ctl_backend_block.c revision 261538
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 261538 2014-02-06 03:54:58Z 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> 52229997Sken#include <sys/lock.h> 53229997Sken#include <sys/mutex.h> 54229997Sken#include <sys/condvar.h> 55229997Sken#include <sys/malloc.h> 56229997Sken#include <sys/conf.h> 57229997Sken#include <sys/ioccom.h> 58229997Sken#include <sys/queue.h> 59229997Sken#include <sys/sbuf.h> 60229997Sken#include <sys/endian.h> 61229997Sken#include <sys/uio.h> 62229997Sken#include <sys/buf.h> 63229997Sken#include <sys/taskqueue.h> 64229997Sken#include <sys/vnode.h> 65229997Sken#include <sys/namei.h> 66229997Sken#include <sys/mount.h> 67229997Sken#include <sys/disk.h> 68229997Sken#include <sys/fcntl.h> 69229997Sken#include <sys/filedesc.h> 70229997Sken#include <sys/proc.h> 71229997Sken#include <sys/pcpu.h> 72229997Sken#include <sys/module.h> 73229997Sken#include <sys/sdt.h> 74229997Sken#include <sys/devicestat.h> 75229997Sken#include <sys/sysctl.h> 76229997Sken 77229997Sken#include <geom/geom.h> 78229997Sken 79229997Sken#include <cam/cam.h> 80229997Sken#include <cam/scsi/scsi_all.h> 81229997Sken#include <cam/scsi/scsi_da.h> 82229997Sken#include <cam/ctl/ctl_io.h> 83229997Sken#include <cam/ctl/ctl.h> 84229997Sken#include <cam/ctl/ctl_backend.h> 85229997Sken#include <cam/ctl/ctl_frontend_internal.h> 86229997Sken#include <cam/ctl/ctl_ioctl.h> 87229997Sken#include <cam/ctl/ctl_scsi_all.h> 88229997Sken#include <cam/ctl/ctl_error.h> 89229997Sken 90229997Sken/* 91229997Sken * The idea here is that we'll allocate enough S/G space to hold a 16MB 92229997Sken * I/O. If we get an I/O larger than that, we'll reject it. 93229997Sken */ 94229997Sken#define CTLBLK_MAX_IO_SIZE (16 * 1024 * 1024) 95229997Sken#define CTLBLK_MAX_SEGS (CTLBLK_MAX_IO_SIZE / MAXPHYS) + 1 96229997Sken 97229997Sken#ifdef CTLBLK_DEBUG 98229997Sken#define DPRINTF(fmt, args...) \ 99229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 100229997Sken#else 101229997Sken#define DPRINTF(fmt, args...) do {} while(0) 102229997Sken#endif 103229997Sken 104229997SkenSDT_PROVIDER_DEFINE(cbb); 105229997Sken 106229997Skentypedef enum { 107229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 108229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 109229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 110229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 111229997Sken} ctl_be_block_lun_flags; 112229997Sken 113229997Skentypedef enum { 114229997Sken CTL_BE_BLOCK_NONE, 115229997Sken CTL_BE_BLOCK_DEV, 116229997Sken CTL_BE_BLOCK_FILE 117229997Sken} ctl_be_block_type; 118229997Sken 119229997Skenstruct ctl_be_block_devdata { 120229997Sken struct cdev *cdev; 121229997Sken struct cdevsw *csw; 122229997Sken int dev_ref; 123229997Sken}; 124229997Sken 125229997Skenstruct ctl_be_block_filedata { 126229997Sken struct ucred *cred; 127229997Sken}; 128229997Sken 129229997Skenunion ctl_be_block_bedata { 130229997Sken struct ctl_be_block_devdata dev; 131229997Sken struct ctl_be_block_filedata file; 132229997Sken}; 133229997Sken 134229997Skenstruct ctl_be_block_io; 135229997Skenstruct ctl_be_block_lun; 136229997Sken 137229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 138229997Sken struct ctl_be_block_io *beio); 139229997Sken 140229997Sken/* 141229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 142229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 143229997Sken */ 144229997Skenstruct ctl_be_block_lun { 145229997Sken struct ctl_block_disk *disk; 146229997Sken char lunname[32]; 147229997Sken char *dev_path; 148229997Sken ctl_be_block_type dev_type; 149229997Sken struct vnode *vn; 150229997Sken union ctl_be_block_bedata backend; 151229997Sken cbb_dispatch_t dispatch; 152229997Sken cbb_dispatch_t lun_flush; 153229997Sken struct mtx lock; 154229997Sken uma_zone_t lun_zone; 155229997Sken uint64_t size_blocks; 156229997Sken uint64_t size_bytes; 157229997Sken uint32_t blocksize; 158229997Sken int blocksize_shift; 159229997Sken struct ctl_be_block_softc *softc; 160229997Sken struct devstat *disk_stats; 161229997Sken ctl_be_block_lun_flags flags; 162229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 163229997Sken struct ctl_be_lun ctl_be_lun; 164229997Sken struct taskqueue *io_taskqueue; 165229997Sken struct task io_task; 166229997Sken int num_threads; 167229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 168229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 169229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 170229997Sken}; 171229997Sken 172229997Sken/* 173229997Sken * Overall softc structure for the block backend module. 174229997Sken */ 175229997Skenstruct ctl_be_block_softc { 176229997Sken STAILQ_HEAD(, ctl_be_block_io) beio_free_queue; 177229997Sken struct mtx lock; 178229997Sken int prealloc_beio; 179229997Sken int num_disks; 180229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 181229997Sken int num_luns; 182229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 183229997Sken}; 184229997Sken 185229997Skenstatic struct ctl_be_block_softc backend_block_softc; 186229997Sken 187229997Sken/* 188229997Sken * Per-I/O information. 189229997Sken */ 190229997Skenstruct ctl_be_block_io { 191229997Sken union ctl_io *io; 192229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 193229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 194229997Sken int bio_cmd; 195229997Sken int bio_flags; 196229997Sken int num_segs; 197229997Sken int num_bios_sent; 198229997Sken int num_bios_done; 199229997Sken int send_complete; 200229997Sken int num_errors; 201229997Sken struct bintime ds_t0; 202229997Sken devstat_tag_type ds_tag_type; 203229997Sken devstat_trans_flags ds_trans_type; 204229997Sken uint64_t io_len; 205229997Sken uint64_t io_offset; 206229997Sken struct ctl_be_block_softc *softc; 207229997Sken struct ctl_be_block_lun *lun; 208229997Sken STAILQ_ENTRY(ctl_be_block_io) links; 209229997Sken}; 210229997Sken 211229997Skenstatic int cbb_num_threads = 14; 212229997SkenTUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 213229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 214229997Sken "CAM Target Layer Block Backend"); 215229997SkenSYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 216229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 217229997Sken 218229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 219229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 220229997Skenstatic int ctl_grow_beio(struct ctl_be_block_softc *softc, int count); 221229997Sken#if 0 222229997Skenstatic void ctl_shrink_beio(struct ctl_be_block_softc *softc); 223229997Sken#endif 224229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 225229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 226229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 227229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 228229997Sken struct ctl_be_block_io *beio); 229229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 230229997Sken struct ctl_be_block_io *beio); 231229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 232229997Sken struct ctl_be_block_io *beio); 233229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 234229997Sken struct ctl_be_block_io *beio); 235229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 236229997Sken union ctl_io *io); 237229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 238229997Sken union ctl_io *io); 239229997Skenstatic void ctl_be_block_worker(void *context, int pending); 240229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 241229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 242229997Sken int flag, struct thread *td); 243229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 244229997Sken struct ctl_lun_req *req); 245229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 246229997Sken struct ctl_lun_req *req); 247229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 248229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 249229997Sken struct ctl_be_block_lun *be_lun, 250229997Sken struct ctl_lun_req *req); 251229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 252229997Sken struct ctl_lun_req *req); 253229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 254229997Sken struct ctl_lun_req *req); 255232604Straszstatic int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 256232604Strasz struct ctl_lun_req *req); 257232604Straszstatic int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 258232604Strasz struct ctl_lun_req *req); 259232604Straszstatic int ctl_be_block_modify(struct ctl_be_block_softc *softc, 260232604Strasz struct ctl_lun_req *req); 261229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 262229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 263229997Sken ctl_lun_config_status status); 264229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 265229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 266229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 267229997Skenint ctl_be_block_init(void); 268229997Sken 269229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 270229997Sken{ 271230334Sken .name = "block", 272230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 273230334Sken .init = ctl_be_block_init, 274230334Sken .data_submit = ctl_be_block_submit, 275230334Sken .data_move_done = ctl_be_block_move_done, 276230334Sken .config_read = ctl_be_block_config_read, 277230334Sken .config_write = ctl_be_block_config_write, 278230334Sken .ioctl = ctl_be_block_ioctl, 279230334Sken .lun_info = ctl_be_block_lun_info 280229997Sken}; 281229997Sken 282229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 283229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 284229997Sken 285229997Skenstatic struct ctl_be_block_io * 286229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 287229997Sken{ 288229997Sken struct ctl_be_block_io *beio; 289229997Sken int count; 290229997Sken 291229997Sken mtx_lock(&softc->lock); 292229997Sken 293229997Sken beio = STAILQ_FIRST(&softc->beio_free_queue); 294229997Sken if (beio != NULL) { 295229997Sken STAILQ_REMOVE(&softc->beio_free_queue, beio, 296229997Sken ctl_be_block_io, links); 297229997Sken } 298229997Sken mtx_unlock(&softc->lock); 299229997Sken 300229997Sken if (beio != NULL) { 301229997Sken bzero(beio, sizeof(*beio)); 302229997Sken beio->softc = softc; 303229997Sken return (beio); 304229997Sken } 305229997Sken 306229997Sken for (;;) { 307229997Sken 308229997Sken count = ctl_grow_beio(softc, /*count*/ 10); 309229997Sken 310229997Sken /* 311229997Sken * This shouldn't be possible, since ctl_grow_beio() uses a 312229997Sken * blocking malloc. 313229997Sken */ 314229997Sken if (count == 0) 315229997Sken return (NULL); 316229997Sken 317229997Sken /* 318229997Sken * Since we have to drop the lock when we're allocating beio 319229997Sken * structures, it's possible someone else can come along and 320229997Sken * allocate the beio's we've just allocated. 