ctl_backend_block.c revision 271794
1229997Sken/*- 2229997Sken * Copyright (c) 2003 Silicon Graphics International Corp. 3229997Sken * Copyright (c) 2009-2011 Spectra Logic Corporation 4232604Strasz * Copyright (c) 2012 The FreeBSD Foundation 5229997Sken * All rights reserved. 6229997Sken * 7232604Strasz * Portions of this software were developed by Edward Tomasz Napierala 8232604Strasz * under sponsorship from the FreeBSD Foundation. 9232604Strasz * 10229997Sken * Redistribution and use in source and binary forms, with or without 11229997Sken * modification, are permitted provided that the following conditions 12229997Sken * are met: 13229997Sken * 1. Redistributions of source code must retain the above copyright 14229997Sken * notice, this list of conditions, and the following disclaimer, 15229997Sken * without modification. 16229997Sken * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17229997Sken * substantially similar to the "NO WARRANTY" disclaimer below 18229997Sken * ("Disclaimer") and any redistribution must be conditioned upon 19229997Sken * including a substantially similar Disclaimer requirement for further 20229997Sken * binary redistribution. 21229997Sken * 22229997Sken * NO WARRANTY 23229997Sken * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24229997Sken * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25229997Sken * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26229997Sken * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27229997Sken * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28229997Sken * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29229997Sken * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30229997Sken * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31229997Sken * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32229997Sken * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33229997Sken * POSSIBILITY OF SUCH DAMAGES. 34229997Sken * 35229997Sken * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36229997Sken */ 37229997Sken/* 38229997Sken * CAM Target Layer driver backend for block devices. 39229997Sken * 40229997Sken * Author: Ken Merry <ken@FreeBSD.org> 41229997Sken */ 42229997Sken#include <sys/cdefs.h> 43229997Sken__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl_backend_block.c 271794 2014-09-18 17:25:20Z mav $"); 44229997Sken 45229997Sken#include <sys/param.h> 46229997Sken#include <sys/systm.h> 47229997Sken#include <sys/kernel.h> 48229997Sken#include <sys/types.h> 49229997Sken#include <sys/kthread.h> 50229997Sken#include <sys/bio.h> 51229997Sken#include <sys/fcntl.h> 52264274Smav#include <sys/limits.h> 53229997Sken#include <sys/lock.h> 54229997Sken#include <sys/mutex.h> 55229997Sken#include <sys/condvar.h> 56229997Sken#include <sys/malloc.h> 57229997Sken#include <sys/conf.h> 58229997Sken#include <sys/ioccom.h> 59229997Sken#include <sys/queue.h> 60229997Sken#include <sys/sbuf.h> 61229997Sken#include <sys/endian.h> 62229997Sken#include <sys/uio.h> 63229997Sken#include <sys/buf.h> 64229997Sken#include <sys/taskqueue.h> 65229997Sken#include <sys/vnode.h> 66229997Sken#include <sys/namei.h> 67229997Sken#include <sys/mount.h> 68229997Sken#include <sys/disk.h> 69229997Sken#include <sys/fcntl.h> 70229997Sken#include <sys/filedesc.h> 71229997Sken#include <sys/proc.h> 72229997Sken#include <sys/pcpu.h> 73229997Sken#include <sys/module.h> 74229997Sken#include <sys/sdt.h> 75229997Sken#include <sys/devicestat.h> 76229997Sken#include <sys/sysctl.h> 77229997Sken 78229997Sken#include <geom/geom.h> 79229997Sken 80229997Sken#include <cam/cam.h> 81229997Sken#include <cam/scsi/scsi_all.h> 82229997Sken#include <cam/scsi/scsi_da.h> 83229997Sken#include <cam/ctl/ctl_io.h> 84229997Sken#include <cam/ctl/ctl.h> 85229997Sken#include <cam/ctl/ctl_backend.h> 86229997Sken#include <cam/ctl/ctl_frontend_internal.h> 87229997Sken#include <cam/ctl/ctl_ioctl.h> 88229997Sken#include <cam/ctl/ctl_scsi_all.h> 89229997Sken#include <cam/ctl/ctl_error.h> 90229997Sken 91229997Sken/* 92264886Smav * The idea here is that we'll allocate enough S/G space to hold a 1MB 93264886Smav * I/O. If we get an I/O larger than that, we'll split it. 94229997Sken */ 95267537Smav#define CTLBLK_HALF_IO_SIZE (512 * 1024) 96267537Smav#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 97264886Smav#define CTLBLK_MAX_SEG MAXPHYS 98267537Smav#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 99267537Smav#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 100229997Sken 101229997Sken#ifdef CTLBLK_DEBUG 102229997Sken#define DPRINTF(fmt, args...) \ 103229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 104229997Sken#else 105229997Sken#define DPRINTF(fmt, args...) do {} while(0) 106229997Sken#endif 107229997Sken 108267519Smav#define PRIV(io) \ 109267519Smav ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 110267537Smav#define ARGS(io) \ 111267537Smav ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 112267519Smav 113229997SkenSDT_PROVIDER_DEFINE(cbb); 114229997Sken 115229997Skentypedef enum { 116229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 117229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 118229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 119229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 120229997Sken} ctl_be_block_lun_flags; 121229997Sken 122229997Skentypedef enum { 123229997Sken CTL_BE_BLOCK_NONE, 124229997Sken CTL_BE_BLOCK_DEV, 125229997Sken CTL_BE_BLOCK_FILE 126229997Sken} ctl_be_block_type; 127229997Sken 128229997Skenstruct ctl_be_block_devdata { 129229997Sken struct cdev *cdev; 130229997Sken struct cdevsw *csw; 131229997Sken int dev_ref; 132229997Sken}; 133229997Sken 134229997Skenstruct ctl_be_block_filedata { 135229997Sken struct ucred *cred; 136229997Sken}; 137229997Sken 138229997Skenunion ctl_be_block_bedata { 139229997Sken struct ctl_be_block_devdata dev; 140229997Sken struct ctl_be_block_filedata file; 141229997Sken}; 142229997Sken 143229997Skenstruct ctl_be_block_io; 144229997Skenstruct ctl_be_block_lun; 145229997Sken 146229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 147229997Sken struct ctl_be_block_io *beio); 148229997Sken 149229997Sken/* 150229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 151229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 152229997Sken */ 153229997Skenstruct ctl_be_block_lun { 154229997Sken struct ctl_block_disk *disk; 155229997Sken char lunname[32]; 156229997Sken char *dev_path; 157229997Sken ctl_be_block_type dev_type; 158229997Sken struct vnode *vn; 159229997Sken union ctl_be_block_bedata backend; 160229997Sken cbb_dispatch_t dispatch; 161229997Sken cbb_dispatch_t lun_flush; 162264274Smav cbb_dispatch_t unmap; 163229997Sken uma_zone_t lun_zone; 164229997Sken uint64_t size_blocks; 165229997Sken uint64_t size_bytes; 166229997Sken uint32_t blocksize; 167229997Sken int blocksize_shift; 168264191Smav uint16_t pblockexp; 169264191Smav uint16_t pblockoff; 170229997Sken struct ctl_be_block_softc *softc; 171229997Sken struct devstat *disk_stats; 172229997Sken ctl_be_block_lun_flags flags; 173229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 174229997Sken struct ctl_be_lun ctl_be_lun; 175229997Sken struct taskqueue *io_taskqueue; 176229997Sken struct task io_task; 177229997Sken int num_threads; 178229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 179229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 180229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 181267877Smav struct mtx_padalign io_lock; 182267877Smav struct mtx_padalign queue_lock; 183229997Sken}; 184229997Sken 185229997Sken/* 186229997Sken * Overall softc structure for the block backend module. 187229997Sken */ 188229997Skenstruct ctl_be_block_softc { 189229997Sken struct mtx lock; 190229997Sken int num_disks; 191229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 192229997Sken int num_luns; 193229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 194229997Sken}; 195229997Sken 196229997Skenstatic struct ctl_be_block_softc backend_block_softc; 197229997Sken 198229997Sken/* 199229997Sken * Per-I/O information. 200229997Sken */ 201229997Skenstruct ctl_be_block_io { 202229997Sken union ctl_io *io; 203229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 204229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 205229997Sken int bio_cmd; 206229997Sken int num_segs; 207229997Sken int num_bios_sent; 208229997Sken int num_bios_done; 209229997Sken int send_complete; 210229997Sken int num_errors; 211229997Sken struct bintime ds_t0; 212229997Sken devstat_tag_type ds_tag_type; 213229997Sken devstat_trans_flags ds_trans_type; 214229997Sken uint64_t io_len; 215229997Sken uint64_t io_offset; 216229997Sken struct ctl_be_block_softc *softc; 217229997Sken struct ctl_be_block_lun *lun; 218264274Smav void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 219229997Sken}; 220229997Sken 221229997Skenstatic int cbb_num_threads = 14; 222229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 223229997Sken "CAM Target Layer Block Backend"); 224267992ShselaskySYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 225229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 226229997Sken 227229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 228229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 229229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 230229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 231229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 232229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 233229997Sken struct ctl_be_block_io *beio); 234229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 235229997Sken struct ctl_be_block_io *beio); 236229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 237229997Sken struct ctl_be_block_io *beio); 238264274Smavstatic void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 239264274Smav struct ctl_be_block_io *beio); 240229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 241229997Sken struct ctl_be_block_io *beio); 242229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 243229997Sken union ctl_io *io); 244229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 245229997Sken union ctl_io *io); 246229997Skenstatic void ctl_be_block_worker(void *context, int pending); 247229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 248229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 249229997Sken int flag, struct thread *td); 250229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 251229997Sken struct ctl_lun_req *req); 252229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 253229997Sken struct ctl_lun_req *req); 254229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 255229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 256229997Sken struct ctl_be_block_lun *be_lun, 257229997Sken struct ctl_lun_req *req); 258229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 259229997Sken struct ctl_lun_req *req); 260229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 261229997Sken struct ctl_lun_req *req); 262232604Straszstatic int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 263232604Strasz struct ctl_lun_req *req); 264232604Straszstatic int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 265232604Strasz struct ctl_lun_req *req); 266232604Straszstatic int ctl_be_block_modify(struct ctl_be_block_softc *softc, 267232604Strasz struct ctl_lun_req *req); 268229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 269229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 270229997Sken ctl_lun_config_status status); 271229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 272229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 273229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 274229997Skenint ctl_be_block_init(void); 275229997Sken 276229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 277229997Sken{ 278230334Sken .name = "block", 279230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 280230334Sken .init = ctl_be_block_init, 281230334Sken .data_submit = ctl_be_block_submit, 282230334Sken .data_move_done = ctl_be_block_move_done, 283230334Sken .config_read = ctl_be_block_config_read, 284230334Sken .config_write = ctl_be_block_config_write, 285230334Sken .ioctl = ctl_be_block_ioctl, 286230334Sken .lun_info = ctl_be_block_lun_info 287229997Sken}; 288229997Sken 289229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 290229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 291229997Sken 292264020Straszstatic uma_zone_t beio_zone; 293264020Strasz 294229997Skenstatic struct ctl_be_block_io * 295229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 296229997Sken{ 297229997Sken struct ctl_be_block_io *beio; 298229997Sken 299264020Strasz beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 300264020Strasz beio->softc = softc; 301229997Sken return (beio); 302229997Sken} 303229997Sken 304229997Skenstatic void 305229997Skenctl_free_beio(struct ctl_be_block_io *beio) 306229997Sken{ 307229997Sken int duplicate_free; 308229997Sken int i; 309229997Sken 310229997Sken duplicate_free = 0; 311229997Sken 312229997Sken for (i = 0; i < beio->num_segs; i++) { 313229997Sken if (beio->sg_segs[i].addr == NULL) 314229997Sken duplicate_free++; 315229997Sken 316229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 317229997Sken beio->sg_segs[i].addr = NULL; 318267537Smav 319267537Smav /* For compare we had two equal S/G lists. */ 320267537Smav if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 321267537Smav uma_zfree(beio->lun->lun_zone, 322267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 323267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 324267537Smav } 325229997Sken } 326229997Sken 327229997Sken if (duplicate_free > 0) { 328229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 329229997Sken duplicate_free, beio->num_segs); 330229997Sken } 331229997Sken 332264020Strasz uma_zfree(beio_zone, beio); 333229997Sken} 334229997Sken 335229997Skenstatic void 336229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 337229997Sken{ 338267877Smav union ctl_io *io = beio->io; 339229997Sken 340264274Smav if (beio->beio_cont != NULL) { 341264274Smav beio->beio_cont(beio); 342264274Smav } else { 343264274Smav ctl_free_beio(beio); 344267537Smav ctl_data_submit_done(io); 345264274Smav } 346229997Sken} 347229997Sken 348229997Skenstatic int 349229997Skenctl_be_block_move_done(union ctl_io *io) 350229997Sken{ 351229997Sken struct ctl_be_block_io *beio; 352229997Sken struct ctl_be_block_lun *be_lun; 353267537Smav struct ctl_lba_len_flags *lbalen; 354229997Sken#ifdef CTL_TIME_IO 355229997Sken struct bintime cur_bt; 356267537Smav#endif 357267537Smav int i; 358229997Sken 359267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 360229997Sken be_lun = beio->lun; 361229997Sken 362229997Sken DPRINTF("entered\n"); 363229997Sken 364229997Sken#ifdef CTL_TIME_IO 365229997Sken getbintime(&cur_bt); 366229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 367229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 368229997Sken io->io_hdr.num_dmas++; 369229997Sken#endif 370267537Smav io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 371229997Sken 372229997Sken /* 373229997Sken * We set status at this point for read commands, and write 374229997Sken * commands with errors. 375229997Sken */ 376267537Smav if ((io->io_hdr.port_status == 0) && 377267537Smav ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 378267537Smav ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 379267537Smav lbalen = ARGS(beio->io); 380267537Smav if (lbalen->flags & CTL_LLF_READ) { 381267537Smav ctl_set_success(&io->scsiio); 382267537Smav } else if (lbalen->flags & CTL_LLF_COMPARE) { 383267537Smav /* We have two data blocks ready for comparison. */ 384267537Smav for (i = 0; i < beio->num_segs; i++) { 385267537Smav if (memcmp(beio->sg_segs[i].addr, 386267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 387267537Smav beio->sg_segs[i].len) != 0) 388267537Smav break; 389267537Smav } 390267537Smav if (i < beio->num_segs) 391267537Smav ctl_set_sense(&io->scsiio, 392267537Smav /*current_error*/ 1, 393267537Smav /*sense_key*/ SSD_KEY_MISCOMPARE, 394267537Smav /*asc*/ 0x1D, 395267537Smav /*ascq*/ 0x00, 396267537Smav SSD_ELEM_NONE); 397267537Smav else 398267537Smav ctl_set_success(&io->scsiio); 399267537Smav } 400267537Smav } 401229997Sken else if ((io->io_hdr.port_status != 0) 402229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 403229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 404229997Sken /* 405229997Sken * For hardware error sense keys, the sense key 406229997Sken * specific value is defined to be a retry count, 407229997Sken * but we use it to pass back an internal FETD 408229997Sken * error code. XXX KDM Hopefully the FETD is only 409229997Sken * using 16 bits for an error code, since that's 410229997Sken * all the space we have in the sks field. 411229997Sken */ 412229997Sken ctl_set_internal_failure(&io->scsiio, 413229997Sken /*sks_valid*/ 1, 414229997Sken /*retry_count*/ 415229997Sken io->io_hdr.port_status); 416229997Sken } 417229997Sken 418229997Sken /* 419229997Sken * If this is a read, or a write with errors, it is done. 420229997Sken */ 421229997Sken if ((beio->bio_cmd == BIO_READ) 422229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 423229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 424229997Sken ctl_complete_beio(beio); 425229997Sken return (0); 426229997Sken } 427229997Sken 428229997Sken /* 429229997Sken * At this point, we have a write and the DMA completed 430229997Sken * successfully. We now have to queue it to the task queue to 431229997Sken * execute the backend I/O. That is because we do blocking 432229997Sken * memory allocations, and in the file backing case, blocking I/O. 433229997Sken * This move done routine is generally called in the SIM's 434229997Sken * interrupt context, and therefore we cannot block. 435229997Sken */ 436267877Smav mtx_lock(&be_lun->queue_lock); 437229997Sken /* 438229997Sken * XXX KDM make sure that links is okay to use at this point. 439229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 440229997Sken * or deal with resource allocation here. 441229997Sken */ 442229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 443267877Smav mtx_unlock(&be_lun->queue_lock); 444229997Sken 445229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 446229997Sken 447229997Sken return (0); 448229997Sken} 449229997Sken 450229997Skenstatic void 451229997Skenctl_be_block_biodone(struct bio *bio) 452229997Sken{ 453229997Sken struct ctl_be_block_io *beio; 454229997Sken struct ctl_be_block_lun *be_lun; 455229997Sken union ctl_io *io; 456261538Smav int error; 457229997Sken 458229997Sken beio = bio->bio_caller1; 459229997Sken be_lun = beio->lun; 460229997Sken io = beio->io; 461229997Sken 462229997Sken DPRINTF("entered\n"); 463229997Sken 464261538Smav error = bio->bio_error; 465267877Smav mtx_lock(&be_lun->io_lock); 466261538Smav if (error != 0) 467229997Sken beio->num_errors++; 468229997Sken 469229997Sken beio->num_bios_done++; 470229997Sken 471229997Sken /* 472229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 473229997Sken * during the free might cause it to complain. 474229997Sken */ 475229997Sken g_destroy_bio(bio); 476229997Sken 477229997Sken /* 478229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 479229997Sken * complete, then we're done. 480229997Sken */ 481229997Sken if ((beio->send_complete == 0) 482229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 483267877Smav mtx_unlock(&be_lun->io_lock); 484229997Sken return; 485229997Sken } 486229997Sken 487229997Sken /* 488229997Sken * At this point, we've verified that we are the last I/O to 489229997Sken * complete, so it's safe to drop the lock. 490229997Sken */ 491267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 492267877Smav beio->ds_tag_type, beio->ds_trans_type, 493267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 494267877Smav mtx_unlock(&be_lun->io_lock); 495229997Sken 496229997Sken /* 497229997Sken * If there are any errors from the backing device, we fail the 498229997Sken * entire I/O with a medium error. 499229997Sken */ 500229997Sken if (beio->num_errors > 0) { 501261538Smav if (error == EOPNOTSUPP) { 502261538Smav ctl_set_invalid_opcode(&io->scsiio); 503261538Smav } else if (beio->bio_cmd == BIO_FLUSH) { 504229997Sken /* XXX KDM is there is a better error here? */ 505229997Sken ctl_set_internal_failure(&io->scsiio, 506229997Sken /*sks_valid*/ 1, 507229997Sken /*retry_count*/ 0xbad2); 508229997Sken } else 509229997Sken ctl_set_medium_error(&io->scsiio); 510229997Sken ctl_complete_beio(beio); 511229997Sken return; 512229997Sken } 513229997Sken 514229997Sken /* 515267537Smav * If this is a write, a flush, a delete or verify, we're all done. 516229997Sken * If this is a read, we can now send the data to the user. 517229997Sken */ 518229997Sken if ((beio->bio_cmd == BIO_WRITE) 519264274Smav || (beio->bio_cmd == BIO_FLUSH) 520267537Smav || (beio->bio_cmd == BIO_DELETE) 521267537Smav || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 522229997Sken ctl_set_success(&io->scsiio); 523229997Sken ctl_complete_beio(beio); 524229997Sken } else { 525229997Sken#ifdef CTL_TIME_IO 526229997Sken getbintime(&io->io_hdr.dma_start_bt); 527229997Sken#endif 528229997Sken ctl_datamove(io); 529229997Sken } 530229997Sken} 531229997Sken 532229997Skenstatic void 533229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 534229997Sken struct ctl_be_block_io *beio) 535229997Sken{ 536267877Smav union ctl_io *io = beio->io; 537229997Sken struct mount *mountpoint; 538241896Skib int error, lock_flags; 539229997Sken 540229997Sken DPRINTF("entered\n"); 541229997Sken 542267877Smav binuptime(&beio->ds_t0); 543267877Smav mtx_lock(&be_lun->io_lock); 544267877Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 545267877Smav mtx_unlock(&be_lun->io_lock); 546229997Sken 547267877Smav (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 548229997Sken 549229997Sken if (MNT_SHARED_WRITES(mountpoint) 550229997Sken || ((mountpoint == NULL) 551229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 552229997Sken lock_flags = LK_SHARED; 553229997Sken else 554229997Sken lock_flags = LK_EXCLUSIVE; 555229997Sken 556229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 557229997Sken 558229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 559229997Sken VOP_UNLOCK(be_lun->vn, 0); 560229997Sken 561229997Sken vn_finished_write(mountpoint); 562229997Sken 563267877Smav mtx_lock(&be_lun->io_lock); 564267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 565267877Smav beio->ds_tag_type, beio->ds_trans_type, 566267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 567267877Smav mtx_unlock(&be_lun->io_lock); 568267877Smav 569229997Sken if (error == 0) 570229997Sken ctl_set_success(&io->scsiio); 571229997Sken else { 572229997Sken /* XXX KDM is there is a better error here? */ 573229997Sken ctl_set_internal_failure(&io->scsiio, 574229997Sken /*sks_valid*/ 1, 575229997Sken /*retry_count*/ 0xbad1); 576229997Sken } 577229997Sken 578229997Sken ctl_complete_beio(beio); 579229997Sken} 580229997Sken 581258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 582258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 583258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 584258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 585229997Sken 586229997Skenstatic void 587229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 588229997Sken struct ctl_be_block_io *beio) 589229997Sken{ 590229997Sken struct ctl_be_block_filedata *file_data; 591229997Sken union ctl_io *io; 592229997Sken struct uio xuio; 593229997Sken struct iovec *xiovec; 594241896Skib int flags; 595229997Sken int error, i; 596229997Sken 597229997Sken DPRINTF("entered\n"); 598229997Sken 599229997Sken file_data = &be_lun->backend.file; 600229997Sken io = beio->io; 601271309Smav flags = 0; 602271309Smav if (ARGS(io)->flags & CTL_LLF_DPO) 603271309Smav flags |= IO_DIRECT; 604271309Smav if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 605271309Smav flags |= IO_SYNC; 606229997Sken 607267537Smav bzero(&xuio, sizeof(xuio)); 608229997Sken if (beio->bio_cmd == BIO_READ) { 609229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 610267537Smav xuio.uio_rw = UIO_READ; 611229997Sken } else { 612229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 613267537Smav 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 627267877Smav binuptime(&beio->ds_t0); 628267877Smav mtx_lock(&be_lun->io_lock); 629267877Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 630267877Smav mtx_unlock(&be_lun->io_lock); 631267877Smav 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 */ 655271309Smav error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 656229997Sken 657229997Sken VOP_UNLOCK(be_lun->vn, 0); 658267537Smav SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 659229997Sken } else { 660229997Sken struct mount *mountpoint; 661229997Sken int lock_flags; 662229997Sken 663229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 664229997Sken 665229997Sken if (MNT_SHARED_WRITES(mountpoint) 666229997Sken || ((mountpoint == NULL) 667229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 668229997Sken lock_flags = LK_SHARED; 669229997Sken else 670229997Sken lock_flags = LK_EXCLUSIVE; 671229997Sken 672229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 673229997Sken 674229997Sken /* 675229997Sken * UFS pays attention to IO_DIRECT for writes. The write 676229997Sken * is done asynchronously. (Normally the write would just 677229997Sken * get put into cache. 