks field. 436 */ 437 ctl_set_internal_failure(&io->scsiio, 438 /*sks_valid*/ 1, 439 /*retry_count*/ 440 io->io_hdr.port_status); 441 } 442 443 /* 444 * If this is a read, or a write with errors, it is done. 445 */ 446 if ((beio->bio_cmd == BIO_READ) 447 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 448 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 449 ctl_complete_beio(beio); 450 return (0); 451 } 452 453 /* 454 * At this point, we have a write and the DMA completed 455 * successfully. We now have to queue it to the task queue to 456 * execute the backend I/O. That is because we do blocking 457 * memory allocations, and in the file backing case, blocking I/O. 458 * This move done routine is generally called in the SIM's 459 * interrupt context, and therefore we cannot block. 460 */ 461 mtx_lock(&be_lun->queue_lock); 462 /* 463 * XXX KDM make sure that links is okay to use at this point. 464 * Otherwise, we either need to add another field to ctl_io_hdr, 465 * or deal with resource allocation here. 466 */ 467 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 468 mtx_unlock(&be_lun->queue_lock); 469 470 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 471 472 return (0); 473} 474 475static void 476ctl_be_block_biodone(struct bio *bio) 477{ 478 struct ctl_be_block_io *beio; 479 struct ctl_be_block_lun *be_lun; 480 union ctl_io *io; 481 int error; 482 483 beio = bio->bio_caller1; 484 be_lun = beio->lun; 485 io = beio->io; 486 487 DPRINTF("entered\n"); 488 489 error = bio->bio_error; 490 mtx_lock(&be_lun->io_lock); 491 if (error != 0) 492 beio->num_errors++; 493 494 beio->num_bios_done++; 495 496 /* 497 * XXX KDM will this cause WITNESS to complain? Holding a lock 498 * during the free might cause it to complain. 499 */ 500 g_destroy_bio(bio); 501 502 /* 503 * If the send complete bit isn't set, or we aren't the last I/O to 504 * complete, then we're done. 505 */ 506 if ((beio->send_complete == 0) 507 || (beio->num_bios_done < beio->num_bios_sent)) { 508 mtx_unlock(&be_lun->io_lock); 509 return; 510 } 511 512 /* 513 * At this point, we've verified that we are the last I/O to 514 * complete, so it's safe to drop the lock. 515 */ 516 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 517 beio->ds_tag_type, beio->ds_trans_type, 518 /*now*/ NULL, /*then*/&beio->ds_t0); 519 mtx_unlock(&be_lun->io_lock); 520 521 /* 522 * If there are any errors from the backing device, we fail the 523 * entire I/O with a medium error. 524 */ 525 if (beio->num_errors > 0) { 526 if (error == EOPNOTSUPP) { 527 ctl_set_invalid_opcode(&io->scsiio); 528 } else if (error == ENOSPC || error == EDQUOT) { 529 ctl_set_space_alloc_fail(&io->scsiio); 530 } else if (beio->bio_cmd == BIO_FLUSH) { 531 /* XXX KDM is there is a better error here? */ 532 ctl_set_internal_failure(&io->scsiio, 533 /*sks_valid*/ 1, 534 /*retry_count*/ 0xbad2); 535 } else 536 ctl_set_medium_error(&io->scsiio); 537 ctl_complete_beio(beio); 538 return; 539 } 540 541 /* 542 * If this is a write, a flush, a delete or verify, we're all done. 543 * If this is a read, we can now send the data to the user. 544 */ 545 if ((beio->bio_cmd == BIO_WRITE) 546 || (beio->bio_cmd == BIO_FLUSH) 547 || (beio->bio_cmd == BIO_DELETE) 548 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 549 ctl_set_success(&io->scsiio); 550 ctl_complete_beio(beio); 551 } else { 552 if ((ARGS(io)->flags & CTL_LLF_READ) && 553 beio->beio_cont == NULL) 554 ctl_set_success(&io->scsiio); 555#ifdef CTL_TIME_IO 556 getbintime(&io->io_hdr.dma_start_bt); 557#endif 558 ctl_datamove(io); 559 } 560} 561 562static void 563ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 564 struct ctl_be_block_io *beio) 565{ 566 union ctl_io *io = beio->io; 567 struct mount *mountpoint; 568 int error, lock_flags; 569 570 DPRINTF("entered\n"); 571 572 binuptime(&beio->ds_t0); 573 mtx_lock(&be_lun->io_lock); 574 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 575 mtx_unlock(&be_lun->io_lock); 576 577 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 578 579 if (MNT_SHARED_WRITES(mountpoint) 580 || ((mountpoint == NULL) 581 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 582 lock_flags = LK_SHARED; 583 else 584 lock_flags = LK_EXCLUSIVE; 585 586 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 587 588 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 589 curthread); 590 VOP_UNLOCK(be_lun->vn, 0); 591 592 vn_finished_write(mountpoint); 593 594 mtx_lock(&be_lun->io_lock); 595 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 596 beio->ds_tag_type, beio->ds_trans_type, 597 /*now*/ NULL, /*then*/&beio->ds_t0); 598 mtx_unlock(&be_lun->io_lock); 599 600 if (error == 0) 601 ctl_set_success(&io->scsiio); 602 else { 603 /* XXX KDM is there is a better error here? */ 604 ctl_set_internal_failure(&io->scsiio, 605 /*sks_valid*/ 1, 606 /*retry_count*/ 0xbad1); 607 } 608 609 ctl_complete_beio(beio); 610} 611 612SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 613SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 614SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 615SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 616 617static void 618ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 619 struct ctl_be_block_io *beio) 620{ 621 struct ctl_be_block_filedata *file_data; 622 union ctl_io *io; 623 struct uio xuio; 624 struct iovec *xiovec; 625 int flags; 626 int error, i; 627 628 DPRINTF("entered\n"); 629 630 file_data = &be_lun->backend.file; 631 io = beio->io; 632 flags = 0; 633 if (ARGS(io)->flags & CTL_LLF_DPO) 634 flags |= IO_DIRECT; 635 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 636 flags |= IO_SYNC; 637 638 bzero(&xuio, sizeof(xuio)); 639 if (beio->bio_cmd == BIO_READ) { 640 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 641 xuio.uio_rw = UIO_READ; 642 } else { 643 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 644 xuio.uio_rw = UIO_WRITE; 645 } 646 xuio.uio_offset = beio->io_offset; 647 xuio.uio_resid = beio->io_len; 648 xuio.uio_segflg = UIO_SYSSPACE; 649 xuio.uio_iov = beio->xiovecs; 650 xuio.uio_iovcnt = beio->num_segs; 651 xuio.uio_td = curthread; 652 653 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 654 xiovec->iov_base = beio->sg_segs[i].addr; 655 xiovec->iov_len = beio->sg_segs[i].len; 656 } 657 658 binuptime(&beio->ds_t0); 659 mtx_lock(&be_lun->io_lock); 660 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 661 mtx_unlock(&be_lun->io_lock); 662 663 if (beio->bio_cmd == BIO_READ) { 664 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 665 666 /* 667 * UFS pays attention to IO_DIRECT for reads. If the 668 * DIRECTIO option is configured into the kernel, it calls 669 * ffs_rawread(). But that only works for single-segment 670 * uios with user space addresses. In our case, with a 671 * kernel uio, it still reads into the buffer cache, but it 672 * will just try to release the buffer from the cache later 673 * on in ffs_read(). 674 * 675 * ZFS does not pay attention to IO_DIRECT for reads. 676 * 677 * UFS does not pay attention to IO_SYNC for reads. 678 * 679 * ZFS pays attention to IO_SYNC (which translates into the 680 * Solaris define FRSYNC for zfs_read()) for reads. It 681 * attempts to sync the file before reading. 682 */ 683 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 684 685 VOP_UNLOCK(be_lun->vn, 0); 686 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 687 } else { 688 struct mount *mountpoint; 689 int lock_flags; 690 691 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 692 693 if (MNT_SHARED_WRITES(mountpoint) 694 || ((mountpoint == NULL) 695 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 696 lock_flags = LK_SHARED; 697 else 698 lock_flags = LK_EXCLUSIVE; 699 700 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 701 702 /* 703 * UFS pays attention to IO_DIRECT for writes. The write 704 * is done asynchronously. (Normally the write would just 705 * get put into cache. 706 * 707 * UFS pays attention to IO_SYNC for writes. It will 708 * attempt to write the buffer out synchronously if that 709 * flag is set. 710 * 711 * ZFS does not pay attention to IO_DIRECT for writes. 712 * 713 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 714 * for writes. It will flush the transaction from the 715 * cache before returning. 