ctl_backend_block.c revision 288706
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 288706 2015-10-05 08:09:34Z 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 uint64_t val; 813 int error; 814 815 val = UINT64_MAX; 816 if (be_lun->vn == NULL) 817 return (val); 818 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 819 if (strcmp(attrname, "blocksused") == 0) { 820 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 821 if (error == 0) 822 val = vattr.va_bytes >> be_lun->blocksize_shift; 823 } 824 if (strcmp(attrname, "blocksavail") == 0 && 825 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 826 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 827 if (error == 0) 828 val = (statfs.f_bavail * statfs.f_bsize) >> 829 be_lun->blocksize_shift; 830 } 831 VOP_UNLOCK(be_lun->vn, 0); 832 return (val); 833} 834 835static void 836ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 837 struct ctl_be_block_io *beio) 838{ 839 struct ctl_be_block_devdata *dev_data; 840 union ctl_io *io; 841 struct uio xuio; 842 struct iovec *xiovec; 843 int flags; 844 int error, i; 845 846 DPRINTF("entered\n"); 847 848 dev_data = &be_lun->backend.dev; 849 io = beio->io; 850 flags = 0; 851 if (ARGS(io)->flags & CTL_LLF_DPO) 852 flags |= IO_DIRECT; 853 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 854 flags |= IO_SYNC; 855 856 bzero(&xuio, sizeof(xuio)); 857 if (beio->bio_cmd == BIO_READ) { 858 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 859 xuio.uio_rw = UIO_READ; 860 } else { 861 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 862 xuio.uio_rw = UIO_WRITE; 863 } 864 xuio.uio_offset = beio->io_offset; 865 xuio.uio_resid = beio->io_len; 866 xuio.uio_segflg = UIO_SYSSPACE; 867 xuio.uio_iov = beio->xiovecs; 868 xuio.uio_iovcnt = beio->num_segs; 869 xuio.uio_td = curthread; 870 871 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 872 xiovec->iov_base = beio->sg_segs[i].addr; 873 xiovec->iov_len = beio->sg_segs[i].len; 874 } 875 876 binuptime(&beio->ds_t0); 877 mtx_lock(&be_lun->io_lock); 878 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 879 mtx_unlock(&be_lun->io_lock); 880 881 if (beio->bio_cmd == BIO_READ) { 882 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 883 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 884 } else { 885 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 886 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 887 } 888 889 mtx_lock(&be_lun->io_lock); 890 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 891 beio->ds_tag_type, beio->ds_trans_type, 892 /*now*/ NULL, /*then*/&beio->ds_t0); 893 mtx_unlock(&be_lun->io_lock); 894 895 /* 896 * If we got an error, set the sense data to "MEDIUM ERROR" and 897 * return the I/O to the user. 898 */ 899 if (error != 0) { 900 if (error == ENOSPC || error == EDQUOT) { 901 ctl_set_space_alloc_fail(&io->scsiio); 902 } else 903 ctl_set_medium_error(&io->scsiio); 904 ctl_complete_beio(beio); 905 return; 906 } 907 908 /* 909 * If this is a write or a verify, we're all done. 910 * If this is a read, we can now send the data to the user. 911 */ 912 if ((beio->bio_cmd == BIO_WRITE) || 913 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 914 ctl_set_success(&io->scsiio); 915 ctl_complete_beio(beio); 916 } else { 917 if ((ARGS(io)->flags & CTL_LLF_READ) && 918 beio->beio_cont == NULL) 919 ctl_set_success(&io->scsiio); 920#ifdef CTL_TIME_IO 921 getbintime(&io->io_hdr.dma_start_bt); 922#endif 923 ctl_datamove(io); 924 } 925} 926 927static void 928ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 929 struct ctl_be_block_io *beio) 930{ 931 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 932 union ctl_io *io = beio->io; 933 struct ctl_lba_len_flags *lbalen = ARGS(io); 934 struct scsi_get_lba_status_data *data; 935 off_t roff, off; 936 int error, status; 937 938 DPRINTF("entered\n"); 939 940 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 941 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE, 942 (caddr_t)&off, FREAD, curthread); 943 if (error == 0 && off > roff) 944 status = 0; /* mapped up to off */ 945 else { 946 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA, 947 (caddr_t)&off, FREAD, curthread); 948 if (error == 0 && off > roff) 949 status = 1; /* deallocated up to off */ 950 else { 951 status = 0; /* unknown up to the end */ 952 off = be_lun->size_bytes; 953 } 954 } 955 956 off >>= be_lun->blocksize_shift; 957 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 958 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 959 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 960 data->descr[0].length); 961 data->descr[0].status = status; 962 963 ctl_complete_beio(beio); 964} 965 966static void 967ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 968 struct ctl_be_block_io *beio) 969{ 970 struct bio *bio; 971 union ctl_io *io; 972 struct ctl_be_block_devdata *dev_data; 973 974 dev_data = &be_lun->backend.dev; 975 io = beio->io; 976 977 DPRINTF("entered\n"); 978 979 /* This can't fail, it's a blocking allocation. */ 980 bio = g_alloc_bio(); 981 982 bio->bio_cmd = BIO_FLUSH; 983 bio->bio_dev = dev_data->cdev; 984 bio->bio_offset = 0; 985 bio->bio_data = 0; 986 bio->bio_done = ctl_be_block_biodone; 987 bio->bio_caller1 = beio; 988 bio->bio_pblkno = 0; 989 990 /* 991 * We don't need to acquire the LUN lock here, because we are only 992 * sending one bio, and so there is no other context to synchronize 993 * with. 994 */ 995 beio->num_bios_sent = 1; 996 beio->send_complete = 1; 997 998 binuptime(&beio->ds_t0); 999 mtx_lock(&be_lun->io_lock); 1000 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1001 mtx_unlock(&be_lun->io_lock); 1002 1003 (*dev_data->csw->d_strategy)(bio); 1004} 1005 1006static void 1007ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1008 struct ctl_be_block_io *beio, 1009 uint64_t off, uint64_t len, int last) 1010{ 1011 struct bio *bio; 1012 struct ctl_be_block_devdata *dev_data; 1013 uint64_t maxlen; 1014 1015 dev_data = &be_lun->backend.dev; 1016 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 1017 while (len > 0) { 1018 bio = g_alloc_bio(); 1019 bio->bio_cmd = BIO_DELETE; 1020 bio->bio_dev = dev_data->cdev; 1021 bio->bio_offset = off; 1022 bio->bio_length = MIN(len, maxlen); 1023 bio->bio_data = 0; 1024 bio->bio_done = ctl_be_block_biodone; 1025 bio->bio_caller1 = beio; 1026 bio->bio_pblkno = off / be_lun->blocksize; 1027 1028 off += bio->bio_length; 1029 len -= bio->bio_length; 1030 1031 mtx_lock(&be_lun->io_lock); 1032 beio->num_bios_sent++; 1033 if (last && len == 0) 1034 beio->send_complete = 1; 1035 mtx_unlock(&be_lun->io_lock); 1036 1037 (*dev_data->csw->d_strategy)(bio); 1038 } 1039} 1040 1041static void 1042ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1043 struct ctl_be_block_io *beio) 1044{ 1045 union ctl_io *io; 1046 struct ctl_be_block_devdata *dev_data; 1047 struct ctl_ptr_len_flags *ptrlen; 1048 struct scsi_unmap_desc *buf, *end; 1049 uint64_t len; 1050 1051 dev_data = &be_lun->backend.dev; 1052 io = beio->io; 1053 1054 DPRINTF("entered\n"); 1055 1056 binuptime(&beio->ds_t0); 1057 mtx_lock(&be_lun->io_lock); 1058 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1059 mtx_unlock(&be_lun->io_lock); 1060 1061 if (beio->io_offset == -1) { 1062 beio->io_len = 0; 1063 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1064 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1065 end = buf + ptrlen->len / sizeof(*buf); 1066 for (; buf < end; buf++) { 1067 len = (uint64_t)scsi_4btoul(buf->length) * 1068 be_lun->blocksize; 1069 beio->io_len += len; 1070 ctl_be_block_unmap_dev_range(be_lun, beio, 1071 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 1072 (end - buf < 2) ? TRUE : FALSE); 1073 } 1074 } else 1075 ctl_be_block_unmap_dev_range(be_lun, beio, 1076 beio->io_offset, beio->io_len, TRUE); 1077} 1078 1079static void 1080ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1081 struct ctl_be_block_io *beio) 1082{ 1083 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1084 int i; 1085 struct bio *bio; 1086 struct ctl_be_block_devdata *dev_data; 1087 off_t cur_offset; 1088 int max_iosize; 1089 1090 DPRINTF("entered\n"); 1091 1092 dev_data = &be_lun->backend.dev; 1093 1094 /* 1095 * We have to limit our I/O size to the maximum supported by the 1096 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1097 * set it properly, use DFLTPHYS. 