ctl_backend_block.c revision 264191
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: head/sys/cam/ctl/ctl_backend_block.c 264191 2014-04-06 10:13:14Z mav $"); 44 45#include <sys/param.h> 46#include <sys/systm.h> 47#include <sys/kernel.h> 48#include <sys/types.h> 49#include <sys/kthread.h> 50#include <sys/bio.h> 51#include <sys/fcntl.h> 52#include <sys/lock.h> 53#include <sys/mutex.h> 54#include <sys/condvar.h> 55#include <sys/malloc.h> 56#include <sys/conf.h> 57#include <sys/ioccom.h> 58#include <sys/queue.h> 59#include <sys/sbuf.h> 60#include <sys/endian.h> 61#include <sys/uio.h> 62#include <sys/buf.h> 63#include <sys/taskqueue.h> 64#include <sys/vnode.h> 65#include <sys/namei.h> 66#include <sys/mount.h> 67#include <sys/disk.h> 68#include <sys/fcntl.h> 69#include <sys/filedesc.h> 70#include <sys/proc.h> 71#include <sys/pcpu.h> 72#include <sys/module.h> 73#include <sys/sdt.h> 74#include <sys/devicestat.h> 75#include <sys/sysctl.h> 76 77#include <geom/geom.h> 78 79#include <cam/cam.h> 80#include <cam/scsi/scsi_all.h> 81#include <cam/scsi/scsi_da.h> 82#include <cam/ctl/ctl_io.h> 83#include <cam/ctl/ctl.h> 84#include <cam/ctl/ctl_backend.h> 85#include <cam/ctl/ctl_frontend_internal.h> 86#include <cam/ctl/ctl_ioctl.h> 87#include <cam/ctl/ctl_scsi_all.h> 88#include <cam/ctl/ctl_error.h> 89 90/* 91 * The idea here is that we'll allocate enough S/G space to hold a 16MB 92 * I/O. If we get an I/O larger than that, we'll reject it. 93 */ 94#define CTLBLK_MAX_IO_SIZE (16 * 1024 * 1024) 95#define CTLBLK_MAX_SEGS (CTLBLK_MAX_IO_SIZE / MAXPHYS) + 1 96 97#ifdef CTLBLK_DEBUG 98#define DPRINTF(fmt, args...) \ 99 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 100#else 101#define DPRINTF(fmt, args...) do {} while(0) 102#endif 103 104SDT_PROVIDER_DEFINE(cbb); 105 106typedef enum { 107 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 108 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 109 CTL_BE_BLOCK_LUN_WAITING = 0x04, 110 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 111} ctl_be_block_lun_flags; 112 113typedef enum { 114 CTL_BE_BLOCK_NONE, 115 CTL_BE_BLOCK_DEV, 116 CTL_BE_BLOCK_FILE 117} ctl_be_block_type; 118 119struct ctl_be_block_devdata { 120 struct cdev *cdev; 121 struct cdevsw *csw; 122 int dev_ref; 123}; 124 125struct ctl_be_block_filedata { 126 struct ucred *cred; 127}; 128 129union ctl_be_block_bedata { 130 struct ctl_be_block_devdata dev; 131 struct ctl_be_block_filedata file; 132}; 133 134struct ctl_be_block_io; 135struct ctl_be_block_lun; 136 137typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 138 struct ctl_be_block_io *beio); 139 140/* 141 * Backend LUN structure. There is a 1:1 mapping between a block device 142 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 143 */ 144struct ctl_be_block_lun { 145 struct ctl_block_disk *disk; 146 char lunname[32]; 147 char *dev_path; 148 ctl_be_block_type dev_type; 149 struct vnode *vn; 150 union ctl_be_block_bedata backend; 151 cbb_dispatch_t dispatch; 152 cbb_dispatch_t lun_flush; 153 struct mtx lock; 154 uma_zone_t lun_zone; 155 uint64_t size_blocks; 156 uint64_t size_bytes; 157 uint32_t blocksize; 158 int blocksize_shift; 159 uint16_t pblockexp; 160 uint16_t pblockoff; 161 struct ctl_be_block_softc *softc; 162 struct devstat *disk_stats; 163 ctl_be_block_lun_flags flags; 164 STAILQ_ENTRY(ctl_be_block_lun) links; 165 struct ctl_be_lun ctl_be_lun; 166 struct taskqueue *io_taskqueue; 167 struct task io_task; 168 int num_threads; 169 STAILQ_HEAD(, ctl_io_hdr) input_queue; 170 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 171 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 172}; 173 174/* 175 * Overall softc structure for the block backend module. 176 */ 177struct ctl_be_block_softc { 178 struct mtx lock; 179 int num_disks; 180 STAILQ_HEAD(, ctl_block_disk) disk_list; 181 int num_luns; 182 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 183}; 184 185static struct ctl_be_block_softc backend_block_softc; 186 187/* 188 * Per-I/O information. 189 */ 190struct ctl_be_block_io { 191 union ctl_io *io; 192 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 193 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 194 int bio_cmd; 195 int bio_flags; 196 int num_segs; 197 int num_bios_sent; 198 int num_bios_done; 199 int send_complete; 200 int num_errors; 201 struct bintime ds_t0; 202 devstat_tag_type ds_tag_type; 203 devstat_trans_flags ds_trans_type; 204 uint64_t io_len; 205 uint64_t io_offset; 206 struct ctl_be_block_softc *softc; 207 struct ctl_be_block_lun *lun; 208}; 209 210static int cbb_num_threads = 14; 211TUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 212SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 213 "CAM Target Layer Block Backend"); 214SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 215 &cbb_num_threads, 0, "Number of threads per backing file"); 216 217static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 218static void ctl_free_beio(struct ctl_be_block_io *beio); 219static void ctl_complete_beio(struct ctl_be_block_io *beio); 220static int ctl_be_block_move_done(union ctl_io *io); 221static void ctl_be_block_biodone(struct bio *bio); 222static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 223 struct ctl_be_block_io *beio); 224static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 225 struct ctl_be_block_io *beio); 226static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 227 struct ctl_be_block_io *beio); 228static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 229 struct ctl_be_block_io *beio); 230static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 231 union ctl_io *io); 232static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 233 union ctl_io *io); 234static void ctl_be_block_worker(void *context, int pending); 235static int ctl_be_block_submit(union ctl_io *io); 236static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 237 int flag, struct thread *td); 238static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 239 struct ctl_lun_req *req); 240static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 241 struct ctl_lun_req *req); 242static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 243static int ctl_be_block_open(struct ctl_be_block_softc *softc, 244 struct ctl_be_block_lun *be_lun, 245 struct ctl_lun_req *req); 246static int ctl_be_block_create(struct ctl_be_block_softc *softc, 247 struct ctl_lun_req *req); 248static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 249 struct ctl_lun_req *req); 250static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 251 struct ctl_lun_req *req); 252static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 253 struct ctl_lun_req *req); 254static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 255 struct ctl_lun_req *req); 256static void ctl_be_block_lun_shutdown(void *be_lun); 257static void ctl_be_block_lun_config_status(void *be_lun, 258 ctl_lun_config_status status); 259static int ctl_be_block_config_write(union ctl_io *io); 260static int ctl_be_block_config_read(union ctl_io *io); 261static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 262int ctl_be_block_init(void); 263 264static struct ctl_backend_driver ctl_be_block_driver = 265{ 266 .name = "block", 267 .flags = CTL_BE_FLAG_HAS_CONFIG, 268 .init = ctl_be_block_init, 269 .data_submit = ctl_be_block_submit, 270 .data_move_done = ctl_be_block_move_done, 271 .config_read = ctl_be_block_config_read, 272 .config_write = ctl_be_block_config_write, 273 .ioctl = ctl_be_block_ioctl, 274 .lun_info = ctl_be_block_lun_info 275}; 276 277MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 278CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 279 280static uma_zone_t beio_zone; 281 282static struct ctl_be_block_io * 283ctl_alloc_beio(struct ctl_be_block_softc *softc) 284{ 285 struct ctl_be_block_io *beio; 286 287 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 288 beio->softc = softc; 289 return (beio); 290} 291 292static void 293ctl_free_beio(struct