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("$FreeBSD$"); 44229997Sken 45229997Sken#include <opt_kdtrace.h> 46229997Sken 47229997Sken#include <sys/param.h> 48229997Sken#include <sys/systm.h> 49229997Sken#include <sys/kernel.h> 50229997Sken#include <sys/types.h> 51229997Sken#include <sys/kthread.h> 52229997Sken#include <sys/bio.h> 53229997Sken#include <sys/fcntl.h> 54265634Smav#include <sys/limits.h> 55229997Sken#include <sys/lock.h> 56229997Sken#include <sys/mutex.h> 57229997Sken#include <sys/condvar.h> 58229997Sken#include <sys/malloc.h> 59229997Sken#include <sys/conf.h> 60229997Sken#include <sys/ioccom.h> 61229997Sken#include <sys/queue.h> 62229997Sken#include <sys/sbuf.h> 63229997Sken#include <sys/endian.h> 64229997Sken#include <sys/uio.h> 65229997Sken#include <sys/buf.h> 66229997Sken#include <sys/taskqueue.h> 67229997Sken#include <sys/vnode.h> 68229997Sken#include <sys/namei.h> 69229997Sken#include <sys/mount.h> 70229997Sken#include <sys/disk.h> 71229997Sken#include <sys/fcntl.h> 72229997Sken#include <sys/filedesc.h> 73229997Sken#include <sys/proc.h> 74229997Sken#include <sys/pcpu.h> 75229997Sken#include <sys/module.h> 76229997Sken#include <sys/sdt.h> 77229997Sken#include <sys/devicestat.h> 78229997Sken#include <sys/sysctl.h> 79229997Sken 80229997Sken#include <geom/geom.h> 81229997Sken 82229997Sken#include <cam/cam.h> 83229997Sken#include <cam/scsi/scsi_all.h> 84229997Sken#include <cam/scsi/scsi_da.h> 85229997Sken#include <cam/ctl/ctl_io.h> 86229997Sken#include <cam/ctl/ctl.h> 87229997Sken#include <cam/ctl/ctl_backend.h> 88229997Sken#include <cam/ctl/ctl_frontend_internal.h> 89229997Sken#include <cam/ctl/ctl_ioctl.h> 90229997Sken#include <cam/ctl/ctl_scsi_all.h> 91229997Sken#include <cam/ctl/ctl_error.h> 92229997Sken 93229997Sken/* 94265642Smav * The idea here is that we'll allocate enough S/G space to hold a 1MB 95265642Smav * I/O. If we get an I/O larger than that, we'll split it. 96229997Sken */ 97268151Smav#define CTLBLK_HALF_IO_SIZE (512 * 1024) 98268151Smav#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 99265642Smav#define CTLBLK_MAX_SEG MAXPHYS 100268151Smav#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 101268151Smav#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 102229997Sken 103229997Sken#ifdef CTLBLK_DEBUG 104229997Sken#define DPRINTF(fmt, args...) \ 105229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 106229997Sken#else 107229997Sken#define DPRINTF(fmt, args...) do {} while(0) 108229997Sken#endif 109229997Sken 110268150Smav#define PRIV(io) \ 111268150Smav ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 112268151Smav#define ARGS(io) \ 113268151Smav ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 114268150Smav 115229997SkenSDT_PROVIDER_DEFINE(cbb); 116229997Sken 117229997Skentypedef enum { 118229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 119229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 120229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 121229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 122229997Sken} ctl_be_block_lun_flags; 123229997Sken 124229997Skentypedef enum { 125229997Sken CTL_BE_BLOCK_NONE, 126229997Sken CTL_BE_BLOCK_DEV, 127229997Sken CTL_BE_BLOCK_FILE 128229997Sken} ctl_be_block_type; 129229997Sken 130229997Skenstruct ctl_be_block_devdata { 131229997Sken struct cdev *cdev; 132229997Sken struct cdevsw *csw; 133229997Sken int dev_ref; 134229997Sken}; 135229997Sken 136229997Skenstruct ctl_be_block_filedata { 137229997Sken struct ucred *cred; 138229997Sken}; 139229997Sken 140229997Skenunion ctl_be_block_bedata { 141229997Sken struct ctl_be_block_devdata dev; 142229997Sken struct ctl_be_block_filedata file; 143229997Sken}; 144229997Sken 145229997Skenstruct ctl_be_block_io; 146229997Skenstruct ctl_be_block_lun; 147229997Sken 148229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 149229997Sken struct ctl_be_block_io *beio); 150229997Sken 151229997Sken/* 152229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 153229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 154229997Sken */ 155229997Skenstruct ctl_be_block_lun { 156229997Sken struct ctl_block_disk *disk; 157229997Sken char lunname[32]; 158229997Sken char *dev_path; 159229997Sken ctl_be_block_type dev_type; 160229997Sken struct vnode *vn; 161229997Sken union ctl_be_block_bedata backend; 162229997Sken cbb_dispatch_t dispatch; 163229997Sken cbb_dispatch_t lun_flush; 164265634Smav cbb_dispatch_t unmap; 165229997Sken uma_zone_t lun_zone; 166229997Sken uint64_t size_blocks; 167229997Sken uint64_t size_bytes; 168229997Sken uint32_t blocksize; 169229997Sken int blocksize_shift; 170264727Smav uint16_t pblockexp; 171264727Smav uint16_t pblockoff; 172229997Sken struct ctl_be_block_softc *softc; 173229997Sken struct devstat *disk_stats; 174229997Sken ctl_be_block_lun_flags flags; 175229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 176229997Sken struct ctl_be_lun ctl_be_lun; 177229997Sken struct taskqueue *io_taskqueue; 178229997Sken struct task io_task; 179229997Sken int num_threads; 180229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 181229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 182229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 183268549Smav struct mtx_padalign io_lock; 184268549Smav struct mtx_padalign queue_lock; 185229997Sken}; 186229997Sken 187229997Sken/* 188229997Sken * Overall softc structure for the block backend module. 189229997Sken */ 190229997Skenstruct ctl_be_block_softc { 191229997Sken struct mtx lock; 192229997Sken int num_disks; 193229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 194229997Sken int num_luns; 195229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 196229997Sken}; 197229997Sken 198229997Skenstatic struct ctl_be_block_softc backend_block_softc; 199229997Sken 200229997Sken/* 201229997Sken * Per-I/O information. 202229997Sken */ 203229997Skenstruct ctl_be_block_io { 204229997Sken union ctl_io *io; 205229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 206229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 207229997Sken int bio_cmd; 208229997Sken int bio_flags; 209229997Sken int num_segs; 210229997Sken int num_bios_sent; 211229997Sken int num_bios_done; 212229997Sken int send_complete; 213229997Sken int num_errors; 214229997Sken struct bintime ds_t0; 215229997Sken devstat_tag_type ds_tag_type; 216229997Sken devstat_trans_flags ds_trans_type; 217229997Sken uint64_t io_len; 218229997Sken uint64_t io_offset; 219229997Sken struct ctl_be_block_softc *softc; 220229997Sken struct ctl_be_block_lun *lun; 221265634Smav void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 222229997Sken}; 223229997Sken 224229997Skenstatic int cbb_num_threads = 14; 225229997SkenTUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 226229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 227229997Sken "CAM Target Layer Block Backend"); 228229997SkenSYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 229229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 230229997Sken 231229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 232229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 233229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 234229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 235229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 236229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 237229997Sken struct ctl_be_block_io *beio); 238229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 239229997Sken struct ctl_be_block_io *beio); 240229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 241229997Sken struct ctl_be_block_io *beio); 242265634Smavstatic void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 243265634Smav struct ctl_be_block_io *beio); 244229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 245229997Sken struct ctl_be_block_io *beio); 246229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 247229997Sken union ctl_io *io); 248229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 249229997Sken union ctl_io *io); 250229997Skenstatic void ctl_be_block_worker(void *context, int pending); 251229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 252229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 253229997Sken int flag, struct thread *td); 254229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 255229997Sken struct ctl_lun_req *req); 256229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 257229997Sken struct ctl_lun_req *req); 258229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 259229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 260229997Sken struct ctl_be_block_lun *be_lun, 261229997Sken struct ctl_lun_req *req); 262229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 263229997Sken struct ctl_lun_req *req); 264229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 265229997Sken struct ctl_lun_req *req); 266232604Straszstatic int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 267232604Strasz struct ctl_lun_req *req); 268232604Straszstatic int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 269232604Strasz struct ctl_lun_req *req); 270232604Straszstatic int ctl_be_block_modify(struct ctl_be_block_softc *softc, 271232604Strasz struct ctl_lun_req *req); 272229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 273229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 274229997Sken ctl_lun_config_status status); 275229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 276229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 277229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 278229997Skenint ctl_be_block_init(void); 279229997Sken 280229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 281229997Sken{ 282230334Sken .name = "block", 283230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 284230334Sken .init = ctl_be_block_init, 285230334Sken .data_submit = ctl_be_block_submit, 286230334Sken .data_move_done = ctl_be_block_move_done, 287230334Sken .config_read = ctl_be_block_config_read, 288230334Sken .config_write = ctl_be_block_config_write, 289230334Sken .ioctl = ctl_be_block_ioctl, 290230334Sken .lun_info = ctl_be_block_lun_info 291229997Sken}; 292229997Sken 293229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 294229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 295229997Sken 296265494Straszstatic uma_zone_t beio_zone; 297265494Strasz 298229997Skenstatic struct ctl_be_block_io * 299229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 300229997Sken{ 301229997Sken struct ctl_be_block_io *beio; 302229997Sken 303265494Strasz beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 304265494Strasz beio->softc = softc; 305229997Sken return (beio); 306229997Sken} 307229997Sken 308229997Skenstatic void 309229997Skenctl_free_beio(struct ctl_be_block_io *beio) 310229997Sken{ 311229997Sken int duplicate_free; 312229997Sken int i; 313229997Sken 314229997Sken duplicate_free = 0; 315229997Sken 316229997Sken for (i = 0; i < beio->num_segs; i++) { 317229997Sken if (beio->sg_segs[i].addr == NULL) 318229997Sken duplicate_free++; 319229997Sken 320229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 321229997Sken beio->sg_segs[i].addr = NULL; 322268151Smav 323268151Smav /* For compare we had two equal S/G lists. */ 324268151Smav if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 325268151Smav uma_zfree(beio->lun->lun_zone, 326268151Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 327268151Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 328268151Smav } 329229997Sken } 330229997Sken 331229997Sken if (duplicate_free > 0) { 332229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 333229997Sken duplicate_free, beio->num_segs); 334229997Sken } 335229997Sken 336265494Strasz uma_zfree(beio_zone, beio); 337229997Sken} 338229997Sken 339229997Skenstatic void 340229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 341229997Sken{ 342268549Smav union ctl_io *io = beio->io; 343229997Sken 344265634Smav if (beio->beio_cont != NULL) { 345265634Smav beio->beio_cont(beio); 346265634Smav } else { 347265634Smav ctl_free_beio(beio); 348268151Smav ctl_data_submit_done(io); 349265634Smav } 350229997Sken} 351229997Sken 352229997Skenstatic int 353229997Skenctl_be_block_move_done(union ctl_io *io) 354229997Sken{ 355229997Sken struct ctl_be_block_io *beio; 356229997Sken struct ctl_be_block_lun *be_lun; 357268151Smav struct ctl_lba_len_flags *lbalen; 358229997Sken#ifdef CTL_TIME_IO 359229997Sken struct bintime cur_bt; 360268151Smav#endif 361268151Smav int i; 362229997Sken 363268150Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 364229997Sken be_lun = beio->lun; 365229997Sken 366229997Sken DPRINTF("entered\n"); 367229997Sken 368229997Sken#ifdef CTL_TIME_IO 369229997Sken getbintime(&cur_bt); 370229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 371229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 372229997Sken io->io_hdr.num_dmas++; 373229997Sken#endif 374268151Smav io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 375229997Sken 376229997Sken /* 377229997Sken * We set status at this point for read commands, and write 378229997Sken * commands with errors. 