1098 */ 1099 max_iosize = dev_data->cdev->si_iosize_max; 1100 if (max_iosize < PAGE_SIZE) 1101 max_iosize = DFLTPHYS; 1102 1103 cur_offset = beio->io_offset; 1104 for (i = 0; i < beio->num_segs; i++) { 1105 size_t cur_size; 1106 uint8_t *cur_ptr; 1107 1108 cur_size = beio->sg_segs[i].len; 1109 cur_ptr = beio->sg_segs[i].addr; 1110 1111 while (cur_size > 0) { 1112 /* This can't fail, it's a blocking allocation. */ 1113 bio = g_alloc_bio(); 1114 1115 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1116 1117 bio->bio_cmd = beio->bio_cmd; 1118 bio->bio_dev = dev_data->cdev; 1119 bio->bio_caller1 = beio; 1120 bio->bio_length = min(cur_size, max_iosize); 1121 bio->bio_offset = cur_offset; 1122 bio->bio_data = cur_ptr; 1123 bio->bio_done = ctl_be_block_biodone; 1124 bio->bio_pblkno = cur_offset / be_lun->blocksize; 1125 1126 cur_offset += bio->bio_length; 1127 cur_ptr += bio->bio_length; 1128 cur_size -= bio->bio_length; 1129 1130 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1131 beio->num_bios_sent++; 1132 } 1133 } 1134 binuptime(&beio->ds_t0); 1135 mtx_lock(&be_lun->io_lock); 1136 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1137 beio->send_complete = 1; 1138 mtx_unlock(&be_lun->io_lock); 1139 1140 /* 1141 * Fire off all allocated requests! 1142 */ 1143 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1144 TAILQ_REMOVE(&queue, bio, bio_queue); 1145 (*dev_data->csw->d_strategy)(bio); 1146 } 1147} 1148 1149static uint64_t 1150ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1151{ 1152 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1153 struct diocgattr_arg arg; 1154 int error; 1155 1156 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1157 return (UINT64_MAX); 1158 strlcpy(arg.name, attrname, sizeof(arg.name)); 1159 arg.len = sizeof(arg.value.off); 1160 error = dev_data->csw->d_ioctl(dev_data->cdev, 1161 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1162 if (error != 0) 1163 return (UINT64_MAX); 1164 return (arg.value.off); 1165} 1166 1167static void 1168ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1169 union ctl_io *io) 1170{ 1171 struct ctl_be_block_io *beio; 1172 struct ctl_lba_len_flags *lbalen; 1173 1174 DPRINTF("entered\n"); 1175 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1176 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1177 1178 beio->io_len = lbalen->len * be_lun->blocksize; 1179 beio->io_offset = lbalen->lba * be_lun->blocksize; 1180 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1181 beio->bio_cmd = BIO_FLUSH; 1182 beio->ds_trans_type = DEVSTAT_NO_DATA; 1183 DPRINTF("SYNC\n"); 1184 be_lun->lun_flush(be_lun, beio); 1185} 1186 1187static void 1188ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1189{ 1190 union ctl_io *io; 1191 1192 io = beio->io; 1193 ctl_free_beio(beio); 1194 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1195 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1196 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1197 ctl_config_write_done(io); 1198 return; 1199 } 1200 1201 ctl_be_block_config_write(io); 1202} 1203 1204static void 1205ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1206 union ctl_io *io) 1207{ 1208 struct ctl_be_block_io *beio; 1209 struct ctl_lba_len_flags *lbalen; 1210 uint64_t len_left, lba; 1211 uint32_t pb, pbo, adj; 1212 int i, seglen; 1213 uint8_t *buf, *end; 1214 1215 DPRINTF("entered\n"); 1216 1217 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1218 lbalen = ARGS(beio->io); 1219 1220 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1221 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1222 ctl_free_beio(beio); 1223 ctl_set_invalid_field(&io->scsiio, 1224 /*sks_valid*/ 1, 1225 /*command*/ 1, 1226 /*field*/ 1, 1227 /*bit_valid*/ 0, 1228 /*bit*/ 0); 1229 ctl_config_write_done(io); 1230 return; 1231 } 1232 1233 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1234 beio->io_offset = lbalen->lba * be_lun->blocksize; 1235 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1236 beio->bio_cmd = BIO_DELETE; 1237 beio->ds_trans_type = DEVSTAT_FREE; 1238 1239 be_lun->unmap(be_lun, beio); 1240 return; 1241 } 1242 1243 beio->bio_cmd = BIO_WRITE; 1244 beio->ds_trans_type = DEVSTAT_WRITE; 1245 1246 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1247 (uintmax_t)lbalen->lba, lbalen->len); 1248 1249 pb = be_lun->blocksize << be_lun->pblockexp; 1250 if (be_lun->pblockoff > 0) 1251 pbo = pb - be_lun->blocksize * be_lun->pblockoff; 1252 else 1253 pbo = 0; 1254 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1255 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1256 1257 /* 1258 * Setup the S/G entry for this chunk. 1259 */ 1260 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1261 if (pb > be_lun->blocksize) { 1262 adj = ((lbalen->lba + lba) * be_lun->blocksize + 1263 seglen - pbo) % pb; 1264 if (seglen > adj) 1265 seglen -= adj; 1266 else 1267 seglen -= seglen % be_lun->blocksize; 1268 } else 1269 seglen -= seglen % be_lun->blocksize; 1270 beio->sg_segs[i].len = seglen; 1271 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1272 1273 DPRINTF("segment %d addr %p len %zd\n", i, 1274 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1275 1276 beio->num_segs++; 1277 len_left -= seglen; 1278 1279 buf = beio->sg_segs[i].addr; 1280 end = buf + seglen; 1281 for (; buf < end; buf += be_lun->blocksize) { 1282 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1283 if (lbalen->flags & SWS_LBDATA) 1284 scsi_ulto4b(lbalen->lba + lba, buf); 1285 lba++; 1286 } 1287 } 1288 1289 beio->io_offset = lbalen->lba * be_lun->blocksize; 1290 beio->io_len = lba * be_lun->blocksize; 1291 1292 /* We can not do all in one run. Correct and schedule rerun. */ 1293 if (len_left > 0) { 1294 lbalen->lba += lba; 1295 lbalen->len -= lba; 1296 beio->beio_cont = ctl_be_block_cw_done_ws; 1297 } 1298 1299 be_lun->dispatch(be_lun, beio); 1300} 1301 1302static void 1303ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1304 union ctl_io *io) 1305{ 1306 struct ctl_be_block_io *beio; 1307 struct ctl_ptr_len_flags *ptrlen; 1308 1309 DPRINTF("entered\n"); 1310 1311 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1312 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1313 1314 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1315 ctl_free_beio(beio); 1316 ctl_set_invalid_field(&io->scsiio, 1317 /*sks_valid*/ 0, 1318 /*command*/ 1, 1319 /*field*/ 0, 1320 /*bit_valid*/ 0, 1321 /*bit*/ 0); 1322 ctl_config_write_done(io); 1323 return; 1324 } 1325 1326 beio->io_len = 0; 1327 beio->io_offset = -1; 1328 beio->bio_cmd = BIO_DELETE; 1329 beio->ds_trans_type = DEVSTAT_FREE; 1330 DPRINTF("UNMAP\n"); 1331 be_lun->unmap(be_lun, beio); 1332} 1333 1334static void 1335ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1336{ 1337 union ctl_io *io; 1338 1339 io = beio->io; 1340 ctl_free_beio(beio); 1341 ctl_config_read_done(io); 1342} 1343 1344static void 1345ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1346 union ctl_io *io) 1347{ 1348 struct ctl_be_block_io *beio; 1349 struct ctl_be_block_softc *softc; 1350 1351 DPRINTF("entered\n"); 1352 1353 softc = be_lun->softc; 1354 beio = ctl_alloc_beio(softc); 