678229997Sken * 679229997Sken * UFS pays attention to IO_SYNC for writes. It will 680229997Sken * attempt to write the buffer out synchronously if that 681229997Sken * flag is set. 682229997Sken * 683229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 684229997Sken * 685229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 686229997Sken * for writes. It will flush the transaction from the 687229997Sken * cache before returning. 688229997Sken * 689229997Sken * So if we've got the BIO_ORDERED flag set, we want 690229997Sken * IO_SYNC in either the UFS or ZFS case. 691229997Sken */ 692271309Smav error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 693229997Sken VOP_UNLOCK(be_lun->vn, 0); 694229997Sken 695229997Sken vn_finished_write(mountpoint); 696267537Smav SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 697229997Sken } 698229997Sken 699267877Smav mtx_lock(&be_lun->io_lock); 700267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 701267877Smav beio->ds_tag_type, beio->ds_trans_type, 702267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 703267877Smav mtx_unlock(&be_lun->io_lock); 704267877Smav 705229997Sken /* 706229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 707229997Sken * return the I/O to the user. 708229997Sken */ 709229997Sken if (error != 0) { 710229997Sken char path_str[32]; 711229997Sken 712229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 713229997Sken /* 714229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 715229997Sken * filesystem fills up. What kind of SCSI error should we 716229997Sken * return for that? 717229997Sken */ 718229997Sken printf("%s%s command returned errno %d\n", path_str, 719229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 720229997Sken ctl_set_medium_error(&io->scsiio); 721229997Sken ctl_complete_beio(beio); 722229997Sken return; 723229997Sken } 724229997Sken 725229997Sken /* 726269122Smav * If this is a write or a verify, we're all done. 727229997Sken * If this is a read, we can now send the data to the user. 728229997Sken */ 729269122Smav if ((beio->bio_cmd == BIO_WRITE) || 730269122Smav (ARGS(io)->flags & CTL_LLF_VERIFY)) { 731229997Sken ctl_set_success(&io->scsiio); 732229997Sken ctl_complete_beio(beio); 733229997Sken } else { 734229997Sken#ifdef CTL_TIME_IO 735229997Sken getbintime(&io->io_hdr.dma_start_bt); 736229997Sken#endif 737229997Sken ctl_datamove(io); 738229997Sken } 739229997Sken} 740229997Sken 741229997Skenstatic void 742269123Smavctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 743269123Smav struct ctl_be_block_io *beio) 744269123Smav{ 745269123Smav struct ctl_be_block_devdata *dev_data; 746269123Smav union ctl_io *io; 747269123Smav struct uio xuio; 748269123Smav struct iovec *xiovec; 749269123Smav int flags; 750269123Smav int error, i; 751269123Smav 752269123Smav DPRINTF("entered\n"); 753269123Smav 754269123Smav dev_data = &be_lun->backend.dev; 755269123Smav io = beio->io; 756271309Smav flags = 0; 757271309Smav if (ARGS(io)->flags & CTL_LLF_DPO) 758271309Smav flags |= IO_DIRECT; 759271309Smav if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 760271309Smav flags |= IO_SYNC; 761269123Smav 762269123Smav bzero(&xuio, sizeof(xuio)); 763269123Smav if (beio->bio_cmd == BIO_READ) { 764269123Smav SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 765269123Smav xuio.uio_rw = UIO_READ; 766269123Smav } else { 767269123Smav SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 768269123Smav xuio.uio_rw = UIO_WRITE; 769269123Smav } 770269123Smav xuio.uio_offset = beio->io_offset; 771269123Smav xuio.uio_resid = beio->io_len; 772269123Smav xuio.uio_segflg = UIO_SYSSPACE; 773269123Smav xuio.uio_iov = beio->xiovecs; 774269123Smav xuio.uio_iovcnt = beio->num_segs; 775269123Smav xuio.uio_td = curthread; 776269123Smav 777269123Smav for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 778269123Smav xiovec->iov_base = beio->sg_segs[i].addr; 779269123Smav xiovec->iov_len = beio->sg_segs[i].len; 780269123Smav } 781269123Smav 782269123Smav binuptime(&beio->ds_t0); 783269123Smav mtx_lock(&be_lun->io_lock); 784269123Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 785269123Smav mtx_unlock(&be_lun->io_lock); 786269123Smav 787269123Smav if (beio->bio_cmd == BIO_READ) { 788271309Smav error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 789269123Smav SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 790269123Smav } else { 791271309Smav error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 792269123Smav SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 793269123Smav } 794269123Smav 795269123Smav mtx_lock(&be_lun->io_lock); 796269123Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 797269123Smav beio->ds_tag_type, beio->ds_trans_type, 798269123Smav /*now*/ NULL, /*then*/&beio->ds_t0); 799269123Smav mtx_unlock(&be_lun->io_lock); 800269123Smav 801269123Smav /* 802269123Smav * If we got an error, set the sense data to "MEDIUM ERROR" and 803269123Smav * return the I/O to the user. 804269123Smav */ 805269123Smav if (error != 0) { 806269123Smav ctl_set_medium_error(&io->scsiio); 807269123Smav ctl_complete_beio(beio); 808269123Smav return; 809269123Smav } 810269123Smav 811269123Smav /* 812269123Smav * If this is a write or a verify, we're all done. 813269123Smav * If this is a read, we can now send the data to the user. 814269123Smav */ 815269123Smav if ((beio->bio_cmd == BIO_WRITE) || 816269123Smav (ARGS(io)->flags & CTL_LLF_VERIFY)) { 817269123Smav ctl_set_success(&io->scsiio); 818269123Smav ctl_complete_beio(beio); 819269123Smav } else { 820269123Smav#ifdef CTL_TIME_IO 821269123Smav getbintime(&io->io_hdr.dma_start_bt); 822269123Smav#endif 823269123Smav ctl_datamove(io); 824269123Smav } 825269123Smav} 826269123Smav 827269123Smavstatic void 828229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 829229997Sken struct ctl_be_block_io *beio) 830229997Sken{ 831229997Sken struct bio *bio; 832229997Sken union ctl_io *io; 833229997Sken struct ctl_be_block_devdata *dev_data; 834229997Sken 835229997Sken dev_data = &be_lun->backend.dev; 836229997Sken io = beio->io; 837229997Sken 838229997Sken DPRINTF("entered\n"); 839229997Sken 840229997Sken /* This can't fail, it's a blocking allocation. */ 841229997Sken bio = g_alloc_bio(); 842229997Sken 843229997Sken bio->bio_cmd = BIO_FLUSH; 844229997Sken bio->bio_flags |= BIO_ORDERED; 845229997Sken bio->bio_dev = dev_data->cdev; 846229997Sken bio->bio_offset = 0; 847229997Sken bio->bio_data = 0; 848229997Sken bio->bio_done = ctl_be_block_biodone; 849229997Sken bio->bio_caller1 = beio; 850229997Sken bio->bio_pblkno = 0; 851229997Sken 852229997Sken /* 853229997Sken * We don't need to acquire the LUN lock here, because we are only 854229997Sken * sending one bio, and so there is no other context to synchronize 855229997Sken * with. 856229997Sken */ 857229997Sken beio->num_bios_sent = 1; 858229997Sken beio->send_complete = 1; 859229997Sken 860229997Sken binuptime(&beio->ds_t0); 861267877Smav mtx_lock(&be_lun->io_lock); 862229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 863267877Smav mtx_unlock(&be_lun->io_lock); 864229997Sken 865229997Sken (*dev_data->csw->d_strategy)(bio); 866229997Sken} 867229997Sken 868229997Skenstatic void 869264274Smavctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 870264274Smav struct ctl_be_block_io *beio, 871264274Smav uint64_t off, uint64_t len, int last) 872264274Smav{ 873264274Smav struct bio *bio; 874264274Smav struct ctl_be_block_devdata *dev_data; 875264296Smav uint64_t maxlen; 876264274Smav 877264274Smav dev_data = &be_lun->backend.dev; 878264296Smav maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 879264274Smav while (len > 0) { 880264274Smav bio = g_alloc_bio(); 881264274Smav bio->bio_cmd = BIO_DELETE; 882264274Smav bio->bio_dev = dev_data->cdev; 883264274Smav bio->bio_offset = off; 884264296Smav bio->bio_length = MIN(len, maxlen); 885264274Smav bio->bio_data = 0; 886264274Smav bio->bio_done = ctl_be_block_biodone; 887264274Smav bio->bio_caller1 = beio; 888264296Smav bio->bio_pblkno = off / be_lun->blocksize; 889264274Smav 890264274Smav off += bio->bio_length; 891264274Smav len -= bio->bio_length; 892264274Smav 893267877Smav mtx_lock(&be_lun->io_lock); 894264274Smav beio->num_bios_sent++; 895264274Smav if (last && len == 0) 896264274Smav beio->send_complete = 1; 897267877Smav mtx_unlock(&be_lun->io_lock); 898264274Smav 899264274Smav (*dev_data->csw->d_strategy)(bio); 900264274Smav } 901264274Smav} 902264274Smav 903264274Smavstatic void 904264274Smavctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 905264274Smav struct ctl_be_block_io *beio) 906264274Smav{ 907264274Smav union ctl_io *io; 908264274Smav struct ctl_be_block_devdata *dev_data; 909267515Smav struct ctl_ptr_len_flags *ptrlen; 910264274Smav struct scsi_unmap_desc *buf, *end; 911264274Smav uint64_t len; 912264274Smav 913264274Smav dev_data = &be_lun->backend.dev; 914264274Smav io = beio->io; 915264274Smav 916264274Smav DPRINTF("entered\n"); 917264274Smav 918264274Smav binuptime(&beio->ds_t0); 919267877Smav mtx_lock(&be_lun->io_lock); 920264274Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 921267877Smav mtx_unlock(&be_lun->io_lock); 922264274Smav 923264274Smav if (beio->io_offset == -1) { 924264274Smav beio->io_len = 0; 925267515Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 926267515Smav buf = (struct scsi_unmap_desc *)ptrlen->ptr; 927267515Smav end = buf + ptrlen->len / sizeof(*buf); 928264274Smav for (; buf < end; buf++) { 929264274Smav len = (uint64_t)scsi_4btoul(buf->length) * 930264274Smav be_lun->blocksize; 931264274Smav beio->io_len += len; 932264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 933264274Smav scsi_8btou64(buf->lba) * be_lun->blocksize, len, 934264283Smav (end - buf < 2) ? TRUE : FALSE); 935264274Smav } 936264274Smav } else 937264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 938264274Smav beio->io_offset, beio->io_len, TRUE); 939264274Smav} 940264274Smav 941264274Smavstatic void 942229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 943229997Sken struct ctl_be_block_io *beio) 944229997Sken{ 945267877Smav TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 946229997Sken int i; 947229997Sken struct bio *bio; 948229997Sken struct ctl_be_block_devdata *dev_data; 949229997Sken off_t cur_offset; 950229997Sken int max_iosize; 951229997Sken 952229997Sken DPRINTF("entered\n"); 953229997Sken 954229997Sken dev_data = &be_lun->backend.dev; 955229997Sken 956229997Sken /* 957229997Sken * We have to limit our I/O size to the maximum supported by the 958229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 959229997Sken * set it properly, use DFLTPHYS. 