ctl_backend_block.c revision 264886
1229997Sken/*- 2229997Sken * Copyright (c) 2003 Silicon Graphics International Corp. 3229997Sken * Copyright (c) 2009-2011 Spectra Logic Corporation 4232604Strasz * Copyright (c) 2012 The FreeBSD Foundation 5229997Sken * All rights reserved. 6229997Sken * 7232604Strasz * Portions of this software were developed by Edward Tomasz Napierala 8232604Strasz * under sponsorship from the FreeBSD Foundation. 9232604Strasz * 10229997Sken * Redistribution and use in source and binary forms, with or without 11229997Sken * modification, are permitted provided that the following conditions 12229997Sken * are met: 13229997Sken * 1. Redistributions of source code must retain the above copyright 14229997Sken * notice, this list of conditions, and the following disclaimer, 15229997Sken * without modification. 16229997Sken * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17229997Sken * substantially similar to the "NO WARRANTY" disclaimer below 18229997Sken * ("Disclaimer") and any redistribution must be conditioned upon 19229997Sken * including a substantially similar Disclaimer requirement for further 20229997Sken * binary redistribution. 21229997Sken * 22229997Sken * NO WARRANTY 23229997Sken * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24229997Sken * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25229997Sken * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26229997Sken * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27229997Sken * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28229997Sken * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29229997Sken * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30229997Sken * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31229997Sken * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32229997Sken * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33229997Sken * POSSIBILITY OF SUCH DAMAGES. 34229997Sken * 35229997Sken * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36229997Sken */ 37229997Sken/* 38229997Sken * CAM Target Layer driver backend for block devices. 39229997Sken * 40229997Sken * Author: Ken Merry <ken@FreeBSD.org> 41229997Sken */ 42229997Sken#include <sys/cdefs.h> 43229997Sken__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl_backend_block.c 264886 2014-04-24 16:19:49Z mav $"); 44229997Sken 45229997Sken#include <sys/param.h> 46229997Sken#include <sys/systm.h> 47229997Sken#include <sys/kernel.h> 48229997Sken#include <sys/types.h> 49229997Sken#include <sys/kthread.h> 50229997Sken#include <sys/bio.h> 51229997Sken#include <sys/fcntl.h> 52264274Smav#include <sys/limits.h> 53229997Sken#include <sys/lock.h> 54229997Sken#include <sys/mutex.h> 55229997Sken#include <sys/condvar.h> 56229997Sken#include <sys/malloc.h> 57229997Sken#include <sys/conf.h> 58229997Sken#include <sys/ioccom.h> 59229997Sken#include <sys/queue.h> 60229997Sken#include <sys/sbuf.h> 61229997Sken#include <sys/endian.h> 62229997Sken#include <sys/uio.h> 63229997Sken#include <sys/buf.h> 64229997Sken#include <sys/taskqueue.h> 65229997Sken#include <sys/vnode.h> 66229997Sken#include <sys/namei.h> 67229997Sken#include <sys/mount.h> 68229997Sken#include <sys/disk.h> 69229997Sken#include <sys/fcntl.h> 70229997Sken#include <sys/filedesc.h> 71229997Sken#include <sys/proc.h> 72229997Sken#include <sys/pcpu.h> 73229997Sken#include <sys/module.h> 74229997Sken#include <sys/sdt.h> 75229997Sken#include <sys/devicestat.h> 76229997Sken#include <sys/sysctl.h> 77229997Sken 78229997Sken#include <geom/geom.h> 79229997Sken 80229997Sken#include <cam/cam.h> 81229997Sken#include <cam/scsi/scsi_all.h> 82229997Sken#include <cam/scsi/scsi_da.h> 83229997Sken#include <cam/ctl/ctl_io.h> 84229997Sken#include <cam/ctl/ctl.h> 85229997Sken#include <cam/ctl/ctl_backend.h> 86229997Sken#include <cam/ctl/ctl_frontend_internal.h> 87229997Sken#include <cam/ctl/ctl_ioctl.h> 88229997Sken#include <cam/ctl/ctl_scsi_all.h> 89229997Sken#include <cam/ctl/ctl_error.h> 90229997Sken 91229997Sken/* 92264886Smav * The idea here is that we'll allocate enough S/G space to hold a 1MB 93264886Smav * I/O. If we get an I/O larger than that, we'll split it. 94229997Sken */ 95264886Smav#define CTLBLK_MAX_IO_SIZE (1024 * 1024) 96264886Smav#define CTLBLK_MAX_SEG MAXPHYS 97264886Smav#define CTLBLK_MAX_SEGS MAX(CTLBLK_MAX_IO_SIZE / CTLBLK_MAX_SEG, 1) 98229997Sken 99229997Sken#ifdef CTLBLK_DEBUG 100229997Sken#define DPRINTF(fmt, args...) \ 101229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 102229997Sken#else 103229997Sken#define DPRINTF(fmt, args...) do {} while(0) 104229997Sken#endif 105229997Sken 106229997SkenSDT_PROVIDER_DEFINE(cbb); 107229997Sken 108229997Skentypedef enum { 109229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 110229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 111229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 112229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 113229997Sken} ctl_be_block_lun_flags; 114229997Sken 115229997Skentypedef enum { 116229997Sken CTL_BE_BLOCK_NONE, 117229997Sken CTL_BE_BLOCK_DEV, 118229997Sken CTL_BE_BLOCK_FILE 119229997Sken} ctl_be_block_type; 120229997Sken 121229997Skenstruct ctl_be_block_devdata { 122229997Sken struct cdev *cdev; 123229997Sken struct cdevsw *csw; 124229997Sken int dev_ref; 125229997Sken}; 126229997Sken 127229997Skenstruct ctl_be_block_filedata { 128229997Sken struct ucred *cred; 129229997Sken}; 130229997Sken 131229997Skenunion ctl_be_block_bedata { 132229997Sken struct ctl_be_block_devdata dev; 133229997Sken struct ctl_be_block_filedata file; 134229997Sken}; 135229997Sken 136229997Skenstruct ctl_be_block_io; 137229997Skenstruct ctl_be_block_lun; 138229997Sken 139229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 140229997Sken struct ctl_be_block_io *beio); 141229997Sken 142229997Sken/* 143229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 144229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 145229997Sken */ 146229997Skenstruct ctl_be_block_lun { 147229997Sken struct ctl_block_disk *disk; 148229997Sken char lunname[32]; 149229997Sken char *dev_path; 150229997Sken ctl_be_block_type dev_type; 151229997Sken struct vnode *vn; 152229997Sken union ctl_be_block_bedata backend; 153229997Sken cbb_dispatch_t dispatch; 154229997Sken cbb_dispatch_t lun_flush; 155264274Smav cbb_dispatch_t unmap; 156229997Sken struct mtx lock; 157229997Sken uma_zone_t lun_zone; 158229997Sken uint64_t size_blocks; 159229997Sken uint64_t size_bytes; 160229997Sken uint32_t blocksize; 161229997Sken int blocksize_shift; 162264191Smav uint16_t pblockexp; 163264191Smav uint16_t pblockoff; 164229997Sken struct ctl_be_block_softc *softc; 165229997Sken struct devstat *disk_stats; 166229997Sken ctl_be_block_lun_flags flags; 167229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 168229997Sken struct ctl_be_lun ctl_be_lun; 169229997Sken struct taskqueue *io_taskqueue; 170229997Sken struct task io_task; 171229997Sken int num_threads; 172229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 173229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 174229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 175229997Sken}; 176229997Sken 177229997Sken/* 178229997Sken * Overall softc structure for the block backend module. 179229997Sken */ 180229997Skenstruct ctl_be_block_softc { 181229997Sken struct mtx lock; 182229997Sken int num_disks; 183229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 184229997Sken int num_luns; 185229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 186229997Sken}; 187229997Sken 188229997Skenstatic struct ctl_be_block_softc backend_block_softc; 189229997Sken 190229997Sken/* 191229997Sken * Per-I/O information. 192229997Sken */ 193229997Skenstruct ctl_be_block_io { 194229997Sken union ctl_io *io; 195229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 196229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 197229997Sken int bio_cmd; 198229997Sken int bio_flags; 199229997Sken int num_segs; 200229997Sken int num_bios_sent; 201229997Sken int num_bios_done; 202229997Sken int send_complete; 203229997Sken int num_errors; 204229997Sken struct bintime ds_t0; 205229997Sken devstat_tag_type ds_tag_type; 206229997Sken devstat_trans_flags ds_trans_type; 207229997Sken uint64_t io_len; 208229997Sken uint64_t io_offset; 209229997Sken struct ctl_be_block_softc *softc; 210229997Sken struct ctl_be_block_lun *lun; 211264274Smav void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 212229997Sken}; 213229997Sken 214229997Skenstatic int cbb_num_threads = 14; 215229997SkenTUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 216229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 217229997Sken "CAM Target Layer Block Backend"); 218229997SkenSYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 219229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 220229997Sken 221229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 222229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 223229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 224229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 225229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 226229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 227229997Sken struct ctl_be_block_io *beio); 228229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 229229997Sken struct ctl_be_block_io *beio); 230229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 231229997Sken struct ctl_be_block_io *beio); 232264274Smavstatic void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 233264274Smav struct ctl_be_block_io *beio); 234229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 235229997Sken struct ctl_be_block_io *beio); 236229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 237229997Sken union ctl_io *io); 238229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 239229997Sken union ctl_io *io); 240229997Skenstatic void ctl_be_block_worker(void *context, int pending); 241229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 242229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 243229997Sken int flag, struct thread *td); 244229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 245229997Sken struct ctl_lun_req *req); 246229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 247229997Sken struct ctl_lun_req *req); 248229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 249229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 250229997Sken struct ctl_be_block_lun *be_lun, 251229997Sken struct ctl_lun_req *req); 252229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 253229997Sken struct ctl_lun_req *req); 254229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 255229997Sken struct ctl_lun_req *req); 256232604Straszstatic int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 257232604Strasz struct ctl_lun_req *req); 258232604Straszstatic int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 259232604Strasz struct ctl_lun_req *req); 260232604Straszstatic int ctl_be_block_modify(struct ctl_be_block_softc *softc, 261232604Strasz struct ctl_lun_req *req); 262229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 263229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 264229997Sken ctl_lun_config_status status); 265229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 266229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 267229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 268229997Skenint ctl_be_block_init(void); 269229997Sken 270229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 271229997Sken{ 272230334Sken .name = "block", 273230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 274230334Sken .init = ctl_be_block_init, 275230334Sken .data_submit = ctl_be_block_submit, 276230334Sken .data_move_done = ctl_be_block_move_done, 277230334Sken .config_read = ctl_be_block_config_read, 278230334Sken .config_write = ctl_be_block_config_write, 279230334Sken .ioctl = ctl_be_block_ioctl, 280230334Sken .lun_info = ctl_be_block_lun_info 281229997Sken}; 282229997Sken 283229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 284229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 285229997Sken 286264020Straszstatic uma_zone_t beio_zone; 287264020Strasz 288229997Skenstatic struct ctl_be_block_io * 