379229997Sken */ 380268151Smav if ((io->io_hdr.port_status == 0) && 381268151Smav ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 382268151Smav ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 383268151Smav lbalen = ARGS(beio->io); 384268151Smav if (lbalen->flags & CTL_LLF_READ) { 385268151Smav ctl_set_success(&io->scsiio); 386268151Smav } else if (lbalen->flags & CTL_LLF_COMPARE) { 387268151Smav /* We have two data blocks ready for comparison. */ 388268151Smav for (i = 0; i < beio->num_segs; i++) { 389268151Smav if (memcmp(beio->sg_segs[i].addr, 390268151Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 391268151Smav beio->sg_segs[i].len) != 0) 392268151Smav break; 393268151Smav } 394268151Smav if (i < beio->num_segs) 395268151Smav ctl_set_sense(&io->scsiio, 396268151Smav /*current_error*/ 1, 397268151Smav /*sense_key*/ SSD_KEY_MISCOMPARE, 398268151Smav /*asc*/ 0x1D, 399268151Smav /*ascq*/ 0x00, 400268151Smav SSD_ELEM_NONE); 401268151Smav else 402268151Smav ctl_set_success(&io->scsiio); 403268151Smav } 404268151Smav } 405229997Sken else if ((io->io_hdr.port_status != 0) 406229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 407229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 408229997Sken /* 409229997Sken * For hardware error sense keys, the sense key 410229997Sken * specific value is defined to be a retry count, 411229997Sken * but we use it to pass back an internal FETD 412229997Sken * error code. XXX KDM Hopefully the FETD is only 413229997Sken * using 16 bits for an error code, since that's 414229997Sken * all the space we have in the sks field. 415229997Sken */ 416229997Sken ctl_set_internal_failure(&io->scsiio, 417229997Sken /*sks_valid*/ 1, 418229997Sken /*retry_count*/ 419229997Sken io->io_hdr.port_status); 420229997Sken } 421229997Sken 422229997Sken /* 423229997Sken * If this is a read, or a write with errors, it is done. 424229997Sken */ 425229997Sken if ((beio->bio_cmd == BIO_READ) 426229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 427229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 428229997Sken ctl_complete_beio(beio); 429229997Sken return (0); 430229997Sken } 431229997Sken 432229997Sken /* 433229997Sken * At this point, we have a write and the DMA completed 434229997Sken * successfully. We now have to queue it to the task queue to 435229997Sken * execute the backend I/O. That is because we do blocking 436229997Sken * memory allocations, and in the file backing case, blocking I/O. 437229997Sken * This move done routine is generally called in the SIM's 438229997Sken * interrupt context, and therefore we cannot block. 439229997Sken */ 440268549Smav mtx_lock(&be_lun->queue_lock); 441229997Sken /* 442229997Sken * XXX KDM make sure that links is okay to use at this point. 443229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 444229997Sken * or deal with resource allocation here. 445229997Sken */ 446229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 447268549Smav mtx_unlock(&be_lun->queue_lock); 448229997Sken 449229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 450229997Sken 451229997Sken return (0); 452229997Sken} 453229997Sken 454229997Skenstatic void 455229997Skenctl_be_block_biodone(struct bio *bio) 456229997Sken{ 457229997Sken struct ctl_be_block_io *beio; 458229997Sken struct ctl_be_block_lun *be_lun; 459229997Sken union ctl_io *io; 460262299Smav int error; 461229997Sken 462229997Sken beio = bio->bio_caller1; 463229997Sken be_lun = beio->lun; 464229997Sken io = beio->io; 465229997Sken 466229997Sken DPRINTF("entered\n"); 467229997Sken 468262299Smav error = bio->bio_error; 469268549Smav mtx_lock(&be_lun->io_lock); 470262299Smav if (error != 0) 471229997Sken beio->num_errors++; 472229997Sken 473229997Sken beio->num_bios_done++; 474229997Sken 475229997Sken /* 476229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 477229997Sken * during the free might cause it to complain. 478229997Sken */ 479229997Sken g_destroy_bio(bio); 480229997Sken 481229997Sken /* 482229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 483229997Sken * complete, then we're done. 484229997Sken */ 485229997Sken if ((beio->send_complete == 0) 486229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 487268549Smav mtx_unlock(&be_lun->io_lock); 488229997Sken return; 489229997Sken } 490229997Sken 491229997Sken /* 492229997Sken * At this point, we've verified that we are the last I/O to 493229997Sken * complete, so it's safe to drop the lock. 494229997Sken */ 495268549Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 496268549Smav beio->ds_tag_type, beio->ds_trans_type, 497268549Smav /*now*/ NULL, /*then*/&beio->ds_t0); 498268549Smav mtx_unlock(&be_lun->io_lock); 499229997Sken 500229997Sken /* 501229997Sken * If there are any errors from the backing device, we fail the 502229997Sken * entire I/O with a medium error. 503229997Sken */ 504229997Sken if (beio->num_errors > 0) { 505262299Smav if (error == EOPNOTSUPP) { 506262299Smav ctl_set_invalid_opcode(&io->scsiio); 507262299Smav } else if (beio->bio_cmd == BIO_FLUSH) { 508229997Sken /* XXX KDM is there is a better error here? */ 509229997Sken ctl_set_internal_failure(&io->scsiio, 510229997Sken /*sks_valid*/ 1, 511229997Sken /*retry_count*/ 0xbad2); 512229997Sken } else 513229997Sken ctl_set_medium_error(&io->scsiio); 514229997Sken ctl_complete_beio(beio); 515229997Sken return; 516229997Sken } 517229997Sken 518229997Sken /* 519268151Smav * If this is a write, a flush, a delete or verify, we're all done. 520229997Sken * If this is a read, we can now send the data to the user. 521229997Sken */ 522229997Sken if ((beio->bio_cmd == BIO_WRITE) 523265634Smav || (beio->bio_cmd == BIO_FLUSH) 524268151Smav || (beio->bio_cmd == BIO_DELETE) 525268151Smav || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 526229997Sken ctl_set_success(&io->scsiio); 527229997Sken ctl_complete_beio(beio); 528229997Sken } else { 529229997Sken#ifdef CTL_TIME_IO 530229997Sken getbintime(&io->io_hdr.dma_start_bt); 531229997Sken#endif 532229997Sken ctl_datamove(io); 533229997Sken } 534229997Sken} 535229997Sken 536229997Skenstatic void 537229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 538229997Sken struct ctl_be_block_io *beio) 539229997Sken{ 540268549Smav union ctl_io *io = beio->io; 541229997Sken struct mount *mountpoint; 542241896Skib int error, lock_flags; 543229997Sken 544229997Sken DPRINTF("entered\n"); 545229997Sken 546268549Smav binuptime(&beio->ds_t0); 547268549Smav mtx_lock(&be_lun->io_lock); 548268549Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 549268549Smav mtx_unlock(&be_lun->io_lock); 550229997Sken 551268549Smav (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 552229997Sken 553229997Sken if (MNT_SHARED_WRITES(mountpoint) 554229997Sken || ((mountpoint == NULL) 555229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 556229997Sken lock_flags = LK_SHARED; 557229997Sken else 558229997Sken lock_flags = LK_EXCLUSIVE; 559229997Sken 560229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 561229997Sken 562229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 563229997Sken VOP_UNLOCK(be_lun->vn, 0); 564229997Sken 565229997Sken vn_finished_write(mountpoint); 566229997Sken 567268549Smav mtx_lock(&be_lun->io_lock); 568268549Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 569268549Smav beio->ds_tag_type, beio->ds_trans_type, 570268549Smav /*now*/ NULL, /*then*/&beio->ds_t0); 571268549Smav mtx_unlock(&be_lun->io_lock); 572268549Smav 573229997Sken if (error == 0) 574229997Sken ctl_set_success(&io->scsiio); 575229997Sken else { 576229997Sken /* XXX KDM is there is a better error here? */ 577229997Sken ctl_set_internal_failure(&io->scsiio, 578229997Sken /*sks_valid*/ 1, 579229997Sken /*retry_count*/ 0xbad1); 580229997Sken } 581229997Sken 582229997Sken ctl_complete_beio(beio); 583229997Sken} 584229997Sken 585260817SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 586260817SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 587260817SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 588260817SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 589229997Sken 590229997Skenstatic void 591229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 592229997Sken struct ctl_be_block_io *beio) 593229997Sken{ 594229997Sken struct ctl_be_block_filedata *file_data; 595229997Sken union ctl_io *io; 596229997Sken struct uio xuio; 597229997Sken struct iovec *xiovec; 598241896Skib int flags; 599229997Sken int error, i; 600229997Sken 601229997Sken DPRINTF("entered\n"); 602229997Sken 603229997Sken file_data = &be_lun->backend.file; 604229997Sken io = beio->io; 605229997Sken flags = beio->bio_flags; 606229997Sken 607268151Smav bzero(&xuio, sizeof(xuio)); 608229997Sken if (beio->bio_cmd == BIO_READ) { 609229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 610268151Smav xuio.uio_rw = UIO_READ; 611229997Sken } else { 612229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 613268151Smav xuio.uio_rw = UIO_WRITE; 614229997Sken } 615229997Sken xuio.uio_offset = beio->io_offset; 616229997Sken xuio.uio_resid = beio->io_len; 617229997Sken xuio.uio_segflg = UIO_SYSSPACE; 618229997Sken xuio.uio_iov = beio->xiovecs; 619229997Sken xuio.uio_iovcnt = beio->num_segs; 620229997Sken xuio.uio_td = curthread; 621229997Sken 622229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 623229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 624229997Sken xiovec->iov_len = beio->sg_segs[i].len; 625229997Sken } 626229997Sken 627268549Smav binuptime(&beio->ds_t0); 628268549Smav mtx_lock(&be_lun->io_lock); 629268549Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 630268549Smav mtx_unlock(&be_lun->io_lock); 631268549Smav 632229997Sken if (beio->bio_cmd == BIO_READ) { 633229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 634229997Sken 635229997Sken /* 636229997Sken * UFS pays attention to IO_DIRECT for reads. If the 637229997Sken * DIRECTIO option is configured into the kernel, it calls 638229997Sken * ffs_rawread(). But that only works for single-segment 639229997Sken * uios with user space addresses. In our case, with a 640229997Sken * kernel uio, it still reads into the buffer cache, but it 641229997Sken * will just try to release the buffer from the cache later 642229997Sken * on in ffs_read(). 643229997Sken * 644229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 645229997Sken * 646229997Sken * UFS does not pay attention to IO_SYNC for reads. 647229997Sken * 648229997Sken * ZFS pays attention to IO_SYNC (which translates into the 649229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 650229997Sken * attempts to sync the file before reading. 651229997Sken * 652229997Sken * So, to attempt to provide some barrier semantics in the 653229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 654229997Sken */ 655229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 656229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 657229997Sken 658229997Sken VOP_UNLOCK(be_lun->vn, 0); 659268151Smav SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 660229997Sken } else { 661229997Sken struct mount *mountpoint; 662229997Sken int lock_flags; 663229997Sken 664229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 665229997Sken 666229997Sken if (MNT_SHARED_WRITES(mountpoint) 667229997Sken || ((mountpoint == NULL) 668229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 669229997Sken lock_flags = LK_SHARED; 670229997Sken else 671229997Sken lock_flags = LK_EXCLUSIVE; 672229997Sken 673229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 674229997Sken 675229997Sken /* 676229997Sken * UFS pays attention to IO_DIRECT for writes. The write 677229997Sken * is done asynchronously. (Normally the write would just 678229997Sken * get put into cache. 