ctl_be_block_io *beio) 294{ 295 int duplicate_free; 296 int i; 297 298 duplicate_free = 0; 299 300 for (i = 0; i < beio->num_segs; i++) { 301 if (beio->sg_segs[i].addr == NULL) 302 duplicate_free++; 303 304 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 305 beio->sg_segs[i].addr = NULL; 306 } 307 308 if (duplicate_free > 0) { 309 printf("%s: %d duplicate frees out of %d segments\n", __func__, 310 duplicate_free, beio->num_segs); 311 } 312 313 uma_zfree(beio_zone, beio); 314} 315 316static void 317ctl_complete_beio(struct ctl_be_block_io *beio) 318{ 319 union ctl_io *io; 320 int io_len; 321 322 io = beio->io; 323 324 if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) 325 io_len = beio->io_len; 326 else 327 io_len = 0; 328 329 devstat_end_transaction(beio->lun->disk_stats, 330 /*bytes*/ io_len, 331 beio->ds_tag_type, 332 beio->ds_trans_type, 333 /*now*/ NULL, 334 /*then*/&beio->ds_t0); 335 336 ctl_free_beio(beio); 337 ctl_done(io); 338} 339 340static int 341ctl_be_block_move_done(union ctl_io *io) 342{ 343 struct ctl_be_block_io *beio; 344 struct ctl_be_block_lun *be_lun; 345#ifdef CTL_TIME_IO 346 struct bintime cur_bt; 347#endif 348 349 beio = (struct ctl_be_block_io *) 350 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 351 352 be_lun = beio->lun; 353 354 DPRINTF("entered\n"); 355 356#ifdef CTL_TIME_IO 357 getbintime(&cur_bt); 358 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 359 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 360 io->io_hdr.num_dmas++; 361#endif 362 363 /* 364 * We set status at this point for read commands, and write 365 * commands with errors. 366 */ 367 if ((beio->bio_cmd == BIO_READ) 368 && (io->io_hdr.port_status == 0) 369 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 370 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) 371 ctl_set_success(&io->scsiio); 372 else if ((io->io_hdr.port_status != 0) 373 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 374 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 375 /* 376 * For hardware error sense keys, the sense key 377 * specific value is defined to be a retry count, 378 * but we use it to pass back an internal FETD 379 * error code. XXX KDM Hopefully the FETD is only 380 * using 16 bits for an error code, since that's 381 * all the space we have in the sks field. 382 */ 383 ctl_set_internal_failure(&io->scsiio, 384 /*sks_valid*/ 1, 385 /*retry_count*/ 386 io->io_hdr.port_status); 387 } 388 389 /* 390 * If this is a read, or a write with errors, it is done. 391 */ 392 if ((beio->bio_cmd == BIO_READ) 393 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 394 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 395 ctl_complete_beio(beio); 396 return (0); 397 } 398 399 /* 400 * At this point, we have a write and the DMA completed 401 * successfully. We now have to queue it to the task queue to 402 * execute the backend I/O. That is because we do blocking 403 * memory allocations, and in the file backing case, blocking I/O. 404 * This move done routine is generally called in the SIM's 405 * interrupt context, and therefore we cannot block. 406 */ 407 mtx_lock(&be_lun->lock); 408 /* 409 * XXX KDM make sure that links is okay to use at this point. 410 * Otherwise, we either need to add another field to ctl_io_hdr, 411 * or deal with resource allocation here. 412 */ 413 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 414 mtx_unlock(&be_lun->lock); 415 416 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 417 418 return (0); 419} 420 421static void 422ctl_be_block_biodone(struct bio *bio) 423{ 424 struct ctl_be_block_io *beio; 425 struct ctl_be_block_lun *be_lun; 426 union ctl_io *io; 427 int error; 428 429 beio = bio->bio_caller1; 430 be_lun = beio->lun; 431 io = beio->io; 432 433 DPRINTF("entered\n"); 434 435 error = bio->bio_error; 436 mtx_lock(&be_lun->lock); 437 if (error != 0) 438 beio->num_errors++; 439 440 beio->num_bios_done++; 441 442 /* 443 * XXX KDM will this cause WITNESS to complain? Holding a lock 444 * during the free might cause it to complain. 445 */ 446 g_destroy_bio(bio); 447 448 /* 449 * If the send complete bit isn't set, or we aren't the last I/O to 450 * complete, then we're done. 451 */ 452 if ((beio->send_complete == 0) 453 || (beio->num_bios_done < beio->num_bios_sent)) { 454 mtx_unlock(&be_lun->lock); 455 return; 456 } 457 458 /* 459 * At this point, we've verified that we are the last I/O to 460 * complete, so it's safe to drop the lock. 461 */ 462 mtx_unlock(&be_lun->lock); 463 464 /* 465 * If there are any errors from the backing device, we fail the 466 * entire I/O with a medium error. 467 */ 468 if (beio->num_errors > 0) { 469 if (error == EOPNOTSUPP) { 470 ctl_set_invalid_opcode(&io->scsiio); 471 } else if (beio->bio_cmd == BIO_FLUSH) { 472 /* XXX KDM is there is a better error here? */ 473 ctl_set_internal_failure(&io->scsiio, 474 /*sks_valid*/ 1, 475 /*retry_count*/ 0xbad2); 476 } else 477 ctl_set_medium_error(&io->scsiio); 478 ctl_complete_beio(beio); 479 return; 480 } 481 482 /* 483 * If this is a write or a flush, we're all done. 484 * If this is a read, we can now send the data to the user. 485 */ 486 if ((beio->bio_cmd == BIO_WRITE) 487 || (beio->bio_cmd == BIO_FLUSH)) { 488 ctl_set_success(&io->scsiio); 489 ctl_complete_beio(beio); 490 } else { 491 io->scsiio.be_move_done = ctl_be_block_move_done; 492 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 493 io->scsiio.kern_data_len = beio->io_len; 494 io->scsiio.kern_total_len = beio->io_len; 495 io->scsiio.kern_rel_offset = 0; 496 io->scsiio.kern_data_resid = 0; 497 io->scsiio.kern_sg_entries = beio->num_segs; 498 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 499#ifdef CTL_TIME_IO 500 getbintime(&io->io_hdr.dma_start_bt); 501#endif 502 ctl_datamove(io); 503 } 504} 505 506static void 507ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 508 struct ctl_be_block_io *beio) 509{ 510 union ctl_io *io; 511 struct mount *mountpoint; 512 int error, lock_flags; 513 514 DPRINTF("entered\n"); 515 516 io = beio->io; 517 518 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 519 520 if (MNT_SHARED_WRITES(mountpoint) 521 || ((mountpoint == NULL) 522 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 523 lock_flags = LK_SHARED; 524 else 525 lock_flags = LK_EXCLUSIVE; 526 527 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 528 529 binuptime(&beio->ds_t0); 530 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 531 532 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 533 VOP_UNLOCK(be_lun->vn, 0); 534 535 vn_finished_write(mountpoint); 536 537 if (error == 0) 538 ctl_set_success(&io->scsiio); 539 else { 540 /* XXX KDM is there is a better error here? */ 541 ctl_set_internal_failure(&io->scsiio, 542 /*sks_valid*/ 1, 543 /*retry_count*/ 0xbad1); 544 } 545 546 ctl_complete_beio(beio); 547} 548 549SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 550SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 551SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 552SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 553 554static void 555ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 556 struct ctl_be_block_io *beio) 557{ 558 struct ctl_be_block_filedata *file_data; 559 union ctl_io *io; 560 struct uio xuio; 561 struct iovec *xiovec; 562 int flags; 563 int error, i; 564 565 DPRINTF("entered\n"); 566 567 file_data = &be_lun->backend.file; 568 io = beio->io; 569 flags = beio->bio_flags; 570 571 if (beio->bio_cmd == BIO_READ) { 572 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 573 } else { 574 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 575 } 576 577 bzero(&xuio, sizeof(xuio)); 578 if (beio->bio_cmd == BIO_READ) 579 xuio.uio_rw = UIO_READ; 580 else 581 xuio.uio_rw = UIO_WRITE; 582 583 xuio.uio_offset = beio->io_offset; 584 xuio.uio_resid = beio->io_len; 585 xuio.uio_segflg = UIO_SYSSPACE; 586 xuio.uio_iov = beio->xiovecs; 587 xuio.uio_iovcnt = beio->num_segs; 588 xuio.uio_td = curthread; 589 590 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 591 xiovec->iov_base = beio->sg_segs[i].addr; 592 xiovec->iov_len = beio->sg_segs[i].len; 593 } 594 595 if (beio->bio_cmd == BIO_READ) { 596 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 597 598 binuptime(&beio->ds_t0); 599 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 600 601 /* 602 * UFS pays attention to IO_DIRECT for reads. If the 603 * DIRECTIO option is configured into the kernel, it calls 604 * ffs_rawread(). But that only works for single-segment 605 * uios with user space addresses. In our case, with a 606 * kernel uio, it still reads into the buffer cache, but it 607 * will just try to release the buffer from the cache later 608 * on in ffs_read(). 