ctl_backend_block.c revision 258871
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 258871 2013-12-03 18:04:14Z trasz $"); 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; 513229997Sken 514229997Sken beio = bio->bio_caller1; 515229997Sken be_lun = beio->lun; 516229997Sken io = beio->io; 517229997Sken 518229997Sken DPRINTF("entered\n"); 519229997Sken 520229997Sken mtx_lock(&be_lun->lock); 521229997Sken if (bio->bio_error != 0) 522229997Sken beio->num_errors++; 523229997Sken 524229997Sken beio->num_bios_done++; 525229997Sken 526229997Sken /* 527229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 528229997Sken * during the free might cause it to complain. 529229997Sken */ 530229997Sken g_destroy_bio(bio); 531229997Sken 532229997Sken /* 533229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 534229997Sken * complete, then we're done. 535229997Sken */ 536229997Sken if ((beio->send_complete == 0) 537229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 538229997Sken mtx_unlock(&be_lun->lock); 539229997Sken return; 540229997Sken } 541229997Sken 542229997Sken /* 543229997Sken * At this point, we've verified that we are the last I/O to 544229997Sken * complete, so it's safe to drop the lock. 545229997Sken */ 546229997Sken mtx_unlock(&be_lun->lock); 547229997Sken 548229997Sken /* 549229997Sken * If there are any errors from the backing device, we fail the 550229997Sken * entire I/O with a medium error. 551229997Sken */ 552229997Sken if (beio->num_errors > 0) { 553229997Sken if (beio->bio_cmd == BIO_FLUSH) { 554229997Sken /* XXX KDM is there is a better error here? */ 555229997Sken ctl_set_internal_failure(&io->scsiio, 556229997Sken /*sks_valid*/ 1, 557229997Sken /*retry_count*/ 0xbad2); 558229997Sken } else 559229997Sken ctl_set_medium_error(&io->scsiio); 560229997Sken ctl_complete_beio(beio); 561229997Sken return; 562229997Sken } 563229997Sken 564229997Sken /* 565229997Sken * If this is a write or a flush, we're all done. 566229997Sken * If this is a read, we can now send the data to the user. 567229997Sken */ 568229997Sken if ((beio->bio_cmd == BIO_WRITE) 569229997Sken || (beio->bio_cmd == BIO_FLUSH)) { 570229997Sken ctl_set_success(&io->scsiio); 571229997Sken ctl_complete_beio(beio); 572229997Sken } else { 573229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 574229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 575229997Sken io->scsiio.kern_data_len = beio->io_len; 576229997Sken io->scsiio.kern_total_len = beio->io_len; 577229997Sken io->scsiio.kern_rel_offset = 0; 578229997Sken io->scsiio.kern_data_resid = 0; 579229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 580229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 581229997Sken#ifdef CTL_TIME_IO 582229997Sken getbintime(&io->io_hdr.dma_start_bt); 583229997Sken#endif 584229997Sken ctl_datamove(io); 585229997Sken } 586229997Sken} 587229997Sken 588229997Skenstatic void 589229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 590229997Sken struct ctl_be_block_io *beio) 591229997Sken{ 592229997Sken union ctl_io *io; 593229997Sken struct mount *mountpoint; 594241896Skib int error, lock_flags; 595229997Sken 596229997Sken DPRINTF("entered\n"); 597229997Sken 598229997Sken io = beio->io; 599229997Sken 600229997Sken (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 601229997Sken 602229997Sken if (MNT_SHARED_WRITES(mountpoint) 603229997Sken || ((mountpoint == NULL) 604229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 605229997Sken lock_flags = LK_SHARED; 606229997Sken else 607229997Sken lock_flags = LK_EXCLUSIVE; 608229997Sken 609229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 610229997Sken 611229997Sken binuptime(&beio->ds_t0); 612229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 613229997Sken 614229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 615229997Sken VOP_UNLOCK(be_lun->vn, 0); 616229997Sken 617229997Sken vn_finished_write(mountpoint); 618229997Sken 619229997Sken if (error == 0) 620229997Sken ctl_set_success(&io->scsiio); 621229997Sken else { 622229997Sken /* XXX KDM is there is a better error here? */ 623229997Sken ctl_set_internal_failure(&io->scsiio, 624229997Sken /*sks_valid*/ 1, 625229997Sken /*retry_count*/ 0xbad1); 626229997Sken } 627229997Sken 628229997Sken ctl_complete_beio(beio); 629229997Sken} 630229997Sken 631258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 632258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 633258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 