679229997Sken * 680229997Sken * UFS pays attention to IO_SYNC for writes. It will 681229997Sken * attempt to write the buffer out synchronously if that 682229997Sken * flag is set. 683229997Sken * 684229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 685229997Sken * 686229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 687229997Sken * for writes. It will flush the transaction from the 688229997Sken * cache before returning. 689229997Sken * 690229997Sken * So if we've got the BIO_ORDERED flag set, we want 691229997Sken * IO_SYNC in either the UFS or ZFS case. 692229997Sken */ 693229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 694229997Sken IO_SYNC : 0, file_data->cred); 695229997Sken VOP_UNLOCK(be_lun->vn, 0); 696229997Sken 697229997Sken vn_finished_write(mountpoint); 698268151Smav SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 699229997Sken } 700229997Sken 701268549Smav mtx_lock(&be_lun->io_lock); 702268549Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 703268549Smav beio->ds_tag_type, beio->ds_trans_type, 704268549Smav /*now*/ NULL, /*then*/&beio->ds_t0); 705268549Smav mtx_unlock(&be_lun->io_lock); 706268549Smav 707229997Sken /* 708229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 709229997Sken * return the I/O to the user. 710229997Sken */ 711229997Sken if (error != 0) { 712229997Sken char path_str[32]; 713229997Sken 714229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 715229997Sken /* 716229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 717229997Sken * filesystem fills up. What kind of SCSI error should we 718229997Sken * return for that? 719229997Sken */ 720229997Sken printf("%s%s command returned errno %d\n", path_str, 721229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 722229997Sken ctl_set_medium_error(&io->scsiio); 723229997Sken ctl_complete_beio(beio); 724229997Sken return; 725229997Sken } 726229997Sken 727229997Sken /* 728269226Smav * If this is a write or a verify, we're all done. 729229997Sken * If this is a read, we can now send the data to the user. 730229997Sken */ 731269226Smav if ((beio->bio_cmd == BIO_WRITE) || 732269226Smav (ARGS(io)->flags & CTL_LLF_VERIFY)) { 733229997Sken ctl_set_success(&io->scsiio); 734229997Sken ctl_complete_beio(beio); 735229997Sken } else { 736229997Sken#ifdef CTL_TIME_IO 737229997Sken getbintime(&io->io_hdr.dma_start_bt); 738229997Sken#endif 739229997Sken ctl_datamove(io); 740229997Sken } 741229997Sken} 742229997Sken 743229997Skenstatic void 744269429Smavctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 745269429Smav struct ctl_be_block_io *beio) 746269429Smav{ 747269429Smav struct ctl_be_block_devdata *dev_data; 748269429Smav union ctl_io *io; 749269429Smav struct uio xuio; 750269429Smav struct iovec *xiovec; 751269429Smav int flags; 752269429Smav int error, i; 753269429Smav 754269429Smav DPRINTF("entered\n"); 755269429Smav 756269429Smav dev_data = &be_lun->backend.dev; 757269429Smav io = beio->io; 758269429Smav flags = beio->bio_flags; 759269429Smav 760269429Smav bzero(&xuio, sizeof(xuio)); 761269429Smav if (beio->bio_cmd == BIO_READ) { 762269429Smav SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 763269429Smav xuio.uio_rw = UIO_READ; 764269429Smav } else { 765269429Smav SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 766269429Smav xuio.uio_rw = UIO_WRITE; 767269429Smav } 768269429Smav xuio.uio_offset = beio->io_offset; 769269429Smav xuio.uio_resid = beio->io_len; 770269429Smav xuio.uio_segflg = UIO_SYSSPACE; 771269429Smav xuio.uio_iov = beio->xiovecs; 772269429Smav xuio.uio_iovcnt = beio->num_segs; 773269429Smav xuio.uio_td = curthread; 774269429Smav 775269429Smav for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 776269429Smav xiovec->iov_base = beio->sg_segs[i].addr; 777269429Smav xiovec->iov_len = beio->sg_segs[i].len; 778269429Smav } 779269429Smav 780269429Smav binuptime(&beio->ds_t0); 781269429Smav mtx_lock(&be_lun->io_lock); 782269429Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 783269429Smav mtx_unlock(&be_lun->io_lock); 784269429Smav 785269429Smav if (beio->bio_cmd == BIO_READ) { 786269429Smav error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, 0); 787269429Smav SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 788269429Smav } else { 789269429Smav error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, 0); 790269429Smav SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 791269429Smav } 792269429Smav 793269429Smav mtx_lock(&be_lun->io_lock); 794269429Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 795269429Smav beio->ds_tag_type, beio->ds_trans_type, 796269429Smav /*now*/ NULL, /*then*/&beio->ds_t0); 797269429Smav mtx_unlock(&be_lun->io_lock); 798269429Smav 799269429Smav /* 800269429Smav * If we got an error, set the sense data to "MEDIUM ERROR" and 801269429Smav * return the I/O to the user. 802269429Smav */ 803269429Smav if (error != 0) { 804269429Smav ctl_set_medium_error(&io->scsiio); 805269429Smav ctl_complete_beio(beio); 806269429Smav return; 807269429Smav } 808269429Smav 809269429Smav /* 810269429Smav * If this is a write or a verify, we're all done. 811269429Smav * If this is a read, we can now send the data to the user. 812269429Smav */ 813269429Smav if ((beio->bio_cmd == BIO_WRITE) || 814269429Smav (ARGS(io)->flags & CTL_LLF_VERIFY)) { 815269429Smav ctl_set_success(&io->scsiio); 816269429Smav ctl_complete_beio(beio); 817269429Smav } else { 818269429Smav#ifdef CTL_TIME_IO 819269429Smav getbintime(&io->io_hdr.dma_start_bt); 820269429Smav#endif 821269429Smav ctl_datamove(io); 822269429Smav } 823269429Smav} 824269429Smav 825269429Smavstatic void 826229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 827229997Sken struct ctl_be_block_io *beio) 828229997Sken{ 829229997Sken struct bio *bio; 830229997Sken union ctl_io *io; 831229997Sken struct ctl_be_block_devdata *dev_data; 832229997Sken 833229997Sken dev_data = &be_lun->backend.dev; 834229997Sken io = beio->io; 835229997Sken 836229997Sken DPRINTF("entered\n"); 837229997Sken 838229997Sken /* This can't fail, it's a blocking allocation. */ 839229997Sken bio = g_alloc_bio(); 840229997Sken 841229997Sken bio->bio_cmd = BIO_FLUSH; 842229997Sken bio->bio_flags |= BIO_ORDERED; 843229997Sken bio->bio_dev = dev_data->cdev; 844229997Sken bio->bio_offset = 0; 845229997Sken bio->bio_data = 0; 846229997Sken bio->bio_done = ctl_be_block_biodone; 847229997Sken bio->bio_caller1 = beio; 848229997Sken bio->bio_pblkno = 0; 849229997Sken 850229997Sken /* 851229997Sken * We don't need to acquire the LUN lock here, because we are only 852229997Sken * sending one bio, and so there is no other context to synchronize 853229997Sken * with. 854229997Sken */ 855229997Sken beio->num_bios_sent = 1; 856229997Sken beio->send_complete = 1; 857229997Sken 858229997Sken binuptime(&beio->ds_t0); 859268549Smav mtx_lock(&be_lun->io_lock); 860229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 861268549Smav mtx_unlock(&be_lun->io_lock); 862229997Sken 863229997Sken (*dev_data->csw->d_strategy)(bio); 864229997Sken} 865229997Sken 866229997Skenstatic void 867265634Smavctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 868265634Smav struct ctl_be_block_io *beio, 869265634Smav uint64_t off, uint64_t len, int last) 870265634Smav{ 871265634Smav struct bio *bio; 872265634Smav struct ctl_be_block_devdata *dev_data; 873265634Smav uint64_t maxlen; 874265634Smav 875265634Smav dev_data = &be_lun->backend.dev; 876265634Smav maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 877265634Smav while (len > 0) { 878265634Smav bio = g_alloc_bio(); 879265634Smav bio->bio_cmd = BIO_DELETE; 880265634Smav bio->bio_flags |= beio->bio_flags; 881265634Smav bio->bio_dev = dev_data->cdev; 882265634Smav bio->bio_offset = off; 883265634Smav bio->bio_length = MIN(len, maxlen); 884265634Smav bio->bio_data = 0; 885265634Smav bio->bio_done = ctl_be_block_biodone; 886265634Smav bio->bio_caller1 = beio; 887265634Smav bio->bio_pblkno = off / be_lun->blocksize; 888265634Smav 889265634Smav off += bio->bio_length; 890265634Smav len -= bio->bio_length; 891265634Smav 892268549Smav mtx_lock(&be_lun->io_lock); 893265634Smav beio->num_bios_sent++; 894265634Smav if (last && len == 0) 895265634Smav beio->send_complete = 1; 896268549Smav mtx_unlock(&be_lun->io_lock); 897265634Smav 898265634Smav (*dev_data->csw->d_strategy)(bio); 899265634Smav } 900265634Smav} 901265634Smav 902265634Smavstatic void 903265634Smavctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 904265634Smav struct ctl_be_block_io *beio) 905265634Smav{ 906265634Smav union ctl_io *io; 907265634Smav struct ctl_be_block_devdata *dev_data; 908268149Smav struct ctl_ptr_len_flags *ptrlen; 909265634Smav struct scsi_unmap_desc *buf, *end; 910265634Smav uint64_t len; 911265634Smav 912265634Smav dev_data = &be_lun->backend.dev; 913265634Smav io = beio->io; 914265634Smav 915265634Smav DPRINTF("entered\n"); 916265634Smav 917265634Smav binuptime(&beio->ds_t0); 918268549Smav mtx_lock(&be_lun->io_lock); 919265634Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 920268549Smav mtx_unlock(&be_lun->io_lock); 921265634Smav 922265634Smav if (beio->io_offset == -1) { 923265634Smav beio->io_len = 0; 924268149Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 925268149Smav buf = (struct scsi_unmap_desc *)ptrlen->ptr; 926268149Smav end = buf + ptrlen->len / sizeof(*buf); 927265634Smav for (; buf < end; buf++) { 928265634Smav len = (uint64_t)scsi_4btoul(buf->length) * 929265634Smav be_lun->blocksize; 930265634Smav beio->io_len += len; 931265634Smav ctl_be_block_unmap_dev_range(be_lun, beio, 932265634Smav scsi_8btou64(buf->lba) * be_lun->blocksize, len, 933265634Smav (end - buf < 2) ? TRUE : FALSE); 934265634Smav } 935265634Smav } else 936265634Smav ctl_be_block_unmap_dev_range(be_lun, beio, 937265634Smav beio->io_offset, beio->io_len, TRUE); 938265634Smav} 939265634Smav 940265634Smavstatic void 941229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 942229997Sken struct ctl_be_block_io *beio) 943229997Sken{ 944268549Smav TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 945229997Sken int i; 946229997Sken struct bio *bio; 947229997Sken struct ctl_be_block_devdata *dev_data; 948229997Sken off_t cur_offset; 949229997Sken int max_iosize; 950229997Sken 951229997Sken DPRINTF("entered\n"); 952229997Sken 953229997Sken dev_data = &be_lun->backend.dev; 954229997Sken 955229997Sken /* 956229997Sken * We have to limit our I/O size to the maximum supported by the 957229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 958229997Sken * set it properly, use DFLTPHYS. 959229997Sken */ 960229997Sken max_iosize = dev_data->cdev->si_iosize_max; 961229997Sken if (max_iosize < PAGE_SIZE) 962229997Sken max_iosize = DFLTPHYS; 963229997Sken 964229997Sken cur_offset = beio->io_offset; 965229997Sken for (i = 0; i < beio->num_segs; i++) { 966229997Sken size_t cur_size; 967229997Sken uint8_t *cur_ptr; 968229997Sken 969229997Sken cur_size = beio->sg_segs[i].len; 970229997Sken cur_ptr = beio->sg_segs[i].addr; 971229997Sken 972229997Sken while (cur_size > 0) { 973229997Sken /* This can't fail, it's a blocking allocation. */ 974229997Sken bio = g_alloc_bio(); 975229997Sken 976229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 977229997Sken 978229997Sken bio->bio_cmd = beio->bio_cmd; 979229997Sken bio->bio_flags |= beio->bio_flags; 980229997Sken bio->bio_dev = dev_data->cdev; 981229997Sken bio->bio_caller1 = beio; 982229997Sken bio->bio_length = min(cur_size, max_iosize); 983229997Sken bio->bio_offset = cur_offset; 984229997Sken bio->bio_data = cur_ptr; 985229997Sken bio->bio_done = ctl_be_block_biodone; 986229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 987229997Sken 988229997Sken cur_offset += bio->bio_length; 989229997Sken cur_ptr += bio->bio_length; 990229997Sken cur_size -= bio->bio_length; 991229997Sken 992268549Smav TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 993229997Sken beio->num_bios_sent++; 994229997Sken } 995229997Sken } 996268549Smav binuptime(&beio->ds_t0); 997268549Smav mtx_lock(&be_lun->io_lock); 998268549Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 999268549Smav beio->send_complete = 1; 1000268549Smav mtx_unlock(&be_lun->io_lock); 1001268549Smav 1002268549Smav /* 1003268549Smav * Fire off all allocated requests! 