609 * 610 * ZFS does not pay attention to IO_DIRECT for reads. 611 * 612 * UFS does not pay attention to IO_SYNC for reads. 613 * 614 * ZFS pays attention to IO_SYNC (which translates into the 615 * Solaris define FRSYNC for zfs_read()) for reads. It 616 * attempts to sync the file before reading. 617 * 618 * So, to attempt to provide some barrier semantics in the 619 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 620 */ 621 error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 622 (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 623 624 VOP_UNLOCK(be_lun->vn, 0); 625 } else { 626 struct mount *mountpoint; 627 int lock_flags; 628 629 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 630 631 if (MNT_SHARED_WRITES(mountpoint) 632 || ((mountpoint == NULL) 633 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 634 lock_flags = LK_SHARED; 635 else 636 lock_flags = LK_EXCLUSIVE; 637 638 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 639 640 binuptime(&beio->ds_t0); 641 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 642 643 /* 644 * UFS pays attention to IO_DIRECT for writes. The write 645 * is done asynchronously. (Normally the write would just 646 * get put into cache. 647 * 648 * UFS pays attention to IO_SYNC for writes. It will 649 * attempt to write the buffer out synchronously if that 650 * flag is set. 651 * 652 * ZFS does not pay attention to IO_DIRECT for writes. 653 * 654 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 655 * for writes. It will flush the transaction from the 656 * cache before returning. 657 * 658 * So if we've got the BIO_ORDERED flag set, we want 659 * IO_SYNC in either the UFS or ZFS case. 660 */ 661 error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 662 IO_SYNC : 0, file_data->cred); 663 VOP_UNLOCK(be_lun->vn, 0); 664 665 vn_finished_write(mountpoint); 666 } 667 668 /* 669 * If we got an error, set the sense data to "MEDIUM ERROR" and 670 * return the I/O to the user. 671 */ 672 if (error != 0) { 673 char path_str[32]; 674 675 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 676 /* 677 * XXX KDM ZFS returns ENOSPC when the underlying 678 * filesystem fills up. What kind of SCSI error should we 679 * return for that? 680 */ 681 printf("%s%s command returned errno %d\n", path_str, 682 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 683 ctl_set_medium_error(&io->scsiio); 684 ctl_complete_beio(beio); 685 return; 686 } 687 688 /* 689 * If this is a write, we're all done. 690 * If this is a read, we can now send the data to the user. 691 */ 692 if (beio->bio_cmd == BIO_WRITE) { 693 ctl_set_success(&io->scsiio); 694 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 695 ctl_complete_beio(beio); 696 } else { 697 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 698 io->scsiio.be_move_done = ctl_be_block_move_done; 699 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 700 io->scsiio.kern_data_len = beio->io_len; 701 io->scsiio.kern_total_len = beio->io_len; 702 io->scsiio.kern_rel_offset = 0; 703 io->scsiio.kern_data_resid = 0; 704 io->scsiio.kern_sg_entries = beio->num_segs; 705 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 706#ifdef CTL_TIME_IO 707 getbintime(&io->io_hdr.dma_start_bt); 708#endif 709 ctl_datamove(io); 710 } 711} 712 713static void 714ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 715 struct ctl_be_block_io *beio) 716{ 717 struct bio *bio; 718 union ctl_io *io; 719 struct ctl_be_block_devdata *dev_data; 720 721 dev_data = &be_lun->backend.dev; 722 io = beio->io; 723 724 DPRINTF("entered\n"); 725 726 /* This can't fail, it's a blocking allocation. */ 727 bio = g_alloc_bio(); 728 729 bio->bio_cmd = BIO_FLUSH; 730 bio->bio_flags |= BIO_ORDERED; 731 bio->bio_dev = dev_data->cdev; 732 bio->bio_offset = 0; 733 bio->bio_data = 0; 734 bio->bio_done = ctl_be_block_biodone; 735 bio->bio_caller1 = beio; 736 bio->bio_pblkno = 0; 737 738 /* 739 * We don't need to acquire the LUN lock here, because we are only 740 * sending one bio, and so there is no other context to synchronize 741 * with. 742 */ 743 beio->num_bios_sent = 1; 744 beio->send_complete = 1; 745 746 binuptime(&beio->ds_t0); 747 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 748 749 (*dev_data->csw->d_strategy)(bio); 750} 751 752static void 753ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 754 struct ctl_be_block_io *beio) 755{ 756 int i; 757 struct bio *bio; 758 struct ctl_be_block_devdata *dev_data; 759 off_t cur_offset; 760 int max_iosize; 761 762 DPRINTF("entered\n"); 763 764 dev_data = &be_lun->backend.dev; 765 766 /* 767 * We have to limit our I/O size to the maximum supported by the 768 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 769 * set it properly, use DFLTPHYS. 770 */ 771 max_iosize = dev_data->cdev->si_iosize_max; 772 if (max_iosize < PAGE_SIZE) 773 max_iosize = DFLTPHYS; 774 775 cur_offset = beio->io_offset; 776 777 /* 778 * XXX KDM need to accurately reflect the number of I/Os outstanding 779 * to a device. 780 */ 781 binuptime(&beio->ds_t0); 782 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 783 784 for (i = 0; i < beio->num_segs; i++) { 785 size_t cur_size; 786 uint8_t *cur_ptr; 787 788 cur_size = beio->sg_segs[i].len; 789 cur_ptr = beio->sg_segs[i].addr; 790 791 while (cur_size > 0) { 792 /* This can't fail, it's a blocking allocation. */ 793 bio = g_alloc_bio(); 794 795 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 796 797 bio->bio_cmd = beio->bio_cmd; 798 bio->bio_flags |= beio->bio_flags; 799 bio->bio_dev = dev_data->cdev; 800 bio->bio_caller1 = beio; 801 bio->bio_length = min(cur_size, max_iosize); 802 bio->bio_offset = cur_offset; 803 bio->bio_data = cur_ptr; 804 bio->bio_done = ctl_be_block_biodone; 805 bio->bio_pblkno = cur_offset / be_lun->blocksize; 806 807 cur_offset += bio->bio_length; 808 cur_ptr += bio->bio_length; 809 cur_size -= bio->bio_length; 810 811 /* 812 * Make sure we set the complete bit just before we 813 * issue the last bio so we don't wind up with a 814 * race. 815 * 816 * Use the LUN mutex here instead of a combination 817 * of atomic variables for simplicity. 