716 */ 717 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 718 VOP_UNLOCK(be_lun->vn, 0); 719 720 vn_finished_write(mountpoint); 721 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 722 } 723 724 mtx_lock(&be_lun->io_lock); 725 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 726 beio->ds_tag_type, beio->ds_trans_type, 727 /*now*/ NULL, /*then*/&beio->ds_t0); 728 mtx_unlock(&be_lun->io_lock); 729 730 /* 731 * If we got an error, set the sense data to "MEDIUM ERROR" and 732 * return the I/O to the user. 733 */ 734 if (error != 0) { 735 char path_str[32]; 736 737 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 738 printf("%s%s command returned errno %d\n", path_str, 739 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 740 if (error == ENOSPC || error == EDQUOT) { 741 ctl_set_space_alloc_fail(&io->scsiio); 742 } else 743 ctl_set_medium_error(&io->scsiio); 744 ctl_complete_beio(beio); 745 return; 746 } 747 748 /* 749 * If this is a write or a verify, we're all done. 750 * If this is a read, we can now send the data to the user. 751 */ 752 if ((beio->bio_cmd == BIO_WRITE) || 753 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 754 ctl_set_success(&io->scsiio); 755 ctl_complete_beio(beio); 756 } else { 757 if ((ARGS(io)->flags & CTL_LLF_READ) && 758 beio->beio_cont == NULL) 759 ctl_set_success(&io->scsiio); 760#ifdef CTL_TIME_IO 761 getbintime(&io->io_hdr.dma_start_bt); 762#endif 763 ctl_datamove(io); 764 } 765} 766 767static void 768ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 769 struct ctl_be_block_io *beio) 770{ 771 union ctl_io *io = beio->io; 772 struct ctl_lba_len_flags *lbalen = ARGS(io); 773 struct scsi_get_lba_status_data *data; 774 off_t roff, off; 775 int error, status; 776 777 DPRINTF("entered\n"); 778 779 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 780 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 781 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 782 0, curthread->td_ucred, curthread); 783 if (error == 0 && off > roff) 784 status = 0; /* mapped up to off */ 785 else { 786 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 787 0, curthread->td_ucred, curthread); 788 if (error == 0 && off > roff) 789 status = 1; /* deallocated up to off */ 790 else { 791 status = 0; /* unknown up to the end */ 792 off = be_lun->size_bytes; 793 } 794 } 795 VOP_UNLOCK(be_lun->vn, 0); 796 797 off >>= be_lun->blocksize_shift; 798 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 799 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 800 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 801 data->descr[0].length); 802 data->descr[0].status = status; 803 804 ctl_complete_beio(beio); 805} 806 807static uint64_t 808ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 809{ 810 struct vattr vattr; 811 struct statfs statfs; 812 int error; 813 814 if (be_lun->vn == NULL) 815 return (UINT64_MAX); 816 if (strcmp(attrname, "blocksused") == 0) { 817 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 818 if (error != 0) 819 return (UINT64_MAX); 820 return (vattr.va_bytes >> be_lun->blocksize_shift); 821 } 822 if (strcmp(attrname, "blocksavail") == 0) { 823 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 824 if (error != 0) 825 return (UINT64_MAX); 826 return ((statfs.f_bavail * statfs.f_bsize) >> 827 be_lun->blocksize_shift); 828 } 829 return (UINT64_MAX); 830} 831 832static void 833ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 834 struct ctl_be_block_io *beio) 835{ 836 struct ctl_be_block_devdata *dev_data; 837 union ctl_io *io; 838 struct uio xuio; 839 struct iovec *xiovec; 840 int flags; 841 int error, i; 842 843 DPRINTF("entered\n"); 844 845 dev_data = &be_lun->backend.dev; 846 io = beio->io; 847 flags = 0; 848 if (ARGS(io)->flags & CTL_LLF_DPO) 849 flags |= IO_DIRECT; 850 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 851 flags |= IO_SYNC; 852 853 bzero(&xuio, sizeof(xuio)); 854 if (beio->bio_cmd == BIO_READ) { 855 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 856 xuio.uio_rw = UIO_READ; 857 } else { 858 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 859 xuio.uio_rw = UIO_WRITE; 860 } 861 xuio.uio_offset = beio->io_offset; 862 xuio.uio_resid = beio->io_len; 863 xuio.uio_segflg = UIO_SYSSPACE; 864 xuio.uio_iov = beio->xiovecs; 865 xuio.uio_iovcnt = beio->num_segs; 866 xuio.uio_td = curthread; 867 868 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 869 xiovec->iov_base = beio->sg_segs[i].addr; 870 xiovec->iov_len = beio->sg_segs[i].len; 871 } 872 873 binuptime(&beio->ds_t0); 874 mtx_lock(&be_lun->io_lock); 875 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 876 mtx_unlock(&be_lun->io_lock); 877 878 if (beio->bio_cmd == BIO_READ) { 879 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 880 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 881 } else { 882 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 883 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 884 } 885 886 mtx_lock(&be_lun->io_lock); 887 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 888 beio->ds_tag_type, beio->ds_trans_type, 889 /*now*/ NULL, /*then*/&beio->ds_t0); 890 mtx_unlock(&be_lun->io_lock); 891 892 /* 893 * If we got an error, set the sense data to "MEDIUM ERROR" and 894 * return the I/O to the user. 895 */ 896 if (error != 0) { 897 if (error == ENOSPC || error == EDQUOT) { 898 ctl_set_space_alloc_fail(&io->scsiio); 899 } else 900 ctl_set_medium_error(&io->scsiio); 901 ctl_complete_beio(beio); 902 return; 903 } 904 905 /* 906 * If this is a write or a verify, we're all done. 907 * If this is a read, we can now send the data to the user. 908 */ 909 if ((beio->bio_cmd == BIO_WRITE) || 910 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 911 ctl_set_success(&io->scsiio); 912 ctl_complete_beio(beio); 913 } else { 914 if ((ARGS(io)->flags & CTL_LLF_READ) && 915 beio->beio_cont == NULL) 916 ctl_set_success(&io->scsiio); 917#ifdef CTL_TIME_IO 918 getbintime(&io->io_hdr.dma_start_bt); 919#endif 920 ctl_datamove(io); 921 } 922} 923 924static void 925ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 926 struct ctl_be_block_io *beio) 927{ 928 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 929 union ctl_io *io = beio->io; 930 struct ctl_lba_len_flags *lbalen = ARGS(io); 931 struct scsi_get_lba_status_data *data; 932 off_t roff, off; 933 int error, status; 934 935 DPRINTF("entered\n"); 936 937 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 938 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE, 939 (caddr_t)&off, FREAD, curthread); 940 if (error == 0 && off > roff) 941 status = 0; /* mapped up to off */ 942 else { 943 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA, 944 (caddr_t)&off, FREAD, curthread); 945 if (error == 0 && off > roff) 946 status = 1; /* deallocated up to off */ 947 else { 948 status = 0; /* unknown up to the end */ 949 off = be_lun->size_bytes; 950 } 951 } 952 953 off >>= be_lun->blocksize_shift; 954 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 955 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 956 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 957 data->descr[0].length); 958 data->descr[0].status = status; 959 960 ctl_complete_beio(beio); 961} 962 963static void 964ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 965 struct ctl_be_block_io *beio) 966{ 967 struct bio *bio; 968 union ctl_io *io; 969 struct ctl_be_block_devdata *dev_data; 970 971 dev_data = &be_lun->backend.dev; 972 io = beio->io; 973 974 DPRINTF("entered\n"); 975 976 /* This can't fail, it's a blocking allocation. */ 977 bio = g_alloc_bio(); 978 979 bio->bio_cmd = BIO_FLUSH; 980 bio->bio_dev = dev_data->cdev; 981 bio->bio_offset = 0; 982 bio->bio_data = 0; 983 bio->bio_done = ctl_be_block_biodone; 984 bio->bio_caller1 = beio; 985 bio->bio_pblkno = 0; 986 987 /* 988 * We don't need to acquire the LUN lock here, because we are only 989 * sending one bio, and so there is no other context to synchronize 990 * with. 991 */ 992 beio->num_bios_sent = 1; 993 beio->send_complete = 1; 994 995 binuptime(&beio->ds_t0); 996 mtx_lock(&be_lun->io_lock); 997 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 998 mtx_unlock(&be_lun->io_lock); 999 1000 (*dev_data->csw->d_strategy)(bio); 1001} 1002 1003static void 1004ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1005 struct ctl_be_block_io *beio, 1006 uint64_t off, uint64_t len, int last) 1007{ 1008 struct bio *bio; 1009 struct ctl_be_block_devdata *dev_data; 1010 uint64_t maxlen; 1011 1012 dev_data = &be_lun->backend.dev; 1013 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 1014 while (len > 0) { 1015 bio = g_alloc_bio(); 1016 bio->bio_cmd = BIO_DELETE; 1017 bio->bio_dev = dev_data->cdev; 1018 bio->bio_offset = off; 1019 bio->bio_length = MIN(len, maxlen); 1020 bio->bio_data = 0; 1021 bio->bio_done = ctl_be_block_biodone; 1022 bio->bio_caller1 = beio; 1023 bio->bio_pblkno = off / be_lun->blocksize; 1024 1025 off += bio->bio_length; 1026 len -= bio->bio_length; 1027 1028 mtx_lock(&be_lun->io_lock); 1029 beio->num_bios_sent++; 1030 if (last && len == 0) 1031 beio->send_complete = 1; 1032 mtx_unlock(&be_lun->io_lock); 1033 1034 (*dev_data->csw->d_strategy)(bio); 1035 } 1036} 1037 1038static void 1039ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1040 struct ctl_be_block_io *beio) 1041{ 1042 union ctl_io *io; 1043 struct ctl_be_block_devdata *dev_data; 1044 struct ctl_ptr_len_flags *ptrlen; 1045 struct scsi_unmap_desc *buf, *end; 1046 uint64_t len; 1047 1048 dev_data = &be_lun->backend.dev; 1049 io = beio->io; 1050 1051 DPRINTF("entered\n"); 1052 1053 binuptime(&beio->ds_t0); 1054 mtx_lock(&be_lun->io_lock); 1055 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1056 mtx_unlock(&be_lun->io_lock); 1057 1058 if (beio->io_offset == -1) { 1059 beio->io_len = 0; 1060 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1061 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1062 end = buf + ptrlen->len / sizeof(*buf); 1063 for (; buf < end; buf++) { 1064 len = (uint64_t)scsi_4btoul(buf->length) * 1065 be_lun->blocksize; 1066 beio->io_len += len; 1067 ctl_be_block_unmap_dev_range(be_lun, beio, 1068 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 1069 (end - buf < 2) ? TRUE : FALSE); 1070 } 1071 } else 1072 ctl_be_block_unmap_dev_range(be_lun, beio, 1073 beio->io_offset, beio->io_len, TRUE); 1074} 1075 1076static void 1077ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1078 struct ctl_be_block_io *beio) 1079{ 1080 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1081 int i; 1082 struct bio *bio; 1083 struct ctl_be_block_devdata *dev_data; 1084 off_t cur_offset; 1085 int max_iosize; 1086 1087 DPRINTF("entered\n"); 1088 1089 dev_data = &be_lun->backend.dev; 1090 1091 /* 1092 * We have to limit our I/O size to the maximum supported by the 1093 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1094 * set it properly, use DFLTPHYS. 