1004268549Smav */ 1005268549Smav while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1006268549Smav TAILQ_REMOVE(&queue, bio, bio_queue); 1007268549Smav (*dev_data->csw->d_strategy)(bio); 1008268549Smav } 1009229997Sken} 1010229997Sken 1011229997Skenstatic void 1012265634Smavctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1013265634Smav{ 1014265634Smav union ctl_io *io; 1015265634Smav 1016265634Smav io = beio->io; 1017265634Smav ctl_free_beio(beio); 1018268261Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1019268261Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1020268261Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1021265634Smav ctl_config_write_done(io); 1022265634Smav return; 1023265634Smav } 1024265634Smav 1025265634Smav ctl_be_block_config_write(io); 1026265634Smav} 1027265634Smav 1028265634Smavstatic void 1029265634Smavctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1030265634Smav union ctl_io *io) 1031265634Smav{ 1032265634Smav struct ctl_be_block_io *beio; 1033265634Smav struct ctl_be_block_softc *softc; 1034268149Smav struct ctl_lba_len_flags *lbalen; 1035265634Smav uint64_t len_left, lba; 1036265634Smav int i, seglen; 1037265634Smav uint8_t *buf, *end; 1038265634Smav 1039265634Smav DPRINTF("entered\n"); 1040265634Smav 1041268150Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1042265634Smav softc = be_lun->softc; 1043268151Smav lbalen = ARGS(beio->io); 1044265634Smav 1045270108Smav if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR) || 1046270108Smav (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1047265634Smav ctl_free_beio(beio); 1048265634Smav ctl_set_invalid_field(&io->scsiio, 1049265634Smav /*sks_valid*/ 1, 1050265634Smav /*command*/ 1, 1051265634Smav /*field*/ 1, 1052265634Smav /*bit_valid*/ 0, 1053265634Smav /*bit*/ 0); 1054265634Smav ctl_config_write_done(io); 1055265634Smav return; 1056265634Smav } 1057265634Smav 1058265634Smav /* 1059265634Smav * If the I/O came down with an ordered or head of queue tag, set 1060265634Smav * the BIO_ORDERED attribute. For head of queue tags, that's 1061265634Smav * pretty much the best we can do. 1062265634Smav */ 1063265634Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1064265634Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1065265634Smav beio->bio_flags = BIO_ORDERED; 1066265634Smav 1067265634Smav switch (io->scsiio.tag_type) { 1068265634Smav case CTL_TAG_ORDERED: 1069265634Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1070265634Smav break; 1071265634Smav case CTL_TAG_HEAD_OF_QUEUE: 1072265634Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1073265634Smav break; 1074265634Smav case CTL_TAG_UNTAGGED: 1075265634Smav case CTL_TAG_SIMPLE: 1076265634Smav case CTL_TAG_ACA: 1077265634Smav default: 1078265634Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1079265634Smav break; 1080265634Smav } 1081265634Smav 1082270108Smav if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1083268149Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 1084268149Smav beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1085265634Smav beio->bio_cmd = BIO_DELETE; 1086265634Smav beio->ds_trans_type = DEVSTAT_FREE; 1087265634Smav 1088265634Smav be_lun->unmap(be_lun, beio); 1089265634Smav return; 1090265634Smav } 1091265634Smav 1092265634Smav beio->bio_cmd = BIO_WRITE; 1093265634Smav beio->ds_trans_type = DEVSTAT_WRITE; 1094265634Smav 1095265634Smav DPRINTF("WRITE SAME at LBA %jx len %u\n", 1096268149Smav (uintmax_t)lbalen->lba, lbalen->len); 1097265634Smav 1098268149Smav len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1099265634Smav for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1100265634Smav 1101265634Smav /* 1102265634Smav * Setup the S/G entry for this chunk. 1103265634Smav */ 1104265642Smav seglen = MIN(CTLBLK_MAX_SEG, len_left); 1105265634Smav seglen -= seglen % be_lun->blocksize; 1106265634Smav beio->sg_segs[i].len = seglen; 1107265634Smav beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1108265634Smav 1109265634Smav DPRINTF("segment %d addr %p len %zd\n", i, 1110265634Smav beio->sg_segs[i].addr, beio->sg_segs[i].len); 1111265634Smav 1112265634Smav beio->num_segs++; 1113265634Smav len_left -= seglen; 1114265634Smav 1115265634Smav buf = beio->sg_segs[i].addr; 1116265634Smav end = buf + seglen; 1117265634Smav for (; buf < end; buf += be_lun->blocksize) { 1118265634Smav memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1119268149Smav if (lbalen->flags & SWS_LBDATA) 1120268149Smav scsi_ulto4b(lbalen->lba + lba, buf); 1121265634Smav lba++; 1122265634Smav } 1123265634Smav } 1124265634Smav 1125268149Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 1126265634Smav beio->io_len = lba * be_lun->blocksize; 1127265634Smav 1128265634Smav /* We can not do all in one run. Correct and schedule rerun. */ 1129265634Smav if (len_left > 0) { 1130268149Smav lbalen->lba += lba; 1131268149Smav lbalen->len -= lba; 1132265634Smav beio->beio_cont = ctl_be_block_cw_done_ws; 1133265634Smav } 1134265634Smav 1135265634Smav be_lun->dispatch(be_lun, beio); 1136265634Smav} 1137265634Smav 1138265634Smavstatic void 1139265634Smavctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1140265634Smav union ctl_io *io) 1141265634Smav{ 1142265634Smav struct ctl_be_block_io *beio; 1143265634Smav struct ctl_be_block_softc *softc; 1144268149Smav struct ctl_ptr_len_flags *ptrlen; 1145265634Smav 1146265634Smav DPRINTF("entered\n"); 1147265634Smav 1148268150Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1149265634Smav softc = be_lun->softc; 1150268149Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1151265634Smav 1152270108Smav if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1153265634Smav ctl_free_beio(beio); 1154265634Smav ctl_set_invalid_field(&io->scsiio, 1155265634Smav /*sks_valid*/ 0, 1156265634Smav /*command*/ 1, 1157265634Smav /*field*/ 0, 1158265634Smav /*bit_valid*/ 0, 1159265634Smav /*bit*/ 0); 1160265634Smav ctl_config_write_done(io); 1161265634Smav return; 1162265634Smav } 1163265634Smav 1164265634Smav /* 1165265634Smav * If the I/O came down with an ordered or head of queue tag, set 1166265634Smav * the BIO_ORDERED attribute. For head of queue tags, that's 1167265634Smav * pretty much the best we can do. 1168265634Smav */ 1169265634Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1170265634Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1171265634Smav beio->bio_flags = BIO_ORDERED; 1172265634Smav 1173265634Smav switch (io->scsiio.tag_type) { 1174265634Smav case CTL_TAG_ORDERED: 1175265634Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1176265634Smav break; 1177265634Smav case CTL_TAG_HEAD_OF_QUEUE: 1178265634Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1179265634Smav break; 1180265634Smav case CTL_TAG_UNTAGGED: 1181265634Smav case CTL_TAG_SIMPLE: 1182265634Smav case CTL_TAG_ACA: 1183265634Smav default: 1184265634Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1185265634Smav break; 1186265634Smav } 1187265634Smav 1188265634Smav beio->io_len = 0; 1189265634Smav beio->io_offset = -1; 1190265634Smav 1191265634Smav beio->bio_cmd = BIO_DELETE; 1192265634Smav beio->ds_trans_type = DEVSTAT_FREE; 1193265634Smav 1194268149Smav DPRINTF("UNMAP\n"); 1195265634Smav 1196265634Smav be_lun->unmap(be_lun, beio); 1197265634Smav} 1198265634Smav 1199265634Smavstatic void 1200265634Smavctl_be_block_cw_done(struct ctl_be_block_io *beio) 1201265634Smav{ 1202265634Smav union ctl_io *io; 1203265634Smav 1204265634Smav io = beio->io; 1205265634Smav ctl_free_beio(beio); 1206265634Smav ctl_config_write_done(io); 1207265634Smav} 1208265634Smav 1209265634Smavstatic void 1210229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1211229997Sken union ctl_io *io) 1212229997Sken{ 1213229997Sken struct ctl_be_block_io *beio; 1214229997Sken struct ctl_be_block_softc *softc; 1215229997Sken 1216229997Sken DPRINTF("entered\n"); 1217229997Sken 1218229997Sken softc = be_lun->softc; 1219229997Sken beio = ctl_alloc_beio(softc); 1220229997Sken beio->io = io; 1221229997Sken beio->lun = be_lun; 1222265634Smav beio->beio_cont = ctl_be_block_cw_done; 1223268150Smav PRIV(io)->ptr = (void *)beio; 1224229997Sken 1225229997Sken switch (io->scsiio.cdb[0]) { 1226229997Sken case SYNCHRONIZE_CACHE: 1227229997Sken case SYNCHRONIZE_CACHE_16: 1228249194Strasz beio->bio_cmd = BIO_FLUSH; 1229229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 1230229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1231229997Sken beio->io_len = 0; 1232229997Sken be_lun->lun_flush(be_lun, beio); 1233229997Sken break; 1234265634Smav case WRITE_SAME_10: 1235265634Smav case WRITE_SAME_16: 1236265634Smav ctl_be_block_cw_dispatch_ws(be_lun, io); 1237265634Smav break; 1238265634Smav case UNMAP: 1239265634Smav ctl_be_block_cw_dispatch_unmap(be_lun, io); 1240265634Smav break; 1241229997Sken default: 1242229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1243229997Sken break; 1244229997Sken } 1245229997Sken} 1246229997Sken 1247260817SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1248260817SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1249260817SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1250260817SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1251229997Sken 1252229997Skenstatic void 1253265642Smavctl_be_block_next(struct ctl_be_block_io *beio) 1254265642Smav{ 1255265642Smav struct ctl_be_block_lun *be_lun; 1256265642Smav union ctl_io *io; 1257265642Smav 1258265642Smav io = beio->io; 1259265642Smav be_lun = beio->lun; 1260265642Smav ctl_free_beio(beio); 1261268261Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1262268261Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1263268261Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1264268151Smav ctl_data_submit_done(io); 1265265642Smav return; 1266265642Smav } 1267265642Smav 1268265642Smav io->io_hdr.status &= ~CTL_STATUS_MASK; 1269265642Smav io->io_hdr.status |= CTL_STATUS_NONE; 1270265642Smav 1271268549Smav mtx_lock(&be_lun->queue_lock); 1272265642Smav /* 1273265642Smav * XXX KDM make sure that links is okay to use at this point. 1274265642Smav * Otherwise, we either need to add another field to ctl_io_hdr, 1275265642Smav * or deal with resource allocation here. 1276265642Smav */ 1277265642Smav STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1278268549Smav mtx_unlock(&be_lun->queue_lock); 1279265642Smav 1280265642Smav taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1281265642Smav} 1282265642Smav 1283265642Smavstatic void 1284229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1285229997Sken union ctl_io *io) 1286229997Sken{ 1287229997Sken struct ctl_be_block_io *beio; 1288229997Sken struct ctl_be_block_softc *softc; 1289268151Smav struct ctl_lba_len_flags *lbalen; 1290268150Smav struct ctl_ptr_len_flags *bptrlen; 1291268150Smav uint64_t len_left, lbas; 1292229997Sken int i; 1293229997Sken 1294229997Sken softc = be_lun->softc; 1295229997Sken 1296229997Sken DPRINTF("entered\n"); 1297229997Sken 1298268151Smav lbalen = ARGS(io); 1299268151Smav if (lbalen->flags & CTL_LLF_WRITE) { 1300268151Smav SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1301268151Smav } else { 1302229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1303229997Sken } 1304229997Sken 1305229997Sken beio = ctl_alloc_beio(softc); 1306229997Sken beio->io = io; 1307229997Sken beio->lun = be_lun; 1308268150Smav bptrlen = PRIV(io); 1309268150Smav bptrlen->ptr = (void *)beio; 1310229997Sken 1311229997Sken /* 1312229997Sken * If the I/O came down with an ordered or head of queue tag, set 1313229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 1314229997Sken * pretty much the best we can do. 1315229997Sken * 1316229997Sken * XXX KDM we don't have a great way to easily know about the FUA 1317229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 1318229997Sken * pass that knowledge to the backend), and in any case we would 1319229997Sken * need to determine how to handle it. 1320229997Sken */ 1321229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1322229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1323229997Sken beio->bio_flags = BIO_ORDERED; 1324229997Sken 1325229997Sken switch (io->scsiio.tag_type) { 1326229997Sken case CTL_TAG_ORDERED: 1327229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1328229997Sken break; 1329229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1330229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1331229997Sken break; 1332229997Sken case CTL_TAG_UNTAGGED: 1333229997Sken case CTL_TAG_SIMPLE: 1334229997Sken case CTL_TAG_ACA: 1335229997Sken default: 1336229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1337229997Sken break; 1338229997Sken } 1339229997Sken 1340268151Smav if (lbalen->flags & CTL_LLF_WRITE) { 1341268151Smav beio->bio_cmd = BIO_WRITE; 1342268151Smav beio->ds_trans_type = DEVSTAT_WRITE; 1343268151Smav } else { 1344229997Sken beio->bio_cmd = BIO_READ; 1345229997Sken beio->ds_trans_type = DEVSTAT_READ; 1346229997Sken } 1347229997Sken 1348265642Smav DPRINTF("%s at LBA %jx len %u @%ju\n", 1349229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1350268150Smav (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1351268151Smav if (lbalen->flags & CTL_LLF_COMPARE) 1352268151Smav lbas = CTLBLK_HALF_IO_SIZE; 1353268151Smav else 1354268151Smav lbas = CTLBLK_MAX_IO_SIZE; 1355268151Smav lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1356268150Smav beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1357268150Smav beio->io_len = lbas * be_lun->blocksize; 1358268150Smav bptrlen->len += lbas; 1359229997Sken 1360265642Smav for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1361265642Smav KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1362265642Smav i, CTLBLK_MAX_SEGS)); 1363229997Sken 1364229997Sken /* 1365229997Sken * Setup the S/G entry for this chunk. 