ctl_backend_block.c revision 230334
1229997Sken/*- 2229997Sken * Copyright (c) 2003 Silicon Graphics International Corp. 3229997Sken * Copyright (c) 2009-2011 Spectra Logic Corporation 4229997Sken * All rights reserved. 5229997Sken * 6229997Sken * Redistribution and use in source and binary forms, with or without 7229997Sken * modification, are permitted provided that the following conditions 8229997Sken * are met: 9229997Sken * 1. Redistributions of source code must retain the above copyright 10229997Sken * notice, this list of conditions, and the following disclaimer, 11229997Sken * without modification. 12229997Sken * 2. Redistributions in binary form must reproduce at minimum a disclaimer 13229997Sken * substantially similar to the "NO WARRANTY" disclaimer below 14229997Sken * ("Disclaimer") and any redistribution must be conditioned upon 15229997Sken * including a substantially similar Disclaimer requirement for further 16229997Sken * binary redistribution. 17229997Sken * 18229997Sken * NO WARRANTY 19229997Sken * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 20229997Sken * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 21229997Sken * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 22229997Sken * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 23229997Sken * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24229997Sken * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25229997Sken * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26229997Sken * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 27229997Sken * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 28229997Sken * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29229997Sken * POSSIBILITY OF SUCH DAMAGES. 30229997Sken * 31229997Sken * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 32229997Sken */ 33229997Sken/* 34229997Sken * CAM Target Layer driver backend for block devices. 35229997Sken * 36229997Sken * Author: Ken Merry <ken@FreeBSD.org> 37229997Sken */ 38229997Sken#include <sys/cdefs.h> 39229997Sken__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl_backend_block.c 230334 2012-01-19 18:42:03Z ken $"); 40229997Sken 41229997Sken#include <opt_kdtrace.h> 42229997Sken 43229997Sken#include <sys/param.h> 44229997Sken#include <sys/systm.h> 45229997Sken#include <sys/kernel.h> 46229997Sken#include <sys/types.h> 47229997Sken#include <sys/kthread.h> 48229997Sken#include <sys/bio.h> 49229997Sken#include <sys/fcntl.h> 50229997Sken#include <sys/lock.h> 51229997Sken#include <sys/mutex.h> 52229997Sken#include <sys/condvar.h> 53229997Sken#include <sys/malloc.h> 54229997Sken#include <sys/conf.h> 55229997Sken#include <sys/ioccom.h> 56229997Sken#include <sys/queue.h> 57229997Sken#include <sys/sbuf.h> 58229997Sken#include <sys/endian.h> 59229997Sken#include <sys/uio.h> 60229997Sken#include <sys/buf.h> 61229997Sken#include <sys/taskqueue.h> 62229997Sken#include <sys/vnode.h> 63229997Sken#include <sys/namei.h> 64229997Sken#include <sys/mount.h> 65229997Sken#include <sys/disk.h> 66229997Sken#include <sys/fcntl.h> 67229997Sken#include <sys/filedesc.h> 68229997Sken#include <sys/proc.h> 69229997Sken#include <sys/pcpu.h> 70229997Sken#include <sys/module.h> 71229997Sken#include <sys/sdt.h> 72229997Sken#include <sys/devicestat.h> 73229997Sken#include <sys/sysctl.h> 74229997Sken 75229997Sken#include <geom/geom.h> 76229997Sken 77229997Sken#include <cam/cam.h> 78229997Sken#include <cam/scsi/scsi_all.h> 79229997Sken#include <cam/scsi/scsi_da.h> 80229997Sken#include <cam/ctl/ctl_io.h> 81229997Sken#include <cam/ctl/ctl.h> 82229997Sken#include <cam/ctl/ctl_backend.h> 83229997Sken#include <cam/ctl/ctl_frontend_internal.h> 84229997Sken#include <cam/ctl/ctl_ioctl.h> 85229997Sken#include <cam/ctl/ctl_scsi_all.h> 86229997Sken#include <cam/ctl/ctl_error.h> 87229997Sken 88229997Sken/* 89229997Sken * The idea here is that we'll allocate enough S/G space to hold a 16MB 90229997Sken * I/O. If we get an I/O larger than that, we'll reject it. 91229997Sken */ 92229997Sken#define CTLBLK_MAX_IO_SIZE (16 * 1024 * 1024) 93229997Sken#define CTLBLK_MAX_SEGS (CTLBLK_MAX_IO_SIZE / MAXPHYS) + 1 94229997Sken 95229997Sken#ifdef CTLBLK_DEBUG 96229997Sken#define DPRINTF(fmt, args...) \ 97229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 98229997Sken#else 99229997Sken#define DPRINTF(fmt, args...) do {} while(0) 100229997Sken#endif 101229997Sken 102229997SkenSDT_PROVIDER_DEFINE(cbb); 103229997Sken 104229997Skentypedef enum { 105229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 106229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 107229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 108229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 109229997Sken} ctl_be_block_lun_flags; 110229997Sken 111229997Skentypedef enum { 112229997Sken CTL_BE_BLOCK_NONE, 113229997Sken CTL_BE_BLOCK_DEV, 114229997Sken CTL_BE_BLOCK_FILE 115229997Sken} ctl_be_block_type; 116229997Sken 117229997Skenstruct ctl_be_block_devdata { 118229997Sken struct cdev *cdev; 119229997Sken struct cdevsw *csw; 120229997Sken int dev_ref; 121229997Sken}; 122229997Sken 123229997Skenstruct ctl_be_block_filedata { 124229997Sken struct ucred *cred; 125229997Sken}; 126229997Sken 127229997Skenunion ctl_be_block_bedata { 128229997Sken struct ctl_be_block_devdata dev; 129229997Sken struct ctl_be_block_filedata file; 130229997Sken}; 131229997Sken 132229997Skenstruct ctl_be_block_io; 133229997Skenstruct ctl_be_block_lun; 134229997Sken 135229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 136229997Sken struct ctl_be_block_io *beio); 137229997Sken 138229997Sken/* 139229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 140229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 141229997Sken */ 142229997Skenstruct ctl_be_block_lun { 143229997Sken struct ctl_block_disk *disk; 144229997Sken char lunname[32]; 145229997Sken char *dev_path; 146229997Sken ctl_be_block_type dev_type; 147229997Sken struct vnode *vn; 148229997Sken union ctl_be_block_bedata backend; 149229997Sken cbb_dispatch_t dispatch; 150229997Sken cbb_dispatch_t lun_flush; 151229997Sken struct mtx lock; 152229997Sken uma_zone_t lun_zone; 153229997Sken uint64_t size_blocks; 154229997Sken uint64_t size_bytes; 155229997Sken uint32_t blocksize; 156229997Sken int blocksize_shift; 157229997Sken struct ctl_be_block_softc *softc; 158229997Sken struct devstat *disk_stats; 159229997Sken ctl_be_block_lun_flags flags; 160229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 161229997Sken struct ctl_be_lun ctl_be_lun; 162229997Sken struct taskqueue *io_taskqueue; 163229997Sken struct task io_task; 164229997Sken int num_threads; 165229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 166229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 167229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 168229997Sken}; 169229997Sken 170229997Sken/* 171229997Sken * Overall softc structure for the block backend module. 172229997Sken */ 173229997Skenstruct ctl_be_block_softc { 174229997Sken STAILQ_HEAD(, ctl_be_block_io) beio_free_queue; 175229997Sken struct mtx lock; 176229997Sken int prealloc_beio; 177229997Sken int num_disks; 178229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 179229997Sken int num_luns; 180229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 181229997Sken}; 182229997Sken 183229997Skenstatic struct ctl_be_block_softc backend_block_softc; 184229997Sken 185229997Sken/* 186229997Sken * Per-I/O information. 187229997Sken */ 188229997Skenstruct ctl_be_block_io { 189229997Sken union ctl_io *io; 190229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 191229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 192229997Sken int bio_cmd; 193229997Sken int bio_flags; 194229997Sken int num_segs; 195229997Sken int num_bios_sent; 196229997Sken int num_bios_done; 197229997Sken int send_complete; 198229997Sken int num_errors; 199229997Sken struct bintime ds_t0; 200229997Sken devstat_tag_type ds_tag_type; 201229997Sken devstat_trans_flags ds_trans_type; 202229997Sken uint64_t io_len; 203229997Sken uint64_t io_offset; 204229997Sken struct ctl_be_block_softc *softc; 205229997Sken struct ctl_be_block_lun *lun; 206229997Sken STAILQ_ENTRY(ctl_be_block_io) links; 207229997Sken}; 208229997Sken 209229997Skenstatic int cbb_num_threads = 14; 210229997SkenTUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 211229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 212229997Sken "CAM Target Layer Block Backend"); 213229997SkenSYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 214229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 215229997Sken 216229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 217229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 218229997Skenstatic int ctl_grow_beio(struct ctl_be_block_softc *softc, int count); 219229997Sken#if 0 220229997Skenstatic void ctl_shrink_beio(struct ctl_be_block_softc *softc); 221229997Sken#endif 222229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 223229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 224229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 225229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 226229997Sken struct ctl_be_block_io *beio); 227229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 228229997Sken struct ctl_be_block_io *beio); 229229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 230229997Sken struct ctl_be_block_io *beio); 231229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 232229997Sken struct ctl_be_block_io *beio); 233229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 234229997Sken union ctl_io *io); 235229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 236229997Sken union ctl_io *io); 237229997Skenstatic void ctl_be_block_worker(void *context, int pending); 238229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 239229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 240229997Sken int flag, struct thread *td); 241229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 242229997Sken struct ctl_lun_req *req); 243229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 244229997Sken struct ctl_lun_req *req); 245229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 246229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 247229997Sken struct ctl_be_block_lun *be_lun, 248229997Sken struct ctl_lun_req *req); 249229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 250229997Sken struct ctl_lun_req *req); 