1095 */ 1096 max_iosize = dev_data->cdev->si_iosize_max; 1097 if (max_iosize < PAGE_SIZE) 1098 max_iosize = DFLTPHYS; 1099 1100 cur_offset = beio->io_offset; 1101 for (i = 0; i < beio->num_segs; i++) { 1102 size_t cur_size; 1103 uint8_t *cur_ptr; 1104 1105 cur_size = beio->sg_segs[i].len; 1106 cur_ptr = beio->sg_segs[i].addr; 1107 1108 while (cur_size > 0) { 1109 /* This can't fail, it's a blocking allocation. */ 1110 bio = g_alloc_bio(); 1111 1112 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1113 1114 bio->bio_cmd = beio->bio_cmd; 1115 bio->bio_dev = dev_data->cdev; 1116 bio->bio_caller1 = beio; 1117 bio->bio_length = min(cur_size, max_iosize); 1118 bio->bio_offset = cur_offset; 1119 bio->bio_data = cur_ptr; 1120 bio->bio_done = ctl_be_block_biodone; 1121 bio->bio_pblkno = cur_offset / be_lun->blocksize; 1122 1123 cur_offset += bio->bio_length; 1124 cur_ptr += bio->bio_length; 1125 cur_size -= bio->bio_length; 1126 1127 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1128 beio->num_bios_sent++; 1129 } 1130 } 1131 binuptime(&beio->ds_t0); 1132 mtx_lock(&be_lun->io_lock); 1133 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1134 beio->send_complete = 1; 1135 mtx_unlock(&be_lun->io_lock); 1136 1137 /* 1138 * Fire off all allocated requests! 1139 */ 1140 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1141 TAILQ_REMOVE(&queue, bio, bio_queue); 1142 (*dev_data->csw->d_strategy)(bio); 1143 } 1144} 1145 1146static uint64_t 1147ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1148{ 1149 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1150 struct diocgattr_arg arg; 1151 int error; 1152 1153 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1154 return (UINT64_MAX); 1155 strlcpy(arg.name, attrname, sizeof(arg.name)); 1156 arg.len = sizeof(arg.value.off); 1157 error = dev_data->csw->d_ioctl(dev_data->cdev, 1158 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1159 if (error != 0) 1160 return (UINT64_MAX); 1161 return (arg.value.off); 1162} 1163 1164static void 1165ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1166 union ctl_io *io) 1167{ 1168 struct ctl_be_block_io *beio; 1169 struct ctl_lba_len_flags *lbalen; 1170 1171 DPRINTF("entered\n"); 1172 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1173 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1174 1175 beio->io_len = lbalen->len * be_lun->blocksize; 1176 beio->io_offset = lbalen->lba * be_lun->blocksize; 1177 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1178 beio->bio_cmd = BIO_FLUSH; 1179 beio->ds_trans_type = DEVSTAT_NO_DATA; 1180 DPRINTF("SYNC\n"); 1181 be_lun->lun_flush(be_lun, beio); 1182} 1183 1184static void 1185ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1186{ 1187 union ctl_io *io; 1188 1189 io = beio->io; 1190 ctl_free_beio(beio); 1191 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1192 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1193 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1194 ctl_config_write_done(io); 1195 return; 1196 } 1197 1198 ctl_be_block_config_write(io); 1199} 1200 1201static void 1202ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1203 union ctl_io *io) 1204{ 1205 struct ctl_be_block_io *beio; 1206 struct ctl_lba_len_flags *lbalen; 1207 uint64_t len_left, lba; 1208 uint32_t pb, pbo, adj; 1209 int i, seglen; 1210 uint8_t *buf, *end; 1211 1212 DPRINTF("entered\n"); 1213 1214 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1215 lbalen = ARGS(beio->io); 1216 1217 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1218 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1219 ctl_free_beio(beio); 1220 ctl_set_invalid_field(&io->scsiio, 1221 /*sks_valid*/ 1, 1222 /*command*/ 1, 1223 /*field*/ 1, 1224 /*bit_valid*/ 0, 1225 /*bit*/ 0); 1226 ctl_config_write_done(io); 1227 return; 1228 } 1229 1230 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1231 beio->io_offset = lbalen->lba * be_lun->blocksize; 1232 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1233 beio->bio_cmd = BIO_DELETE; 1234 beio->ds_trans_type = DEVSTAT_FREE; 1235 1236 be_lun->unmap(be_lun, beio); 1237 return; 1238 } 1239 1240 beio->bio_cmd = BIO_WRITE; 1241 beio->ds_trans_type = DEVSTAT_WRITE; 1242 1243 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1244 (uintmax_t)lbalen->lba, lbalen->len); 1245 1246 pb = be_lun->blocksize << be_lun->pblockexp; 1247 if (be_lun->pblockoff > 0) 1248 pbo = pb - be_lun->blocksize * be_lun->pblockoff; 1249 else 1250 pbo = 0; 1251 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1252 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1253 1254 /* 1255 * Setup the S/G entry for this chunk. 1256 */ 1257 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1258 if (pb > be_lun->blocksize) { 1259 adj = ((lbalen->lba + lba) * be_lun->blocksize + 1260 seglen - pbo) % pb; 1261 if (seglen > adj) 1262 seglen -= adj; 1263 else 1264 seglen -= seglen % be_lun->blocksize; 1265 } else 1266 seglen -= seglen % be_lun->blocksize; 1267 beio->sg_segs[i].len = seglen; 1268 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1269 1270 DPRINTF("segment %d addr %p len %zd\n", i, 1271 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1272 1273 beio->num_segs++; 1274 len_left -= seglen; 1275 1276 buf = beio->sg_segs[i].addr; 1277 end = buf + seglen; 1278 for (; buf < end; buf += be_lun->blocksize) { 1279 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1280 if (lbalen->flags & SWS_LBDATA) 1281 scsi_ulto4b(lbalen->lba + lba, buf); 1282 lba++; 1283 } 1284 } 1285 1286 beio->io_offset = lbalen->lba * be_lun->blocksize; 1287 beio->io_len = lba * be_lun->blocksize; 1288 1289 /* We can not do all in one run. Correct and schedule rerun. */ 1290 if (len_left > 0) { 1291 lbalen->lba += lba; 1292 lbalen->len -= lba; 1293 beio->beio_cont = ctl_be_block_cw_done_ws; 1294 } 1295 1296 be_lun->dispatch(be_lun, beio); 1297} 1298 1299static void 1300ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1301 union ctl_io *io) 1302{ 1303 struct ctl_be_block_io *beio; 1304 struct ctl_ptr_len_flags *ptrlen; 1305 1306 DPRINTF("entered\n"); 1307 1308 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1309 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1310 1311 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1312 ctl_free_beio(beio); 1313 ctl_set_invalid_field(&io->scsiio, 1314 /*sks_valid*/ 0, 1315 /*command*/ 1, 1316 /*field*/ 0, 1317 /*bit_valid*/ 0, 1318 /*bit*/ 0); 1319 ctl_config_write_done(io); 1320 return; 1321 } 1322 1323 beio->io_len = 0; 1324 beio->io_offset = -1; 1325 beio->bio_cmd = BIO_DELETE; 1326 beio->ds_trans_type = DEVSTAT_FREE; 1327 DPRINTF("UNMAP\n"); 1328 be_lun->unmap(be_lun, beio); 1329} 1330 1331static void 1332ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1333{ 1334 union ctl_io *io; 1335 1336 io = beio->io; 1337 ctl_free_beio(beio); 1338 ctl_config_read_done(io); 1339} 1340 1341static void 1342ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1343 union ctl_io *io) 1344{ 1345 struct ctl_be_block_io *beio; 1346 struct ctl_be_block_softc *softc; 1347 1348 DPRINTF("entered\n"); 1349 1350 softc = be_lun->softc; 1351 beio = ctl_alloc_beio(softc); 1352 beio->io = io; 1353 beio->lun = be_lun; 1354 beio->beio_cont = ctl_be_block_cr_done; 1355 PRIV(io)->ptr = (void *)beio; 1356 1357 switch (io->scsiio.cdb[0]) { 1358 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1359 beio->bio_cmd = -1; 1360 beio->ds_trans_type = DEVSTAT_NO_DATA; 1361 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1362 beio->io_len = 0; 1363 if (be_lun->get_lba_status) 1364 be_lun->get_lba_status(be_lun, beio); 1365 else 1366 ctl_be_block_cr_done(beio); 1367 break; 1368 default: 1369 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1370 break; 1371 } 1372} 1373 1374static void 1375ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1376{ 1377 union ctl_io *io; 1378 1379 io = beio->io; 1380 ctl_free_beio(beio); 1381 ctl_config_write_done(io); 1382} 1383 1384static void 1385ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1386 union ctl_io *io) 1387{ 1388 struct ctl_be_block_io *beio; 1389 struct ctl_be_block_softc *softc; 1390 1391 DPRINTF("entered\n"); 1392 1393 softc = be_lun->softc; 1394 beio = ctl_alloc_beio(softc); 1395 beio->io = io; 1396 beio->lun = be_lun; 1397 beio->beio_cont = ctl_be_block_cw_done; 1398 switch (io->scsiio.tag_type) { 1399 case CTL_TAG_ORDERED: 1400 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1401 break; 1402 case CTL_TAG_HEAD_OF_QUEUE: 1403 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1404 break; 1405 case CTL_TAG_UNTAGGED: 1406 case CTL_TAG_SIMPLE: 1407 case CTL_TAG_ACA: 1408 default: 1409 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1410 break; 1411 } 1412 PRIV(io)->ptr = (void *)beio; 1413 1414 switch (io->scsiio.cdb[0]) { 1415 case SYNCHRONIZE_CACHE: 1416 case SYNCHRONIZE_CACHE_16: 1417 ctl_be_block_cw_dispatch_sync(be_lun, io); 1418 break; 1419 case WRITE_SAME_10: 1420 case WRITE_SAME_16: 1421 ctl_be_block_cw_dispatch_ws(be_lun, io); 1422 break; 1423 case UNMAP: 1424 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1425 break; 1426 default: 1427 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1428 break; 1429 } 1430} 1431 1432SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1433SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1434SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1435SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1436 1437static void 1438ctl_be_block_next(struct ctl_be_block_io *beio) 1439{ 1440 struct ctl_be_block_lun *be_lun; 1441 union ctl_io *io; 1442 1443 io = beio->io; 1444 be_lun = beio->lun; 1445 ctl_free_beio(beio); 1446 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1447 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1448 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1449 ctl_data_submit_done(io); 1450 return; 1451 } 1452 1453 io->io_hdr.status &= ~CTL_STATUS_MASK; 1454 io->io_hdr.status |= CTL_STATUS_NONE; 1455 1456 mtx_lock(&be_lun->queue_lock); 1457 /* 1458 * XXX KDM make sure that links is okay to use at this point. 1459 * Otherwise, we either need to add another field to ctl_io_hdr, 1460 * or deal with resource allocation here. 1461 */ 1462 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1463 mtx_unlock(&be_lun->queue_lock); 1464 1465 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1466} 1467 1468static void 1469ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1470 union ctl_io *io) 1471{ 1472 struct ctl_be_block_io *beio; 1473 struct ctl_be_block_softc *softc; 1474 struct ctl_lba_len_flags *lbalen; 1475 struct ctl_ptr_len_flags *bptrlen; 1476 uint64_t len_left, lbas; 1477 int i; 1478 1479 softc = be_lun->softc; 1480 1481 DPRINTF("entered\n"); 1482 1483 lbalen = ARGS(io); 1484 if (lbalen->flags & CTL_LLF_WRITE) { 1485 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1486 } else { 1487 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1488 } 1489 1490 beio = ctl_alloc_beio(softc); 1491 beio->io = io; 1492 beio->lun = be_lun; 1493 bptrlen = PRIV(io); 1494 bptrlen->ptr = (void *)beio; 1495 1496 switch (io->scsiio.tag_type) { 1497 case CTL_TAG_ORDERED: 1498 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1499 break; 1500 case CTL_TAG_HEAD_OF_QUEUE: 1501 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1502 break; 1503 case CTL_TAG_UNTAGGED: 1504 case CTL_TAG_SIMPLE: 1505 case CTL_TAG_ACA: 1506 default: 1507 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1508 break; 1509 } 1510 1511 if (lbalen->flags & CTL_LLF_WRITE) { 1512 beio->bio_cmd = BIO_WRITE; 1513 beio->ds_trans_type = DEVSTAT_WRITE; 1514 } else { 1515 beio->bio_cmd = BIO_READ; 1516 beio->ds_trans_type = DEVSTAT_READ; 1517 } 1518 1519 DPRINTF("%s at LBA %jx len %u @%ju\n", 1520 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1521 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1522 if (lbalen->flags & CTL_LLF_COMPARE) 1523 lbas = CTLBLK_HALF_IO_SIZE; 1524 else 1525 lbas = CTLBLK_MAX_IO_SIZE; 1526 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1527 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1528 beio->io_len = lbas * be_lun->blocksize; 1529 bptrlen->len += lbas; 1530 1531 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1532 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1533 i, CTLBLK_MAX_SEGS)); 1534 1535 /* 1536 * Setup the S/G entry for this chunk. 1537 */ 1538 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1539 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1540 1541 DPRINTF("segment %d addr %p len %zd\n", i, 1542 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1543 1544 /* Set up second segment for compare operation. */ 1545 if (lbalen->flags & CTL_LLF_COMPARE) { 1546 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1547 beio->sg_segs[i].len; 1548 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1549 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1550 } 1551 1552 beio->num_segs++; 1553 len_left -= beio->sg_segs[i].len; 1554 } 1555 if (bptrlen->len < lbalen->len) 1556 beio->beio_cont = ctl_be_block_next; 1557 io->scsiio.be_move_done = ctl_be_block_move_done; 1558 /* For compare we have separate S/G lists for read and datamove. */ 1559 if (lbalen->flags & CTL_LLF_COMPARE) 1560 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1561 else 1562 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1563 io->scsiio.kern_data_len = beio->io_len; 1564 io->scsiio.kern_data_resid = 0; 1565 io->scsiio.kern_sg_entries = beio->num_segs; 1566 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1567 1568 /* 1569 * For the read case, we need to read the data into our buffers and 1570 * then we can send it back to the user. For the write case, we 1571 * need to get the data from the user first. 1572 */ 1573 if (beio->bio_cmd == BIO_READ) { 1574 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1575 be_lun->dispatch(be_lun, beio); 1576 } else { 1577 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1578#ifdef CTL_TIME_IO 1579 getbintime(&io->io_hdr.dma_start_bt); 1580#endif 1581 ctl_datamove(io); 1582 } 1583} 1584 1585static void 1586ctl_be_block_worker(void *context, int pending) 1587{ 1588 struct ctl_be_block_lun *be_lun; 1589 struct ctl_be_block_softc *softc; 1590 union ctl_io *io; 1591 1592 be_lun = (struct ctl_be_block_lun *)context; 1593 softc = be_lun->softc; 1594 1595 DPRINTF("entered\n"); 1596 1597 mtx_lock(&be_lun->queue_lock); 1598 for (;;) { 1599 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1600 if (io != NULL) { 1601 struct ctl_be_block_io *beio; 1602 1603 DPRINTF("datamove queue\n"); 1604 1605 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1606 ctl_io_hdr, links); 1607 1608 mtx_unlock(&be_lun->queue_lock); 1609 1610 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1611 1612 be_lun->dispatch(be_lun, beio); 1613 1614 mtx_lock(&be_lun->queue_lock); 1615 continue; 1616 } 1617 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1618 if (io != NULL) { 1619 DPRINTF("config write queue\n"); 1620 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1621 ctl_io_hdr, links); 1622 mtx_unlock(&be_lun->queue_lock); 1623 ctl_be_block_cw_dispatch(be_lun, io); 1624 mtx_lock(&be_lun->queue_lock); 1625 continue; 1626 } 1627 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1628 if (io != NULL) { 1629 DPRINTF("config read queue\n"); 1630 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1631 ctl_io_hdr, links); 1632 mtx_unlock(&be_lun->queue_lock); 1633 ctl_be_block_cr_dispatch(be_lun, io); 1634 mtx_lock(&be_lun->queue_lock); 1635 continue; 1636 } 1637 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1638 if (io != NULL) { 1639 DPRINTF("input queue\n"); 1640 1641 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1642 ctl_io_hdr, links); 1643 mtx_unlock(&be_lun->queue_lock); 1644 1645 /* 1646 * We must drop the lock, since this routine and 1647 * its children may sleep. 1648 */ 1649 ctl_be_block_dispatch(be_lun, io); 1650 1651 mtx_lock(&be_lun->queue_lock); 1652 continue; 1653 } 1654 1655 /* 1656 * If we get here, there is no work left in the queues, so 1657 * just break out and let the task queue go to sleep. 1658 */ 1659 break; 1660 } 1661 mtx_unlock(&be_lun->queue_lock); 1662} 1663 1664/* 1665 * Entry point from CTL to the backend for I/O. We queue everything to a 1666 * work thread, so this just puts the I/O on a queue and wakes up the 1667 * thread. 1668 */ 1669static int 1670ctl_be_block_submit(union ctl_io *io) 1671{ 1672 struct ctl_be_block_lun *be_lun; 1673 struct ctl_be_lun *ctl_be_lun; 1674 1675 DPRINTF("entered\n"); 1676 1677 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1678 CTL_PRIV_BACKEND_LUN].ptr; 1679 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1680 1681 /* 1682 * Make sure we only get SCSI I/O. 1683 */ 1684 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1685 "%#x) encountered", io->io_hdr.io_type)); 1686 1687 PRIV(io)->len = 0; 1688 1689 mtx_lock(&be_lun->queue_lock); 1690 /* 1691 * XXX KDM make sure that links is okay to use at this point. 