960229997Sken */ 961229997Sken max_iosize = dev_data->cdev->si_iosize_max; 962229997Sken if (max_iosize < PAGE_SIZE) 963229997Sken max_iosize = DFLTPHYS; 964229997Sken 965229997Sken cur_offset = beio->io_offset; 966229997Sken for (i = 0; i < beio->num_segs; i++) { 967229997Sken size_t cur_size; 968229997Sken uint8_t *cur_ptr; 969229997Sken 970229997Sken cur_size = beio->sg_segs[i].len; 971229997Sken cur_ptr = beio->sg_segs[i].addr; 972229997Sken 973229997Sken while (cur_size > 0) { 974229997Sken /* This can't fail, it's a blocking allocation. */ 975229997Sken bio = g_alloc_bio(); 976229997Sken 977229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 978229997Sken 979229997Sken bio->bio_cmd = beio->bio_cmd; 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 992267877Smav TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 993229997Sken beio->num_bios_sent++; 994229997Sken } 995229997Sken } 996267877Smav binuptime(&beio->ds_t0); 997267877Smav mtx_lock(&be_lun->io_lock); 998267877Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 999267877Smav beio->send_complete = 1; 1000267877Smav mtx_unlock(&be_lun->io_lock); 1001267877Smav 1002267877Smav /* 1003267877Smav * Fire off all allocated requests! 1004267877Smav */ 1005267877Smav while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1006267877Smav TAILQ_REMOVE(&queue, bio, bio_queue); 1007267877Smav (*dev_data->csw->d_strategy)(bio); 1008267877Smav } 1009229997Sken} 1010229997Sken 1011229997Skenstatic void 1012264274Smavctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1013264274Smav{ 1014264274Smav union ctl_io *io; 1015264274Smav 1016264274Smav io = beio->io; 1017264274Smav ctl_free_beio(beio); 1018267641Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1019267641Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1020267641Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1021264274Smav ctl_config_write_done(io); 1022264274Smav return; 1023264274Smav } 1024264274Smav 1025264274Smav ctl_be_block_config_write(io); 1026264274Smav} 1027264274Smav 1028264274Smavstatic void 1029264274Smavctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1030264274Smav union ctl_io *io) 1031264274Smav{ 1032264274Smav struct ctl_be_block_io *beio; 1033264274Smav struct ctl_be_block_softc *softc; 1034267515Smav struct ctl_lba_len_flags *lbalen; 1035264274Smav uint64_t len_left, lba; 1036264274Smav int i, seglen; 1037264274Smav uint8_t *buf, *end; 1038264274Smav 1039264274Smav DPRINTF("entered\n"); 1040264274Smav 1041267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1042264274Smav softc = be_lun->softc; 1043267537Smav lbalen = ARGS(beio->io); 1044264274Smav 1045269622Smav if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR) || 1046269622Smav (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1047264274Smav ctl_free_beio(beio); 1048264274Smav ctl_set_invalid_field(&io->scsiio, 1049264274Smav /*sks_valid*/ 1, 1050264274Smav /*command*/ 1, 1051264274Smav /*field*/ 1, 1052264274Smav /*bit_valid*/ 0, 1053264274Smav /*bit*/ 0); 1054264274Smav ctl_config_write_done(io); 1055264274Smav return; 1056264274Smav } 1057264274Smav 1058264274Smav switch (io->scsiio.tag_type) { 1059264274Smav case CTL_TAG_ORDERED: 1060264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1061264274Smav break; 1062264274Smav case CTL_TAG_HEAD_OF_QUEUE: 1063264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1064264274Smav break; 1065264274Smav case CTL_TAG_UNTAGGED: 1066264274Smav case CTL_TAG_SIMPLE: 1067264274Smav case CTL_TAG_ACA: 1068264274Smav default: 1069264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1070264274Smav break; 1071264274Smav } 1072264274Smav 1073269622Smav if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1074267515Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 1075267515Smav beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1076264274Smav beio->bio_cmd = BIO_DELETE; 1077264274Smav beio->ds_trans_type = DEVSTAT_FREE; 1078264274Smav 1079264274Smav be_lun->unmap(be_lun, beio); 1080264274Smav return; 1081264274Smav } 1082264274Smav 1083264274Smav beio->bio_cmd = BIO_WRITE; 1084264274Smav beio->ds_trans_type = DEVSTAT_WRITE; 1085264274Smav 1086264274Smav DPRINTF("WRITE SAME at LBA %jx len %u\n", 1087267515Smav (uintmax_t)lbalen->lba, lbalen->len); 1088264274Smav 1089267515Smav len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1090264274Smav for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1091264274Smav 1092264274Smav /* 1093264274Smav * Setup the S/G entry for this chunk. 1094264274Smav */ 1095264886Smav seglen = MIN(CTLBLK_MAX_SEG, len_left); 1096264274Smav seglen -= seglen % be_lun->blocksize; 1097264274Smav beio->sg_segs[i].len = seglen; 1098264274Smav beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1099264274Smav 1100264274Smav DPRINTF("segment %d addr %p len %zd\n", i, 1101264274Smav beio->sg_segs[i].addr, beio->sg_segs[i].len); 1102264274Smav 1103264274Smav beio->num_segs++; 1104264274Smav len_left -= seglen; 1105264274Smav 1106264274Smav buf = beio->sg_segs[i].addr; 1107264274Smav end = buf + seglen; 1108264274Smav for (; buf < end; buf += be_lun->blocksize) { 1109264274Smav memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1110267515Smav if (lbalen->flags & SWS_LBDATA) 1111267515Smav scsi_ulto4b(lbalen->lba + lba, buf); 1112264274Smav lba++; 1113264274Smav } 1114264274Smav } 1115264274Smav 1116267515Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 1117264274Smav beio->io_len = lba * be_lun->blocksize; 1118264274Smav 1119264274Smav /* We can not do all in one run. Correct and schedule rerun. */ 1120264274Smav if (len_left > 0) { 1121267515Smav lbalen->lba += lba; 1122267515Smav lbalen->len -= lba; 1123264274Smav beio->beio_cont = ctl_be_block_cw_done_ws; 1124264274Smav } 1125264274Smav 1126264274Smav be_lun->dispatch(be_lun, beio); 1127264274Smav} 1128264274Smav 1129264274Smavstatic void 1130264274Smavctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1131264274Smav union ctl_io *io) 1132264274Smav{ 1133264274Smav struct ctl_be_block_io *beio; 1134264274Smav struct ctl_be_block_softc *softc; 1135267515Smav struct ctl_ptr_len_flags *ptrlen; 1136264274Smav 1137264274Smav DPRINTF("entered\n"); 1138264274Smav 1139267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1140264274Smav softc = be_lun->softc; 1141267515Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1142264274Smav 1143269622Smav if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1144264274Smav ctl_free_beio(beio); 1145264274Smav ctl_set_invalid_field(&io->scsiio, 1146264274Smav /*sks_valid*/ 0, 1147264274Smav /*command*/ 1, 1148264274Smav /*field*/ 0, 1149264274Smav /*bit_valid*/ 0, 1150264274Smav /*bit*/ 0); 1151264274Smav ctl_config_write_done(io); 1152264274Smav return; 1153264274Smav } 1154264274Smav 1155264274Smav switch (io->scsiio.tag_type) { 1156264274Smav case CTL_TAG_ORDERED: 1157264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1158264274Smav break; 1159264274Smav case CTL_TAG_HEAD_OF_QUEUE: 1160264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1161264274Smav break; 1162264274Smav case CTL_TAG_UNTAGGED: 1163264274Smav case CTL_TAG_SIMPLE: 1164264274Smav case CTL_TAG_ACA: 1165264274Smav default: 1166264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1167264274Smav break; 1168264274Smav } 1169264274Smav 1170264274Smav beio->io_len = 0; 1171264274Smav beio->io_offset = -1; 1172264274Smav 1173264274Smav beio->bio_cmd = BIO_DELETE; 1174264274Smav beio->ds_trans_type = DEVSTAT_FREE; 1175264274Smav 1176267515Smav DPRINTF("UNMAP\n"); 1177264274Smav 1178264274Smav be_lun->unmap(be_lun, beio); 1179264274Smav} 1180264274Smav 1181264274Smavstatic void 1182264274Smavctl_be_block_cw_done(struct ctl_be_block_io *beio) 1183264274Smav{ 1184264274Smav union ctl_io *io; 1185264274Smav 1186264274Smav io = beio->io; 1187264274Smav ctl_free_beio(beio); 1188264274Smav ctl_config_write_done(io); 1189264274Smav} 1190264274Smav 1191264274Smavstatic void 1192229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1193229997Sken union ctl_io *io) 1194229997Sken{ 1195229997Sken struct ctl_be_block_io *beio; 1196229997Sken struct ctl_be_block_softc *softc; 1197229997Sken 1198229997Sken DPRINTF("entered\n"); 1199229997Sken 1200229997Sken softc = be_lun->softc; 1201229997Sken beio = ctl_alloc_beio(softc); 1202229997Sken beio->io = io; 1203229997Sken beio->lun = be_lun; 1204264274Smav beio->beio_cont = ctl_be_block_cw_done; 1205267519Smav PRIV(io)->ptr = (void *)beio; 1206229997Sken 1207229997Sken switch (io->scsiio.cdb[0]) { 1208229997Sken case SYNCHRONIZE_CACHE: 1209229997Sken case SYNCHRONIZE_CACHE_16: 1210249194Strasz beio->bio_cmd = BIO_FLUSH; 1211229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 1212229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1213229997Sken beio->io_len = 0; 1214229997Sken be_lun->lun_flush(be_lun, beio); 1215229997Sken break; 1216264274Smav case WRITE_SAME_10: 1217264274Smav case WRITE_SAME_16: 1218264274Smav ctl_be_block_cw_dispatch_ws(be_lun, io); 1219264274Smav break; 1220264274Smav case UNMAP: 1221264274Smav ctl_be_block_cw_dispatch_unmap(be_lun, io); 1222264274Smav break; 1223229997Sken default: 1224229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1225229997Sken break; 1226229997Sken } 1227229997Sken} 1228229997Sken 1229258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1230258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1231258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1232258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1233229997Sken 1234229997Skenstatic void 1235264886Smavctl_be_block_next(struct ctl_be_block_io *beio) 1236264886Smav{ 1237264886Smav struct ctl_be_block_lun *be_lun; 1238264886Smav union ctl_io *io; 1239264886Smav 1240264886Smav io = beio->io; 1241264886Smav be_lun = beio->lun; 1242264886Smav ctl_free_beio(beio); 1243267641Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1244267641Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1245267641Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1246267537Smav ctl_data_submit_done(io); 1247264886Smav return; 1248264886Smav } 1249264886Smav 1250264886Smav io->io_hdr.status &= ~CTL_STATUS_MASK; 1251264886Smav io->io_hdr.status |= CTL_STATUS_NONE; 1252264886Smav 1253267877Smav mtx_lock(&be_lun->queue_lock); 1254264886Smav /* 1255264886Smav * XXX KDM make sure that links is okay to use at this point. 1256264886Smav * Otherwise, we either need to add another field to ctl_io_hdr, 1257264886Smav * or deal with resource allocation here. 1258264886Smav */ 1259264886Smav STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1260267877Smav mtx_unlock(&be_lun->queue_lock); 1261264886Smav 1262264886Smav taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1263264886Smav} 1264264886Smav 1265264886Smavstatic void 1266229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1267229997Sken union ctl_io *io) 1268229997Sken{ 1269229997Sken struct ctl_be_block_io *beio; 1270229997Sken struct ctl_be_block_softc *softc; 1271267537Smav struct ctl_lba_len_flags *lbalen; 1272267519Smav struct ctl_ptr_len_flags *bptrlen; 1273267519Smav uint64_t len_left, lbas; 1274229997Sken int i; 1275229997Sken 1276229997Sken softc = be_lun->softc; 1277229997Sken 1278229997Sken DPRINTF("entered\n"); 1279229997Sken 1280267537Smav lbalen = ARGS(io); 1281267537Smav if (lbalen->flags & CTL_LLF_WRITE) { 1282267537Smav SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1283267537Smav } else { 1284229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1285229997Sken } 1286229997Sken 1287229997Sken beio = ctl_alloc_beio(softc); 1288229997Sken beio->io = io; 1289229997Sken beio->lun = be_lun; 1290267519Smav bptrlen = PRIV(io); 1291267519Smav bptrlen->ptr = (void *)beio; 1292229997Sken 1293229997Sken switch (io->scsiio.tag_type) { 1294229997Sken case CTL_TAG_ORDERED: 1295229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1296229997Sken break; 1297229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1298229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1299229997Sken break; 1300229997Sken case CTL_TAG_UNTAGGED: 1301229997Sken case CTL_TAG_SIMPLE: 1302229997Sken case CTL_TAG_ACA: 1303229997Sken default: 1304229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1305229997Sken break; 1306229997Sken } 1307229997Sken 1308267537Smav if (lbalen->flags & CTL_LLF_WRITE) { 1309267537Smav beio->bio_cmd = BIO_WRITE; 1310267537Smav beio->ds_trans_type = DEVSTAT_WRITE; 1311267537Smav } else { 1312229997Sken beio->bio_cmd = BIO_READ; 1313229997Sken beio->ds_trans_type = DEVSTAT_READ; 1314229997Sken } 1315229997Sken 1316264886Smav DPRINTF("%s at LBA %jx len %u @%ju\n", 1317229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1318267519Smav (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1319267537Smav if (lbalen->flags & CTL_LLF_COMPARE) 1320267537Smav lbas = CTLBLK_HALF_IO_SIZE; 1321267537Smav else 1322267537Smav lbas = CTLBLK_MAX_IO_SIZE; 1323267537Smav lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1324267519Smav beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1325267519Smav beio->io_len = lbas * be_lun->blocksize; 1326267519Smav bptrlen->len += lbas; 1327229997Sken 1328264886Smav for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1329264886Smav KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1330264886Smav i, CTLBLK_MAX_SEGS)); 1331229997Sken 1332229997Sken /* 1333229997Sken * Setup the S/G entry for this chunk. 