ctl_backend_block.c revision 268283
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 268283 2014-07-05 04:10:24Z 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 */ 95267537Smav#define CTLBLK_HALF_IO_SIZE (512 * 1024) 96267537Smav#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 97264886Smav#define CTLBLK_MAX_SEG MAXPHYS 98267537Smav#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 99267537Smav#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 100229997Sken 101229997Sken#ifdef CTLBLK_DEBUG 102229997Sken#define DPRINTF(fmt, args...) \ 103229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 104229997Sken#else 105229997Sken#define DPRINTF(fmt, args...) do {} while(0) 106229997Sken#endif 107229997Sken 108267519Smav#define PRIV(io) \ 109267519Smav ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 110267537Smav#define ARGS(io) \ 111267537Smav ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 112267519Smav 113229997SkenSDT_PROVIDER_DEFINE(cbb); 114229997Sken 115229997Skentypedef enum { 116229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 117229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 118229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 119229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 120229997Sken} ctl_be_block_lun_flags; 121229997Sken 122229997Skentypedef enum { 123229997Sken CTL_BE_BLOCK_NONE, 124229997Sken CTL_BE_BLOCK_DEV, 125229997Sken CTL_BE_BLOCK_FILE 126229997Sken} ctl_be_block_type; 127229997Sken 128229997Skenstruct ctl_be_block_devdata { 129229997Sken struct cdev *cdev; 130229997Sken struct cdevsw *csw; 131229997Sken int dev_ref; 132229997Sken}; 133229997Sken 134229997Skenstruct ctl_be_block_filedata { 135229997Sken struct ucred *cred; 136229997Sken}; 137229997Sken 138229997Skenunion ctl_be_block_bedata { 139229997Sken struct ctl_be_block_devdata dev; 140229997Sken struct ctl_be_block_filedata file; 141229997Sken}; 142229997Sken 143229997Skenstruct ctl_be_block_io; 144229997Skenstruct ctl_be_block_lun; 145229997Sken 146229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 147229997Sken struct ctl_be_block_io *beio); 148229997Sken 149229997Sken/* 150229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 151229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 152229997Sken */ 153229997Skenstruct ctl_be_block_lun { 154229997Sken struct ctl_block_disk *disk; 155229997Sken char lunname[32]; 156229997Sken char *dev_path; 157229997Sken ctl_be_block_type dev_type; 158229997Sken struct vnode *vn; 159229997Sken union ctl_be_block_bedata backend; 160229997Sken cbb_dispatch_t dispatch; 161229997Sken cbb_dispatch_t lun_flush; 162264274Smav cbb_dispatch_t unmap; 163229997Sken uma_zone_t lun_zone; 164229997Sken uint64_t size_blocks; 165229997Sken uint64_t size_bytes; 166229997Sken uint32_t blocksize; 167229997Sken int blocksize_shift; 168264191Smav uint16_t pblockexp; 169264191Smav uint16_t pblockoff; 170229997Sken struct ctl_be_block_softc *softc; 171229997Sken struct devstat *disk_stats; 172229997Sken ctl_be_block_lun_flags flags; 173229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 174229997Sken struct ctl_be_lun ctl_be_lun; 175229997Sken struct taskqueue *io_taskqueue; 176229997Sken struct task io_task; 177229997Sken int num_threads; 178229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 179229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 180229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 181267877Smav struct mtx_padalign io_lock; 182267877Smav struct mtx_padalign queue_lock; 183229997Sken}; 184229997Sken 185229997Sken/* 186229997Sken * Overall softc structure for the block backend module. 187229997Sken */ 188229997Skenstruct ctl_be_block_softc { 189229997Sken struct mtx lock; 190229997Sken int num_disks; 191229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 192229997Sken int num_luns; 193229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 194229997Sken}; 195229997Sken 196229997Skenstatic struct ctl_be_block_softc backend_block_softc; 197229997Sken 198229997Sken/* 199229997Sken * Per-I/O information. 200229997Sken */ 201229997Skenstruct ctl_be_block_io { 202229997Sken union ctl_io *io; 203229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 204229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 205229997Sken int bio_cmd; 206229997Sken int bio_flags; 207229997Sken int num_segs; 208229997Sken int num_bios_sent; 209229997Sken int num_bios_done; 210229997Sken int send_complete; 211229997Sken int num_errors; 212229997Sken struct bintime ds_t0; 213229997Sken devstat_tag_type ds_tag_type; 214229997Sken devstat_trans_flags ds_trans_type; 215229997Sken uint64_t io_len; 216229997Sken uint64_t io_offset; 217229997Sken struct ctl_be_block_softc *softc; 218229997Sken struct ctl_be_block_lun *lun; 219264274Smav void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 220229997Sken}; 221229997Sken 222229997Skenstatic int cbb_num_threads = 14; 223229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 224229997Sken "CAM Target Layer Block Backend"); 225267992ShselaskySYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 226229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 227229997Sken 228229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 229229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 230229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 231229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 232229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 233229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 234229997Sken struct ctl_be_block_io *beio); 235229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 236229997Sken struct ctl_be_block_io *beio); 237229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 238229997Sken struct ctl_be_block_io *beio); 239264274Smavstatic void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 240264274Smav struct ctl_be_block_io *beio); 241229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 242229997Sken struct ctl_be_block_io *beio); 243229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 244229997Sken union ctl_io *io); 245229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 246229997Sken union ctl_io *io); 247229997Skenstatic void ctl_be_block_worker(void *context, int pending); 248229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 249229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 250229997Sken int flag, struct thread *td); 251229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 252229997Sken struct ctl_lun_req *req); 253229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 254229997Sken struct ctl_lun_req *req); 255229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 256229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 257229997Sken struct ctl_be_block_lun *be_lun, 258229997Sken struct ctl_lun_req *req); 259229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 260229997Sken struct ctl_lun_req *req); 261229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 262229997Sken struct ctl_lun_req *req); 263232604Straszstatic int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 264232604Strasz struct ctl_lun_req *req); 265232604Straszstatic int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 266232604Strasz struct ctl_lun_req *req); 267232604Straszstatic int ctl_be_block_modify(struct ctl_be_block_softc *softc, 268232604Strasz struct ctl_lun_req *req); 269229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 270229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 271229997Sken ctl_lun_config_status status); 272229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 273229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 274229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 275229997Skenint ctl_be_block_init(void); 276229997Sken 277229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 278229997Sken{ 279230334Sken .name = "block", 280230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 281230334Sken .init = ctl_be_block_init, 282230334Sken .data_submit = ctl_be_block_submit, 283230334Sken .data_move_done = ctl_be_block_move_done, 284230334Sken .config_read = ctl_be_block_config_read, 285230334Sken .config_write = ctl_be_block_config_write, 286230334Sken .ioctl = ctl_be_block_ioctl, 287230334Sken .lun_info = ctl_be_block_lun_info 288229997Sken}; 289229997Sken 290229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 291229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 292229997Sken 293264020Straszstatic uma_zone_t beio_zone; 294264020Strasz 295229997Skenstatic struct ctl_be_block_io * 296229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 297229997Sken{ 298229997Sken struct ctl_be_block_io *beio; 299229997Sken 300264020Strasz beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 301264020Strasz beio->softc = softc; 302229997Sken return (beio); 303229997Sken} 304229997Sken 305229997Skenstatic void 306229997Skenctl_free_beio(struct ctl_be_block_io *beio) 307229997Sken{ 308229997Sken int duplicate_free; 309229997Sken int i; 310229997Sken 311229997Sken duplicate_free = 0; 312229997Sken 313229997Sken for (i = 0; i < beio->num_segs; i++) { 314229997Sken if (beio->sg_segs[i].addr == NULL) 315229997Sken duplicate_free++; 316229997Sken 317229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 318229997Sken beio->sg_segs[i].addr = NULL; 319267537Smav 320267537Smav /* For compare we had two equal S/G lists. */ 321267537Smav if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 322267537Smav uma_zfree(beio->lun->lun_zone, 323267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 324267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 325267537Smav } 326229997Sken } 327229997Sken 328229997Sken if (duplicate_free > 0) { 329229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 330229997Sken duplicate_free, beio->num_segs); 331229997Sken } 332229997Sken 333264020Strasz uma_zfree(beio_zone, beio); 334229997Sken} 335229997Sken 336229997Skenstatic void 337229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 338229997Sken{ 339267877Smav union ctl_io *io = beio->io; 340229997Sken 341264274Smav if (beio->beio_cont != NULL) { 342264274Smav beio->beio_cont(beio); 343264274Smav } else { 344264274Smav ctl_free_beio(beio); 345267537Smav ctl_data_submit_done(io); 346264274Smav } 347229997Sken} 348229997Sken 349229997Skenstatic int 350229997Skenctl_be_block_move_done(union ctl_io *io) 351229997Sken{ 352229997Sken struct ctl_be_block_io *beio; 353229997Sken struct ctl_be_block_lun *be_lun; 354267537Smav struct ctl_lba_len_flags *lbalen; 355229997Sken#ifdef CTL_TIME_IO 356229997Sken struct bintime cur_bt; 357267537Smav#endif 358267537Smav int i; 359229997Sken 360267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 361229997Sken be_lun = beio->lun; 362229997Sken 363229997Sken DPRINTF("entered\n"); 364229997Sken 365229997Sken#ifdef CTL_TIME_IO 366229997Sken getbintime(&cur_bt); 367229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 368229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 369229997Sken io->io_hdr.num_dmas++; 370229997Sken#endif 371267537Smav io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 372229997Sken 373229997Sken /* 374229997Sken * We set status at this point for read commands, and write 375229997Sken * commands with errors. 