289229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 290229997Sken{ 291229997Sken struct ctl_be_block_io *beio; 292229997Sken 293264020Strasz beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 294264020Strasz beio->softc = softc; 295229997Sken return (beio); 296229997Sken} 297229997Sken 298229997Skenstatic void 299229997Skenctl_free_beio(struct ctl_be_block_io *beio) 300229997Sken{ 301229997Sken int duplicate_free; 302229997Sken int i; 303229997Sken 304229997Sken duplicate_free = 0; 305229997Sken 306229997Sken for (i = 0; i < beio->num_segs; i++) { 307229997Sken if (beio->sg_segs[i].addr == NULL) 308229997Sken duplicate_free++; 309229997Sken 310229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 311229997Sken beio->sg_segs[i].addr = NULL; 312229997Sken } 313229997Sken 314229997Sken if (duplicate_free > 0) { 315229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 316229997Sken duplicate_free, beio->num_segs); 317229997Sken } 318229997Sken 319264020Strasz uma_zfree(beio_zone, beio); 320229997Sken} 321229997Sken 322229997Skenstatic void 323229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 324229997Sken{ 325229997Sken union ctl_io *io; 326229997Sken int io_len; 327229997Sken 328229997Sken io = beio->io; 329229997Sken 330229997Sken if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) 331229997Sken io_len = beio->io_len; 332229997Sken else 333229997Sken io_len = 0; 334229997Sken 335229997Sken devstat_end_transaction(beio->lun->disk_stats, 336229997Sken /*bytes*/ io_len, 337229997Sken beio->ds_tag_type, 338229997Sken beio->ds_trans_type, 339229997Sken /*now*/ NULL, 340229997Sken /*then*/&beio->ds_t0); 341229997Sken 342264274Smav if (beio->beio_cont != NULL) { 343264274Smav beio->beio_cont(beio); 344264274Smav } else { 345264274Smav ctl_free_beio(beio); 346264274Smav ctl_done(io); 347264274Smav } 348229997Sken} 349229997Sken 350229997Skenstatic int 351229997Skenctl_be_block_move_done(union ctl_io *io) 352229997Sken{ 353229997Sken struct ctl_be_block_io *beio; 354229997Sken struct ctl_be_block_lun *be_lun; 355229997Sken#ifdef CTL_TIME_IO 356229997Sken struct bintime cur_bt; 357229997Sken#endif 358229997Sken 359229997Sken beio = (struct ctl_be_block_io *) 360229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 361229997Sken 362229997Sken be_lun = beio->lun; 363229997Sken 364229997Sken DPRINTF("entered\n"); 365229997Sken 366229997Sken#ifdef CTL_TIME_IO 367229997Sken getbintime(&cur_bt); 368229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 369229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 370229997Sken io->io_hdr.num_dmas++; 371229997Sken#endif 372229997Sken 373229997Sken /* 374229997Sken * We set status at this point for read commands, and write 375229997Sken * commands with errors. 376229997Sken */ 377229997Sken if ((beio->bio_cmd == BIO_READ) 378229997Sken && (io->io_hdr.port_status == 0) 379229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 380229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) 381229997Sken ctl_set_success(&io->scsiio); 382229997Sken else if ((io->io_hdr.port_status != 0) 383229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 384229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 385229997Sken /* 386229997Sken * For hardware error sense keys, the sense key 387229997Sken * specific value is defined to be a retry count, 388229997Sken * but we use it to pass back an internal FETD 389229997Sken * error code. XXX KDM Hopefully the FETD is only 390229997Sken * using 16 bits for an error code, since that's 391229997Sken * all the space we have in the sks field. 392229997Sken */ 393229997Sken ctl_set_internal_failure(&io->scsiio, 394229997Sken /*sks_valid*/ 1, 395229997Sken /*retry_count*/ 396229997Sken io->io_hdr.port_status); 397229997Sken } 398229997Sken 399229997Sken /* 400229997Sken * If this is a read, or a write with errors, it is done. 401229997Sken */ 402229997Sken if ((beio->bio_cmd == BIO_READ) 403229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 404229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 405229997Sken ctl_complete_beio(beio); 406229997Sken return (0); 407229997Sken } 408229997Sken 409229997Sken /* 410229997Sken * At this point, we have a write and the DMA completed 411229997Sken * successfully. We now have to queue it to the task queue to 412229997Sken * execute the backend I/O. That is because we do blocking 413229997Sken * memory allocations, and in the file backing case, blocking I/O. 414229997Sken * This move done routine is generally called in the SIM's 415229997Sken * interrupt context, and therefore we cannot block. 416229997Sken */ 417229997Sken mtx_lock(&be_lun->lock); 418229997Sken /* 419229997Sken * XXX KDM make sure that links is okay to use at this point. 420229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 421229997Sken * or deal with resource allocation here. 422229997Sken */ 423229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 424229997Sken mtx_unlock(&be_lun->lock); 425229997Sken 426229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 427229997Sken 428229997Sken return (0); 429229997Sken} 430229997Sken 431229997Skenstatic void 432229997Skenctl_be_block_biodone(struct bio *bio) 433229997Sken{ 434229997Sken struct ctl_be_block_io *beio; 435229997Sken struct ctl_be_block_lun *be_lun; 436229997Sken union ctl_io *io; 437261538Smav int error; 438229997Sken 439229997Sken beio = bio->bio_caller1; 440229997Sken be_lun = beio->lun; 441229997Sken io = beio->io; 442229997Sken 443229997Sken DPRINTF("entered\n"); 444229997Sken 445261538Smav error = bio->bio_error; 446229997Sken mtx_lock(&be_lun->lock); 447261538Smav if (error != 0) 448229997Sken beio->num_errors++; 449229997Sken 450229997Sken beio->num_bios_done++; 451229997Sken 452229997Sken /* 453229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 454229997Sken * during the free might cause it to complain. 455229997Sken */ 456229997Sken g_destroy_bio(bio); 457229997Sken 458229997Sken /* 459229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 460229997Sken * complete, then we're done. 461229997Sken */ 462229997Sken if ((beio->send_complete == 0) 463229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 464229997Sken mtx_unlock(&be_lun->lock); 465229997Sken return; 466229997Sken } 467229997Sken 468229997Sken /* 469229997Sken * At this point, we've verified that we are the last I/O to 470229997Sken * complete, so it's safe to drop the lock. 471229997Sken */ 472229997Sken mtx_unlock(&be_lun->lock); 473229997Sken 474229997Sken /* 475229997Sken * If there are any errors from the backing device, we fail the 476229997Sken * entire I/O with a medium error. 477229997Sken */ 478229997Sken if (beio->num_errors > 0) { 479261538Smav if (error == EOPNOTSUPP) { 480261538Smav ctl_set_invalid_opcode(&io->scsiio); 481261538Smav } else if (beio->bio_cmd == BIO_FLUSH) { 482229997Sken /* XXX KDM is there is a better error here? */ 483229997Sken ctl_set_internal_failure(&io->scsiio, 484229997Sken /*sks_valid*/ 1, 485229997Sken /*retry_count*/ 0xbad2); 486229997Sken } else 487229997Sken ctl_set_medium_error(&io->scsiio); 488229997Sken ctl_complete_beio(beio); 489229997Sken return; 490229997Sken } 491229997Sken 492229997Sken /* 493264274Smav * If this is a write, a flush or a delete, we're all done. 494229997Sken * If this is a read, we can now send the data to the user. 495229997Sken */ 496229997Sken if ((beio->bio_cmd == BIO_WRITE) 497264274Smav || (beio->bio_cmd == BIO_FLUSH) 498264274Smav || (beio->bio_cmd == BIO_DELETE)) { 499229997Sken ctl_set_success(&io->scsiio); 500229997Sken ctl_complete_beio(beio); 501229997Sken } else { 502229997Sken#ifdef CTL_TIME_IO 503229997Sken getbintime(&io->io_hdr.dma_start_bt); 504229997Sken#endif 505229997Sken ctl_datamove(io); 506229997Sken } 507229997Sken} 508229997Sken 509229997Skenstatic void 510229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 511229997Sken struct ctl_be_block_io *beio) 512229997Sken{ 513229997Sken union ctl_io *io; 514229997Sken struct mount *mountpoint; 515241896Skib int error, lock_flags; 516229997Sken 517229997Sken DPRINTF("entered\n"); 518229997Sken 519229997Sken io = beio->io; 520229997Sken 521229997Sken (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 522229997Sken 523229997Sken if (MNT_SHARED_WRITES(mountpoint) 524229997Sken || ((mountpoint == NULL) 525229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 526229997Sken lock_flags = LK_SHARED; 527229997Sken else 528229997Sken lock_flags = LK_EXCLUSIVE; 529229997Sken 530229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 531229997Sken 532229997Sken binuptime(&beio->ds_t0); 533229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 534229997Sken 535229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 536229997Sken VOP_UNLOCK(be_lun->vn, 0); 537229997Sken 538229997Sken vn_finished_write(mountpoint); 539229997Sken 540229997Sken if (error == 0) 541229997Sken ctl_set_success(&io->scsiio); 542229997Sken else { 543229997Sken /* XXX KDM is there is a better error here? */ 544229997Sken ctl_set_internal_failure(&io->scsiio, 545229997Sken /*sks_valid*/ 1, 546229997Sken /*retry_count*/ 0xbad1); 547229997Sken } 548229997Sken 549229997Sken ctl_complete_beio(beio); 550229997Sken} 551229997Sken 552258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 553258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 554258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 555258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 556229997Sken 557229997Skenstatic void 558229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 559229997Sken struct ctl_be_block_io *beio) 560229997Sken{ 561229997Sken struct ctl_be_block_filedata *file_data; 562229997Sken union ctl_io *io; 563229997Sken struct uio xuio; 564229997Sken struct iovec *xiovec; 565241896Skib int flags; 566229997Sken int error, i; 567229997Sken 568229997Sken DPRINTF("entered\n"); 569229997Sken 570229997Sken file_data = &be_lun->backend.file; 571229997Sken io = beio->io; 572229997Sken flags = beio->bio_flags; 573229997Sken 574229997Sken if (beio->bio_cmd == BIO_READ) { 575229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 576229997Sken } else { 577229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 578229997Sken } 579229997Sken 580229997Sken bzero(&xuio, sizeof(xuio)); 581229997Sken if (beio->bio_cmd == BIO_READ) 582229997Sken xuio.uio_rw = UIO_READ; 583229997Sken else 584229997Sken xuio.uio_rw = UIO_WRITE; 585229997Sken 586229997Sken xuio.uio_offset = beio->io_offset; 587229997Sken xuio.uio_resid = beio->io_len; 588229997Sken xuio.uio_segflg = UIO_SYSSPACE; 589229997Sken xuio.uio_iov = beio->xiovecs; 590229997Sken xuio.uio_iovcnt = beio->num_segs; 591229997Sken xuio.uio_td = curthread; 592229997Sken 593229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 594229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 595229997Sken xiovec->iov_len = beio->sg_segs[i].len; 596229997Sken } 597229997Sken 598229997Sken if (beio->bio_cmd == BIO_READ) { 599229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 600229997Sken 601229997Sken binuptime(&beio->ds_t0); 602229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 603229997Sken 604229997Sken /* 605229997Sken * UFS pays attention to IO_DIRECT for reads. If the 606229997Sken * DIRECTIO option is configured into the kernel, it calls 607229997Sken * ffs_rawread(). But that only works for single-segment 608229997Sken * uios with user space addresses. In our case, with a 609229997Sken * kernel uio, it still reads into the buffer cache, but it 610229997Sken * will just try to release the buffer from the cache later 611229997Sken * on in ffs_read(). 612229997Sken * 613229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 614229997Sken * 615229997Sken * UFS does not pay attention to IO_SYNC for reads. 616229997Sken * 617229997Sken * ZFS pays attention to IO_SYNC (which translates into the 618229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 619229997Sken * attempts to sync the file before reading. 