1334229997Sken */ 1335264886Smav beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1336229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1337229997Sken 1338229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1339229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1340229997Sken 1341267537Smav /* Set up second segment for compare operation. */ 1342267537Smav if (lbalen->flags & CTL_LLF_COMPARE) { 1343267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1344267537Smav beio->sg_segs[i].len; 1345267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1346267537Smav uma_zalloc(be_lun->lun_zone, M_WAITOK); 1347267537Smav } 1348267537Smav 1349229997Sken beio->num_segs++; 1350229997Sken len_left -= beio->sg_segs[i].len; 1351229997Sken } 1352267519Smav if (bptrlen->len < lbalen->len) 1353264886Smav beio->beio_cont = ctl_be_block_next; 1354264886Smav io->scsiio.be_move_done = ctl_be_block_move_done; 1355267537Smav /* For compare we have separate S/G lists for read and datamove. */ 1356267537Smav if (lbalen->flags & CTL_LLF_COMPARE) 1357267537Smav io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1358267537Smav else 1359267537Smav io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1360264886Smav io->scsiio.kern_data_len = beio->io_len; 1361264886Smav io->scsiio.kern_data_resid = 0; 1362264886Smav io->scsiio.kern_sg_entries = beio->num_segs; 1363264886Smav io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1364229997Sken 1365229997Sken /* 1366229997Sken * For the read case, we need to read the data into our buffers and 1367229997Sken * then we can send it back to the user. For the write case, we 1368229997Sken * need to get the data from the user first. 1369229997Sken */ 1370229997Sken if (beio->bio_cmd == BIO_READ) { 1371229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1372229997Sken be_lun->dispatch(be_lun, beio); 1373229997Sken } else { 1374229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1375229997Sken#ifdef CTL_TIME_IO 1376229997Sken getbintime(&io->io_hdr.dma_start_bt); 1377229997Sken#endif 1378229997Sken ctl_datamove(io); 1379229997Sken } 1380229997Sken} 1381229997Sken 1382229997Skenstatic void 1383229997Skenctl_be_block_worker(void *context, int pending) 1384229997Sken{ 1385229997Sken struct ctl_be_block_lun *be_lun; 1386229997Sken struct ctl_be_block_softc *softc; 1387229997Sken union ctl_io *io; 1388229997Sken 1389229997Sken be_lun = (struct ctl_be_block_lun *)context; 1390229997Sken softc = be_lun->softc; 1391229997Sken 1392229997Sken DPRINTF("entered\n"); 1393229997Sken 1394267877Smav mtx_lock(&be_lun->queue_lock); 1395229997Sken for (;;) { 1396229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1397229997Sken if (io != NULL) { 1398229997Sken struct ctl_be_block_io *beio; 1399229997Sken 1400229997Sken DPRINTF("datamove queue\n"); 1401229997Sken 1402229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1403229997Sken ctl_io_hdr, links); 1404229997Sken 1405267877Smav mtx_unlock(&be_lun->queue_lock); 1406229997Sken 1407267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1408229997Sken 1409229997Sken be_lun->dispatch(be_lun, beio); 1410229997Sken 1411267877Smav mtx_lock(&be_lun->queue_lock); 1412229997Sken continue; 1413229997Sken } 1414229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1415229997Sken if (io != NULL) { 1416229997Sken 1417229997Sken DPRINTF("config write queue\n"); 1418229997Sken 1419229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1420229997Sken ctl_io_hdr, links); 1421229997Sken 1422267877Smav mtx_unlock(&be_lun->queue_lock); 1423229997Sken 1424229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1425229997Sken 1426267877Smav mtx_lock(&be_lun->queue_lock); 1427229997Sken continue; 1428229997Sken } 1429229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1430229997Sken if (io != NULL) { 1431229997Sken DPRINTF("input queue\n"); 1432229997Sken 1433229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1434229997Sken ctl_io_hdr, links); 1435267877Smav mtx_unlock(&be_lun->queue_lock); 1436229997Sken 1437229997Sken /* 1438229997Sken * We must drop the lock, since this routine and 1439229997Sken * its children may sleep. 1440229997Sken */ 1441229997Sken ctl_be_block_dispatch(be_lun, io); 1442229997Sken 1443267877Smav mtx_lock(&be_lun->queue_lock); 1444229997Sken continue; 1445229997Sken } 1446229997Sken 1447229997Sken /* 1448229997Sken * If we get here, there is no work left in the queues, so 1449229997Sken * just break out and let the task queue go to sleep. 1450229997Sken */ 1451229997Sken break; 1452229997Sken } 1453267877Smav mtx_unlock(&be_lun->queue_lock); 1454229997Sken} 1455229997Sken 1456229997Sken/* 1457229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1458229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1459229997Sken * thread. 1460229997Sken */ 1461229997Skenstatic int 1462229997Skenctl_be_block_submit(union ctl_io *io) 1463229997Sken{ 1464229997Sken struct ctl_be_block_lun *be_lun; 1465229997Sken struct ctl_be_lun *ctl_be_lun; 1466229997Sken 1467229997Sken DPRINTF("entered\n"); 1468229997Sken 1469229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1470229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1471229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1472229997Sken 1473229997Sken /* 1474229997Sken * Make sure we only get SCSI I/O. 1475229997Sken */ 1476229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1477229997Sken "%#x) encountered", io->io_hdr.io_type)); 1478229997Sken 1479267519Smav PRIV(io)->len = 0; 1480267519Smav 1481267877Smav mtx_lock(&be_lun->queue_lock); 1482229997Sken /* 1483229997Sken * XXX KDM make sure that links is okay to use at this point. 1484229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1485229997Sken * or deal with resource allocation here. 1486229997Sken */ 1487229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1488267877Smav mtx_unlock(&be_lun->queue_lock); 1489229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1490229997Sken 1491267514Smav return (CTL_RETVAL_COMPLETE); 1492229997Sken} 1493229997Sken 1494229997Skenstatic int 1495229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1496229997Sken int flag, struct thread *td) 1497229997Sken{ 1498229997Sken struct ctl_be_block_softc *softc; 1499229997Sken int error; 1500229997Sken 1501229997Sken softc = &backend_block_softc; 1502229997Sken 1503229997Sken error = 0; 1504229997Sken 1505229997Sken switch (cmd) { 1506229997Sken case CTL_LUN_REQ: { 1507229997Sken struct ctl_lun_req *lun_req; 1508229997Sken 1509229997Sken lun_req = (struct ctl_lun_req *)addr; 1510229997Sken 1511229997Sken switch (lun_req->reqtype) { 1512229997Sken case CTL_LUNREQ_CREATE: 1513229997Sken error = ctl_be_block_create(softc, lun_req); 1514229997Sken break; 1515229997Sken case CTL_LUNREQ_RM: 1516229997Sken error = ctl_be_block_rm(softc, lun_req); 1517229997Sken break; 1518232604Strasz case CTL_LUNREQ_MODIFY: 1519232604Strasz error = ctl_be_block_modify(softc, lun_req); 1520232604Strasz break; 1521229997Sken default: 1522229997Sken lun_req->status = CTL_LUN_ERROR; 1523229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1524229997Sken "%s: invalid LUN request type %d", __func__, 1525229997Sken lun_req->reqtype); 1526229997Sken break; 1527229997Sken } 1528229997Sken break; 1529229997Sken } 1530229997Sken default: 1531229997Sken error = ENOTTY; 1532229997Sken break; 1533229997Sken } 1534229997Sken 1535229997Sken return (error); 1536229997Sken} 1537229997Sken 1538229997Skenstatic int 1539229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1540229997Sken{ 1541229997Sken struct ctl_be_block_filedata *file_data; 1542229997Sken struct ctl_lun_create_params *params; 1543229997Sken struct vattr vattr; 1544229997Sken int error; 1545229997Sken 1546229997Sken error = 0; 1547229997Sken file_data = &be_lun->backend.file; 1548229997Sken params = &req->reqdata.create; 1549229997Sken 1550229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1551229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1552229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1553229997Sken 1554229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1555229997Sken if (error != 0) { 1556229997Sken snprintf(req->error_str, sizeof(req->error_str), 1557229997Sken "error calling VOP_GETATTR() for file %s", 1558229997Sken be_lun->dev_path); 1559229997Sken return (error); 1560229997Sken } 1561229997Sken 1562229997Sken /* 1563229997Sken * Verify that we have the ability to upgrade to exclusive 1564229997Sken * access on this file so we can trap errors at open instead 1565229997Sken * of reporting them during first access. 1566229997Sken */ 1567229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1568229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1569229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1570229997Sken error = EBADF; 1571229997Sken snprintf(req->error_str, sizeof(req->error_str), 1572229997Sken "error locking file %s", be_lun->dev_path); 1573229997Sken return (error); 1574229997Sken } 1575229997Sken } 1576229997Sken 1577229997Sken 1578229997Sken file_data->cred = crhold(curthread->td_ucred); 1579232604Strasz if (params->lun_size_bytes != 0) 1580232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1581232604Strasz else 1582232604Strasz be_lun->size_bytes = vattr.va_size; 1583229997Sken /* 1584229997Sken * We set the multi thread flag for file operations because all 1585229997Sken * filesystems (in theory) are capable of allowing multiple readers 1586229997Sken * of a file at once. So we want to get the maximum possible 1587229997Sken * concurrency. 1588229997Sken */ 1589229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1590229997Sken 1591229997Sken /* 1592229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1593229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1594229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1595229997Sken * may not work with other OSes as well. So just export a sector 1596229997Sken * size of 512 bytes, which should work with any OS or 1597229997Sken * application. Since our backing is a file, any block size will 1598229997Sken * work fine for the backing store. 