376229997Sken */ 377267537Smav if ((io->io_hdr.port_status == 0) && 378267537Smav ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 379267537Smav ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 380267537Smav lbalen = ARGS(beio->io); 381267537Smav if (lbalen->flags & CTL_LLF_READ) { 382267537Smav ctl_set_success(&io->scsiio); 383267537Smav } else if (lbalen->flags & CTL_LLF_COMPARE) { 384267537Smav /* We have two data blocks ready for comparison. */ 385267537Smav for (i = 0; i < beio->num_segs; i++) { 386267537Smav if (memcmp(beio->sg_segs[i].addr, 387267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 388267537Smav beio->sg_segs[i].len) != 0) 389267537Smav break; 390267537Smav } 391267537Smav if (i < beio->num_segs) 392267537Smav ctl_set_sense(&io->scsiio, 393267537Smav /*current_error*/ 1, 394267537Smav /*sense_key*/ SSD_KEY_MISCOMPARE, 395267537Smav /*asc*/ 0x1D, 396267537Smav /*ascq*/ 0x00, 397267537Smav SSD_ELEM_NONE); 398267537Smav else 399267537Smav ctl_set_success(&io->scsiio); 400267537Smav } 401267537Smav } 402229997Sken else if ((io->io_hdr.port_status != 0) 403229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 404229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 405229997Sken /* 406229997Sken * For hardware error sense keys, the sense key 407229997Sken * specific value is defined to be a retry count, 408229997Sken * but we use it to pass back an internal FETD 409229997Sken * error code. XXX KDM Hopefully the FETD is only 410229997Sken * using 16 bits for an error code, since that's 411229997Sken * all the space we have in the sks field. 412229997Sken */ 413229997Sken ctl_set_internal_failure(&io->scsiio, 414229997Sken /*sks_valid*/ 1, 415229997Sken /*retry_count*/ 416229997Sken io->io_hdr.port_status); 417229997Sken } 418229997Sken 419229997Sken /* 420229997Sken * If this is a read, or a write with errors, it is done. 421229997Sken */ 422229997Sken if ((beio->bio_cmd == BIO_READ) 423229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 424229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 425229997Sken ctl_complete_beio(beio); 426229997Sken return (0); 427229997Sken } 428229997Sken 429229997Sken /* 430229997Sken * At this point, we have a write and the DMA completed 431229997Sken * successfully. We now have to queue it to the task queue to 432229997Sken * execute the backend I/O. That is because we do blocking 433229997Sken * memory allocations, and in the file backing case, blocking I/O. 434229997Sken * This move done routine is generally called in the SIM's 435229997Sken * interrupt context, and therefore we cannot block. 436229997Sken */ 437267877Smav mtx_lock(&be_lun->queue_lock); 438229997Sken /* 439229997Sken * XXX KDM make sure that links is okay to use at this point. 440229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 441229997Sken * or deal with resource allocation here. 442229997Sken */ 443229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 444267877Smav mtx_unlock(&be_lun->queue_lock); 445229997Sken 446229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 447229997Sken 448229997Sken return (0); 449229997Sken} 450229997Sken 451229997Skenstatic void 452229997Skenctl_be_block_biodone(struct bio *bio) 453229997Sken{ 454229997Sken struct ctl_be_block_io *beio; 455229997Sken struct ctl_be_block_lun *be_lun; 456229997Sken union ctl_io *io; 457261538Smav int error; 458229997Sken 459229997Sken beio = bio->bio_caller1; 460229997Sken be_lun = beio->lun; 461229997Sken io = beio->io; 462229997Sken 463229997Sken DPRINTF("entered\n"); 464229997Sken 465261538Smav error = bio->bio_error; 466267877Smav mtx_lock(&be_lun->io_lock); 467261538Smav if (error != 0) 468229997Sken beio->num_errors++; 469229997Sken 470229997Sken beio->num_bios_done++; 471229997Sken 472229997Sken /* 473229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 474229997Sken * during the free might cause it to complain. 475229997Sken */ 476229997Sken g_destroy_bio(bio); 477229997Sken 478229997Sken /* 479229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 480229997Sken * complete, then we're done. 481229997Sken */ 482229997Sken if ((beio->send_complete == 0) 483229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 484267877Smav mtx_unlock(&be_lun->io_lock); 485229997Sken return; 486229997Sken } 487229997Sken 488229997Sken /* 489229997Sken * At this point, we've verified that we are the last I/O to 490229997Sken * complete, so it's safe to drop the lock. 491229997Sken */ 492267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 493267877Smav beio->ds_tag_type, beio->ds_trans_type, 494267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 495267877Smav mtx_unlock(&be_lun->io_lock); 496229997Sken 497229997Sken /* 498229997Sken * If there are any errors from the backing device, we fail the 499229997Sken * entire I/O with a medium error. 500229997Sken */ 501229997Sken if (beio->num_errors > 0) { 502261538Smav if (error == EOPNOTSUPP) { 503261538Smav ctl_set_invalid_opcode(&io->scsiio); 504261538Smav } else if (beio->bio_cmd == BIO_FLUSH) { 505229997Sken /* XXX KDM is there is a better error here? */ 506229997Sken ctl_set_internal_failure(&io->scsiio, 507229997Sken /*sks_valid*/ 1, 508229997Sken /*retry_count*/ 0xbad2); 509229997Sken } else 510229997Sken ctl_set_medium_error(&io->scsiio); 511229997Sken ctl_complete_beio(beio); 512229997Sken return; 513229997Sken } 514229997Sken 515229997Sken /* 516267537Smav * If this is a write, a flush, a delete or verify, we're all done. 517229997Sken * If this is a read, we can now send the data to the user. 518229997Sken */ 519229997Sken if ((beio->bio_cmd == BIO_WRITE) 520264274Smav || (beio->bio_cmd == BIO_FLUSH) 521267537Smav || (beio->bio_cmd == BIO_DELETE) 522267537Smav || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 523229997Sken ctl_set_success(&io->scsiio); 524229997Sken ctl_complete_beio(beio); 525229997Sken } else { 526229997Sken#ifdef CTL_TIME_IO 527229997Sken getbintime(&io->io_hdr.dma_start_bt); 528229997Sken#endif 529229997Sken ctl_datamove(io); 530229997Sken } 531229997Sken} 532229997Sken 533229997Skenstatic void 534229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 535229997Sken struct ctl_be_block_io *beio) 536229997Sken{ 537267877Smav union ctl_io *io = beio->io; 538229997Sken struct mount *mountpoint; 539241896Skib int error, lock_flags; 540229997Sken 541229997Sken DPRINTF("entered\n"); 542229997Sken 543267877Smav binuptime(&beio->ds_t0); 544267877Smav mtx_lock(&be_lun->io_lock); 545267877Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 546267877Smav mtx_unlock(&be_lun->io_lock); 547229997Sken 548267877Smav (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 549229997Sken 550229997Sken if (MNT_SHARED_WRITES(mountpoint) 551229997Sken || ((mountpoint == NULL) 552229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 553229997Sken lock_flags = LK_SHARED; 554229997Sken else 555229997Sken lock_flags = LK_EXCLUSIVE; 556229997Sken 557229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 558229997Sken 559229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 560229997Sken VOP_UNLOCK(be_lun->vn, 0); 561229997Sken 562229997Sken vn_finished_write(mountpoint); 563229997Sken 564267877Smav mtx_lock(&be_lun->io_lock); 565267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 566267877Smav beio->ds_tag_type, beio->ds_trans_type, 567267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 568267877Smav mtx_unlock(&be_lun->io_lock); 569267877Smav 570229997Sken if (error == 0) 571229997Sken ctl_set_success(&io->scsiio); 572229997Sken else { 573229997Sken /* XXX KDM is there is a better error here? */ 574229997Sken ctl_set_internal_failure(&io->scsiio, 575229997Sken /*sks_valid*/ 1, 576229997Sken /*retry_count*/ 0xbad1); 577229997Sken } 578229997Sken 579229997Sken ctl_complete_beio(beio); 580229997Sken} 581229997Sken 582258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 583258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 584258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 585258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 586229997Sken 587229997Skenstatic void 588229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 589229997Sken struct ctl_be_block_io *beio) 590229997Sken{ 591229997Sken struct ctl_be_block_filedata *file_data; 592229997Sken union ctl_io *io; 593229997Sken struct uio xuio; 594229997Sken struct iovec *xiovec; 595241896Skib int flags; 596229997Sken int error, i; 597229997Sken 598229997Sken DPRINTF("entered\n"); 599229997Sken 600229997Sken file_data = &be_lun->backend.file; 601229997Sken io = beio->io; 602229997Sken flags = beio->bio_flags; 603229997Sken 604267537Smav bzero(&xuio, sizeof(xuio)); 605229997Sken if (beio->bio_cmd == BIO_READ) { 606229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 607267537Smav xuio.uio_rw = UIO_READ; 608229997Sken } else { 609229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 610267537Smav xuio.uio_rw = UIO_WRITE; 611229997Sken } 612229997Sken xuio.uio_offset = beio->io_offset; 613229997Sken xuio.uio_resid = beio->io_len; 614229997Sken xuio.uio_segflg = UIO_SYSSPACE; 615229997Sken xuio.uio_iov = beio->xiovecs; 616229997Sken xuio.uio_iovcnt = beio->num_segs; 617229997Sken xuio.uio_td = curthread; 618229997Sken 619229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 620229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 621229997Sken xiovec->iov_len = beio->sg_segs[i].len; 622229997Sken } 623229997Sken 624267877Smav binuptime(&beio->ds_t0); 625267877Smav mtx_lock(&be_lun->io_lock); 626267877Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 627267877Smav mtx_unlock(&be_lun->io_lock); 628267877Smav 629229997Sken if (beio->bio_cmd == BIO_READ) { 630229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 631229997Sken 632229997Sken /* 633229997Sken * UFS pays attention to IO_DIRECT for reads. If the 634229997Sken * DIRECTIO option is configured into the kernel, it calls 635229997Sken * ffs_rawread(). But that only works for single-segment 636229997Sken * uios with user space addresses. In our case, with a 637229997Sken * kernel uio, it still reads into the buffer cache, but it 638229997Sken * will just try to release the buffer from the cache later 639229997Sken * on in ffs_read(). 640229997Sken * 641229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 642229997Sken * 643229997Sken * UFS does not pay attention to IO_SYNC for reads. 644229997Sken * 645229997Sken * ZFS pays attention to IO_SYNC (which translates into the 646229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 647229997Sken * attempts to sync the file before reading. 648229997Sken * 649229997Sken * So, to attempt to provide some barrier semantics in the 650229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 651229997Sken */ 652229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 653229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 654229997Sken 655229997Sken VOP_UNLOCK(be_lun->vn, 0); 656267537Smav SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 657229997Sken } else { 658229997Sken struct mount *mountpoint; 659229997Sken int lock_flags; 660229997Sken 661229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 662229997Sken 663229997Sken if (MNT_SHARED_WRITES(mountpoint) 664229997Sken || ((mountpoint == NULL) 665229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 666229997Sken lock_flags = LK_SHARED; 667229997Sken else 668229997Sken lock_flags = LK_EXCLUSIVE; 669229997Sken 670229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 671229997Sken 672229997Sken /* 673229997Sken * UFS pays attention to IO_DIRECT for writes. The write 674229997Sken * is done asynchronously. (Normally the write would just 675229997Sken * get put into cache. 676229997Sken * 677229997Sken * UFS pays attention to IO_SYNC for writes. It will 678229997Sken * attempt to write the buffer out synchronously if that 679229997Sken * flag is set. 680229997Sken * 681229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 682229997Sken * 683229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 684229997Sken * for writes. It will flush the transaction from the 685229997Sken * cache before returning. 