620229997Sken * 621229997Sken * So, to attempt to provide some barrier semantics in the 622229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 623229997Sken */ 624229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 625229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 626229997Sken 627229997Sken VOP_UNLOCK(be_lun->vn, 0); 628229997Sken } else { 629229997Sken struct mount *mountpoint; 630229997Sken int lock_flags; 631229997Sken 632229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 633229997Sken 634229997Sken if (MNT_SHARED_WRITES(mountpoint) 635229997Sken || ((mountpoint == NULL) 636229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 637229997Sken lock_flags = LK_SHARED; 638229997Sken else 639229997Sken lock_flags = LK_EXCLUSIVE; 640229997Sken 641229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 642229997Sken 643229997Sken binuptime(&beio->ds_t0); 644229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 645229997Sken 646229997Sken /* 647229997Sken * UFS pays attention to IO_DIRECT for writes. The write 648229997Sken * is done asynchronously. (Normally the write would just 649229997Sken * get put into cache. 650229997Sken * 651229997Sken * UFS pays attention to IO_SYNC for writes. It will 652229997Sken * attempt to write the buffer out synchronously if that 653229997Sken * flag is set. 654229997Sken * 655229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 656229997Sken * 657229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 658229997Sken * for writes. It will flush the transaction from the 659229997Sken * cache before returning. 660229997Sken * 661229997Sken * So if we've got the BIO_ORDERED flag set, we want 662229997Sken * IO_SYNC in either the UFS or ZFS case. 663229997Sken */ 664229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 665229997Sken IO_SYNC : 0, file_data->cred); 666229997Sken VOP_UNLOCK(be_lun->vn, 0); 667229997Sken 668229997Sken vn_finished_write(mountpoint); 669229997Sken } 670229997Sken 671229997Sken /* 672229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 673229997Sken * return the I/O to the user. 674229997Sken */ 675229997Sken if (error != 0) { 676229997Sken char path_str[32]; 677229997Sken 678229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 679229997Sken /* 680229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 681229997Sken * filesystem fills up. What kind of SCSI error should we 682229997Sken * return for that? 683229997Sken */ 684229997Sken printf("%s%s command returned errno %d\n", path_str, 685229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 686229997Sken ctl_set_medium_error(&io->scsiio); 687229997Sken ctl_complete_beio(beio); 688229997Sken return; 689229997Sken } 690229997Sken 691229997Sken /* 692229997Sken * If this is a write, we're all done. 693229997Sken * If this is a read, we can now send the data to the user. 694229997Sken */ 695229997Sken if (beio->bio_cmd == BIO_WRITE) { 696229997Sken ctl_set_success(&io->scsiio); 697229997Sken SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 698229997Sken ctl_complete_beio(beio); 699229997Sken } else { 700229997Sken SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 701229997Sken#ifdef CTL_TIME_IO 702229997Sken getbintime(&io->io_hdr.dma_start_bt); 703229997Sken#endif 704229997Sken ctl_datamove(io); 705229997Sken } 706229997Sken} 707229997Sken 708229997Skenstatic void 709229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 710229997Sken struct ctl_be_block_io *beio) 711229997Sken{ 712229997Sken struct bio *bio; 713229997Sken union ctl_io *io; 714229997Sken struct ctl_be_block_devdata *dev_data; 715229997Sken 716229997Sken dev_data = &be_lun->backend.dev; 717229997Sken io = beio->io; 718229997Sken 719229997Sken DPRINTF("entered\n"); 720229997Sken 721229997Sken /* This can't fail, it's a blocking allocation. */ 722229997Sken bio = g_alloc_bio(); 723229997Sken 724229997Sken bio->bio_cmd = BIO_FLUSH; 725229997Sken bio->bio_flags |= BIO_ORDERED; 726229997Sken bio->bio_dev = dev_data->cdev; 727229997Sken bio->bio_offset = 0; 728229997Sken bio->bio_data = 0; 729229997Sken bio->bio_done = ctl_be_block_biodone; 730229997Sken bio->bio_caller1 = beio; 731229997Sken bio->bio_pblkno = 0; 732229997Sken 733229997Sken /* 734229997Sken * We don't need to acquire the LUN lock here, because we are only 735229997Sken * sending one bio, and so there is no other context to synchronize 736229997Sken * with. 737229997Sken */ 738229997Sken beio->num_bios_sent = 1; 739229997Sken beio->send_complete = 1; 740229997Sken 741229997Sken binuptime(&beio->ds_t0); 742229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 743229997Sken 744229997Sken (*dev_data->csw->d_strategy)(bio); 745229997Sken} 746229997Sken 747229997Skenstatic void 748264274Smavctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 749264274Smav struct ctl_be_block_io *beio, 750264274Smav uint64_t off, uint64_t len, int last) 751264274Smav{ 752264274Smav struct bio *bio; 753264274Smav struct ctl_be_block_devdata *dev_data; 754264296Smav uint64_t maxlen; 755264274Smav 756264274Smav dev_data = &be_lun->backend.dev; 757264296Smav maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 758264274Smav while (len > 0) { 759264274Smav bio = g_alloc_bio(); 760264274Smav bio->bio_cmd = BIO_DELETE; 761264274Smav bio->bio_flags |= beio->bio_flags; 762264274Smav bio->bio_dev = dev_data->cdev; 763264274Smav bio->bio_offset = off; 764264296Smav bio->bio_length = MIN(len, maxlen); 765264274Smav bio->bio_data = 0; 766264274Smav bio->bio_done = ctl_be_block_biodone; 767264274Smav bio->bio_caller1 = beio; 768264296Smav bio->bio_pblkno = off / be_lun->blocksize; 769264274Smav 770264274Smav off += bio->bio_length; 771264274Smav len -= bio->bio_length; 772264274Smav 773264274Smav mtx_lock(&be_lun->lock); 774264274Smav beio->num_bios_sent++; 775264274Smav if (last && len == 0) 776264274Smav beio->send_complete = 1; 777264274Smav mtx_unlock(&be_lun->lock); 778264274Smav 779264274Smav (*dev_data->csw->d_strategy)(bio); 780264274Smav } 781264274Smav} 782264274Smav 783264274Smavstatic void 784264274Smavctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 785264274Smav struct ctl_be_block_io *beio) 786264274Smav{ 787264274Smav union ctl_io *io; 788264274Smav struct ctl_be_block_devdata *dev_data; 789264274Smav struct ctl_ptr_len_flags ptrlen; 790264274Smav struct scsi_unmap_desc *buf, *end; 791264274Smav uint64_t len; 792264274Smav 793264274Smav dev_data = &be_lun->backend.dev; 794264274Smav io = beio->io; 795264274Smav 796264274Smav DPRINTF("entered\n"); 797264274Smav 798264274Smav binuptime(&beio->ds_t0); 799264274Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 800264274Smav 801264274Smav if (beio->io_offset == -1) { 802264274Smav beio->io_len = 0; 803264274Smav memcpy(&ptrlen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 804264274Smav sizeof(ptrlen)); 805264274Smav buf = (struct scsi_unmap_desc *)ptrlen.ptr; 806264274Smav end = buf + ptrlen.len / sizeof(*buf); 807264274Smav for (; buf < end; buf++) { 808264274Smav len = (uint64_t)scsi_4btoul(buf->length) * 809264274Smav be_lun->blocksize; 810264274Smav beio->io_len += len; 811264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 812264274Smav scsi_8btou64(buf->lba) * be_lun->blocksize, len, 813264283Smav (end - buf < 2) ? TRUE : FALSE); 814264274Smav } 815264274Smav } else 816264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 817264274Smav beio->io_offset, beio->io_len, TRUE); 818264274Smav} 819264274Smav 820264274Smavstatic void 821229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 822229997Sken struct ctl_be_block_io *beio) 823229997Sken{ 824229997Sken int i; 825229997Sken struct bio *bio; 826229997Sken struct ctl_be_block_devdata *dev_data; 827229997Sken off_t cur_offset; 828229997Sken int max_iosize; 829229997Sken 830229997Sken DPRINTF("entered\n"); 831229997Sken 832229997Sken dev_data = &be_lun->backend.dev; 833229997Sken 834229997Sken /* 835229997Sken * We have to limit our I/O size to the maximum supported by the 836229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 837229997Sken * set it properly, use DFLTPHYS. 838229997Sken */ 839229997Sken max_iosize = dev_data->cdev->si_iosize_max; 840229997Sken if (max_iosize < PAGE_SIZE) 841229997Sken max_iosize = DFLTPHYS; 842229997Sken 843229997Sken cur_offset = beio->io_offset; 844229997Sken 845229997Sken /* 846229997Sken * XXX KDM need to accurately reflect the number of I/Os outstanding 847229997Sken * to a device. 848229997Sken */ 849229997Sken binuptime(&beio->ds_t0); 850229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 851229997Sken 852229997Sken for (i = 0; i < beio->num_segs; i++) { 853229997Sken size_t cur_size; 854229997Sken uint8_t *cur_ptr; 855229997Sken 856229997Sken cur_size = beio->sg_segs[i].len; 857229997Sken cur_ptr = beio->sg_segs[i].addr; 858229997Sken 859229997Sken while (cur_size > 0) { 860229997Sken /* This can't fail, it's a blocking allocation. */ 861229997Sken bio = g_alloc_bio(); 862229997Sken 863229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 864229997Sken 865229997Sken bio->bio_cmd = beio->bio_cmd; 866229997Sken bio->bio_flags |= beio->bio_flags; 867229997Sken bio->bio_dev = dev_data->cdev; 868229997Sken bio->bio_caller1 = beio; 869229997Sken bio->bio_length = min(cur_size, max_iosize); 870229997Sken bio->bio_offset = cur_offset; 871229997Sken bio->bio_data = cur_ptr; 872229997Sken bio->bio_done = ctl_be_block_biodone; 873229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 874229997Sken 875229997Sken cur_offset += bio->bio_length; 876229997Sken cur_ptr += bio->bio_length; 877229997Sken cur_size -= bio->bio_length; 878229997Sken 879229997Sken /* 880229997Sken * Make sure we set the complete bit just before we 881229997Sken * issue the last bio so we don't wind up with a 882229997Sken * race. 883229997Sken * 884229997Sken * Use the LUN mutex here instead of a combination 885229997Sken * of atomic variables for simplicity. 886229997Sken * 887229997Sken * XXX KDM we could have a per-IO lock, but that 888229997Sken * would cause additional per-IO setup and teardown 889229997Sken * overhead. Hopefully there won't be too much 890229997Sken * contention on the LUN lock. 891229997Sken */ 892229997Sken mtx_lock(&be_lun->lock); 893229997Sken 894229997Sken beio->num_bios_sent++; 895229997Sken 896229997Sken if ((i == beio->num_segs - 1) 897229997Sken && (cur_size == 0)) 898229997Sken beio->send_complete = 1; 899229997Sken 900229997Sken mtx_unlock(&be_lun->lock); 901229997Sken 902229997Sken (*dev_data->csw->d_strategy)(bio); 903229997Sken } 904229997Sken } 905229997Sken} 906229997Sken 907229997Skenstatic void 908264274Smavctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 909264274Smav{ 910264274Smav union ctl_io *io; 911264274Smav 912264274Smav io = beio->io; 913264274Smav ctl_free_beio(beio); 914264274Smav if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) 915264274Smav && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 916264274Smav ctl_config_write_done(io); 917264274Smav return; 918264274Smav } 919264274Smav 920264274Smav ctl_be_block_config_write(io); 921264274Smav} 922264274Smav 923264274Smavstatic void 924264274Smavctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 925264274Smav union ctl_io *io) 926264274Smav{ 927264274Smav struct ctl_be_block_io *beio; 928264274Smav struct ctl_be_block_softc *softc; 929264274Smav struct ctl_lba_len_flags lbalen; 930264274Smav uint64_t len_left, lba; 931264274Smav int i, seglen; 932264274Smav uint8_t *buf, *end; 933264274Smav 934264274Smav DPRINTF("entered\n"); 935264274Smav 936264274Smav beio = io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 937264274Smav softc = be_lun->softc; 938264274Smav memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 939264274Smav sizeof(lbalen)); 940264274Smav 941264274Smav if (lbalen.flags & ~(SWS_LBDATA | SWS_UNMAP) || 942264274Smav (lbalen.flags & SWS_UNMAP && be_lun->unmap == NULL)) { 943264274Smav ctl_free_beio(beio); 944264274Smav ctl_set_invalid_field(&io->scsiio, 945264274Smav /*sks_valid*/ 1, 946264274Smav /*command*/ 1, 947264274Smav /*field*/ 1, 948264274Smav /*bit_valid*/ 0, 949264274Smav /*bit*/ 0); 950264274Smav ctl_config_write_done(io); 951264274Smav return; 952264274Smav } 953264274Smav 954264274Smav /* 955264274Smav * If the I/O came down with an ordered or head of queue tag, set 956264274Smav * the BIO_ORDERED attribute. For head of queue tags, that's 957264274Smav * pretty much the best we can do. 958264274Smav */ 959264274Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 960264274Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 961264274Smav beio->bio_flags = BIO_ORDERED; 962264274Smav 963264274Smav switch (io->scsiio.tag_type) { 964264274Smav case CTL_TAG_ORDERED: 965264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 966264274Smav break; 967264274Smav case CTL_TAG_HEAD_OF_QUEUE: 968264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 969264274Smav break; 970264274Smav case CTL_TAG_UNTAGGED: 971264274Smav case CTL_TAG_SIMPLE: 972264274Smav case CTL_TAG_ACA: 973264274Smav default: 974264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 975264274Smav break; 976264274Smav } 977264274Smav 978264274Smav if (lbalen.flags & SWS_UNMAP) { 979264274Smav beio->io_offset = lbalen.lba * be_lun->blocksize; 980264274Smav beio->io_len = (uint64_t)lbalen.len * be_lun->blocksize; 981264274Smav beio->bio_cmd = BIO_DELETE; 982264274Smav beio->ds_trans_type = DEVSTAT_FREE; 983264274Smav 984264274Smav be_lun->unmap(be_lun, beio); 985264274Smav return; 986264274Smav } 987264274Smav 988264274Smav beio->bio_cmd = BIO_WRITE; 989264274Smav beio->ds_trans_type = DEVSTAT_WRITE; 990264274Smav 991264274Smav DPRINTF("WRITE SAME at LBA %jx len %u\n", 992264274Smav (uintmax_t)lbalen.lba, lbalen.len); 993264274Smav 994264274Smav len_left = (uint64_t)lbalen.len * be_lun->blocksize; 995264274Smav for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 996264274Smav 997264274Smav /* 998264274Smav * Setup the S/G entry for this chunk. 