251229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 252229997Sken struct ctl_lun_req *req); 253229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 254229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 255229997Sken ctl_lun_config_status status); 256229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 257229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 258229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 259229997Skenint ctl_be_block_init(void); 260229997Sken 261229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 262229997Sken{ 263230334Sken .name = "block", 264230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 265230334Sken .init = ctl_be_block_init, 266230334Sken .data_submit = ctl_be_block_submit, 267230334Sken .data_move_done = ctl_be_block_move_done, 268230334Sken .config_read = ctl_be_block_config_read, 269230334Sken .config_write = ctl_be_block_config_write, 270230334Sken .ioctl = ctl_be_block_ioctl, 271230334Sken .lun_info = ctl_be_block_lun_info 272229997Sken}; 273229997Sken 274229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 275229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 276229997Sken 277229997Skenstatic struct ctl_be_block_io * 278229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 279229997Sken{ 280229997Sken struct ctl_be_block_io *beio; 281229997Sken int count; 282229997Sken 283229997Sken mtx_lock(&softc->lock); 284229997Sken 285229997Sken beio = STAILQ_FIRST(&softc->beio_free_queue); 286229997Sken if (beio != NULL) { 287229997Sken STAILQ_REMOVE(&softc->beio_free_queue, beio, 288229997Sken ctl_be_block_io, links); 289229997Sken } 290229997Sken mtx_unlock(&softc->lock); 291229997Sken 292229997Sken if (beio != NULL) { 293229997Sken bzero(beio, sizeof(*beio)); 294229997Sken beio->softc = softc; 295229997Sken return (beio); 296229997Sken } 297229997Sken 298229997Sken for (;;) { 299229997Sken 300229997Sken count = ctl_grow_beio(softc, /*count*/ 10); 301229997Sken 302229997Sken /* 303229997Sken * This shouldn't be possible, since ctl_grow_beio() uses a 304229997Sken * blocking malloc. 305229997Sken */ 306229997Sken if (count == 0) 307229997Sken return (NULL); 308229997Sken 309229997Sken /* 310229997Sken * Since we have to drop the lock when we're allocating beio 311229997Sken * structures, it's possible someone else can come along and 312229997Sken * allocate the beio's we've just allocated. 313229997Sken */ 314229997Sken mtx_lock(&softc->lock); 315229997Sken beio = STAILQ_FIRST(&softc->beio_free_queue); 316229997Sken if (beio != NULL) { 317229997Sken STAILQ_REMOVE(&softc->beio_free_queue, beio, 318229997Sken ctl_be_block_io, links); 319229997Sken } 320229997Sken mtx_unlock(&softc->lock); 321229997Sken 322229997Sken if (beio != NULL) { 323229997Sken bzero(beio, sizeof(*beio)); 324229997Sken beio->softc = softc; 325229997Sken break; 326229997Sken } 327229997Sken } 328229997Sken return (beio); 329229997Sken} 330229997Sken 331229997Skenstatic void 332229997Skenctl_free_beio(struct ctl_be_block_io *beio) 333229997Sken{ 334229997Sken struct ctl_be_block_softc *softc; 335229997Sken int duplicate_free; 336229997Sken int i; 337229997Sken 338229997Sken softc = beio->softc; 339229997Sken duplicate_free = 0; 340229997Sken 341229997Sken for (i = 0; i < beio->num_segs; i++) { 342229997Sken if (beio->sg_segs[i].addr == NULL) 343229997Sken duplicate_free++; 344229997Sken 345229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 346229997Sken beio->sg_segs[i].addr = NULL; 347229997Sken } 348229997Sken 349229997Sken if (duplicate_free > 0) { 350229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 351229997Sken duplicate_free, beio->num_segs); 352229997Sken } 353229997Sken mtx_lock(&softc->lock); 354229997Sken STAILQ_INSERT_TAIL(&softc->beio_free_queue, beio, links); 355229997Sken mtx_unlock(&softc->lock); 356229997Sken} 357229997Sken 358229997Skenstatic int 359229997Skenctl_grow_beio(struct ctl_be_block_softc *softc, int count) 360229997Sken{ 361229997Sken int i; 362229997Sken 363229997Sken for (i = 0; i < count; i++) { 364229997Sken struct ctl_be_block_io *beio; 365229997Sken 366229997Sken beio = (struct ctl_be_block_io *)malloc(sizeof(*beio), 367229997Sken M_CTLBLK, 368229997Sken M_WAITOK | M_ZERO); 369229997Sken if (beio == NULL) 370229997Sken break; 371229997Sken 372229997Sken bzero(beio, sizeof(*beio)); 373229997Sken beio->softc = softc; 374229997Sken mtx_lock(&softc->lock); 375229997Sken STAILQ_INSERT_TAIL(&softc->beio_free_queue, beio, links); 376229997Sken mtx_unlock(&softc->lock); 377229997Sken } 378229997Sken 379229997Sken return (i); 380229997Sken} 381229997Sken 382229997Sken#if 0 383229997Skenstatic void 384229997Skenctl_shrink_beio(struct ctl_be_block_softc *softc) 385229997Sken{ 386229997Sken struct ctl_be_block_io *beio, *beio_tmp; 387229997Sken 388229997Sken mtx_lock(&softc->lock); 389229997Sken STAILQ_FOREACH_SAFE(beio, &softc->beio_free_queue, links, beio_tmp) { 390229997Sken STAILQ_REMOVE(&softc->beio_free_queue, beio, 391229997Sken ctl_be_block_io, links); 392229997Sken free(beio, M_CTLBLK); 393229997Sken } 394229997Sken mtx_unlock(&softc->lock); 395229997Sken} 396229997Sken#endif 397229997Sken 398229997Skenstatic void 399229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 400229997Sken{ 401229997Sken union ctl_io *io; 402229997Sken int io_len; 403229997Sken 404229997Sken io = beio->io; 405229997Sken 406229997Sken if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) 407229997Sken io_len = beio->io_len; 408229997Sken else 409229997Sken io_len = 0; 410229997Sken 411229997Sken devstat_end_transaction(beio->lun->disk_stats, 412229997Sken /*bytes*/ io_len, 413229997Sken beio->ds_tag_type, 414229997Sken beio->ds_trans_type, 415229997Sken /*now*/ NULL, 416229997Sken /*then*/&beio->ds_t0); 417229997Sken 418229997Sken ctl_free_beio(beio); 419229997Sken ctl_done(io); 420229997Sken} 421229997Sken 422229997Skenstatic int 423229997Skenctl_be_block_move_done(union ctl_io *io) 424229997Sken{ 425229997Sken struct ctl_be_block_io *beio; 426229997Sken struct ctl_be_block_lun *be_lun; 427229997Sken#ifdef CTL_TIME_IO 428229997Sken struct bintime cur_bt; 429229997Sken#endif 430229997Sken 431229997Sken beio = (struct ctl_be_block_io *) 432229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 433229997Sken 434229997Sken be_lun = beio->lun; 435229997Sken 436229997Sken DPRINTF("entered\n"); 437229997Sken 438229997Sken#ifdef CTL_TIME_IO 439229997Sken getbintime(&cur_bt); 440229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 441229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 442229997Sken io->io_hdr.num_dmas++; 443229997Sken#endif 444229997Sken 445229997Sken /* 446229997Sken * We set status at this point for read commands, and write 447229997Sken * commands with errors. 448229997Sken */ 449229997Sken if ((beio->bio_cmd == BIO_READ) 450229997Sken && (io->io_hdr.port_status == 0) 451229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 452229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) 453229997Sken ctl_set_success(&io->scsiio); 454229997Sken else if ((io->io_hdr.port_status != 0) 455229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 456229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 457229997Sken /* 458229997Sken * For hardware error sense keys, the sense key 459229997Sken * specific value is defined to be a retry count, 460229997Sken * but we use it to pass back an internal FETD 461229997Sken * error code. XXX KDM Hopefully the FETD is only 462229997Sken * using 16 bits for an error code, since that's 463229997Sken * all the space we have in the sks field. 464229997Sken */ 465229997Sken ctl_set_internal_failure(&io->scsiio, 466229997Sken /*sks_valid*/ 1, 467229997Sken /*retry_count*/ 468229997Sken io->io_hdr.port_status); 469229997Sken } 470229997Sken 471229997Sken /* 472229997Sken * If this is a read, or a write with errors, it is done. 473229997Sken */ 474229997Sken if ((beio->bio_cmd == BIO_READ) 475229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 476229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 477229997Sken ctl_complete_beio(beio); 478229997Sken return (0); 479229997Sken } 480229997Sken 481229997Sken /* 482229997Sken * At this point, we have a write and the DMA completed 483229997Sken * successfully. We now have to queue it to the task queue to 484229997Sken * execute the backend I/O. That is because we do blocking 485229997Sken * memory allocations, and in the file backing case, blocking I/O. 486229997Sken * This move done routine is generally called in the SIM's 487229997Sken * interrupt context, and therefore we cannot block. 488229997Sken */ 489229997Sken mtx_lock(&be_lun->lock); 490229997Sken /* 491229997Sken * XXX KDM make sure that links is okay to use at this point. 492229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 493229997Sken * or deal with resource allocation here. 494229997Sken */ 495229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 496229997Sken mtx_unlock(&be_lun->lock); 497229997Sken 498229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 499229997Sken 500229997Sken return (0); 501229997Sken} 502229997Sken 503229997Skenstatic void 504229997Skenctl_be_block_biodone(struct bio *bio) 505229997Sken{ 506229997Sken struct ctl_be_block_io *beio; 507229997Sken struct ctl_be_block_lun *be_lun; 508229997Sken union ctl_io *io; 509229997Sken 510229997Sken beio = bio->bio_caller1; 511229997Sken be_lun = beio->lun; 512229997Sken io = beio->io; 513229997Sken 514229997Sken DPRINTF("entered\n"); 515229997Sken 516229997Sken mtx_lock(&be_lun->lock); 517229997Sken if (bio->bio_error != 0) 518229997Sken beio->num_errors++; 519229997Sken 520229997Sken beio->num_bios_done++; 521229997Sken 522229997Sken /* 523229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 524229997Sken * during the free might cause it to complain. 525229997Sken */ 526229997Sken g_destroy_bio(bio); 527229997Sken 528229997Sken /* 529229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 530229997Sken * complete, then we're done. 531229997Sken */ 532229997Sken if ((beio->send_complete == 0) 533229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 534229997Sken mtx_unlock(&be_lun->lock); 535229997Sken return; 536229997Sken } 537229997Sken 538229997Sken /* 539229997Sken * At this point, we've verified that we are the last I/O to 540229997Sken * complete, so it's safe to drop the lock. 541229997Sken */ 542229997Sken mtx_unlock(&be_lun->lock); 543229997Sken 544229997Sken /* 545229997Sken * If there are any errors from the backing device, we fail the 546229997Sken * entire I/O with a medium error. 