1366229997Sken */ 1367265642Smav beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1368229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1369229997Sken 1370229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1371229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1372229997Sken 1373268151Smav /* Set up second segment for compare operation. */ 1374268151Smav if (lbalen->flags & CTL_LLF_COMPARE) { 1375268151Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1376268151Smav beio->sg_segs[i].len; 1377268151Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1378268151Smav uma_zalloc(be_lun->lun_zone, M_WAITOK); 1379268151Smav } 1380268151Smav 1381229997Sken beio->num_segs++; 1382229997Sken len_left -= beio->sg_segs[i].len; 1383229997Sken } 1384268150Smav if (bptrlen->len < lbalen->len) 1385265642Smav beio->beio_cont = ctl_be_block_next; 1386265642Smav io->scsiio.be_move_done = ctl_be_block_move_done; 1387268151Smav /* For compare we have separate S/G lists for read and datamove. */ 1388268151Smav if (lbalen->flags & CTL_LLF_COMPARE) 1389268151Smav io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1390268151Smav else 1391268151Smav io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1392265642Smav io->scsiio.kern_data_len = beio->io_len; 1393265642Smav io->scsiio.kern_data_resid = 0; 1394265642Smav io->scsiio.kern_sg_entries = beio->num_segs; 1395265642Smav io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1396229997Sken 1397229997Sken /* 1398229997Sken * For the read case, we need to read the data into our buffers and 1399229997Sken * then we can send it back to the user. For the write case, we 1400229997Sken * need to get the data from the user first. 1401229997Sken */ 1402229997Sken if (beio->bio_cmd == BIO_READ) { 1403229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1404229997Sken be_lun->dispatch(be_lun, beio); 1405229997Sken } else { 1406229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1407229997Sken#ifdef CTL_TIME_IO 1408229997Sken getbintime(&io->io_hdr.dma_start_bt); 1409229997Sken#endif 1410229997Sken ctl_datamove(io); 1411229997Sken } 1412229997Sken} 1413229997Sken 1414229997Skenstatic void 1415229997Skenctl_be_block_worker(void *context, int pending) 1416229997Sken{ 1417229997Sken struct ctl_be_block_lun *be_lun; 1418229997Sken struct ctl_be_block_softc *softc; 1419229997Sken union ctl_io *io; 1420229997Sken 1421229997Sken be_lun = (struct ctl_be_block_lun *)context; 1422229997Sken softc = be_lun->softc; 1423229997Sken 1424229997Sken DPRINTF("entered\n"); 1425229997Sken 1426268549Smav mtx_lock(&be_lun->queue_lock); 1427229997Sken for (;;) { 1428229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1429229997Sken if (io != NULL) { 1430229997Sken struct ctl_be_block_io *beio; 1431229997Sken 1432229997Sken DPRINTF("datamove queue\n"); 1433229997Sken 1434229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1435229997Sken ctl_io_hdr, links); 1436229997Sken 1437268549Smav mtx_unlock(&be_lun->queue_lock); 1438229997Sken 1439268150Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1440229997Sken 1441229997Sken be_lun->dispatch(be_lun, beio); 1442229997Sken 1443268549Smav mtx_lock(&be_lun->queue_lock); 1444229997Sken continue; 1445229997Sken } 1446229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1447229997Sken if (io != NULL) { 1448229997Sken 1449229997Sken DPRINTF("config write queue\n"); 1450229997Sken 1451229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1452229997Sken ctl_io_hdr, links); 1453229997Sken 1454268549Smav mtx_unlock(&be_lun->queue_lock); 1455229997Sken 1456229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1457229997Sken 1458268549Smav mtx_lock(&be_lun->queue_lock); 1459229997Sken continue; 1460229997Sken } 1461229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1462229997Sken if (io != NULL) { 1463229997Sken DPRINTF("input queue\n"); 1464229997Sken 1465229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1466229997Sken ctl_io_hdr, links); 1467268549Smav mtx_unlock(&be_lun->queue_lock); 1468229997Sken 1469229997Sken /* 1470229997Sken * We must drop the lock, since this routine and 1471229997Sken * its children may sleep. 1472229997Sken */ 1473229997Sken ctl_be_block_dispatch(be_lun, io); 1474229997Sken 1475268549Smav mtx_lock(&be_lun->queue_lock); 1476229997Sken continue; 1477229997Sken } 1478229997Sken 1479229997Sken /* 1480229997Sken * If we get here, there is no work left in the queues, so 1481229997Sken * just break out and let the task queue go to sleep. 1482229997Sken */ 1483229997Sken break; 1484229997Sken } 1485268549Smav mtx_unlock(&be_lun->queue_lock); 1486229997Sken} 1487229997Sken 1488229997Sken/* 1489229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1490229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1491229997Sken * thread. 1492229997Sken */ 1493229997Skenstatic int 1494229997Skenctl_be_block_submit(union ctl_io *io) 1495229997Sken{ 1496229997Sken struct ctl_be_block_lun *be_lun; 1497229997Sken struct ctl_be_lun *ctl_be_lun; 1498229997Sken 1499229997Sken DPRINTF("entered\n"); 1500229997Sken 1501229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1502229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1503229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1504229997Sken 1505229997Sken /* 1506229997Sken * Make sure we only get SCSI I/O. 1507229997Sken */ 1508229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1509229997Sken "%#x) encountered", io->io_hdr.io_type)); 1510229997Sken 1511268150Smav PRIV(io)->len = 0; 1512268150Smav 1513268549Smav mtx_lock(&be_lun->queue_lock); 1514229997Sken /* 1515229997Sken * XXX KDM make sure that links is okay to use at this point. 1516229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1517229997Sken * or deal with resource allocation here. 1518229997Sken */ 1519229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1520268549Smav mtx_unlock(&be_lun->queue_lock); 1521229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1522229997Sken 1523268148Smav return (CTL_RETVAL_COMPLETE); 1524229997Sken} 1525229997Sken 1526229997Skenstatic int 1527229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1528229997Sken int flag, struct thread *td) 1529229997Sken{ 1530229997Sken struct ctl_be_block_softc *softc; 1531229997Sken int error; 1532229997Sken 1533229997Sken softc = &backend_block_softc; 1534229997Sken 1535229997Sken error = 0; 1536229997Sken 1537229997Sken switch (cmd) { 1538229997Sken case CTL_LUN_REQ: { 1539229997Sken struct ctl_lun_req *lun_req; 1540229997Sken 1541229997Sken lun_req = (struct ctl_lun_req *)addr; 1542229997Sken 1543229997Sken switch (lun_req->reqtype) { 1544229997Sken case CTL_LUNREQ_CREATE: 1545229997Sken error = ctl_be_block_create(softc, lun_req); 1546229997Sken break; 1547229997Sken case CTL_LUNREQ_RM: 1548229997Sken error = ctl_be_block_rm(softc, lun_req); 1549229997Sken break; 1550232604Strasz case CTL_LUNREQ_MODIFY: 1551232604Strasz error = ctl_be_block_modify(softc, lun_req); 1552232604Strasz break; 1553229997Sken default: 1554229997Sken lun_req->status = CTL_LUN_ERROR; 1555229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1556229997Sken "%s: invalid LUN request type %d", __func__, 1557229997Sken lun_req->reqtype); 1558229997Sken break; 1559229997Sken } 1560229997Sken break; 1561229997Sken } 1562229997Sken default: 1563229997Sken error = ENOTTY; 1564229997Sken break; 1565229997Sken } 1566229997Sken 1567229997Sken return (error); 1568229997Sken} 1569229997Sken 1570229997Skenstatic int 1571229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1572229997Sken{ 1573229997Sken struct ctl_be_block_filedata *file_data; 1574229997Sken struct ctl_lun_create_params *params; 1575229997Sken struct vattr vattr; 1576229997Sken int error; 1577229997Sken 1578229997Sken error = 0; 1579229997Sken file_data = &be_lun->backend.file; 1580229997Sken params = &req->reqdata.create; 1581229997Sken 1582229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1583229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1584229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1585229997Sken 1586229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1587229997Sken if (error != 0) { 1588229997Sken snprintf(req->error_str, sizeof(req->error_str), 1589229997Sken "error calling VOP_GETATTR() for file %s", 1590229997Sken be_lun->dev_path); 1591229997Sken return (error); 1592229997Sken } 1593229997Sken 1594229997Sken /* 1595229997Sken * Verify that we have the ability to upgrade to exclusive 1596229997Sken * access on this file so we can trap errors at open instead 1597229997Sken * of reporting them during first access. 1598229997Sken */ 1599229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1600229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1601229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1602229997Sken error = EBADF; 1603229997Sken snprintf(req->error_str, sizeof(req->error_str), 1604229997Sken "error locking file %s", be_lun->dev_path); 1605229997Sken return (error); 1606229997Sken } 1607229997Sken } 1608229997Sken 1609229997Sken 1610229997Sken file_data->cred = crhold(curthread->td_ucred); 1611232604Strasz if (params->lun_size_bytes != 0) 1612232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1613232604Strasz else 1614232604Strasz be_lun->size_bytes = vattr.va_size; 1615229997Sken /* 1616229997Sken * We set the multi thread flag for file operations because all 1617229997Sken * filesystems (in theory) are capable of allowing multiple readers 1618229997Sken * of a file at once. So we want to get the maximum possible 1619229997Sken * concurrency. 1620229997Sken */ 1621229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1622229997Sken 1623229997Sken /* 1624229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1625229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1626229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1627229997Sken * may not work with other OSes as well. So just export a sector 1628229997Sken * size of 512 bytes, which should work with any OS or 1629229997Sken * application. Since our backing is a file, any block size will 1630229997Sken * work fine for the backing store. 