999264274Smav */ 1000264886Smav seglen = MIN(CTLBLK_MAX_SEG, len_left); 1001264274Smav seglen -= seglen % be_lun->blocksize; 1002264274Smav beio->sg_segs[i].len = seglen; 1003264274Smav beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1004264274Smav 1005264274Smav DPRINTF("segment %d addr %p len %zd\n", i, 1006264274Smav beio->sg_segs[i].addr, beio->sg_segs[i].len); 1007264274Smav 1008264274Smav beio->num_segs++; 1009264274Smav len_left -= seglen; 1010264274Smav 1011264274Smav buf = beio->sg_segs[i].addr; 1012264274Smav end = buf + seglen; 1013264274Smav for (; buf < end; buf += be_lun->blocksize) { 1014264274Smav memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1015264274Smav if (lbalen.flags & SWS_LBDATA) 1016264274Smav scsi_ulto4b(lbalen.lba + lba, buf); 1017264274Smav lba++; 1018264274Smav } 1019264274Smav } 1020264274Smav 1021264274Smav beio->io_offset = lbalen.lba * be_lun->blocksize; 1022264274Smav beio->io_len = lba * be_lun->blocksize; 1023264274Smav 1024264274Smav /* We can not do all in one run. Correct and schedule rerun. */ 1025264274Smav if (len_left > 0) { 1026264274Smav lbalen.lba += lba; 1027264274Smav lbalen.len -= lba; 1028264274Smav memcpy(io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen, 1029264274Smav sizeof(lbalen)); 1030264274Smav beio->beio_cont = ctl_be_block_cw_done_ws; 1031264274Smav } 1032264274Smav 1033264274Smav be_lun->dispatch(be_lun, beio); 1034264274Smav} 1035264274Smav 1036264274Smavstatic void 1037264274Smavctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1038264274Smav union ctl_io *io) 1039264274Smav{ 1040264274Smav struct ctl_be_block_io *beio; 1041264274Smav struct ctl_be_block_softc *softc; 1042264274Smav struct ctl_ptr_len_flags ptrlen; 1043264274Smav 1044264274Smav DPRINTF("entered\n"); 1045264274Smav 1046264274Smav beio = io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1047264274Smav softc = be_lun->softc; 1048264274Smav memcpy(&ptrlen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 1049264274Smav sizeof(ptrlen)); 1050264274Smav 1051264274Smav if (ptrlen.flags != 0 || be_lun->unmap == NULL) { 1052264274Smav ctl_free_beio(beio); 1053264274Smav ctl_set_invalid_field(&io->scsiio, 1054264274Smav /*sks_valid*/ 0, 1055264274Smav /*command*/ 1, 1056264274Smav /*field*/ 0, 1057264274Smav /*bit_valid*/ 0, 1058264274Smav /*bit*/ 0); 1059264274Smav ctl_config_write_done(io); 1060264274Smav return; 1061264274Smav } 1062264274Smav 1063264274Smav /* 1064264274Smav * If the I/O came down with an ordered or head of queue tag, set 1065264274Smav * the BIO_ORDERED attribute. For head of queue tags, that's 1066264274Smav * pretty much the best we can do. 1067264274Smav */ 1068264274Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1069264274Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1070264274Smav beio->bio_flags = BIO_ORDERED; 1071264274Smav 1072264274Smav switch (io->scsiio.tag_type) { 1073264274Smav case CTL_TAG_ORDERED: 1074264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1075264274Smav break; 1076264274Smav case CTL_TAG_HEAD_OF_QUEUE: 1077264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1078264274Smav break; 1079264274Smav case CTL_TAG_UNTAGGED: 1080264274Smav case CTL_TAG_SIMPLE: 1081264274Smav case CTL_TAG_ACA: 1082264274Smav default: 1083264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1084264274Smav break; 1085264274Smav } 1086264274Smav 1087264274Smav beio->io_len = 0; 1088264274Smav beio->io_offset = -1; 1089264274Smav 1090264274Smav beio->bio_cmd = BIO_DELETE; 1091264274Smav beio->ds_trans_type = DEVSTAT_FREE; 1092264274Smav 1093264274Smav DPRINTF("WRITE SAME at LBA %jx len %u\n", 1094264274Smav (uintmax_t)lbalen.lba, lbalen.len); 1095264274Smav 1096264274Smav be_lun->unmap(be_lun, beio); 1097264274Smav} 1098264274Smav 1099264274Smavstatic void 1100264274Smavctl_be_block_cw_done(struct ctl_be_block_io *beio) 1101264274Smav{ 1102264274Smav union ctl_io *io; 1103264274Smav 1104264274Smav io = beio->io; 1105264274Smav ctl_free_beio(beio); 1106264274Smav ctl_config_write_done(io); 1107264274Smav} 1108264274Smav 1109264274Smavstatic void 1110229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1111229997Sken union ctl_io *io) 1112229997Sken{ 1113229997Sken struct ctl_be_block_io *beio; 1114229997Sken struct ctl_be_block_softc *softc; 1115229997Sken 1116229997Sken DPRINTF("entered\n"); 1117229997Sken 1118229997Sken softc = be_lun->softc; 1119229997Sken beio = ctl_alloc_beio(softc); 1120229997Sken beio->io = io; 1121229997Sken beio->lun = be_lun; 1122264274Smav beio->beio_cont = ctl_be_block_cw_done; 1123229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 1124229997Sken 1125229997Sken switch (io->scsiio.cdb[0]) { 1126229997Sken case SYNCHRONIZE_CACHE: 1127229997Sken case SYNCHRONIZE_CACHE_16: 1128249194Strasz beio->bio_cmd = BIO_FLUSH; 1129229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 1130229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1131229997Sken beio->io_len = 0; 1132229997Sken be_lun->lun_flush(be_lun, beio); 1133229997Sken break; 1134264274Smav case WRITE_SAME_10: 1135264274Smav case WRITE_SAME_16: 1136264274Smav ctl_be_block_cw_dispatch_ws(be_lun, io); 1137264274Smav break; 1138264274Smav case UNMAP: 1139264274Smav ctl_be_block_cw_dispatch_unmap(be_lun, io); 1140264274Smav break; 1141229997Sken default: 1142229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1143229997Sken break; 1144229997Sken } 1145229997Sken} 1146229997Sken 1147258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1148258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1149258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1150258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1151229997Sken 1152229997Skenstatic void 1153264886Smavctl_be_block_next(struct ctl_be_block_io *beio) 1154264886Smav{ 1155264886Smav struct ctl_be_block_lun *be_lun; 1156264886Smav union ctl_io *io; 1157264886Smav 1158264886Smav io = beio->io; 1159264886Smav be_lun = beio->lun; 1160264886Smav ctl_free_beio(beio); 1161264886Smav if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) 1162264886Smav && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1163264886Smav ctl_done(io); 1164264886Smav return; 1165264886Smav } 1166264886Smav 1167264886Smav io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 1168264886Smav io->io_hdr.status &= ~CTL_STATUS_MASK; 1169264886Smav io->io_hdr.status |= CTL_STATUS_NONE; 1170264886Smav 1171264886Smav mtx_lock(&be_lun->lock); 1172264886Smav /* 1173264886Smav * XXX KDM make sure that links is okay to use at this point. 1174264886Smav * Otherwise, we either need to add another field to ctl_io_hdr, 1175264886Smav * or deal with resource allocation here. 1176264886Smav */ 1177264886Smav STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1178264886Smav mtx_unlock(&be_lun->lock); 1179264886Smav 1180264886Smav taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1181264886Smav} 1182264886Smav 1183264886Smavstatic void 1184229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1185229997Sken union ctl_io *io) 1186229997Sken{ 1187229997Sken struct ctl_be_block_io *beio; 1188229997Sken struct ctl_be_block_softc *softc; 1189229997Sken struct ctl_lba_len lbalen; 1190264886Smav uint64_t len_left, lbaoff; 1191229997Sken int i; 1192229997Sken 1193229997Sken softc = be_lun->softc; 1194229997Sken 1195229997Sken DPRINTF("entered\n"); 1196229997Sken 1197229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 1198229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1199229997Sken } else { 1200229997Sken SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1201229997Sken } 1202229997Sken 1203229997Sken beio = ctl_alloc_beio(softc); 1204229997Sken beio->io = io; 1205229997Sken beio->lun = be_lun; 1206229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 1207229997Sken 1208229997Sken /* 1209229997Sken * If the I/O came down with an ordered or head of queue tag, set 1210229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 1211229997Sken * pretty much the best we can do. 1212229997Sken * 1213229997Sken * XXX KDM we don't have a great way to easily know about the FUA 1214229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 1215229997Sken * pass that knowledge to the backend), and in any case we would 1216229997Sken * need to determine how to handle it. 1217229997Sken */ 1218229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1219229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1220229997Sken beio->bio_flags = BIO_ORDERED; 1221229997Sken 1222229997Sken switch (io->scsiio.tag_type) { 1223229997Sken case CTL_TAG_ORDERED: 1224229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1225229997Sken break; 1226229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1227229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1228229997Sken break; 1229229997Sken case CTL_TAG_UNTAGGED: 1230229997Sken case CTL_TAG_SIMPLE: 1231229997Sken case CTL_TAG_ACA: 1232229997Sken default: 1233229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1234229997Sken break; 1235229997Sken } 1236229997Sken 1237229997Sken /* 1238229997Sken * This path handles read and write only. The config write path 1239229997Sken * handles flush operations. 1240229997Sken */ 1241229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 1242229997Sken beio->bio_cmd = BIO_READ; 1243229997Sken beio->ds_trans_type = DEVSTAT_READ; 1244229997Sken } else { 1245229997Sken beio->bio_cmd = BIO_WRITE; 1246229997Sken beio->ds_trans_type = DEVSTAT_WRITE; 1247229997Sken } 1248229997Sken 1249264886Smav memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 1250264886Smav sizeof(lbalen)); 1251264886Smav DPRINTF("%s at LBA %jx len %u @%ju\n", 1252229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1253264886Smav (uintmax_t)lbalen.lba, lbalen.len, lbaoff); 1254264886Smav lbaoff = io->scsiio.kern_rel_offset / be_lun->blocksize; 1255264886Smav beio->io_offset = (lbalen.lba + lbaoff) * be_lun->blocksize; 1256264886Smav beio->io_len = MIN((lbalen.len - lbaoff) * be_lun->blocksize, 1257264886Smav CTLBLK_MAX_IO_SIZE); 1258264886Smav beio->io_len -= beio->io_len % be_lun->blocksize; 1259229997Sken 1260264886Smav for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1261264886Smav KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1262264886Smav i, CTLBLK_MAX_SEGS)); 1263229997Sken 1264229997Sken /* 1265229997Sken * Setup the S/G entry for this chunk. 