321229997Sken */ 322229997Sken mtx_lock(&softc->lock); 323229997Sken beio = STAILQ_FIRST(&softc->beio_free_queue); 324229997Sken if (beio != NULL) { 325229997Sken STAILQ_REMOVE(&softc->beio_free_queue, beio, 326229997Sken ctl_be_block_io, links); 327229997Sken } 328229997Sken mtx_unlock(&softc->lock); 329229997Sken 330229997Sken if (beio != NULL) { 331229997Sken bzero(beio, sizeof(*beio)); 332229997Sken beio->softc = softc; 333229997Sken break; 334229997Sken } 335229997Sken } 336229997Sken return (beio); 337229997Sken} 338229997Sken 339229997Skenstatic void 340229997Skenctl_free_beio(struct ctl_be_block_io *beio) 341229997Sken{ 342229997Sken struct ctl_be_block_softc *softc; 343229997Sken int duplicate_free; 344229997Sken int i; 345229997Sken 346229997Sken softc = beio->softc; 347229997Sken duplicate_free = 0; 348229997Sken 349229997Sken for (i = 0; i < beio->num_segs; i++) { 350229997Sken if (beio->sg_segs[i].addr == NULL) 351229997Sken duplicate_free++; 352229997Sken 353229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 354229997Sken beio->sg_segs[i].addr = NULL; 355229997Sken } 356229997Sken 357229997Sken if (duplicate_free > 0) { 358229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 359229997Sken duplicate_free, beio->num_segs); 360229997Sken } 361229997Sken mtx_lock(&softc->lock); 362229997Sken STAILQ_INSERT_TAIL(&softc->beio_free_queue, beio, links); 363229997Sken mtx_unlock(&softc->lock); 364229997Sken} 365229997Sken 366229997Skenstatic int 367229997Skenctl_grow_beio(struct ctl_be_block_softc *softc, int count) 368229997Sken{ 369229997Sken int i; 370229997Sken 371229997Sken for (i = 0; i < count; i++) { 372229997Sken struct ctl_be_block_io *beio; 373229997Sken 374229997Sken beio = (struct ctl_be_block_io *)malloc(sizeof(*beio), 375229997Sken M_CTLBLK, 376229997Sken M_WAITOK | M_ZERO); 377229997Sken beio->softc = softc; 378229997Sken mtx_lock(&softc->lock); 379229997Sken STAILQ_INSERT_TAIL(&softc->beio_free_queue, beio, links); 380229997Sken mtx_unlock(&softc->lock); 381229997Sken } 382229997Sken 383229997Sken return (i); 384229997Sken} 385229997Sken 386229997Sken#if 0 387229997Skenstatic void 388229997Skenctl_shrink_beio(struct ctl_be_block_softc *softc) 389229997Sken{ 390229997Sken struct ctl_be_block_io *beio, *beio_tmp; 391229997Sken 392229997Sken mtx_lock(&softc->lock); 393229997Sken STAILQ_FOREACH_SAFE(beio, &softc->beio_free_queue, links, beio_tmp) { 394229997Sken STAILQ_REMOVE(&softc->beio_free_queue, beio, 395229997Sken ctl_be_block_io, links); 396229997Sken free(beio, M_CTLBLK); 397229997Sken } 398229997Sken mtx_unlock(&softc->lock); 399229997Sken} 400229997Sken#endif 401229997Sken 402229997Skenstatic void 403229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 404229997Sken{ 405229997Sken union ctl_io *io; 406229997Sken int io_len; 407229997Sken 408229997Sken io = beio->io; 409229997Sken 410229997Sken if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) 411229997Sken io_len = beio->io_len; 412229997Sken else 413229997Sken io_len = 0; 414229997Sken 415229997Sken devstat_end_transaction(beio->lun->disk_stats, 416229997Sken /*bytes*/ io_len, 417229997Sken beio->ds_tag_type, 418229997Sken beio->ds_trans_type, 419229997Sken /*now*/ NULL, 420229997Sken /*then*/&beio->ds_t0); 421229997Sken 422229997Sken ctl_free_beio(beio); 423229997Sken ctl_done(io); 424229997Sken} 425229997Sken 426229997Skenstatic int 427229997Skenctl_be_block_move_done(union ctl_io *io) 428229997Sken{ 429229997Sken struct ctl_be_block_io *beio; 430229997Sken struct ctl_be_block_lun *be_lun; 431229997Sken#ifdef CTL_TIME_IO 432229997Sken struct bintime cur_bt; 433229997Sken#endif 434229997Sken 435229997Sken beio = (struct ctl_be_block_io *) 436229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 437229997Sken 438229997Sken be_lun = beio->lun; 439229997Sken 440229997Sken DPRINTF("entered\n"); 441229997Sken 442229997Sken#ifdef CTL_TIME_IO 443229997Sken getbintime(&cur_bt); 444229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 445229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 446229997Sken io->io_hdr.num_dmas++; 447229997Sken#endif 448229997Sken 449229997Sken /* 450229997Sken * We set status at this point for read commands, and write 451229997Sken * commands with errors. 452229997Sken */ 453229997Sken if ((beio->bio_cmd == BIO_READ) 454229997Sken && (io->io_hdr.port_status == 0) 455229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 456229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) 457229997Sken ctl_set_success(&io->scsiio); 458229997Sken else if ((io->io_hdr.port_status != 0) 459229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 460229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 461229997Sken /* 462229997Sken * For hardware error sense keys, the sense key 463229997Sken * specific value is defined to be a retry count, 464229997Sken * but we use it to pass back an internal FETD 465229997Sken * error code. XXX KDM Hopefully the FETD is only 466229997Sken * using 16 bits for an error code, since that's 467229997Sken * all the space we have in the sks field. 468229997Sken */ 469229997Sken ctl_set_internal_failure(&io->scsiio, 470229997Sken /*sks_valid*/ 1, 471229997Sken /*retry_count*/ 472229997Sken io->io_hdr.port_status); 473229997Sken } 474229997Sken 475229997Sken /* 476229997Sken * If this is a read, or a write with errors, it is done. 477229997Sken */ 478229997Sken if ((beio->bio_cmd == BIO_READ) 479229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 480229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 481229997Sken ctl_complete_beio(beio); 482229997Sken return (0); 483229997Sken } 484229997Sken 485229997Sken /* 486229997Sken * At this point, we have a write and the DMA completed 487229997Sken * successfully. We now have to queue it to the task queue to 488229997Sken * execute the backend I/O. That is because we do blocking 489229997Sken * memory allocations, and in the file backing case, blocking I/O. 490229997Sken * This move done routine is generally called in the SIM's 491229997Sken * interrupt context, and therefore we cannot block. 492229997Sken */ 493229997Sken mtx_lock(&be_lun->lock); 494229997Sken /* 495229997Sken * XXX KDM make sure that links is okay to use at this point. 496229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 497229997Sken * or deal with resource allocation here. 498229997Sken */ 499229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 500229997Sken mtx_unlock(&be_lun->lock); 501229997Sken 502229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 503229997Sken 504229997Sken return (0); 505229997Sken} 506229997Sken 507229997Skenstatic void 508229997Skenctl_be_block_biodone(struct bio *bio) 509229997Sken{ 510229997Sken struct ctl_be_block_io *beio; 511229997Sken struct ctl_be_block_lun *be_lun; 512229997Sken union ctl_io *io; 513261538Smav int error; 514229997Sken 515229997Sken beio = bio->bio_caller1; 516229997Sken be_lun = beio->lun; 517229997Sken io = beio->io; 518229997Sken 519229997Sken DPRINTF("entered\n"); 520229997Sken 521261538Smav error = bio->bio_error; 522229997Sken mtx_lock(&be_lun->lock); 523261538Smav if (error != 0) 524229997Sken beio->num_errors++; 525229997Sken 526229997Sken beio->num_bios_done++; 527229997Sken 528229997Sken /* 529229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 530229997Sken * during the free might cause it to complain. 531229997Sken */ 532229997Sken g_destroy_bio(bio); 533229997Sken 534229997Sken /* 535229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 536229997Sken * complete, then we're done. 537229997Sken */ 538229997Sken if ((beio->send_complete == 0) 539229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 540229997Sken mtx_unlock(&be_lun->lock); 541229997Sken return; 542229997Sken } 543229997Sken 544229997Sken /* 545229997Sken * At this point, we've verified that we are the last I/O to 546229997Sken * complete, so it's safe to drop the lock. 547229997Sken */ 548229997Sken mtx_unlock(&be_lun->lock); 549229997Sken 550229997Sken /* 551229997Sken * If there are any errors from the backing device, we fail the 552229997Sken * entire I/O with a medium error. 553229997Sken */ 554229997Sken if (beio->num_errors > 0) { 555261538Smav if (error == EOPNOTSUPP) { 556261538Smav ctl_set_invalid_opcode(&io->scsiio); 557261538Smav } else if (beio->bio_cmd == BIO_FLUSH) { 558229997Sken /* XXX KDM is there is a better error here? */ 559229997Sken ctl_set_internal_failure(&io->scsiio, 560229997Sken /*sks_valid*/ 1, 561229997Sken /*retry_count*/ 0xbad2); 562229997Sken } else 563229997Sken ctl_set_medium_error(&io->scsiio); 564229997Sken ctl_complete_beio(beio); 565229997Sken return; 566229997Sken } 567229997Sken 568229997Sken /* 569229997Sken * If this is a write or a flush, we're all done. 570229997Sken * If this is a read, we can now send the data to the user. 571229997Sken */ 572229997Sken if ((beio->bio_cmd == BIO_WRITE) 573229997Sken || (beio->bio_cmd == BIO_FLUSH)) { 574229997Sken ctl_set_success(&io->scsiio); 575229997Sken ctl_complete_beio(beio); 576229997Sken } else { 577229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 578229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 579229997Sken io->scsiio.kern_data_len = beio->io_len; 580229997Sken io->scsiio.kern_total_len = beio->io_len; 581229997Sken io->scsiio.kern_rel_offset = 0; 582229997Sken io->scsiio.kern_data_resid = 0; 583229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 584229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 585229997Sken#ifdef CTL_TIME_IO 586229997Sken getbintime(&io->io_hdr.dma_start_bt); 587229997Sken#endif 588229997Sken ctl_datamove(io); 589229997Sken } 590229997Sken} 591229997Sken 592229997Skenstatic void 593229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 594229997Sken struct ctl_be_block_io *beio) 595229997Sken{ 596229997Sken union ctl_io *io; 597229997Sken struct mount *mountpoint; 598241896Skib int error, lock_flags; 599229997Sken 600229997Sken DPRINTF("entered\n"); 601229997Sken 602229997Sken io = beio->io; 603229997Sken 604229997Sken (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 605229997Sken 606229997Sken if (MNT_SHARED_WRITES(mountpoint) 607229997Sken || ((mountpoint == NULL) 608229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 609229997Sken lock_flags = LK_SHARED; 610229997Sken else 611229997Sken lock_flags = LK_EXCLUSIVE; 612229997Sken 613229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 614229997Sken 615229997Sken binuptime(&beio->ds_t0); 616229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 617229997Sken 618229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 619229997Sken VOP_UNLOCK(be_lun->vn, 0); 620229997Sken 621229997Sken vn_finished_write(mountpoint); 622229997Sken 623229997Sken if (error == 0) 624229997Sken ctl_set_success(&io->scsiio); 