1692 * Otherwise, we either need to add another field to ctl_io_hdr, 1693 * or deal with resource allocation here. 1694 */ 1695 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1696 mtx_unlock(&be_lun->queue_lock); 1697 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1698 1699 return (CTL_RETVAL_COMPLETE); 1700} 1701 1702static int 1703ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1704 int flag, struct thread *td) 1705{ 1706 struct ctl_be_block_softc *softc; 1707 int error; 1708 1709 softc = &backend_block_softc; 1710 1711 error = 0; 1712 1713 switch (cmd) { 1714 case CTL_LUN_REQ: { 1715 struct ctl_lun_req *lun_req; 1716 1717 lun_req = (struct ctl_lun_req *)addr; 1718 1719 switch (lun_req->reqtype) { 1720 case CTL_LUNREQ_CREATE: 1721 error = ctl_be_block_create(softc, lun_req); 1722 break; 1723 case CTL_LUNREQ_RM: 1724 error = ctl_be_block_rm(softc, lun_req); 1725 break; 1726 case CTL_LUNREQ_MODIFY: 1727 error = ctl_be_block_modify(softc, lun_req); 1728 break; 1729 default: 1730 lun_req->status = CTL_LUN_ERROR; 1731 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1732 "invalid LUN request type %d", 1733 lun_req->reqtype); 1734 break; 1735 } 1736 break; 1737 } 1738 default: 1739 error = ENOTTY; 1740 break; 1741 } 1742 1743 return (error); 1744} 1745 1746static int 1747ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1748{ 1749 struct ctl_be_block_filedata *file_data; 1750 struct ctl_lun_create_params *params; 1751 char *value; 1752 struct vattr vattr; 1753 off_t ps, pss, po, pos, us, uss, uo, uos; 1754 int error; 1755 1756 error = 0; 1757 file_data = &be_lun->backend.file; 1758 params = &be_lun->params; 1759 1760 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1761 be_lun->dispatch = ctl_be_block_dispatch_file; 1762 be_lun->lun_flush = ctl_be_block_flush_file; 1763 be_lun->get_lba_status = ctl_be_block_gls_file; 1764 be_lun->getattr = ctl_be_block_getattr_file; 1765 1766 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1767 if (error != 0) { 1768 snprintf(req->error_str, sizeof(req->error_str), 1769 "error calling VOP_GETATTR() for file %s", 1770 be_lun->dev_path); 1771 return (error); 1772 } 1773 1774 /* 1775 * Verify that we have the ability to upgrade to exclusive 1776 * access on this file so we can trap errors at open instead 1777 * of reporting them during first access. 1778 */ 1779 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1780 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1781 if (be_lun->vn->v_iflag & VI_DOOMED) { 1782 error = EBADF; 1783 snprintf(req->error_str, sizeof(req->error_str), 1784 "error locking file %s", be_lun->dev_path); 1785 return (error); 1786 } 1787 } 1788 1789 1790 file_data->cred = crhold(curthread->td_ucred); 1791 if (params->lun_size_bytes != 0) 1792 be_lun->size_bytes = params->lun_size_bytes; 1793 else 1794 be_lun->size_bytes = vattr.va_size; 1795 /* 1796 * We set the multi thread flag for file operations because all 1797 * filesystems (in theory) are capable of allowing multiple readers 1798 * of a file at once. So we want to get the maximum possible 1799 * concurrency. 1800 */ 1801 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1802 1803 /* 1804 * For files we can use any logical block size. Prefer 512 bytes 1805 * for compatibility reasons. If file's vattr.va_blocksize 1806 * (preferred I/O block size) is bigger and multiple to chosen 1807 * logical block size -- report it as physical block size. 1808 */ 1809 if (params->blocksize_bytes != 0) 1810 be_lun->blocksize = params->blocksize_bytes; 1811 else 1812 be_lun->blocksize = 512; 1813 1814 us = ps = vattr.va_blocksize; 1815 uo = po = 0; 1816 1817 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 1818 if (value != NULL) 1819 ctl_expand_number(value, &ps); 1820 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 1821 if (value != NULL) 1822 ctl_expand_number(value, &po); 1823 pss = ps / be_lun->blocksize; 1824 pos = po / be_lun->blocksize; 1825 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1826 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1827 be_lun->pblockexp = fls(pss) - 1; 1828 be_lun->pblockoff = (pss - pos) % pss; 1829 } 1830 1831 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 1832 if (value != NULL) 1833 ctl_expand_number(value, &us); 1834 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 1835 if (value != NULL) 1836 ctl_expand_number(value, &uo); 1837 uss = us / be_lun->blocksize; 1838 uos = uo / be_lun->blocksize; 1839 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 1840 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 1841 be_lun->ublockexp = fls(uss) - 1; 1842 be_lun->ublockoff = (uss - uos) % uss; 1843 } 1844 1845 /* 1846 * Sanity check. The media size has to be at least one 1847 * sector long. 1848 */ 1849 if (be_lun->size_bytes < be_lun->blocksize) { 1850 error = EINVAL; 1851 snprintf(req->error_str, sizeof(req->error_str), 1852 "file %s size %ju < block size %u", be_lun->dev_path, 1853 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1854 } 1855 1856 be_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / be_lun->blocksize; 1857 return (error); 1858} 1859 1860static int 1861ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1862{ 1863 struct ctl_lun_create_params *params; 1864 struct vattr vattr; 1865 struct cdev *dev; 1866 struct cdevsw *devsw; 1867 char *value; 1868 int error, atomic, maxio, unmap; 1869 off_t ps, pss, po, pos, us, uss, uo, uos; 1870 1871 params = &be_lun->params; 1872 1873 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1874 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1875 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1876 &be_lun->backend.dev.dev_ref); 1877 if (be_lun->backend.dev.csw == NULL) 1878 panic("Unable to retrieve device switch"); 1879 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) { 1880 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1881 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1882 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1883 } else { 1884 be_lun->dispatch = ctl_be_block_dispatch_dev; 1885 atomic = 0; 1886 maxio = be_lun->backend.dev.cdev->si_iosize_max; 1887 if (maxio <= 0) 1888 maxio = DFLTPHYS; 1889 if (maxio > CTLBLK_MAX_IO_SIZE) 1890 maxio = CTLBLK_MAX_IO_SIZE; 1891 } 1892 be_lun->lun_flush = ctl_be_block_flush_dev; 1893 be_lun->getattr = ctl_be_block_getattr_dev; 1894 1895 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1896 if (error) { 1897 snprintf(req->error_str, sizeof(req->error_str), 1898 "error getting vnode attributes for device %s", 1899 be_lun->dev_path); 1900 return (error); 1901 } 1902 1903 dev = be_lun->vn->v_rdev; 1904 devsw = dev->si_devsw; 1905 if (!devsw->d_ioctl) { 1906 snprintf(req->error_str, sizeof(req->error_str), 1907 "no d_ioctl for device %s!", 1908 be_lun->dev_path); 1909 return (ENODEV); 1910 } 1911 1912 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1913 (caddr_t)&be_lun->blocksize, FREAD, 1914 curthread); 1915 if (error) { 1916 snprintf(req->error_str, sizeof(req->error_str), 1917 "error %d returned for DIOCGSECTORSIZE ioctl " 1918 "on %s!", error, be_lun->dev_path); 1919 return (error); 1920 } 1921 1922 /* 1923 * If the user has asked for a blocksize that is greater than the 1924 * backing device's blocksize, we can do it only if the blocksize 1925 * the user is asking for is an even multiple of the underlying 1926 * device's blocksize. 1927 */ 1928 if ((params->blocksize_bytes != 0) 1929 && (params->blocksize_bytes > be_lun->blocksize)) { 1930 uint32_t bs_multiple, tmp_blocksize; 1931 1932 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1933 1934 tmp_blocksize = bs_multiple * be_lun->blocksize; 1935 1936 if (tmp_blocksize == params->blocksize_bytes) { 1937 be_lun->blocksize = params->blocksize_bytes; 1938 } else { 1939 snprintf(req->error_str, sizeof(req->error_str), 1940 "requested blocksize %u is not an even " 1941 "multiple of backing device blocksize %u", 1942 params->blocksize_bytes, 1943 be_lun->blocksize); 1944 return (EINVAL); 1945 1946 } 1947 } else if ((params->blocksize_bytes != 0) 1948 && (params->blocksize_bytes != be_lun->blocksize)) { 1949 snprintf(req->error_str, sizeof(req->error_str), 1950 "requested blocksize %u < backing device " 1951 "blocksize %u", params->blocksize_bytes, 1952 be_lun->blocksize); 1953 return (EINVAL); 1954 } 1955 1956 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1957 (caddr_t)&be_lun->size_bytes, FREAD, 1958 curthread); 1959 if (error) { 1960 snprintf(req->error_str, sizeof(req->error_str), 1961 "error %d returned for DIOCGMEDIASIZE " 1962 " ioctl on %s!", error, 1963 be_lun->dev_path); 1964 return (error); 1965 } 1966 1967 if (params->lun_size_bytes != 0) { 1968 if (params->lun_size_bytes > be_lun->size_bytes) { 1969 snprintf(req->error_str, sizeof(req->error_str), 1970 "requested LUN size %ju > backing device " 1971 "size %ju", 1972 (uintmax_t)params->lun_size_bytes, 1973 (uintmax_t)be_lun->size_bytes); 1974 return (EINVAL); 1975 } 1976 1977 be_lun->size_bytes = params->lun_size_bytes; 1978 } 1979 1980 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1981 (caddr_t)&ps, FREAD, curthread); 1982 if (error) 1983 ps = po = 0; 1984 else { 1985 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1986 (caddr_t)&po, FREAD, curthread); 1987 if (error) 1988 po = 0; 1989 } 1990 us = ps; 1991 uo = po; 1992 1993 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 1994 if (value != NULL) 1995 ctl_expand_number(value, &ps); 1996 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 