818 * 819 * XXX KDM we could have a per-IO lock, but that 820 * would cause additional per-IO setup and teardown 821 * overhead. Hopefully there won't be too much 822 * contention on the LUN lock. 823 */ 824 mtx_lock(&be_lun->lock); 825 826 beio->num_bios_sent++; 827 828 if ((i == beio->num_segs - 1) 829 && (cur_size == 0)) 830 beio->send_complete = 1; 831 832 mtx_unlock(&be_lun->lock); 833 834 (*dev_data->csw->d_strategy)(bio); 835 } 836 } 837} 838 839static void 840ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 841 union ctl_io *io) 842{ 843 struct ctl_be_block_io *beio; 844 struct ctl_be_block_softc *softc; 845 846 DPRINTF("entered\n"); 847 848 softc = be_lun->softc; 849 beio = ctl_alloc_beio(softc); 850 KASSERT(beio != NULL, ("ctl_alloc_beio() failed")); 851 852 beio->io = io; 853 beio->softc = softc; 854 beio->lun = be_lun; 855 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 856 857 switch (io->scsiio.cdb[0]) { 858 case SYNCHRONIZE_CACHE: 859 case SYNCHRONIZE_CACHE_16: 860 beio->bio_cmd = BIO_FLUSH; 861 beio->ds_trans_type = DEVSTAT_NO_DATA; 862 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 863 beio->io_len = 0; 864 be_lun->lun_flush(be_lun, beio); 865 break; 866 default: 867 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 868 break; 869 } 870} 871 872SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 873SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 874SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 875SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 876 877static void 878ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 879 union ctl_io *io) 880{ 881 struct ctl_be_block_io *beio; 882 struct ctl_be_block_softc *softc; 883 struct ctl_lba_len lbalen; 884 uint64_t len_left, io_size_bytes; 885 int i; 886 887 softc = be_lun->softc; 888 889 DPRINTF("entered\n"); 890 891 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 892 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 893 } else { 894 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 895 } 896 897 memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 898 sizeof(lbalen)); 899 900 io_size_bytes = lbalen.len * be_lun->blocksize; 901 902 /* 903 * XXX KDM this is temporary, until we implement chaining of beio 904 * structures and multiple datamove calls to move all the data in 905 * or out. 906 */ 907 if (io_size_bytes > CTLBLK_MAX_IO_SIZE) { 908 printf("%s: IO length %ju > max io size %u\n", __func__, 909 io_size_bytes, CTLBLK_MAX_IO_SIZE); 910 ctl_set_invalid_field(&io->scsiio, 911 /*sks_valid*/ 0, 912 /*command*/ 1, 913 /*field*/ 0, 914 /*bit_valid*/ 0, 915 /*bit*/ 0); 916 ctl_done(io); 917 return; 918 } 919 920 beio = ctl_alloc_beio(softc); 921 KASSERT(beio != NULL, ("ctl_alloc_beio() failed")); 922 923 beio->io = io; 924 beio->softc = softc; 925 beio->lun = be_lun; 926 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 927 928 /* 929 * If the I/O came down with an ordered or head of queue tag, set 930 * the BIO_ORDERED attribute. For head of queue tags, that's 931 * pretty much the best we can do. 932 * 933 * XXX KDM we don't have a great way to easily know about the FUA 934 * bit right now (it is decoded in ctl_read_write(), but we don't 935 * pass that knowledge to the backend), and in any case we would 936 * need to determine how to handle it. 937 */ 938 if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 939 || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 940 beio->bio_flags = BIO_ORDERED; 941 942 switch (io->scsiio.tag_type) { 943 case CTL_TAG_ORDERED: 944 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 945 break; 946 case CTL_TAG_HEAD_OF_QUEUE: 947 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 948 break; 949 case CTL_TAG_UNTAGGED: 950 case CTL_TAG_SIMPLE: 951 case CTL_TAG_ACA: 952 default: 953 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 954 break; 955 } 956 957 /* 958 * This path handles read and write only. The config write path 959 * handles flush operations. 960 */ 961 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 962 beio->bio_cmd = BIO_READ; 963 beio->ds_trans_type = DEVSTAT_READ; 964 } else { 965 beio->bio_cmd = BIO_WRITE; 966 beio->ds_trans_type = DEVSTAT_WRITE; 967 } 968 969 beio->io_len = lbalen.len * be_lun->blocksize; 970 beio->io_offset = lbalen.lba * be_lun->blocksize; 971 972 DPRINTF("%s at LBA %jx len %u\n", 973 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 974 (uintmax_t)lbalen.lba, lbalen.len); 975 976 for (i = 0, len_left = io_size_bytes; i < CTLBLK_MAX_SEGS && 977 len_left > 0; i++) { 978 979 /* 980 * Setup the S/G entry for this chunk. 981 */ 982 beio->sg_segs[i].len = min(MAXPHYS, len_left); 983 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 984 985 DPRINTF("segment %d addr %p len %zd\n", i, 986 beio->sg_segs[i].addr, beio->sg_segs[i].len); 987 988 beio->num_segs++; 989 len_left -= beio->sg_segs[i].len; 990 } 991 992 /* 993 * For the read case, we need to read the data into our buffers and 994 * then we can send it back to the user. For the write case, we 995 * need to get the data from the user first. 996 */ 997 if (beio->bio_cmd == BIO_READ) { 998 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 999 be_lun->dispatch(be_lun, beio); 1000 } else { 1001 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1002 io->scsiio.be_move_done = ctl_be_block_move_done; 1003 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1004 io->scsiio.kern_data_len = beio->io_len; 1005 io->scsiio.kern_total_len = beio->io_len; 1006 io->scsiio.kern_rel_offset = 0; 1007 io->scsiio.kern_data_resid = 0; 1008 io->scsiio.kern_sg_entries = beio->num_segs; 1009 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1010#ifdef CTL_TIME_IO 1011 getbintime(&io->io_hdr.dma_start_bt); 1012#endif 1013 ctl_datamove(io); 1014 } 1015} 1016 1017static void 1018ctl_be_block_worker(void *context, int pending) 1019{ 1020 struct ctl_be_block_lun *be_lun; 1021 struct ctl_be_block_softc *softc; 1022 union ctl_io *io; 1023 1024 be_lun = (struct ctl_be_block_lun *)context; 1025 softc = be_lun->softc; 1026 1027 DPRINTF("entered\n"); 1028 1029 mtx_lock(&be_lun->lock); 1030 for (;;) { 1031 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1032 if (io != NULL) { 1033 struct ctl_be_block_io *beio; 1034 1035 DPRINTF("datamove queue\n"); 1036 1037 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1038 ctl_io_hdr, links); 1039 1040 mtx_unlock(&be_lun->lock); 1041 1042 beio = (struct ctl_be_block_io *) 1043 io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1044 1045 be_lun->dispatch(be_lun, beio); 1046 1047 mtx_lock(&be_lun->lock); 1048 continue; 1049 } 1050 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1051 if (io != NULL) { 1052 1053 DPRINTF("config write queue\n"); 1054 1055 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1056 ctl_io_hdr, links); 1057 1058 mtx_unlock(&be_lun->lock); 1059 1060 ctl_be_block_cw_dispatch(be_lun, io); 1061 1062 mtx_lock(&be_lun->lock); 1063 continue; 1064 } 1065 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1066 if (io != NULL) { 1067 DPRINTF("input queue\n"); 1068 1069 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1070 ctl_io_hdr, links); 1071 mtx_unlock(&be_lun->lock); 1072 1073 /* 1074 * We must drop the lock, since this routine and 1075 * its children may sleep. 1076 */ 1077 ctl_be_block_dispatch(be_lun, io); 1078 1079 mtx_lock(&be_lun->lock); 1080 continue; 1081 } 1082 1083 /* 1084 * If we get here, there is no work left in the queues, so 1085 * just break out and let the task queue go to sleep. 1086 */ 1087 break; 1088 } 1089 mtx_unlock(&be_lun->lock); 1090} 1091 1092/* 1093 * Entry point from CTL to the backend for I/O. We queue everything to a 1094 * work thread, so this just puts the I/O on a queue and wakes up the 1095 * thread. 1096 */ 1097static int 1098ctl_be_block_submit(union ctl_io *io) 1099{ 1100 struct ctl_be_block_lun *be_lun; 1101 struct ctl_be_lun *ctl_be_lun; 1102 int retval; 1103 1104 DPRINTF("entered\n"); 1105 1106 retval = CTL_RETVAL_COMPLETE; 1107 1108 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1109 CTL_PRIV_BACKEND_LUN].ptr; 1110 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1111 1112 /* 1113 * Make sure we only get SCSI I/O. 1114 */ 1115 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1116 "%#x) encountered", io->io_hdr.io_type)); 1117 1118 mtx_lock(&be_lun->lock); 1119 /* 1120 * XXX KDM make sure that links is okay to use at this point. 