547229997Sken */ 548229997Sken if (beio->num_errors > 0) { 549229997Sken if (beio->bio_cmd == BIO_FLUSH) { 550229997Sken /* XXX KDM is there is a better error here? */ 551229997Sken ctl_set_internal_failure(&io->scsiio, 552229997Sken /*sks_valid*/ 1, 553229997Sken /*retry_count*/ 0xbad2); 554229997Sken } else 555229997Sken ctl_set_medium_error(&io->scsiio); 556229997Sken ctl_complete_beio(beio); 557229997Sken return; 558229997Sken } 559229997Sken 560229997Sken /* 561229997Sken * If this is a write or a flush, we're all done. 562229997Sken * If this is a read, we can now send the data to the user. 563229997Sken */ 564229997Sken if ((beio->bio_cmd == BIO_WRITE) 565229997Sken || (beio->bio_cmd == BIO_FLUSH)) { 566229997Sken ctl_set_success(&io->scsiio); 567229997Sken ctl_complete_beio(beio); 568229997Sken } else { 569229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 570229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 571229997Sken io->scsiio.kern_data_len = beio->io_len; 572229997Sken io->scsiio.kern_total_len = beio->io_len; 573229997Sken io->scsiio.kern_rel_offset = 0; 574229997Sken io->scsiio.kern_data_resid = 0; 575229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 576229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 577229997Sken#ifdef CTL_TIME_IO 578229997Sken getbintime(&io->io_hdr.dma_start_bt); 579229997Sken#endif 580229997Sken ctl_datamove(io); 581229997Sken } 582229997Sken} 583229997Sken 584229997Skenstatic void 585229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 586229997Sken struct ctl_be_block_io *beio) 587229997Sken{ 588229997Sken union ctl_io *io; 589229997Sken struct mount *mountpoint; 590229997Sken int vfs_is_locked, error, lock_flags; 591229997Sken 592229997Sken DPRINTF("entered\n"); 593229997Sken 594229997Sken io = beio->io; 595229997Sken 596229997Sken vfs_is_locked = VFS_LOCK_GIANT(be_lun->vn->v_mount); 597229997Sken 598229997Sken (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 599229997Sken 600229997Sken if (MNT_SHARED_WRITES(mountpoint) 601229997Sken || ((mountpoint == NULL) 602229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 603229997Sken lock_flags = LK_SHARED; 604229997Sken else 605229997Sken lock_flags = LK_EXCLUSIVE; 606229997Sken 607229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 608229997Sken 609229997Sken binuptime(&beio->ds_t0); 610229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 611229997Sken 612229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 613229997Sken VOP_UNLOCK(be_lun->vn, 0); 614229997Sken 615229997Sken vn_finished_write(mountpoint); 616229997Sken 617229997Sken VFS_UNLOCK_GIANT(vfs_is_locked); 618229997Sken 619229997Sken if (error == 0) 620229997Sken ctl_set_success(&io->scsiio); 621229997Sken else { 622229997Sken /* XXX KDM is there is a better error here? */ 623229997Sken ctl_set_internal_failure(&io->scsiio, 624229997Sken /*sks_valid*/ 1, 625229997Sken /*retry_count*/ 0xbad1); 626229997Sken } 627229997Sken 628229997Sken ctl_complete_beio(beio); 629229997Sken} 630229997Sken 631229997SkenSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, file_start, "uint64_t"); 632229997SkenSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, file_start, "uint64_t"); 633229997SkenSDT_PROBE_DEFINE1(cbb, kernel, read, file_done, file_done,"uint64_t"); 634229997SkenSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, file_done, "uint64_t"); 635229997Sken 636229997Skenstatic void 637229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 638229997Sken struct ctl_be_block_io *beio) 639229997Sken{ 640229997Sken struct ctl_be_block_filedata *file_data; 641229997Sken union ctl_io *io; 642229997Sken struct uio xuio; 643229997Sken struct iovec *xiovec; 644229997Sken int vfs_is_locked, flags; 645229997Sken int error, i; 646229997Sken 647229997Sken DPRINTF("entered\n"); 648229997Sken 649229997Sken file_data = &be_lun->backend.file; 650229997Sken io = beio->io; 651229997Sken flags = beio->bio_flags; 652229997Sken 653229997Sken if (beio->bio_cmd == BIO_READ) { 654229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 655229997Sken } else { 656229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 657229997Sken } 658229997Sken 659229997Sken bzero(&xuio, sizeof(xuio)); 660229997Sken if (beio->bio_cmd == BIO_READ) 661229997Sken xuio.uio_rw = UIO_READ; 662229997Sken else 663229997Sken xuio.uio_rw = UIO_WRITE; 664229997Sken 665229997Sken xuio.uio_offset = beio->io_offset; 666229997Sken xuio.uio_resid = beio->io_len; 667229997Sken xuio.uio_segflg = UIO_SYSSPACE; 668229997Sken xuio.uio_iov = beio->xiovecs; 669229997Sken xuio.uio_iovcnt = beio->num_segs; 670229997Sken xuio.uio_td = curthread; 671229997Sken 672229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 673229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 674229997Sken xiovec->iov_len = beio->sg_segs[i].len; 675229997Sken } 676229997Sken 677229997Sken vfs_is_locked = VFS_LOCK_GIANT(be_lun->vn->v_mount); 678229997Sken if (beio->bio_cmd == BIO_READ) { 679229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 680229997Sken 681229997Sken binuptime(&beio->ds_t0); 682229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 683229997Sken 684229997Sken /* 685229997Sken * UFS pays attention to IO_DIRECT for reads. If the 686229997Sken * DIRECTIO option is configured into the kernel, it calls 687229997Sken * ffs_rawread(). But that only works for single-segment 688229997Sken * uios with user space addresses. In our case, with a 689229997Sken * kernel uio, it still reads into the buffer cache, but it 690229997Sken * will just try to release the buffer from the cache later 691229997Sken * on in ffs_read(). 692229997Sken * 693229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 694229997Sken * 695229997Sken * UFS does not pay attention to IO_SYNC for reads. 696229997Sken * 697229997Sken * ZFS pays attention to IO_SYNC (which translates into the 698229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 699229997Sken * attempts to sync the file before reading. 700229997Sken * 701229997Sken * So, to attempt to provide some barrier semantics in the 702229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 703229997Sken */ 704229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 705229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 706229997Sken 707229997Sken VOP_UNLOCK(be_lun->vn, 0); 708229997Sken } else { 709229997Sken struct mount *mountpoint; 710229997Sken int lock_flags; 711229997Sken 712229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 713229997Sken 714229997Sken if (MNT_SHARED_WRITES(mountpoint) 715229997Sken || ((mountpoint == NULL) 716229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 717229997Sken lock_flags = LK_SHARED; 718229997Sken else 719229997Sken lock_flags = LK_EXCLUSIVE; 720229997Sken 721229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 722229997Sken 723229997Sken binuptime(&beio->ds_t0); 724229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 725229997Sken 726229997Sken /* 727229997Sken * UFS pays attention to IO_DIRECT for writes. The write 728229997Sken * is done asynchronously. (Normally the write would just 729229997Sken * get put into cache. 730229997Sken * 731229997Sken * UFS pays attention to IO_SYNC for writes. It will 732229997Sken * attempt to write the buffer out synchronously if that 733229997Sken * flag is set. 734229997Sken * 735229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 736229997Sken * 737229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 738229997Sken * for writes. It will flush the transaction from the 739229997Sken * cache before returning. 740229997Sken * 741229997Sken * So if we've got the BIO_ORDERED flag set, we want 742229997Sken * IO_SYNC in either the UFS or ZFS case. 743229997Sken */ 744229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 745229997Sken IO_SYNC : 0, file_data->cred); 746229997Sken VOP_UNLOCK(be_lun->vn, 0); 747229997Sken 748229997Sken vn_finished_write(mountpoint); 749229997Sken } 750229997Sken VFS_UNLOCK_GIANT(vfs_is_locked); 751229997Sken 752229997Sken /* 753229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 754229997Sken * return the I/O to the user. 755229997Sken */ 756229997Sken if (error != 0) { 757229997Sken char path_str[32]; 758229997Sken 759229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 760229997Sken /* 761229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 762229997Sken * filesystem fills up. What kind of SCSI error should we 763229997Sken * return for that? 764229997Sken */ 765229997Sken printf("%s%s command returned errno %d\n", path_str, 766229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 767229997Sken ctl_set_medium_error(&io->scsiio); 768229997Sken ctl_complete_beio(beio); 769229997Sken return; 770229997Sken } 771229997Sken 772229997Sken /* 773229997Sken * If this is a write, we're all done. 774229997Sken * If this is a read, we can now send the data to the user. 775229997Sken */ 776229997Sken if (beio->bio_cmd == BIO_WRITE) { 777229997Sken ctl_set_success(&io->scsiio); 778229997Sken SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 779229997Sken ctl_complete_beio(beio); 780229997Sken } else { 781229997Sken SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 782229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 783229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 784229997Sken io->scsiio.kern_data_len = beio->io_len; 785229997Sken io->scsiio.kern_total_len = beio->io_len; 786229997Sken io->scsiio.kern_rel_offset = 0; 787229997Sken io->scsiio.kern_data_resid = 0; 788229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 789229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 790229997Sken#ifdef CTL_TIME_IO 791229997Sken getbintime(&io->io_hdr.dma_start_bt); 792229997Sken#endif 793229997Sken ctl_datamove(io); 794229997Sken } 795229997Sken} 796229997Sken 797229997Skenstatic void 798229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 799229997Sken struct ctl_be_block_io *beio) 800229997Sken{ 801229997Sken struct bio *bio; 802229997Sken union ctl_io *io; 803229997Sken struct ctl_be_block_devdata *dev_data; 804229997Sken 805229997Sken dev_data = &be_lun->backend.dev; 806229997Sken io = beio->io; 807229997Sken 808229997Sken DPRINTF("entered\n"); 809229997Sken 810229997Sken /* This can't fail, it's a blocking allocation. */ 811229997Sken bio = g_alloc_bio(); 812229997Sken 813229997Sken bio->bio_cmd = BIO_FLUSH; 814229997Sken bio->bio_flags |= BIO_ORDERED; 815229997Sken bio->bio_dev = dev_data->cdev; 816229997Sken bio->bio_offset = 0; 817229997Sken bio->bio_data = 0; 818229997Sken bio->bio_done = ctl_be_block_biodone; 819229997Sken bio->bio_caller1 = beio; 820229997Sken bio->bio_pblkno = 0; 821229997Sken 822229997Sken /* 823229997Sken * We don't need to acquire the LUN lock here, because we are only 824229997Sken * sending one bio, and so there is no other context to synchronize 825229997Sken * with. 826229997Sken */ 827229997Sken beio->num_bios_sent = 1; 828229997Sken beio->send_complete = 1; 829229997Sken 830229997Sken binuptime(&beio->ds_t0); 831229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 832229997Sken 833229997Sken (*dev_data->csw->d_strategy)(bio); 834229997Sken} 835229997Sken 836229997Skenstatic void 837229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 838229997Sken struct ctl_be_block_io *beio) 839229997Sken{ 840229997Sken int i; 841229997Sken struct bio *bio; 842229997Sken struct ctl_be_block_devdata *dev_data; 843229997Sken off_t cur_offset; 844229997Sken int max_iosize; 845229997Sken 846229997Sken DPRINTF("entered\n"); 847229997Sken 848229997Sken dev_data = &be_lun->backend.dev; 849229997Sken 850229997Sken /* 851229997Sken * We have to limit our I/O size to the maximum supported by the 852229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 853229997Sken * set it properly, use DFLTPHYS. 