686229997Sken * 687229997Sken * So if we've got the BIO_ORDERED flag set, we want 688229997Sken * IO_SYNC in either the UFS or ZFS case. 689229997Sken */ 690229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 691229997Sken IO_SYNC : 0, file_data->cred); 692229997Sken VOP_UNLOCK(be_lun->vn, 0); 693229997Sken 694229997Sken vn_finished_write(mountpoint); 695267537Smav SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 696229997Sken } 697229997Sken 698267877Smav mtx_lock(&be_lun->io_lock); 699267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 700267877Smav beio->ds_tag_type, beio->ds_trans_type, 701267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 702267877Smav mtx_unlock(&be_lun->io_lock); 703267877Smav 704229997Sken /* 705229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 706229997Sken * return the I/O to the user. 707229997Sken */ 708229997Sken if (error != 0) { 709229997Sken char path_str[32]; 710229997Sken 711229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 712229997Sken /* 713229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 714229997Sken * filesystem fills up. What kind of SCSI error should we 715229997Sken * return for that? 716229997Sken */ 717229997Sken printf("%s%s command returned errno %d\n", path_str, 718229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 719229997Sken ctl_set_medium_error(&io->scsiio); 720229997Sken ctl_complete_beio(beio); 721229997Sken return; 722229997Sken } 723229997Sken 724229997Sken /* 725229997Sken * If this is a write, we're all done. 726229997Sken * If this is a read, we can now send the data to the user. 727229997Sken */ 728267537Smav if (ARGS(io)->flags & (CTL_LLF_WRITE | CTL_LLF_VERIFY)) { 729229997Sken ctl_set_success(&io->scsiio); 730229997Sken ctl_complete_beio(beio); 731229997Sken } else { 732229997Sken#ifdef CTL_TIME_IO 733229997Sken getbintime(&io->io_hdr.dma_start_bt); 734229997Sken#endif 735229997Sken ctl_datamove(io); 736229997Sken } 737229997Sken} 738229997Sken 739229997Skenstatic void 740229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 741229997Sken struct ctl_be_block_io *beio) 742229997Sken{ 743229997Sken struct bio *bio; 744229997Sken union ctl_io *io; 745229997Sken struct ctl_be_block_devdata *dev_data; 746229997Sken 747229997Sken dev_data = &be_lun->backend.dev; 748229997Sken io = beio->io; 749229997Sken 750229997Sken DPRINTF("entered\n"); 751229997Sken 752229997Sken /* This can't fail, it's a blocking allocation. */ 753229997Sken bio = g_alloc_bio(); 754229997Sken 755229997Sken bio->bio_cmd = BIO_FLUSH; 756229997Sken bio->bio_flags |= BIO_ORDERED; 757229997Sken bio->bio_dev = dev_data->cdev; 758229997Sken bio->bio_offset = 0; 759229997Sken bio->bio_data = 0; 760229997Sken bio->bio_done = ctl_be_block_biodone; 761229997Sken bio->bio_caller1 = beio; 762229997Sken bio->bio_pblkno = 0; 763229997Sken 764229997Sken /* 765229997Sken * We don't need to acquire the LUN lock here, because we are only 766229997Sken * sending one bio, and so there is no other context to synchronize 767229997Sken * with. 768229997Sken */ 769229997Sken beio->num_bios_sent = 1; 770229997Sken beio->send_complete = 1; 771229997Sken 772229997Sken binuptime(&beio->ds_t0); 773267877Smav mtx_lock(&be_lun->io_lock); 774229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 775267877Smav mtx_unlock(&be_lun->io_lock); 776229997Sken 777229997Sken (*dev_data->csw->d_strategy)(bio); 778229997Sken} 779229997Sken 780229997Skenstatic void 781264274Smavctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 782264274Smav struct ctl_be_block_io *beio, 783264274Smav uint64_t off, uint64_t len, int last) 784264274Smav{ 785264274Smav struct bio *bio; 786264274Smav struct ctl_be_block_devdata *dev_data; 787264296Smav uint64_t maxlen; 788264274Smav 789264274Smav dev_data = &be_lun->backend.dev; 790264296Smav maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 791264274Smav while (len > 0) { 792264274Smav bio = g_alloc_bio(); 793264274Smav bio->bio_cmd = BIO_DELETE; 794264274Smav bio->bio_flags |= beio->bio_flags; 795264274Smav bio->bio_dev = dev_data->cdev; 796264274Smav bio->bio_offset = off; 797264296Smav bio->bio_length = MIN(len, maxlen); 798264274Smav bio->bio_data = 0; 799264274Smav bio->bio_done = ctl_be_block_biodone; 800264274Smav bio->bio_caller1 = beio; 801264296Smav bio->bio_pblkno = off / be_lun->blocksize; 802264274Smav 803264274Smav off += bio->bio_length; 804264274Smav len -= bio->bio_length; 805264274Smav 806267877Smav mtx_lock(&be_lun->io_lock); 807264274Smav beio->num_bios_sent++; 808264274Smav if (last && len == 0) 809264274Smav beio->send_complete = 1; 810267877Smav mtx_unlock(&be_lun->io_lock); 811264274Smav 812264274Smav (*dev_data->csw->d_strategy)(bio); 813264274Smav } 814264274Smav} 815264274Smav 816264274Smavstatic void 817264274Smavctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 818264274Smav struct ctl_be_block_io *beio) 819264274Smav{ 820264274Smav union ctl_io *io; 821264274Smav struct ctl_be_block_devdata *dev_data; 822267515Smav struct ctl_ptr_len_flags *ptrlen; 823264274Smav struct scsi_unmap_desc *buf, *end; 824264274Smav uint64_t len; 825264274Smav 826264274Smav dev_data = &be_lun->backend.dev; 827264274Smav io = beio->io; 828264274Smav 829264274Smav DPRINTF("entered\n"); 830264274Smav 831264274Smav binuptime(&beio->ds_t0); 832267877Smav mtx_lock(&be_lun->io_lock); 833264274Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 834267877Smav mtx_unlock(&be_lun->io_lock); 835264274Smav 836264274Smav if (beio->io_offset == -1) { 837264274Smav beio->io_len = 0; 838267515Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 839267515Smav buf = (struct scsi_unmap_desc *)ptrlen->ptr; 840267515Smav end = buf + ptrlen->len / sizeof(*buf); 841264274Smav for (; buf < end; buf++) { 842264274Smav len = (uint64_t)scsi_4btoul(buf->length) * 843264274Smav be_lun->blocksize; 844264274Smav beio->io_len += len; 845264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 846264274Smav scsi_8btou64(buf->lba) * be_lun->blocksize, len, 847264283Smav (end - buf < 2) ? TRUE : FALSE); 848264274Smav } 849264274Smav } else 850264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 851264274Smav beio->io_offset, beio->io_len, TRUE); 852264274Smav} 853264274Smav 854264274Smavstatic void 855229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 856229997Sken struct ctl_be_block_io *beio) 857229997Sken{ 858267877Smav TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 859229997Sken int i; 860229997Sken struct bio *bio; 861229997Sken struct ctl_be_block_devdata *dev_data; 862229997Sken off_t cur_offset; 863229997Sken int max_iosize; 864229997Sken 865229997Sken DPRINTF("entered\n"); 866229997Sken 867229997Sken dev_data = &be_lun->backend.dev; 868229997Sken 869229997Sken /* 870229997Sken * We have to limit our I/O size to the maximum supported by the 871229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 872229997Sken * set it properly, use DFLTPHYS. 873229997Sken */ 874229997Sken max_iosize = dev_data->cdev->si_iosize_max; 875229997Sken if (max_iosize < PAGE_SIZE) 876229997Sken max_iosize = DFLTPHYS; 877229997Sken 878229997Sken cur_offset = beio->io_offset; 879229997Sken for (i = 0; i < beio->num_segs; i++) { 880229997Sken size_t cur_size; 881229997Sken uint8_t *cur_ptr; 882229997Sken 883229997Sken cur_size = beio->sg_segs[i].len; 884229997Sken cur_ptr = beio->sg_segs[i].addr; 885229997Sken 886229997Sken while (cur_size > 0) { 887229997Sken /* This can't fail, it's a blocking allocation. */ 888229997Sken bio = g_alloc_bio(); 889229997Sken 890229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 891229997Sken 892229997Sken bio->bio_cmd = beio->bio_cmd; 893229997Sken bio->bio_flags |= beio->bio_flags; 894229997Sken bio->bio_dev = dev_data->cdev; 895229997Sken bio->bio_caller1 = beio; 896229997Sken bio->bio_length = min(cur_size, max_iosize); 897229997Sken bio->bio_offset = cur_offset; 898229997Sken bio->bio_data = cur_ptr; 899229997Sken bio->bio_done = ctl_be_block_biodone; 900229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 901229997Sken 902229997Sken cur_offset += bio->bio_length; 903229997Sken cur_ptr += bio->bio_length; 904229997Sken cur_size -= bio->bio_length; 905229997Sken 906267877Smav TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 907229997Sken beio->num_bios_sent++; 908229997Sken } 909229997Sken } 910267877Smav binuptime(&beio->ds_t0); 911267877Smav mtx_lock(&be_lun->io_lock); 912267877Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 913267877Smav beio->send_complete = 1; 914267877Smav mtx_unlock(&be_lun->io_lock); 915267877Smav 916267877Smav /* 917267877Smav * Fire off all allocated requests! 918267877Smav */ 919267877Smav while ((bio = TAILQ_FIRST(&queue)) != NULL) { 920267877Smav TAILQ_REMOVE(&queue, bio, bio_queue); 921267877Smav (*dev_data->csw->d_strategy)(bio); 922267877Smav } 923229997Sken} 924229997Sken 925229997Skenstatic void 926264274Smavctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 927264274Smav{ 928264274Smav union ctl_io *io; 929264274Smav 930264274Smav io = beio->io; 931264274Smav ctl_free_beio(beio); 932267641Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 933267641Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 934267641Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 935264274Smav ctl_config_write_done(io); 936264274Smav return; 937264274Smav } 938264274Smav 939264274Smav ctl_be_block_config_write(io); 940264274Smav} 941264274Smav 942264274Smavstatic void 943264274Smavctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 944264274Smav union ctl_io *io) 945264274Smav{ 946264274Smav struct ctl_be_block_io *beio; 947264274Smav struct ctl_be_block_softc *softc; 948267515Smav struct ctl_lba_len_flags *lbalen; 949264274Smav uint64_t len_left, lba; 950264274Smav int i, seglen; 951264274Smav uint8_t *buf, *end; 952264274Smav 953264274Smav DPRINTF("entered\n"); 954264274Smav 955267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 956264274Smav softc = be_lun->softc; 957267537Smav lbalen = ARGS(beio->io); 958264274Smav 959267515Smav if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP) || 960267515Smav (lbalen->flags & SWS_UNMAP && be_lun->unmap == NULL)) { 961264274Smav ctl_free_beio(beio); 962264274Smav ctl_set_invalid_field(&io->scsiio, 963264274Smav /*sks_valid*/ 1, 964264274Smav /*command*/ 1, 965264274Smav /*field*/ 1, 966264274Smav /*bit_valid*/ 0, 967264274Smav /*bit*/ 0); 968264274Smav ctl_config_write_done(io); 969264274Smav return; 970264274Smav } 971264274Smav 972264274Smav /* 973264274Smav * If the I/O came down with an ordered or head of queue tag, set 974264274Smav * the BIO_ORDERED attribute. For head of queue tags, that's 975264274Smav * pretty much the best we can do. 976264274Smav */ 977264274Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 978264274Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 979264274Smav beio->bio_flags = BIO_ORDERED; 980264274Smav 981264274Smav switch (io->scsiio.tag_type) { 982264274Smav case CTL_TAG_ORDERED: 983264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 984264274Smav break; 985264274Smav case CTL_TAG_HEAD_OF_QUEUE: 986264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 987264274Smav break; 988264274Smav case CTL_TAG_UNTAGGED: 989264274Smav case CTL_TAG_SIMPLE: 990264274Smav case CTL_TAG_ACA: 991264274Smav default: 992264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 993264274Smav break; 994264274Smav } 995264274Smav 996267515Smav if (lbalen->flags & SWS_UNMAP) { 997267515Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 998267515Smav beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 999264274Smav beio->bio_cmd = BIO_DELETE; 1000264274Smav beio->ds_trans_type = DEVSTAT_FREE; 1001264274Smav 1002264274Smav be_lun->unmap(be_lun, beio); 1003264274Smav return; 1004264274Smav } 1005264274Smav 1006264274Smav beio->bio_cmd = BIO_WRITE; 1007264274Smav beio->ds_trans_type = DEVSTAT_WRITE; 1008264274Smav 1009264274Smav DPRINTF("WRITE SAME at LBA %jx len %u\n", 1010267515Smav (uintmax_t)lbalen->lba, lbalen->len); 1011264274Smav 1012267515Smav len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1013264274Smav for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1014264274Smav 1015264274Smav /* 1016264274Smav * Setup the S/G entry for this chunk. 