1599229997Sken */ 1600229997Sken#if 0 1601229997Sken be_lun->blocksize= vattr.va_blocksize; 1602229997Sken#endif 1603229997Sken if (params->blocksize_bytes != 0) 1604229997Sken be_lun->blocksize = params->blocksize_bytes; 1605229997Sken else 1606229997Sken be_lun->blocksize = 512; 1607229997Sken 1608229997Sken /* 1609229997Sken * Sanity check. The media size has to be at least one 1610229997Sken * sector long. 1611229997Sken */ 1612229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1613229997Sken error = EINVAL; 1614229997Sken snprintf(req->error_str, sizeof(req->error_str), 1615229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1616229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1617229997Sken } 1618229997Sken return (error); 1619229997Sken} 1620229997Sken 1621229997Skenstatic int 1622229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1623229997Sken{ 1624229997Sken struct ctl_lun_create_params *params; 1625229997Sken struct vattr vattr; 1626229997Sken struct cdev *dev; 1627229997Sken struct cdevsw *devsw; 1628229997Sken int error; 1629264191Smav off_t ps, pss, po, pos; 1630229997Sken 1631229997Sken params = &req->reqdata.create; 1632229997Sken 1633229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1634229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1635229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1636229997Sken &be_lun->backend.dev.dev_ref); 1637229997Sken if (be_lun->backend.dev.csw == NULL) 1638229997Sken panic("Unable to retrieve device switch"); 1639269123Smav if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) 1640269123Smav be_lun->dispatch = ctl_be_block_dispatch_zvol; 1641269123Smav else 1642269123Smav be_lun->dispatch = ctl_be_block_dispatch_dev; 1643269123Smav be_lun->lun_flush = ctl_be_block_flush_dev; 1644269123Smav be_lun->unmap = ctl_be_block_unmap_dev; 1645229997Sken 1646229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1647229997Sken if (error) { 1648229997Sken snprintf(req->error_str, sizeof(req->error_str), 1649229997Sken "%s: error getting vnode attributes for device %s", 1650229997Sken __func__, be_lun->dev_path); 1651229997Sken return (error); 1652229997Sken } 1653229997Sken 1654229997Sken dev = be_lun->vn->v_rdev; 1655229997Sken devsw = dev->si_devsw; 1656229997Sken if (!devsw->d_ioctl) { 1657229997Sken snprintf(req->error_str, sizeof(req->error_str), 1658229997Sken "%s: no d_ioctl for device %s!", __func__, 1659229997Sken be_lun->dev_path); 1660229997Sken return (ENODEV); 1661229997Sken } 1662229997Sken 1663229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1664229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1665229997Sken curthread); 1666229997Sken if (error) { 1667229997Sken snprintf(req->error_str, sizeof(req->error_str), 1668229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1669229997Sken "on %s!", __func__, error, be_lun->dev_path); 1670229997Sken return (error); 1671229997Sken } 1672229997Sken 1673229997Sken /* 1674229997Sken * If the user has asked for a blocksize that is greater than the 1675229997Sken * backing device's blocksize, we can do it only if the blocksize 1676229997Sken * the user is asking for is an even multiple of the underlying 1677229997Sken * device's blocksize. 1678229997Sken */ 1679229997Sken if ((params->blocksize_bytes != 0) 1680229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1681229997Sken uint32_t bs_multiple, tmp_blocksize; 1682229997Sken 1683229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1684229997Sken 1685229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1686229997Sken 1687229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1688229997Sken be_lun->blocksize = params->blocksize_bytes; 1689229997Sken } else { 1690229997Sken snprintf(req->error_str, sizeof(req->error_str), 1691229997Sken "%s: requested blocksize %u is not an even " 1692229997Sken "multiple of backing device blocksize %u", 1693229997Sken __func__, params->blocksize_bytes, 1694229997Sken be_lun->blocksize); 1695229997Sken return (EINVAL); 1696229997Sken 1697229997Sken } 1698229997Sken } else if ((params->blocksize_bytes != 0) 1699229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1700229997Sken snprintf(req->error_str, sizeof(req->error_str), 1701229997Sken "%s: requested blocksize %u < backing device " 1702229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1703229997Sken be_lun->blocksize); 1704229997Sken return (EINVAL); 1705229997Sken } 1706229997Sken 1707229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1708229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1709229997Sken curthread); 1710229997Sken if (error) { 1711229997Sken snprintf(req->error_str, sizeof(req->error_str), 1712232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1713232604Strasz " ioctl on %s!", __func__, error, 1714232604Strasz be_lun->dev_path); 1715229997Sken return (error); 1716229997Sken } 1717229997Sken 1718232604Strasz if (params->lun_size_bytes != 0) { 1719232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1720232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1721232604Strasz "%s: requested LUN size %ju > backing device " 1722232604Strasz "size %ju", __func__, 1723232604Strasz (uintmax_t)params->lun_size_bytes, 1724232604Strasz (uintmax_t)be_lun->size_bytes); 1725232604Strasz return (EINVAL); 1726232604Strasz } 1727232604Strasz 1728232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1729232604Strasz } 1730232604Strasz 1731264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1732264191Smav (caddr_t)&ps, FREAD, curthread); 1733264191Smav if (error) 1734264191Smav ps = po = 0; 1735264191Smav else { 1736264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1737264191Smav (caddr_t)&po, FREAD, curthread); 1738264191Smav if (error) 1739264191Smav po = 0; 1740264191Smav } 1741264191Smav pss = ps / be_lun->blocksize; 1742264191Smav pos = po / be_lun->blocksize; 1743264191Smav if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1744264191Smav ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1745264191Smav be_lun->pblockexp = fls(pss) - 1; 1746264191Smav be_lun->pblockoff = (pss - pos) % pss; 1747264191Smav } 1748264191Smav 1749229997Sken return (0); 1750229997Sken} 1751229997Sken 1752229997Skenstatic int 1753229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1754229997Sken{ 1755229997Sken DROP_GIANT(); 1756229997Sken if (be_lun->vn) { 1757229997Sken int flags = FREAD | FWRITE; 1758229997Sken 1759229997Sken switch (be_lun->dev_type) { 1760229997Sken case CTL_BE_BLOCK_DEV: 1761229997Sken if (be_lun->backend.dev.csw) { 1762229997Sken dev_relthread(be_lun->backend.dev.cdev, 1763229997Sken be_lun->backend.dev.dev_ref); 1764229997Sken be_lun->backend.dev.csw = NULL; 1765229997Sken be_lun->backend.dev.cdev = NULL; 1766229997Sken } 1767229997Sken break; 1768229997Sken case CTL_BE_BLOCK_FILE: 1769229997Sken break; 1770229997Sken case CTL_BE_BLOCK_NONE: 1771258871Strasz break; 1772229997Sken default: 1773229997Sken panic("Unexpected backend type."); 1774229997Sken break; 1775229997Sken } 1776229997Sken 1777229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1778229997Sken be_lun->vn = NULL; 1779229997Sken 1780229997Sken switch (be_lun->dev_type) { 1781229997Sken case CTL_BE_BLOCK_DEV: 1782229997Sken break; 1783229997Sken case CTL_BE_BLOCK_FILE: 1784229997Sken if (be_lun->backend.file.cred != NULL) { 1785229997Sken crfree(be_lun->backend.file.cred); 1786229997Sken be_lun->backend.file.cred = NULL; 1787229997Sken } 1788229997Sken break; 1789229997Sken case CTL_BE_BLOCK_NONE: 1790258871Strasz break; 1791229997Sken default: 1792229997Sken panic("Unexpected backend type."); 1793229997Sken break; 1794229997Sken } 1795229997Sken } 1796229997Sken PICKUP_GIANT(); 1797229997Sken 1798229997Sken return (0); 1799229997Sken} 1800229997Sken 1801229997Skenstatic int 1802229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1803229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1804229997Sken{ 1805229997Sken struct nameidata nd; 1806229997Sken int flags; 1807229997Sken int error; 1808229997Sken 1809229997Sken /* 1810229997Sken * XXX KDM allow a read-only option? 1811229997Sken */ 1812229997Sken flags = FREAD | FWRITE; 1813229997Sken error = 0; 1814229997Sken 1815229997Sken if (rootvnode == NULL) { 1816229997Sken snprintf(req->error_str, sizeof(req->error_str), 1817229997Sken "%s: Root filesystem is not mounted", __func__); 1818229997Sken return (1); 1819229997Sken } 1820229997Sken 1821229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1822229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1823229997Sken VREF(rootvnode); 1824229997Sken } 1825229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1826229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1827229997Sken VREF(rootvnode); 1828229997Sken } 1829229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1830229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1831229997Sken VREF(rootvnode); 1832229997Sken } 1833229997Sken 1834229997Sken again: 1835229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1836229997Sken error = vn_open(&nd, &flags, 0, NULL); 1837229997Sken if (error) { 1838229997Sken /* 1839229997Sken * This is the only reasonable guess we can make as far as 1840229997Sken * path if the user doesn't give us a fully qualified path. 1841229997Sken * If they want to specify a file, they need to specify the 1842229997Sken * full path. 1843229997Sken */ 1844229997Sken if (be_lun->dev_path[0] != '/') { 1845229997Sken char *dev_path = "/dev/"; 1846229997Sken char *dev_name; 1847229997Sken 1848229997Sken /* Try adding device path at beginning of name */ 1849229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1850229997Sken + strlen(dev_path) + 1, 1851229997Sken M_CTLBLK, M_WAITOK); 1852229997Sken if (dev_name) { 1853229997Sken sprintf(dev_name, "%s%s", dev_path, 1854229997Sken be_lun->dev_path); 1855229997Sken free(be_lun->dev_path, M_CTLBLK); 1856229997Sken be_lun->dev_path = dev_name; 1857229997Sken goto again; 1858229997Sken } 1859229997Sken } 1860229997Sken snprintf(req->error_str, sizeof(req->error_str), 1861229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1862229997Sken return (error); 1863229997Sken } 1864229997Sken 1865229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1866229997Sken 1867229997Sken be_lun->vn = nd.ni_vp; 1868229997Sken 1869229997Sken /* We only support disks and files. */ 1870229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1871229997Sken error = ctl_be_block_open_dev(be_lun, req); 1872229997Sken } else if (be_lun->vn->v_type == VREG) { 1873229997Sken error = ctl_be_block_open_file(be_lun, req); 1874229997Sken } else { 1875229997Sken error = EINVAL; 1876229997Sken snprintf(req->error_str, sizeof(req->error_str), 1877258871Strasz "%s is not a disk or plain file", be_lun->dev_path); 1878229997Sken } 1879229997Sken VOP_UNLOCK(be_lun->vn, 0); 1880229997Sken 1881229997Sken if (error != 0) { 1882229997Sken ctl_be_block_close(be_lun); 1883229997Sken return (error); 1884229997Sken } 1885229997Sken 1886229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1887229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1888229997Sken 1889229997Sken return (0); 1890229997Sken} 1891229997Sken 1892229997Skenstatic int 1893229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1894229997Sken{ 1895229997Sken struct ctl_be_block_lun *be_lun; 1896229997Sken struct ctl_lun_create_params *params; 1897267481Smav char num_thread_str[16]; 1898229997Sken char tmpstr[32]; 1899267481Smav char *value; 1900264274Smav int retval, num_threads, unmap; 1901267481Smav int tmp_num_threads; 1902229997Sken 1903229997Sken params = &req->reqdata.create; 1904229997Sken retval = 0; 1905229997Sken 1906229997Sken num_threads = cbb_num_threads; 1907229997Sken 1908229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1909229997Sken 1910229997Sken be_lun->softc = softc; 1911229997Sken STAILQ_INIT(&be_lun->input_queue); 1912229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1913229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1914229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1915267877Smav mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1916267877Smav mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1917268280Smav ctl_init_opts(&be_lun->ctl_be_lun.options, 1918268280Smav req->num_be_args, req->kern_be_args); 1919229997Sken 1920264886Smav be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1921256995Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1922229997Sken 1923229997Sken if (be_lun->lun_zone == NULL) { 1924229997Sken snprintf(req->error_str, sizeof(req->error_str), 1925229997Sken "%s: error allocating UMA zone", __func__); 1926229997Sken goto bailout_error; 1927229997Sken } 1928229997Sken 1929229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1930229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1931229997Sken else 1932229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1933229997Sken 1934229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1935268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1936267499Smav if (value == NULL) { 1937229997Sken snprintf(req->error_str, sizeof(req->error_str), 1938229997Sken "%s: no file argument specified", __func__); 1939229997Sken goto bailout_error; 1940229997Sken } 1941267499Smav be_lun->dev_path = strdup(value, M_CTLBLK); 1942229997Sken 1943229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1944229997Sken if (retval != 0) { 1945229997Sken retval = 0; 1946229997Sken goto bailout_error; 1947229997Sken } 1948229997Sken 1949229997Sken /* 1950229997Sken * Tell the user the size of the file/device. 1951229997Sken */ 1952229997Sken params->lun_size_bytes = be_lun->size_bytes; 1953229997Sken 1954229997Sken /* 1955229997Sken * The maximum LBA is the size - 1. 