1266229997Sken */ 1267264886Smav beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1268229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1269229997Sken 1270229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1271229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1272229997Sken 1273229997Sken beio->num_segs++; 1274229997Sken len_left -= beio->sg_segs[i].len; 1275229997Sken } 1276264886Smav if (io->scsiio.kern_rel_offset + beio->io_len < 1277264886Smav io->scsiio.kern_total_len) 1278264886Smav beio->beio_cont = ctl_be_block_next; 1279264886Smav io->scsiio.be_move_done = ctl_be_block_move_done; 1280264886Smav io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1281264886Smav io->scsiio.kern_data_len = beio->io_len; 1282264886Smav io->scsiio.kern_data_resid = 0; 1283264886Smav io->scsiio.kern_sg_entries = beio->num_segs; 1284264886Smav io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1285229997Sken 1286229997Sken /* 1287229997Sken * For the read case, we need to read the data into our buffers and 1288229997Sken * then we can send it back to the user. For the write case, we 1289229997Sken * need to get the data from the user first. 1290229997Sken */ 1291229997Sken if (beio->bio_cmd == BIO_READ) { 1292229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1293229997Sken be_lun->dispatch(be_lun, beio); 1294229997Sken } else { 1295229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1296229997Sken#ifdef CTL_TIME_IO 1297229997Sken getbintime(&io->io_hdr.dma_start_bt); 1298229997Sken#endif 1299229997Sken ctl_datamove(io); 1300229997Sken } 1301229997Sken} 1302229997Sken 1303229997Skenstatic void 1304229997Skenctl_be_block_worker(void *context, int pending) 1305229997Sken{ 1306229997Sken struct ctl_be_block_lun *be_lun; 1307229997Sken struct ctl_be_block_softc *softc; 1308229997Sken union ctl_io *io; 1309229997Sken 1310229997Sken be_lun = (struct ctl_be_block_lun *)context; 1311229997Sken softc = be_lun->softc; 1312229997Sken 1313229997Sken DPRINTF("entered\n"); 1314229997Sken 1315229997Sken mtx_lock(&be_lun->lock); 1316229997Sken for (;;) { 1317229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1318229997Sken if (io != NULL) { 1319229997Sken struct ctl_be_block_io *beio; 1320229997Sken 1321229997Sken DPRINTF("datamove queue\n"); 1322229997Sken 1323229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1324229997Sken ctl_io_hdr, links); 1325229997Sken 1326229997Sken mtx_unlock(&be_lun->lock); 1327229997Sken 1328229997Sken beio = (struct ctl_be_block_io *) 1329229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1330229997Sken 1331229997Sken be_lun->dispatch(be_lun, beio); 1332229997Sken 1333229997Sken mtx_lock(&be_lun->lock); 1334229997Sken continue; 1335229997Sken } 1336229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1337229997Sken if (io != NULL) { 1338229997Sken 1339229997Sken DPRINTF("config write queue\n"); 1340229997Sken 1341229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1342229997Sken ctl_io_hdr, links); 1343229997Sken 1344229997Sken mtx_unlock(&be_lun->lock); 1345229997Sken 1346229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1347229997Sken 1348229997Sken mtx_lock(&be_lun->lock); 1349229997Sken continue; 1350229997Sken } 1351229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1352229997Sken if (io != NULL) { 1353229997Sken DPRINTF("input queue\n"); 1354229997Sken 1355229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1356229997Sken ctl_io_hdr, links); 1357229997Sken mtx_unlock(&be_lun->lock); 1358229997Sken 1359229997Sken /* 1360229997Sken * We must drop the lock, since this routine and 1361229997Sken * its children may sleep. 1362229997Sken */ 1363229997Sken ctl_be_block_dispatch(be_lun, io); 1364229997Sken 1365229997Sken mtx_lock(&be_lun->lock); 1366229997Sken continue; 1367229997Sken } 1368229997Sken 1369229997Sken /* 1370229997Sken * If we get here, there is no work left in the queues, so 1371229997Sken * just break out and let the task queue go to sleep. 1372229997Sken */ 1373229997Sken break; 1374229997Sken } 1375229997Sken mtx_unlock(&be_lun->lock); 1376229997Sken} 1377229997Sken 1378229997Sken/* 1379229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1380229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1381229997Sken * thread. 1382229997Sken */ 1383229997Skenstatic int 1384229997Skenctl_be_block_submit(union ctl_io *io) 1385229997Sken{ 1386264886Smav struct ctl_lba_len lbalen; 1387229997Sken struct ctl_be_block_lun *be_lun; 1388229997Sken struct ctl_be_lun *ctl_be_lun; 1389229997Sken int retval; 1390229997Sken 1391229997Sken DPRINTF("entered\n"); 1392229997Sken 1393229997Sken retval = CTL_RETVAL_COMPLETE; 1394229997Sken 1395229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1396229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1397229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1398229997Sken 1399229997Sken /* 1400229997Sken * Make sure we only get SCSI I/O. 1401229997Sken */ 1402229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1403229997Sken "%#x) encountered", io->io_hdr.io_type)); 1404229997Sken 1405264886Smav memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 1406264886Smav sizeof(lbalen)); 1407264886Smav io->scsiio.kern_total_len = lbalen.len * be_lun->blocksize; 1408264886Smav io->scsiio.kern_rel_offset = 0; 1409264886Smav 1410229997Sken mtx_lock(&be_lun->lock); 1411229997Sken /* 1412229997Sken * XXX KDM make sure that links is okay to use at this point. 1413229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1414229997Sken * or deal with resource allocation here. 1415229997Sken */ 1416229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1417229997Sken mtx_unlock(&be_lun->lock); 1418229997Sken 1419229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1420229997Sken 1421229997Sken return (retval); 1422229997Sken} 1423229997Sken 1424229997Skenstatic int 1425229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1426229997Sken int flag, struct thread *td) 1427229997Sken{ 1428229997Sken struct ctl_be_block_softc *softc; 1429229997Sken int error; 1430229997Sken 1431229997Sken softc = &backend_block_softc; 1432229997Sken 1433229997Sken error = 0; 1434229997Sken 1435229997Sken switch (cmd) { 1436229997Sken case CTL_LUN_REQ: { 1437229997Sken struct ctl_lun_req *lun_req; 1438229997Sken 1439229997Sken lun_req = (struct ctl_lun_req *)addr; 1440229997Sken 1441229997Sken switch (lun_req->reqtype) { 1442229997Sken case CTL_LUNREQ_CREATE: 1443229997Sken error = ctl_be_block_create(softc, lun_req); 1444229997Sken break; 1445229997Sken case CTL_LUNREQ_RM: 1446229997Sken error = ctl_be_block_rm(softc, lun_req); 1447229997Sken break; 1448232604Strasz case CTL_LUNREQ_MODIFY: 1449232604Strasz error = ctl_be_block_modify(softc, lun_req); 1450232604Strasz break; 1451229997Sken default: 1452229997Sken lun_req->status = CTL_LUN_ERROR; 1453229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1454229997Sken "%s: invalid LUN request type %d", __func__, 1455229997Sken lun_req->reqtype); 1456229997Sken break; 1457229997Sken } 1458229997Sken break; 1459229997Sken } 1460229997Sken default: 1461229997Sken error = ENOTTY; 1462229997Sken break; 1463229997Sken } 1464229997Sken 1465229997Sken return (error); 1466229997Sken} 1467229997Sken 1468229997Skenstatic int 1469229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1470229997Sken{ 1471229997Sken struct ctl_be_block_filedata *file_data; 1472229997Sken struct ctl_lun_create_params *params; 1473229997Sken struct vattr vattr; 1474229997Sken int error; 1475229997Sken 1476229997Sken error = 0; 1477229997Sken file_data = &be_lun->backend.file; 1478229997Sken params = &req->reqdata.create; 1479229997Sken 1480229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1481229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1482229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1483229997Sken 1484229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1485229997Sken if (error != 0) { 1486229997Sken snprintf(req->error_str, sizeof(req->error_str), 1487229997Sken "error calling VOP_GETATTR() for file %s", 1488229997Sken be_lun->dev_path); 1489229997Sken return (error); 1490229997Sken } 1491229997Sken 1492229997Sken /* 1493229997Sken * Verify that we have the ability to upgrade to exclusive 1494229997Sken * access on this file so we can trap errors at open instead 1495229997Sken * of reporting them during first access. 1496229997Sken */ 1497229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1498229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1499229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1500229997Sken error = EBADF; 1501229997Sken snprintf(req->error_str, sizeof(req->error_str), 1502229997Sken "error locking file %s", be_lun->dev_path); 1503229997Sken return (error); 1504229997Sken } 1505229997Sken } 1506229997Sken 1507229997Sken 1508229997Sken file_data->cred = crhold(curthread->td_ucred); 1509232604Strasz if (params->lun_size_bytes != 0) 1510232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1511232604Strasz else 1512232604Strasz be_lun->size_bytes = vattr.va_size; 1513229997Sken /* 1514229997Sken * We set the multi thread flag for file operations because all 1515229997Sken * filesystems (in theory) are capable of allowing multiple readers 1516229997Sken * of a file at once. So we want to get the maximum possible 1517229997Sken * concurrency. 1518229997Sken */ 1519229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1520229997Sken 1521229997Sken /* 1522229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1523229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1524229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1525229997Sken * may not work with other OSes as well. So just export a sector 1526229997Sken * size of 512 bytes, which should work with any OS or 1527229997Sken * application. Since our backing is a file, any block size will 1528229997Sken * work fine for the backing store. 1529229997Sken */ 1530229997Sken#if 0 1531229997Sken be_lun->blocksize= vattr.va_blocksize; 1532229997Sken#endif 1533229997Sken if (params->blocksize_bytes != 0) 1534229997Sken be_lun->blocksize = params->blocksize_bytes; 1535229997Sken else 1536229997Sken be_lun->blocksize = 512; 1537229997Sken 1538229997Sken /* 1539229997Sken * Sanity check. The media size has to be at least one 1540229997Sken * sector long. 1541229997Sken */ 1542229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1543229997Sken error = EINVAL; 1544229997Sken snprintf(req->error_str, sizeof(req->error_str), 1545229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1546229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1547229997Sken } 1548229997Sken return (error); 1549229997Sken} 1550229997Sken 1551229997Skenstatic int 1552229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1553229997Sken{ 1554229997Sken struct ctl_lun_create_params *params; 1555229997Sken struct vattr vattr; 1556229997Sken struct cdev *dev; 1557229997Sken struct cdevsw *devsw; 1558229997Sken int error; 1559264191Smav off_t ps, pss, po, pos; 1560229997Sken 1561229997Sken params = &req->reqdata.create; 1562229997Sken 1563229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1564229997Sken be_lun->dispatch = ctl_be_block_dispatch_dev; 1565229997Sken be_lun->lun_flush = ctl_be_block_flush_dev; 1566264274Smav be_lun->unmap = ctl_be_block_unmap_dev; 1567229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1568229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1569229997Sken &be_lun->backend.dev.dev_ref); 1570229997Sken if (be_lun->backend.dev.csw == NULL) 1571229997Sken panic("Unable to retrieve device switch"); 1572229997Sken 1573229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1574229997Sken if (error) { 1575229997Sken snprintf(req->error_str, sizeof(req->error_str), 1576229997Sken "%s: error getting vnode attributes for device %s", 1577229997Sken __func__, be_lun->dev_path); 1578229997Sken return (error); 1579229997Sken } 1580229997Sken 1581229997Sken dev = be_lun->vn->v_rdev; 1582229997Sken devsw = dev->si_devsw; 1583229997Sken if (!devsw->d_ioctl) { 1584229997Sken snprintf(req->error_str, sizeof(req->error_str), 1585229997Sken "%s: no d_ioctl for device %s!", __func__, 1586229997Sken be_lun->dev_path); 1587229997Sken return (ENODEV); 1588229997Sken } 1589229997Sken 1590229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1591229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1592229997Sken curthread); 1593229997Sken if (error) { 1594229997Sken snprintf(req->error_str, sizeof(req->error_str), 1595229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1596229997Sken "on %s!", __func__, error, be_lun->dev_path); 1597229997Sken return (error); 1598229997Sken } 1599229997Sken 1600229997Sken /* 1601229997Sken * If the user has asked for a blocksize that is greater than the 1602229997Sken * backing device's blocksize, we can do it only if the blocksize 1603229997Sken * the user is asking for is an even multiple of the underlying 1604229997Sken * device's blocksize. 