1121 * Otherwise, we either need to add another field to ctl_io_hdr, 1122 * or deal with resource allocation here. 1123 */ 1124 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1125 mtx_unlock(&be_lun->lock); 1126 1127 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1128 1129 return (retval); 1130} 1131 1132static int 1133ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1134 int flag, struct thread *td) 1135{ 1136 struct ctl_be_block_softc *softc; 1137 int error; 1138 1139 softc = &backend_block_softc; 1140 1141 error = 0; 1142 1143 switch (cmd) { 1144 case CTL_LUN_REQ: { 1145 struct ctl_lun_req *lun_req; 1146 1147 lun_req = (struct ctl_lun_req *)addr; 1148 1149 switch (lun_req->reqtype) { 1150 case CTL_LUNREQ_CREATE: 1151 error = ctl_be_block_create(softc, lun_req); 1152 break; 1153 case CTL_LUNREQ_RM: 1154 error = ctl_be_block_rm(softc, lun_req); 1155 break; 1156 case CTL_LUNREQ_MODIFY: 1157 error = ctl_be_block_modify(softc, lun_req); 1158 break; 1159 default: 1160 lun_req->status = CTL_LUN_ERROR; 1161 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1162 "%s: invalid LUN request type %d", __func__, 1163 lun_req->reqtype); 1164 break; 1165 } 1166 break; 1167 } 1168 default: 1169 error = ENOTTY; 1170 break; 1171 } 1172 1173 return (error); 1174} 1175 1176static int 1177ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1178{ 1179 struct ctl_be_block_filedata *file_data; 1180 struct ctl_lun_create_params *params; 1181 struct vattr vattr; 1182 int error; 1183 1184 error = 0; 1185 file_data = &be_lun->backend.file; 1186 params = &req->reqdata.create; 1187 1188 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1189 be_lun->dispatch = ctl_be_block_dispatch_file; 1190 be_lun->lun_flush = ctl_be_block_flush_file; 1191 1192 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1193 if (error != 0) { 1194 snprintf(req->error_str, sizeof(req->error_str), 1195 "error calling VOP_GETATTR() for file %s", 1196 be_lun->dev_path); 1197 return (error); 1198 } 1199 1200 /* 1201 * Verify that we have the ability to upgrade to exclusive 1202 * access on this file so we can trap errors at open instead 1203 * of reporting them during first access. 1204 */ 1205 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1206 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1207 if (be_lun->vn->v_iflag & VI_DOOMED) { 1208 error = EBADF; 1209 snprintf(req->error_str, sizeof(req->error_str), 1210 "error locking file %s", be_lun->dev_path); 1211 return (error); 1212 } 1213 } 1214 1215 1216 file_data->cred = crhold(curthread->td_ucred); 1217 if (params->lun_size_bytes != 0) 1218 be_lun->size_bytes = params->lun_size_bytes; 1219 else 1220 be_lun->size_bytes = vattr.va_size; 1221 /* 1222 * We set the multi thread flag for file operations because all 1223 * filesystems (in theory) are capable of allowing multiple readers 1224 * of a file at once. So we want to get the maximum possible 1225 * concurrency. 1226 */ 1227 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1228 1229 /* 1230 * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1231 * With ZFS, it is 131072 bytes. Block sizes that large don't work 1232 * with disklabel and UFS on FreeBSD at least. Large block sizes 1233 * may not work with other OSes as well. So just export a sector 1234 * size of 512 bytes, which should work with any OS or 1235 * application. Since our backing is a file, any block size will 1236 * work fine for the backing store. 1237 */ 1238#if 0 1239 be_lun->blocksize= vattr.va_blocksize; 1240#endif 1241 if (params->blocksize_bytes != 0) 1242 be_lun->blocksize = params->blocksize_bytes; 1243 else 1244 be_lun->blocksize = 512; 1245 1246 /* 1247 * Sanity check. The media size has to be at least one 1248 * sector long. 1249 */ 1250 if (be_lun->size_bytes < be_lun->blocksize) { 1251 error = EINVAL; 1252 snprintf(req->error_str, sizeof(req->error_str), 1253 "file %s size %ju < block size %u", be_lun->dev_path, 1254 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1255 } 1256 return (error); 1257} 1258 1259static int 1260ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1261{ 1262 struct ctl_lun_create_params *params; 1263 struct vattr vattr; 1264 struct cdev *dev; 1265 struct cdevsw *devsw; 1266 int error; 1267 off_t ps, pss, po, pos; 1268 1269 params = &req->reqdata.create; 1270 1271 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1272 be_lun->dispatch = ctl_be_block_dispatch_dev; 1273 be_lun->lun_flush = ctl_be_block_flush_dev; 1274 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1275 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1276 &be_lun->backend.dev.dev_ref); 1277 if (be_lun->backend.dev.csw == NULL) 1278 panic("Unable to retrieve device switch"); 1279 1280 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1281 if (error) { 1282 snprintf(req->error_str, sizeof(req->error_str), 1283 "%s: error getting vnode attributes for device %s", 1284 __func__, be_lun->dev_path); 1285 return (error); 1286 } 1287 1288 dev = be_lun->vn->v_rdev; 1289 devsw = dev->si_devsw; 1290 if (!devsw->d_ioctl) { 1291 snprintf(req->error_str, sizeof(req->error_str), 1292 "%s: no d_ioctl for device %s!", __func__, 1293 be_lun->dev_path); 1294 return (ENODEV); 1295 } 1296 1297 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1298 (caddr_t)&be_lun->blocksize, FREAD, 1299 curthread); 1300 if (error) { 1301 snprintf(req->error_str, sizeof(req->error_str), 1302 "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1303 "on %s!", __func__, error, be_lun->dev_path); 1304 return (error); 1305 } 1306 1307 /* 1308 * If the user has asked for a blocksize that is greater than the 1309 * backing device's blocksize, we can do it only if the blocksize 1310 * the user is asking for is an even multiple of the underlying 1311 * device's blocksize. 1312 */ 1313 if ((params->blocksize_bytes != 0) 1314 && (params->blocksize_bytes > be_lun->blocksize)) { 1315 uint32_t bs_multiple, tmp_blocksize; 1316 1317 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1318 1319 tmp_blocksize = bs_multiple * be_lun->blocksize; 1320 1321 if (tmp_blocksize == params->blocksize_bytes) { 1322 be_lun->blocksize = params->blocksize_bytes; 1323 } else { 1324 snprintf(req->error_str, sizeof(req->error_str), 1325 "%s: requested blocksize %u is not an even " 1326 "multiple of backing device blocksize %u", 1327 __func__, params->blocksize_bytes, 1328 be_lun->blocksize); 1329 return (EINVAL); 1330 1331 } 1332 } else if ((params->blocksize_bytes != 0) 1333 && (params->blocksize_bytes != be_lun->blocksize)) { 1334 snprintf(req->error_str, sizeof(req->error_str), 1335 "%s: requested blocksize %u < backing device " 1336 "blocksize %u", __func__, params->blocksize_bytes, 1337 be_lun->blocksize); 1338 return (EINVAL); 1339 } 1340 1341 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1342 (caddr_t)&be_lun->size_bytes, FREAD, 1343 curthread); 1344 if (error) { 1345 snprintf(req->error_str, sizeof(req->error_str), 1346 "%s: error %d returned for DIOCGMEDIASIZE " 1347 " ioctl on %s!", __func__, error, 1348 be_lun->dev_path); 1349 return (error); 1350 } 1351 1352 if (params->lun_size_bytes != 0) { 1353 if (params->lun_size_bytes > be_lun->size_bytes) { 1354 snprintf(req->error_str, sizeof(req->error_str), 1355 "%s: requested LUN size %ju > backing device " 1356 "size %ju", __func__, 1357 (uintmax_t)params->lun_size_bytes, 1358 (uintmax_t)be_lun->size_bytes); 1359 return (EINVAL); 1360 } 1361 1362 be_lun->size_bytes = params->lun_size_bytes; 1363 } 1364 1365 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1366 (caddr_t)&ps, FREAD, curthread); 1367 if (error) 1368 ps = po = 0; 1369 else { 1370 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1371 (caddr_t)&po, FREAD, curthread); 1372 if (error) 1373 po = 0; 1374 } 1375 pss = ps / be_lun->blocksize; 1376 pos = po / be_lun->blocksize; 1377 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1378 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1379 be_lun->pblockexp = fls(pss) - 1; 1380 be_lun->pblockoff = (pss - pos) % pss; 1381 } 1382 1383 return (0); 1384} 1385 1386static int 1387ctl_be_block_close(struct ctl_be_block_lun *be_lun) 1388{ 1389 DROP_GIANT(); 1390 if (be_lun->vn) { 1391 int flags = FREAD | FWRITE; 1392 1393 switch (be_lun->dev_type) { 1394 case CTL_BE_BLOCK_DEV: 1395 if (be_lun->backend.dev.csw) { 1396 dev_relthread(be_lun->backend.dev.cdev, 1397 be_lun->backend.dev.dev_ref); 1398 be_lun->backend.dev.csw = NULL; 1399 be_lun->backend.dev.cdev = NULL; 1400 } 1401 break; 1402 case CTL_BE_BLOCK_FILE: 1403 break; 1404 case CTL_BE_BLOCK_NONE: 1405 break; 1406 default: 1407 panic("Unexpected backend type."); 1408 break; 1409 } 1410 1411 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1412 be_lun->vn = NULL; 1413 1414 switch (be_lun->dev_type) { 1415 case CTL_BE_BLOCK_DEV: 1416 break; 1417 case CTL_BE_BLOCK_FILE: 1418 if (be_lun->backend.file.cred != NULL) { 1419 crfree(be_lun->backend.file.cred); 1420 be_lun->backend.file.cred = NULL; 1421 } 1422 break; 1423 case CTL_BE_BLOCK_NONE: 1424 break; 1425 default: 1426 panic("Unexpected backend type."); 1427 break; 1428 } 1429 } 1430 PICKUP_GIANT(); 1431 1432 return (0); 1433} 1434 1435static int 1436ctl_be_block_open(struct ctl_be_block_softc *softc, 1437 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1438{ 1439 struct nameidata nd; 1440 int flags; 1441 int error; 1442 1443 /* 1444 * XXX KDM allow a read-only option? 