625229997Sken else { 626229997Sken /* XXX KDM is there is a better error here? */ 627229997Sken ctl_set_internal_failure(&io->scsiio, 628229997Sken /*sks_valid*/ 1, 629229997Sken /*retry_count*/ 0xbad1); 630229997Sken } 631229997Sken 632229997Sken ctl_complete_beio(beio); 633229997Sken} 634229997Sken 635258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 636258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 637258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 638258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 639229997Sken 640229997Skenstatic void 641229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 642229997Sken struct ctl_be_block_io *beio) 643229997Sken{ 644229997Sken struct ctl_be_block_filedata *file_data; 645229997Sken union ctl_io *io; 646229997Sken struct uio xuio; 647229997Sken struct iovec *xiovec; 648241896Skib int flags; 649229997Sken int error, i; 650229997Sken 651229997Sken DPRINTF("entered\n"); 652229997Sken 653229997Sken file_data = &be_lun->backend.file; 654229997Sken io = beio->io; 655229997Sken flags = beio->bio_flags; 656229997Sken 657229997Sken if (beio->bio_cmd == BIO_READ) { 658229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 659229997Sken } else { 660229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 661229997Sken } 662229997Sken 663229997Sken bzero(&xuio, sizeof(xuio)); 664229997Sken if (beio->bio_cmd == BIO_READ) 665229997Sken xuio.uio_rw = UIO_READ; 666229997Sken else 667229997Sken xuio.uio_rw = UIO_WRITE; 668229997Sken 669229997Sken xuio.uio_offset = beio->io_offset; 670229997Sken xuio.uio_resid = beio->io_len; 671229997Sken xuio.uio_segflg = UIO_SYSSPACE; 672229997Sken xuio.uio_iov = beio->xiovecs; 673229997Sken xuio.uio_iovcnt = beio->num_segs; 674229997Sken xuio.uio_td = curthread; 675229997Sken 676229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 677229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 678229997Sken xiovec->iov_len = beio->sg_segs[i].len; 679229997Sken } 680229997Sken 681229997Sken if (beio->bio_cmd == BIO_READ) { 682229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 683229997Sken 684229997Sken binuptime(&beio->ds_t0); 685229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 686229997Sken 687229997Sken /* 688229997Sken * UFS pays attention to IO_DIRECT for reads. If the 689229997Sken * DIRECTIO option is configured into the kernel, it calls 690229997Sken * ffs_rawread(). But that only works for single-segment 691229997Sken * uios with user space addresses. In our case, with a 692229997Sken * kernel uio, it still reads into the buffer cache, but it 693229997Sken * will just try to release the buffer from the cache later 694229997Sken * on in ffs_read(). 695229997Sken * 696229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 697229997Sken * 698229997Sken * UFS does not pay attention to IO_SYNC for reads. 699229997Sken * 700229997Sken * ZFS pays attention to IO_SYNC (which translates into the 701229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 702229997Sken * attempts to sync the file before reading. 703229997Sken * 704229997Sken * So, to attempt to provide some barrier semantics in the 705229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 706229997Sken */ 707229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 708229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 709229997Sken 710229997Sken VOP_UNLOCK(be_lun->vn, 0); 711229997Sken } else { 712229997Sken struct mount *mountpoint; 713229997Sken int lock_flags; 714229997Sken 715229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 716229997Sken 717229997Sken if (MNT_SHARED_WRITES(mountpoint) 718229997Sken || ((mountpoint == NULL) 719229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 720229997Sken lock_flags = LK_SHARED; 721229997Sken else 722229997Sken lock_flags = LK_EXCLUSIVE; 723229997Sken 724229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 725229997Sken 726229997Sken binuptime(&beio->ds_t0); 727229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 728229997Sken 729229997Sken /* 730229997Sken * UFS pays attention to IO_DIRECT for writes. The write 731229997Sken * is done asynchronously. (Normally the write would just 732229997Sken * get put into cache. 733229997Sken * 734229997Sken * UFS pays attention to IO_SYNC for writes. It will 735229997Sken * attempt to write the buffer out synchronously if that 736229997Sken * flag is set. 737229997Sken * 738229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 739229997Sken * 740229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 741229997Sken * for writes. It will flush the transaction from the 742229997Sken * cache before returning. 743229997Sken * 744229997Sken * So if we've got the BIO_ORDERED flag set, we want 745229997Sken * IO_SYNC in either the UFS or ZFS case. 746229997Sken */ 747229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 748229997Sken IO_SYNC : 0, file_data->cred); 749229997Sken VOP_UNLOCK(be_lun->vn, 0); 750229997Sken 751229997Sken vn_finished_write(mountpoint); 752229997Sken } 753229997Sken 754229997Sken /* 755229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 756229997Sken * return the I/O to the user. 757229997Sken */ 758229997Sken if (error != 0) { 759229997Sken char path_str[32]; 760229997Sken 761229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 762229997Sken /* 763229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 764229997Sken * filesystem fills up. What kind of SCSI error should we 765229997Sken * return for that? 766229997Sken */ 767229997Sken printf("%s%s command returned errno %d\n", path_str, 768229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 769229997Sken ctl_set_medium_error(&io->scsiio); 770229997Sken ctl_complete_beio(beio); 771229997Sken return; 772229997Sken } 773229997Sken 774229997Sken /* 775229997Sken * If this is a write, we're all done. 776229997Sken * If this is a read, we can now send the data to the user. 777229997Sken */ 778229997Sken if (beio->bio_cmd == BIO_WRITE) { 779229997Sken ctl_set_success(&io->scsiio); 780229997Sken SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 781229997Sken ctl_complete_beio(beio); 782229997Sken } else { 783229997Sken SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 784229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 785229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 786229997Sken io->scsiio.kern_data_len = beio->io_len; 787229997Sken io->scsiio.kern_total_len = beio->io_len; 788229997Sken io->scsiio.kern_rel_offset = 0; 789229997Sken io->scsiio.kern_data_resid = 0; 790229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 791229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 792229997Sken#ifdef CTL_TIME_IO 793229997Sken getbintime(&io->io_hdr.dma_start_bt); 794229997Sken#endif 795229997Sken ctl_datamove(io); 796229997Sken } 797229997Sken} 798229997Sken 799229997Skenstatic void 800229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 801229997Sken struct ctl_be_block_io *beio) 802229997Sken{ 803229997Sken struct bio *bio; 804229997Sken union ctl_io *io; 805229997Sken struct ctl_be_block_devdata *dev_data; 806229997Sken 807229997Sken dev_data = &be_lun->backend.dev; 808229997Sken io = beio->io; 809229997Sken 810229997Sken DPRINTF("entered\n"); 811229997Sken 812229997Sken /* This can't fail, it's a blocking allocation. */ 813229997Sken bio = g_alloc_bio(); 814229997Sken 815229997Sken bio->bio_cmd = BIO_FLUSH; 816229997Sken bio->bio_flags |= BIO_ORDERED; 817229997Sken bio->bio_dev = dev_data->cdev; 818229997Sken bio->bio_offset = 0; 819229997Sken bio->bio_data = 0; 820229997Sken bio->bio_done = ctl_be_block_biodone; 821229997Sken bio->bio_caller1 = beio; 822229997Sken bio->bio_pblkno = 0; 823229997Sken 824229997Sken /* 825229997Sken * We don't need to acquire the LUN lock here, because we are only 826229997Sken * sending one bio, and so there is no other context to synchronize 827229997Sken * with. 828229997Sken */ 829229997Sken beio->num_bios_sent = 1; 830229997Sken beio->send_complete = 1; 831229997Sken 832229997Sken binuptime(&beio->ds_t0); 833229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 834229997Sken 835229997Sken (*dev_data->csw->d_strategy)(bio); 836229997Sken} 837229997Sken 838229997Skenstatic void 839229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 840229997Sken struct ctl_be_block_io *beio) 841229997Sken{ 842229997Sken int i; 843229997Sken struct bio *bio; 844229997Sken struct ctl_be_block_devdata *dev_data; 845229997Sken off_t cur_offset; 846229997Sken int max_iosize; 847229997Sken 848229997Sken DPRINTF("entered\n"); 849229997Sken 850229997Sken dev_data = &be_lun->backend.dev; 851229997Sken 852229997Sken /* 853229997Sken * We have to limit our I/O size to the maximum supported by the 854229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 855229997Sken * set it properly, use DFLTPHYS. 856229997Sken */ 857229997Sken max_iosize = dev_data->cdev->si_iosize_max; 858229997Sken if (max_iosize < PAGE_SIZE) 859229997Sken max_iosize = DFLTPHYS; 860229997Sken 861229997Sken cur_offset = beio->io_offset; 862229997Sken 863229997Sken /* 864229997Sken * XXX KDM need to accurately reflect the number of I/Os outstanding 865229997Sken * to a device. 866229997Sken */ 867229997Sken binuptime(&beio->ds_t0); 868229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 869229997Sken 870229997Sken for (i = 0; i < beio->num_segs; i++) { 871229997Sken size_t cur_size; 872229997Sken uint8_t *cur_ptr; 873229997Sken 874229997Sken cur_size = beio->sg_segs[i].len; 875229997Sken cur_ptr = beio->sg_segs[i].addr; 876229997Sken 877229997Sken while (cur_size > 0) { 878229997Sken /* This can't fail, it's a blocking allocation. */ 879229997Sken bio = g_alloc_bio(); 880229997Sken 881229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 882229997Sken 883229997Sken bio->bio_cmd = beio->bio_cmd; 884229997Sken bio->bio_flags |= beio->bio_flags; 885229997Sken bio->bio_dev = dev_data->cdev; 886229997Sken bio->bio_caller1 = beio; 887229997Sken bio->bio_length = min(cur_size, max_iosize); 888229997Sken bio->bio_offset = cur_offset; 889229997Sken bio->bio_data = cur_ptr; 890229997Sken bio->bio_done = ctl_be_block_biodone; 891229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 892229997Sken 893229997Sken cur_offset += bio->bio_length; 894229997Sken cur_ptr += bio->bio_length; 895229997Sken cur_size -= bio->bio_length; 896229997Sken 897229997Sken /* 898229997Sken * Make sure we set the complete bit just before we 899229997Sken * issue the last bio so we don't wind up with a 900229997Sken * race. 901229997Sken * 902229997Sken * Use the LUN mutex here instead of a combination 903229997Sken * of atomic variables for simplicity. 