1355 beio->io = io; 1356 beio->lun = be_lun; 1357 beio->beio_cont = ctl_be_block_cr_done; 1358 PRIV(io)->ptr = (void *)beio; 1359 1360 switch (io->scsiio.cdb[0]) { 1361 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1362 beio->bio_cmd = -1; 1363 beio->ds_trans_type = DEVSTAT_NO_DATA; 1364 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1365 beio->io_len = 0; 1366 if (be_lun->get_lba_status) 1367 be_lun->get_lba_status(be_lun, beio); 1368 else 1369 ctl_be_block_cr_done(beio); 1370 break; 1371 default: 1372 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1373 break; 1374 } 1375} 1376 1377static void 1378ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1379{ 1380 union ctl_io *io; 1381 1382 io = beio->io; 1383 ctl_free_beio(beio); 1384 ctl_config_write_done(io); 1385} 1386 1387static void 1388ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1389 union ctl_io *io) 1390{ 1391 struct ctl_be_block_io *beio; 1392 struct ctl_be_block_softc *softc; 1393 1394 DPRINTF("entered\n"); 1395 1396 softc = be_lun->softc; 1397 beio = ctl_alloc_beio(softc); 1398 beio->io = io; 1399 beio->lun = be_lun; 1400 beio->beio_cont = ctl_be_block_cw_done; 1401 switch (io->scsiio.tag_type) { 1402 case CTL_TAG_ORDERED: 1403 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1404 break; 1405 case CTL_TAG_HEAD_OF_QUEUE: 1406 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1407 break; 1408 case CTL_TAG_UNTAGGED: 1409 case CTL_TAG_SIMPLE: 1410 case CTL_TAG_ACA: 1411 default: 1412 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1413 break; 1414 } 1415 PRIV(io)->ptr = (void *)beio; 1416 1417 switch (io->scsiio.cdb[0]) { 1418 case SYNCHRONIZE_CACHE: 1419 case SYNCHRONIZE_CACHE_16: 1420 ctl_be_block_cw_dispatch_sync(be_lun, io); 1421 break; 1422 case WRITE_SAME_10: 1423 case WRITE_SAME_16: 1424 ctl_be_block_cw_dispatch_ws(be_lun, io); 1425 break; 1426 case UNMAP: 1427 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1428 break; 1429 default: 1430 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1431 break; 1432 } 1433} 1434 1435SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1436SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1437SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1438SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1439 1440static void 1441ctl_be_block_next(struct ctl_be_block_io *beio) 1442{ 1443 struct ctl_be_block_lun *be_lun; 1444 union ctl_io *io; 1445 1446 io = beio->io; 1447 be_lun = beio->lun; 1448 ctl_free_beio(beio); 1449 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1450 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1451 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1452 ctl_data_submit_done(io); 1453 return; 1454 } 1455 1456 io->io_hdr.status &= ~CTL_STATUS_MASK; 1457 io->io_hdr.status |= CTL_STATUS_NONE; 1458 1459 mtx_lock(&be_lun->queue_lock); 1460 /* 1461 * XXX KDM make sure that links is okay to use at this point. 1462 * Otherwise, we either need to add another field to ctl_io_hdr, 1463 * or deal with resource allocation here. 1464 */ 1465 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1466 mtx_unlock(&be_lun->queue_lock); 1467 1468 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1469} 1470 1471static void 1472ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1473 union ctl_io *io) 1474{ 1475 struct ctl_be_block_io *beio; 1476 struct ctl_be_block_softc *softc; 1477 struct ctl_lba_len_flags *lbalen; 1478 struct ctl_ptr_len_flags *bptrlen; 1479 uint64_t len_left, lbas; 1480 int i; 1481 1482 softc = be_lun->softc; 1483 1484 DPRINTF("entered\n"); 1485 1486 lbalen = ARGS(io); 1487 if (lbalen->flags & CTL_LLF_WRITE) { 1488 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1489 } else { 1490 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1491 } 1492 1493 beio = ctl_alloc_beio(softc); 1494 beio->io = io; 1495 beio->lun = be_lun; 1496 bptrlen = PRIV(io); 1497 bptrlen->ptr = (void *)beio; 1498 1499 switch (io->scsiio.tag_type) { 1500 case CTL_TAG_ORDERED: 1501 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1502 break; 1503 case CTL_TAG_HEAD_OF_QUEUE: 1504 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1505 break; 1506 case CTL_TAG_UNTAGGED: 1507 case CTL_TAG_SIMPLE: 1508 case CTL_TAG_ACA: 1509 default: 1510 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1511 break; 1512 } 1513 1514 if (lbalen->flags & CTL_LLF_WRITE) { 1515 beio->bio_cmd = BIO_WRITE; 1516 beio->ds_trans_type = DEVSTAT_WRITE; 1517 } else { 1518 beio->bio_cmd = BIO_READ; 1519 beio->ds_trans_type = DEVSTAT_READ; 1520 } 1521 1522 DPRINTF("%s at LBA %jx len %u @%ju\n", 1523 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1524 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1525 if (lbalen->flags & CTL_LLF_COMPARE) 1526 lbas = CTLBLK_HALF_IO_SIZE; 1527 else 1528 lbas = CTLBLK_MAX_IO_SIZE; 1529 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1530 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1531 beio->io_len = lbas * be_lun->blocksize; 1532 bptrlen->len += lbas; 1533 1534 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1535 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1536 i, CTLBLK_MAX_SEGS)); 1537 1538 /* 1539 * Setup the S/G entry for this chunk. 1540 */ 1541 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1542 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1543 1544 DPRINTF("segment %d addr %p len %zd\n", i, 1545 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1546 1547 /* Set up second segment for compare operation. */ 1548 if (lbalen->flags & CTL_LLF_COMPARE) { 1549 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1550 beio->sg_segs[i].len; 1551 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1552 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1553 } 1554 1555 beio->num_segs++; 1556 len_left -= beio->sg_segs[i].len; 1557 } 1558 if (bptrlen->len < lbalen->len) 1559 beio->beio_cont = ctl_be_block_next; 1560 io->scsiio.be_move_done = ctl_be_block_move_done; 1561 /* For compare we have separate S/G lists for read and datamove. */ 1562 if (lbalen->flags & CTL_LLF_COMPARE) 1563 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1564 else 1565 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1566 io->scsiio.kern_data_len = beio->io_len; 1567 io->scsiio.kern_data_resid = 0; 1568 io->scsiio.kern_sg_entries = beio->num_segs; 1569 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1570 1571 /* 1572 * For the read case, we need to read the data into our buffers and 1573 * then we can send it back to the user. For the write case, we 1574 * need to get the data from the user first. 1575 */ 1576 if (beio->bio_cmd == BIO_READ) { 1577 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1578 be_lun->dispatch(be_lun, beio); 1579 } else { 1580 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1581#ifdef CTL_TIME_IO 1582 getbintime(&io->io_hdr.dma_start_bt); 1583#endif 1584 ctl_datamove(io); 1585 } 1586} 1587 1588static void 1589ctl_be_block_worker(void *context, int pending) 1590{ 1591 struct ctl_be_block_lun *be_lun; 1592 struct ctl_be_block_softc *softc; 1593 union ctl_io *io; 1594 1595 be_lun = (struct ctl_be_block_lun *)context; 1596 softc = be_lun->softc; 1597 1598 DPRINTF("entered\n"); 1599 1600 mtx_lock(&be_lun->queue_lock); 1601 for (;;) { 1602 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1603 if (io != NULL) { 1604 struct ctl_be_block_io *beio; 1605 1606 DPRINTF("datamove queue\n"); 1607 1608 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1609 ctl_io_hdr, links); 1610 1611 mtx_unlock(&be_lun->queue_lock); 1612 1613 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1614 1615 be_lun->dispatch(be_lun, beio); 1616 1617 mtx_lock(&be_lun->queue_lock); 1618 continue; 1619 } 1620 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1621 if (io != NULL) { 1622 DPRINTF("config write queue\n"); 1623 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1624 ctl_io_hdr, links); 1625 mtx_unlock(&be_lun->queue_lock); 1626 ctl_be_block_cw_dispatch(be_lun, io); 1627 mtx_lock(&be_lun->queue_lock); 1628 continue; 1629 } 1630 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1631 if (io != NULL) { 1632 DPRINTF("config read queue\n"); 1633 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1634 ctl_io_hdr, links); 1635 mtx_unlock(&be_lun->queue_lock); 1636 ctl_be_block_cr_dispatch(be_lun, io); 1637 mtx_lock(&be_lun->queue_lock); 1638 continue; 1639 } 1640 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1641 if (io != NULL) { 1642 DPRINTF("input queue\n"); 1643 1644 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1645 ctl_io_hdr, links); 1646 mtx_unlock(&be_lun->queue_lock); 1647 1648 /* 1649 * We must drop the lock, since this routine and 1650 * its children may sleep. 