1605229997Sken */ 1606229997Sken if ((params->blocksize_bytes != 0) 1607229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1608229997Sken uint32_t bs_multiple, tmp_blocksize; 1609229997Sken 1610229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1611229997Sken 1612229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1613229997Sken 1614229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1615229997Sken be_lun->blocksize = params->blocksize_bytes; 1616229997Sken } else { 1617229997Sken snprintf(req->error_str, sizeof(req->error_str), 1618229997Sken "%s: requested blocksize %u is not an even " 1619229997Sken "multiple of backing device blocksize %u", 1620229997Sken __func__, params->blocksize_bytes, 1621229997Sken be_lun->blocksize); 1622229997Sken return (EINVAL); 1623229997Sken 1624229997Sken } 1625229997Sken } else if ((params->blocksize_bytes != 0) 1626229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1627229997Sken snprintf(req->error_str, sizeof(req->error_str), 1628229997Sken "%s: requested blocksize %u < backing device " 1629229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1630229997Sken be_lun->blocksize); 1631229997Sken return (EINVAL); 1632229997Sken } 1633229997Sken 1634229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1635229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1636229997Sken curthread); 1637229997Sken if (error) { 1638229997Sken snprintf(req->error_str, sizeof(req->error_str), 1639232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1640232604Strasz " ioctl on %s!", __func__, error, 1641232604Strasz be_lun->dev_path); 1642229997Sken return (error); 1643229997Sken } 1644229997Sken 1645232604Strasz if (params->lun_size_bytes != 0) { 1646232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1647232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1648232604Strasz "%s: requested LUN size %ju > backing device " 1649232604Strasz "size %ju", __func__, 1650232604Strasz (uintmax_t)params->lun_size_bytes, 1651232604Strasz (uintmax_t)be_lun->size_bytes); 1652232604Strasz return (EINVAL); 1653232604Strasz } 1654232604Strasz 1655232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1656232604Strasz } 1657232604Strasz 1658264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1659264191Smav (caddr_t)&ps, FREAD, curthread); 1660264191Smav if (error) 1661264191Smav ps = po = 0; 1662264191Smav else { 1663264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1664264191Smav (caddr_t)&po, FREAD, curthread); 1665264191Smav if (error) 1666264191Smav po = 0; 1667264191Smav } 1668264191Smav pss = ps / be_lun->blocksize; 1669264191Smav pos = po / be_lun->blocksize; 1670264191Smav if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1671264191Smav ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1672264191Smav be_lun->pblockexp = fls(pss) - 1; 1673264191Smav be_lun->pblockoff = (pss - pos) % pss; 1674264191Smav } 1675264191Smav 1676229997Sken return (0); 1677229997Sken} 1678229997Sken 1679229997Skenstatic int 1680229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1681229997Sken{ 1682229997Sken DROP_GIANT(); 1683229997Sken if (be_lun->vn) { 1684229997Sken int flags = FREAD | FWRITE; 1685229997Sken 1686229997Sken switch (be_lun->dev_type) { 1687229997Sken case CTL_BE_BLOCK_DEV: 1688229997Sken if (be_lun->backend.dev.csw) { 1689229997Sken dev_relthread(be_lun->backend.dev.cdev, 1690229997Sken be_lun->backend.dev.dev_ref); 1691229997Sken be_lun->backend.dev.csw = NULL; 1692229997Sken be_lun->backend.dev.cdev = NULL; 1693229997Sken } 1694229997Sken break; 1695229997Sken case CTL_BE_BLOCK_FILE: 1696229997Sken break; 1697229997Sken case CTL_BE_BLOCK_NONE: 1698258871Strasz break; 1699229997Sken default: 1700229997Sken panic("Unexpected backend type."); 1701229997Sken break; 1702229997Sken } 1703229997Sken 1704229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1705229997Sken be_lun->vn = NULL; 1706229997Sken 1707229997Sken switch (be_lun->dev_type) { 1708229997Sken case CTL_BE_BLOCK_DEV: 1709229997Sken break; 1710229997Sken case CTL_BE_BLOCK_FILE: 1711229997Sken if (be_lun->backend.file.cred != NULL) { 1712229997Sken crfree(be_lun->backend.file.cred); 1713229997Sken be_lun->backend.file.cred = NULL; 1714229997Sken } 1715229997Sken break; 1716229997Sken case CTL_BE_BLOCK_NONE: 1717258871Strasz break; 1718229997Sken default: 1719229997Sken panic("Unexpected backend type."); 1720229997Sken break; 1721229997Sken } 1722229997Sken } 1723229997Sken PICKUP_GIANT(); 1724229997Sken 1725229997Sken return (0); 1726229997Sken} 1727229997Sken 1728229997Skenstatic int 1729229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1730229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1731229997Sken{ 1732229997Sken struct nameidata nd; 1733229997Sken int flags; 1734229997Sken int error; 1735229997Sken 1736229997Sken /* 1737229997Sken * XXX KDM allow a read-only option? 1738229997Sken */ 1739229997Sken flags = FREAD | FWRITE; 1740229997Sken error = 0; 1741229997Sken 1742229997Sken if (rootvnode == NULL) { 1743229997Sken snprintf(req->error_str, sizeof(req->error_str), 1744229997Sken "%s: Root filesystem is not mounted", __func__); 1745229997Sken return (1); 1746229997Sken } 1747229997Sken 1748229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1749229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1750229997Sken VREF(rootvnode); 1751229997Sken } 1752229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1753229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1754229997Sken VREF(rootvnode); 1755229997Sken } 1756229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1757229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1758229997Sken VREF(rootvnode); 1759229997Sken } 1760229997Sken 1761229997Sken again: 1762229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1763229997Sken error = vn_open(&nd, &flags, 0, NULL); 1764229997Sken if (error) { 1765229997Sken /* 1766229997Sken * This is the only reasonable guess we can make as far as 1767229997Sken * path if the user doesn't give us a fully qualified path. 1768229997Sken * If they want to specify a file, they need to specify the 1769229997Sken * full path. 1770229997Sken */ 1771229997Sken if (be_lun->dev_path[0] != '/') { 1772229997Sken char *dev_path = "/dev/"; 1773229997Sken char *dev_name; 1774229997Sken 1775229997Sken /* Try adding device path at beginning of name */ 1776229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1777229997Sken + strlen(dev_path) + 1, 1778229997Sken M_CTLBLK, M_WAITOK); 1779229997Sken if (dev_name) { 1780229997Sken sprintf(dev_name, "%s%s", dev_path, 1781229997Sken be_lun->dev_path); 1782229997Sken free(be_lun->dev_path, M_CTLBLK); 1783229997Sken be_lun->dev_path = dev_name; 1784229997Sken goto again; 1785229997Sken } 1786229997Sken } 1787229997Sken snprintf(req->error_str, sizeof(req->error_str), 1788229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1789229997Sken return (error); 1790229997Sken } 1791229997Sken 1792229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1793229997Sken 1794229997Sken be_lun->vn = nd.ni_vp; 1795229997Sken 1796229997Sken /* We only support disks and files. */ 1797229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1798229997Sken error = ctl_be_block_open_dev(be_lun, req); 1799229997Sken } else if (be_lun->vn->v_type == VREG) { 1800229997Sken error = ctl_be_block_open_file(be_lun, req); 1801229997Sken } else { 1802229997Sken error = EINVAL; 1803229997Sken snprintf(req->error_str, sizeof(req->error_str), 1804258871Strasz "%s is not a disk or plain file", be_lun->dev_path); 1805229997Sken } 1806229997Sken VOP_UNLOCK(be_lun->vn, 0); 1807229997Sken 1808229997Sken if (error != 0) { 1809229997Sken ctl_be_block_close(be_lun); 1810229997Sken return (error); 1811229997Sken } 1812229997Sken 1813229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1814229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1815229997Sken 1816229997Sken return (0); 1817229997Sken} 1818229997Sken 1819229997Skenstatic int 1820229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1821229997Sken{ 1822229997Sken struct ctl_be_block_lun *be_lun; 1823229997Sken struct ctl_lun_create_params *params; 1824229997Sken struct ctl_be_arg *file_arg; 1825229997Sken char tmpstr[32]; 1826264274Smav int retval, num_threads, unmap; 1827229997Sken int i; 1828229997Sken 1829229997Sken params = &req->reqdata.create; 1830229997Sken retval = 0; 1831229997Sken 1832229997Sken num_threads = cbb_num_threads; 1833229997Sken 1834229997Sken file_arg = NULL; 1835229997Sken 1836229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1837229997Sken 1838229997Sken be_lun->softc = softc; 1839229997Sken STAILQ_INIT(&be_lun->input_queue); 1840229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1841229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1842254759Strasz STAILQ_INIT(&be_lun->ctl_be_lun.options); 1843229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1844229997Sken mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF); 1845229997Sken 1846264886Smav be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1847256995Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1848229997Sken 1849229997Sken if (be_lun->lun_zone == NULL) { 1850229997Sken snprintf(req->error_str, sizeof(req->error_str), 1851229997Sken "%s: error allocating UMA zone", __func__); 1852229997Sken goto bailout_error; 1853229997Sken } 1854229997Sken 1855229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1856229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1857229997Sken else 1858229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1859229997Sken 1860229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1861229997Sken for (i = 0; i < req->num_be_args; i++) { 1862249026Strasz if (strcmp(req->kern_be_args[i].kname, "file") == 0) { 1863229997Sken file_arg = &req->kern_be_args[i]; 1864229997Sken break; 1865229997Sken } 1866229997Sken } 1867229997Sken 1868229997Sken if (file_arg == NULL) { 1869229997Sken snprintf(req->error_str, sizeof(req->error_str), 1870229997Sken "%s: no file argument specified", __func__); 1871229997Sken goto bailout_error; 1872229997Sken } 1873229997Sken 1874229997Sken be_lun->dev_path = malloc(file_arg->vallen, M_CTLBLK, 1875229997Sken M_WAITOK | M_ZERO); 1876229997Sken 1877249026Strasz strlcpy(be_lun->dev_path, (char *)file_arg->kvalue, 1878229997Sken file_arg->vallen); 1879229997Sken 1880229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1881229997Sken if (retval != 0) { 1882229997Sken retval = 0; 1883229997Sken goto bailout_error; 1884229997Sken } 1885229997Sken 1886229997Sken /* 1887229997Sken * Tell the user the size of the file/device. 1888229997Sken */ 1889229997Sken params->lun_size_bytes = be_lun->size_bytes; 1890229997Sken 1891229997Sken /* 1892229997Sken * The maximum LBA is the size - 1. 1893229997Sken */ 1894229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1895229997Sken } else { 1896229997Sken /* 1897229997Sken * For processor devices, we don't have any size. 1898229997Sken */ 1899229997Sken be_lun->blocksize = 0; 1900264191Smav be_lun->pblockexp = 0; 1901264191Smav be_lun->pblockoff = 0; 1902229997Sken be_lun->size_blocks = 0; 1903229997Sken be_lun->size_bytes = 0; 1904229997Sken be_lun->ctl_be_lun.maxlba = 0; 1905229997Sken params->lun_size_bytes = 0; 1906229997Sken 1907229997Sken /* 1908229997Sken * Default to just 1 thread for processor devices. 1909229997Sken */ 1910229997Sken num_threads = 1; 1911229997Sken } 1912229997Sken 1913229997Sken /* 1914229997Sken * XXX This searching loop might be refactored to be combined with 1915229997Sken * the loop above, 1916229997Sken */ 1917264274Smav unmap = 0; 1918229997Sken for (i = 0; i < req->num_be_args; i++) { 1919249026Strasz if (strcmp(req->kern_be_args[i].kname, "num_threads") == 0) { 1920229997Sken struct ctl_be_arg *thread_arg; 1921229997Sken char num_thread_str[16]; 1922229997Sken int tmp_num_threads; 1923229997Sken 1924229997Sken 1925229997Sken thread_arg = &req->kern_be_args[i]; 1926229997Sken 1927249026Strasz strlcpy(num_thread_str, (char *)thread_arg->kvalue, 1928229997Sken min(thread_arg->vallen, 1929229997Sken sizeof(num_thread_str))); 1930229997Sken 1931229997Sken tmp_num_threads = strtol(num_thread_str, NULL, 0); 1932229997Sken 1933229997Sken /* 1934229997Sken * We don't let the user specify less than one 1935229997Sken * thread, but hope he's clueful enough not to 1936229997Sken * specify 1000 threads. 