1017264274Smav */ 1018264886Smav seglen = MIN(CTLBLK_MAX_SEG, len_left); 1019264274Smav seglen -= seglen % be_lun->blocksize; 1020264274Smav beio->sg_segs[i].len = seglen; 1021264274Smav beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1022264274Smav 1023264274Smav DPRINTF("segment %d addr %p len %zd\n", i, 1024264274Smav beio->sg_segs[i].addr, beio->sg_segs[i].len); 1025264274Smav 1026264274Smav beio->num_segs++; 1027264274Smav len_left -= seglen; 1028264274Smav 1029264274Smav buf = beio->sg_segs[i].addr; 1030264274Smav end = buf + seglen; 1031264274Smav for (; buf < end; buf += be_lun->blocksize) { 1032264274Smav memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1033267515Smav if (lbalen->flags & SWS_LBDATA) 1034267515Smav scsi_ulto4b(lbalen->lba + lba, buf); 1035264274Smav lba++; 1036264274Smav } 1037264274Smav } 1038264274Smav 1039267515Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 1040264274Smav beio->io_len = lba * be_lun->blocksize; 1041264274Smav 1042264274Smav /* We can not do all in one run. Correct and schedule rerun. */ 1043264274Smav if (len_left > 0) { 1044267515Smav lbalen->lba += lba; 1045267515Smav lbalen->len -= lba; 1046264274Smav beio->beio_cont = ctl_be_block_cw_done_ws; 1047264274Smav } 1048264274Smav 1049264274Smav be_lun->dispatch(be_lun, beio); 1050264274Smav} 1051264274Smav 1052264274Smavstatic void 1053264274Smavctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1054264274Smav union ctl_io *io) 1055264274Smav{ 1056264274Smav struct ctl_be_block_io *beio; 1057264274Smav struct ctl_be_block_softc *softc; 1058267515Smav struct ctl_ptr_len_flags *ptrlen; 1059264274Smav 1060264274Smav DPRINTF("entered\n"); 1061264274Smav 1062267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1063264274Smav softc = be_lun->softc; 1064267515Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1065264274Smav 1066267515Smav if (ptrlen->flags != 0 || be_lun->unmap == NULL) { 1067264274Smav ctl_free_beio(beio); 1068264274Smav ctl_set_invalid_field(&io->scsiio, 1069264274Smav /*sks_valid*/ 0, 1070264274Smav /*command*/ 1, 1071264274Smav /*field*/ 0, 1072264274Smav /*bit_valid*/ 0, 1073264274Smav /*bit*/ 0); 1074264274Smav ctl_config_write_done(io); 1075264274Smav return; 1076264274Smav } 1077264274Smav 1078264274Smav /* 1079264274Smav * If the I/O came down with an ordered or head of queue tag, set 1080264274Smav * the BIO_ORDERED attribute. For head of queue tags, that's 1081264274Smav * pretty much the best we can do. 1082264274Smav */ 1083264274Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1084264274Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1085264274Smav beio->bio_flags = BIO_ORDERED; 1086264274Smav 1087264274Smav switch (io->scsiio.tag_type) { 1088264274Smav case CTL_TAG_ORDERED: 1089264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1090264274Smav break; 1091264274Smav case CTL_TAG_HEAD_OF_QUEUE: 1092264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1093264274Smav break; 1094264274Smav case CTL_TAG_UNTAGGED: 1095264274Smav case CTL_TAG_SIMPLE: 1096264274Smav case CTL_TAG_ACA: 1097264274Smav default: 1098264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1099264274Smav break; 1100264274Smav } 1101264274Smav 1102264274Smav beio->io_len = 0; 1103264274Smav beio->io_offset = -1; 1104264274Smav 1105264274Smav beio->bio_cmd = BIO_DELETE; 1106264274Smav beio->ds_trans_type = DEVSTAT_FREE; 1107264274Smav 1108267515Smav DPRINTF("UNMAP\n"); 1109264274Smav 1110264274Smav be_lun->unmap(be_lun, beio); 1111264274Smav} 1112264274Smav 1113264274Smavstatic void 1114264274Smavctl_be_block_cw_done(struct ctl_be_block_io *beio) 1115264274Smav{ 1116264274Smav union ctl_io *io; 1117264274Smav 1118264274Smav io = beio->io; 1119264274Smav ctl_free_beio(beio); 1120264274Smav ctl_config_write_done(io); 1121264274Smav} 1122264274Smav 1123264274Smavstatic void 1124229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1125229997Sken union ctl_io *io) 1126229997Sken{ 1127229997Sken struct ctl_be_block_io *beio; 1128229997Sken struct ctl_be_block_softc *softc; 1129229997Sken 1130229997Sken DPRINTF("entered\n"); 1131229997Sken 1132229997Sken softc = be_lun->softc; 1133229997Sken beio = ctl_alloc_beio(softc); 1134229997Sken beio->io = io; 1135229997Sken beio->lun = be_lun; 1136264274Smav beio->beio_cont = ctl_be_block_cw_done; 1137267519Smav PRIV(io)->ptr = (void *)beio; 1138229997Sken 1139229997Sken switch (io->scsiio.cdb[0]) { 1140229997Sken case SYNCHRONIZE_CACHE: 1141229997Sken case SYNCHRONIZE_CACHE_16: 1142249194Strasz beio->bio_cmd = BIO_FLUSH; 1143229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 1144229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1145229997Sken beio->io_len = 0; 1146229997Sken be_lun->lun_flush(be_lun, beio); 1147229997Sken break; 1148264274Smav case WRITE_SAME_10: 1149264274Smav case WRITE_SAME_16: 1150264274Smav ctl_be_block_cw_dispatch_ws(be_lun, io); 1151264274Smav break; 1152264274Smav case UNMAP: 1153264274Smav ctl_be_block_cw_dispatch_unmap(be_lun, io); 1154264274Smav break; 1155229997Sken default: 1156229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1157229997Sken break; 1158229997Sken } 1159229997Sken} 1160229997Sken 1161258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1162258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1163258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1164258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1165229997Sken 1166229997Skenstatic void 1167264886Smavctl_be_block_next(struct ctl_be_block_io *beio) 1168264886Smav{ 1169264886Smav struct ctl_be_block_lun *be_lun; 1170264886Smav union ctl_io *io; 1171264886Smav 1172264886Smav io = beio->io; 1173264886Smav be_lun = beio->lun; 1174264886Smav ctl_free_beio(beio); 1175267641Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1176267641Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1177267641Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1178267537Smav ctl_data_submit_done(io); 1179264886Smav return; 1180264886Smav } 1181264886Smav 1182264886Smav io->io_hdr.status &= ~CTL_STATUS_MASK; 1183264886Smav io->io_hdr.status |= CTL_STATUS_NONE; 1184264886Smav 1185267877Smav mtx_lock(&be_lun->queue_lock); 1186264886Smav /* 1187264886Smav * XXX KDM make sure that links is okay to use at this point. 1188264886Smav * Otherwise, we either need to add another field to ctl_io_hdr, 1189264886Smav * or deal with resource allocation here. 1190264886Smav */ 1191264886Smav STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1192267877Smav mtx_unlock(&be_lun->queue_lock); 1193264886Smav 1194264886Smav taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1195264886Smav} 1196264886Smav 1197264886Smavstatic void 1198229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1199229997Sken union ctl_io *io) 1200229997Sken{ 1201229997Sken struct ctl_be_block_io *beio; 1202229997Sken struct ctl_be_block_softc *softc; 1203267537Smav struct ctl_lba_len_flags *lbalen; 1204267519Smav struct ctl_ptr_len_flags *bptrlen; 1205267519Smav uint64_t len_left, lbas; 1206229997Sken int i; 1207229997Sken 1208229997Sken softc = be_lun->softc; 1209229997Sken 1210229997Sken DPRINTF("entered\n"); 1211229997Sken 1212267537Smav lbalen = ARGS(io); 1213267537Smav if (lbalen->flags & CTL_LLF_WRITE) { 1214267537Smav SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1215267537Smav } else { 1216229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1217229997Sken } 1218229997Sken 1219229997Sken beio = ctl_alloc_beio(softc); 1220229997Sken beio->io = io; 1221229997Sken beio->lun = be_lun; 1222267519Smav bptrlen = PRIV(io); 1223267519Smav bptrlen->ptr = (void *)beio; 1224229997Sken 1225229997Sken /* 1226229997Sken * If the I/O came down with an ordered or head of queue tag, set 1227229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 1228229997Sken * pretty much the best we can do. 1229229997Sken * 1230229997Sken * XXX KDM we don't have a great way to easily know about the FUA 1231229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 1232229997Sken * pass that knowledge to the backend), and in any case we would 1233229997Sken * need to determine how to handle it. 1234229997Sken */ 1235229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1236229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1237229997Sken beio->bio_flags = BIO_ORDERED; 1238229997Sken 1239229997Sken switch (io->scsiio.tag_type) { 1240229997Sken case CTL_TAG_ORDERED: 1241229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1242229997Sken break; 1243229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1244229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1245229997Sken break; 1246229997Sken case CTL_TAG_UNTAGGED: 1247229997Sken case CTL_TAG_SIMPLE: 1248229997Sken case CTL_TAG_ACA: 1249229997Sken default: 1250229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1251229997Sken break; 1252229997Sken } 1253229997Sken 1254267537Smav if (lbalen->flags & CTL_LLF_WRITE) { 1255267537Smav beio->bio_cmd = BIO_WRITE; 1256267537Smav beio->ds_trans_type = DEVSTAT_WRITE; 1257267537Smav } else { 1258229997Sken beio->bio_cmd = BIO_READ; 1259229997Sken beio->ds_trans_type = DEVSTAT_READ; 1260229997Sken } 1261229997Sken 1262264886Smav DPRINTF("%s at LBA %jx len %u @%ju\n", 1263229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1264267519Smav (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1265267537Smav if (lbalen->flags & CTL_LLF_COMPARE) 1266267537Smav lbas = CTLBLK_HALF_IO_SIZE; 1267267537Smav else 1268267537Smav lbas = CTLBLK_MAX_IO_SIZE; 1269267537Smav lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1270267519Smav beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1271267519Smav beio->io_len = lbas * be_lun->blocksize; 1272267519Smav bptrlen->len += lbas; 1273229997Sken 1274264886Smav for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1275264886Smav KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1276264886Smav i, CTLBLK_MAX_SEGS)); 1277229997Sken 1278229997Sken /* 1279229997Sken * Setup the S/G entry for this chunk. 1280229997Sken */ 1281264886Smav beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1282229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1283229997Sken 1284229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1285229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1286229997Sken 1287267537Smav /* Set up second segment for compare operation. */ 1288267537Smav if (lbalen->flags & CTL_LLF_COMPARE) { 1289267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1290267537Smav beio->sg_segs[i].len; 1291267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1292267537Smav uma_zalloc(be_lun->lun_zone, M_WAITOK); 1293267537Smav } 1294267537Smav 1295229997Sken beio->num_segs++; 1296229997Sken len_left -= beio->sg_segs[i].len; 1297229997Sken } 1298267519Smav if (bptrlen->len < lbalen->len) 1299264886Smav beio->beio_cont = ctl_be_block_next; 1300264886Smav io->scsiio.be_move_done = ctl_be_block_move_done; 1301267537Smav /* For compare we have separate S/G lists for read and datamove. */ 1302267537Smav if (lbalen->flags & CTL_LLF_COMPARE) 1303267537Smav io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1304267537Smav else 1305267537Smav io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1306264886Smav io->scsiio.kern_data_len = beio->io_len; 1307264886Smav io->scsiio.kern_data_resid = 0; 1308264886Smav io->scsiio.kern_sg_entries = beio->num_segs; 1309264886Smav io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1310229997Sken 1311229997Sken /* 1312229997Sken * For the read case, we need to read the data into our buffers and 1313229997Sken * then we can send it back to the user. For the write case, we 1314229997Sken * need to get the data from the user first. 