854229997Sken */ 855229997Sken max_iosize = dev_data->cdev->si_iosize_max; 856229997Sken if (max_iosize < PAGE_SIZE) 857229997Sken max_iosize = DFLTPHYS; 858229997Sken 859229997Sken cur_offset = beio->io_offset; 860229997Sken 861229997Sken /* 862229997Sken * XXX KDM need to accurately reflect the number of I/Os outstanding 863229997Sken * to a device. 864229997Sken */ 865229997Sken binuptime(&beio->ds_t0); 866229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 867229997Sken 868229997Sken for (i = 0; i < beio->num_segs; i++) { 869229997Sken size_t cur_size; 870229997Sken uint8_t *cur_ptr; 871229997Sken 872229997Sken cur_size = beio->sg_segs[i].len; 873229997Sken cur_ptr = beio->sg_segs[i].addr; 874229997Sken 875229997Sken while (cur_size > 0) { 876229997Sken /* This can't fail, it's a blocking allocation. */ 877229997Sken bio = g_alloc_bio(); 878229997Sken 879229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 880229997Sken 881229997Sken bio->bio_cmd = beio->bio_cmd; 882229997Sken bio->bio_flags |= beio->bio_flags; 883229997Sken bio->bio_dev = dev_data->cdev; 884229997Sken bio->bio_caller1 = beio; 885229997Sken bio->bio_length = min(cur_size, max_iosize); 886229997Sken bio->bio_offset = cur_offset; 887229997Sken bio->bio_data = cur_ptr; 888229997Sken bio->bio_done = ctl_be_block_biodone; 889229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 890229997Sken 891229997Sken cur_offset += bio->bio_length; 892229997Sken cur_ptr += bio->bio_length; 893229997Sken cur_size -= bio->bio_length; 894229997Sken 895229997Sken /* 896229997Sken * Make sure we set the complete bit just before we 897229997Sken * issue the last bio so we don't wind up with a 898229997Sken * race. 899229997Sken * 900229997Sken * Use the LUN mutex here instead of a combination 901229997Sken * of atomic variables for simplicity. 902229997Sken * 903229997Sken * XXX KDM we could have a per-IO lock, but that 904229997Sken * would cause additional per-IO setup and teardown 905229997Sken * overhead. Hopefully there won't be too much 906229997Sken * contention on the LUN lock. 907229997Sken */ 908229997Sken mtx_lock(&be_lun->lock); 909229997Sken 910229997Sken beio->num_bios_sent++; 911229997Sken 912229997Sken if ((i == beio->num_segs - 1) 913229997Sken && (cur_size == 0)) 914229997Sken beio->send_complete = 1; 915229997Sken 916229997Sken mtx_unlock(&be_lun->lock); 917229997Sken 918229997Sken (*dev_data->csw->d_strategy)(bio); 919229997Sken } 920229997Sken } 921229997Sken} 922229997Sken 923229997Skenstatic void 924229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 925229997Sken union ctl_io *io) 926229997Sken{ 927229997Sken struct ctl_be_block_io *beio; 928229997Sken struct ctl_be_block_softc *softc; 929229997Sken 930229997Sken DPRINTF("entered\n"); 931229997Sken 932229997Sken softc = be_lun->softc; 933229997Sken beio = ctl_alloc_beio(softc); 934229997Sken if (beio == NULL) { 935229997Sken /* 936229997Sken * This should not happen. ctl_alloc_beio() will call 937229997Sken * ctl_grow_beio() with a blocking malloc as needed. 938229997Sken * A malloc with M_WAITOK should not fail. 939229997Sken */ 940229997Sken ctl_set_busy(&io->scsiio); 941229997Sken ctl_done(io); 942229997Sken return; 943229997Sken } 944229997Sken 945229997Sken beio->io = io; 946229997Sken beio->softc = softc; 947229997Sken beio->lun = be_lun; 948229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 949229997Sken 950229997Sken switch (io->scsiio.cdb[0]) { 951229997Sken case SYNCHRONIZE_CACHE: 952229997Sken case SYNCHRONIZE_CACHE_16: 953229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 954229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 955229997Sken beio->io_len = 0; 956229997Sken be_lun->lun_flush(be_lun, beio); 957229997Sken break; 958229997Sken default: 959229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 960229997Sken break; 961229997Sken } 962229997Sken} 963229997Sken 964229997SkenSDT_PROBE_DEFINE1(cbb, kernel, read, start, start, "uint64_t"); 965229997SkenSDT_PROBE_DEFINE1(cbb, kernel, write, start, start, "uint64_t"); 966229997SkenSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, alloc_done, "uint64_t"); 967229997SkenSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, alloc_done, "uint64_t"); 968229997Sken 969229997Skenstatic void 970229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 971229997Sken union ctl_io *io) 972229997Sken{ 973229997Sken struct ctl_be_block_io *beio; 974229997Sken struct ctl_be_block_softc *softc; 975229997Sken struct ctl_lba_len lbalen; 976229997Sken uint64_t len_left, io_size_bytes; 977229997Sken int i; 978229997Sken 979229997Sken softc = be_lun->softc; 980229997Sken 981229997Sken DPRINTF("entered\n"); 982229997Sken 983229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 984229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 985229997Sken } else { 986229997Sken SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 987229997Sken } 988229997Sken 989229997Sken memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 990229997Sken sizeof(lbalen)); 991229997Sken 992229997Sken io_size_bytes = lbalen.len * be_lun->blocksize; 993229997Sken 994229997Sken /* 995229997Sken * XXX KDM this is temporary, until we implement chaining of beio 996229997Sken * structures and multiple datamove calls to move all the data in 997229997Sken * or out. 998229997Sken */ 999229997Sken if (io_size_bytes > CTLBLK_MAX_IO_SIZE) { 1000229997Sken printf("%s: IO length %ju > max io size %u\n", __func__, 1001229997Sken io_size_bytes, CTLBLK_MAX_IO_SIZE); 1002229997Sken ctl_set_invalid_field(&io->scsiio, 1003229997Sken /*sks_valid*/ 0, 1004229997Sken /*command*/ 1, 1005229997Sken /*field*/ 0, 1006229997Sken /*bit_valid*/ 0, 1007229997Sken /*bit*/ 0); 1008229997Sken ctl_done(io); 1009229997Sken return; 1010229997Sken } 1011229997Sken 1012229997Sken beio = ctl_alloc_beio(softc); 1013229997Sken if (beio == NULL) { 1014229997Sken /* 1015229997Sken * This should not happen. ctl_alloc_beio() will call 1016229997Sken * ctl_grow_beio() with a blocking malloc as needed. 1017229997Sken * A malloc with M_WAITOK should not fail. 1018229997Sken */ 1019229997Sken ctl_set_busy(&io->scsiio); 1020229997Sken ctl_done(io); 1021229997Sken return; 1022229997Sken } 1023229997Sken 1024229997Sken beio->io = io; 1025229997Sken beio->softc = softc; 1026229997Sken beio->lun = be_lun; 1027229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 1028229997Sken 1029229997Sken /* 1030229997Sken * If the I/O came down with an ordered or head of queue tag, set 1031229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 1032229997Sken * pretty much the best we can do. 1033229997Sken * 1034229997Sken * XXX KDM we don't have a great way to easily know about the FUA 1035229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 1036229997Sken * pass that knowledge to the backend), and in any case we would 1037229997Sken * need to determine how to handle it. 1038229997Sken */ 1039229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1040229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1041229997Sken beio->bio_flags = BIO_ORDERED; 1042229997Sken 1043229997Sken switch (io->scsiio.tag_type) { 1044229997Sken case CTL_TAG_ORDERED: 1045229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1046229997Sken break; 1047229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1048229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1049229997Sken break; 1050229997Sken case CTL_TAG_UNTAGGED: 1051229997Sken case CTL_TAG_SIMPLE: 1052229997Sken case CTL_TAG_ACA: 1053229997Sken default: 1054229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1055229997Sken break; 1056229997Sken } 1057229997Sken 1058229997Sken /* 1059229997Sken * This path handles read and write only. The config write path 1060229997Sken * handles flush operations. 1061229997Sken */ 1062229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 1063229997Sken beio->bio_cmd = BIO_READ; 1064229997Sken beio->ds_trans_type = DEVSTAT_READ; 1065229997Sken } else { 1066229997Sken beio->bio_cmd = BIO_WRITE; 1067229997Sken beio->ds_trans_type = DEVSTAT_WRITE; 1068229997Sken } 1069229997Sken 1070229997Sken beio->io_len = lbalen.len * be_lun->blocksize; 1071229997Sken beio->io_offset = lbalen.lba * be_lun->blocksize; 1072229997Sken 1073229997Sken DPRINTF("%s at LBA %jx len %u\n", 1074229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1075229997Sken (uintmax_t)lbalen.lba, lbalen.len); 1076229997Sken 1077229997Sken for (i = 0, len_left = io_size_bytes; i < CTLBLK_MAX_SEGS && 1078229997Sken len_left > 0; i++) { 1079229997Sken 1080229997Sken /* 1081229997Sken * Setup the S/G entry for this chunk. 