634258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 635229997Sken 636229997Skenstatic void 637229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 638229997Sken struct ctl_be_block_io *beio) 639229997Sken{ 640229997Sken struct ctl_be_block_filedata *file_data; 641229997Sken union ctl_io *io; 642229997Sken struct uio xuio; 643229997Sken struct iovec *xiovec; 644241896Skib int flags; 645229997Sken int error, i; 646229997Sken 647229997Sken DPRINTF("entered\n"); 648229997Sken 649229997Sken file_data = &be_lun->backend.file; 650229997Sken io = beio->io; 651229997Sken flags = beio->bio_flags; 652229997Sken 653229997Sken if (beio->bio_cmd == BIO_READ) { 654229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 655229997Sken } else { 656229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 657229997Sken } 658229997Sken 659229997Sken bzero(&xuio, sizeof(xuio)); 660229997Sken if (beio->bio_cmd == BIO_READ) 661229997Sken xuio.uio_rw = UIO_READ; 662229997Sken else 663229997Sken xuio.uio_rw = UIO_WRITE; 664229997Sken 665229997Sken xuio.uio_offset = beio->io_offset; 666229997Sken xuio.uio_resid = beio->io_len; 667229997Sken xuio.uio_segflg = UIO_SYSSPACE; 668229997Sken xuio.uio_iov = beio->xiovecs; 669229997Sken xuio.uio_iovcnt = beio->num_segs; 670229997Sken xuio.uio_td = curthread; 671229997Sken 672229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 673229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 674229997Sken xiovec->iov_len = beio->sg_segs[i].len; 675229997Sken } 676229997Sken 677229997Sken if (beio->bio_cmd == BIO_READ) { 678229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 679229997Sken 680229997Sken binuptime(&beio->ds_t0); 681229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 682229997Sken 683229997Sken /* 684229997Sken * UFS pays attention to IO_DIRECT for reads. If the 685229997Sken * DIRECTIO option is configured into the kernel, it calls 686229997Sken * ffs_rawread(). But that only works for single-segment 687229997Sken * uios with user space addresses. In our case, with a 688229997Sken * kernel uio, it still reads into the buffer cache, but it 689229997Sken * will just try to release the buffer from the cache later 690229997Sken * on in ffs_read(). 691229997Sken * 692229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 693229997Sken * 694229997Sken * UFS does not pay attention to IO_SYNC for reads. 695229997Sken * 696229997Sken * ZFS pays attention to IO_SYNC (which translates into the 697229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 698229997Sken * attempts to sync the file before reading. 699229997Sken * 700229997Sken * So, to attempt to provide some barrier semantics in the 701229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 702229997Sken */ 703229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 704229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 705229997Sken 706229997Sken VOP_UNLOCK(be_lun->vn, 0); 707229997Sken } else { 708229997Sken struct mount *mountpoint; 709229997Sken int lock_flags; 710229997Sken 711229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 712229997Sken 713229997Sken if (MNT_SHARED_WRITES(mountpoint) 714229997Sken || ((mountpoint == NULL) 715229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 716229997Sken lock_flags = LK_SHARED; 717229997Sken else 718229997Sken lock_flags = LK_EXCLUSIVE; 719229997Sken 720229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 721229997Sken 722229997Sken binuptime(&beio->ds_t0); 723229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 724229997Sken 725229997Sken /* 726229997Sken * UFS pays attention to IO_DIRECT for writes. The write 727229997Sken * is done asynchronously. (Normally the write would just 728229997Sken * get put into cache. 729229997Sken * 730229997Sken * UFS pays attention to IO_SYNC for writes. It will 731229997Sken * attempt to write the buffer out synchronously if that 732229997Sken * flag is set. 733229997Sken * 734229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 735229997Sken * 736229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 737229997Sken * for writes. It will flush the transaction from the 738229997Sken * cache before returning. 