904229997Sken * 905229997Sken * XXX KDM we could have a per-IO lock, but that 906229997Sken * would cause additional per-IO setup and teardown 907229997Sken * overhead. Hopefully there won't be too much 908229997Sken * contention on the LUN lock. 909229997Sken */ 910229997Sken mtx_lock(&be_lun->lock); 911229997Sken 912229997Sken beio->num_bios_sent++; 913229997Sken 914229997Sken if ((i == beio->num_segs - 1) 915229997Sken && (cur_size == 0)) 916229997Sken beio->send_complete = 1; 917229997Sken 918229997Sken mtx_unlock(&be_lun->lock); 919229997Sken 920229997Sken (*dev_data->csw->d_strategy)(bio); 921229997Sken } 922229997Sken } 923229997Sken} 924229997Sken 925229997Skenstatic void 926229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 927229997Sken union ctl_io *io) 928229997Sken{ 929229997Sken struct ctl_be_block_io *beio; 930229997Sken struct ctl_be_block_softc *softc; 931229997Sken 932229997Sken DPRINTF("entered\n"); 933229997Sken 934229997Sken softc = be_lun->softc; 935229997Sken beio = ctl_alloc_beio(softc); 936229997Sken if (beio == NULL) { 937229997Sken /* 938229997Sken * This should not happen. ctl_alloc_beio() will call 939229997Sken * ctl_grow_beio() with a blocking malloc as needed. 940229997Sken * A malloc with M_WAITOK should not fail. 941229997Sken */ 942229997Sken ctl_set_busy(&io->scsiio); 943229997Sken ctl_done(io); 944229997Sken return; 945229997Sken } 946229997Sken 947229997Sken beio->io = io; 948229997Sken beio->softc = softc; 949229997Sken beio->lun = be_lun; 950229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 951229997Sken 952229997Sken switch (io->scsiio.cdb[0]) { 953229997Sken case SYNCHRONIZE_CACHE: 954229997Sken case SYNCHRONIZE_CACHE_16: 955249194Strasz beio->bio_cmd = BIO_FLUSH; 956229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 957229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 958229997Sken beio->io_len = 0; 959229997Sken be_lun->lun_flush(be_lun, beio); 960229997Sken break; 961229997Sken default: 962229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 963229997Sken break; 964229997Sken } 965229997Sken} 966229997Sken 967258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 968258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 969258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 970258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 971229997Sken 972229997Skenstatic void 973229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 974229997Sken union ctl_io *io) 975229997Sken{ 976229997Sken struct ctl_be_block_io *beio; 977229997Sken struct ctl_be_block_softc *softc; 978229997Sken struct ctl_lba_len lbalen; 979229997Sken uint64_t len_left, io_size_bytes; 980229997Sken int i; 981229997Sken 982229997Sken softc = be_lun->softc; 983229997Sken 984229997Sken DPRINTF("entered\n"); 985229997Sken 986229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 987229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 988229997Sken } else { 989229997Sken SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 990229997Sken } 991229997Sken 992229997Sken memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 993229997Sken sizeof(lbalen)); 994229997Sken 995229997Sken io_size_bytes = lbalen.len * be_lun->blocksize; 996229997Sken 997229997Sken /* 998229997Sken * XXX KDM this is temporary, until we implement chaining of beio 999229997Sken * structures and multiple datamove calls to move all the data in 1000229997Sken * or out. 1001229997Sken */ 1002229997Sken if (io_size_bytes > CTLBLK_MAX_IO_SIZE) { 1003229997Sken printf("%s: IO length %ju > max io size %u\n", __func__, 1004229997Sken io_size_bytes, CTLBLK_MAX_IO_SIZE); 1005229997Sken ctl_set_invalid_field(&io->scsiio, 1006229997Sken /*sks_valid*/ 0, 1007229997Sken /*command*/ 1, 1008229997Sken /*field*/ 0, 1009229997Sken /*bit_valid*/ 0, 1010229997Sken /*bit*/ 0); 1011229997Sken ctl_done(io); 1012229997Sken return; 1013229997Sken } 1014229997Sken 1015229997Sken beio = ctl_alloc_beio(softc); 1016229997Sken if (beio == NULL) { 1017229997Sken /* 1018229997Sken * This should not happen. ctl_alloc_beio() will call 1019229997Sken * ctl_grow_beio() with a blocking malloc as needed. 1020229997Sken * A malloc with M_WAITOK should not fail. 1021229997Sken */ 1022229997Sken ctl_set_busy(&io->scsiio); 1023229997Sken ctl_done(io); 1024229997Sken return; 1025229997Sken } 1026229997Sken 1027229997Sken beio->io = io; 1028229997Sken beio->softc = softc; 1029229997Sken beio->lun = be_lun; 1030229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 1031229997Sken 1032229997Sken /* 1033229997Sken * If the I/O came down with an ordered or head of queue tag, set 1034229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 1035229997Sken * pretty much the best we can do. 1036229997Sken * 1037229997Sken * XXX KDM we don't have a great way to easily know about the FUA 1038229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 1039229997Sken * pass that knowledge to the backend), and in any case we would 1040229997Sken * need to determine how to handle it. 1041229997Sken */ 1042229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1043229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1044229997Sken beio->bio_flags = BIO_ORDERED; 1045229997Sken 1046229997Sken switch (io->scsiio.tag_type) { 1047229997Sken case CTL_TAG_ORDERED: 1048229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1049229997Sken break; 1050229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1051229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1052229997Sken break; 1053229997Sken case CTL_TAG_UNTAGGED: 1054229997Sken case CTL_TAG_SIMPLE: 1055229997Sken case CTL_TAG_ACA: 1056229997Sken default: 1057229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1058229997Sken break; 1059229997Sken } 1060229997Sken 1061229997Sken /* 1062229997Sken * This path handles read and write only. The config write path 1063229997Sken * handles flush operations. 1064229997Sken */ 1065229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 1066229997Sken beio->bio_cmd = BIO_READ; 1067229997Sken beio->ds_trans_type = DEVSTAT_READ; 1068229997Sken } else { 1069229997Sken beio->bio_cmd = BIO_WRITE; 1070229997Sken beio->ds_trans_type = DEVSTAT_WRITE; 1071229997Sken } 1072229997Sken 1073229997Sken beio->io_len = lbalen.len * be_lun->blocksize; 1074229997Sken beio->io_offset = lbalen.lba * be_lun->blocksize; 1075229997Sken 1076229997Sken DPRINTF("%s at LBA %jx len %u\n", 1077229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1078229997Sken (uintmax_t)lbalen.lba, lbalen.len); 1079229997Sken 1080229997Sken for (i = 0, len_left = io_size_bytes; i < CTLBLK_MAX_SEGS && 1081229997Sken len_left > 0; i++) { 1082229997Sken 1083229997Sken /* 1084229997Sken * Setup the S/G entry for this chunk. 1085229997Sken */ 1086229997Sken beio->sg_segs[i].len = min(MAXPHYS, len_left); 1087229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1088229997Sken 1089229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1090229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1091229997Sken 1092229997Sken beio->num_segs++; 1093229997Sken len_left -= beio->sg_segs[i].len; 1094229997Sken } 1095229997Sken 1096229997Sken /* 1097229997Sken * For the read case, we need to read the data into our buffers and 1098229997Sken * then we can send it back to the user. For the write case, we 1099229997Sken * need to get the data from the user first. 1100229997Sken */ 1101229997Sken if (beio->bio_cmd == BIO_READ) { 1102229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1103229997Sken be_lun->dispatch(be_lun, beio); 1104229997Sken } else { 1105229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1106229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 1107229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1108229997Sken io->scsiio.kern_data_len = beio->io_len; 1109229997Sken io->scsiio.kern_total_len = beio->io_len; 1110229997Sken io->scsiio.kern_rel_offset = 0; 1111229997Sken io->scsiio.kern_data_resid = 0; 1112229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 1113229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1114229997Sken#ifdef CTL_TIME_IO 1115229997Sken getbintime(&io->io_hdr.dma_start_bt); 1116229997Sken#endif 1117229997Sken ctl_datamove(io); 1118229997Sken } 1119229997Sken} 1120229997Sken 1121229997Skenstatic void 1122229997Skenctl_be_block_worker(void *context, int pending) 1123229997Sken{ 1124229997Sken struct ctl_be_block_lun *be_lun; 1125229997Sken struct ctl_be_block_softc *softc; 1126229997Sken union ctl_io *io; 1127229997Sken 1128229997Sken be_lun = (struct ctl_be_block_lun *)context; 1129229997Sken softc = be_lun->softc; 1130229997Sken 1131229997Sken DPRINTF("entered\n"); 1132229997Sken 1133229997Sken mtx_lock(&be_lun->lock); 1134229997Sken for (;;) { 1135229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1136229997Sken if (io != NULL) { 1137229997Sken struct ctl_be_block_io *beio; 1138229997Sken 1139229997Sken DPRINTF("datamove queue\n"); 1140229997Sken 1141229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1142229997Sken ctl_io_hdr, links); 1143229997Sken 1144229997Sken mtx_unlock(&be_lun->lock); 1145229997Sken 1146229997Sken beio = (struct ctl_be_block_io *) 1147229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1148229997Sken 1149229997Sken be_lun->dispatch(be_lun, beio); 1150229997Sken 1151229997Sken mtx_lock(&be_lun->lock); 1152229997Sken continue; 1153229997Sken } 1154229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1155229997Sken if (io != NULL) { 1156229997Sken 1157229997Sken DPRINTF("config write queue\n"); 1158229997Sken 1159229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1160229997Sken ctl_io_hdr, links); 1161229997Sken 1162229997Sken mtx_unlock(&be_lun->lock); 1163229997Sken 1164229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1165229997Sken 1166229997Sken mtx_lock(&be_lun->lock); 1167229997Sken continue; 1168229997Sken } 1169229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1170229997Sken if (io != NULL) { 1171229997Sken DPRINTF("input queue\n"); 1172229997Sken 1173229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1174229997Sken ctl_io_hdr, links); 1175229997Sken mtx_unlock(&be_lun->lock); 1176229997Sken 1177229997Sken /* 1178229997Sken * We must drop the lock, since this routine and 1179229997Sken * its children may sleep. 1180229997Sken */ 1181229997Sken ctl_be_block_dispatch(be_lun, io); 1182229997Sken 1183229997Sken mtx_lock(&be_lun->lock); 1184229997Sken continue; 1185229997Sken } 1186229997Sken 1187229997Sken /* 1188229997Sken * If we get here, there is no work left in the queues, so 1189229997Sken * just break out and let the task queue go to sleep. 1190229997Sken */ 1191229997Sken break; 1192229997Sken } 1193229997Sken mtx_unlock(&be_lun->lock); 1194229997Sken} 1195229997Sken 1196229997Sken/* 1197229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1198229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1199229997Sken * thread. 