1651 */ 1652 ctl_be_block_dispatch(be_lun, io); 1653 1654 mtx_lock(&be_lun->queue_lock); 1655 continue; 1656 } 1657 1658 /* 1659 * If we get here, there is no work left in the queues, so 1660 * just break out and let the task queue go to sleep. 1661 */ 1662 break; 1663 } 1664 mtx_unlock(&be_lun->queue_lock); 1665} 1666 1667/* 1668 * Entry point from CTL to the backend for I/O. We queue everything to a 1669 * work thread, so this just puts the I/O on a queue and wakes up the 1670 * thread. 1671 */ 1672static int 1673ctl_be_block_submit(union ctl_io *io) 1674{ 1675 struct ctl_be_block_lun *be_lun; 1676 struct ctl_be_lun *ctl_be_lun; 1677 1678 DPRINTF("entered\n"); 1679 1680 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1681 CTL_PRIV_BACKEND_LUN].ptr; 1682 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1683 1684 /* 1685 * Make sure we only get SCSI I/O. 1686 */ 1687 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1688 "%#x) encountered", io->io_hdr.io_type)); 1689 1690 PRIV(io)->len = 0; 1691 1692 mtx_lock(&be_lun->queue_lock); 1693 /* 1694 * XXX KDM make sure that links is okay to use at this point. 1695 * Otherwise, we either need to add another field to ctl_io_hdr, 1696 * or deal with resource allocation here. 1697 */ 1698 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1699 mtx_unlock(&be_lun->queue_lock); 1700 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1701 1702 return (CTL_RETVAL_COMPLETE); 1703} 1704 1705static int 1706ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1707 int flag, struct thread *td) 1708{ 1709 struct ctl_be_block_softc *softc; 1710 int error; 1711 1712 softc = &backend_block_softc; 1713 1714 error = 0; 1715 1716 switch (cmd) { 1717 case CTL_LUN_REQ: { 1718 struct ctl_lun_req *lun_req; 1719 1720 lun_req = (struct ctl_lun_req *)addr; 1721 1722 switch (lun_req->reqtype) { 1723 case CTL_LUNREQ_CREATE: 1724 error = ctl_be_block_create(softc, lun_req); 1725 break; 1726 case CTL_LUNREQ_RM: 1727 error = ctl_be_block_rm(softc, lun_req); 1728 break; 1729 case CTL_LUNREQ_MODIFY: 1730 error = ctl_be_block_modify(softc, lun_req); 1731 break; 1732 default: 1733 lun_req->status = CTL_LUN_ERROR; 1734 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1735 "invalid LUN request type %d", 1736 lun_req->reqtype); 1737 break; 1738 } 1739 break; 1740 } 1741 default: 1742 error = ENOTTY; 1743 break; 1744 } 1745 1746 return (error); 1747} 1748 1749static int 1750ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1751{ 1752 struct ctl_be_block_filedata *file_data; 1753 struct ctl_lun_create_params *params; 1754 char *value; 1755 struct vattr vattr; 1756 off_t ps, pss, po, pos, us, uss, uo, uos; 1757 int error; 1758 1759 error = 0; 1760 file_data = &be_lun->backend.file; 1761 params = &be_lun->params; 1762 1763 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1764 be_lun->dispatch = ctl_be_block_dispatch_file; 1765 be_lun->lun_flush = ctl_be_block_flush_file; 1766 be_lun->get_lba_status = ctl_be_block_gls_file; 1767 be_lun->getattr = ctl_be_block_getattr_file; 1768 1769 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1770 if (error != 0) { 1771 snprintf(req->error_str, sizeof(req->error_str), 1772 "error calling VOP_GETATTR() for file %s", 1773 be_lun->dev_path); 1774 return (error); 1775 } 1776 1777 /* 1778 * Verify that we have the ability to upgrade to exclusive 1779 * access on this file so we can trap errors at open instead 1780 * of reporting them during first access. 1781 */ 1782 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1783 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1784 if (be_lun->vn->v_iflag & VI_DOOMED) { 1785 error = EBADF; 1786 snprintf(req->error_str, sizeof(req->error_str), 1787 "error locking file %s", be_lun->dev_path); 1788 return (error); 1789 } 1790 } 1791 1792 1793 file_data->cred = crhold(curthread->td_ucred); 1794 if (params->lun_size_bytes != 0) 1795 be_lun->size_bytes = params->lun_size_bytes; 1796 else 1797 be_lun->size_bytes = vattr.va_size; 1798 /* 1799 * We set the multi thread flag for file operations because all 1800 * filesystems (in theory) are capable of allowing multiple readers 1801 * of a file at once. So we want to get the maximum possible 1802 * concurrency. 1803 */ 1804 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1805 1806 /* 1807 * For files we can use any logical block size. Prefer 512 bytes 1808 * for compatibility reasons. If file's vattr.va_blocksize 1809 * (preferred I/O block size) is bigger and multiple to chosen 1810 * logical block size -- report it as physical block size. 1811 */ 1812 if (params->blocksize_bytes != 0) 1813 be_lun->blocksize = params->blocksize_bytes; 1814 else 1815 be_lun->blocksize = 512; 1816 1817 us = ps = vattr.va_blocksize; 1818 uo = po = 0; 1819 1820 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 1821 if (value != NULL) 1822 ctl_expand_number(value, &ps); 1823 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 1824 if (value != NULL) 1825 ctl_expand_number(value, &po); 1826 pss = ps / be_lun->blocksize; 1827 pos = po / be_lun->blocksize; 1828 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1829 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1830 be_lun->pblockexp = fls(pss) - 1; 1831 be_lun->pblockoff = (pss - pos) % pss; 1832 } 1833 1834 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 1835 if (value != NULL) 1836 ctl_expand_number(value, &us); 1837 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 1838 if (value != NULL) 1839 ctl_expand_number(value, &uo); 1840 uss = us / be_lun->blocksize; 1841 uos = uo / be_lun->blocksize; 1842 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 1843 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 1844 be_lun->ublockexp = fls(uss) - 1; 1845 be_lun->ublockoff = (uss - uos) % uss; 1846 } 1847 1848 /* 1849 * Sanity check. The media size has to be at least one 1850 * sector long. 1851 */ 1852 if (be_lun->size_bytes < be_lun->blocksize) { 1853 error = EINVAL; 1854 snprintf(req->error_str, sizeof(req->error_str), 1855 "file %s size %ju < block size %u", be_lun->dev_path, 1856 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1857 } 1858 1859 be_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / be_lun->blocksize; 1860 return (error); 1861} 1862 1863static int 1864ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1865{ 1866 struct ctl_lun_create_params *params; 1867 struct vattr vattr; 1868 struct cdev *dev; 1869 struct cdevsw *devsw; 1870 char *value; 1871 int error, atomic, maxio, unmap; 1872 off_t ps, pss, po, pos, us, uss, uo, uos; 1873 1874 params = &be_lun->params; 1875 1876 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1877 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1878 