1082229997Sken */ 1083229997Sken beio->sg_segs[i].len = min(MAXPHYS, len_left); 1084229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1085229997Sken /* 1086229997Sken * uma_zalloc() can in theory return NULL even with M_WAITOK 1087229997Sken * if it can't pull more memory into the zone. 1088229997Sken */ 1089229997Sken if (beio->sg_segs[i].addr == NULL) { 1090229997Sken ctl_set_busy(&io->scsiio); 1091229997Sken ctl_complete_beio(beio); 1092229997Sken return; 1093229997Sken } 1094229997Sken 1095229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1096229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1097229997Sken 1098229997Sken beio->num_segs++; 1099229997Sken len_left -= beio->sg_segs[i].len; 1100229997Sken } 1101229997Sken 1102229997Sken /* 1103229997Sken * For the read case, we need to read the data into our buffers and 1104229997Sken * then we can send it back to the user. For the write case, we 1105229997Sken * need to get the data from the user first. 1106229997Sken */ 1107229997Sken if (beio->bio_cmd == BIO_READ) { 1108229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1109229997Sken be_lun->dispatch(be_lun, beio); 1110229997Sken } else { 1111229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1112229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 1113229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1114229997Sken io->scsiio.kern_data_len = beio->io_len; 1115229997Sken io->scsiio.kern_total_len = beio->io_len; 1116229997Sken io->scsiio.kern_rel_offset = 0; 1117229997Sken io->scsiio.kern_data_resid = 0; 1118229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 1119229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1120229997Sken#ifdef CTL_TIME_IO 1121229997Sken getbintime(&io->io_hdr.dma_start_bt); 1122229997Sken#endif 1123229997Sken ctl_datamove(io); 1124229997Sken } 1125229997Sken} 1126229997Sken 1127229997Skenstatic void 1128229997Skenctl_be_block_worker(void *context, int pending) 1129229997Sken{ 1130229997Sken struct ctl_be_block_lun *be_lun; 1131229997Sken struct ctl_be_block_softc *softc; 1132229997Sken union ctl_io *io; 1133229997Sken 1134229997Sken be_lun = (struct ctl_be_block_lun *)context; 1135229997Sken softc = be_lun->softc; 1136229997Sken 1137229997Sken DPRINTF("entered\n"); 1138229997Sken 1139229997Sken mtx_lock(&be_lun->lock); 1140229997Sken for (;;) { 1141229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1142229997Sken if (io != NULL) { 1143229997Sken struct ctl_be_block_io *beio; 1144229997Sken 1145229997Sken DPRINTF("datamove queue\n"); 1146229997Sken 1147229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1148229997Sken ctl_io_hdr, links); 1149229997Sken 1150229997Sken mtx_unlock(&be_lun->lock); 1151229997Sken 1152229997Sken beio = (struct ctl_be_block_io *) 1153229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1154229997Sken 1155229997Sken be_lun->dispatch(be_lun, beio); 1156229997Sken 1157229997Sken mtx_lock(&be_lun->lock); 1158229997Sken continue; 1159229997Sken } 1160229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1161229997Sken if (io != NULL) { 1162229997Sken 1163229997Sken DPRINTF("config write queue\n"); 1164229997Sken 1165229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1166229997Sken ctl_io_hdr, links); 1167229997Sken 1168229997Sken mtx_unlock(&be_lun->lock); 1169229997Sken 1170229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1171229997Sken 1172229997Sken mtx_lock(&be_lun->lock); 1173229997Sken continue; 1174229997Sken } 1175229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1176229997Sken if (io != NULL) { 1177229997Sken DPRINTF("input queue\n"); 1178229997Sken 1179229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1180229997Sken ctl_io_hdr, links); 1181229997Sken mtx_unlock(&be_lun->lock); 1182229997Sken 1183229997Sken /* 1184229997Sken * We must drop the lock, since this routine and 1185229997Sken * its children may sleep. 1186229997Sken */ 1187229997Sken ctl_be_block_dispatch(be_lun, io); 1188229997Sken 1189229997Sken mtx_lock(&be_lun->lock); 1190229997Sken continue; 1191229997Sken } 1192229997Sken 1193229997Sken /* 1194229997Sken * If we get here, there is no work left in the queues, so 1195229997Sken * just break out and let the task queue go to sleep. 1196229997Sken */ 1197229997Sken break; 1198229997Sken } 1199229997Sken mtx_unlock(&be_lun->lock); 1200229997Sken} 1201229997Sken 1202229997Sken/* 1203229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1204229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1205229997Sken * thread. 1206229997Sken */ 1207229997Skenstatic int 1208229997Skenctl_be_block_submit(union ctl_io *io) 1209229997Sken{ 1210229997Sken struct ctl_be_block_lun *be_lun; 1211229997Sken struct ctl_be_lun *ctl_be_lun; 1212229997Sken int retval; 1213229997Sken 1214229997Sken DPRINTF("entered\n"); 1215229997Sken 1216229997Sken retval = CTL_RETVAL_COMPLETE; 1217229997Sken 1218229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1219229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1220229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1221229997Sken 1222229997Sken /* 1223229997Sken * Make sure we only get SCSI I/O. 1224229997Sken */ 1225229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1226229997Sken "%#x) encountered", io->io_hdr.io_type)); 1227229997Sken 1228229997Sken mtx_lock(&be_lun->lock); 1229229997Sken /* 1230229997Sken * XXX KDM make sure that links is okay to use at this point. 1231229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1232229997Sken * or deal with resource allocation here. 1233229997Sken */ 1234229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1235229997Sken mtx_unlock(&be_lun->lock); 1236229997Sken 1237229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1238229997Sken 1239229997Sken return (retval); 1240229997Sken} 1241229997Sken 1242229997Skenstatic int 1243229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1244229997Sken int flag, struct thread *td) 1245229997Sken{ 1246229997Sken struct ctl_be_block_softc *softc; 1247229997Sken int error; 1248229997Sken 1249229997Sken softc = &backend_block_softc; 1250229997Sken 1251229997Sken error = 0; 1252229997Sken 1253229997Sken switch (cmd) { 1254229997Sken case CTL_LUN_REQ: { 1255229997Sken struct ctl_lun_req *lun_req; 1256229997Sken 1257229997Sken lun_req = (struct ctl_lun_req *)addr; 1258229997Sken 1259229997Sken switch (lun_req->reqtype) { 1260229997Sken case CTL_LUNREQ_CREATE: 1261229997Sken error = ctl_be_block_create(softc, lun_req); 1262229997Sken break; 1263229997Sken case CTL_LUNREQ_RM: 1264229997Sken error = ctl_be_block_rm(softc, lun_req); 1265229997Sken break; 1266229997Sken default: 1267229997Sken lun_req->status = CTL_LUN_ERROR; 1268229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1269229997Sken "%s: invalid LUN request type %d", __func__, 1270229997Sken lun_req->reqtype); 1271229997Sken break; 1272229997Sken } 1273229997Sken break; 1274229997Sken } 1275229997Sken default: 1276229997Sken error = ENOTTY; 1277229997Sken break; 1278229997Sken } 1279229997Sken 1280229997Sken return (error); 1281229997Sken} 1282229997Sken 1283229997Skenstatic int 1284229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1285229997Sken{ 1286229997Sken struct ctl_be_block_filedata *file_data; 1287229997Sken struct ctl_lun_create_params *params; 1288229997Sken struct vattr vattr; 1289229997Sken int error; 1290229997Sken 1291229997Sken error = 0; 1292229997Sken file_data = &be_lun->backend.file; 1293229997Sken params = &req->reqdata.create; 1294229997Sken 1295229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1296229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1297229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1298229997Sken 1299229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1300229997Sken if (error != 0) { 1301229997Sken snprintf(req->error_str, sizeof(req->error_str), 1302229997Sken "error calling VOP_GETATTR() for file %s", 1303229997Sken be_lun->dev_path); 1304229997Sken return (error); 1305229997Sken } 1306229997Sken 1307229997Sken /* 1308229997Sken * Verify that we have the ability to upgrade to exclusive 1309229997Sken * access on this file so we can trap errors at open instead 1310229997Sken * of reporting them during first access. 1311229997Sken */ 1312229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1313229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1314229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1315229997Sken error = EBADF; 1316229997Sken snprintf(req->error_str, sizeof(req->error_str), 1317229997Sken "error locking file %s", be_lun->dev_path); 1318229997Sken return (error); 1319229997Sken } 1320229997Sken } 1321229997Sken 1322229997Sken 1323229997Sken file_data->cred = crhold(curthread->td_ucred); 1324229997Sken be_lun->size_bytes = vattr.va_size; 1325229997Sken /* 1326229997Sken * We set the multi thread flag for file operations because all 1327229997Sken * filesystems (in theory) are capable of allowing multiple readers 1328229997Sken * of a file at once. So we want to get the maximum possible 1329229997Sken * concurrency. 1330229997Sken */ 1331229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1332229997Sken 1333229997Sken /* 1334229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1335229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1336229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1337229997Sken * may not work with other OSes as well. So just export a sector 1338229997Sken * size of 512 bytes, which should work with any OS or 1339229997Sken * application. Since our backing is a file, any block size will 1340229997Sken * work fine for the backing store. 