1315229997Sken */ 1316229997Sken if (beio->bio_cmd == BIO_READ) { 1317229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1318229997Sken be_lun->dispatch(be_lun, beio); 1319229997Sken } else { 1320229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1321229997Sken#ifdef CTL_TIME_IO 1322229997Sken getbintime(&io->io_hdr.dma_start_bt); 1323229997Sken#endif 1324229997Sken ctl_datamove(io); 1325229997Sken } 1326229997Sken} 1327229997Sken 1328229997Skenstatic void 1329229997Skenctl_be_block_worker(void *context, int pending) 1330229997Sken{ 1331229997Sken struct ctl_be_block_lun *be_lun; 1332229997Sken struct ctl_be_block_softc *softc; 1333229997Sken union ctl_io *io; 1334229997Sken 1335229997Sken be_lun = (struct ctl_be_block_lun *)context; 1336229997Sken softc = be_lun->softc; 1337229997Sken 1338229997Sken DPRINTF("entered\n"); 1339229997Sken 1340267877Smav mtx_lock(&be_lun->queue_lock); 1341229997Sken for (;;) { 1342229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1343229997Sken if (io != NULL) { 1344229997Sken struct ctl_be_block_io *beio; 1345229997Sken 1346229997Sken DPRINTF("datamove queue\n"); 1347229997Sken 1348229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1349229997Sken ctl_io_hdr, links); 1350229997Sken 1351267877Smav mtx_unlock(&be_lun->queue_lock); 1352229997Sken 1353267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1354229997Sken 1355229997Sken be_lun->dispatch(be_lun, beio); 1356229997Sken 1357267877Smav mtx_lock(&be_lun->queue_lock); 1358229997Sken continue; 1359229997Sken } 1360229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1361229997Sken if (io != NULL) { 1362229997Sken 1363229997Sken DPRINTF("config write queue\n"); 1364229997Sken 1365229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1366229997Sken ctl_io_hdr, links); 1367229997Sken 1368267877Smav mtx_unlock(&be_lun->queue_lock); 1369229997Sken 1370229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1371229997Sken 1372267877Smav mtx_lock(&be_lun->queue_lock); 1373229997Sken continue; 1374229997Sken } 1375229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1376229997Sken if (io != NULL) { 1377229997Sken DPRINTF("input queue\n"); 1378229997Sken 1379229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1380229997Sken ctl_io_hdr, links); 1381267877Smav mtx_unlock(&be_lun->queue_lock); 1382229997Sken 1383229997Sken /* 1384229997Sken * We must drop the lock, since this routine and 1385229997Sken * its children may sleep. 1386229997Sken */ 1387229997Sken ctl_be_block_dispatch(be_lun, io); 1388229997Sken 1389267877Smav mtx_lock(&be_lun->queue_lock); 1390229997Sken continue; 1391229997Sken } 1392229997Sken 1393229997Sken /* 1394229997Sken * If we get here, there is no work left in the queues, so 1395229997Sken * just break out and let the task queue go to sleep. 1396229997Sken */ 1397229997Sken break; 1398229997Sken } 1399267877Smav mtx_unlock(&be_lun->queue_lock); 1400229997Sken} 1401229997Sken 1402229997Sken/* 1403229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1404229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1405229997Sken * thread. 1406229997Sken */ 1407229997Skenstatic int 1408229997Skenctl_be_block_submit(union ctl_io *io) 1409229997Sken{ 1410229997Sken struct ctl_be_block_lun *be_lun; 1411229997Sken struct ctl_be_lun *ctl_be_lun; 1412229997Sken 1413229997Sken DPRINTF("entered\n"); 1414229997Sken 1415229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1416229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1417229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1418229997Sken 1419229997Sken /* 1420229997Sken * Make sure we only get SCSI I/O. 1421229997Sken */ 1422229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1423229997Sken "%#x) encountered", io->io_hdr.io_type)); 1424229997Sken 1425267519Smav PRIV(io)->len = 0; 1426267519Smav 1427267877Smav mtx_lock(&be_lun->queue_lock); 1428229997Sken /* 1429229997Sken * XXX KDM make sure that links is okay to use at this point. 1430229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1431229997Sken * or deal with resource allocation here. 1432229997Sken */ 1433229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1434267877Smav mtx_unlock(&be_lun->queue_lock); 1435229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1436229997Sken 1437267514Smav return (CTL_RETVAL_COMPLETE); 1438229997Sken} 1439229997Sken 1440229997Skenstatic int 1441229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1442229997Sken int flag, struct thread *td) 1443229997Sken{ 1444229997Sken struct ctl_be_block_softc *softc; 1445229997Sken int error; 1446229997Sken 1447229997Sken softc = &backend_block_softc; 1448229997Sken 1449229997Sken error = 0; 1450229997Sken 1451229997Sken switch (cmd) { 1452229997Sken case CTL_LUN_REQ: { 1453229997Sken struct ctl_lun_req *lun_req; 1454229997Sken 1455229997Sken lun_req = (struct ctl_lun_req *)addr; 1456229997Sken 1457229997Sken switch (lun_req->reqtype) { 1458229997Sken case CTL_LUNREQ_CREATE: 1459229997Sken error = ctl_be_block_create(softc, lun_req); 1460229997Sken break; 1461229997Sken case CTL_LUNREQ_RM: 1462229997Sken error = ctl_be_block_rm(softc, lun_req); 1463229997Sken break; 1464232604Strasz case CTL_LUNREQ_MODIFY: 1465232604Strasz error = ctl_be_block_modify(softc, lun_req); 1466232604Strasz break; 1467229997Sken default: 1468229997Sken lun_req->status = CTL_LUN_ERROR; 1469229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1470229997Sken "%s: invalid LUN request type %d", __func__, 1471229997Sken lun_req->reqtype); 1472229997Sken break; 1473229997Sken } 1474229997Sken break; 1475229997Sken } 1476229997Sken default: 1477229997Sken error = ENOTTY; 1478229997Sken break; 1479229997Sken } 1480229997Sken 1481229997Sken return (error); 1482229997Sken} 1483229997Sken 1484229997Skenstatic int 1485229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1486229997Sken{ 1487229997Sken struct ctl_be_block_filedata *file_data; 1488229997Sken struct ctl_lun_create_params *params; 1489229997Sken struct vattr vattr; 1490229997Sken int error; 1491229997Sken 1492229997Sken error = 0; 1493229997Sken file_data = &be_lun->backend.file; 1494229997Sken params = &req->reqdata.create; 1495229997Sken 1496229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1497229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1498229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1499229997Sken 1500229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1501229997Sken if (error != 0) { 1502229997Sken snprintf(req->error_str, sizeof(req->error_str), 1503229997Sken "error calling VOP_GETATTR() for file %s", 1504229997Sken be_lun->dev_path); 1505229997Sken return (error); 1506229997Sken } 1507229997Sken 1508229997Sken /* 1509229997Sken * Verify that we have the ability to upgrade to exclusive 1510229997Sken * access on this file so we can trap errors at open instead 1511229997Sken * of reporting them during first access. 1512229997Sken */ 1513229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1514229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1515229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1516229997Sken error = EBADF; 1517229997Sken snprintf(req->error_str, sizeof(req->error_str), 1518229997Sken "error locking file %s", be_lun->dev_path); 1519229997Sken return (error); 1520229997Sken } 1521229997Sken } 1522229997Sken 1523229997Sken 1524229997Sken file_data->cred = crhold(curthread->td_ucred); 1525232604Strasz if (params->lun_size_bytes != 0) 1526232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1527232604Strasz else 1528232604Strasz be_lun->size_bytes = vattr.va_size; 1529229997Sken /* 1530229997Sken * We set the multi thread flag for file operations because all 1531229997Sken * filesystems (in theory) are capable of allowing multiple readers 1532229997Sken * of a file at once. So we want to get the maximum possible 1533229997Sken * concurrency. 1534229997Sken */ 1535229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1536229997Sken 1537229997Sken /* 1538229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1539229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1540229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1541229997Sken * may not work with other OSes as well. So just export a sector 1542229997Sken * size of 512 bytes, which should work with any OS or 1543229997Sken * application. Since our backing is a file, any block size will 1544229997Sken * work fine for the backing store. 1545229997Sken */ 1546229997Sken#if 0 1547229997Sken be_lun->blocksize= vattr.va_blocksize; 1548229997Sken#endif 1549229997Sken if (params->blocksize_bytes != 0) 1550229997Sken be_lun->blocksize = params->blocksize_bytes; 1551229997Sken else 1552229997Sken be_lun->blocksize = 512; 1553229997Sken 1554229997Sken /* 1555229997Sken * Sanity check. The media size has to be at least one 1556229997Sken * sector long. 1557229997Sken */ 1558229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1559229997Sken error = EINVAL; 1560229997Sken snprintf(req->error_str, sizeof(req->error_str), 1561229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1562229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1563229997Sken } 1564229997Sken return (error); 1565229997Sken} 1566229997Sken 1567229997Skenstatic int 1568229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1569229997Sken{ 1570229997Sken struct ctl_lun_create_params *params; 1571229997Sken struct vattr vattr; 1572229997Sken struct cdev *dev; 1573229997Sken struct cdevsw *devsw; 1574229997Sken int error; 1575264191Smav off_t ps, pss, po, pos; 1576229997Sken 1577229997Sken params = &req->reqdata.create; 1578229997Sken 1579229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1580229997Sken be_lun->dispatch = ctl_be_block_dispatch_dev; 1581229997Sken be_lun->lun_flush = ctl_be_block_flush_dev; 1582264274Smav be_lun->unmap = ctl_be_block_unmap_dev; 1583229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1584229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1585229997Sken &be_lun->backend.dev.dev_ref); 1586229997Sken if (be_lun->backend.dev.csw == NULL) 1587229997Sken panic("Unable to retrieve device switch"); 1588229997Sken 1589229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1590229997Sken if (error) { 1591229997Sken snprintf(req->error_str, sizeof(req->error_str), 1592229997Sken "%s: error getting vnode attributes for device %s", 1593229997Sken __func__, be_lun->dev_path); 1594229997Sken return (error); 1595229997Sken } 1596229997Sken 1597229997Sken dev = be_lun->vn->v_rdev; 1598229997Sken devsw = dev->si_devsw; 1599229997Sken if (!devsw->d_ioctl) { 1600229997Sken snprintf(req->error_str, sizeof(req->error_str), 1601229997Sken "%s: no d_ioctl for device %s!", __func__, 1602229997Sken be_lun->dev_path); 1603229997Sken return (ENODEV); 1604229997Sken } 1605229997Sken 1606229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1607229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1608229997Sken curthread); 1609229997Sken if (error) { 1610229997Sken snprintf(req->error_str, sizeof(req->error_str), 1611229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1612229997Sken "on %s!", __func__, error, be_lun->dev_path); 1613229997Sken return (error); 1614229997Sken } 1615229997Sken 1616229997Sken /* 1617229997Sken * If the user has asked for a blocksize that is greater than the 1618229997Sken * backing device's blocksize, we can do it only if the blocksize 1619229997Sken * the user is asking for is an even multiple of the underlying 1620229997Sken * device's blocksize. 