1445 */ 1446 flags = FREAD | FWRITE; 1447 error = 0; 1448 1449 if (rootvnode == NULL) { 1450 snprintf(req->error_str, sizeof(req->error_str), 1451 "%s: Root filesystem is not mounted", __func__); 1452 return (1); 1453 } 1454 1455 if (!curthread->td_proc->p_fd->fd_cdir) { 1456 curthread->td_proc->p_fd->fd_cdir = rootvnode; 1457 VREF(rootvnode); 1458 } 1459 if (!curthread->td_proc->p_fd->fd_rdir) { 1460 curthread->td_proc->p_fd->fd_rdir = rootvnode; 1461 VREF(rootvnode); 1462 } 1463 if (!curthread->td_proc->p_fd->fd_jdir) { 1464 curthread->td_proc->p_fd->fd_jdir = rootvnode; 1465 VREF(rootvnode); 1466 } 1467 1468 again: 1469 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1470 error = vn_open(&nd, &flags, 0, NULL); 1471 if (error) { 1472 /* 1473 * This is the only reasonable guess we can make as far as 1474 * path if the user doesn't give us a fully qualified path. 1475 * If they want to specify a file, they need to specify the 1476 * full path. 1477 */ 1478 if (be_lun->dev_path[0] != '/') { 1479 char *dev_path = "/dev/"; 1480 char *dev_name; 1481 1482 /* Try adding device path at beginning of name */ 1483 dev_name = malloc(strlen(be_lun->dev_path) 1484 + strlen(dev_path) + 1, 1485 M_CTLBLK, M_WAITOK); 1486 if (dev_name) { 1487 sprintf(dev_name, "%s%s", dev_path, 1488 be_lun->dev_path); 1489 free(be_lun->dev_path, M_CTLBLK); 1490 be_lun->dev_path = dev_name; 1491 goto again; 1492 } 1493 } 1494 snprintf(req->error_str, sizeof(req->error_str), 1495 "%s: error opening %s", __func__, be_lun->dev_path); 1496 return (error); 1497 } 1498 1499 NDFREE(&nd, NDF_ONLY_PNBUF); 1500 1501 be_lun->vn = nd.ni_vp; 1502 1503 /* We only support disks and files. */ 1504 if (vn_isdisk(be_lun->vn, &error)) { 1505 error = ctl_be_block_open_dev(be_lun, req); 1506 } else if (be_lun->vn->v_type == VREG) { 1507 error = ctl_be_block_open_file(be_lun, req); 1508 } else { 1509 error = EINVAL; 1510 snprintf(req->error_str, sizeof(req->error_str), 1511 "%s is not a disk or plain file", be_lun->dev_path); 1512 } 1513 VOP_UNLOCK(be_lun->vn, 0); 1514 1515 if (error != 0) { 1516 ctl_be_block_close(be_lun); 1517 return (error); 1518 } 1519 1520 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1521 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1522 1523 return (0); 1524} 1525 1526static int 1527ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1528{ 1529 struct ctl_be_block_lun *be_lun; 1530 struct ctl_lun_create_params *params; 1531 struct ctl_be_arg *file_arg; 1532 char tmpstr[32]; 1533 int retval, num_threads; 1534 int i; 1535 1536 params = &req->reqdata.create; 1537 retval = 0; 1538 1539 num_threads = cbb_num_threads; 1540 1541 file_arg = NULL; 1542 1543 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1544 1545 be_lun->softc = softc; 1546 STAILQ_INIT(&be_lun->input_queue); 1547 STAILQ_INIT(&be_lun->config_write_queue); 1548 STAILQ_INIT(&be_lun->datamove_queue); 1549 STAILQ_INIT(&be_lun->ctl_be_lun.options); 1550 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1551 mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF); 1552 1553 be_lun->lun_zone = uma_zcreate(be_lun->lunname, MAXPHYS, 1554 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1555 1556 if (be_lun->lun_zone == NULL) { 1557 snprintf(req->error_str, sizeof(req->error_str), 1558 "%s: error allocating UMA zone", __func__); 1559 goto bailout_error; 1560 } 1561 1562 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1563 be_lun->ctl_be_lun.lun_type = params->device_type; 1564 else 1565 be_lun->ctl_be_lun.lun_type = T_DIRECT; 1566 1567 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1568 for (i = 0; i < req->num_be_args; i++) { 1569 if (strcmp(req->kern_be_args[i].kname, "file") == 0) { 1570 file_arg = &req->kern_be_args[i]; 1571 break; 1572 } 1573 } 1574 1575 if (file_arg == NULL) { 1576 snprintf(req->error_str, sizeof(req->error_str), 1577 "%s: no file argument specified", __func__); 1578 goto bailout_error; 1579 } 1580 1581 be_lun->dev_path = malloc(file_arg->vallen, M_CTLBLK, 1582 M_WAITOK | M_ZERO); 1583 1584 strlcpy(be_lun->dev_path, (char *)file_arg->kvalue, 1585 file_arg->vallen); 1586 1587 retval = ctl_be_block_open(softc, be_lun, req); 1588 if (retval != 0) { 1589 retval = 0; 1590 goto bailout_error; 1591 } 1592 1593 /* 1594 * Tell the user the size of the file/device. 1595 */ 1596 params->lun_size_bytes = be_lun->size_bytes; 1597 1598 /* 1599 * The maximum LBA is the size - 1. 1600 */ 1601 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1602 } else { 1603 /* 1604 * For processor devices, we don't have any size. 1605 */ 1606 be_lun->blocksize = 0; 1607 be_lun->pblockexp = 0; 1608 be_lun->pblockoff = 0; 1609 be_lun->size_blocks = 0; 1610 be_lun->size_bytes = 0; 1611 be_lun->ctl_be_lun.maxlba = 0; 1612 params->lun_size_bytes = 0; 1613 1614 /* 1615 * Default to just 1 thread for processor devices. 1616 */ 1617 num_threads = 1; 1618 } 1619 1620 /* 1621 * XXX This searching loop might be refactored to be combined with 1622 * the loop above, 1623 */ 1624 for (i = 0; i < req->num_be_args; i++) { 1625 if (strcmp(req->kern_be_args[i].kname, "num_threads") == 0) { 1626 struct ctl_be_arg *thread_arg; 1627 char num_thread_str[16]; 1628 int tmp_num_threads; 1629 1630 1631 thread_arg = &req->kern_be_args[i]; 1632 1633 strlcpy(num_thread_str, (char *)thread_arg->kvalue, 1634 min(thread_arg->vallen, 1635 sizeof(num_thread_str))); 1636 1637 tmp_num_threads = strtol(num_thread_str, NULL, 0); 1638 1639 /* 1640 * We don't let the user specify less than one 1641 * thread, but hope he's clueful enough not to 1642 * specify 1000 threads. 1643 */ 1644 if (tmp_num_threads < 1) { 1645 snprintf(req->error_str, sizeof(req->error_str), 1646 "%s: invalid number of threads %s", 1647 __func__, num_thread_str); 1648 goto bailout_error; 1649 } 1650 1651 num_threads = tmp_num_threads; 1652 } else if (strcmp(req->kern_be_args[i].kname, "file") != 0 && 1653 strcmp(req->kern_be_args[i].kname, "dev") != 0) { 1654 struct ctl_be_lun_option *opt; 1655 1656 opt = malloc(sizeof(*opt), M_CTLBLK, M_WAITOK); 1657 opt->name = malloc(strlen(req->kern_be_args[i].kname) + 1, M_CTLBLK, M_WAITOK); 1658 strcpy(opt->name, req->kern_be_args[i].kname); 1659 opt->value = malloc(strlen(req->kern_be_args[i].kvalue) + 1, M_CTLBLK, M_WAITOK); 1660 strcpy(opt->value, req->kern_be_args[i].kvalue); 1661 STAILQ_INSERT_TAIL(&be_lun->ctl_be_lun.options, opt, links); 1662 } 1663 } 1664 1665 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1666 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1667 be_lun->ctl_be_lun.be_lun = be_lun; 1668 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1669 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 1670 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 1671 /* Tell the user the blocksize we ended up using */ 1672 params->blocksize_bytes = be_lun->blocksize; 1673 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1674 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1675 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1676 } else 1677 be_lun->ctl_be_lun.req_lun_id = 0; 1678 1679 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1680 be_lun->ctl_be_lun.lun_config_status = 1681 ctl_be_block_lun_config_status; 1682 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1683 1684 