1937229997Sken */ 1938229997Sken if (tmp_num_threads < 1) { 1939229997Sken snprintf(req->error_str, sizeof(req->error_str), 1940229997Sken "%s: invalid number of threads %s", 1941229997Sken __func__, num_thread_str); 1942229997Sken goto bailout_error; 1943229997Sken } 1944229997Sken 1945229997Sken num_threads = tmp_num_threads; 1946264274Smav } else if (strcmp(req->kern_be_args[i].kname, "unmap") == 0 && 1947264274Smav strcmp(req->kern_be_args[i].kvalue, "on") == 0) { 1948264274Smav unmap = 1; 1949254759Strasz } else if (strcmp(req->kern_be_args[i].kname, "file") != 0 && 1950254759Strasz strcmp(req->kern_be_args[i].kname, "dev") != 0) { 1951254759Strasz struct ctl_be_lun_option *opt; 1952254759Strasz 1953254759Strasz opt = malloc(sizeof(*opt), M_CTLBLK, M_WAITOK); 1954254759Strasz opt->name = malloc(strlen(req->kern_be_args[i].kname) + 1, M_CTLBLK, M_WAITOK); 1955254759Strasz strcpy(opt->name, req->kern_be_args[i].kname); 1956254759Strasz opt->value = malloc(strlen(req->kern_be_args[i].kvalue) + 1, M_CTLBLK, M_WAITOK); 1957254759Strasz strcpy(opt->value, req->kern_be_args[i].kvalue); 1958254759Strasz STAILQ_INSERT_TAIL(&be_lun->ctl_be_lun.options, opt, links); 1959229997Sken } 1960229997Sken } 1961229997Sken 1962229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1963229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1964264274Smav if (unmap) 1965264274Smav be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 1966229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 1967229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1968264191Smav be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 1969264191Smav be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 1970229997Sken /* Tell the user the blocksize we ended up using */ 1971229997Sken params->blocksize_bytes = be_lun->blocksize; 1972229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1973229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1974229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1975229997Sken } else 1976229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 1977229997Sken 1978229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1979229997Sken be_lun->ctl_be_lun.lun_config_status = 1980229997Sken ctl_be_block_lun_config_status; 1981229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1982229997Sken 1983229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1984229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1985229997Sken softc->num_luns); 1986229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1987229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1988229997Sken sizeof(tmpstr))); 1989229997Sken 1990229997Sken /* Tell the user what we used for a serial number */ 1991229997Sken strncpy((char *)params->serial_num, tmpstr, 1992229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1993229997Sken } else { 1994229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 1995229997Sken params->serial_num, 1996229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1997229997Sken sizeof(params->serial_num))); 1998229997Sken } 1999229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2000229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2001229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2002229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2003229997Sken sizeof(tmpstr))); 2004229997Sken 2005229997Sken /* Tell the user what we used for a device ID */ 2006229997Sken strncpy((char *)params->device_id, tmpstr, 2007229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 2008229997Sken } else { 2009229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 2010229997Sken params->device_id, 2011229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2012229997Sken sizeof(params->device_id))); 2013229997Sken } 2014229997Sken 2015229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2016229997Sken 2017229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2018229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2019229997Sken 2020229997Sken if (be_lun->io_taskqueue == NULL) { 2021229997Sken snprintf(req->error_str, sizeof(req->error_str), 2022229997Sken "%s: Unable to create taskqueue", __func__); 2023229997Sken goto bailout_error; 2024229997Sken } 2025229997Sken 2026229997Sken /* 2027229997Sken * Note that we start the same number of threads by default for 2028229997Sken * both the file case and the block device case. For the file 2029229997Sken * case, we need multiple threads to allow concurrency, because the 2030229997Sken * vnode interface is designed to be a blocking interface. For the 2031229997Sken * block device case, ZFS zvols at least will block the caller's 2032229997Sken * context in many instances, and so we need multiple threads to 2033229997Sken * overcome that problem. Other block devices don't need as many 2034229997Sken * threads, but they shouldn't cause too many problems. 2035229997Sken * 2036229997Sken * If the user wants to just have a single thread for a block 2037229997Sken * device, he can specify that when the LUN is created, or change 2038229997Sken * the tunable/sysctl to alter the default number of threads. 2039229997Sken */ 2040229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2041229997Sken /*num threads*/num_threads, 2042229997Sken /*priority*/PWAIT, 2043229997Sken /*thread name*/ 2044229997Sken "%s taskq", be_lun->lunname); 2045229997Sken 2046229997Sken if (retval != 0) 2047229997Sken goto bailout_error; 2048229997Sken 2049229997Sken be_lun->num_threads = num_threads; 2050229997Sken 2051229997Sken mtx_lock(&softc->lock); 2052229997Sken softc->num_luns++; 2053229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2054229997Sken 2055229997Sken mtx_unlock(&softc->lock); 2056229997Sken 2057229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 2058229997Sken if (retval != 0) { 2059229997Sken mtx_lock(&softc->lock); 2060229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2061229997Sken links); 2062229997Sken softc->num_luns--; 2063229997Sken mtx_unlock(&softc->lock); 2064229997Sken snprintf(req->error_str, sizeof(req->error_str), 2065229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 2066229997Sken "details", __func__, retval); 2067229997Sken retval = 0; 2068229997Sken goto bailout_error; 2069229997Sken } 2070229997Sken 2071229997Sken mtx_lock(&softc->lock); 2072229997Sken 2073229997Sken /* 2074229997Sken * Tell the config_status routine that we're waiting so it won't 2075229997Sken * clean up the LUN in the event of an error. 2076229997Sken */ 2077229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2078229997Sken 2079229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2080229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2081229997Sken if (retval == EINTR) 2082229997Sken break; 2083229997Sken } 2084229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2085229997Sken 2086229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2087229997Sken snprintf(req->error_str, sizeof(req->error_str), 2088229997Sken "%s: LUN configuration error, see dmesg for details", 2089229997Sken __func__); 2090229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2091229997Sken links); 2092229997Sken softc->num_luns--; 2093229997Sken mtx_unlock(&softc->lock); 2094229997Sken goto bailout_error; 2095229997Sken } else { 2096229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2097229997Sken } 2098229997Sken 2099229997Sken mtx_unlock(&softc->lock); 2100229997Sken 2101229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2102229997Sken be_lun->blocksize, 2103229997Sken DEVSTAT_ALL_SUPPORTED, 2104229997Sken be_lun->ctl_be_lun.lun_type 2105229997Sken | DEVSTAT_TYPE_IF_OTHER, 2106229997Sken DEVSTAT_PRIORITY_OTHER); 2107229997Sken 2108229997Sken 2109229997Sken req->status = CTL_LUN_OK; 2110229997Sken 2111229997Sken return (retval); 2112229997Sken 2113229997Skenbailout_error: 2114229997Sken req->status = CTL_LUN_ERROR; 2115229997Sken 2116229997Sken ctl_be_block_close(be_lun); 2117229997Sken 2118229997Sken free(be_lun->dev_path, M_CTLBLK); 2119229997Sken free(be_lun, M_CTLBLK); 2120229997Sken 2121229997Sken return (retval); 2122229997Sken} 2123229997Sken 2124229997Skenstatic int 2125229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2126229997Sken{ 2127229997Sken struct ctl_lun_rm_params *params; 2128229997Sken struct ctl_be_block_lun *be_lun; 2129229997Sken int retval; 2130229997Sken 2131229997Sken params = &req->reqdata.rm; 2132229997Sken 2133229997Sken mtx_lock(&softc->lock); 2134229997Sken 2135229997Sken be_lun = NULL; 2136229997Sken 2137229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2138229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2139229997Sken break; 2140229997Sken } 2141229997Sken mtx_unlock(&softc->lock); 2142229997Sken 2143229997Sken if (be_lun == NULL) { 2144229997Sken snprintf(req->error_str, sizeof(req->error_str), 2145229997Sken "%s: LUN %u is not managed by the block backend", 2146229997Sken __func__, params->lun_id); 2147229997Sken goto bailout_error; 2148229997Sken } 2149229997Sken 2150229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2151229997Sken 2152229997Sken if (retval != 0) { 2153229997Sken snprintf(req->error_str, sizeof(req->error_str), 2154229997Sken "%s: error %d returned from ctl_disable_lun() for " 2155229997Sken "LUN %d", __func__, retval, params->lun_id); 2156229997Sken goto bailout_error; 2157229997Sken 2158229997Sken } 2159229997Sken 2160229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2161229997Sken if (retval != 0) { 2162229997Sken snprintf(req->error_str, sizeof(req->error_str), 2163229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 2164229997Sken "LUN %d", __func__, retval, params->lun_id); 2165229997Sken goto bailout_error; 2166229997Sken } 2167229997Sken 2168229997Sken mtx_lock(&softc->lock); 2169229997Sken 2170229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2171229997Sken 2172229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2173229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2174229997Sken if (retval == EINTR) 2175229997Sken break; 2176229997Sken } 2177229997Sken 2178229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2179229997Sken 2180229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2181229997Sken snprintf(req->error_str, sizeof(req->error_str), 2182229997Sken "%s: interrupted waiting for LUN to be freed", 2183229997Sken __func__); 2184229997Sken mtx_unlock(&softc->lock); 2185229997Sken goto bailout_error; 2186229997Sken } 2187229997Sken 2188229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2189229997Sken 2190229997Sken softc->num_luns--; 2191229997Sken mtx_unlock(&softc->lock); 2192229997Sken 2193229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2194229997Sken 2195229997Sken taskqueue_free(be_lun->io_taskqueue); 2196229997Sken 2197229997Sken ctl_be_block_close(be_lun); 2198229997Sken 2199229997Sken if (be_lun->disk_stats != NULL) 2200229997Sken devstat_remove_entry(be_lun->disk_stats); 2201229997Sken 2202229997Sken uma_zdestroy(be_lun->lun_zone); 2203229997Sken 2204229997Sken free(be_lun->dev_path, M_CTLBLK); 2205229997Sken 2206229997Sken free(be_lun, M_CTLBLK); 2207229997Sken 2208229997Sken req->status = CTL_LUN_OK; 2209229997Sken 2210229997Sken return (0); 2211229997Sken 2212229997Skenbailout_error: 2213229997Sken 2214229997Sken req->status = CTL_LUN_ERROR; 2215229997Sken 2216229997Sken return (0); 2217229997Sken} 2218229997Sken 2219232604Straszstatic int 2220232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2221232604Strasz struct ctl_lun_req *req) 2222232604Strasz{ 2223232604Strasz struct vattr vattr; 2224232604Strasz int error; 2225232604Strasz struct ctl_lun_modify_params *params; 2226232604Strasz 2227232604Strasz params = &req->reqdata.modify; 2228232604Strasz 