1341229997Sken */ 1342229997Sken#if 0 1343229997Sken be_lun->blocksize= vattr.va_blocksize; 1344229997Sken#endif 1345229997Sken if (params->blocksize_bytes != 0) 1346229997Sken be_lun->blocksize = params->blocksize_bytes; 1347229997Sken else 1348229997Sken be_lun->blocksize = 512; 1349229997Sken 1350229997Sken /* 1351229997Sken * Sanity check. The media size has to be at least one 1352229997Sken * sector long. 1353229997Sken */ 1354229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1355229997Sken error = EINVAL; 1356229997Sken snprintf(req->error_str, sizeof(req->error_str), 1357229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1358229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1359229997Sken } 1360229997Sken return (error); 1361229997Sken} 1362229997Sken 1363229997Skenstatic int 1364229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1365229997Sken{ 1366229997Sken struct ctl_lun_create_params *params; 1367229997Sken struct vattr vattr; 1368229997Sken struct cdev *dev; 1369229997Sken struct cdevsw *devsw; 1370229997Sken int error; 1371229997Sken 1372229997Sken params = &req->reqdata.create; 1373229997Sken 1374229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1375229997Sken be_lun->dispatch = ctl_be_block_dispatch_dev; 1376229997Sken be_lun->lun_flush = ctl_be_block_flush_dev; 1377229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1378229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1379229997Sken &be_lun->backend.dev.dev_ref); 1380229997Sken if (be_lun->backend.dev.csw == NULL) 1381229997Sken panic("Unable to retrieve device switch"); 1382229997Sken 1383229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1384229997Sken if (error) { 1385229997Sken snprintf(req->error_str, sizeof(req->error_str), 1386229997Sken "%s: error getting vnode attributes for device %s", 1387229997Sken __func__, be_lun->dev_path); 1388229997Sken return (error); 1389229997Sken } 1390229997Sken 1391229997Sken dev = be_lun->vn->v_rdev; 1392229997Sken devsw = dev->si_devsw; 1393229997Sken if (!devsw->d_ioctl) { 1394229997Sken snprintf(req->error_str, sizeof(req->error_str), 1395229997Sken "%s: no d_ioctl for device %s!", __func__, 1396229997Sken be_lun->dev_path); 1397229997Sken return (ENODEV); 1398229997Sken } 1399229997Sken 1400229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1401229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1402229997Sken curthread); 1403229997Sken if (error) { 1404229997Sken snprintf(req->error_str, sizeof(req->error_str), 1405229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1406229997Sken "on %s!", __func__, error, be_lun->dev_path); 1407229997Sken return (error); 1408229997Sken } 1409229997Sken 1410229997Sken /* 1411229997Sken * If the user has asked for a blocksize that is greater than the 1412229997Sken * backing device's blocksize, we can do it only if the blocksize 1413229997Sken * the user is asking for is an even multiple of the underlying 1414229997Sken * device's blocksize. 1415229997Sken */ 1416229997Sken if ((params->blocksize_bytes != 0) 1417229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1418229997Sken uint32_t bs_multiple, tmp_blocksize; 1419229997Sken 1420229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1421229997Sken 1422229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1423229997Sken 1424229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1425229997Sken be_lun->blocksize = params->blocksize_bytes; 1426229997Sken } else { 1427229997Sken snprintf(req->error_str, sizeof(req->error_str), 1428229997Sken "%s: requested blocksize %u is not an even " 1429229997Sken "multiple of backing device blocksize %u", 1430229997Sken __func__, params->blocksize_bytes, 1431229997Sken be_lun->blocksize); 1432229997Sken return (EINVAL); 1433229997Sken 1434229997Sken } 1435229997Sken } else if ((params->blocksize_bytes != 0) 1436229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1437229997Sken snprintf(req->error_str, sizeof(req->error_str), 1438229997Sken "%s: requested blocksize %u < backing device " 1439229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1440229997Sken be_lun->blocksize); 1441229997Sken return (EINVAL); 1442229997Sken } 1443229997Sken 1444229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1445229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1446229997Sken curthread); 1447229997Sken if (error) { 1448229997Sken snprintf(req->error_str, sizeof(req->error_str), 1449229997Sken "%s: error %d returned for DIOCGMEDIASIZE ioctl " 1450229997Sken "on %s!", __func__, error, be_lun->dev_path); 1451229997Sken return (error); 1452229997Sken } 1453229997Sken 1454229997Sken return (0); 1455229997Sken 1456229997Sken} 1457229997Sken 1458229997Sken 1459229997Skenstatic int 1460229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1461229997Sken{ 1462229997Sken DROP_GIANT(); 1463229997Sken if (be_lun->vn) { 1464229997Sken int flags = FREAD | FWRITE; 1465229997Sken int vfs_is_locked = 0; 1466229997Sken 1467229997Sken switch (be_lun->dev_type) { 1468229997Sken case CTL_BE_BLOCK_DEV: 1469229997Sken if (be_lun->backend.dev.csw) { 1470229997Sken dev_relthread(be_lun->backend.dev.cdev, 1471229997Sken be_lun->backend.dev.dev_ref); 1472229997Sken be_lun->backend.dev.csw = NULL; 1473229997Sken be_lun->backend.dev.cdev = NULL; 1474229997Sken } 1475229997Sken break; 1476229997Sken case CTL_BE_BLOCK_FILE: 1477229997Sken vfs_is_locked = VFS_LOCK_GIANT(be_lun->vn->v_mount); 1478229997Sken break; 1479229997Sken case CTL_BE_BLOCK_NONE: 1480229997Sken default: 1481229997Sken panic("Unexpected backend type."); 1482229997Sken break; 1483229997Sken } 1484229997Sken 1485229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1486229997Sken be_lun->vn = NULL; 1487229997Sken 1488229997Sken switch (be_lun->dev_type) { 1489229997Sken case CTL_BE_BLOCK_DEV: 1490229997Sken break; 1491229997Sken case CTL_BE_BLOCK_FILE: 1492229997Sken VFS_UNLOCK_GIANT(vfs_is_locked); 1493229997Sken if (be_lun->backend.file.cred != NULL) { 1494229997Sken crfree(be_lun->backend.file.cred); 1495229997Sken be_lun->backend.file.cred = NULL; 1496229997Sken } 1497229997Sken break; 1498229997Sken case CTL_BE_BLOCK_NONE: 1499229997Sken default: 1500229997Sken panic("Unexpected backend type."); 1501229997Sken break; 1502229997Sken } 1503229997Sken } 1504229997Sken PICKUP_GIANT(); 1505229997Sken 1506229997Sken return (0); 1507229997Sken} 1508229997Sken 1509229997Skenstatic int 1510229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1511229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1512229997Sken{ 1513229997Sken struct nameidata nd; 1514229997Sken int flags; 1515229997Sken int error; 1516229997Sken int vfs_is_locked; 1517229997Sken 1518229997Sken /* 1519229997Sken * XXX KDM allow a read-only option? 1520229997Sken */ 1521229997Sken flags = FREAD | FWRITE; 1522229997Sken error = 0; 1523229997Sken 1524229997Sken if (rootvnode == NULL) { 1525229997Sken snprintf(req->error_str, sizeof(req->error_str), 1526229997Sken "%s: Root filesystem is not mounted", __func__); 1527229997Sken return (1); 1528229997Sken } 1529229997Sken 1530229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1531229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1532229997Sken VREF(rootvnode); 1533229997Sken } 1534229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1535229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1536229997Sken VREF(rootvnode); 1537229997Sken } 1538229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1539229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1540229997Sken VREF(rootvnode); 1541229997Sken } 1542229997Sken 1543229997Sken again: 1544229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1545229997Sken error = vn_open(&nd, &flags, 0, NULL); 1546229997Sken if (error) { 1547229997Sken /* 1548229997Sken * This is the only reasonable guess we can make as far as 1549229997Sken * path if the user doesn't give us a fully qualified path. 1550229997Sken * If they want to specify a file, they need to specify the 1551229997Sken * full path. 1552229997Sken */ 1553229997Sken if (be_lun->dev_path[0] != '/') { 1554229997Sken char *dev_path = "/dev/"; 1555229997Sken char *dev_name; 1556229997Sken 1557229997Sken /* Try adding device path at beginning of name */ 1558229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1559229997Sken + strlen(dev_path) + 1, 1560229997Sken M_CTLBLK, M_WAITOK); 1561229997Sken if (dev_name) { 1562229997Sken sprintf(dev_name, "%s%s", dev_path, 1563229997Sken be_lun->dev_path); 1564229997Sken free(be_lun->dev_path, M_CTLBLK); 1565229997Sken be_lun->dev_path = dev_name; 1566229997Sken goto again; 1567229997Sken } 1568229997Sken } 1569229997Sken snprintf(req->error_str, sizeof(req->error_str), 1570229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1571229997Sken return (error); 1572229997Sken } 1573229997Sken 1574229997Sken vfs_is_locked = NDHASGIANT(&nd); 1575229997Sken 1576229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1577229997Sken 1578229997Sken be_lun->vn = nd.ni_vp; 1579229997Sken 1580229997Sken /* We only support disks and files. */ 1581229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1582229997Sken error = ctl_be_block_open_dev(be_lun, req); 1583229997Sken } else if (be_lun->vn->v_type == VREG) { 1584229997Sken error = ctl_be_block_open_file(be_lun, req); 1585229997Sken } else { 1586229997Sken error = EINVAL; 1587229997Sken snprintf(req->error_str, sizeof(req->error_str), 1588229997Sken "%s is not a disk or file", be_lun->dev_path); 1589229997Sken } 1590229997Sken VOP_UNLOCK(be_lun->vn, 0); 1591229997Sken VFS_UNLOCK_GIANT(vfs_is_locked); 1592229997Sken 1593229997Sken if (error != 0) { 1594229997Sken ctl_be_block_close(be_lun); 1595229997Sken return (error); 1596229997Sken } 1597229997Sken 1598229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1599229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1600229997Sken 1601229997Sken return (0); 1602229997Sken 1603229997Sken} 1604229997Sken 1605229997Skenstatic int 1606229997Skenctl_be_block_mem_ctor(void *mem, int size, void *arg, int flags) 1607229997Sken{ 1608229997Sken return (0); 1609229997Sken} 1610229997Sken 1611229997Skenstatic void 1612229997Skenctl_be_block_mem_dtor(void *mem, int size, void *arg) 1613229997Sken{ 1614229997Sken bzero(mem, size); 1615229997Sken} 1616229997Sken 1617229997Skenstatic int 1618229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1619229997Sken{ 1620229997Sken struct ctl_be_block_lun *be_lun; 1621229997Sken struct ctl_lun_create_params *params; 1622229997Sken struct ctl_be_arg *file_arg; 1623229997Sken char tmpstr[32]; 1624229997Sken int retval, num_threads; 1625229997Sken int i; 1626229997Sken 1627229997Sken params = &req->reqdata.create; 1628229997Sken retval = 0; 1629229997Sken 1630229997Sken num_threads = cbb_num_threads; 1631229997Sken 1632229997Sken file_arg = NULL; 1633229997Sken 1634229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1635229997Sken 1636229997Sken if (be_lun == NULL) { 1637229997Sken snprintf(req->error_str, sizeof(req->error_str), 1638229997Sken "%s: error allocating %zd bytes", __func__, 1639229997Sken sizeof(*be_lun)); 1640229997Sken goto bailout_error; 1641229997Sken } 1642229997Sken 1643229997Sken be_lun->softc = softc; 1644229997Sken STAILQ_INIT(&be_lun->input_queue); 1645229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1646229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1647229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1648229997Sken mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF); 1649229997Sken 1650229997Sken