1200229997Sken */ 1201229997Skenstatic int 1202229997Skenctl_be_block_submit(union ctl_io *io) 1203229997Sken{ 1204229997Sken struct ctl_be_block_lun *be_lun; 1205229997Sken struct ctl_be_lun *ctl_be_lun; 1206229997Sken int retval; 1207229997Sken 1208229997Sken DPRINTF("entered\n"); 1209229997Sken 1210229997Sken retval = CTL_RETVAL_COMPLETE; 1211229997Sken 1212229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1213229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1214229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1215229997Sken 1216229997Sken /* 1217229997Sken * Make sure we only get SCSI I/O. 1218229997Sken */ 1219229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1220229997Sken "%#x) encountered", io->io_hdr.io_type)); 1221229997Sken 1222229997Sken mtx_lock(&be_lun->lock); 1223229997Sken /* 1224229997Sken * XXX KDM make sure that links is okay to use at this point. 1225229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1226229997Sken * or deal with resource allocation here. 1227229997Sken */ 1228229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1229229997Sken mtx_unlock(&be_lun->lock); 1230229997Sken 1231229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1232229997Sken 1233229997Sken return (retval); 1234229997Sken} 1235229997Sken 1236229997Skenstatic int 1237229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1238229997Sken int flag, struct thread *td) 1239229997Sken{ 1240229997Sken struct ctl_be_block_softc *softc; 1241229997Sken int error; 1242229997Sken 1243229997Sken softc = &backend_block_softc; 1244229997Sken 1245229997Sken error = 0; 1246229997Sken 1247229997Sken switch (cmd) { 1248229997Sken case CTL_LUN_REQ: { 1249229997Sken struct ctl_lun_req *lun_req; 1250229997Sken 1251229997Sken lun_req = (struct ctl_lun_req *)addr; 1252229997Sken 1253229997Sken switch (lun_req->reqtype) { 1254229997Sken case CTL_LUNREQ_CREATE: 1255229997Sken error = ctl_be_block_create(softc, lun_req); 1256229997Sken break; 1257229997Sken case CTL_LUNREQ_RM: 1258229997Sken error = ctl_be_block_rm(softc, lun_req); 1259229997Sken break; 1260232604Strasz case CTL_LUNREQ_MODIFY: 1261232604Strasz error = ctl_be_block_modify(softc, lun_req); 1262232604Strasz break; 1263229997Sken default: 1264229997Sken lun_req->status = CTL_LUN_ERROR; 1265229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1266229997Sken "%s: invalid LUN request type %d", __func__, 1267229997Sken lun_req->reqtype); 1268229997Sken break; 1269229997Sken } 1270229997Sken break; 1271229997Sken } 1272229997Sken default: 1273229997Sken error = ENOTTY; 1274229997Sken break; 1275229997Sken } 1276229997Sken 1277229997Sken return (error); 1278229997Sken} 1279229997Sken 1280229997Skenstatic int 1281229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1282229997Sken{ 1283229997Sken struct ctl_be_block_filedata *file_data; 1284229997Sken struct ctl_lun_create_params *params; 1285229997Sken struct vattr vattr; 1286229997Sken int error; 1287229997Sken 1288229997Sken error = 0; 1289229997Sken file_data = &be_lun->backend.file; 1290229997Sken params = &req->reqdata.create; 1291229997Sken 1292229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1293229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1294229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1295229997Sken 1296229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1297229997Sken if (error != 0) { 1298229997Sken snprintf(req->error_str, sizeof(req->error_str), 1299229997Sken "error calling VOP_GETATTR() for file %s", 1300229997Sken be_lun->dev_path); 1301229997Sken return (error); 1302229997Sken } 1303229997Sken 1304229997Sken /* 1305229997Sken * Verify that we have the ability to upgrade to exclusive 1306229997Sken * access on this file so we can trap errors at open instead 1307229997Sken * of reporting them during first access. 1308229997Sken */ 1309229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1310229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1311229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1312229997Sken error = EBADF; 1313229997Sken snprintf(req->error_str, sizeof(req->error_str), 1314229997Sken "error locking file %s", be_lun->dev_path); 1315229997Sken return (error); 1316229997Sken } 1317229997Sken } 1318229997Sken 1319229997Sken 1320229997Sken file_data->cred = crhold(curthread->td_ucred); 1321232604Strasz if (params->lun_size_bytes != 0) 1322232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1323232604Strasz else 1324232604Strasz be_lun->size_bytes = vattr.va_size; 1325229997Sken /* 1326229997Sken * We set the multi thread flag for file operations because all 1327229997Sken * filesystems (in theory) are capable of allowing multiple readers 1328229997Sken * of a file at once. So we want to get the maximum possible 1329229997Sken * concurrency. 1330229997Sken */ 1331229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1332229997Sken 1333229997Sken /* 1334229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1335229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1336229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1337229997Sken * may not work with other OSes as well. So just export a sector 1338229997Sken * size of 512 bytes, which should work with any OS or 1339229997Sken * application. Since our backing is a file, any block size will 1340229997Sken * work fine for the backing store. 1341229997Sken */ 1342229997Sken#if 0 1343229997Sken be_lun->blocksize= vattr.va_blocksize; 1344229997Sken#endif 1345229997Sken if (params->blocksize_bytes != 0) 1346229997Sken be_lun->blocksize = params->blocksize_bytes; 1347229997Sken else 1348229997Sken be_lun->blocksize = 512; 1349229997Sken 1350229997Sken /* 1351229997Sken * Sanity check. The media size has to be at least one 1352229997Sken * sector long. 1353229997Sken */ 1354229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1355229997Sken error = EINVAL; 1356229997Sken snprintf(req->error_str, sizeof(req->error_str), 1357229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1358229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1359229997Sken } 1360229997Sken return (error); 1361229997Sken} 1362229997Sken 1363229997Skenstatic int 1364229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1365229997Sken{ 1366229997Sken struct ctl_lun_create_params *params; 1367229997Sken struct vattr vattr; 1368229997Sken struct cdev *dev; 1369229997Sken struct cdevsw *devsw; 1370229997Sken int error; 1371229997Sken 1372229997Sken params = &req->reqdata.create; 1373229997Sken 1374229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1375229997Sken be_lun->dispatch = ctl_be_block_dispatch_dev; 1376229997Sken be_lun->lun_flush = ctl_be_block_flush_dev; 1377229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1378229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1379229997Sken &be_lun->backend.dev.dev_ref); 1380229997Sken if (be_lun->backend.dev.csw == NULL) 1381229997Sken panic("Unable to retrieve device switch"); 1382229997Sken 1383229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1384229997Sken if (error) { 1385229997Sken snprintf(req->error_str, sizeof(req->error_str), 1386229997Sken "%s: error getting vnode attributes for device %s", 1387229997Sken __func__, be_lun->dev_path); 1388229997Sken return (error); 1389229997Sken } 1390229997Sken 1391229997Sken dev = be_lun->vn->v_rdev; 1392229997Sken devsw = dev->si_devsw; 1393229997Sken if (!devsw->d_ioctl) { 1394229997Sken snprintf(req->error_str, sizeof(req->error_str), 1395229997Sken "%s: no d_ioctl for device %s!", __func__, 1396229997Sken be_lun->dev_path); 1397229997Sken return (ENODEV); 1398229997Sken } 1399229997Sken 1400229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1401229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1402229997Sken curthread); 1403229997Sken if (error) { 1404229997Sken snprintf(req->error_str, sizeof(req->error_str), 1405229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1406229997Sken "on %s!", __func__, error, be_lun->dev_path); 1407229997Sken return (error); 1408229997Sken } 1409229997Sken 1410229997Sken /* 1411229997Sken * If the user has asked for a blocksize that is greater than the 1412229997Sken * backing device's blocksize, we can do it only if the blocksize 1413229997Sken * the user is asking for is an even multiple of the underlying 1414229997Sken * device's blocksize. 1415229997Sken */ 1416229997Sken if ((params->blocksize_bytes != 0) 1417229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1418229997Sken uint32_t bs_multiple, tmp_blocksize; 1419229997Sken 1420229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1421229997Sken 1422229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1423229997Sken 1424229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1425229997Sken be_lun->blocksize = params->blocksize_bytes; 1426229997Sken } else { 1427229997Sken snprintf(req->error_str, sizeof(req->error_str), 1428229997Sken "%s: requested blocksize %u is not an even " 1429229997Sken "multiple of backing device blocksize %u", 1430229997Sken __func__, params->blocksize_bytes, 1431229997Sken be_lun->blocksize); 1432229997Sken return (EINVAL); 1433229997Sken 1434229997Sken } 1435229997Sken } else if ((params->blocksize_bytes != 0) 1436229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1437229997Sken snprintf(req->error_str, sizeof(req->error_str), 1438229997Sken "%s: requested blocksize %u < backing device " 1439229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1440229997Sken be_lun->blocksize); 1441229997Sken return (EINVAL); 1442229997Sken } 1443229997Sken 1444229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1445229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1446229997Sken curthread); 1447229997Sken if (error) { 1448229997Sken snprintf(req->error_str, sizeof(req->error_str), 1449232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1450232604Strasz " ioctl on %s!", __func__, error, 1451232604Strasz be_lun->dev_path); 1452229997Sken return (error); 1453229997Sken } 1454229997Sken 1455232604Strasz if (params->lun_size_bytes != 0) { 1456232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1457232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1458232604Strasz "%s: requested LUN size %ju > backing device " 1459232604Strasz "size %ju", __func__, 1460232604Strasz (uintmax_t)params->lun_size_bytes, 1461232604Strasz (uintmax_t)be_lun->size_bytes); 1462232604Strasz return (EINVAL); 1463232604Strasz } 1464232604Strasz 1465232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1466232604Strasz } 1467232604Strasz 1468229997Sken return (0); 1469229997Sken} 1470229997Sken 1471229997Skenstatic int 1472229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1473229997Sken{ 1474229997Sken DROP_GIANT(); 1475229997Sken if (be_lun->vn) { 1476229997Sken int flags = FREAD | FWRITE; 1477229997Sken 1478229997Sken switch (be_lun->dev_type) { 1479229997Sken case CTL_BE_BLOCK_DEV: 1480229997Sken if (be_lun->backend.dev.csw) { 1481229997Sken dev_relthread(be_lun->backend.dev.cdev, 1482229997Sken be_lun->backend.dev.dev_ref); 1483229997Sken be_lun->backend.dev.csw = NULL; 1484229997Sken be_lun->backend.dev.cdev = NULL; 1485229997Sken } 1486229997Sken break; 1487229997Sken case CTL_BE_BLOCK_FILE: 1488229997Sken break; 1489229997Sken case CTL_BE_BLOCK_NONE: 1490258871Strasz break; 1491229997Sken default: 1492229997Sken panic("Unexpected backend type."); 1493229997Sken break; 1494229997Sken } 1495229997Sken 1496229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1497229997Sken be_lun->vn = NULL; 1498229997Sken 1499229997Sken switch (be_lun->dev_type) { 1500229997Sken case CTL_BE_BLOCK_DEV: 1501229997Sken break; 1502229997Sken case CTL_BE_BLOCK_FILE: 1503229997Sken if (be_lun->backend.file.cred != NULL) { 1504229997Sken crfree(be_lun->backend.file.cred); 1505229997Sken be_lun->backend.file.cred = NULL; 1506229997Sken } 1507229997Sken break; 1508229997Sken case CTL_BE_BLOCK_NONE: 1509258871Strasz break; 1510229997Sken default: 1511229997Sken panic("Unexpected backend type."); 1512229997Sken break; 1513229997Sken } 1514229997Sken } 1515229997Sken PICKUP_GIANT(); 1516229997Sken 1517229997Sken return (0); 1518229997Sken} 1519229997Sken 1520229997Skenstatic int 1521229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1522229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1523229997Sken{ 1524229997Sken struct nameidata nd; 1525229997Sken int flags; 1526229997Sken int error; 1527229997Sken 1528229997Sken /* 1529229997Sken * XXX KDM allow a read-only option? 