1631229997Sken */ 1632229997Sken#if 0 1633229997Sken be_lun->blocksize= vattr.va_blocksize; 1634229997Sken#endif 1635229997Sken if (params->blocksize_bytes != 0) 1636229997Sken be_lun->blocksize = params->blocksize_bytes; 1637229997Sken else 1638229997Sken be_lun->blocksize = 512; 1639229997Sken 1640229997Sken /* 1641229997Sken * Sanity check. The media size has to be at least one 1642229997Sken * sector long. 1643229997Sken */ 1644229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1645229997Sken error = EINVAL; 1646229997Sken snprintf(req->error_str, sizeof(req->error_str), 1647229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1648229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1649229997Sken } 1650229997Sken return (error); 1651229997Sken} 1652229997Sken 1653229997Skenstatic int 1654229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1655229997Sken{ 1656229997Sken struct ctl_lun_create_params *params; 1657229997Sken struct vattr vattr; 1658229997Sken struct cdev *dev; 1659229997Sken struct cdevsw *devsw; 1660229997Sken int error; 1661264727Smav off_t ps, pss, po, pos; 1662229997Sken 1663229997Sken params = &req->reqdata.create; 1664229997Sken 1665229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1666229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1667229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1668229997Sken &be_lun->backend.dev.dev_ref); 1669229997Sken if (be_lun->backend.dev.csw == NULL) 1670229997Sken panic("Unable to retrieve device switch"); 1671269429Smav if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) 1672269429Smav be_lun->dispatch = ctl_be_block_dispatch_zvol; 1673269429Smav else 1674269429Smav be_lun->dispatch = ctl_be_block_dispatch_dev; 1675269429Smav be_lun->lun_flush = ctl_be_block_flush_dev; 1676269429Smav be_lun->unmap = ctl_be_block_unmap_dev; 1677229997Sken 1678229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1679229997Sken if (error) { 1680229997Sken snprintf(req->error_str, sizeof(req->error_str), 1681229997Sken "%s: error getting vnode attributes for device %s", 1682229997Sken __func__, be_lun->dev_path); 1683229997Sken return (error); 1684229997Sken } 1685229997Sken 1686229997Sken dev = be_lun->vn->v_rdev; 1687229997Sken devsw = dev->si_devsw; 1688229997Sken if (!devsw->d_ioctl) { 1689229997Sken snprintf(req->error_str, sizeof(req->error_str), 1690229997Sken "%s: no d_ioctl for device %s!", __func__, 1691229997Sken be_lun->dev_path); 1692229997Sken return (ENODEV); 1693229997Sken } 1694229997Sken 1695229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1696229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1697229997Sken curthread); 1698229997Sken if (error) { 1699229997Sken snprintf(req->error_str, sizeof(req->error_str), 1700229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1701229997Sken "on %s!", __func__, error, be_lun->dev_path); 1702229997Sken return (error); 1703229997Sken } 1704229997Sken 1705229997Sken /* 1706229997Sken * If the user has asked for a blocksize that is greater than the 1707229997Sken * backing device's blocksize, we can do it only if the blocksize 1708229997Sken * the user is asking for is an even multiple of the underlying 1709229997Sken * device's blocksize. 1710229997Sken */ 1711229997Sken if ((params->blocksize_bytes != 0) 1712229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1713229997Sken uint32_t bs_multiple, tmp_blocksize; 1714229997Sken 1715229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1716229997Sken 1717229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1718229997Sken 1719229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1720229997Sken be_lun->blocksize = params->blocksize_bytes; 1721229997Sken } else { 1722229997Sken snprintf(req->error_str, sizeof(req->error_str), 1723229997Sken "%s: requested blocksize %u is not an even " 1724229997Sken "multiple of backing device blocksize %u", 1725229997Sken __func__, params->blocksize_bytes, 1726229997Sken be_lun->blocksize); 1727229997Sken return (EINVAL); 1728229997Sken 1729229997Sken } 1730229997Sken } else if ((params->blocksize_bytes != 0) 1731229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1732229997Sken snprintf(req->error_str, sizeof(req->error_str), 1733229997Sken "%s: requested blocksize %u < backing device " 1734229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1735229997Sken be_lun->blocksize); 1736229997Sken return (EINVAL); 1737229997Sken } 1738229997Sken 1739229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1740229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1741229997Sken curthread); 1742229997Sken if (error) { 1743229997Sken snprintf(req->error_str, sizeof(req->error_str), 1744232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1745232604Strasz " ioctl on %s!", __func__, error, 1746232604Strasz be_lun->dev_path); 1747229997Sken return (error); 1748229997Sken } 1749229997Sken 1750232604Strasz if (params->lun_size_bytes != 0) { 1751232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1752232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1753232604Strasz "%s: requested LUN size %ju > backing device " 1754232604Strasz "size %ju", __func__, 1755232604Strasz (uintmax_t)params->lun_size_bytes, 1756232604Strasz (uintmax_t)be_lun->size_bytes); 1757232604Strasz return (EINVAL); 1758232604Strasz } 1759232604Strasz 1760232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1761232604Strasz } 1762232604Strasz 1763264727Smav error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1764264727Smav (caddr_t)&ps, FREAD, curthread); 1765264727Smav if (error) 1766264727Smav ps = po = 0; 1767264727Smav else { 1768264727Smav error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1769264727Smav (caddr_t)&po, FREAD, curthread); 1770264727Smav if (error) 1771264727Smav po = 0; 1772264727Smav } 1773264727Smav pss = ps / be_lun->blocksize; 1774264727Smav pos = po / be_lun->blocksize; 1775264727Smav if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1776264727Smav ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1777264727Smav be_lun->pblockexp = fls(pss) - 1; 1778264727Smav be_lun->pblockoff = (pss - pos) % pss; 1779264727Smav } 1780264727Smav 1781229997Sken return (0); 1782229997Sken} 1783229997Sken 1784229997Skenstatic int 1785229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1786229997Sken{ 1787229997Sken DROP_GIANT(); 1788229997Sken if (be_lun->vn) { 1789229997Sken int flags = FREAD | FWRITE; 1790229997Sken 1791229997Sken switch (be_lun->dev_type) { 1792229997Sken case CTL_BE_BLOCK_DEV: 1793229997Sken if (be_lun->backend.dev.csw) { 1794229997Sken dev_relthread(be_lun->backend.dev.cdev, 1795229997Sken be_lun->backend.dev.dev_ref); 1796229997Sken be_lun->backend.dev.csw = NULL; 1797229997Sken be_lun->backend.dev.cdev = NULL; 1798229997Sken } 1799229997Sken break; 1800229997Sken case CTL_BE_BLOCK_FILE: 1801229997Sken break; 1802229997Sken case CTL_BE_BLOCK_NONE: 1803259304Strasz break; 1804229997Sken default: 1805229997Sken panic("Unexpected backend type."); 1806229997Sken break; 1807229997Sken } 1808229997Sken 1809229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1810229997Sken be_lun->vn = NULL; 1811229997Sken 1812229997Sken switch (be_lun->dev_type) { 1813229997Sken case CTL_BE_BLOCK_DEV: 1814229997Sken break; 1815229997Sken case CTL_BE_BLOCK_FILE: 1816229997Sken if (be_lun->backend.file.cred != NULL) { 1817229997Sken crfree(be_lun->backend.file.cred); 1818229997Sken be_lun->backend.file.cred = NULL; 1819229997Sken } 1820229997Sken break; 1821229997Sken case CTL_BE_BLOCK_NONE: 1822259304Strasz break; 1823229997Sken default: 1824229997Sken panic("Unexpected backend type."); 1825229997Sken break; 1826229997Sken } 1827229997Sken } 1828229997Sken PICKUP_GIANT(); 1829229997Sken 1830229997Sken return (0); 1831229997Sken} 1832229997Sken 1833229997Skenstatic int 1834229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1835229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1836229997Sken{ 1837229997Sken struct nameidata nd; 1838229997Sken int flags; 1839229997Sken int error; 1840229997Sken 1841229997Sken /* 1842229997Sken * XXX KDM allow a read-only option? 1843229997Sken */ 1844229997Sken flags = FREAD | FWRITE; 1845229997Sken error = 0; 1846229997Sken 1847229997Sken if (rootvnode == NULL) { 1848229997Sken snprintf(req->error_str, sizeof(req->error_str), 1849229997Sken "%s: Root filesystem is not mounted", __func__); 1850229997Sken return (1); 1851229997Sken } 1852229997Sken 1853229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1854229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1855229997Sken VREF(rootvnode); 1856229997Sken } 1857229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1858229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1859229997Sken VREF(rootvnode); 1860229997Sken } 1861229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1862229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1863229997Sken VREF(rootvnode); 1864229997Sken } 1865229997Sken 1866229997Sken again: 1867229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1868229997Sken error = vn_open(&nd, &flags, 0, NULL); 1869229997Sken if (error) { 1870229997Sken /* 1871229997Sken * This is the only reasonable guess we can make as far as 1872229997Sken * path if the user doesn't give us a fully qualified path. 1873229997Sken * If they want to specify a file, they need to specify the 1874229997Sken * full path. 1875229997Sken */ 1876229997Sken if (be_lun->dev_path[0] != '/') { 1877229997Sken char *dev_path = "/dev/"; 1878229997Sken char *dev_name; 1879229997Sken 1880229997Sken /* Try adding device path at beginning of name */ 1881229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1882229997Sken + strlen(dev_path) + 1, 1883229997Sken M_CTLBLK, M_WAITOK); 1884229997Sken if (dev_name) { 1885229997Sken sprintf(dev_name, "%s%s", dev_path, 1886229997Sken be_lun->dev_path); 1887229997Sken free(be_lun->dev_path, M_CTLBLK); 1888229997Sken be_lun->dev_path = dev_name; 1889229997Sken goto again; 1890229997Sken } 1891229997Sken } 1892229997Sken snprintf(req->error_str, sizeof(req->error_str), 1893229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1894229997Sken return (error); 1895229997Sken } 1896229997Sken 1897229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1898229997Sken 1899229997Sken be_lun->vn = nd.ni_vp; 1900229997Sken 1901229997Sken /* We only support disks and files. */ 1902229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1903229997Sken error = ctl_be_block_open_dev(be_lun, req); 1904229997Sken } else if (be_lun->vn->v_type == VREG) { 1905229997Sken error = ctl_be_block_open_file(be_lun, req); 1906229997Sken } else { 1907229997Sken error = EINVAL; 1908229997Sken snprintf(req->error_str, sizeof(req->error_str), 1909259304Strasz "%s is not a disk or plain file", be_lun->dev_path); 1910229997Sken } 1911229997Sken VOP_UNLOCK(be_lun->vn, 0); 1912229997Sken 1913229997Sken if (error != 0) { 1914229997Sken ctl_be_block_close(be_lun); 1915229997Sken return (error); 1916229997Sken } 1917229997Sken 1918229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1919229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1920229997Sken 1921229997Sken return (0); 1922229997Sken} 1923229997Sken 1924229997Skenstatic int 1925229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1926229997Sken{ 1927229997Sken struct ctl_be_block_lun *be_lun; 1928229997Sken struct ctl_lun_create_params *params; 1929268143Smav char num_thread_str[16]; 1930229997Sken char tmpstr[32]; 1931268143Smav char *value; 1932265634Smav int retval, num_threads, unmap; 1933268143Smav int tmp_num_threads; 1934229997Sken 1935229997Sken params = &req->reqdata.create; 1936229997Sken retval = 0; 1937229997Sken 1938229997Sken num_threads = cbb_num_threads; 1939229997Sken 1940229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1941229997Sken 1942229997Sken be_lun->softc = softc; 1943229997Sken STAILQ_INIT(&be_lun->input_queue); 1944229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1945229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1946229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1947268549Smav mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1948268549Smav mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1949268678Smav ctl_init_opts(&be_lun->ctl_be_lun.options, 1950268678Smav req->num_be_args, req->kern_be_args); 1951229997Sken 1952265642Smav be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1953260476Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1954229997Sken 1955229997Sken if (be_lun->lun_zone == NULL) { 1956229997Sken snprintf(req->error_str, sizeof(req->error_str), 1957229997Sken "%s: error allocating UMA zone", __func__); 1958229997Sken goto bailout_error; 1959229997Sken } 1960229997Sken 1961229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1962229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1963229997Sken else 1964229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1965229997Sken 1966229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1967268678Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1968268146Smav if (value == NULL) { 1969229997Sken snprintf(req->error_str, sizeof(req->error_str), 1970229997Sken "%s: no file argument specified", __func__); 1971229997Sken goto bailout_error; 1972229997Sken } 1973268146Smav be_lun->dev_path = strdup(value, M_CTLBLK); 1974229997Sken 1975229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1976229997Sken if (retval != 0) { 1977229997Sken retval = 0; 1978229997Sken goto bailout_error; 1979229997Sken } 1980229997Sken 1981229997Sken /* 1982229997Sken * Tell the user the size of the file/device. 