be_lun->lun_zone = uma_zcreate(be_lun->lunname, MAXPHYS, 1651229997Sken ctl_be_block_mem_ctor, ctl_be_block_mem_dtor, NULL, NULL, 1652229997Sken /*align*/ 0, /*flags*/0); 1653229997Sken 1654229997Sken if (be_lun->lun_zone == NULL) { 1655229997Sken snprintf(req->error_str, sizeof(req->error_str), 1656229997Sken "%s: error allocating UMA zone", __func__); 1657229997Sken goto bailout_error; 1658229997Sken } 1659229997Sken 1660229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1661229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1662229997Sken else 1663229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1664229997Sken 1665229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1666229997Sken for (i = 0; i < req->num_be_args; i++) { 1667229997Sken if (strcmp(req->kern_be_args[i].name, "file") == 0) { 1668229997Sken file_arg = &req->kern_be_args[i]; 1669229997Sken break; 1670229997Sken } 1671229997Sken } 1672229997Sken 1673229997Sken if (file_arg == NULL) { 1674229997Sken snprintf(req->error_str, sizeof(req->error_str), 1675229997Sken "%s: no file argument specified", __func__); 1676229997Sken goto bailout_error; 1677229997Sken } 1678229997Sken 1679229997Sken be_lun->dev_path = malloc(file_arg->vallen, M_CTLBLK, 1680229997Sken M_WAITOK | M_ZERO); 1681229997Sken if (be_lun->dev_path == NULL) { 1682229997Sken snprintf(req->error_str, sizeof(req->error_str), 1683229997Sken "%s: error allocating %d bytes", __func__, 1684229997Sken file_arg->vallen); 1685229997Sken goto bailout_error; 1686229997Sken } 1687229997Sken 1688229997Sken strlcpy(be_lun->dev_path, (char *)file_arg->value, 1689229997Sken file_arg->vallen); 1690229997Sken 1691229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1692229997Sken if (retval != 0) { 1693229997Sken retval = 0; 1694229997Sken goto bailout_error; 1695229997Sken } 1696229997Sken 1697229997Sken /* 1698229997Sken * Tell the user the size of the file/device. 1699229997Sken */ 1700229997Sken params->lun_size_bytes = be_lun->size_bytes; 1701229997Sken 1702229997Sken /* 1703229997Sken * The maximum LBA is the size - 1. 1704229997Sken */ 1705229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1706229997Sken } else { 1707229997Sken /* 1708229997Sken * For processor devices, we don't have any size. 1709229997Sken */ 1710229997Sken be_lun->blocksize = 0; 1711229997Sken be_lun->size_blocks = 0; 1712229997Sken be_lun->size_bytes = 0; 1713229997Sken be_lun->ctl_be_lun.maxlba = 0; 1714229997Sken params->lun_size_bytes = 0; 1715229997Sken 1716229997Sken /* 1717229997Sken * Default to just 1 thread for processor devices. 1718229997Sken */ 1719229997Sken num_threads = 1; 1720229997Sken } 1721229997Sken 1722229997Sken /* 1723229997Sken * XXX This searching loop might be refactored to be combined with 1724229997Sken * the loop above, 1725229997Sken */ 1726229997Sken for (i = 0; i < req->num_be_args; i++) { 1727229997Sken if (strcmp(req->kern_be_args[i].name, "num_threads") == 0) { 1728229997Sken struct ctl_be_arg *thread_arg; 1729229997Sken char num_thread_str[16]; 1730229997Sken int tmp_num_threads; 1731229997Sken 1732229997Sken 1733229997Sken thread_arg = &req->kern_be_args[i]; 1734229997Sken 1735229997Sken strlcpy(num_thread_str, (char *)thread_arg->value, 1736229997Sken min(thread_arg->vallen, 1737229997Sken sizeof(num_thread_str))); 1738229997Sken 1739229997Sken tmp_num_threads = strtol(num_thread_str, NULL, 0); 1740229997Sken 1741229997Sken /* 1742229997Sken * We don't let the user specify less than one 1743229997Sken * thread, but hope he's clueful enough not to 1744229997Sken * specify 1000 threads. 1745229997Sken */ 1746229997Sken if (tmp_num_threads < 1) { 1747229997Sken snprintf(req->error_str, sizeof(req->error_str), 1748229997Sken "%s: invalid number of threads %s", 1749229997Sken __func__, num_thread_str); 1750229997Sken goto bailout_error; 1751229997Sken } 1752229997Sken 1753229997Sken num_threads = tmp_num_threads; 1754229997Sken } 1755229997Sken } 1756229997Sken 1757229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1758229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1759229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 1760229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1761229997Sken /* Tell the user the blocksize we ended up using */ 1762229997Sken params->blocksize_bytes = be_lun->blocksize; 1763229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1764229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1765229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1766229997Sken } else 1767229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 1768229997Sken 1769229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1770229997Sken be_lun->ctl_be_lun.lun_config_status = 1771229997Sken ctl_be_block_lun_config_status; 1772229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1773229997Sken 1774229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1775229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1776229997Sken softc->num_luns); 1777229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1778229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1779229997Sken sizeof(tmpstr))); 1780229997Sken 1781229997Sken /* Tell the user what we used for a serial number */ 1782229997Sken strncpy((char *)params->serial_num, tmpstr, 1783229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1784229997Sken } else { 1785229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 1786229997Sken params->serial_num, 1787229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1788229997Sken sizeof(params->serial_num))); 1789229997Sken } 1790229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1791229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1792229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1793229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1794229997Sken sizeof(tmpstr))); 1795229997Sken 1796229997Sken /* Tell the user what we used for a device ID */ 1797229997Sken strncpy((char *)params->device_id, tmpstr, 1798229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1799229997Sken } else { 1800229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 1801229997Sken params->device_id, 1802229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1803229997Sken sizeof(params->device_id))); 1804229997Sken } 1805229997Sken 1806229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 1807229997Sken 1808229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 1809229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 1810229997Sken 1811229997Sken if (be_lun->io_taskqueue == NULL) { 1812229997Sken snprintf(req->error_str, sizeof(req->error_str), 1813229997Sken "%s: Unable to create taskqueue", __func__); 1814229997Sken goto bailout_error; 1815229997Sken } 1816229997Sken 1817229997Sken /* 1818229997Sken * Note that we start the same number of threads by default for 1819229997Sken * both the file case and the block device case. For the file 1820229997Sken * case, we need multiple threads to allow concurrency, because the 1821229997Sken * vnode interface is designed to be a blocking interface. For the 1822229997Sken * block device case, ZFS zvols at least will block the caller's 1823229997Sken * context in many instances, and so we need multiple threads to 1824229997Sken * overcome that problem. Other block devices don't need as many 1825229997Sken * threads, but they shouldn't cause too many problems. 1826229997Sken * 1827229997Sken * If the user wants to just have a single thread for a block 1828229997Sken * device, he can specify that when the LUN is created, or change 1829229997Sken * the tunable/sysctl to alter the default number of threads. 1830229997Sken */ 1831229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 1832229997Sken /*num threads*/num_threads, 1833229997Sken /*priority*/PWAIT, 1834229997Sken /*thread name*/ 1835229997Sken "%s taskq", be_lun->lunname); 1836229997Sken 1837229997Sken if (retval != 0) 1838229997Sken goto bailout_error; 1839229997Sken 1840229997Sken be_lun->num_threads = num_threads; 1841229997Sken 1842229997Sken mtx_lock(&softc->lock); 1843229997Sken softc->num_luns++; 1844229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 1845229997Sken 1846229997Sken mtx_unlock(&softc->lock); 1847229997Sken 1848229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 1849229997Sken if (retval != 0) { 1850229997Sken mtx_lock(&softc->lock); 1851229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1852229997Sken links); 1853229997Sken softc->num_luns--; 1854229997Sken mtx_unlock(&softc->lock); 1855229997Sken snprintf(req->error_str, sizeof(req->error_str), 1856229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 1857229997Sken "details", __func__, retval); 1858229997Sken retval = 0; 1859229997Sken goto bailout_error; 1860229997Sken } 1861229997Sken 1862229997Sken mtx_lock(&softc->lock); 1863229997Sken 1864229997Sken /* 1865229997Sken * Tell the config_status routine that we're waiting so it won't 1866229997Sken * clean up the LUN in the event of an error. 