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1685 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1686 softc->num_luns); 1687 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1688 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1689 sizeof(tmpstr))); 1690 1691 /* Tell the user what we used for a serial number */ 1692 strncpy((char *)params->serial_num, tmpstr, 1693 ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1694 } else { 1695 strncpy((char *)be_lun->ctl_be_lun.serial_num, 1696 params->serial_num, 1697 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1698 sizeof(params->serial_num))); 1699 } 1700 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1701 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1702 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1703 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1704 sizeof(tmpstr))); 1705 1706 /* Tell the user what we used for a device ID */ 1707 strncpy((char *)params->device_id, tmpstr, 1708 ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1709 } else { 1710 strncpy((char *)be_lun->ctl_be_lun.device_id, 1711 params->device_id, 1712 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1713 sizeof(params->device_id))); 1714 } 1715 1716 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 1717 1718 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 1719 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 1720 1721 if (be_lun->io_taskqueue == NULL) { 1722 snprintf(req->error_str, sizeof(req->error_str), 1723 "%s: Unable to create taskqueue", __func__); 1724 goto bailout_error; 1725 } 1726 1727 /* 1728 * Note that we start the same number of threads by default for 1729 * both the file case and the block device case. For the file 1730 * case, we need multiple threads to allow concurrency, because the 1731 * vnode interface is designed to be a blocking interface. For the 1732 * block device case, ZFS zvols at least will block the caller's 1733 * context in many instances, and so we need multiple threads to 1734 * overcome that problem. Other block devices don't need as many 1735 * threads, but they shouldn't cause too many problems. 1736 * 1737 * If the user wants to just have a single thread for a block 1738 * device, he can specify that when the LUN is created, or change 1739 * the tunable/sysctl to alter the default number of threads. 1740 */ 1741 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 1742 /*num threads*/num_threads, 1743 /*priority*/PWAIT, 1744 /*thread name*/ 1745 "%s taskq", be_lun->lunname); 1746 1747 if (retval != 0) 1748 goto bailout_error; 1749 1750 be_lun->num_threads = num_threads; 1751 1752 mtx_lock(&softc->lock); 1753 softc->num_luns++; 1754 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 1755 1756 mtx_unlock(&softc->lock); 1757 1758 retval = ctl_add_lun(&be_lun->ctl_be_lun); 1759 if (retval != 0) { 1760 mtx_lock(&softc->lock); 1761 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1762 links); 1763 softc->num_luns--; 1764 mtx_unlock(&softc->lock); 1765 snprintf(req->error_str, sizeof(req->error_str), 1766 "%s: ctl_add_lun() returned error %d, see dmesg for " 1767 "details", __func__, retval); 1768 retval = 0; 1769 goto bailout_error; 1770 } 1771 1772 mtx_lock(&softc->lock); 1773 1774 /* 1775 * Tell the config_status routine that we're waiting so it won't 1776 * clean up the LUN in the event of an error. 1777 */ 1778 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1779 1780 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 1781 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1782 if (retval == EINTR) 1783 break; 1784 } 1785 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1786 1787 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 1788 snprintf(req->error_str, sizeof(req->error_str), 1789 "%s: LUN configuration error, see dmesg for details", 1790 __func__); 1791 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1792 links); 1793 softc->num_luns--; 1794 mtx_unlock(&softc->lock); 1795 goto bailout_error; 1796 } else { 1797 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 1798 } 1799 1800 mtx_unlock(&softc->lock); 1801 1802 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 1803 be_lun->blocksize, 1804 DEVSTAT_ALL_SUPPORTED, 1805 be_lun->ctl_be_lun.lun_type 1806 | DEVSTAT_TYPE_IF_OTHER, 1807 DEVSTAT_PRIORITY_OTHER); 1808 1809 1810 req->status = CTL_LUN_OK; 1811 1812 return (retval); 1813 1814bailout_error: 1815 req->status = CTL_LUN_ERROR; 1816 1817 ctl_be_block_close(be_lun); 1818 1819 free(be_lun->dev_path, M_CTLBLK); 1820 free(be_lun, M_CTLBLK); 1821 1822 return (retval); 1823} 1824 1825static int 1826ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1827{ 1828 struct ctl_lun_rm_params *params; 1829 struct ctl_be_block_lun *be_lun; 1830 int retval; 1831 1832 params = &req->reqdata.rm; 1833 1834 mtx_lock(&softc->lock); 1835 1836 be_lun = NULL; 1837 1838 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 1839 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 1840 break; 1841 } 1842 mtx_unlock(&softc->lock); 1843 1844 if (be_lun == NULL) { 1845 snprintf(req->error_str, sizeof(req->error_str), 1846 "%s: LUN %u is not managed by the block backend", 1847 __func__, params->lun_id); 1848 goto bailout_error; 1849 } 1850 1851 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 1852 1853 if (retval != 0) { 1854 snprintf(req->error_str, sizeof(req->error_str), 1855 "%s: error %d returned from ctl_disable_lun() for " 1856 "LUN %d", __func__, retval, params->lun_id); 1857 goto bailout_error; 1858 1859 } 1860 1861 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 1862 if (retval != 0) { 1863 snprintf(req->error_str, sizeof(req->error_str), 1864 "%s: error %d returned from ctl_invalidate_lun() for " 1865 "LUN %d", __func__, retval, params->lun_id); 1866 goto bailout_error; 1867 } 1868 1869 mtx_lock(&softc->lock); 1870 1871 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1872 1873 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1874 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1875 if (retval == EINTR) 1876 break; 1877 } 1878 1879 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1880 1881 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1882 snprintf(req->error_str, sizeof(req->error_str), 1883 "%s: interrupted waiting for LUN to be freed", 1884 __func__); 1885 mtx_unlock(&softc->lock); 1886 goto bailout_error; 1887 } 1888 1889 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 1890 1891 softc->num_luns--; 1892 mtx_unlock(&softc->lock); 1893 1894 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 1895 1896 taskqueue_free(be_lun->io_taskqueue); 1897 1898 ctl_be_block_close(be_lun); 1899 1900 if (be_lun->disk_stats != NULL) 1901 devstat_remove_entry(be_lun->disk_stats); 1902 1903 uma_zdestroy(be_lun->lun_zone); 1904 1905 free(be_lun->dev_path, M_CTLBLK); 1906 1907 free(be_lun, M_CTLBLK); 1908 1909 req->status = CTL_LUN_OK; 1910 1911 return (0); 1912 1913bailout_error: 1914 1915 req->status = CTL_LUN_ERROR; 1916 1917 return (0); 1918} 1919 1920static int 1921ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 1922 struct ctl_lun_req *req) 1923{ 1924 struct vattr vattr; 1925 int error; 1926 struct ctl_lun_modify_params *params; 1927 1928 params = &req->reqdata.modify; 1929 1930 if (params->lun_size_bytes != 0) { 1931 be_lun->size_bytes = params->lun_size_bytes; 1932 } else { 1933 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1934 if (error != 0) { 1935 snprintf(req->error_str, sizeof(req->error_str), 1936 "error calling VOP_GETATTR() for file %s", 1937 be_lun->dev_path); 1938 return (error); 1939 } 1940 1941 be_lun->size_bytes = vattr.va_size; 1942 } 1943 1944 return (0); 1945} 1946 1947static int 1948ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 1949 struct ctl_lun_req *req) 1950{ 1951 struct cdev *dev; 1952 struct cdevsw *devsw; 1953 int error; 1954 struct ctl_lun_modify_params *params; 1955 uint64_t size_bytes; 1956 1957 params = &req->reqdata.modify; 1958 