1530229997Sken */ 1531229997Sken flags = FREAD | FWRITE; 1532229997Sken error = 0; 1533229997Sken 1534229997Sken if (rootvnode == NULL) { 1535229997Sken snprintf(req->error_str, sizeof(req->error_str), 1536229997Sken "%s: Root filesystem is not mounted", __func__); 1537229997Sken return (1); 1538229997Sken } 1539229997Sken 1540229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1541229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1542229997Sken VREF(rootvnode); 1543229997Sken } 1544229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1545229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1546229997Sken VREF(rootvnode); 1547229997Sken } 1548229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1549229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1550229997Sken VREF(rootvnode); 1551229997Sken } 1552229997Sken 1553229997Sken again: 1554229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1555229997Sken error = vn_open(&nd, &flags, 0, NULL); 1556229997Sken if (error) { 1557229997Sken /* 1558229997Sken * This is the only reasonable guess we can make as far as 1559229997Sken * path if the user doesn't give us a fully qualified path. 1560229997Sken * If they want to specify a file, they need to specify the 1561229997Sken * full path. 1562229997Sken */ 1563229997Sken if (be_lun->dev_path[0] != '/') { 1564229997Sken char *dev_path = "/dev/"; 1565229997Sken char *dev_name; 1566229997Sken 1567229997Sken /* Try adding device path at beginning of name */ 1568229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1569229997Sken + strlen(dev_path) + 1, 1570229997Sken M_CTLBLK, M_WAITOK); 1571229997Sken if (dev_name) { 1572229997Sken sprintf(dev_name, "%s%s", dev_path, 1573229997Sken be_lun->dev_path); 1574229997Sken free(be_lun->dev_path, M_CTLBLK); 1575229997Sken be_lun->dev_path = dev_name; 1576229997Sken goto again; 1577229997Sken } 1578229997Sken } 1579229997Sken snprintf(req->error_str, sizeof(req->error_str), 1580229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1581229997Sken return (error); 1582229997Sken } 1583229997Sken 1584229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1585229997Sken 1586229997Sken be_lun->vn = nd.ni_vp; 1587229997Sken 1588229997Sken /* We only support disks and files. */ 1589229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1590229997Sken error = ctl_be_block_open_dev(be_lun, req); 1591229997Sken } else if (be_lun->vn->v_type == VREG) { 1592229997Sken error = ctl_be_block_open_file(be_lun, req); 1593229997Sken } else { 1594229997Sken error = EINVAL; 1595229997Sken snprintf(req->error_str, sizeof(req->error_str), 1596258871Strasz "%s is not a disk or plain file", be_lun->dev_path); 1597229997Sken } 1598229997Sken VOP_UNLOCK(be_lun->vn, 0); 1599229997Sken 1600229997Sken if (error != 0) { 1601229997Sken ctl_be_block_close(be_lun); 1602229997Sken return (error); 1603229997Sken } 1604229997Sken 1605229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1606229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1607229997Sken 1608229997Sken return (0); 1609229997Sken} 1610229997Sken 1611229997Skenstatic int 1612229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1613229997Sken{ 1614229997Sken struct ctl_be_block_lun *be_lun; 1615229997Sken struct ctl_lun_create_params *params; 1616229997Sken struct ctl_be_arg *file_arg; 1617229997Sken char tmpstr[32]; 1618229997Sken int retval, num_threads; 1619229997Sken int i; 1620229997Sken 1621229997Sken params = &req->reqdata.create; 1622229997Sken retval = 0; 1623229997Sken 1624229997Sken num_threads = cbb_num_threads; 1625229997Sken 1626229997Sken file_arg = NULL; 1627229997Sken 1628229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1629229997Sken 1630229997Sken be_lun->softc = softc; 1631229997Sken STAILQ_INIT(&be_lun->input_queue); 1632229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1633229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1634254759Strasz STAILQ_INIT(&be_lun->ctl_be_lun.options); 1635229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1636229997Sken mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF); 1637229997Sken 1638229997Sken be_lun->lun_zone = uma_zcreate(be_lun->lunname, MAXPHYS, 1639256995Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1640229997Sken 1641229997Sken if (be_lun->lun_zone == NULL) { 1642229997Sken snprintf(req->error_str, sizeof(req->error_str), 1643229997Sken "%s: error allocating UMA zone", __func__); 1644229997Sken goto bailout_error; 1645229997Sken } 1646229997Sken 1647229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1648229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1649229997Sken else 1650229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1651229997Sken 1652229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1653229997Sken for (i = 0; i < req->num_be_args; i++) { 1654249026Strasz if (strcmp(req->kern_be_args[i].kname, "file") == 0) { 1655229997Sken file_arg = &req->kern_be_args[i]; 1656229997Sken break; 1657229997Sken } 1658229997Sken } 1659229997Sken 1660229997Sken if (file_arg == NULL) { 1661229997Sken snprintf(req->error_str, sizeof(req->error_str), 1662229997Sken "%s: no file argument specified", __func__); 1663229997Sken goto bailout_error; 1664229997Sken } 1665229997Sken 1666229997Sken be_lun->dev_path = malloc(file_arg->vallen, M_CTLBLK, 1667229997Sken M_WAITOK | M_ZERO); 1668229997Sken 1669249026Strasz strlcpy(be_lun->dev_path, (char *)file_arg->kvalue, 1670229997Sken file_arg->vallen); 1671229997Sken 1672229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1673229997Sken if (retval != 0) { 1674229997Sken retval = 0; 1675229997Sken goto bailout_error; 1676229997Sken } 1677229997Sken 1678229997Sken /* 1679229997Sken * Tell the user the size of the file/device. 1680229997Sken */ 1681229997Sken params->lun_size_bytes = be_lun->size_bytes; 1682229997Sken 1683229997Sken /* 1684229997Sken * The maximum LBA is the size - 1. 1685229997Sken */ 1686229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1687229997Sken } else { 1688229997Sken /* 1689229997Sken * For processor devices, we don't have any size. 1690229997Sken */ 1691229997Sken be_lun->blocksize = 0; 1692229997Sken be_lun->size_blocks = 0; 1693229997Sken be_lun->size_bytes = 0; 1694229997Sken be_lun->ctl_be_lun.maxlba = 0; 1695229997Sken params->lun_size_bytes = 0; 1696229997Sken 1697229997Sken /* 1698229997Sken * Default to just 1 thread for processor devices. 1699229997Sken */ 1700229997Sken num_threads = 1; 1701229997Sken } 1702229997Sken 1703229997Sken /* 1704229997Sken * XXX This searching loop might be refactored to be combined with 1705229997Sken * the loop above, 1706229997Sken */ 1707229997Sken for (i = 0; i < req->num_be_args; i++) { 1708249026Strasz if (strcmp(req->kern_be_args[i].kname, "num_threads") == 0) { 1709229997Sken struct ctl_be_arg *thread_arg; 1710229997Sken char num_thread_str[16]; 1711229997Sken int tmp_num_threads; 1712229997Sken 1713229997Sken 1714229997Sken thread_arg = &req->kern_be_args[i]; 1715229997Sken 1716249026Strasz strlcpy(num_thread_str, (char *)thread_arg->kvalue, 1717229997Sken min(thread_arg->vallen, 1718229997Sken sizeof(num_thread_str))); 1719229997Sken 1720229997Sken tmp_num_threads = strtol(num_thread_str, NULL, 0); 1721229997Sken 1722229997Sken /* 1723229997Sken * We don't let the user specify less than one 1724229997Sken * thread, but hope he's clueful enough not to 1725229997Sken * specify 1000 threads. 1726229997Sken */ 1727229997Sken if (tmp_num_threads < 1) { 1728229997Sken snprintf(req->error_str, sizeof(req->error_str), 1729229997Sken "%s: invalid number of threads %s", 1730229997Sken __func__, num_thread_str); 1731229997Sken goto bailout_error; 1732229997Sken } 1733229997Sken 1734229997Sken num_threads = tmp_num_threads; 1735254759Strasz } else if (strcmp(req->kern_be_args[i].kname, "file") != 0 && 1736254759Strasz strcmp(req->kern_be_args[i].kname, "dev") != 0) { 1737254759Strasz struct ctl_be_lun_option *opt; 1738254759Strasz 1739254759Strasz opt = malloc(sizeof(*opt), M_CTLBLK, M_WAITOK); 1740254759Strasz opt->name = malloc(strlen(req->kern_be_args[i].kname) + 1, M_CTLBLK, M_WAITOK); 1741254759Strasz strcpy(opt->name, req->kern_be_args[i].kname); 1742254759Strasz opt->value = malloc(strlen(req->kern_be_args[i].kvalue) + 1, M_CTLBLK, M_WAITOK); 1743254759Strasz strcpy(opt->value, req->kern_be_args[i].kvalue); 1744254759Strasz STAILQ_INSERT_TAIL(&be_lun->ctl_be_lun.options, opt, links); 1745229997Sken } 1746229997Sken } 1747229997Sken 1748229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1749229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1750229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 1751229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1752229997Sken /* Tell the user the blocksize we ended up using */ 1753229997Sken params->blocksize_bytes = be_lun->blocksize; 1754229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1755229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1756229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1757229997Sken } else 1758229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 1759229997Sken 1760229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1761229997Sken be_lun->ctl_be_lun.lun_config_status = 1762229997Sken ctl_be_block_lun_config_status; 1763229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1764229997Sken 1765229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1766229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1767229997Sken softc->num_luns); 1768229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1769229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1770229997Sken sizeof(tmpstr))); 1771229997Sken 1772229997Sken /* Tell the user what we used for a serial number */ 1773229997Sken strncpy((char *)params->serial_num, tmpstr, 1774229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1775229997Sken } else { 1776229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 1777229997Sken params->serial_num, 1778229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1779229997Sken sizeof(params->serial_num))); 1780229997Sken } 1781229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1782229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1783229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1784229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1785229997Sken sizeof(tmpstr))); 1786229997Sken 1787229997Sken /* Tell the user what we used for a device ID */ 1788229997Sken strncpy((char *)params->device_id, tmpstr, 1789229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1790229997Sken } else { 1791229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 1792229997Sken params->device_id, 1793229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1794229997Sken sizeof(params->device_id))); 1795229997Sken } 1796229997Sken 1797229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 1798229997Sken 1799229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 1800229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 1801229997Sken 1802229997Sken if (be_lun->io_taskqueue == NULL) { 1803229997Sken snprintf(req->error_str, sizeof(req->error_str), 1804229997Sken "%s: Unable to create taskqueue", __func__); 1805229997Sken goto bailout_error; 1806229997Sken } 1807229997Sken 1808229997Sken /* 1809229997Sken * Note that we start the same number of threads by default for 1810229997Sken * both the file case and the block device case. For the file 1811229997Sken * case, we need multiple threads to allow concurrency, because the 1812229997Sken * vnode interface is designed to be a blocking interface. For the 1813229997Sken * block device case, ZFS zvols at least will block the caller's 1814229997Sken * context in many instances, and so we need multiple threads to 1815229997Sken * overcome that problem. Other block devices don't need as many 1816229997Sken * threads, but they shouldn't cause too many problems. 1817229997Sken * 1818229997Sken * If the user wants to just have a single thread for a block 1819229997Sken * device, he can specify that when the LUN is created, or change 1820229997Sken * the tunable/sysctl to alter the default number of threads. 1821229997Sken */ 1822229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 1823229997Sken /*num threads*/num_threads, 1824229997Sken /*priority*/PWAIT, 1825229997Sken /*thread name*/ 1826229997Sken "%s taskq", be_lun->lunname); 1827229997Sken 1828229997Sken if (retval != 0) 1829229997Sken goto bailout_error; 1830229997Sken 1831229997Sken be_lun->num_threads = num_threads; 1832229997Sken 1833229997Sken mtx_lock(&softc->lock); 1834229997Sken softc->num_luns++; 1835229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 1836229997Sken 1837229997Sken mtx_unlock(&softc->lock); 1838229997Sken 1839229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 1840229997Sken if (retval != 0) { 1841229997Sken mtx_lock(&softc->lock); 1842229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1843229997Sken links); 1844229997Sken softc->num_luns--; 1845229997Sken mtx_unlock(&softc->lock); 1846229997Sken snprintf(req->error_str, sizeof(req->error_str), 1847229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 1848229997Sken "details", __func__, retval); 1849229997Sken retval = 0; 1850229997Sken goto bailout_error; 1851229997Sken } 1852229997Sken 1853229997Sken mtx_lock(&softc->lock); 1854229997Sken 1855229997Sken /* 1856229997Sken * Tell the config_status routine that we're waiting so it won't 1857229997Sken * clean up the LUN in the event of an error. 1858229997Sken */ 1859229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1860229997Sken 1861229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 1862229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1863229997Sken if (retval == EINTR) 1864229997Sken break; 1865229997Sken } 1866229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1867229997Sken 1868229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 1869229997Sken snprintf(req->error_str, sizeof(req->error_str), 1870229997Sken "%s: LUN configuration error, see dmesg for details", 1871229997Sken __func__); 1872229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1873229997Sken links); 1874229997Sken softc->num_luns--; 1875229997Sken mtx_unlock(&softc->lock); 1876229997Sken goto bailout_error; 1877229997Sken } else { 1878229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 1879229997Sken } 1880229997Sken 1881229997Sken mtx_unlock(&softc->lock); 1882229997Sken 1883229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 1884229997Sken be_lun->blocksize, 1885229997Sken DEVSTAT_ALL_SUPPORTED, 1886229997Sken be_lun->ctl_be_lun.lun_type 1887229997Sken | DEVSTAT_TYPE_IF_OTHER, 1888229997Sken DEVSTAT_PRIORITY_OTHER); 1889229997Sken 1890229997Sken 1891229997Sken req->status = CTL_LUN_OK; 1892229997Sken 1893229997Sken return (retval); 1894229997Sken 1895229997Skenbailout_error: 1896229997Sken req->status = CTL_LUN_ERROR; 1897229997Sken 1898229997Sken ctl_be_block_close(be_lun); 1899229997Sken 1900229997Sken free(be_lun->dev_path, M_CTLBLK); 1901229997Sken free(be_lun, M_CTLBLK); 1902229997Sken 1903229997Sken return (retval); 1904229997Sken} 1905229997Sken 1906229997Skenstatic int 1907229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1908229997Sken{ 1909229997Sken struct ctl_lun_rm_params *params; 1910229997Sken struct ctl_be_block_lun *be_lun; 1911229997Sken int retval; 1912229997Sken 1913229997Sken params = &req->reqdata.rm; 1914229997Sken 1915229997Sken mtx_lock(&softc->lock); 1916229997Sken 1917229997Sken be_lun = NULL; 1918229997Sken 1919229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 1920229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 1921229997Sken break; 1922229997Sken } 1923229997Sken mtx_unlock(&softc->lock); 1924229997Sken 1925229997Sken if (be_lun == NULL) { 1926229997Sken snprintf(req->error_str, sizeof(req->error_str), 1927229997Sken "%s: LUN %u is not managed by the block backend", 1928229997Sken __func__, params->lun_id); 1929229997Sken goto bailout_error; 1930229997Sken } 1931229997Sken 1932229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 1933229997Sken 1934229997Sken if (retval != 0) { 1935229997Sken snprintf(req->error_str, sizeof(req->error_str), 1936229997Sken "%s: error %d returned from ctl_disable_lun() for " 1937229997Sken "LUN %d", __func__, retval, params->lun_id); 1938229997Sken goto bailout_error; 1939229997Sken 1940229997Sken } 1941229997Sken 1942229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 1943229997Sken if (retval != 0) { 1944229997Sken snprintf(req->error_str, sizeof(req->error_str), 1945229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 1946229997Sken "LUN %d", __func__, retval, params->lun_id); 1947229997Sken goto bailout_error; 1948229997Sken } 1949229997Sken 1950229997Sken mtx_lock(&softc->lock); 1951229997Sken 1952229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1953229997Sken 1954229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1955229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1956229997Sken if (retval == EINTR) 1957229997Sken break; 1958229997Sken } 1959229997Sken 1960229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1961229997Sken 1962229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1963229997Sken snprintf(req->error_str, sizeof(req->error_str), 1964229997Sken "%s: interrupted waiting for LUN to be freed", 1965229997Sken __func__); 1966229997Sken mtx_unlock(&softc->lock); 1967229997Sken goto bailout_error; 1968229997Sken } 1969229997Sken 1970229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 1971229997Sken 1972229997Sken softc->num_luns--; 1973229997Sken mtx_unlock(&softc->lock); 1974229997Sken 1975229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 1976229997Sken 1977229997Sken taskqueue_free(be_lun->io_taskqueue); 1978229997Sken 1979229997Sken ctl_be_block_close(be_lun); 1980229997Sken 1981229997Sken if (be_lun->disk_stats != NULL) 1982229997Sken devstat_remove_entry(be_lun->disk_stats); 1983229997Sken 1984229997Sken uma_zdestroy(be_lun->lun_zone); 1985229997Sken 1986229997Sken free(be_lun->dev_path, M_CTLBLK); 1987229997Sken 1988229997Sken free(be_lun, M_CTLBLK); 1989229997Sken 1990229997Sken req->status = CTL_LUN_OK; 1991229997Sken 1992229997Sken return (0); 1993229997Sken 1994229997Skenbailout_error: 1995229997Sken 1996229997Sken req->status = CTL_LUN_ERROR; 1997229997Sken 1998229997Sken return (0); 1999229997Sken} 2000229997Sken 2001232604Straszstatic int 2002232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2003232604Strasz struct ctl_lun_req *req) 2004232604Strasz{ 2005232604Strasz struct vattr vattr; 2006232604Strasz int error; 2007232604Strasz struct ctl_lun_modify_params *params; 2008232604Strasz 2009232604Strasz params = &req->reqdata.modify; 2010232604Strasz 2011232604Strasz if (params->lun_size_bytes != 0) { 2012232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2013232604Strasz } else { 2014232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2015232604Strasz if (error != 0) { 2016232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2017232604Strasz "error calling VOP_GETATTR() for file %s", 2018232604Strasz be_lun->dev_path); 2019232604Strasz return (error); 2020232604Strasz } 2021232604Strasz 2022232604Strasz be_lun->size_bytes = vattr.va_size; 2023232604Strasz } 2024232604Strasz 2025232604Strasz return (0); 2026232604Strasz} 2027232604Strasz 2028232604Straszstatic int 2029232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2030232604Strasz struct ctl_lun_req *req) 2031232604Strasz{ 2032232604Strasz struct cdev *dev; 2033232604Strasz struct cdevsw *devsw; 2034232604Strasz int error; 2035232604Strasz struct ctl_lun_modify_params *params; 2036232604Strasz uint64_t size_bytes; 2037232604Strasz 2038232604Strasz params = &req->reqdata.modify; 2039232604Strasz 2040232604Strasz dev = be_lun->vn->v_rdev; 2041232604Strasz devsw = dev->si_devsw; 2042232604Strasz if (!devsw->d_ioctl) { 2043232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2044232604Strasz "%s: no d_ioctl for device %s!", __func__, 