1621229997Sken */ 1622229997Sken if ((params->blocksize_bytes != 0) 1623229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1624229997Sken uint32_t bs_multiple, tmp_blocksize; 1625229997Sken 1626229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1627229997Sken 1628229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1629229997Sken 1630229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1631229997Sken be_lun->blocksize = params->blocksize_bytes; 1632229997Sken } else { 1633229997Sken snprintf(req->error_str, sizeof(req->error_str), 1634229997Sken "%s: requested blocksize %u is not an even " 1635229997Sken "multiple of backing device blocksize %u", 1636229997Sken __func__, params->blocksize_bytes, 1637229997Sken be_lun->blocksize); 1638229997Sken return (EINVAL); 1639229997Sken 1640229997Sken } 1641229997Sken } else if ((params->blocksize_bytes != 0) 1642229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1643229997Sken snprintf(req->error_str, sizeof(req->error_str), 1644229997Sken "%s: requested blocksize %u < backing device " 1645229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1646229997Sken be_lun->blocksize); 1647229997Sken return (EINVAL); 1648229997Sken } 1649229997Sken 1650229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1651229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1652229997Sken curthread); 1653229997Sken if (error) { 1654229997Sken snprintf(req->error_str, sizeof(req->error_str), 1655232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1656232604Strasz " ioctl on %s!", __func__, error, 1657232604Strasz be_lun->dev_path); 1658229997Sken return (error); 1659229997Sken } 1660229997Sken 1661232604Strasz if (params->lun_size_bytes != 0) { 1662232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1663232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1664232604Strasz "%s: requested LUN size %ju > backing device " 1665232604Strasz "size %ju", __func__, 1666232604Strasz (uintmax_t)params->lun_size_bytes, 1667232604Strasz (uintmax_t)be_lun->size_bytes); 1668232604Strasz return (EINVAL); 1669232604Strasz } 1670232604Strasz 1671232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1672232604Strasz } 1673232604Strasz 1674264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1675264191Smav (caddr_t)&ps, FREAD, curthread); 1676264191Smav if (error) 1677264191Smav ps = po = 0; 1678264191Smav else { 1679264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1680264191Smav (caddr_t)&po, FREAD, curthread); 1681264191Smav if (error) 1682264191Smav po = 0; 1683264191Smav } 1684264191Smav pss = ps / be_lun->blocksize; 1685264191Smav pos = po / be_lun->blocksize; 1686264191Smav if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1687264191Smav ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1688264191Smav be_lun->pblockexp = fls(pss) - 1; 1689264191Smav be_lun->pblockoff = (pss - pos) % pss; 1690264191Smav } 1691264191Smav 1692229997Sken return (0); 1693229997Sken} 1694229997Sken 1695229997Skenstatic int 1696229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1697229997Sken{ 1698229997Sken DROP_GIANT(); 1699229997Sken if (be_lun->vn) { 1700229997Sken int flags = FREAD | FWRITE; 1701229997Sken 1702229997Sken switch (be_lun->dev_type) { 1703229997Sken case CTL_BE_BLOCK_DEV: 1704229997Sken if (be_lun->backend.dev.csw) { 1705229997Sken dev_relthread(be_lun->backend.dev.cdev, 1706229997Sken be_lun->backend.dev.dev_ref); 1707229997Sken be_lun->backend.dev.csw = NULL; 1708229997Sken be_lun->backend.dev.cdev = NULL; 1709229997Sken } 1710229997Sken break; 1711229997Sken case CTL_BE_BLOCK_FILE: 1712229997Sken break; 1713229997Sken case CTL_BE_BLOCK_NONE: 1714258871Strasz break; 1715229997Sken default: 1716229997Sken panic("Unexpected backend type."); 1717229997Sken break; 1718229997Sken } 1719229997Sken 1720229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1721229997Sken be_lun->vn = NULL; 1722229997Sken 1723229997Sken switch (be_lun->dev_type) { 1724229997Sken case CTL_BE_BLOCK_DEV: 1725229997Sken break; 1726229997Sken case CTL_BE_BLOCK_FILE: 1727229997Sken if (be_lun->backend.file.cred != NULL) { 1728229997Sken crfree(be_lun->backend.file.cred); 1729229997Sken be_lun->backend.file.cred = NULL; 1730229997Sken } 1731229997Sken break; 1732229997Sken case CTL_BE_BLOCK_NONE: 1733258871Strasz break; 1734229997Sken default: 1735229997Sken panic("Unexpected backend type."); 1736229997Sken break; 1737229997Sken } 1738229997Sken } 1739229997Sken PICKUP_GIANT(); 1740229997Sken 1741229997Sken return (0); 1742229997Sken} 1743229997Sken 1744229997Skenstatic int 1745229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1746229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1747229997Sken{ 1748229997Sken struct nameidata nd; 1749229997Sken int flags; 1750229997Sken int error; 1751229997Sken 1752229997Sken /* 1753229997Sken * XXX KDM allow a read-only option? 1754229997Sken */ 1755229997Sken flags = FREAD | FWRITE; 1756229997Sken error = 0; 1757229997Sken 1758229997Sken if (rootvnode == NULL) { 1759229997Sken snprintf(req->error_str, sizeof(req->error_str), 1760229997Sken "%s: Root filesystem is not mounted", __func__); 1761229997Sken return (1); 1762229997Sken } 1763229997Sken 1764229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1765229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1766229997Sken VREF(rootvnode); 1767229997Sken } 1768229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1769229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1770229997Sken VREF(rootvnode); 1771229997Sken } 1772229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1773229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1774229997Sken VREF(rootvnode); 1775229997Sken } 1776229997Sken 1777229997Sken again: 1778229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1779229997Sken error = vn_open(&nd, &flags, 0, NULL); 1780229997Sken if (error) { 1781229997Sken /* 1782229997Sken * This is the only reasonable guess we can make as far as 1783229997Sken * path if the user doesn't give us a fully qualified path. 1784229997Sken * If they want to specify a file, they need to specify the 1785229997Sken * full path. 1786229997Sken */ 1787229997Sken if (be_lun->dev_path[0] != '/') { 1788229997Sken char *dev_path = "/dev/"; 1789229997Sken char *dev_name; 1790229997Sken 1791229997Sken /* Try adding device path at beginning of name */ 1792229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1793229997Sken + strlen(dev_path) + 1, 1794229997Sken M_CTLBLK, M_WAITOK); 1795229997Sken if (dev_name) { 1796229997Sken sprintf(dev_name, "%s%s", dev_path, 1797229997Sken be_lun->dev_path); 1798229997Sken free(be_lun->dev_path, M_CTLBLK); 1799229997Sken be_lun->dev_path = dev_name; 1800229997Sken goto again; 1801229997Sken } 1802229997Sken } 1803229997Sken snprintf(req->error_str, sizeof(req->error_str), 1804229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1805229997Sken return (error); 1806229997Sken } 1807229997Sken 1808229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1809229997Sken 1810229997Sken be_lun->vn = nd.ni_vp; 1811229997Sken 1812229997Sken /* We only support disks and files. */ 1813229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1814229997Sken error = ctl_be_block_open_dev(be_lun, req); 1815229997Sken } else if (be_lun->vn->v_type == VREG) { 1816229997Sken error = ctl_be_block_open_file(be_lun, req); 1817229997Sken } else { 1818229997Sken error = EINVAL; 1819229997Sken snprintf(req->error_str, sizeof(req->error_str), 1820258871Strasz "%s is not a disk or plain file", be_lun->dev_path); 1821229997Sken } 1822229997Sken VOP_UNLOCK(be_lun->vn, 0); 1823229997Sken 1824229997Sken if (error != 0) { 1825229997Sken ctl_be_block_close(be_lun); 1826229997Sken return (error); 1827229997Sken } 1828229997Sken 1829229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1830229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1831229997Sken 1832229997Sken return (0); 1833229997Sken} 1834229997Sken 1835229997Skenstatic int 1836229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1837229997Sken{ 1838229997Sken struct ctl_be_block_lun *be_lun; 1839229997Sken struct ctl_lun_create_params *params; 1840267481Smav char num_thread_str[16]; 1841229997Sken char tmpstr[32]; 1842267481Smav char *value; 1843264274Smav int retval, num_threads, unmap; 1844267481Smav int tmp_num_threads; 1845229997Sken 1846229997Sken params = &req->reqdata.create; 1847229997Sken retval = 0; 1848229997Sken 1849229997Sken num_threads = cbb_num_threads; 1850229997Sken 1851229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1852229997Sken 1853229997Sken be_lun->softc = softc; 1854229997Sken STAILQ_INIT(&be_lun->input_queue); 1855229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1856229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1857229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1858267877Smav mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1859267877Smav mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1860268280Smav ctl_init_opts(&be_lun->ctl_be_lun.options, 1861268280Smav req->num_be_args, req->kern_be_args); 1862229997Sken 1863264886Smav be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1864256995Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1865229997Sken 1866229997Sken if (be_lun->lun_zone == NULL) { 1867229997Sken snprintf(req->error_str, sizeof(req->error_str), 1868229997Sken "%s: error allocating UMA zone", __func__); 1869229997Sken goto bailout_error; 1870229997Sken } 1871229997Sken 1872229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1873229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1874229997Sken else 1875229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1876229997Sken 1877229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1878268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1879267499Smav if (value == NULL) { 1880229997Sken snprintf(req->error_str, sizeof(req->error_str), 1881229997Sken "%s: no file argument specified", __func__); 1882229997Sken goto bailout_error; 1883229997Sken } 1884267499Smav be_lun->dev_path = strdup(value, M_CTLBLK); 1885229997Sken 1886229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1887229997Sken if (retval != 0) { 1888229997Sken retval = 0; 1889229997Sken goto bailout_error; 1890229997Sken } 1891229997Sken 1892229997Sken /* 1893229997Sken * Tell the user the size of the file/device. 1894229997Sken */ 1895229997Sken params->lun_size_bytes = be_lun->size_bytes; 1896229997Sken 1897229997Sken /* 1898229997Sken * The maximum LBA is the size - 1. 1899229997Sken */ 1900229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1901229997Sken } else { 1902229997Sken /* 1903229997Sken * For processor devices, we don't have any size. 1904229997Sken */ 1905229997Sken be_lun->blocksize = 0; 1906264191Smav be_lun->pblockexp = 0; 1907264191Smav be_lun->pblockoff = 0; 1908229997Sken be_lun->size_blocks = 0; 1909229997Sken be_lun->size_bytes = 0; 1910229997Sken be_lun->ctl_be_lun.maxlba = 0; 1911229997Sken params->lun_size_bytes = 0; 1912229997Sken 1913229997Sken /* 1914229997Sken * Default to just 1 thread for processor devices. 1915229997Sken */ 1916229997Sken num_threads = 1; 1917229997Sken } 1918229997Sken 1919229997Sken /* 1920229997Sken * XXX This searching loop might be refactored to be combined with 1921229997Sken * the loop above, 1922229997Sken */ 1923268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 1924267481Smav if (value != NULL) { 1925267481Smav tmp_num_threads = strtol(value, NULL, 0); 1926229997Sken 1927267481Smav /* 1928267481Smav * We don't let the user specify less than one 1929267481Smav * thread, but hope he's clueful enough not to 1930267481Smav * specify 1000 threads. 