1867229997Sken */ 1868229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1869229997Sken 1870229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 1871229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1872229997Sken if (retval == EINTR) 1873229997Sken break; 1874229997Sken } 1875229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1876229997Sken 1877229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 1878229997Sken snprintf(req->error_str, sizeof(req->error_str), 1879229997Sken "%s: LUN configuration error, see dmesg for details", 1880229997Sken __func__); 1881229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1882229997Sken links); 1883229997Sken softc->num_luns--; 1884229997Sken mtx_unlock(&softc->lock); 1885229997Sken goto bailout_error; 1886229997Sken } else { 1887229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 1888229997Sken } 1889229997Sken 1890229997Sken mtx_unlock(&softc->lock); 1891229997Sken 1892229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 1893229997Sken be_lun->blocksize, 1894229997Sken DEVSTAT_ALL_SUPPORTED, 1895229997Sken be_lun->ctl_be_lun.lun_type 1896229997Sken | DEVSTAT_TYPE_IF_OTHER, 1897229997Sken DEVSTAT_PRIORITY_OTHER); 1898229997Sken 1899229997Sken 1900229997Sken req->status = CTL_LUN_OK; 1901229997Sken 1902229997Sken return (retval); 1903229997Sken 1904229997Skenbailout_error: 1905229997Sken req->status = CTL_LUN_ERROR; 1906229997Sken 1907229997Sken ctl_be_block_close(be_lun); 1908229997Sken 1909229997Sken free(be_lun->dev_path, M_CTLBLK); 1910229997Sken free(be_lun, M_CTLBLK); 1911229997Sken 1912229997Sken return (retval); 1913229997Sken} 1914229997Sken 1915229997Skenstatic int 1916229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1917229997Sken{ 1918229997Sken struct ctl_lun_rm_params *params; 1919229997Sken struct ctl_be_block_lun *be_lun; 1920229997Sken int retval; 1921229997Sken 1922229997Sken params = &req->reqdata.rm; 1923229997Sken 1924229997Sken mtx_lock(&softc->lock); 1925229997Sken 1926229997Sken be_lun = NULL; 1927229997Sken 1928229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 1929229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 1930229997Sken break; 1931229997Sken } 1932229997Sken mtx_unlock(&softc->lock); 1933229997Sken 1934229997Sken if (be_lun == NULL) { 1935229997Sken snprintf(req->error_str, sizeof(req->error_str), 1936229997Sken "%s: LUN %u is not managed by the block backend", 1937229997Sken __func__, params->lun_id); 1938229997Sken goto bailout_error; 1939229997Sken } 1940229997Sken 1941229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 1942229997Sken 1943229997Sken if (retval != 0) { 1944229997Sken snprintf(req->error_str, sizeof(req->error_str), 1945229997Sken "%s: error %d returned from ctl_disable_lun() for " 1946229997Sken "LUN %d", __func__, retval, params->lun_id); 1947229997Sken goto bailout_error; 1948229997Sken 1949229997Sken } 1950229997Sken 1951229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 1952229997Sken if (retval != 0) { 1953229997Sken snprintf(req->error_str, sizeof(req->error_str), 1954229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 1955229997Sken "LUN %d", __func__, retval, params->lun_id); 1956229997Sken goto bailout_error; 1957229997Sken } 1958229997Sken 1959229997Sken mtx_lock(&softc->lock); 1960229997Sken 1961229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1962229997Sken 1963229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1964229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1965229997Sken if (retval == EINTR) 1966229997Sken break; 1967229997Sken } 1968229997Sken 1969229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1970229997Sken 1971229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1972229997Sken snprintf(req->error_str, sizeof(req->error_str), 1973229997Sken "%s: interrupted waiting for LUN to be freed", 1974229997Sken __func__); 1975229997Sken mtx_unlock(&softc->lock); 1976229997Sken goto bailout_error; 1977229997Sken } 1978229997Sken 1979229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 1980229997Sken 1981229997Sken softc->num_luns--; 1982229997Sken mtx_unlock(&softc->lock); 1983229997Sken 1984229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 1985229997Sken 1986229997Sken taskqueue_free(be_lun->io_taskqueue); 1987229997Sken 1988229997Sken ctl_be_block_close(be_lun); 1989229997Sken 1990229997Sken if (be_lun->disk_stats != NULL) 1991229997Sken devstat_remove_entry(be_lun->disk_stats); 1992229997Sken 1993229997Sken uma_zdestroy(be_lun->lun_zone); 1994229997Sken 1995229997Sken free(be_lun->dev_path, M_CTLBLK); 1996229997Sken 1997229997Sken free(be_lun, M_CTLBLK); 1998229997Sken 1999229997Sken req->status = CTL_LUN_OK; 2000229997Sken 2001229997Sken return (0); 2002229997Sken 2003229997Skenbailout_error: 2004229997Sken 2005229997Sken req->status = CTL_LUN_ERROR; 2006229997Sken 2007229997Sken return (0); 2008229997Sken} 2009229997Sken 2010229997Skenstatic void 2011229997Skenctl_be_block_lun_shutdown(void *be_lun) 2012229997Sken{ 2013229997Sken struct ctl_be_block_lun *lun; 2014229997Sken struct ctl_be_block_softc *softc; 2015229997Sken 2016229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2017229997Sken 2018229997Sken softc = lun->softc; 2019229997Sken 2020229997Sken mtx_lock(&softc->lock); 2021229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2022229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2023229997Sken wakeup(lun); 2024229997Sken mtx_unlock(&softc->lock); 2025229997Sken 2026229997Sken} 2027229997Sken 2028229997Skenstatic void 2029229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2030229997Sken{ 2031229997Sken struct ctl_be_block_lun *lun; 2032229997Sken struct ctl_be_block_softc *softc; 2033229997Sken 2034229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2035229997Sken softc = lun->softc; 2036229997Sken 2037229997Sken if (status == CTL_LUN_CONFIG_OK) { 2038229997Sken mtx_lock(&softc->lock); 2039229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2040229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2041229997Sken wakeup(lun); 2042229997Sken mtx_unlock(&softc->lock); 2043229997Sken 2044229997Sken /* 2045229997Sken * We successfully added the LUN, attempt to enable it. 2046229997Sken */ 2047229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2048229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2049229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2050229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2051229997Sken __func__); 2052229997Sken } 2053229997Sken } 2054229997Sken 2055229997Sken return; 2056229997Sken } 2057229997Sken 2058229997Sken 2059229997Sken mtx_lock(&softc->lock); 2060229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2061229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2062229997Sken wakeup(lun); 2063229997Sken mtx_unlock(&softc->lock); 2064229997Sken} 2065229997Sken 2066229997Sken 2067229997Skenstatic int 2068229997Skenctl_be_block_config_write(union ctl_io *io) 2069229997Sken{ 2070229997Sken struct ctl_be_block_lun *be_lun; 2071229997Sken struct ctl_be_lun *ctl_be_lun; 2072229997Sken int retval; 2073229997Sken 2074229997Sken retval = 0; 2075229997Sken 2076229997Sken DPRINTF("entered\n"); 2077229997Sken 2078229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2079229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2080229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2081229997Sken 2082229997Sken switch (io->scsiio.cdb[0]) { 2083229997Sken case SYNCHRONIZE_CACHE: 2084229997Sken case SYNCHRONIZE_CACHE_16: 2085229997Sken /* 2086229997Sken * The upper level CTL code will filter out any CDBs with 2087229997Sken * the immediate bit set and return the proper error. 2088229997Sken * 2089229997Sken * We don't really need to worry about what LBA range the 2090229997Sken * user asked to be synced out. When they issue a sync 2091229997Sken * cache command, we'll sync out the whole thing. 2092229997Sken */ 2093229997Sken mtx_lock(&be_lun->lock); 2094229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2095229997Sken links); 2096229997Sken mtx_unlock(&be_lun->lock); 2097229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2098229997Sken break; 2099229997Sken case START_STOP_UNIT: { 2100229997Sken struct scsi_start_stop_unit *cdb; 2101229997Sken 2102229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2103229997Sken 2104229997Sken if (cdb->how & SSS_START) 2105229997Sken retval = ctl_start_lun(ctl_be_lun); 2106229997Sken else { 2107229997Sken retval = ctl_stop_lun(ctl_be_lun); 2108229997Sken /* 2109229997Sken * XXX KDM Copan-specific offline behavior. 2110229997Sken * Figure out a reasonable way to port this? 2111229997Sken */ 2112229997Sken#ifdef NEEDTOPORT 2113229997Sken if ((retval == 0) 2114229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2115229997Sken retval = ctl_lun_offline(ctl_be_lun); 2116229997Sken#endif 2117229997Sken } 2118229997Sken 2119229997Sken /* 2120229997Sken * In general, the above routines should not fail. They 2121229997Sken * just set state for the LUN. So we've got something 2122229997Sken * pretty wrong here if we can't start or stop the LUN. 2123229997Sken */ 2124229997Sken if (retval != 0) { 2125229997Sken ctl_set_internal_failure(&io->scsiio, 2126229997Sken /*sks_valid*/ 1, 2127229997Sken /*retry_count*/ 0xf051); 2128229997Sken retval = CTL_RETVAL_COMPLETE; 2129229997Sken } else { 2130229997Sken ctl_set_success(&io->scsiio); 2131229997Sken } 2132229997Sken ctl_config_write_done(io); 2133229997Sken break; 2134229997Sken } 2135229997Sken default: 2136229997Sken ctl_set_invalid_opcode(&io->scsiio); 2137229997Sken ctl_config_write_done(io); 2138229997Sken retval = CTL_RETVAL_COMPLETE; 2139229997Sken break; 2140229997Sken } 2141229997Sken 2142229997Sken return (retval); 2143229997Sken 2144229997Sken} 2145229997Sken 2146229997Skenstatic int 2147229997Skenctl_be_block_config_read(union ctl_io *io) 2148229997Sken{ 2149229997Sken return (0); 2150229997Sken} 2151229997Sken 2152229997Skenstatic int 2153229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2154229997Sken{ 2155229997Sken struct ctl_be_block_lun *lun; 2156229997Sken int retval; 2157229997Sken 2158229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2159229997Sken retval = 0; 2160229997Sken 2161229997Sken retval = sbuf_printf(sb, "<num_threads>"); 2162229997Sken 2163229997Sken if (retval != 0) 2164229997Sken goto bailout; 2165229997Sken 2166229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2167229997Sken 2168229997Sken if (retval != 0) 2169229997Sken goto bailout; 2170229997Sken 2171229997Sken retval = sbuf_printf(sb, "</num_threads>"); 2172229997Sken 2173229997Sken /* 2174229997Sken * For processor devices, we don't have a path variable. 2175229997Sken */ 2176229997Sken if ((retval != 0) 2177229997Sken || (lun->dev_path == NULL)) 2178229997Sken goto bailout; 2179229997Sken 2180229997Sken retval = sbuf_printf(sb, "<file>"); 2181229997Sken 2182229997Sken if (retval != 0) 2183229997Sken goto bailout; 2184229997Sken 2185229997Sken retval = ctl_sbuf_printf_esc(sb, lun->dev_path); 2186229997Sken 2187229997Sken if (retval != 0) 2188229997Sken goto bailout; 2189229997Sken 2190229997Sken retval = sbuf_printf(sb, "</file>\n"); 2191229997Sken 2192229997Skenbailout: 2193229997Sken 2194229997Sken return (retval); 2195229997Sken} 2196229997Sken 2197229997Skenint 2198229997Skenctl_be_block_init(void) 2199229997Sken{ 2200229997Sken struct ctl_be_block_softc *softc; 2201229997Sken int retval; 2202229997Sken 2203229997Sken softc = &backend_block_softc; 2204229997Sken retval = 0; 2205229997Sken 2206229997Sken mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF); 2207229997Sken STAILQ_INIT(&softc->beio_free_queue); 2208229997Sken STAILQ_INIT(&softc->disk_list); 2209229997Sken STAILQ_INIT(&softc->lun_list); 2210229997Sken ctl_grow_beio(softc, 200); 2211229997Sken 2212229997Sken return (retval); 2213229997Sken} 2214