2045232604Strasz be_lun->dev_path); 2046232604Strasz return (ENODEV); 2047232604Strasz } 2048232604Strasz 2049232604Strasz error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2050232604Strasz (caddr_t)&size_bytes, FREAD, 2051232604Strasz curthread); 2052232604Strasz if (error) { 2053232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2054232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2055232604Strasz "on %s!", __func__, error, be_lun->dev_path); 2056232604Strasz return (error); 2057232604Strasz } 2058232604Strasz 2059232604Strasz if (params->lun_size_bytes != 0) { 2060232604Strasz if (params->lun_size_bytes > size_bytes) { 2061232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2062232604Strasz "%s: requested LUN size %ju > backing device " 2063232604Strasz "size %ju", __func__, 2064232604Strasz (uintmax_t)params->lun_size_bytes, 2065232604Strasz (uintmax_t)size_bytes); 2066232604Strasz return (EINVAL); 2067232604Strasz } 2068232604Strasz 2069232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2070232604Strasz } else { 2071232604Strasz be_lun->size_bytes = size_bytes; 2072232604Strasz } 2073232604Strasz 2074232604Strasz return (0); 2075232604Strasz} 2076232604Strasz 2077232604Straszstatic int 2078232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2079232604Strasz{ 2080232604Strasz struct ctl_lun_modify_params *params; 2081232604Strasz struct ctl_be_block_lun *be_lun; 2082241896Skib int error; 2083232604Strasz 2084232604Strasz params = &req->reqdata.modify; 2085232604Strasz 2086232604Strasz mtx_lock(&softc->lock); 2087232604Strasz 2088232604Strasz be_lun = NULL; 2089232604Strasz 2090232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2091232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2092232604Strasz break; 2093232604Strasz } 2094232604Strasz mtx_unlock(&softc->lock); 2095232604Strasz 2096232604Strasz if (be_lun == NULL) { 2097232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2098232604Strasz "%s: LUN %u is not managed by the block backend", 2099232604Strasz __func__, params->lun_id); 2100232604Strasz goto bailout_error; 2101232604Strasz } 2102232604Strasz 2103232604Strasz if (params->lun_size_bytes != 0) { 2104232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2105232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2106232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2107232604Strasz params->lun_size_bytes, be_lun->blocksize); 2108232604Strasz goto bailout_error; 2109232604Strasz } 2110232604Strasz } 2111232604Strasz 2112232604Strasz vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2113232604Strasz 2114232604Strasz if (be_lun->vn->v_type == VREG) 2115232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2116232604Strasz else 2117232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2118232604Strasz 2119232604Strasz VOP_UNLOCK(be_lun->vn, 0); 2120232604Strasz 2121232604Strasz if (error != 0) 2122232604Strasz goto bailout_error; 2123232604Strasz 2124232604Strasz be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2125232604Strasz 2126232604Strasz /* 2127232604Strasz * The maximum LBA is the size - 1. 2128232604Strasz * 2129232604Strasz * XXX: Note that this field is being updated without locking, 2130232604Strasz * which might cause problems on 32-bit architectures. 2131232604Strasz */ 2132232604Strasz be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2133232604Strasz ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2134232604Strasz 2135232604Strasz /* Tell the user the exact size we ended up using */ 2136232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2137232604Strasz 2138232604Strasz req->status = CTL_LUN_OK; 2139232604Strasz 2140232604Strasz return (0); 2141232604Strasz 2142232604Straszbailout_error: 2143232604Strasz req->status = CTL_LUN_ERROR; 2144232604Strasz 2145232604Strasz return (0); 2146232604Strasz} 2147232604Strasz 2148229997Skenstatic void 2149229997Skenctl_be_block_lun_shutdown(void *be_lun) 2150229997Sken{ 2151229997Sken struct ctl_be_block_lun *lun; 2152229997Sken struct ctl_be_block_softc *softc; 2153229997Sken 2154229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2155229997Sken 2156229997Sken softc = lun->softc; 2157229997Sken 2158229997Sken mtx_lock(&softc->lock); 2159229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2160229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2161229997Sken wakeup(lun); 2162229997Sken mtx_unlock(&softc->lock); 2163229997Sken 2164229997Sken} 2165229997Sken 2166229997Skenstatic void 2167229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2168229997Sken{ 2169229997Sken struct ctl_be_block_lun *lun; 2170229997Sken struct ctl_be_block_softc *softc; 2171229997Sken 2172229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2173229997Sken softc = lun->softc; 2174229997Sken 2175229997Sken if (status == CTL_LUN_CONFIG_OK) { 2176229997Sken mtx_lock(&softc->lock); 2177229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2178229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2179229997Sken wakeup(lun); 2180229997Sken mtx_unlock(&softc->lock); 2181229997Sken 2182229997Sken /* 2183229997Sken * We successfully added the LUN, attempt to enable it. 2184229997Sken */ 2185229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2186229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2187229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2188229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2189229997Sken __func__); 2190229997Sken } 2191229997Sken } 2192229997Sken 2193229997Sken return; 2194229997Sken } 2195229997Sken 2196229997Sken 2197229997Sken mtx_lock(&softc->lock); 2198229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2199229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2200229997Sken wakeup(lun); 2201229997Sken mtx_unlock(&softc->lock); 2202229997Sken} 2203229997Sken 2204229997Sken 2205229997Skenstatic int 2206229997Skenctl_be_block_config_write(union ctl_io *io) 2207229997Sken{ 2208229997Sken struct ctl_be_block_lun *be_lun; 2209229997Sken struct ctl_be_lun *ctl_be_lun; 2210229997Sken int retval; 2211229997Sken 2212229997Sken retval = 0; 2213229997Sken 2214229997Sken DPRINTF("entered\n"); 2215229997Sken 2216229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2217229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2218229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2219229997Sken 2220229997Sken switch (io->scsiio.cdb[0]) { 2221229997Sken case SYNCHRONIZE_CACHE: 2222229997Sken case SYNCHRONIZE_CACHE_16: 2223229997Sken /* 2224229997Sken * The upper level CTL code will filter out any CDBs with 2225229997Sken * the immediate bit set and return the proper error. 2226229997Sken * 2227229997Sken * We don't really need to worry about what LBA range the 2228229997Sken * user asked to be synced out. When they issue a sync 2229229997Sken * cache command, we'll sync out the whole thing. 2230229997Sken */ 2231229997Sken mtx_lock(&be_lun->lock); 2232229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2233229997Sken links); 2234229997Sken mtx_unlock(&be_lun->lock); 2235229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2236229997Sken break; 2237229997Sken case START_STOP_UNIT: { 2238229997Sken struct scsi_start_stop_unit *cdb; 2239229997Sken 2240229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2241229997Sken 2242229997Sken if (cdb->how & SSS_START) 2243229997Sken retval = ctl_start_lun(ctl_be_lun); 2244229997Sken else { 2245229997Sken retval = ctl_stop_lun(ctl_be_lun); 2246229997Sken /* 2247229997Sken * XXX KDM Copan-specific offline behavior. 2248229997Sken * Figure out a reasonable way to port this? 2249229997Sken */ 2250229997Sken#ifdef NEEDTOPORT 2251229997Sken if ((retval == 0) 2252229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2253229997Sken retval = ctl_lun_offline(ctl_be_lun); 2254229997Sken#endif 2255229997Sken } 2256229997Sken 2257229997Sken /* 2258229997Sken * In general, the above routines should not fail. They 2259229997Sken * just set state for the LUN. So we've got something 2260229997Sken * pretty wrong here if we can't start or stop the LUN. 2261229997Sken */ 2262229997Sken if (retval != 0) { 2263229997Sken ctl_set_internal_failure(&io->scsiio, 2264229997Sken /*sks_valid*/ 1, 2265229997Sken /*retry_count*/ 0xf051); 2266229997Sken retval = CTL_RETVAL_COMPLETE; 2267229997Sken } else { 2268229997Sken ctl_set_success(&io->scsiio); 2269229997Sken } 2270229997Sken ctl_config_write_done(io); 2271229997Sken break; 2272229997Sken } 2273229997Sken default: 2274229997Sken ctl_set_invalid_opcode(&io->scsiio); 2275229997Sken ctl_config_write_done(io); 2276229997Sken retval = CTL_RETVAL_COMPLETE; 2277229997Sken break; 2278229997Sken } 2279229997Sken 2280229997Sken return (retval); 2281229997Sken 2282229997Sken} 2283229997Sken 2284229997Skenstatic int 2285229997Skenctl_be_block_config_read(union ctl_io *io) 2286229997Sken{ 2287229997Sken return (0); 2288229997Sken} 2289229997Sken 2290229997Skenstatic int 2291229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2292229997Sken{ 2293229997Sken struct ctl_be_block_lun *lun; 2294229997Sken int retval; 2295229997Sken 2296229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2297229997Sken retval = 0; 2298229997Sken 2299229997Sken retval = sbuf_printf(sb, "<num_threads>"); 2300229997Sken 2301229997Sken if (retval != 0) 2302229997Sken goto bailout; 2303229997Sken 2304229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2305229997Sken 2306229997Sken if (retval != 0) 2307229997Sken goto bailout; 2308229997Sken 2309229997Sken retval = sbuf_printf(sb, "</num_threads>"); 2310229997Sken 2311229997Sken /* 2312229997Sken * For processor devices, we don't have a path variable. 2313229997Sken */ 2314229997Sken if ((retval != 0) 2315229997Sken || (lun->dev_path == NULL)) 2316229997Sken goto bailout; 2317229997Sken 2318229997Sken retval = sbuf_printf(sb, "<file>"); 2319229997Sken 2320229997Sken if (retval != 0) 2321229997Sken goto bailout; 2322229997Sken 2323229997Sken retval = ctl_sbuf_printf_esc(sb, lun->dev_path); 2324229997Sken 2325229997Sken if (retval != 0) 2326229997Sken goto bailout; 2327229997Sken 2328229997Sken retval = sbuf_printf(sb, "</file>\n"); 2329229997Sken 2330229997Skenbailout: 2331229997Sken 2332229997Sken return (retval); 2333229997Sken} 2334229997Sken 2335229997Skenint 2336229997Skenctl_be_block_init(void) 2337229997Sken{ 2338229997Sken struct ctl_be_block_softc *softc; 2339229997Sken int retval; 2340229997Sken 2341229997Sken softc = &backend_block_softc; 2342229997Sken retval = 0; 2343229997Sken 2344229997Sken mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF); 2345229997Sken STAILQ_INIT(&softc->beio_free_queue); 2346229997Sken STAILQ_INIT(&softc->disk_list); 2347229997Sken STAILQ_INIT(&softc->lun_list); 2348229997Sken ctl_grow_beio(softc, 200); 2349229997Sken 2350229997Sken return (retval); 2351229997Sken} 2352