2229232604Strasz if (params->lun_size_bytes != 0) { 2230232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2231232604Strasz } else { 2232232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2233232604Strasz if (error != 0) { 2234232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2235232604Strasz "error calling VOP_GETATTR() for file %s", 2236232604Strasz be_lun->dev_path); 2237232604Strasz return (error); 2238232604Strasz } 2239232604Strasz 2240232604Strasz be_lun->size_bytes = vattr.va_size; 2241232604Strasz } 2242232604Strasz 2243232604Strasz return (0); 2244232604Strasz} 2245232604Strasz 2246232604Straszstatic int 2247232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2248232604Strasz struct ctl_lun_req *req) 2249232604Strasz{ 2250232604Strasz struct cdev *dev; 2251232604Strasz struct cdevsw *devsw; 2252232604Strasz int error; 2253232604Strasz struct ctl_lun_modify_params *params; 2254232604Strasz uint64_t size_bytes; 2255232604Strasz 2256232604Strasz params = &req->reqdata.modify; 2257232604Strasz 2258232604Strasz dev = be_lun->vn->v_rdev; 2259232604Strasz devsw = dev->si_devsw; 2260232604Strasz if (!devsw->d_ioctl) { 2261232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2262232604Strasz "%s: no d_ioctl for device %s!", __func__, 2263232604Strasz be_lun->dev_path); 2264232604Strasz return (ENODEV); 2265232604Strasz } 2266232604Strasz 2267232604Strasz error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2268232604Strasz (caddr_t)&size_bytes, FREAD, 2269232604Strasz curthread); 2270232604Strasz if (error) { 2271232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2272232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2273232604Strasz "on %s!", __func__, error, be_lun->dev_path); 2274232604Strasz return (error); 2275232604Strasz } 2276232604Strasz 2277232604Strasz if (params->lun_size_bytes != 0) { 2278232604Strasz if (params->lun_size_bytes > size_bytes) { 2279232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2280232604Strasz "%s: requested LUN size %ju > backing device " 2281232604Strasz "size %ju", __func__, 2282232604Strasz (uintmax_t)params->lun_size_bytes, 2283232604Strasz (uintmax_t)size_bytes); 2284232604Strasz return (EINVAL); 2285232604Strasz } 2286232604Strasz 2287232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2288232604Strasz } else { 2289232604Strasz be_lun->size_bytes = size_bytes; 2290232604Strasz } 2291232604Strasz 2292232604Strasz return (0); 2293232604Strasz} 2294232604Strasz 2295232604Straszstatic int 2296232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2297232604Strasz{ 2298232604Strasz struct ctl_lun_modify_params *params; 2299232604Strasz struct ctl_be_block_lun *be_lun; 2300241896Skib int error; 2301232604Strasz 2302232604Strasz params = &req->reqdata.modify; 2303232604Strasz 2304232604Strasz mtx_lock(&softc->lock); 2305232604Strasz 2306232604Strasz be_lun = NULL; 2307232604Strasz 2308232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2309232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2310232604Strasz break; 2311232604Strasz } 2312232604Strasz mtx_unlock(&softc->lock); 2313232604Strasz 2314232604Strasz if (be_lun == NULL) { 2315232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2316232604Strasz "%s: LUN %u is not managed by the block backend", 2317232604Strasz __func__, params->lun_id); 2318232604Strasz goto bailout_error; 2319232604Strasz } 2320232604Strasz 2321232604Strasz if (params->lun_size_bytes != 0) { 2322232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2323232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2324232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2325232604Strasz params->lun_size_bytes, be_lun->blocksize); 2326232604Strasz goto bailout_error; 2327232604Strasz } 2328232604Strasz } 2329232604Strasz 2330232604Strasz vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2331232604Strasz 2332232604Strasz if (be_lun->vn->v_type == VREG) 2333232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2334232604Strasz else 2335232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2336232604Strasz 2337232604Strasz VOP_UNLOCK(be_lun->vn, 0); 2338232604Strasz 2339232604Strasz if (error != 0) 2340232604Strasz goto bailout_error; 2341232604Strasz 2342232604Strasz be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2343232604Strasz 2344232604Strasz /* 2345232604Strasz * The maximum LBA is the size - 1. 2346232604Strasz * 2347232604Strasz * XXX: Note that this field is being updated without locking, 2348232604Strasz * which might cause problems on 32-bit architectures. 2349232604Strasz */ 2350232604Strasz be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2351232604Strasz ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2352232604Strasz 2353232604Strasz /* Tell the user the exact size we ended up using */ 2354232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2355232604Strasz 2356232604Strasz req->status = CTL_LUN_OK; 2357232604Strasz 2358232604Strasz return (0); 2359232604Strasz 2360232604Straszbailout_error: 2361232604Strasz req->status = CTL_LUN_ERROR; 2362232604Strasz 2363232604Strasz return (0); 2364232604Strasz} 2365232604Strasz 2366229997Skenstatic void 2367229997Skenctl_be_block_lun_shutdown(void *be_lun) 2368229997Sken{ 2369229997Sken struct ctl_be_block_lun *lun; 2370229997Sken struct ctl_be_block_softc *softc; 2371229997Sken 2372229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2373229997Sken 2374229997Sken softc = lun->softc; 2375229997Sken 2376229997Sken mtx_lock(&softc->lock); 2377229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2378229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2379229997Sken wakeup(lun); 2380229997Sken mtx_unlock(&softc->lock); 2381229997Sken 2382229997Sken} 2383229997Sken 2384229997Skenstatic void 2385229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2386229997Sken{ 2387229997Sken struct ctl_be_block_lun *lun; 2388229997Sken struct ctl_be_block_softc *softc; 2389229997Sken 2390229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2391229997Sken softc = lun->softc; 2392229997Sken 2393229997Sken if (status == CTL_LUN_CONFIG_OK) { 2394229997Sken mtx_lock(&softc->lock); 2395229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2396229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2397229997Sken wakeup(lun); 2398229997Sken mtx_unlock(&softc->lock); 2399229997Sken 2400229997Sken /* 2401229997Sken * We successfully added the LUN, attempt to enable it. 2402229997Sken */ 2403229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2404229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2405229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2406229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2407229997Sken __func__); 2408229997Sken } 2409229997Sken } 2410229997Sken 2411229997Sken return; 2412229997Sken } 2413229997Sken 2414229997Sken 2415229997Sken mtx_lock(&softc->lock); 2416229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2417229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2418229997Sken wakeup(lun); 2419229997Sken mtx_unlock(&softc->lock); 2420229997Sken} 2421229997Sken 2422229997Sken 2423229997Skenstatic int 2424229997Skenctl_be_block_config_write(union ctl_io *io) 2425229997Sken{ 2426229997Sken struct ctl_be_block_lun *be_lun; 2427229997Sken struct ctl_be_lun *ctl_be_lun; 2428229997Sken int retval; 2429229997Sken 2430229997Sken retval = 0; 2431229997Sken 2432229997Sken DPRINTF("entered\n"); 2433229997Sken 2434229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2435229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2436229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2437229997Sken 2438229997Sken switch (io->scsiio.cdb[0]) { 2439229997Sken case SYNCHRONIZE_CACHE: 2440229997Sken case SYNCHRONIZE_CACHE_16: 2441264274Smav case WRITE_SAME_10: 2442264274Smav case WRITE_SAME_16: 2443264274Smav case UNMAP: 2444229997Sken /* 2445229997Sken * The upper level CTL code will filter out any CDBs with 2446229997Sken * the immediate bit set and return the proper error. 2447229997Sken * 2448229997Sken * We don't really need to worry about what LBA range the 2449229997Sken * user asked to be synced out. When they issue a sync 2450229997Sken * cache command, we'll sync out the whole thing. 2451229997Sken */ 2452229997Sken mtx_lock(&be_lun->lock); 2453229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2454229997Sken links); 2455229997Sken mtx_unlock(&be_lun->lock); 2456229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2457229997Sken break; 2458229997Sken case START_STOP_UNIT: { 2459229997Sken struct scsi_start_stop_unit *cdb; 2460229997Sken 2461229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2462229997Sken 2463229997Sken if (cdb->how & SSS_START) 2464229997Sken retval = ctl_start_lun(ctl_be_lun); 2465229997Sken else { 2466229997Sken retval = ctl_stop_lun(ctl_be_lun); 2467229997Sken /* 2468229997Sken * XXX KDM Copan-specific offline behavior. 2469229997Sken * Figure out a reasonable way to port this? 2470229997Sken */ 2471229997Sken#ifdef NEEDTOPORT 2472229997Sken if ((retval == 0) 2473229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2474229997Sken retval = ctl_lun_offline(ctl_be_lun); 2475229997Sken#endif 2476229997Sken } 2477229997Sken 2478229997Sken /* 2479229997Sken * In general, the above routines should not fail. They 2480229997Sken * just set state for the LUN. So we've got something 2481229997Sken * pretty wrong here if we can't start or stop the LUN. 2482229997Sken */ 2483229997Sken if (retval != 0) { 2484229997Sken ctl_set_internal_failure(&io->scsiio, 2485229997Sken /*sks_valid*/ 1, 2486229997Sken /*retry_count*/ 0xf051); 2487229997Sken retval = CTL_RETVAL_COMPLETE; 2488229997Sken } else { 2489229997Sken ctl_set_success(&io->scsiio); 2490229997Sken } 2491229997Sken ctl_config_write_done(io); 2492229997Sken break; 2493229997Sken } 2494229997Sken default: 2495229997Sken ctl_set_invalid_opcode(&io->scsiio); 2496229997Sken ctl_config_write_done(io); 2497229997Sken retval = CTL_RETVAL_COMPLETE; 2498229997Sken break; 2499229997Sken } 2500229997Sken 2501229997Sken return (retval); 2502229997Sken 2503229997Sken} 2504229997Sken 2505229997Skenstatic int 2506229997Skenctl_be_block_config_read(union ctl_io *io) 2507229997Sken{ 2508229997Sken return (0); 2509229997Sken} 2510229997Sken 2511229997Skenstatic int 2512229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2513229997Sken{ 2514229997Sken struct ctl_be_block_lun *lun; 2515229997Sken int retval; 2516229997Sken 2517229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2518229997Sken retval = 0; 2519229997Sken 2520229997Sken retval = sbuf_printf(sb, "<num_threads>"); 2521229997Sken 2522229997Sken if (retval != 0) 2523229997Sken goto bailout; 2524229997Sken 2525229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2526229997Sken 2527229997Sken if (retval != 0) 2528229997Sken goto bailout; 2529229997Sken 2530229997Sken retval = sbuf_printf(sb, "</num_threads>"); 2531229997Sken 2532229997Sken /* 2533229997Sken * For processor devices, we don't have a path variable. 2534229997Sken */ 2535229997Sken if ((retval != 0) 2536229997Sken || (lun->dev_path == NULL)) 2537229997Sken goto bailout; 2538229997Sken 2539229997Sken retval = sbuf_printf(sb, "<file>"); 2540229997Sken 2541229997Sken if (retval != 0) 2542229997Sken goto bailout; 2543229997Sken 2544229997Sken retval = ctl_sbuf_printf_esc(sb, lun->dev_path); 2545229997Sken 2546229997Sken if (retval != 0) 2547229997Sken goto bailout; 2548229997Sken 2549229997Sken retval = sbuf_printf(sb, "</file>\n"); 2550229997Sken 2551229997Skenbailout: 2552229997Sken 2553229997Sken return (retval); 2554229997Sken} 2555229997Sken 2556229997Skenint 2557229997Skenctl_be_block_init(void) 2558229997Sken{ 2559229997Sken struct ctl_be_block_softc *softc; 2560229997Sken int retval; 2561229997Sken 2562229997Sken softc = &backend_block_softc; 2563229997Sken retval = 0; 2564229997Sken 2565229997Sken mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF); 2566264020Strasz beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2567264020Strasz NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2568229997Sken STAILQ_INIT(&softc->disk_list); 2569229997Sken STAILQ_INIT(&softc->lun_list); 2570229997Sken 2571229997Sken return (retval); 2572229997Sken} 2573