1956229997Sken */ 1957229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1958229997Sken } else { 1959229997Sken /* 1960229997Sken * For processor devices, we don't have any size. 1961229997Sken */ 1962229997Sken be_lun->blocksize = 0; 1963264191Smav be_lun->pblockexp = 0; 1964264191Smav be_lun->pblockoff = 0; 1965229997Sken be_lun->size_blocks = 0; 1966229997Sken be_lun->size_bytes = 0; 1967229997Sken be_lun->ctl_be_lun.maxlba = 0; 1968229997Sken params->lun_size_bytes = 0; 1969229997Sken 1970229997Sken /* 1971229997Sken * Default to just 1 thread for processor devices. 1972229997Sken */ 1973229997Sken num_threads = 1; 1974229997Sken } 1975229997Sken 1976229997Sken /* 1977229997Sken * XXX This searching loop might be refactored to be combined with 1978229997Sken * the loop above, 1979229997Sken */ 1980268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 1981267481Smav if (value != NULL) { 1982267481Smav tmp_num_threads = strtol(value, NULL, 0); 1983229997Sken 1984267481Smav /* 1985267481Smav * We don't let the user specify less than one 1986267481Smav * thread, but hope he's clueful enough not to 1987267481Smav * specify 1000 threads. 1988267481Smav */ 1989267481Smav if (tmp_num_threads < 1) { 1990267481Smav snprintf(req->error_str, sizeof(req->error_str), 1991267481Smav "%s: invalid number of threads %s", 1992267481Smav __func__, num_thread_str); 1993267481Smav goto bailout_error; 1994229997Sken } 1995267481Smav num_threads = tmp_num_threads; 1996229997Sken } 1997267481Smav unmap = 0; 1998268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 1999267481Smav if (value != NULL && strcmp(value, "on") == 0) 2000267481Smav unmap = 1; 2001229997Sken 2002229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2003229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2004264274Smav if (unmap) 2005264274Smav be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2006229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 2007229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2008264191Smav be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2009264191Smav be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2010229997Sken /* Tell the user the blocksize we ended up using */ 2011229997Sken params->blocksize_bytes = be_lun->blocksize; 2012229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2013229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2014229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2015229997Sken } else 2016229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 2017229997Sken 2018229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2019229997Sken be_lun->ctl_be_lun.lun_config_status = 2020229997Sken ctl_be_block_lun_config_status; 2021229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2022229997Sken 2023229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2024229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2025229997Sken softc->num_luns); 2026229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2027229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2028229997Sken sizeof(tmpstr))); 2029229997Sken 2030229997Sken /* Tell the user what we used for a serial number */ 2031229997Sken strncpy((char *)params->serial_num, tmpstr, 2032229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 2033229997Sken } else { 2034229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 2035229997Sken params->serial_num, 2036229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2037229997Sken sizeof(params->serial_num))); 2038229997Sken } 2039229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2040229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2041229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2042229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2043229997Sken sizeof(tmpstr))); 2044229997Sken 2045229997Sken /* Tell the user what we used for a device ID */ 2046229997Sken strncpy((char *)params->device_id, tmpstr, 2047229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 2048229997Sken } else { 2049229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 2050229997Sken params->device_id, 2051229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2052229997Sken sizeof(params->device_id))); 2053229997Sken } 2054229997Sken 2055229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2056229997Sken 2057229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2058229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2059229997Sken 2060229997Sken if (be_lun->io_taskqueue == NULL) { 2061229997Sken snprintf(req->error_str, sizeof(req->error_str), 2062229997Sken "%s: Unable to create taskqueue", __func__); 2063229997Sken goto bailout_error; 2064229997Sken } 2065229997Sken 2066229997Sken /* 2067229997Sken * Note that we start the same number of threads by default for 2068229997Sken * both the file case and the block device case. For the file 2069229997Sken * case, we need multiple threads to allow concurrency, because the 2070229997Sken * vnode interface is designed to be a blocking interface. For the 2071229997Sken * block device case, ZFS zvols at least will block the caller's 2072229997Sken * context in many instances, and so we need multiple threads to 2073229997Sken * overcome that problem. Other block devices don't need as many 2074229997Sken * threads, but they shouldn't cause too many problems. 2075229997Sken * 2076229997Sken * If the user wants to just have a single thread for a block 2077229997Sken * device, he can specify that when the LUN is created, or change 2078229997Sken * the tunable/sysctl to alter the default number of threads. 2079229997Sken */ 2080229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2081229997Sken /*num threads*/num_threads, 2082229997Sken /*priority*/PWAIT, 2083229997Sken /*thread name*/ 2084229997Sken "%s taskq", be_lun->lunname); 2085229997Sken 2086229997Sken if (retval != 0) 2087229997Sken goto bailout_error; 2088229997Sken 2089229997Sken be_lun->num_threads = num_threads; 2090229997Sken 2091229997Sken mtx_lock(&softc->lock); 2092229997Sken softc->num_luns++; 2093229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2094229997Sken 2095229997Sken mtx_unlock(&softc->lock); 2096229997Sken 2097229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 2098229997Sken if (retval != 0) { 2099229997Sken mtx_lock(&softc->lock); 2100229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2101229997Sken links); 2102229997Sken softc->num_luns--; 2103229997Sken mtx_unlock(&softc->lock); 2104229997Sken snprintf(req->error_str, sizeof(req->error_str), 2105229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 2106229997Sken "details", __func__, retval); 2107229997Sken retval = 0; 2108229997Sken goto bailout_error; 2109229997Sken } 2110229997Sken 2111229997Sken mtx_lock(&softc->lock); 2112229997Sken 2113229997Sken /* 2114229997Sken * Tell the config_status routine that we're waiting so it won't 2115229997Sken * clean up the LUN in the event of an error. 2116229997Sken */ 2117229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2118229997Sken 2119229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2120229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2121229997Sken if (retval == EINTR) 2122229997Sken break; 2123229997Sken } 2124229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2125229997Sken 2126229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2127229997Sken snprintf(req->error_str, sizeof(req->error_str), 2128229997Sken "%s: LUN configuration error, see dmesg for details", 2129229997Sken __func__); 2130229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2131229997Sken links); 2132229997Sken softc->num_luns--; 2133229997Sken mtx_unlock(&softc->lock); 2134229997Sken goto bailout_error; 2135229997Sken } else { 2136229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2137229997Sken } 2138229997Sken 2139229997Sken mtx_unlock(&softc->lock); 2140229997Sken 2141229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2142229997Sken be_lun->blocksize, 2143229997Sken DEVSTAT_ALL_SUPPORTED, 2144229997Sken be_lun->ctl_be_lun.lun_type 2145229997Sken | DEVSTAT_TYPE_IF_OTHER, 2146229997Sken DEVSTAT_PRIORITY_OTHER); 2147229997Sken 2148229997Sken 2149229997Sken req->status = CTL_LUN_OK; 2150229997Sken 2151229997Sken return (retval); 2152229997Sken 2153229997Skenbailout_error: 2154229997Sken req->status = CTL_LUN_ERROR; 2155229997Sken 2156267429Smav if (be_lun->io_taskqueue != NULL) 2157267429Smav taskqueue_free(be_lun->io_taskqueue); 2158229997Sken ctl_be_block_close(be_lun); 2159267429Smav if (be_lun->dev_path != NULL) 2160267429Smav free(be_lun->dev_path, M_CTLBLK); 2161267429Smav if (be_lun->lun_zone != NULL) 2162267429Smav uma_zdestroy(be_lun->lun_zone); 2163268280Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2164267877Smav mtx_destroy(&be_lun->queue_lock); 2165267877Smav mtx_destroy(&be_lun->io_lock); 2166229997Sken free(be_lun, M_CTLBLK); 2167229997Sken 2168229997Sken return (retval); 2169229997Sken} 2170229997Sken 2171229997Skenstatic int 2172229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2173229997Sken{ 2174229997Sken struct ctl_lun_rm_params *params; 2175229997Sken struct ctl_be_block_lun *be_lun; 2176229997Sken int retval; 2177229997Sken 2178229997Sken params = &req->reqdata.rm; 2179229997Sken 2180229997Sken mtx_lock(&softc->lock); 2181229997Sken 2182229997Sken be_lun = NULL; 2183229997Sken 2184229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2185229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2186229997Sken break; 2187229997Sken } 2188229997Sken mtx_unlock(&softc->lock); 2189229997Sken 2190229997Sken if (be_lun == NULL) { 2191229997Sken snprintf(req->error_str, sizeof(req->error_str), 2192229997Sken "%s: LUN %u is not managed by the block backend", 2193229997Sken __func__, params->lun_id); 2194229997Sken goto bailout_error; 2195229997Sken } 2196229997Sken 2197229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2198229997Sken 2199229997Sken if (retval != 0) { 2200229997Sken snprintf(req->error_str, sizeof(req->error_str), 2201229997Sken "%s: error %d returned from ctl_disable_lun() for " 2202229997Sken "LUN %d", __func__, retval, params->lun_id); 2203229997Sken goto bailout_error; 2204229997Sken 2205229997Sken } 2206229997Sken 2207229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2208229997Sken if (retval != 0) { 2209229997Sken snprintf(req->error_str, sizeof(req->error_str), 2210229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 2211229997Sken "LUN %d", __func__, retval, params->lun_id); 2212229997Sken goto bailout_error; 2213229997Sken } 2214229997Sken 2215229997Sken mtx_lock(&softc->lock); 2216229997Sken 2217229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2218229997Sken 2219229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2220229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2221229997Sken if (retval == EINTR) 2222229997Sken break; 2223229997Sken } 2224229997Sken 2225229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2226229997Sken 2227229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2228229997Sken snprintf(req->error_str, sizeof(req->error_str), 2229229997Sken "%s: interrupted waiting for LUN to be freed", 2230229997Sken __func__); 2231229997Sken mtx_unlock(&softc->lock); 2232229997Sken goto bailout_error; 2233229997Sken } 2234229997Sken 2235229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2236229997Sken 2237229997Sken softc->num_luns--; 2238229997Sken mtx_unlock(&softc->lock); 2239229997Sken 2240229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2241229997Sken 2242229997Sken taskqueue_free(be_lun->io_taskqueue); 2243229997Sken 2244229997Sken ctl_be_block_close(be_lun); 2245229997Sken 2246229997Sken if (be_lun->disk_stats != NULL) 2247229997Sken devstat_remove_entry(be_lun->disk_stats); 2248229997Sken 2249229997Sken uma_zdestroy(be_lun->lun_zone); 2250229997Sken 2251268280Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2252229997Sken free(be_lun->dev_path, M_CTLBLK); 2253267877Smav mtx_destroy(&be_lun->queue_lock); 2254267877Smav mtx_destroy(&be_lun->io_lock); 2255229997Sken free(be_lun, M_CTLBLK); 2256229997Sken 2257229997Sken req->status = CTL_LUN_OK; 2258229997Sken 2259229997Sken return (0); 2260229997Sken 