1931267481Smav */ 1932267481Smav if (tmp_num_threads < 1) { 1933267481Smav snprintf(req->error_str, sizeof(req->error_str), 1934267481Smav "%s: invalid number of threads %s", 1935267481Smav __func__, num_thread_str); 1936267481Smav goto bailout_error; 1937229997Sken } 1938267481Smav num_threads = tmp_num_threads; 1939229997Sken } 1940267481Smav unmap = 0; 1941268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 1942267481Smav if (value != NULL && strcmp(value, "on") == 0) 1943267481Smav unmap = 1; 1944229997Sken 1945229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1946229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1947264274Smav if (unmap) 1948264274Smav be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 1949229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 1950229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1951264191Smav be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 1952264191Smav be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 1953229997Sken /* Tell the user the blocksize we ended up using */ 1954229997Sken params->blocksize_bytes = be_lun->blocksize; 1955229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1956229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1957229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1958229997Sken } else 1959229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 1960229997Sken 1961229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1962229997Sken be_lun->ctl_be_lun.lun_config_status = 1963229997Sken ctl_be_block_lun_config_status; 1964229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1965229997Sken 1966229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1967229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1968229997Sken softc->num_luns); 1969229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1970229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1971229997Sken sizeof(tmpstr))); 1972229997Sken 1973229997Sken /* Tell the user what we used for a serial number */ 1974229997Sken strncpy((char *)params->serial_num, tmpstr, 1975229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1976229997Sken } else { 1977229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 1978229997Sken params->serial_num, 1979229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1980229997Sken sizeof(params->serial_num))); 1981229997Sken } 1982229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1983229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1984229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1985229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1986229997Sken sizeof(tmpstr))); 1987229997Sken 1988229997Sken /* Tell the user what we used for a device ID */ 1989229997Sken strncpy((char *)params->device_id, tmpstr, 1990229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1991229997Sken } else { 1992229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 1993229997Sken params->device_id, 1994229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1995229997Sken sizeof(params->device_id))); 1996229997Sken } 1997229997Sken 1998229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 1999229997Sken 2000229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2001229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2002229997Sken 2003229997Sken if (be_lun->io_taskqueue == NULL) { 2004229997Sken snprintf(req->error_str, sizeof(req->error_str), 2005229997Sken "%s: Unable to create taskqueue", __func__); 2006229997Sken goto bailout_error; 2007229997Sken } 2008229997Sken 2009229997Sken /* 2010229997Sken * Note that we start the same number of threads by default for 2011229997Sken * both the file case and the block device case. For the file 2012229997Sken * case, we need multiple threads to allow concurrency, because the 2013229997Sken * vnode interface is designed to be a blocking interface. For the 2014229997Sken * block device case, ZFS zvols at least will block the caller's 2015229997Sken * context in many instances, and so we need multiple threads to 2016229997Sken * overcome that problem. Other block devices don't need as many 2017229997Sken * threads, but they shouldn't cause too many problems. 2018229997Sken * 2019229997Sken * If the user wants to just have a single thread for a block 2020229997Sken * device, he can specify that when the LUN is created, or change 2021229997Sken * the tunable/sysctl to alter the default number of threads. 2022229997Sken */ 2023229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2024229997Sken /*num threads*/num_threads, 2025229997Sken /*priority*/PWAIT, 2026229997Sken /*thread name*/ 2027229997Sken "%s taskq", be_lun->lunname); 2028229997Sken 2029229997Sken if (retval != 0) 2030229997Sken goto bailout_error; 2031229997Sken 2032229997Sken be_lun->num_threads = num_threads; 2033229997Sken 2034229997Sken mtx_lock(&softc->lock); 2035229997Sken softc->num_luns++; 2036229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2037229997Sken 2038229997Sken mtx_unlock(&softc->lock); 2039229997Sken 2040229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 2041229997Sken if (retval != 0) { 2042229997Sken mtx_lock(&softc->lock); 2043229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2044229997Sken links); 2045229997Sken softc->num_luns--; 2046229997Sken mtx_unlock(&softc->lock); 2047229997Sken snprintf(req->error_str, sizeof(req->error_str), 2048229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 2049229997Sken "details", __func__, retval); 2050229997Sken retval = 0; 2051229997Sken goto bailout_error; 2052229997Sken } 2053229997Sken 2054229997Sken mtx_lock(&softc->lock); 2055229997Sken 2056229997Sken /* 2057229997Sken * Tell the config_status routine that we're waiting so it won't 2058229997Sken * clean up the LUN in the event of an error. 2059229997Sken */ 2060229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2061229997Sken 2062229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2063229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2064229997Sken if (retval == EINTR) 2065229997Sken break; 2066229997Sken } 2067229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2068229997Sken 2069229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2070229997Sken snprintf(req->error_str, sizeof(req->error_str), 2071229997Sken "%s: LUN configuration error, see dmesg for details", 2072229997Sken __func__); 2073229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2074229997Sken links); 2075229997Sken softc->num_luns--; 2076229997Sken mtx_unlock(&softc->lock); 2077229997Sken goto bailout_error; 2078229997Sken } else { 2079229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2080229997Sken } 2081229997Sken 2082229997Sken mtx_unlock(&softc->lock); 2083229997Sken 2084229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2085229997Sken be_lun->blocksize, 2086229997Sken DEVSTAT_ALL_SUPPORTED, 2087229997Sken be_lun->ctl_be_lun.lun_type 2088229997Sken | DEVSTAT_TYPE_IF_OTHER, 2089229997Sken DEVSTAT_PRIORITY_OTHER); 2090229997Sken 2091229997Sken 2092229997Sken req->status = CTL_LUN_OK; 2093229997Sken 2094229997Sken return (retval); 2095229997Sken 2096229997Skenbailout_error: 2097229997Sken req->status = CTL_LUN_ERROR; 2098229997Sken 2099267429Smav if (be_lun->io_taskqueue != NULL) 2100267429Smav taskqueue_free(be_lun->io_taskqueue); 2101229997Sken ctl_be_block_close(be_lun); 2102267429Smav if (be_lun->dev_path != NULL) 2103267429Smav free(be_lun->dev_path, M_CTLBLK); 2104267429Smav if (be_lun->lun_zone != NULL) 2105267429Smav uma_zdestroy(be_lun->lun_zone); 2106268280Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2107267877Smav mtx_destroy(&be_lun->queue_lock); 2108267877Smav mtx_destroy(&be_lun->io_lock); 2109229997Sken free(be_lun, M_CTLBLK); 2110229997Sken 2111229997Sken return (retval); 2112229997Sken} 2113229997Sken 2114229997Skenstatic int 2115229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2116229997Sken{ 2117229997Sken struct ctl_lun_rm_params *params; 2118229997Sken struct ctl_be_block_lun *be_lun; 2119229997Sken int retval; 2120229997Sken 2121229997Sken params = &req->reqdata.rm; 2122229997Sken 2123229997Sken mtx_lock(&softc->lock); 2124229997Sken 2125229997Sken be_lun = NULL; 2126229997Sken 2127229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2128229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2129229997Sken break; 2130229997Sken } 2131229997Sken mtx_unlock(&softc->lock); 2132229997Sken 2133229997Sken if (be_lun == NULL) { 2134229997Sken snprintf(req->error_str, sizeof(req->error_str), 2135229997Sken "%s: LUN %u is not managed by the block backend", 2136229997Sken __func__, params->lun_id); 2137229997Sken goto bailout_error; 2138229997Sken } 2139229997Sken 2140229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2141229997Sken 2142229997Sken if (retval != 0) { 2143229997Sken snprintf(req->error_str, sizeof(req->error_str), 2144229997Sken "%s: error %d returned from ctl_disable_lun() for " 2145229997Sken "LUN %d", __func__, retval, params->lun_id); 2146229997Sken goto bailout_error; 2147229997Sken 2148229997Sken } 2149229997Sken 2150229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2151229997Sken if (retval != 0) { 2152229997Sken snprintf(req->error_str, sizeof(req->error_str), 2153229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 2154229997Sken "LUN %d", __func__, retval, params->lun_id); 2155229997Sken goto bailout_error; 2156229997Sken } 2157229997Sken 2158229997Sken mtx_lock(&softc->lock); 2159229997Sken 2160229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2161229997Sken 2162229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2163229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2164229997Sken if (retval == EINTR) 2165229997Sken break; 2166229997Sken } 2167229997Sken 2168229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2169229997Sken 2170229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2171229997Sken snprintf(req->error_str, sizeof(req->error_str), 2172229997Sken "%s: interrupted waiting for LUN to be freed", 2173229997Sken __func__); 2174229997Sken mtx_unlock(&softc->lock); 2175229997Sken goto bailout_error; 2176229997Sken } 2177229997Sken 2178229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2179229997Sken 2180229997Sken softc->num_luns--; 2181229997Sken mtx_unlock(&softc->lock); 2182229997Sken 2183229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2184229997Sken 2185229997Sken taskqueue_free(be_lun->io_taskqueue); 2186229997Sken 2187229997Sken ctl_be_block_close(be_lun); 2188229997Sken 2189229997Sken if (be_lun->disk_stats != NULL) 2190229997Sken devstat_remove_entry(be_lun->disk_stats); 2191229997Sken 2192229997Sken uma_zdestroy(be_lun->lun_zone); 2193229997Sken 2194268280Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2195229997Sken free(be_lun->dev_path, M_CTLBLK); 2196267877Smav mtx_destroy(&be_lun->queue_lock); 2197267877Smav mtx_destroy(&be_lun->io_lock); 2198229997Sken free(be_lun, M_CTLBLK); 2199229997Sken 2200229997Sken req->status = CTL_LUN_OK; 2201229997Sken 2202229997Sken return (0); 2203229997Sken 2204229997Skenbailout_error: 2205229997Sken 2206229997Sken req->status = CTL_LUN_ERROR; 2207229997Sken 2208229997Sken return (0); 2209229997Sken} 2210229997Sken 2211232604Straszstatic int 2212232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2213232604Strasz struct ctl_lun_req *req) 2214232604Strasz{ 2215232604Strasz struct vattr vattr; 2216232604Strasz int error; 2217232604Strasz struct ctl_lun_modify_params *params; 2218232604Strasz 2219232604Strasz