1983229997Sken */ 1984229997Sken params->lun_size_bytes = be_lun->size_bytes; 1985229997Sken 1986229997Sken /* 1987229997Sken * The maximum LBA is the size - 1. 1988229997Sken */ 1989229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1990229997Sken } else { 1991229997Sken /* 1992229997Sken * For processor devices, we don't have any size. 1993229997Sken */ 1994229997Sken be_lun->blocksize = 0; 1995264727Smav be_lun->pblockexp = 0; 1996264727Smav be_lun->pblockoff = 0; 1997229997Sken be_lun->size_blocks = 0; 1998229997Sken be_lun->size_bytes = 0; 1999229997Sken be_lun->ctl_be_lun.maxlba = 0; 2000229997Sken params->lun_size_bytes = 0; 2001229997Sken 2002229997Sken /* 2003229997Sken * Default to just 1 thread for processor devices. 2004229997Sken */ 2005229997Sken num_threads = 1; 2006229997Sken } 2007229997Sken 2008229997Sken /* 2009229997Sken * XXX This searching loop might be refactored to be combined with 2010229997Sken * the loop above, 2011229997Sken */ 2012268678Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2013268143Smav if (value != NULL) { 2014268143Smav tmp_num_threads = strtol(value, NULL, 0); 2015229997Sken 2016268143Smav /* 2017268143Smav * We don't let the user specify less than one 2018268143Smav * thread, but hope he's clueful enough not to 2019268143Smav * specify 1000 threads. 2020268143Smav */ 2021268143Smav if (tmp_num_threads < 1) { 2022268143Smav snprintf(req->error_str, sizeof(req->error_str), 2023268143Smav "%s: invalid number of threads %s", 2024268143Smav __func__, num_thread_str); 2025268143Smav goto bailout_error; 2026229997Sken } 2027268143Smav num_threads = tmp_num_threads; 2028229997Sken } 2029268143Smav unmap = 0; 2030268678Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2031268143Smav if (value != NULL && strcmp(value, "on") == 0) 2032268143Smav unmap = 1; 2033229997Sken 2034229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2035229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2036265634Smav if (unmap) 2037265634Smav be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2038229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 2039229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2040264727Smav be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2041264727Smav be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2042229997Sken /* Tell the user the blocksize we ended up using */ 2043229997Sken params->blocksize_bytes = be_lun->blocksize; 2044229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2045229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2046229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2047229997Sken } else 2048229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 2049229997Sken 2050229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2051229997Sken be_lun->ctl_be_lun.lun_config_status = 2052229997Sken ctl_be_block_lun_config_status; 2053229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2054229997Sken 2055229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2056229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2057229997Sken softc->num_luns); 2058229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2059229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2060229997Sken sizeof(tmpstr))); 2061229997Sken 2062229997Sken /* Tell the user what we used for a serial number */ 2063229997Sken strncpy((char *)params->serial_num, tmpstr, 2064229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 2065229997Sken } else { 2066229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 2067229997Sken params->serial_num, 2068229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2069229997Sken sizeof(params->serial_num))); 2070229997Sken } 2071229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2072229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2073229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2074229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2075229997Sken sizeof(tmpstr))); 2076229997Sken 2077229997Sken /* Tell the user what we used for a device ID */ 2078229997Sken strncpy((char *)params->device_id, tmpstr, 2079229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 2080229997Sken } else { 2081229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 2082229997Sken params->device_id, 2083229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2084229997Sken sizeof(params->device_id))); 2085229997Sken } 2086229997Sken 2087229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2088229997Sken 2089229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2090229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2091229997Sken 2092229997Sken if (be_lun->io_taskqueue == NULL) { 2093229997Sken snprintf(req->error_str, sizeof(req->error_str), 2094229997Sken "%s: Unable to create taskqueue", __func__); 2095229997Sken goto bailout_error; 2096229997Sken } 2097229997Sken 2098229997Sken /* 2099229997Sken * Note that we start the same number of threads by default for 2100229997Sken * both the file case and the block device case. For the file 2101229997Sken * case, we need multiple threads to allow concurrency, because the 2102229997Sken * vnode interface is designed to be a blocking interface. For the 2103229997Sken * block device case, ZFS zvols at least will block the caller's 2104229997Sken * context in many instances, and so we need multiple threads to 2105229997Sken * overcome that problem. Other block devices don't need as many 2106229997Sken * threads, but they shouldn't cause too many problems. 2107229997Sken * 2108229997Sken * If the user wants to just have a single thread for a block 2109229997Sken * device, he can specify that when the LUN is created, or change 2110229997Sken * the tunable/sysctl to alter the default number of threads. 2111229997Sken */ 2112229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2113229997Sken /*num threads*/num_threads, 2114229997Sken /*priority*/PWAIT, 2115229997Sken /*thread name*/ 2116229997Sken "%s taskq", be_lun->lunname); 2117229997Sken 2118229997Sken if (retval != 0) 2119229997Sken goto bailout_error; 2120229997Sken 2121229997Sken be_lun->num_threads = num_threads; 2122229997Sken 2123229997Sken mtx_lock(&softc->lock); 2124229997Sken softc->num_luns++; 2125229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2126229997Sken 2127229997Sken mtx_unlock(&softc->lock); 2128229997Sken 2129229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 2130229997Sken if (retval != 0) { 2131229997Sken mtx_lock(&softc->lock); 2132229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2133229997Sken links); 2134229997Sken softc->num_luns--; 2135229997Sken mtx_unlock(&softc->lock); 2136229997Sken snprintf(req->error_str, sizeof(req->error_str), 2137229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 2138229997Sken "details", __func__, retval); 2139229997Sken retval = 0; 2140229997Sken goto bailout_error; 2141229997Sken } 2142229997Sken 2143229997Sken mtx_lock(&softc->lock); 2144229997Sken 2145229997Sken /* 2146229997Sken * Tell the config_status routine that we're waiting so it won't 2147229997Sken * clean up the LUN in the event of an error. 2148229997Sken */ 2149229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2150229997Sken 2151229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2152229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2153229997Sken if (retval == EINTR) 2154229997Sken break; 2155229997Sken } 2156229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2157229997Sken 2158229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2159229997Sken snprintf(req->error_str, sizeof(req->error_str), 2160229997Sken "%s: LUN configuration error, see dmesg for details", 2161229997Sken __func__); 2162229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2163229997Sken links); 2164229997Sken softc->num_luns--; 2165229997Sken mtx_unlock(&softc->lock); 2166229997Sken goto bailout_error; 2167229997Sken } else { 2168229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2169229997Sken } 2170229997Sken 2171229997Sken mtx_unlock(&softc->lock); 2172229997Sken 2173229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2174229997Sken be_lun->blocksize, 2175229997Sken DEVSTAT_ALL_SUPPORTED, 2176229997Sken be_lun->ctl_be_lun.lun_type 2177229997Sken | DEVSTAT_TYPE_IF_OTHER, 2178229997Sken DEVSTAT_PRIORITY_OTHER); 2179229997Sken 2180229997Sken 2181229997Sken req->status = CTL_LUN_OK; 2182229997Sken 2183229997Sken return (retval); 2184229997Sken 2185229997Skenbailout_error: 2186229997Sken req->status = CTL_LUN_ERROR; 2187229997Sken 2188267754Smav if (be_lun->io_taskqueue != NULL) 2189267754Smav taskqueue_free(be_lun->io_taskqueue); 2190229997Sken ctl_be_block_close(be_lun); 2191267754Smav if (be_lun->dev_path != NULL) 2192267754Smav free(be_lun->dev_path, M_CTLBLK); 2193267754Smav if (be_lun->lun_zone != NULL) 2194267754Smav uma_zdestroy(be_lun->lun_zone); 2195268678Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2196268549Smav mtx_destroy(&be_lun->queue_lock); 2197268549Smav mtx_destroy(&be_lun->io_lock); 2198229997Sken free(be_lun, M_CTLBLK); 2199229997Sken 2200229997Sken return (retval); 2201229997Sken} 2202229997Sken 2203229997Skenstatic int 2204229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2205229997Sken{ 2206229997Sken struct ctl_lun_rm_params *params; 2207229997Sken struct ctl_be_block_lun *be_lun; 2208229997Sken int retval; 2209229997Sken 2210229997Sken params = &req->reqdata.rm; 2211229997Sken 2212229997Sken mtx_lock(&softc->lock); 2213229997Sken 2214229997Sken be_lun = NULL; 2215229997Sken 2216229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2217229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2218229997Sken break; 2219229997Sken } 2220229997Sken mtx_unlock(&softc->lock); 2221229997Sken 2222229997Sken if (be_lun == NULL) { 2223229997Sken snprintf(req->error_str, sizeof(req->error_str), 2224229997Sken "%s: LUN %u is not managed by the block backend", 2225229997Sken __func__, params->lun_id); 2226229997Sken goto bailout_error; 2227229997Sken } 2228229997Sken 2229229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2230229997Sken 2231229997Sken if (retval != 0) { 2232229997Sken snprintf(req->error_str, sizeof(req->error_str), 2233229997Sken "%s: error %d returned from ctl_disable_lun() for " 2234229997Sken "LUN %d", __func__, retval, params->lun_id); 2235229997Sken goto bailout_error; 2236229997Sken 2237229997Sken } 2238229997Sken 2239229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2240229997Sken if (retval != 0) { 2241229997Sken snprintf(req->error_str, sizeof(req->error_str), 2242229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 2243229997Sken "LUN %d", __func__, retval, params->lun_id); 2244229997Sken goto bailout_error; 2245229997Sken } 2246229997Sken 2247229997Sken mtx_lock(&softc->lock); 2248229997Sken 2249229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2250229997Sken 2251229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2252229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2253229997Sken if (retval == EINTR) 2254229997Sken break; 2255229997Sken } 2256229997Sken 2257229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2258229997Sken 2259229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2260229997Sken snprintf(req->error_str, sizeof(req->error_str), 2261229997Sken "%s: interrupted waiting for LUN to be freed", 2262229997Sken __func__); 2263229997Sken mtx_unlock(&softc->lock); 2264229997Sken goto bailout_error; 2265229997Sken } 2266229997Sken 2267229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2268229997Sken 2269229997Sken softc->num_luns--; 2270229997Sken mtx_unlock(&softc->lock); 2271229997Sken 2272229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2273229997Sken 2274229997Sken taskqueue_free(be_lun->io_taskqueue); 2275229997Sken 2276229997Sken ctl_be_block_close(be_lun); 2277229997Sken 2278229997Sken if (be_lun->disk_stats != NULL) 2279229997Sken devstat_remove_entry(be_lun->disk_stats); 2280229997Sken 2281229997Sken uma_zdestroy(be_lun->lun_zone); 2282229997Sken 2283268678Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2284229997Sken free(be_lun->dev_path, M_CTLBLK); 2285268549Smav mtx_destroy(&be_lun->queue_lock); 2286268549Smav