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1879 &be_lun->backend.dev.dev_ref); 1880 if (be_lun->backend.dev.csw == NULL) 1881 panic("Unable to retrieve device switch"); 1882 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) { 1883 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1884 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1885 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1886 } else { 1887 be_lun->dispatch = ctl_be_block_dispatch_dev; 1888 atomic = 0; 1889 maxio = be_lun->backend.dev.cdev->si_iosize_max; 1890 if (maxio <= 0) 1891 maxio = DFLTPHYS; 1892 if (maxio > CTLBLK_MAX_IO_SIZE) 1893 maxio = CTLBLK_MAX_IO_SIZE; 1894 } 1895 be_lun->lun_flush = ctl_be_block_flush_dev; 1896 be_lun->getattr = ctl_be_block_getattr_dev; 1897 1898 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1899 if (error) { 1900 snprintf(req->error_str, sizeof(req->error_str), 1901 "error getting vnode attributes for device %s", 1902 be_lun->dev_path); 1903 return (error); 1904 } 1905 1906 dev = be_lun->vn->v_rdev; 1907 devsw = dev->si_devsw; 1908 if (!devsw->d_ioctl) { 1909 snprintf(req->error_str, sizeof(req->error_str), 1910 "no d_ioctl for device %s!", 1911 be_lun->dev_path); 1912 return (ENODEV); 1913 } 1914 1915 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1916 (caddr_t)&be_lun->blocksize, FREAD, 1917 curthread); 1918 if (error) { 1919 snprintf(req->error_str, sizeof(req->error_str), 1920 "error %d returned for DIOCGSECTORSIZE ioctl " 1921 "on %s!", error, be_lun->dev_path); 1922 return (error); 1923 } 1924 1925 /* 1926 * If the user has asked for a blocksize that is greater than the 1927 * backing device's blocksize, we can do it only if the blocksize 1928 * the user is asking for is an even multiple of the underlying 1929 * device's blocksize. 1930 */ 1931 if ((params->blocksize_bytes != 0) 1932 && (params->blocksize_bytes > be_lun->blocksize)) { 1933 uint32_t bs_multiple, tmp_blocksize; 1934 1935 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1936 1937 tmp_blocksize = bs_multiple * be_lun->blocksize; 1938 1939 if (tmp_blocksize == params->blocksize_bytes) { 1940 be_lun->blocksize = params->blocksize_bytes; 1941 } else { 1942 snprintf(req->error_str, sizeof(req->error_str), 1943 "requested blocksize %u is not an even " 1944 "multiple of backing device blocksize %u", 1945 params->blocksize_bytes, 1946 be_lun->blocksize); 1947 return (EINVAL); 1948 1949 } 1950 } else if ((params->blocksize_bytes != 0) 1951 && (params->blocksize_bytes != be_lun->blocksize)) { 1952 snprintf(req->error_str, sizeof(req->error_str), 1953 "requested blocksize %u < backing device " 1954 "blocksize %u", params->blocksize_bytes, 1955 be_lun->blocksize); 1956 return (EINVAL); 1957 } 1958 1959 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1960 (caddr_t)&be_lun->size_bytes, FREAD, 1961 curthread); 1962 if (error) { 1963 snprintf(req->error_str, sizeof(req->error_str), 1964 "error %d returned for DIOCGMEDIASIZE " 1965 " ioctl on %s!", error, 1966 be_lun->dev_path); 1967 return (error); 1968 } 1969 1970 if (params->lun_size_bytes != 0) { 1971 if (params->lun_size_bytes > be_lun->size_bytes) { 1972 snprintf(req->error_str, sizeof(req->error_str), 1973 "requested LUN size %ju > backing device " 1974 "size %ju", 1975 (uintmax_t)params->lun_size_bytes, 1976 (uintmax_t)be_lun->size_bytes); 1977 return (EINVAL); 1978 } 1979 1980 be_lun->size_bytes = params->lun_size_bytes; 1981 } 1982 1983 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1984 (caddr_t)&ps, FREAD, curthread); 1985 if (error) 1986 ps = po = 0; 1987 else { 1988 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1989 (caddr_t)&po, FREAD, curthread); 1990 if (error) 1991 po = 0; 1992 } 1993 us = ps; 1994 uo = po; 1995 1996 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 1997 if (value != NULL) 1998 ctl_expand_number(value, &ps); 1999 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 2000 if (value != NULL) 2001 ctl_expand_number(value, &po); 2002 pss = ps / be_lun->blocksize; 2003 pos = po / be_lun->blocksize; 2004 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 2005 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 2006 be_lun->pblockexp = fls(pss) - 1; 2007 be_lun->pblockoff = (pss - pos) % pss; 2008 } 2009 2010 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 2011 if (value != NULL) 2012 ctl_expand_number(value, &us); 2013 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 2014 if (value != NULL) 2015 ctl_expand_number(value, &uo); 2016 uss = us / be_lun->blocksize; 2017 uos = uo / be_lun->blocksize; 2018 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 2019 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 2020 be_lun->ublockexp = fls(uss) - 1; 2021 be_lun->ublockoff = (uss - uos) % uss; 2022 } 2023 2024 be_lun->atomicblock = atomic / be_lun->blocksize; 2025 be_lun->opttxferlen = maxio / be_lun->blocksize; 2026 2027 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2028 unmap = 1; 2029 } else { 2030 struct diocgattr_arg arg; 2031 2032 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2033 arg.len = sizeof(arg.value.i); 2034 error = devsw->d_ioctl(dev, DIOCGATTR, 2035 (caddr_t)&arg, FREAD, curthread); 2036 unmap = (error == 0) ? arg.value.i : 0; 2037 } 2038 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2039 if (value != NULL) 2040 unmap = (strcmp(value, "on") == 0); 2041 if (unmap) 2042 be_lun->unmap = ctl_be_block_unmap_dev; 2043 2044 return (0); 2045} 2046 2047static int 2048ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2049{ 2050 DROP_GIANT(); 2051 if (be_lun->vn) { 2052 int flags = FREAD | FWRITE; 2053 2054 switch (be_lun->dev_type) { 2055 case CTL_BE_BLOCK_DEV: 2056 if (be_lun->backend.dev.csw) { 2057 dev_relthread(be_lun->backend.dev.cdev, 2058 be_lun->backend.dev.dev_ref); 2059 be_lun->backend.dev.csw = NULL; 2060 be_lun->backend.dev.cdev = NULL; 2061 } 2062 break; 2063 case CTL_BE_BLOCK_FILE: 2064 break; 2065 case CTL_BE_BLOCK_NONE: 2066 break; 2067 default: 2068 panic("Unexpected backend type."); 2069 break; 2070 } 2071 2072 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2073 be_lun->vn = NULL; 2074 2075 switch (be_lun->dev_type) { 2076 case CTL_BE_BLOCK_DEV: 2077 break; 2078 case CTL_BE_BLOCK_FILE: 2079 if (be_lun->backend.file.cred != NULL) { 2080 crfree(be_lun->backend.file.cred); 2081 be_lun->backend.file.cred = NULL; 2082 } 2083 break; 2084 case CTL_BE_BLOCK_NONE: 2085 break; 2086 default: 2087 panic("Unexpected backend type."); 2088 break; 2089 } 2090 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2091 } 2092 PICKUP_GIANT(); 2093 2094 return (0); 2095} 2096 2097static int 2098ctl_be_block_open(struct ctl_be_block_softc *softc, 2099 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2100{ 2101 struct nameidata nd; 2102 int flags; 2103 int error; 2104 2105 /* 2106 * XXX KDM allow a read-only option? 