1959 dev = be_lun->vn->v_rdev; 1960 devsw = dev->si_devsw; 1961 if (!devsw->d_ioctl) { 1962 snprintf(req->error_str, sizeof(req->error_str), 1963 "%s: no d_ioctl for device %s!", __func__, 1964 be_lun->dev_path); 1965 return (ENODEV); 1966 } 1967 1968 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1969 (caddr_t)&size_bytes, FREAD, 1970 curthread); 1971 if (error) { 1972 snprintf(req->error_str, sizeof(req->error_str), 1973 "%s: error %d returned for DIOCGMEDIASIZE ioctl " 1974 "on %s!", __func__, error, be_lun->dev_path); 1975 return (error); 1976 } 1977 1978 if (params->lun_size_bytes != 0) { 1979 if (params->lun_size_bytes > size_bytes) { 1980 snprintf(req->error_str, sizeof(req->error_str), 1981 "%s: requested LUN size %ju > backing device " 1982 "size %ju", __func__, 1983 (uintmax_t)params->lun_size_bytes, 1984 (uintmax_t)size_bytes); 1985 return (EINVAL); 1986 } 1987 1988 be_lun->size_bytes = params->lun_size_bytes; 1989 } else { 1990 be_lun->size_bytes = size_bytes; 1991 } 1992 1993 return (0); 1994} 1995 1996static int 1997ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1998{ 1999 struct ctl_lun_modify_params *params; 2000 struct ctl_be_block_lun *be_lun; 2001 int error; 2002 2003 params = &req->reqdata.modify; 2004 2005 mtx_lock(&softc->lock); 2006 2007 be_lun = NULL; 2008 2009 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2010 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2011 break; 2012 } 2013 mtx_unlock(&softc->lock); 2014 2015 if (be_lun == NULL) { 2016 snprintf(req->error_str, sizeof(req->error_str), 2017 "%s: LUN %u is not managed by the block backend", 2018 __func__, params->lun_id); 2019 goto bailout_error; 2020 } 2021 2022 if (params->lun_size_bytes != 0) { 2023 if (params->lun_size_bytes < be_lun->blocksize) { 2024 snprintf(req->error_str, sizeof(req->error_str), 2025 "%s: LUN size %ju < blocksize %u", __func__, 2026 params->lun_size_bytes, be_lun->blocksize); 2027 goto bailout_error; 2028 } 2029 } 2030 2031 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2032 2033 if (be_lun->vn->v_type == VREG) 2034 error = ctl_be_block_modify_file(be_lun, req); 2035 else 2036 error = ctl_be_block_modify_dev(be_lun, req); 2037 2038 VOP_UNLOCK(be_lun->vn, 0); 2039 2040 if (error != 0) 2041 goto bailout_error; 2042 2043 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2044 2045 /* 2046 * The maximum LBA is the size - 1. 2047 * 2048 * XXX: Note that this field is being updated without locking, 2049 * which might cause problems on 32-bit architectures. 2050 */ 2051 be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2052 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2053 2054 /* Tell the user the exact size we ended up using */ 2055 params->lun_size_bytes = be_lun->size_bytes; 2056 2057 req->status = CTL_LUN_OK; 2058 2059 return (0); 2060 2061bailout_error: 2062 req->status = CTL_LUN_ERROR; 2063 2064 return (0); 2065} 2066 2067static void 2068ctl_be_block_lun_shutdown(void *be_lun) 2069{ 2070 struct ctl_be_block_lun *lun; 2071 struct ctl_be_block_softc *softc; 2072 2073 lun = (struct ctl_be_block_lun *)be_lun; 2074 2075 softc = lun->softc; 2076 2077 mtx_lock(&softc->lock); 2078 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2079 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2080 wakeup(lun); 2081 mtx_unlock(&softc->lock); 2082 2083} 2084 2085static void 2086ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2087{ 2088 struct ctl_be_block_lun *lun; 2089 struct ctl_be_block_softc *softc; 2090 2091 lun = (struct ctl_be_block_lun *)be_lun; 2092 softc = lun->softc; 2093 2094 if (status == CTL_LUN_CONFIG_OK) { 2095 mtx_lock(&softc->lock); 2096 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2097 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2098 wakeup(lun); 2099 mtx_unlock(&softc->lock); 2100 2101 /* 2102 * We successfully added the LUN, attempt to enable it. 2103 */ 2104 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2105 printf("%s: ctl_enable_lun() failed!\n", __func__); 2106 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2107 printf("%s: ctl_invalidate_lun() failed!\n", 2108 __func__); 2109 } 2110 } 2111 2112 return; 2113 } 2114 2115 2116 mtx_lock(&softc->lock); 2117 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2118 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2119 wakeup(lun); 2120 mtx_unlock(&softc->lock); 2121} 2122 2123 2124static int 2125ctl_be_block_config_write(union ctl_io *io) 2126{ 2127 struct ctl_be_block_lun *be_lun; 2128 struct ctl_be_lun *ctl_be_lun; 2129 int retval; 2130 2131 retval = 0; 2132 2133 DPRINTF("entered\n"); 2134 2135 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2136 CTL_PRIV_BACKEND_LUN].ptr; 2137 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2138 2139 switch (io->scsiio.cdb[0]) { 2140 case SYNCHRONIZE_CACHE: 2141 case SYNCHRONIZE_CACHE_16: 2142 /* 2143 * The upper level CTL code will filter out any CDBs with 2144 * the immediate bit set and return the proper error. 2145 * 2146 * We don't really need to worry about what LBA range the 2147 * user asked to be synced out. When they issue a sync 2148 * cache command, we'll sync out the whole thing. 2149 */ 2150 mtx_lock(&be_lun->lock); 2151 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2152 links); 2153 mtx_unlock(&be_lun->lock); 2154 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2155 break; 2156 case START_STOP_UNIT: { 2157 struct scsi_start_stop_unit *cdb; 2158 2159 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2160 2161 if (cdb->how & SSS_START) 2162 retval = ctl_start_lun(ctl_be_lun); 2163 else { 2164 retval = ctl_stop_lun(ctl_be_lun); 2165 /* 2166 * XXX KDM Copan-specific offline behavior. 2167 * Figure out a reasonable way to port this? 2168 */ 2169#ifdef NEEDTOPORT 2170 if ((retval == 0) 2171 && (cdb->byte2 & SSS_ONOFFLINE)) 2172 retval = ctl_lun_offline(ctl_be_lun); 2173#endif 2174 } 2175 2176 /* 2177 * In general, the above routines should not fail. They 2178 * just set state for the LUN. So we've got something 2179 * pretty wrong here if we can't start or stop the LUN. 2180 */ 2181 if (retval != 0) { 2182 ctl_set_internal_failure(&io->scsiio, 2183 /*sks_valid*/ 1, 2184 /*retry_count*/ 0xf051); 2185 retval = CTL_RETVAL_COMPLETE; 2186 } else { 2187 ctl_set_success(&io->scsiio); 2188 } 2189 ctl_config_write_done(io); 2190 break; 2191 } 2192 default: 2193 ctl_set_invalid_opcode(&io->scsiio); 2194 ctl_config_write_done(io); 2195 retval = CTL_RETVAL_COMPLETE; 2196 break; 2197 } 2198 2199 return (retval); 2200 2201} 2202 2203static int 2204ctl_be_block_config_read(union ctl_io *io) 2205{ 2206 return (0); 2207} 2208 2209static int 2210ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2211{ 2212 struct ctl_be_block_lun *lun; 2213 int retval; 2214 2215 lun = (struct ctl_be_block_lun *)be_lun; 2216 retval = 0; 2217 2218 retval = sbuf_printf(sb, "<num_threads>"); 2219 2220 if (retval != 0) 2221 goto bailout; 2222 2223 retval = sbuf_printf(sb, "%d", lun->num_threads); 2224 2225 if (retval != 0) 2226 goto bailout; 2227 2228 retval = sbuf_printf(sb, "</num_threads>"); 2229 2230 /* 2231 * For processor devices, we don't have a path variable. 2232 */ 2233 if ((retval != 0) 2234 || (lun->dev_path == NULL)) 2235 goto bailout; 2236 2237 retval = sbuf_printf(sb, "<file>"); 2238 2239 if (retval != 0) 2240 goto bailout; 2241 2242 retval = ctl_sbuf_printf_esc(sb, lun->dev_path); 2243 2244 if (retval != 0) 2245 goto bailout; 2246 2247 retval = sbuf_printf(sb, "</file>\n"); 2248 2249bailout: 2250 2251 return (retval); 2252} 2253 2254int 2255ctl_be_block_init(void) 2256{ 2257 struct ctl_be_block_softc *softc; 2258 int retval; 2259 2260 softc = &backend_block_softc; 2261 retval = 0; 2262 2263 mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF); 2264 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2265 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2266 STAILQ_INIT(&softc->disk_list); 2267 STAILQ_INIT(&softc->lun_list); 2268 2269 return (retval); 2270} 2271