mtx_destroy(&be_lun->io_lock); 2287229997Sken free(be_lun, M_CTLBLK); 2288229997Sken 2289229997Sken req->status = CTL_LUN_OK; 2290229997Sken 2291229997Sken return (0); 2292229997Sken 2293229997Skenbailout_error: 2294229997Sken 2295229997Sken req->status = CTL_LUN_ERROR; 2296229997Sken 2297229997Sken return (0); 2298229997Sken} 2299229997Sken 2300232604Straszstatic int 2301232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2302232604Strasz struct ctl_lun_req *req) 2303232604Strasz{ 2304232604Strasz struct vattr vattr; 2305232604Strasz int error; 2306232604Strasz struct ctl_lun_modify_params *params; 2307232604Strasz 2308232604Strasz params = &req->reqdata.modify; 2309232604Strasz 2310232604Strasz if (params->lun_size_bytes != 0) { 2311232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2312232604Strasz } else { 2313271928Smav vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2314232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2315271928Smav VOP_UNLOCK(be_lun->vn, 0); 2316232604Strasz if (error != 0) { 2317232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2318232604Strasz "error calling VOP_GETATTR() for file %s", 2319232604Strasz be_lun->dev_path); 2320232604Strasz return (error); 2321232604Strasz } 2322232604Strasz 2323232604Strasz be_lun->size_bytes = vattr.va_size; 2324232604Strasz } 2325232604Strasz 2326232604Strasz return (0); 2327232604Strasz} 2328232604Strasz 2329232604Straszstatic int 2330232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2331232604Strasz struct ctl_lun_req *req) 2332232604Strasz{ 2333271928Smav struct ctl_be_block_devdata *dev_data; 2334232604Strasz int error; 2335232604Strasz struct ctl_lun_modify_params *params; 2336232604Strasz uint64_t size_bytes; 2337232604Strasz 2338232604Strasz params = &req->reqdata.modify; 2339232604Strasz 2340271928Smav dev_data = &be_lun->backend.dev; 2341271928Smav if (!dev_data->csw->d_ioctl) { 2342232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2343232604Strasz "%s: no d_ioctl for device %s!", __func__, 2344232604Strasz be_lun->dev_path); 2345232604Strasz return (ENODEV); 2346232604Strasz } 2347232604Strasz 2348271928Smav error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2349232604Strasz (caddr_t)&size_bytes, FREAD, 2350232604Strasz curthread); 2351232604Strasz if (error) { 2352232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2353232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2354232604Strasz "on %s!", __func__, error, be_lun->dev_path); 2355232604Strasz return (error); 2356232604Strasz } 2357232604Strasz 2358232604Strasz if (params->lun_size_bytes != 0) { 2359232604Strasz if (params->lun_size_bytes > size_bytes) { 2360232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2361232604Strasz "%s: requested LUN size %ju > backing device " 2362232604Strasz "size %ju", __func__, 2363232604Strasz (uintmax_t)params->lun_size_bytes, 2364232604Strasz (uintmax_t)size_bytes); 2365232604Strasz return (EINVAL); 2366232604Strasz } 2367232604Strasz 2368232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2369232604Strasz } else { 2370232604Strasz be_lun->size_bytes = size_bytes; 2371232604Strasz } 2372232604Strasz 2373232604Strasz return (0); 2374232604Strasz} 2375232604Strasz 2376232604Straszstatic int 2377232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2378232604Strasz{ 2379232604Strasz struct ctl_lun_modify_params *params; 2380232604Strasz struct ctl_be_block_lun *be_lun; 2381271928Smav uint64_t oldsize; 2382241896Skib int error; 2383232604Strasz 2384232604Strasz params = &req->reqdata.modify; 2385232604Strasz 2386232604Strasz mtx_lock(&softc->lock); 2387232604Strasz 2388232604Strasz be_lun = NULL; 2389232604Strasz 2390232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2391232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2392232604Strasz break; 2393232604Strasz } 2394232604Strasz mtx_unlock(&softc->lock); 2395232604Strasz 2396232604Strasz if (be_lun == NULL) { 2397232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2398232604Strasz "%s: LUN %u is not managed by the block backend", 2399232604Strasz __func__, params->lun_id); 2400232604Strasz goto bailout_error; 2401232604Strasz } 2402232604Strasz 2403232604Strasz if (params->lun_size_bytes != 0) { 2404232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2405232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2406232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2407232604Strasz params->lun_size_bytes, be_lun->blocksize); 2408232604Strasz goto bailout_error; 2409232604Strasz } 2410232604Strasz } 2411232604Strasz 2412271928Smav oldsize = be_lun->size_bytes; 2413232604Strasz if (be_lun->vn->v_type == VREG) 2414232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2415232604Strasz else 2416232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2417232604Strasz if (error != 0) 2418232604Strasz goto bailout_error; 2419232604Strasz 2420271928Smav if (be_lun->size_bytes != oldsize) { 2421271928Smav be_lun->size_blocks = be_lun->size_bytes >> 2422271928Smav be_lun->blocksize_shift; 2423232604Strasz 2424271928Smav /* 2425271928Smav * The maximum LBA is the size - 1. 2426271928Smav * 2427271928Smav * XXX: Note that this field is being updated without locking, 2428271928Smav * which might cause problems on 32-bit architectures. 2429271928Smav */ 2430271928Smav be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2431271928Smav ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2432271928Smav } 2433232604Strasz 2434232604Strasz /* Tell the user the exact size we ended up using */ 2435232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2436232604Strasz 2437232604Strasz req->status = CTL_LUN_OK; 2438232604Strasz 2439232604Strasz return (0); 2440232604Strasz 2441232604Straszbailout_error: 2442232604Strasz req->status = CTL_LUN_ERROR; 2443232604Strasz 2444232604Strasz return (0); 2445232604Strasz} 2446232604Strasz 2447229997Skenstatic void 2448229997Skenctl_be_block_lun_shutdown(void *be_lun) 2449229997Sken{ 2450229997Sken struct ctl_be_block_lun *lun; 2451229997Sken struct ctl_be_block_softc *softc; 2452229997Sken 2453229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2454229997Sken 2455229997Sken softc = lun->softc; 2456229997Sken 2457229997Sken mtx_lock(&softc->lock); 2458229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2459229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2460229997Sken wakeup(lun); 2461229997Sken mtx_unlock(&softc->lock); 2462229997Sken 2463229997Sken} 2464229997Sken 2465229997Skenstatic void 2466229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2467229997Sken{ 2468229997Sken struct ctl_be_block_lun *lun; 2469229997Sken struct ctl_be_block_softc *softc; 2470229997Sken 2471229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2472229997Sken softc = lun->softc; 2473229997Sken 2474229997Sken if (status == CTL_LUN_CONFIG_OK) { 2475229997Sken mtx_lock(&softc->lock); 2476229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2477229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2478229997Sken wakeup(lun); 2479229997Sken mtx_unlock(&softc->lock); 2480229997Sken 2481229997Sken /* 2482229997Sken * We successfully added the LUN, attempt to enable it. 2483229997Sken */ 2484229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2485229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2486229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2487229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2488229997Sken __func__); 2489229997Sken } 2490229997Sken } 2491229997Sken 2492229997Sken return; 2493229997Sken } 2494229997Sken 2495229997Sken 2496229997Sken mtx_lock(&softc->lock); 2497229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2498229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2499229997Sken wakeup(lun); 2500229997Sken mtx_unlock(&softc->lock); 2501229997Sken} 2502229997Sken 2503229997Sken 2504229997Skenstatic int 2505229997Skenctl_be_block_config_write(union ctl_io *io) 2506229997Sken{ 2507229997Sken struct ctl_be_block_lun *be_lun; 2508229997Sken struct ctl_be_lun *ctl_be_lun; 2509229997Sken int retval; 2510229997Sken 2511229997Sken retval = 0; 2512229997Sken 2513229997Sken DPRINTF("entered\n"); 2514229997Sken 2515229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2516229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2517229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2518229997Sken 2519229997Sken switch (io->scsiio.cdb[0]) { 2520229997Sken case SYNCHRONIZE_CACHE: 2521229997Sken case SYNCHRONIZE_CACHE_16: 2522265634Smav case WRITE_SAME_10: 2523265634Smav case WRITE_SAME_16: 2524265634Smav case UNMAP: 2525229997Sken /* 2526229997Sken * The upper level CTL code will filter out any CDBs with 2527229997Sken * the immediate bit set and return the proper error. 2528229997Sken * 2529229997Sken * We don't really need to worry about what LBA range the 2530229997Sken * user asked to be synced out. When they issue a sync 2531229997Sken * cache command, we'll sync out the whole thing. 2532229997Sken */ 2533268549Smav mtx_lock(&be_lun->queue_lock); 2534229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2535229997Sken links); 2536268549Smav mtx_unlock(&be_lun->queue_lock); 2537229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2538229997Sken break; 2539229997Sken case START_STOP_UNIT: { 2540229997Sken struct scsi_start_stop_unit *cdb; 2541229997Sken 2542229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2543229997Sken 2544229997Sken if (cdb->how & SSS_START) 2545229997Sken retval = ctl_start_lun(ctl_be_lun); 2546229997Sken else { 2547229997Sken retval = ctl_stop_lun(ctl_be_lun); 2548229997Sken /* 2549229997Sken * XXX KDM Copan-specific offline behavior. 2550229997Sken * Figure out a reasonable way to port this? 2551229997Sken */ 2552229997Sken#ifdef NEEDTOPORT 2553229997Sken if ((retval == 0) 2554229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2555229997Sken retval = ctl_lun_offline(ctl_be_lun); 2556229997Sken#endif 2557229997Sken } 2558229997Sken 2559229997Sken /* 2560229997Sken * In general, the above routines should not fail. They 2561229997Sken * just set state for the LUN. So we've got something 2562229997Sken * pretty wrong here if we can't start or stop the LUN. 2563229997Sken */ 2564229997Sken if (retval != 0) { 2565229997Sken ctl_set_internal_failure(&io->scsiio, 2566229997Sken /*sks_valid*/ 1, 2567229997Sken /*retry_count*/ 0xf051); 2568229997Sken retval = CTL_RETVAL_COMPLETE; 2569229997Sken } else { 2570229997Sken ctl_set_success(&io->scsiio); 2571229997Sken } 2572229997Sken ctl_config_write_done(io); 2573229997Sken break; 2574229997Sken } 2575229997Sken default: 2576229997Sken ctl_set_invalid_opcode(&io->scsiio); 2577229997Sken ctl_config_write_done(io); 2578229997Sken retval = CTL_RETVAL_COMPLETE; 2579229997Sken break; 2580229997Sken } 2581229997Sken 2582229997Sken return (retval); 2583229997Sken 2584229997Sken} 2585229997Sken 2586229997Skenstatic int 2587229997Skenctl_be_block_config_read(union ctl_io *io) 2588229997Sken{ 2589229997Sken return (0); 2590229997Sken} 2591229997Sken 2592229997Skenstatic int 2593229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2594229997Sken{ 2595229997Sken struct ctl_be_block_lun *lun; 2596229997Sken int retval; 2597229997Sken 2598229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2599229997Sken retval = 0; 2600229997Sken 2601268555Smav retval = sbuf_printf(sb, "\t<num_threads>"); 2602229997Sken 2603229997Sken if (retval != 0) 2604229997Sken goto bailout; 2605229997Sken 2606229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2607229997Sken 2608229997Sken if (retval != 0) 2609229997Sken goto bailout; 2610229997Sken 2611268555Smav retval = sbuf_printf(sb, "</num_threads>\n"); 2612229997Sken 2613229997Skenbailout: 2614229997Sken 2615229997Sken return (retval); 2616229997Sken} 2617229997Sken 2618229997Skenint 2619229997Skenctl_be_block_init(void) 2620229997Sken{ 2621229997Sken struct ctl_be_block_softc *softc; 2622229997Sken int retval; 2623229997Sken 2624229997Sken softc = &backend_block_softc; 2625229997Sken retval = 0; 2626229997Sken 2627268549Smav mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2628265494Strasz beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2629265494Strasz NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2630229997Sken STAILQ_INIT(&softc->disk_list); 2631229997Sken STAILQ_INIT(&softc->lun_list); 2632229997Sken 2633229997Sken return (retval); 2634229997Sken} 2635