2107 */ 2108 flags = FREAD | FWRITE; 2109 error = 0; 2110 2111 if (rootvnode == NULL) { 2112 snprintf(req->error_str, sizeof(req->error_str), 2113 "Root filesystem is not mounted"); 2114 return (1); 2115 } 2116 2117 if (!curthread->td_proc->p_fd->fd_cdir) { 2118 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2119 VREF(rootvnode); 2120 } 2121 if (!curthread->td_proc->p_fd->fd_rdir) { 2122 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2123 VREF(rootvnode); 2124 } 2125 if (!curthread->td_proc->p_fd->fd_jdir) { 2126 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2127 VREF(rootvnode); 2128 } 2129 2130 again: 2131 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2132 error = vn_open(&nd, &flags, 0, NULL); 2133 if (error) { 2134 /* 2135 * This is the only reasonable guess we can make as far as 2136 * path if the user doesn't give us a fully qualified path. 2137 * If they want to specify a file, they need to specify the 2138 * full path. 2139 */ 2140 if (be_lun->dev_path[0] != '/') { 2141 char *dev_path = "/dev/"; 2142 char *dev_name; 2143 2144 /* Try adding device path at beginning of name */ 2145 dev_name = malloc(strlen(be_lun->dev_path) 2146 + strlen(dev_path) + 1, 2147 M_CTLBLK, M_WAITOK); 2148 if (dev_name) { 2149 sprintf(dev_name, "%s%s", dev_path, 2150 be_lun->dev_path); 2151 free(be_lun->dev_path, M_CTLBLK); 2152 be_lun->dev_path = dev_name; 2153 goto again; 2154 } 2155 } 2156 snprintf(req->error_str, sizeof(req->error_str), 2157 "error opening %s: %d", be_lun->dev_path, error); 2158 return (error); 2159 } 2160 2161 NDFREE(&nd, NDF_ONLY_PNBUF); 2162 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 return (error); 2180 } 2181 2182 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2183 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2184 2185 return (0); 2186} 2187 2188static int 2189ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2190{ 2191 struct ctl_be_block_lun *be_lun; 2192 struct ctl_lun_create_params *params; 2193 char num_thread_str[16]; 2194 char tmpstr[32]; 2195 char *value; 2196 int retval, num_threads; 2197 int tmp_num_threads; 2198 2199 params = &req->reqdata.create; 2200 retval = 0; 2201 req->status = CTL_LUN_OK; 2202 2203 num_threads = cbb_num_threads; 2204 2205 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2206 2207 be_lun->params = req->reqdata.create; 2208 be_lun->softc = softc; 2209 STAILQ_INIT(&be_lun->input_queue); 2210 STAILQ_INIT(&be_lun->config_read_queue); 2211 STAILQ_INIT(&be_lun->config_write_queue); 2212 STAILQ_INIT(&be_lun->datamove_queue); 2213 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2214 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2215 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2216 ctl_init_opts(&be_lun->ctl_be_lun.options, 2217 req->num_be_args, req->kern_be_args); 2218 2219 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2220 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2221 2222 if (be_lun->lun_zone == NULL) { 2223 snprintf(req->error_str, sizeof(req->error_str), 2224 "error allocating UMA zone"); 2225 goto bailout_error; 2226 } 2227 2228 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2229 be_lun->ctl_be_lun.lun_type = params->device_type; 2230 else 2231 be_lun->ctl_be_lun.lun_type = T_DIRECT; 2232 2233 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 2234 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 2235 if (value == NULL) { 2236 snprintf(req->error_str, sizeof(req->error_str), 2237 "no file argument specified"); 2238 goto bailout_error; 2239 } 2240 be_lun->dev_path = strdup(value, M_CTLBLK); 2241 be_lun->blocksize = 512; 2242 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2243 2244 retval = ctl_be_block_open(softc, be_lun, req); 2245 if (retval != 0) { 2246 retval = 0; 2247 req->status = CTL_LUN_WARNING; 2248 } 2249 } else { 2250 /* 2251 * For processor devices, we don't have any size. 2252 */ 2253 be_lun->blocksize = 0; 2254 be_lun->pblockexp = 0; 2255 be_lun->pblockoff = 0; 2256 be_lun->ublockexp = 0; 2257 be_lun->ublockoff = 0; 2258 be_lun->size_blocks = 0; 2259 be_lun->size_bytes = 0; 2260 be_lun->ctl_be_lun.maxlba = 0; 2261 2262 /* 2263 * Default to just 1 thread for processor devices. 2264 */ 2265 num_threads = 1; 2266 } 2267 2268 /* 2269 * XXX This searching loop might be refactored to be combined with 2270 * the loop above, 2271 */ 2272 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2273 if (value != NULL) { 2274 tmp_num_threads = strtol(value, NULL, 0); 2275 2276 /* 2277 * We don't let the user specify less than one 2278 * thread, but hope he's clueful enough not to 2279 * specify 1000 threads. 2280 */ 2281 if (tmp_num_threads < 1) { 2282 snprintf(req->error_str, sizeof(req->error_str), 2283 "invalid number of threads %s", 2284 num_thread_str); 2285 goto bailout_error; 2286 } 2287 num_threads = tmp_num_threads; 2288 } 2289 2290 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2291 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2292 if (be_lun->vn == NULL) 2293 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2294 if (be_lun->unmap != NULL) 2295 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2296 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2297 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_SERSEQ_READ; 2298 be_lun->ctl_be_lun.be_lun = be_lun; 2299 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2300 0 : (be_lun->size_blocks - 1); 2301 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2302 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2303 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2304 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2305 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2306 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2307 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2308 /* Tell the user the blocksize we ended up using */ 2309 params->lun_size_bytes = be_lun->size_bytes; 2310 params->blocksize_bytes = be_lun->blocksize; 2311 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2312 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2313 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2314 } else 2315 be_lun->ctl_be_lun.req_lun_id = 0; 2316 2317 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2318 be_lun->ctl_be_lun.lun_config_status = 2319 ctl_be_block_lun_config_status; 2320 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2321 2322 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2323 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2324 softc->num_luns); 2325 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2326 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2327 sizeof(tmpstr))); 2328 2329 /* Tell the user what we used for a serial number */ 2330 strncpy((char *)params->serial_num, tmpstr, 2331 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2332 } else { 2333 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2334 params->serial_num, 2335 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2336 sizeof(params->serial_num))); 2337 } 2338 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2339 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2340 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2341 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2342 sizeof(tmpstr))); 2343 2344 /* Tell the user what we used for a device ID */ 2345 strncpy((char *)params->device_id, tmpstr, 2346 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2347 } else { 2348 strncpy((char *)be_lun->ctl_be_lun.device_id, 2349 params->device_id, 2350 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2351 sizeof(params->device_id))); 2352 } 2353 2354 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2355 2356 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2357 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2358 2359 if (be_lun->io_taskqueue == NULL) { 2360 snprintf(req->error_str, sizeof(req->error_str), 2361 "unable to create taskqueue"); 2362 goto bailout_error; 2363 } 2364 2365 /* 2366 * Note that we start the same number of threads by default for 2367 * both the file case and the block device case. For the file 2368 * case, we need multiple threads to allow concurrency, because the 2369 * vnode interface is designed to be a blocking interface. For the 2370 * block device case, ZFS zvols at least will block the caller's 2371 * context in many instances, and so we need multiple threads to 2372 * overcome that problem. Other block devices don't need as many 2373 * threads, but they shouldn't cause too many problems. 