739229997Sken * 740229997Sken * So if we've got the BIO_ORDERED flag set, we want 741229997Sken * IO_SYNC in either the UFS or ZFS case. 742229997Sken */ 743229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 744229997Sken IO_SYNC : 0, file_data->cred); 745229997Sken VOP_UNLOCK(be_lun->vn, 0); 746229997Sken 747229997Sken vn_finished_write(mountpoint); 748229997Sken } 749229997Sken 750229997Sken /* 751229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 752229997Sken * return the I/O to the user. 753229997Sken */ 754229997Sken if (error != 0) { 755229997Sken char path_str[32]; 756229997Sken 757229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 758229997Sken /* 759229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 760229997Sken * filesystem fills up. What kind of SCSI error should we 761229997Sken * return for that? 762229997Sken */ 763229997Sken printf("%s%s command returned errno %d\n", path_str, 764229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 765229997Sken ctl_set_medium_error(&io->scsiio); 766229997Sken ctl_complete_beio(beio); 767229997Sken return; 768229997Sken } 769229997Sken 770229997Sken /* 771229997Sken * If this is a write, we're all done. 772229997Sken * If this is a read, we can now send the data to the user. 773229997Sken */ 774229997Sken if (beio->bio_cmd == BIO_WRITE) { 775229997Sken ctl_set_success(&io->scsiio); 776229997Sken SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 777229997Sken ctl_complete_beio(beio); 778229997Sken } else { 779229997Sken SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 780229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 781229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 782229997Sken io->scsiio.kern_data_len = beio->io_len; 783229997Sken io->scsiio.kern_total_len = beio->io_len; 784229997Sken io->scsiio.kern_rel_offset = 0; 785229997Sken io->scsiio.kern_data_resid = 0; 786229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 787229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 788229997Sken#ifdef CTL_TIME_IO 789229997Sken getbintime(&io->io_hdr.dma_start_bt); 790229997Sken#endif 791229997Sken ctl_datamove(io); 792229997Sken } 793229997Sken} 794229997Sken 795229997Skenstatic void 796229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 797229997Sken struct ctl_be_block_io *beio) 798229997Sken{ 799229997Sken struct bio *bio; 800229997Sken union ctl_io *io; 801229997Sken struct ctl_be_block_devdata *dev_data; 802229997Sken 803229997Sken dev_data = &be_lun->backend.dev; 804229997Sken io = beio->io; 805229997Sken 806229997Sken DPRINTF("entered\n"); 807229997Sken 808229997Sken /* This can't fail, it's a blocking allocation. */ 809229997Sken bio = g_alloc_bio(); 810229997Sken 811229997Sken bio->bio_cmd = BIO_FLUSH; 812229997Sken bio->bio_flags |= BIO_ORDERED; 813229997Sken bio->bio_dev = dev_data->cdev; 814229997Sken bio->bio_offset = 0; 815229997Sken bio->bio_data = 0; 816229997Sken bio->bio_done = ctl_be_block_biodone; 817229997Sken bio->bio_caller1 = beio; 818229997Sken bio->bio_pblkno = 0; 819229997Sken 820229997Sken /* 821229997Sken * We don't need to acquire the LUN lock here, because we are only 822229997Sken * sending one bio, and so there is no other context to synchronize 823229997Sken * with. 824229997Sken */ 825229997Sken beio->num_bios_sent = 1; 826229997Sken beio->send_complete = 1; 827229997Sken 828229997Sken binuptime(&beio->ds_t0); 829229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 830229997Sken 831229997Sken (*dev_data->csw->d_strategy)(bio); 832229997Sken} 833229997Sken 834229997Skenstatic void 835229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 836229997Sken struct ctl_be_block_io *beio) 837229997Sken{ 838229997Sken int i; 839229997Sken struct bio *bio; 840229997Sken struct ctl_be_block_devdata *dev_data; 841229997Sken off_t cur_offset; 842229997Sken int max_iosize; 843229997Sken 844229997Sken DPRINTF("entered\n"); 845229997Sken 846229997Sken dev_data = &be_lun->backend.dev; 847229997Sken 848229997Sken /* 849229997Sken * We have to limit our I/O size to the maximum supported by the 850229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 851229997Sken * set it properly, use DFLTPHYS. 852229997Sken */ 853229997Sken max_iosize = dev_data->cdev->si_iosize_max; 854229997Sken if (max_iosize < PAGE_SIZE) 855229997Sken max_iosize = DFLTPHYS; 856229997Sken 857229997Sken cur_offset = beio->io_offset; 858229997Sken 859229997Sken /* 860229997Sken * XXX KDM need to accurately reflect the number of I/Os outstanding 861229997Sken * to a device. 