1997 if (value != NULL) 1998 ctl_expand_number(value, &po); 1999 pss = ps / be_lun->blocksize; 2000 pos = po / be_lun->blocksize; 2001 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 2002 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 2003 be_lun->pblockexp = fls(pss) - 1; 2004 be_lun->pblockoff = (pss - pos) % pss; 2005 } 2006 2007 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 2008 if (value != NULL) 2009 ctl_expand_number(value, &us); 2010 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 2011 if (value != NULL) 2012 ctl_expand_number(value, &uo); 2013 uss = us / be_lun->blocksize; 2014 uos = uo / be_lun->blocksize; 2015 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 2016 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 2017 be_lun->ublockexp = fls(uss) - 1; 2018 be_lun->ublockoff = (uss - uos) % uss; 2019 } 2020 2021 be_lun->atomicblock = atomic / be_lun->blocksize; 2022 be_lun->opttxferlen = maxio / be_lun->blocksize; 2023 2024 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2025 unmap = 1; 2026 } else { 2027 struct diocgattr_arg arg; 2028 2029 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2030 arg.len = sizeof(arg.value.i); 2031 error = devsw->d_ioctl(dev, DIOCGATTR, 2032 (caddr_t)&arg, FREAD, curthread); 2033 unmap = (error == 0) ? arg.value.i : 0; 2034 } 2035 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2036 if (value != NULL) 2037 unmap = (strcmp(value, "on") == 0); 2038 if (unmap) 2039 be_lun->unmap = ctl_be_block_unmap_dev; 2040 2041 return (0); 2042} 2043 2044static int 2045ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2046{ 2047 DROP_GIANT(); 2048 if (be_lun->vn) { 2049 int flags = FREAD | FWRITE; 2050 2051 switch (be_lun->dev_type) { 2052 case CTL_BE_BLOCK_DEV: 2053 if (be_lun->backend.dev.csw) { 2054 dev_relthread(be_lun->backend.dev.cdev, 2055 be_lun->backend.dev.dev_ref); 2056 be_lun->backend.dev.csw = NULL; 2057 be_lun->backend.dev.cdev = NULL; 2058 } 2059 break; 2060 case CTL_BE_BLOCK_FILE: 2061 break; 2062 case CTL_BE_BLOCK_NONE: 2063 break; 2064 default: 2065 panic("Unexpected backend type."); 2066 break; 2067 } 2068 2069 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2070 be_lun->vn = NULL; 2071 2072 switch (be_lun->dev_type) { 2073 case CTL_BE_BLOCK_DEV: 2074 break; 2075 case CTL_BE_BLOCK_FILE: 2076 if (be_lun->backend.file.cred != NULL) { 2077 crfree(be_lun->backend.file.cred); 2078 be_lun->backend.file.cred = NULL; 2079 } 2080 break; 2081 case CTL_BE_BLOCK_NONE: 2082 break; 2083 default: 2084 panic("Unexpected backend type."); 2085 break; 2086 } 2087 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2088 } 2089 PICKUP_GIANT(); 2090 2091 return (0); 2092} 2093 2094static int 2095ctl_be_block_open(struct ctl_be_block_softc *softc, 2096 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2097{ 2098 struct nameidata nd; 2099 int flags; 2100 int error; 2101 2102 /* 2103 * XXX KDM allow a read-only option? 2104 */ 2105 flags = FREAD | FWRITE; 2106 error = 0; 2107 2108 if (rootvnode == NULL) { 2109 snprintf(req->error_str, sizeof(req->error_str), 2110 "Root filesystem is not mounted"); 2111 return (1); 2112 } 2113 2114 if (!curthread->td_proc->p_fd->fd_cdir) { 2115 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2116 VREF(rootvnode); 2117 } 2118 if (!curthread->td_proc->p_fd->fd_rdir) { 2119 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2120 VREF(rootvnode); 2121 } 2122 if (!curthread->td_proc->p_fd->fd_jdir) { 2123 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2124 VREF(rootvnode); 2125 } 2126 2127 again: 2128 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2129 error = vn_open(&nd, &flags, 0, NULL); 2130 if (error) { 2131 /* 2132 * This is the only reasonable guess we can make as far as 2133 * path if the user doesn't give us a fully qualified path. 2134 * If they want to specify a file, they need to specify the 2135 * full path. 2136 */ 2137 if (be_lun->dev_path[0] != '/') { 2138 char *dev_path = "/dev/"; 2139 char *dev_name; 2140 2141 /* Try adding device path at beginning of name */ 2142 dev_name = malloc(strlen(be_lun->dev_path) 2143 + strlen(dev_path) + 1, 2144 M_CTLBLK, M_WAITOK); 2145 if (dev_name) { 2146 sprintf(dev_name, "%s%s", dev_path, 2147 be_lun->dev_path); 2148 free(be_lun->dev_path, M_CTLBLK); 2149 be_lun->dev_path = dev_name; 2150 goto again; 2151 } 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 2158 NDFREE(&nd, NDF_ONLY_PNBUF); 2159 2160 be_lun->vn = nd.ni_vp; 2161 2162 /* We only support disks and files. */ 2163 if (vn_isdisk(be_lun->vn, &error)) { 2164 error = ctl_be_block_open_dev(be_lun, req); 2165 } else if (be_lun->vn->v_type == VREG) { 2166 error = ctl_be_block_open_file(be_lun, req); 2167 } else { 2168 error = EINVAL; 2169 snprintf(req->error_str, sizeof(req->error_str), 2170 "%s is not a disk or plain file", be_lun->dev_path); 2171 } 2172 VOP_UNLOCK(be_lun->vn, 0); 2173 2174 if (error != 0) { 2175 ctl_be_block_close(be_lun); 2176 return (error); 2177 } 2178 2179 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2180 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2181 2182 return (0); 2183} 2184 2185static int 2186ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2187{ 2188 struct ctl_be_block_lun *be_lun; 2189 struct ctl_lun_create_params *params; 2190 char num_thread_str[16]; 2191 char tmpstr[32]; 2192 char *value; 2193 int retval, num_threads; 2194 int tmp_num_threads; 2195 2196 params = &req->reqdata.create; 2197 retval = 0; 2198 req->status = CTL_LUN_OK; 2199 2200 num_threads = cbb_num_threads; 2201 2202 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2203 2204 be_lun->params = req->reqdata.create; 2205 be_lun->softc = softc; 2206 STAILQ_INIT(&be_lun->input_queue); 2207 STAILQ_INIT(&be_lun->config_read_queue); 2208 STAILQ_INIT(&be_lun->config_write_queue); 2209 STAILQ_INIT(&be_lun->datamove_queue); 2210 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2211 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2212 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2213 ctl_init_opts(&be_lun->ctl_be_lun.options, 2214 req->num_be_args, req->kern_be_args); 2215 2216 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2217 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2218 2219 if (be_lun->lun_zone == NULL) { 2220 snprintf(req->error_str, sizeof(req->error_str), 2221 "error allocating UMA zone"); 2222 goto bailout_error; 2223 } 2224 2225 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2226 be_lun->ctl_be_lun.lun_type = params->device_type; 2227 else 2228 be_lun->ctl_be_lun.lun_type = T_DIRECT; 2229 2230 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 2231 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 2232 if (value == NULL) { 2233 snprintf(req->error_str, sizeof(req->error_str), 2234 "no file argument specified"); 2235 goto bailout_error; 2236 } 2237 be_lun->dev_path = strdup(value, M_CTLBLK); 2238 be_lun->blocksize = 512; 2239 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2240 2241 retval = ctl_be_block_open(softc, be_lun, req); 2242 if (retval != 0) { 2243 retval = 0; 2244 req->status = CTL_LUN_WARNING; 2245 } 2246 } else { 2247 /* 2248 * For processor devices, we don't have any size. 2249 */ 2250 be_lun->blocksize = 0; 2251 be_lun->pblockexp = 0; 2252 be_lun->pblockoff = 0; 2253 be_lun->ublockexp = 0; 2254 be_lun->ublockoff = 0; 2255 be_lun->size_blocks = 0; 2256 be_lun->size_bytes = 0; 2257 be_lun->ctl_be_lun.maxlba = 0; 2258 2259 /* 2260 * Default to just 1 thread for processor devices. 2261 */ 2262 num_threads = 1; 2263 } 2264 2265 /* 2266 * XXX This searching loop might be refactored to be combined with 2267 * the loop above, 2268 */ 2269 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2270 if (value != NULL) { 2271 tmp_num_threads = strtol(value, NULL, 0); 2272 2273 /* 2274 * We don't let the user specify less than one 2275 * thread, but hope he's clueful enough not to 2276 * specify 1000 threads. 