2261229997Skenbailout_error: 2262229997Sken 2263229997Sken req->status = CTL_LUN_ERROR; 2264229997Sken 2265229997Sken return (0); 2266229997Sken} 2267229997Sken 2268232604Straszstatic int 2269232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2270232604Strasz struct ctl_lun_req *req) 2271232604Strasz{ 2272232604Strasz struct vattr vattr; 2273232604Strasz int error; 2274232604Strasz struct ctl_lun_modify_params *params; 2275232604Strasz 2276232604Strasz params = &req->reqdata.modify; 2277232604Strasz 2278232604Strasz if (params->lun_size_bytes != 0) { 2279232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2280232604Strasz } else { 2281271794Smav vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2282232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2283271794Smav VOP_UNLOCK(be_lun->vn, 0); 2284232604Strasz if (error != 0) { 2285232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2286232604Strasz "error calling VOP_GETATTR() for file %s", 2287232604Strasz be_lun->dev_path); 2288232604Strasz return (error); 2289232604Strasz } 2290232604Strasz 2291232604Strasz be_lun->size_bytes = vattr.va_size; 2292232604Strasz } 2293232604Strasz 2294232604Strasz return (0); 2295232604Strasz} 2296232604Strasz 2297232604Straszstatic int 2298232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2299232604Strasz struct ctl_lun_req *req) 2300232604Strasz{ 2301271794Smav struct ctl_be_block_devdata *dev_data; 2302232604Strasz int error; 2303232604Strasz struct ctl_lun_modify_params *params; 2304232604Strasz uint64_t size_bytes; 2305232604Strasz 2306232604Strasz params = &req->reqdata.modify; 2307232604Strasz 2308271794Smav dev_data = &be_lun->backend.dev; 2309271794Smav if (!dev_data->csw->d_ioctl) { 2310232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2311232604Strasz "%s: no d_ioctl for device %s!", __func__, 2312232604Strasz be_lun->dev_path); 2313232604Strasz return (ENODEV); 2314232604Strasz } 2315232604Strasz 2316271794Smav error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2317232604Strasz (caddr_t)&size_bytes, FREAD, 2318232604Strasz curthread); 2319232604Strasz if (error) { 2320232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2321232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2322232604Strasz "on %s!", __func__, error, be_lun->dev_path); 2323232604Strasz return (error); 2324232604Strasz } 2325232604Strasz 2326232604Strasz if (params->lun_size_bytes != 0) { 2327232604Strasz if (params->lun_size_bytes > size_bytes) { 2328232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2329232604Strasz "%s: requested LUN size %ju > backing device " 2330232604Strasz "size %ju", __func__, 2331232604Strasz (uintmax_t)params->lun_size_bytes, 2332232604Strasz (uintmax_t)size_bytes); 2333232604Strasz return (EINVAL); 2334232604Strasz } 2335232604Strasz 2336232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2337232604Strasz } else { 2338232604Strasz be_lun->size_bytes = size_bytes; 2339232604Strasz } 2340232604Strasz 2341232604Strasz return (0); 2342232604Strasz} 2343232604Strasz 2344232604Straszstatic int 2345232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2346232604Strasz{ 2347232604Strasz struct ctl_lun_modify_params *params; 2348232604Strasz struct ctl_be_block_lun *be_lun; 2349271794Smav uint64_t oldsize; 2350241896Skib int error; 2351232604Strasz 2352232604Strasz params = &req->reqdata.modify; 2353232604Strasz 2354232604Strasz mtx_lock(&softc->lock); 2355232604Strasz 2356232604Strasz be_lun = NULL; 2357232604Strasz 2358232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2359232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2360232604Strasz break; 2361232604Strasz } 2362232604Strasz mtx_unlock(&softc->lock); 2363232604Strasz 2364232604Strasz if (be_lun == NULL) { 2365232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2366232604Strasz "%s: LUN %u is not managed by the block backend", 2367232604Strasz __func__, params->lun_id); 2368232604Strasz goto bailout_error; 2369232604Strasz } 2370232604Strasz 2371232604Strasz if (params->lun_size_bytes != 0) { 2372232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2373232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2374232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2375232604Strasz params->lun_size_bytes, be_lun->blocksize); 2376232604Strasz goto bailout_error; 2377232604Strasz } 2378232604Strasz } 2379232604Strasz 2380271794Smav oldsize = be_lun->size_bytes; 2381232604Strasz if (be_lun->vn->v_type == VREG) 2382232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2383232604Strasz else 2384232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2385232604Strasz if (error != 0) 2386232604Strasz goto bailout_error; 2387232604Strasz 2388271794Smav if (be_lun->size_bytes != oldsize) { 2389271794Smav be_lun->size_blocks = be_lun->size_bytes >> 2390271794Smav be_lun->blocksize_shift; 2391232604Strasz 2392271794Smav /* 2393271794Smav * The maximum LBA is the size - 1. 2394271794Smav * 2395271794Smav * XXX: Note that this field is being updated without locking, 2396271794Smav * which might cause problems on 32-bit architectures. 2397271794Smav */ 2398271794Smav be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2399271794Smav ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2400271794Smav } 2401232604Strasz 2402232604Strasz /* Tell the user the exact size we ended up using */ 2403232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2404232604Strasz 2405232604Strasz req->status = CTL_LUN_OK; 2406232604Strasz 2407232604Strasz return (0); 2408232604Strasz 2409232604Straszbailout_error: 2410232604Strasz req->status = CTL_LUN_ERROR; 2411232604Strasz 2412232604Strasz return (0); 2413232604Strasz} 2414232604Strasz 2415229997Skenstatic void 2416229997Skenctl_be_block_lun_shutdown(void *be_lun) 2417229997Sken{ 2418229997Sken struct ctl_be_block_lun *lun; 2419229997Sken struct ctl_be_block_softc *softc; 2420229997Sken 2421229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2422229997Sken 2423229997Sken softc = lun->softc; 2424229997Sken 2425229997Sken mtx_lock(&softc->lock); 2426229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2427229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2428229997Sken wakeup(lun); 2429229997Sken mtx_unlock(&softc->lock); 2430229997Sken 2431229997Sken} 2432229997Sken 2433229997Skenstatic void 2434229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2435229997Sken{ 2436229997Sken struct ctl_be_block_lun *lun; 2437229997Sken struct ctl_be_block_softc *softc; 2438229997Sken 2439229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2440229997Sken softc = lun->softc; 2441229997Sken 2442229997Sken if (status == CTL_LUN_CONFIG_OK) { 2443229997Sken mtx_lock(&softc->lock); 2444229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2445229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2446229997Sken wakeup(lun); 2447229997Sken mtx_unlock(&softc->lock); 2448229997Sken 2449229997Sken /* 2450229997Sken * We successfully added the LUN, attempt to enable it. 2451229997Sken */ 2452229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2453229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2454229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2455229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2456229997Sken __func__); 2457229997Sken } 2458229997Sken } 2459229997Sken 2460229997Sken return; 2461229997Sken } 2462229997Sken 2463229997Sken 2464229997Sken mtx_lock(&softc->lock); 2465229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2466229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2467229997Sken wakeup(lun); 2468229997Sken mtx_unlock(&softc->lock); 2469229997Sken} 2470229997Sken 2471229997Sken 2472229997Skenstatic int 2473229997Skenctl_be_block_config_write(union ctl_io *io) 2474229997Sken{ 2475229997Sken struct ctl_be_block_lun *be_lun; 2476229997Sken struct ctl_be_lun *ctl_be_lun; 2477229997Sken int retval; 2478229997Sken 2479229997Sken retval = 0; 2480229997Sken 2481229997Sken DPRINTF("entered\n"); 2482229997Sken 2483229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2484229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2485229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2486229997Sken 2487229997Sken switch (io->scsiio.cdb[0]) { 2488229997Sken case SYNCHRONIZE_CACHE: 2489229997Sken case SYNCHRONIZE_CACHE_16: 2490264274Smav case WRITE_SAME_10: 2491264274Smav case WRITE_SAME_16: 2492264274Smav case UNMAP: 2493229997Sken /* 2494229997Sken * The upper level CTL code will filter out any CDBs with 2495229997Sken * the immediate bit set and return the proper error. 2496229997Sken * 2497229997Sken * We don't really need to worry about what LBA range the 2498229997Sken * user asked to be synced out. When they issue a sync 2499229997Sken * cache command, we'll sync out the whole thing. 2500229997Sken */ 2501267877Smav mtx_lock(&be_lun->queue_lock); 2502229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2503229997Sken links); 2504267877Smav mtx_unlock(&be_lun->queue_lock); 2505229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2506229997Sken break; 2507229997Sken case START_STOP_UNIT: { 2508229997Sken struct scsi_start_stop_unit *cdb; 2509229997Sken 2510229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2511229997Sken 2512229997Sken if (cdb->how & SSS_START) 2513229997Sken retval = ctl_start_lun(ctl_be_lun); 2514229997Sken else { 2515229997Sken retval = ctl_stop_lun(ctl_be_lun); 2516229997Sken /* 2517229997Sken * XXX KDM Copan-specific offline behavior. 2518229997Sken * Figure out a reasonable way to port this? 2519229997Sken */ 2520229997Sken#ifdef NEEDTOPORT 2521229997Sken if ((retval == 0) 2522229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2523229997Sken retval = ctl_lun_offline(ctl_be_lun); 2524229997Sken#endif 2525229997Sken } 2526229997Sken 2527229997Sken /* 2528229997Sken * In general, the above routines should not fail. They 2529229997Sken * just set state for the LUN. So we've got something 2530229997Sken * pretty wrong here if we can't start or stop the LUN. 2531229997Sken */ 2532229997Sken if (retval != 0) { 2533229997Sken ctl_set_internal_failure(&io->scsiio, 2534229997Sken /*sks_valid*/ 1, 2535229997Sken /*retry_count*/ 0xf051); 2536229997Sken retval = CTL_RETVAL_COMPLETE; 2537229997Sken } else { 2538229997Sken ctl_set_success(&io->scsiio); 2539229997Sken } 2540229997Sken ctl_config_write_done(io); 2541229997Sken break; 2542229997Sken } 2543229997Sken default: 2544229997Sken ctl_set_invalid_opcode(&io->scsiio); 2545229997Sken ctl_config_write_done(io); 2546229997Sken retval = CTL_RETVAL_COMPLETE; 2547229997Sken break; 2548229997Sken } 2549229997Sken 2550229997Sken return (retval); 2551229997Sken 2552229997Sken} 2553229997Sken 2554229997Skenstatic int 2555229997Skenctl_be_block_config_read(union ctl_io *io) 2556229997Sken{ 2557229997Sken return (0); 2558229997Sken} 2559229997Sken 2560229997Skenstatic int 2561229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2562229997Sken{ 2563229997Sken struct ctl_be_block_lun *lun; 2564229997Sken int retval; 2565229997Sken 2566229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2567229997Sken retval = 0; 2568229997Sken 2569268283Smav retval = sbuf_printf(sb, "\t<num_threads>"); 2570229997Sken 2571229997Sken if (retval != 0) 2572229997Sken goto bailout; 2573229997Sken 2574229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2575229997Sken 2576229997Sken if (retval != 0) 2577229997Sken goto bailout; 2578229997Sken 2579268283Smav retval = sbuf_printf(sb, "</num_threads>\n"); 2580229997Sken 2581229997Skenbailout: 2582229997Sken 2583229997Sken return (retval); 2584229997Sken} 2585229997Sken 2586229997Skenint 2587229997Skenctl_be_block_init(void) 2588229997Sken{ 2589229997Sken struct ctl_be_block_softc *softc; 2590229997Sken int retval; 2591229997Sken 2592229997Sken softc = &backend_block_softc; 2593229997Sken retval = 0; 2594229997Sken 2595267877Smav mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2596264020Strasz beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2597264020Strasz NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2598229997Sken STAILQ_INIT(&softc->disk_list); 2599229997Sken STAILQ_INIT(&softc->lun_list); 2600229997Sken 2601229997Sken return (retval); 2602229997Sken} 2603