2374 * 2375 * If the user wants to just have a single thread for a block 2376 * device, he can specify that when the LUN is created, or change 2377 * the tunable/sysctl to alter the default number of threads. 2378 */ 2379 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2380 /*num threads*/num_threads, 2381 /*priority*/PWAIT, 2382 /*thread name*/ 2383 "%s taskq", be_lun->lunname); 2384 2385 if (retval != 0) 2386 goto bailout_error; 2387 2388 be_lun->num_threads = num_threads; 2389 2390 mtx_lock(&softc->lock); 2391 softc->num_luns++; 2392 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2393 2394 mtx_unlock(&softc->lock); 2395 2396 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2397 if (retval != 0) { 2398 mtx_lock(&softc->lock); 2399 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2400 links); 2401 softc->num_luns--; 2402 mtx_unlock(&softc->lock); 2403 snprintf(req->error_str, sizeof(req->error_str), 2404 "ctl_add_lun() returned error %d, see dmesg for " 2405 "details", retval); 2406 retval = 0; 2407 goto bailout_error; 2408 } 2409 2410 mtx_lock(&softc->lock); 2411 2412 /* 2413 * Tell the config_status routine that we're waiting so it won't 2414 * clean up the LUN in the event of an error. 2415 */ 2416 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2417 2418 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2419 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2420 if (retval == EINTR) 2421 break; 2422 } 2423 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2424 2425 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2426 snprintf(req->error_str, sizeof(req->error_str), 2427 "LUN configuration error, see dmesg for details"); 2428 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2429 links); 2430 softc->num_luns--; 2431 mtx_unlock(&softc->lock); 2432 goto bailout_error; 2433 } else { 2434 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2435 } 2436 2437 mtx_unlock(&softc->lock); 2438 2439 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2440 be_lun->blocksize, 2441 DEVSTAT_ALL_SUPPORTED, 2442 be_lun->ctl_be_lun.lun_type 2443 | DEVSTAT_TYPE_IF_OTHER, 2444 DEVSTAT_PRIORITY_OTHER); 2445 2446 return (retval); 2447 2448bailout_error: 2449 req->status = CTL_LUN_ERROR; 2450 2451 if (be_lun->io_taskqueue != NULL) 2452 taskqueue_free(be_lun->io_taskqueue); 2453 ctl_be_block_close(be_lun); 2454 if (be_lun->dev_path != NULL) 2455 free(be_lun->dev_path, M_CTLBLK); 2456 if (be_lun->lun_zone != NULL) 2457 uma_zdestroy(be_lun->lun_zone); 2458 ctl_free_opts(&be_lun->ctl_be_lun.options); 2459 mtx_destroy(&be_lun->queue_lock); 2460 mtx_destroy(&be_lun->io_lock); 2461 free(be_lun, M_CTLBLK); 2462 2463 return (retval); 2464} 2465 2466static int 2467ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2468{ 2469 struct ctl_lun_rm_params *params; 2470 struct ctl_be_block_lun *be_lun; 2471 int retval; 2472 2473 params = &req->reqdata.rm; 2474 2475 mtx_lock(&softc->lock); 2476 2477 be_lun = NULL; 2478 2479 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2480 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2481 break; 2482 } 2483 mtx_unlock(&softc->lock); 2484 2485 if (be_lun == NULL) { 2486 snprintf(req->error_str, sizeof(req->error_str), 2487 "LUN %u is not managed by the block backend", 2488 params->lun_id); 2489 goto bailout_error; 2490 } 2491 2492 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2493 2494 if (retval != 0) { 2495 snprintf(req->error_str, sizeof(req->error_str), 2496 "error %d returned from ctl_disable_lun() for " 2497 "LUN %d", retval, params->lun_id); 2498 goto bailout_error; 2499 2500 } 2501 2502 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2503 if (retval != 0) { 2504 snprintf(req->error_str, sizeof(req->error_str), 2505 "error %d returned from ctl_invalidate_lun() for " 2506 "LUN %d", retval, params->lun_id); 2507 goto bailout_error; 2508 } 2509 2510 mtx_lock(&softc->lock); 2511 2512 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2513 2514 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2515 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2516 if (retval == EINTR) 2517 break; 2518 } 2519 2520 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2521 2522 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2523 snprintf(req->error_str, sizeof(req->error_str), 2524 "interrupted waiting for LUN to be freed"); 2525 mtx_unlock(&softc->lock); 2526 goto bailout_error; 2527 } 2528 2529 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2530 2531 softc->num_luns--; 2532 mtx_unlock(&softc->lock); 2533 2534 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2535 2536 taskqueue_free(be_lun->io_taskqueue); 2537 2538 ctl_be_block_close(be_lun); 2539 2540 if (be_lun->disk_stats != NULL) 2541 devstat_remove_entry(be_lun->disk_stats); 2542 2543 uma_zdestroy(be_lun->lun_zone); 2544 2545 ctl_free_opts(&be_lun->ctl_be_lun.options); 2546 free(be_lun->dev_path, M_CTLBLK); 2547 mtx_destroy(&be_lun->queue_lock); 2548 mtx_destroy(&be_lun->io_lock); 2549 free(be_lun, M_CTLBLK); 2550 2551 req->status = CTL_LUN_OK; 2552 2553 return (0); 2554 2555bailout_error: 2556 2557 req->status = CTL_LUN_ERROR; 2558 2559 return (0); 2560} 2561 2562static int 2563ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2564 struct ctl_lun_req *req) 2565{ 2566 struct vattr vattr; 2567 int error; 2568 struct ctl_lun_create_params *params = &be_lun->params; 2569 2570 if (params->lun_size_bytes != 0) { 2571 be_lun->size_bytes = params->lun_size_bytes; 2572 } else { 2573 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2574 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2575 VOP_UNLOCK(be_lun->vn, 0); 2576 if (error != 0) { 2577 snprintf(req->error_str, sizeof(req->error_str), 2578 "error calling VOP_GETATTR() for file %s", 2579 be_lun->dev_path); 2580 return (error); 2581 } 2582 2583 be_lun->size_bytes = vattr.va_size; 2584 } 2585 2586 return (0); 2587} 2588 2589static int 2590ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2591 struct ctl_lun_req *req) 2592{ 2593 struct ctl_be_block_devdata *dev_data; 2594 int error; 2595 struct ctl_lun_create_params *params = &be_lun->params; 2596 uint64_t size_bytes; 2597 2598 dev_data = &be_lun->backend.dev; 2599 if (!dev_data->csw->d_ioctl) { 2600 snprintf(req->error_str, sizeof(req->error_str), 2601 "no d_ioctl for device %s!", be_lun->dev_path); 2602 return (ENODEV); 2603 } 2604 2605 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2606 (caddr_t)&size_bytes, FREAD, 2607 curthread); 2608 if (error) { 2609 snprintf(req->error_str, sizeof(req->error_str), 2610 "error %d returned for DIOCGMEDIASIZE ioctl " 2611 "on %s!", error, be_lun->dev_path); 2612 return (error); 2613 } 2614 2615 if (params->lun_size_bytes != 0) { 2616 if (params->lun_size_bytes > size_bytes) { 2617 snprintf(req->error_str, sizeof(req->error_str), 2618 "requested LUN size %ju > backing device " 2619 "size %ju", 2620 (uintmax_t)params->lun_size_bytes, 2621 (uintmax_t)size_bytes); 2622 return (EINVAL); 2623 } 2624 2625 be_lun->size_bytes = params->lun_size_bytes; 2626 } else { 2627 be_lun->size_bytes = size_bytes; 2628 } 2629 2630 return (0); 2631} 2632 2633static int 2634ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2635{ 2636 struct ctl_lun_modify_params *params; 2637 struct ctl_be_block_lun *be_lun; 2638 uint64_t oldsize; 2639 int error; 2640 2641 params = &req->reqdata.modify; 2642 2643 mtx_lock(&softc->lock); 2644 be_lun = NULL; 2645 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2646 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2647 break; 2648 } 2649 mtx_unlock(&softc->lock); 2650 2651 if (be_lun == NULL) { 2652 snprintf(req->error_str, sizeof(req->error_str), 2653 "LUN %u is not managed by the block backend", 2654 params->lun_id); 2655 goto bailout_error; 2656 } 2657 2658 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2659 2660 oldsize = be_lun->size_bytes; 2661 if (be_lun->vn == NULL) 2662 error = ctl_be_block_open(softc, be_lun, req); 2663 else if (vn_isdisk(be_lun->vn, &error)) 2664 error = ctl_be_block_modify_dev(be_lun, req); 2665 else if (be_lun->vn->v_type == VREG) 2666 error = ctl_be_block_modify_file(be_lun, req); 2667 else 2668 error = EINVAL; 2669 2670 if (error == 0 && be_lun->size_bytes != oldsize) { 2671 be_lun->size_blocks = be_lun->size_bytes >> 2672 be_lun->blocksize_shift; 2673 2674 /* 2675 * The maximum LBA is the size - 1. 