2277 */ 2278 if (tmp_num_threads < 1) { 2279 snprintf(req->error_str, sizeof(req->error_str), 2280 "invalid number of threads %s", 2281 num_thread_str); 2282 goto bailout_error; 2283 } 2284 num_threads = tmp_num_threads; 2285 } 2286 2287 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2288 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2289 if (be_lun->vn == NULL) 2290 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2291 if (be_lun->unmap != NULL) 2292 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2293 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2294 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_SERSEQ_READ; 2295 be_lun->ctl_be_lun.be_lun = be_lun; 2296 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2297 0 : (be_lun->size_blocks - 1); 2298 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2299 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2300 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2301 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2302 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2303 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2304 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2305 /* Tell the user the blocksize we ended up using */ 2306 params->lun_size_bytes = be_lun->size_bytes; 2307 params->blocksize_bytes = be_lun->blocksize; 2308 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2309 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2310 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2311 } else 2312 be_lun->ctl_be_lun.req_lun_id = 0; 2313 2314 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2315 be_lun->ctl_be_lun.lun_config_status = 2316 ctl_be_block_lun_config_status; 2317 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2318 2319 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2320 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2321 softc->num_luns); 2322 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2323 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2324 sizeof(tmpstr))); 2325 2326 /* Tell the user what we used for a serial number */ 2327 strncpy((char *)params->serial_num, tmpstr, 2328 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2329 } else { 2330 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2331 params->serial_num, 2332 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2333 sizeof(params->serial_num))); 2334 } 2335 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2336 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2337 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2338 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2339 sizeof(tmpstr))); 2340 2341 /* Tell the user what we used for a device ID */ 2342 strncpy((char *)params->device_id, tmpstr, 2343 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2344 } else { 2345 strncpy((char *)be_lun->ctl_be_lun.device_id, 2346 params->device_id, 2347 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2348 sizeof(params->device_id))); 2349 } 2350 2351 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2352 2353 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2354 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2355 2356 if (be_lun->io_taskqueue == NULL) { 2357 snprintf(req->error_str, sizeof(req->error_str), 2358 "unable to create taskqueue"); 2359 goto bailout_error; 2360 } 2361 2362 /* 2363 * Note that we start the same number of threads by default for 2364 * both the file case and the block device case. For the file 2365 * case, we need multiple threads to allow concurrency, because the 2366 * vnode interface is designed to be a blocking interface. For the 2367 * block device case, ZFS zvols at least will block the caller's 2368 * context in many instances, and so we need multiple threads to 2369 * overcome that problem. Other block devices don't need as many 2370 * threads, but they shouldn't cause too many problems. 2371 * 2372 * If the user wants to just have a single thread for a block 2373 * device, he can specify that when the LUN is created, or change 2374 * the tunable/sysctl to alter the default number of threads. 2375 */ 2376 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2377 /*num threads*/num_threads, 2378 /*priority*/PWAIT, 2379 /*thread name*/ 2380 "%s taskq", be_lun->lunname); 2381 2382 if (retval != 0) 2383 goto bailout_error; 2384 2385 be_lun->num_threads = num_threads; 2386 2387 mtx_lock(&softc->lock); 2388 softc->num_luns++; 2389 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2390 2391 mtx_unlock(&softc->lock); 2392 2393 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2394 if (retval != 0) { 2395 mtx_lock(&softc->lock); 2396 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2397 links); 2398 softc->num_luns--; 2399 mtx_unlock(&softc->lock); 2400 snprintf(req->error_str, sizeof(req->error_str), 2401 "ctl_add_lun() returned error %d, see dmesg for " 2402 "details", retval); 2403 retval = 0; 2404 goto bailout_error; 2405 } 2406 2407 mtx_lock(&softc->lock); 2408 2409 /* 2410 * Tell the config_status routine that we're waiting so it won't 2411 * clean up the LUN in the event of an error. 2412 */ 2413 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2414 2415 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2416 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2417 if (retval == EINTR) 2418 break; 2419 } 2420 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2421 2422 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2423 snprintf(req->error_str, sizeof(req->error_str), 2424 "LUN configuration error, see dmesg for details"); 2425 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2426 links); 2427 softc->num_luns--; 2428 mtx_unlock(&softc->lock); 2429 goto bailout_error; 2430 } else { 2431 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2432 } 2433 2434 mtx_unlock(&softc->lock); 2435 2436 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2437 be_lun->blocksize, 2438 DEVSTAT_ALL_SUPPORTED, 2439 be_lun->ctl_be_lun.lun_type 2440 | DEVSTAT_TYPE_IF_OTHER, 2441 DEVSTAT_PRIORITY_OTHER); 2442 2443 return (retval); 2444 2445bailout_error: 2446 req->status = CTL_LUN_ERROR; 2447 2448 if (be_lun->io_taskqueue != NULL) 2449 taskqueue_free(be_lun->io_taskqueue); 2450 ctl_be_block_close(be_lun); 2451 if (be_lun->dev_path != NULL) 2452 free(be_lun->dev_path, M_CTLBLK); 2453 if (be_lun->lun_zone != NULL) 2454 uma_zdestroy(be_lun->lun_zone); 2455 ctl_free_opts(&be_lun->ctl_be_lun.options); 2456 mtx_destroy(&be_lun->queue_lock); 2457 mtx_destroy(&be_lun->io_lock); 2458 free(be_lun, M_CTLBLK); 2459 2460 return (retval); 2461} 2462 2463static int 2464ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2465{ 2466 struct ctl_lun_rm_params *params; 2467 struct ctl_be_block_lun *be_lun; 2468 int retval; 2469 2470 params = &req->reqdata.rm; 2471 2472 mtx_lock(&softc->lock); 2473 2474 be_lun = NULL; 2475 2476 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2477 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2478 break; 2479 } 2480 mtx_unlock(&softc->lock); 2481 2482 if (be_lun == NULL) { 2483 snprintf(req->error_str, sizeof(req->error_str), 2484 "LUN %u is not managed by the block backend", 2485 params->lun_id); 2486 goto bailout_error; 2487 } 2488 2489 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2490 2491 if (retval != 0) { 2492 snprintf(req->error_str, sizeof(req->error_str), 2493 "error %d returned from ctl_disable_lun() for " 2494 "LUN %d", retval, params->lun_id); 2495 goto bailout_error; 2496 2497 } 2498 2499 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2500 if (retval != 0) { 2501 snprintf(req->error_str, sizeof(req->error_str), 2502 "error %d returned from ctl_invalidate_lun() for " 2503 "LUN %d", retval, params->lun_id); 2504 goto bailout_error; 2505 } 2506 2507 mtx_lock(&softc->lock); 2508 2509 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2510 2511 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2512 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2513 if (retval == EINTR) 2514 break; 2515 } 2516 2517 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2518 2519 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2520 snprintf(req->error_str, sizeof(req->error_str), 2521 "interrupted waiting for LUN to be freed"); 2522 mtx_unlock(&softc->lock); 2523 goto bailout_error; 2524 } 2525 2526 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2527 2528 softc->num_luns--; 2529 mtx_unlock(&softc->lock); 2530 2531 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2532 2533 taskqueue_free(be_lun->io_taskqueue); 2534 2535 ctl_be_block_close(be_lun); 2536 2537 if (be_lun->disk_stats != NULL) 2538 devstat_remove_entry(be_lun->disk_stats); 2539 2540 uma_zdestroy(be_lun->lun_zone); 2541 2542 ctl_free_opts(&be_lun->ctl_be_lun.options); 2543 free(be_lun->dev_path, M_CTLBLK); 2544 mtx_destroy(&be_lun->queue_lock); 2545 mtx_destroy(&be_lun->io_lock); 2546 free(be_lun, M_CTLBLK); 2547 2548 req->status = CTL_LUN_OK; 2549 2550 return (0); 2551 2552bailout_error: 2553 2554 req->status = CTL_LUN_ERROR; 2555 2556 return (0); 2557} 2558 2559static int 2560ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2561 struct ctl_lun_req *req) 2562{ 2563 struct vattr vattr; 2564 int error; 2565 struct ctl_lun_create_params *params = &be_lun->params; 2566 2567 if (params->lun_size_bytes != 0) { 2568 be_lun->size_bytes = params->lun_size_bytes; 2569 } else { 2570 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2571 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2572 VOP_UNLOCK(be_lun->vn, 0); 2573 if (error != 0) { 2574 snprintf(req->error_str, sizeof(req->error_str), 2575 "error calling VOP_GETATTR() for file %s", 2576 be_lun->dev_path); 2577 return (error); 2578 } 2579 2580 be_lun->size_bytes = vattr.va_size; 2581 } 2582 2583 return (0); 2584} 2585 2586static int 2587ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2588 struct ctl_lun_req *req) 2589{ 2590 struct ctl_be_block_devdata *dev_data; 2591 int error; 2592 struct ctl_lun_create_params *params = &be_lun->params; 2593 uint64_t size_bytes; 2594 2595 dev_data = &be_lun->backend.dev; 2596 if (!dev_data->csw->d_ioctl) { 2597 snprintf(req->error_str, sizeof(req->error_str), 2598 "no d_ioctl for device %s!", be_lun->dev_path); 2599 return (ENODEV); 2600 } 2601 2602 