params = &req->reqdata.modify; 2220232604Strasz 2221232604Strasz if (params->lun_size_bytes != 0) { 2222232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2223232604Strasz } else { 2224232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2225232604Strasz if (error != 0) { 2226232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2227232604Strasz "error calling VOP_GETATTR() for file %s", 2228232604Strasz be_lun->dev_path); 2229232604Strasz return (error); 2230232604Strasz } 2231232604Strasz 2232232604Strasz be_lun->size_bytes = vattr.va_size; 2233232604Strasz } 2234232604Strasz 2235232604Strasz return (0); 2236232604Strasz} 2237232604Strasz 2238232604Straszstatic int 2239232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2240232604Strasz struct ctl_lun_req *req) 2241232604Strasz{ 2242232604Strasz struct cdev *dev; 2243232604Strasz struct cdevsw *devsw; 2244232604Strasz int error; 2245232604Strasz struct ctl_lun_modify_params *params; 2246232604Strasz uint64_t size_bytes; 2247232604Strasz 2248232604Strasz params = &req->reqdata.modify; 2249232604Strasz 2250232604Strasz dev = be_lun->vn->v_rdev; 2251232604Strasz devsw = dev->si_devsw; 2252232604Strasz if (!devsw->d_ioctl) { 2253232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2254232604Strasz "%s: no d_ioctl for device %s!", __func__, 2255232604Strasz be_lun->dev_path); 2256232604Strasz return (ENODEV); 2257232604Strasz } 2258232604Strasz 2259232604Strasz error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2260232604Strasz (caddr_t)&size_bytes, FREAD, 2261232604Strasz curthread); 2262232604Strasz if (error) { 2263232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2264232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2265232604Strasz "on %s!", __func__, error, be_lun->dev_path); 2266232604Strasz return (error); 2267232604Strasz } 2268232604Strasz 2269232604Strasz if (params->lun_size_bytes != 0) { 2270232604Strasz if (params->lun_size_bytes > size_bytes) { 2271232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2272232604Strasz "%s: requested LUN size %ju > backing device " 2273232604Strasz "size %ju", __func__, 2274232604Strasz (uintmax_t)params->lun_size_bytes, 2275232604Strasz (uintmax_t)size_bytes); 2276232604Strasz return (EINVAL); 2277232604Strasz } 2278232604Strasz 2279232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2280232604Strasz } else { 2281232604Strasz be_lun->size_bytes = size_bytes; 2282232604Strasz } 2283232604Strasz 2284232604Strasz return (0); 2285232604Strasz} 2286232604Strasz 2287232604Straszstatic int 2288232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2289232604Strasz{ 2290232604Strasz struct ctl_lun_modify_params *params; 2291232604Strasz struct ctl_be_block_lun *be_lun; 2292241896Skib int error; 2293232604Strasz 2294232604Strasz params = &req->reqdata.modify; 2295232604Strasz 2296232604Strasz mtx_lock(&softc->lock); 2297232604Strasz 2298232604Strasz be_lun = NULL; 2299232604Strasz 2300232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2301232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2302232604Strasz break; 2303232604Strasz } 2304232604Strasz mtx_unlock(&softc->lock); 2305232604Strasz 2306232604Strasz if (be_lun == NULL) { 2307232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2308232604Strasz "%s: LUN %u is not managed by the block backend", 2309232604Strasz __func__, params->lun_id); 2310232604Strasz goto bailout_error; 2311232604Strasz } 2312232604Strasz 2313232604Strasz if (params->lun_size_bytes != 0) { 2314232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2315232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2316232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2317232604Strasz params->lun_size_bytes, be_lun->blocksize); 2318232604Strasz goto bailout_error; 2319232604Strasz } 2320232604Strasz } 2321232604Strasz 2322232604Strasz vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2323232604Strasz 2324232604Strasz if (be_lun->vn->v_type == VREG) 2325232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2326232604Strasz else 2327232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2328232604Strasz 2329232604Strasz VOP_UNLOCK(be_lun->vn, 0); 2330232604Strasz 2331232604Strasz if (error != 0) 2332232604Strasz goto bailout_error; 2333232604Strasz 2334232604Strasz be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2335232604Strasz 2336232604Strasz /* 2337232604Strasz * The maximum LBA is the size - 1. 2338232604Strasz * 2339232604Strasz * XXX: Note that this field is being updated without locking, 2340232604Strasz * which might cause problems on 32-bit architectures. 2341232604Strasz */ 2342232604Strasz be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2343232604Strasz ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2344232604Strasz 2345232604Strasz /* Tell the user the exact size we ended up using */ 2346232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2347232604Strasz 2348232604Strasz req->status = CTL_LUN_OK; 2349232604Strasz 2350232604Strasz return (0); 2351232604Strasz 2352232604Straszbailout_error: 2353232604Strasz req->status = CTL_LUN_ERROR; 2354232604Strasz 2355232604Strasz return (0); 2356232604Strasz} 2357232604Strasz 2358229997Skenstatic void 2359229997Skenctl_be_block_lun_shutdown(void *be_lun) 2360229997Sken{ 2361229997Sken struct ctl_be_block_lun *lun; 2362229997Sken struct ctl_be_block_softc *softc; 2363229997Sken 2364229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2365229997Sken 2366229997Sken softc = lun->softc; 2367229997Sken 2368229997Sken mtx_lock(&softc->lock); 2369229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2370229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2371229997Sken wakeup(lun); 2372229997Sken mtx_unlock(&softc->lock); 2373229997Sken 2374229997Sken} 2375229997Sken 2376229997Skenstatic void 2377229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2378229997Sken{ 2379229997Sken struct ctl_be_block_lun *lun; 2380229997Sken struct ctl_be_block_softc *softc; 2381229997Sken 2382229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2383229997Sken softc = lun->softc; 2384229997Sken 2385229997Sken if (status == CTL_LUN_CONFIG_OK) { 2386229997Sken mtx_lock(&softc->lock); 2387229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2388229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2389229997Sken wakeup(lun); 2390229997Sken mtx_unlock(&softc->lock); 2391229997Sken 2392229997Sken /* 2393229997Sken * We successfully added the LUN, attempt to enable it. 2394229997Sken */ 2395229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2396229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2397229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2398229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2399229997Sken __func__); 2400229997Sken } 2401229997Sken } 2402229997Sken 2403229997Sken return; 2404229997Sken } 2405229997Sken 2406229997Sken 2407229997Sken mtx_lock(&softc->lock); 2408229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2409229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2410229997Sken wakeup(lun); 2411229997Sken mtx_unlock(&softc->lock); 2412229997Sken} 2413229997Sken 2414229997Sken 2415229997Skenstatic int 2416229997Skenctl_be_block_config_write(union ctl_io *io) 2417229997Sken{ 2418229997Sken struct ctl_be_block_lun *be_lun; 2419229997Sken struct ctl_be_lun *ctl_be_lun; 2420229997Sken int retval; 2421229997Sken 2422229997Sken retval = 0; 2423229997Sken 2424229997Sken DPRINTF("entered\n"); 2425229997Sken 2426229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2427229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2428229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2429229997Sken 2430229997Sken switch (io->scsiio.cdb[0]) { 2431229997Sken case SYNCHRONIZE_CACHE: 2432229997Sken case SYNCHRONIZE_CACHE_16: 2433264274Smav case WRITE_SAME_10: 2434264274Smav case WRITE_SAME_16: 2435264274Smav case UNMAP: 2436229997Sken /* 2437229997Sken * The upper level CTL code will filter out any CDBs with 2438229997Sken * the immediate bit set and return the proper error. 2439229997Sken * 2440229997Sken * We don't really need to worry about what LBA range the 2441229997Sken * user asked to be synced out. When they issue a sync 2442229997Sken * cache command, we'll sync out the whole thing. 2443229997Sken */ 2444267877Smav mtx_lock(&be_lun->queue_lock); 2445229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2446229997Sken links); 2447267877Smav mtx_unlock(&be_lun->queue_lock); 2448229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2449229997Sken break; 2450229997Sken case START_STOP_UNIT: { 2451229997Sken struct scsi_start_stop_unit *cdb; 2452229997Sken 2453229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2454229997Sken 2455229997Sken if (cdb->how & SSS_START) 2456229997Sken retval = ctl_start_lun(ctl_be_lun); 2457229997Sken else { 2458229997Sken retval = ctl_stop_lun(ctl_be_lun); 2459229997Sken /* 2460229997Sken * XXX KDM Copan-specific offline behavior. 2461229997Sken * Figure out a reasonable way to port this? 2462229997Sken */ 2463229997Sken#ifdef NEEDTOPORT 2464229997Sken if ((retval == 0) 2465229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2466229997Sken retval = ctl_lun_offline(ctl_be_lun); 2467229997Sken#endif 2468229997Sken } 2469229997Sken 2470229997Sken /* 2471229997Sken * In general, the above routines should not fail. They 2472229997Sken * just set state for the LUN. So we've got something 2473229997Sken * pretty wrong here if we can't start or stop the LUN. 2474229997Sken */ 2475229997Sken if (retval != 0) { 2476229997Sken ctl_set_internal_failure(&io->scsiio, 2477229997Sken /*sks_valid*/ 1, 2478229997Sken /*retry_count*/ 0xf051); 2479229997Sken retval = CTL_RETVAL_COMPLETE; 2480229997Sken } else { 2481229997Sken ctl_set_success(&io->scsiio); 2482229997Sken } 2483229997Sken ctl_config_write_done(io); 2484229997Sken break; 2485229997Sken } 2486229997Sken default: 2487229997Sken ctl_set_invalid_opcode(&io->scsiio); 2488229997Sken ctl_config_write_done(io); 2489229997Sken retval = CTL_RETVAL_COMPLETE; 2490229997Sken break; 2491229997Sken } 2492229997Sken 2493229997Sken return (retval); 2494229997Sken 2495229997Sken} 2496229997Sken 2497229997Skenstatic int 2498229997Skenctl_be_block_config_read(union ctl_io *io) 2499229997Sken{ 2500229997Sken return (0); 2501229997Sken} 2502229997Sken 2503229997Skenstatic int 2504229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2505229997Sken{ 2506229997Sken struct ctl_be_block_lun *lun; 2507229997Sken int retval; 2508229997Sken 2509229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2510229997Sken retval = 0; 2511229997Sken 2512268283Smav retval = sbuf_printf(sb, "\t<num_threads>"); 2513229997Sken 2514229997Sken if (retval != 0) 2515229997Sken goto bailout; 2516229997Sken 2517229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2518229997Sken 2519229997Sken if (retval != 0) 2520229997Sken goto bailout; 2521229997Sken 2522268283Smav retval = sbuf_printf(sb, "</num_threads>\n"); 2523229997Sken 2524229997Skenbailout: 2525229997Sken 2526229997Sken return (retval); 2527229997Sken} 2528229997Sken 2529229997Skenint 2530229997Skenctl_be_block_init(void) 2531229997Sken{ 2532229997Sken struct ctl_be_block_softc *softc; 2533229997Sken int retval; 2534229997Sken 2535229997Sken softc = &backend_block_softc; 2536229997Sken retval = 0; 2537229997Sken 2538267877Smav mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2539264020Strasz beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2540264020Strasz NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2541229997Sken STAILQ_INIT(&softc->disk_list); 2542229997Sken STAILQ_INIT(&softc->lun_list); 2543229997Sken 2544229997Sken return (retval); 2545229997Sken} 2546