2676 * 2677 * XXX: Note that this field is being updated without locking, 2678 * which might cause problems on 32-bit architectures. 2679 */ 2680 if (be_lun->unmap != NULL) 2681 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2682 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2683 0 : (be_lun->size_blocks - 1); 2684 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2685 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2686 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2687 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2688 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2689 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2690 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2691 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2692 if (oldsize == 0 && be_lun->size_blocks != 0) 2693 ctl_lun_online(&be_lun->ctl_be_lun); 2694 } 2695 2696 /* Tell the user the exact size we ended up using */ 2697 params->lun_size_bytes = be_lun->size_bytes; 2698 2699 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2700 2701 return (0); 2702 2703bailout_error: 2704 req->status = CTL_LUN_ERROR; 2705 2706 return (0); 2707} 2708 2709static void 2710ctl_be_block_lun_shutdown(void *be_lun) 2711{ 2712 struct ctl_be_block_lun *lun; 2713 struct ctl_be_block_softc *softc; 2714 2715 lun = (struct ctl_be_block_lun *)be_lun; 2716 2717 softc = lun->softc; 2718 2719 mtx_lock(&softc->lock); 2720 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2721 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2722 wakeup(lun); 2723 mtx_unlock(&softc->lock); 2724 2725} 2726 2727static void 2728ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2729{ 2730 struct ctl_be_block_lun *lun; 2731 struct ctl_be_block_softc *softc; 2732 2733 lun = (struct ctl_be_block_lun *)be_lun; 2734 softc = lun->softc; 2735 2736 if (status == CTL_LUN_CONFIG_OK) { 2737 mtx_lock(&softc->lock); 2738 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2739 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2740 wakeup(lun); 2741 mtx_unlock(&softc->lock); 2742 2743 /* 2744 * We successfully added the LUN, attempt to enable it. 2745 */ 2746 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2747 printf("%s: ctl_enable_lun() failed!\n", __func__); 2748 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2749 printf("%s: ctl_invalidate_lun() failed!\n", 2750 __func__); 2751 } 2752 } 2753 2754 return; 2755 } 2756 2757 2758 mtx_lock(&softc->lock); 2759 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2760 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2761 wakeup(lun); 2762 mtx_unlock(&softc->lock); 2763} 2764 2765 2766static int 2767ctl_be_block_config_write(union ctl_io *io) 2768{ 2769 struct ctl_be_block_lun *be_lun; 2770 struct ctl_be_lun *ctl_be_lun; 2771 int retval; 2772 2773 retval = 0; 2774 2775 DPRINTF("entered\n"); 2776 2777 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2778 CTL_PRIV_BACKEND_LUN].ptr; 2779 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2780 2781 switch (io->scsiio.cdb[0]) { 2782 case SYNCHRONIZE_CACHE: 2783 case SYNCHRONIZE_CACHE_16: 2784 case WRITE_SAME_10: 2785 case WRITE_SAME_16: 2786 case UNMAP: 2787 /* 2788 * The upper level CTL code will filter out any CDBs with 2789 * the immediate bit set and return the proper error. 2790 * 2791 * We don't really need to worry about what LBA range the 2792 * user asked to be synced out. When they issue a sync 2793 * cache command, we'll sync out the whole thing. 2794 */ 2795 mtx_lock(&be_lun->queue_lock); 2796 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2797 links); 2798 mtx_unlock(&be_lun->queue_lock); 2799 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2800 break; 2801 case START_STOP_UNIT: { 2802 struct scsi_start_stop_unit *cdb; 2803 2804 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2805 2806 if (cdb->how & SSS_START) 2807 retval = ctl_start_lun(ctl_be_lun); 2808 else { 2809 retval = ctl_stop_lun(ctl_be_lun); 2810 /* 2811 * XXX KDM Copan-specific offline behavior. 2812 * Figure out a reasonable way to port this? 2813 */ 2814#ifdef NEEDTOPORT 2815 if ((retval == 0) 2816 && (cdb->byte2 & SSS_ONOFFLINE)) 2817 retval = ctl_lun_offline(ctl_be_lun); 2818#endif 2819 } 2820 2821 /* 2822 * In general, the above routines should not fail. They 2823 * just set state for the LUN. So we've got something 2824 * pretty wrong here if we can't start or stop the LUN. 2825 */ 2826 if (retval != 0) { 2827 ctl_set_internal_failure(&io->scsiio, 2828 /*sks_valid*/ 1, 2829 /*retry_count*/ 0xf051); 2830 retval = CTL_RETVAL_COMPLETE; 2831 } else { 2832 ctl_set_success(&io->scsiio); 2833 } 2834 ctl_config_write_done(io); 2835 break; 2836 } 2837 default: 2838 ctl_set_invalid_opcode(&io->scsiio); 2839 ctl_config_write_done(io); 2840 retval = CTL_RETVAL_COMPLETE; 2841 break; 2842 } 2843 2844 return (retval); 2845} 2846 2847static int 2848ctl_be_block_config_read(union ctl_io *io) 2849{ 2850 struct ctl_be_block_lun *be_lun; 2851 struct ctl_be_lun *ctl_be_lun; 2852 int retval = 0; 2853 2854 DPRINTF("entered\n"); 2855 2856 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2857 CTL_PRIV_BACKEND_LUN].ptr; 2858 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2859 2860 switch (io->scsiio.cdb[0]) { 2861 case SERVICE_ACTION_IN: 2862 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2863 mtx_lock(&be_lun->queue_lock); 2864 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2865 &io->io_hdr, links); 2866 mtx_unlock(&be_lun->queue_lock); 2867 taskqueue_enqueue(be_lun->io_taskqueue, 2868 &be_lun->io_task); 2869 retval = CTL_RETVAL_QUEUED; 2870 break; 2871 } 2872 ctl_set_invalid_field(&io->scsiio, 2873 /*sks_valid*/ 1, 2874 /*command*/ 1, 2875 /*field*/ 1, 2876 /*bit_valid*/ 1, 2877 /*bit*/ 4); 2878 ctl_config_read_done(io); 2879 retval = CTL_RETVAL_COMPLETE; 2880 break; 2881 default: 2882 ctl_set_invalid_opcode(&io->scsiio); 2883 ctl_config_read_done(io); 2884 retval = CTL_RETVAL_COMPLETE; 2885 break; 2886 } 2887 2888 return (retval); 2889} 2890 2891static int 2892ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2893{ 2894 struct ctl_be_block_lun *lun; 2895 int retval; 2896 2897 lun = (struct ctl_be_block_lun *)be_lun; 2898 retval = 0; 2899 2900 retval = sbuf_printf(sb, "\t<num_threads>"); 2901 2902 if (retval != 0) 2903 goto bailout; 2904 2905 retval = sbuf_printf(sb, "%d", lun->num_threads); 2906 2907 if (retval != 0) 2908 goto bailout; 2909 2910 retval = sbuf_printf(sb, "</num_threads>\n"); 2911 2912bailout: 2913 2914 return (retval); 2915} 2916 2917static uint64_t 2918ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2919{ 2920 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2921 2922 if (lun->getattr == NULL) 2923 return (UINT64_MAX); 2924 return (lun->getattr(lun, attrname)); 2925} 2926 2927int 2928ctl_be_block_init(void) 2929{ 2930 struct ctl_be_block_softc *softc; 2931 int retval; 2932 2933 softc = &backend_block_softc; 2934 retval = 0; 2935 2936 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2937 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2938 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2939 STAILQ_INIT(&softc->disk_list); 2940 STAILQ_INIT(&softc->lun_list); 2941 2942 return (retval); 2943} 2944