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2603 (caddr_t)&size_bytes, FREAD, 2604 curthread); 2605 if (error) { 2606 snprintf(req->error_str, sizeof(req->error_str), 2607 "error %d returned for DIOCGMEDIASIZE ioctl " 2608 "on %s!", error, be_lun->dev_path); 2609 return (error); 2610 } 2611 2612 if (params->lun_size_bytes != 0) { 2613 if (params->lun_size_bytes > size_bytes) { 2614 snprintf(req->error_str, sizeof(req->error_str), 2615 "requested LUN size %ju > backing device " 2616 "size %ju", 2617 (uintmax_t)params->lun_size_bytes, 2618 (uintmax_t)size_bytes); 2619 return (EINVAL); 2620 } 2621 2622 be_lun->size_bytes = params->lun_size_bytes; 2623 } else { 2624 be_lun->size_bytes = size_bytes; 2625 } 2626 2627 return (0); 2628} 2629 2630static int 2631ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2632{ 2633 struct ctl_lun_modify_params *params; 2634 struct ctl_be_block_lun *be_lun; 2635 uint64_t oldsize; 2636 int error; 2637 2638 params = &req->reqdata.modify; 2639 2640 mtx_lock(&softc->lock); 2641 be_lun = NULL; 2642 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2643 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2644 break; 2645 } 2646 mtx_unlock(&softc->lock); 2647 2648 if (be_lun == NULL) { 2649 snprintf(req->error_str, sizeof(req->error_str), 2650 "LUN %u is not managed by the block backend", 2651 params->lun_id); 2652 goto bailout_error; 2653 } 2654 2655 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2656 2657 oldsize = be_lun->size_bytes; 2658 if (be_lun->vn == NULL) 2659 error = ctl_be_block_open(softc, be_lun, req); 2660 else if (be_lun->vn->v_type == VREG) 2661 error = ctl_be_block_modify_file(be_lun, req); 2662 else 2663 error = ctl_be_block_modify_dev(be_lun, req); 2664 2665 if (error == 0 && be_lun->size_bytes != oldsize) { 2666 be_lun->size_blocks = be_lun->size_bytes >> 2667 be_lun->blocksize_shift; 2668 2669 /* 2670 * The maximum LBA is the size - 1. 2671 * 2672 * XXX: Note that this field is being updated without locking, 2673 * which might cause problems on 32-bit architectures. 2674 */ 2675 if (be_lun->unmap != NULL) 2676 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2677 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2678 0 : (be_lun->size_blocks - 1); 2679 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2680 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2681 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2682 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2683 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2684 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2685 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2686 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2687 if (oldsize == 0 && be_lun->size_blocks != 0) 2688 ctl_lun_online(&be_lun->ctl_be_lun); 2689 } 2690 2691 /* Tell the user the exact size we ended up using */ 2692 params->lun_size_bytes = be_lun->size_bytes; 2693 2694 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2695 2696 return (0); 2697 2698bailout_error: 2699 req->status = CTL_LUN_ERROR; 2700 2701 return (0); 2702} 2703 2704static void 2705ctl_be_block_lun_shutdown(void *be_lun) 2706{ 2707 struct ctl_be_block_lun *lun; 2708 struct ctl_be_block_softc *softc; 2709 2710 lun = (struct ctl_be_block_lun *)be_lun; 2711 2712 softc = lun->softc; 2713 2714 mtx_lock(&softc->lock); 2715 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2716 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2717 wakeup(lun); 2718 mtx_unlock(&softc->lock); 2719 2720} 2721 2722static void 2723ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2724{ 2725 struct ctl_be_block_lun *lun; 2726 struct ctl_be_block_softc *softc; 2727 2728 lun = (struct ctl_be_block_lun *)be_lun; 2729 softc = lun->softc; 2730 2731 if (status == CTL_LUN_CONFIG_OK) { 2732 mtx_lock(&softc->lock); 2733 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2734 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2735 wakeup(lun); 2736 mtx_unlock(&softc->lock); 2737 2738 /* 2739 * We successfully added the LUN, attempt to enable it. 2740 */ 2741 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2742 printf("%s: ctl_enable_lun() failed!\n", __func__); 2743 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2744 printf("%s: ctl_invalidate_lun() failed!\n", 2745 __func__); 2746 } 2747 } 2748 2749 return; 2750 } 2751 2752 2753 mtx_lock(&softc->lock); 2754 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2755 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2756 wakeup(lun); 2757 mtx_unlock(&softc->lock); 2758} 2759 2760 2761static int 2762ctl_be_block_config_write(union ctl_io *io) 2763{ 2764 struct ctl_be_block_lun *be_lun; 2765 struct ctl_be_lun *ctl_be_lun; 2766 int retval; 2767 2768 retval = 0; 2769 2770 DPRINTF("entered\n"); 2771 2772 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2773 CTL_PRIV_BACKEND_LUN].ptr; 2774 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2775 2776 switch (io->scsiio.cdb[0]) { 2777 case SYNCHRONIZE_CACHE: 2778 case SYNCHRONIZE_CACHE_16: 2779 case WRITE_SAME_10: 2780 case WRITE_SAME_16: 2781 case UNMAP: 2782 /* 2783 * The upper level CTL code will filter out any CDBs with 2784 * the immediate bit set and return the proper error. 2785 * 2786 * We don't really need to worry about what LBA range the 2787 * user asked to be synced out. When they issue a sync 2788 * cache command, we'll sync out the whole thing. 2789 */ 2790 mtx_lock(&be_lun->queue_lock); 2791 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2792 links); 2793 mtx_unlock(&be_lun->queue_lock); 2794 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2795 break; 2796 case START_STOP_UNIT: { 2797 struct scsi_start_stop_unit *cdb; 2798 2799 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2800 2801 if (cdb->how & SSS_START) 2802 retval = ctl_start_lun(ctl_be_lun); 2803 else { 2804 retval = ctl_stop_lun(ctl_be_lun); 2805 /* 2806 * XXX KDM Copan-specific offline behavior. 2807 * Figure out a reasonable way to port this? 2808 */ 2809#ifdef NEEDTOPORT 2810 if ((retval == 0) 2811 && (cdb->byte2 & SSS_ONOFFLINE)) 2812 retval = ctl_lun_offline(ctl_be_lun); 2813#endif 2814 } 2815 2816 /* 2817 * In general, the above routines should not fail. They 2818 * just set state for the LUN. So we've got something 2819 * pretty wrong here if we can't start or stop the LUN. 2820 */ 2821 if (retval != 0) { 2822 ctl_set_internal_failure(&io->scsiio, 2823 /*sks_valid*/ 1, 2824 /*retry_count*/ 0xf051); 2825 retval = CTL_RETVAL_COMPLETE; 2826 } else { 2827 ctl_set_success(&io->scsiio); 2828 } 2829 ctl_config_write_done(io); 2830 break; 2831 } 2832 default: 2833 ctl_set_invalid_opcode(&io->scsiio); 2834 ctl_config_write_done(io); 2835 retval = CTL_RETVAL_COMPLETE; 2836 break; 2837 } 2838 2839 return (retval); 2840} 2841 2842static int 2843ctl_be_block_config_read(union ctl_io *io) 2844{ 2845 struct ctl_be_block_lun *be_lun; 2846 struct ctl_be_lun *ctl_be_lun; 2847 int retval = 0; 2848 2849 DPRINTF("entered\n"); 2850 2851 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2852 CTL_PRIV_BACKEND_LUN].ptr; 2853 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2854 2855 switch (io->scsiio.cdb[0]) { 2856 case SERVICE_ACTION_IN: 2857 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2858 mtx_lock(&be_lun->queue_lock); 2859 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2860 &io->io_hdr, links); 2861 mtx_unlock(&be_lun->queue_lock); 2862 taskqueue_enqueue(be_lun->io_taskqueue, 2863 &be_lun->io_task); 2864 retval = CTL_RETVAL_QUEUED; 2865 break; 2866 } 2867 ctl_set_invalid_field(&io->scsiio, 2868 /*sks_valid*/ 1, 2869 /*command*/ 1, 2870 /*field*/ 1, 2871 /*bit_valid*/ 1, 2872 /*bit*/ 4); 2873 ctl_config_read_done(io); 2874 retval = CTL_RETVAL_COMPLETE; 2875 break; 2876 default: 2877 ctl_set_invalid_opcode(&io->scsiio); 2878 ctl_config_read_done(io); 2879 retval = CTL_RETVAL_COMPLETE; 2880 break; 2881 } 2882 2883 return (retval); 2884} 2885 2886static int 2887ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2888{ 2889 struct ctl_be_block_lun *lun; 2890 int retval; 2891 2892 lun = (struct ctl_be_block_lun *)be_lun; 2893 retval = 0; 2894 2895 retval = sbuf_printf(sb, "\t<num_threads>"); 2896 2897 if (retval != 0) 2898 goto bailout; 2899 2900 retval = sbuf_printf(sb, "%d", lun->num_threads); 2901 2902 if (retval != 0) 2903 goto bailout; 2904 2905 retval = sbuf_printf(sb, "</num_threads>\n"); 2906 2907bailout: 2908 2909 return (retval); 2910} 2911 2912static uint64_t 2913ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2914{ 2915 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2916 2917 if (lun->getattr == NULL) 2918 return (UINT64_MAX); 2919 return (lun->getattr(lun, attrname)); 2920} 2921 2922int 2923ctl_be_block_init(void) 2924{ 2925 struct ctl_be_block_softc *softc; 2926 int retval; 2927 2928 softc = &backend_block_softc; 2929 retval = 0; 2930 2931 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2932 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2933 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2934 STAILQ_INIT(&softc->disk_list); 2935 STAILQ_INIT(&softc->lun_list); 2936 2937 return (retval); 2938} 2939