862229997Sken */ 863229997Sken binuptime(&beio->ds_t0); 864229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 865229997Sken 866229997Sken for (i = 0; i < beio->num_segs; i++) { 867229997Sken size_t cur_size; 868229997Sken uint8_t *cur_ptr; 869229997Sken 870229997Sken cur_size = beio->sg_segs[i].len; 871229997Sken cur_ptr = beio->sg_segs[i].addr; 872229997Sken 873229997Sken while (cur_size > 0) { 874229997Sken /* This can't fail, it's a blocking allocation. */ 875229997Sken bio = g_alloc_bio(); 876229997Sken 877229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 878229997Sken 879229997Sken bio->bio_cmd = beio->bio_cmd; 880229997Sken bio->bio_flags |= beio->bio_flags; 881229997Sken bio->bio_dev = dev_data->cdev; 882229997Sken bio->bio_caller1 = beio; 883229997Sken bio->bio_length = min(cur_size, max_iosize); 884229997Sken bio->bio_offset = cur_offset; 885229997Sken bio->bio_data = cur_ptr; 886229997Sken bio->bio_done = ctl_be_block_biodone; 887229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 888229997Sken 889229997Sken cur_offset += bio->bio_length; 890229997Sken cur_ptr += bio->bio_length; 891229997Sken cur_size -= bio->bio_length; 892229997Sken 893229997Sken /* 894229997Sken * Make sure we set the complete bit just before we 895229997Sken * issue the last bio so we don't wind up with a 896229997Sken * race. 897229997Sken * 898229997Sken * Use the LUN mutex here instead of a combination 899229997Sken * of atomic variables for simplicity. 900229997Sken * 901229997Sken * XXX KDM we could have a per-IO lock, but that 902229997Sken * would cause additional per-IO setup and teardown 903229997Sken * overhead. Hopefully there won't be too much 904229997Sken * contention on the LUN lock. 905229997Sken */ 906229997Sken mtx_lock(&be_lun->lock); 907229997Sken 908229997Sken beio->num_bios_sent++; 909229997Sken 910229997Sken if ((i == beio->num_segs - 1) 911229997Sken && (cur_size == 0)) 912229997Sken beio->send_complete = 1; 913229997Sken 914229997Sken mtx_unlock(&be_lun->lock); 915229997Sken 916229997Sken (*dev_data->csw->d_strategy)(bio); 917229997Sken } 918229997Sken } 919229997Sken} 920229997Sken 921229997Skenstatic void 922229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 923229997Sken union ctl_io *io) 924229997Sken{ 925229997Sken struct ctl_be_block_io *beio; 926229997Sken struct ctl_be_block_softc *softc; 927229997Sken 928229997Sken DPRINTF("entered\n"); 929229997Sken 930229997Sken softc = be_lun->softc; 931229997Sken beio = ctl_alloc_beio(softc); 932229997Sken if (beio == NULL) { 933229997Sken /* 934229997Sken * This should not happen. ctl_alloc_beio() will call 935229997Sken * ctl_grow_beio() with a blocking malloc as needed. 936229997Sken * A malloc with M_WAITOK should not fail. 937229997Sken */ 938229997Sken ctl_set_busy(&io->scsiio); 939229997Sken ctl_done(io); 940229997Sken return; 941229997Sken } 942229997Sken 943229997Sken beio->io = io; 944229997Sken beio->softc = softc; 945229997Sken beio->lun = be_lun; 946229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 947229997Sken 948229997Sken switch (io->scsiio.cdb[0]) { 949229997Sken case SYNCHRONIZE_CACHE: 950229997Sken case SYNCHRONIZE_CACHE_16: 951249194Strasz beio->bio_cmd = BIO_FLUSH; 952229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 953229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 954229997Sken beio->io_len = 0; 955229997Sken be_lun->lun_flush(be_lun, beio); 956229997Sken break; 957229997Sken default: 958229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 959229997Sken break; 960229997Sken } 961229997Sken} 962229997Sken 963258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 964258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 965258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 966258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 967229997Sken 968229997Skenstatic void 969229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 970229997Sken union ctl_io *io) 971229997Sken{ 972229997Sken struct ctl_be_block_io *beio; 973229997Sken struct ctl_be_block_softc *softc; 974229997Sken struct ctl_lba_len lbalen; 975229997Sken uint64_t len_left, io_size_bytes; 976229997Sken int i; 977229997Sken 978229997Sken softc = be_lun->softc; 979229997Sken 980229997Sken DPRINTF("entered\n"); 981229997Sken 982229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 983229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 984229997Sken } else { 985229997Sken SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 986229997Sken } 987229997Sken 988229997Sken memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 989229997Sken sizeof(lbalen)); 990229997Sken 991229997Sken io_size_bytes = lbalen.len * be_lun->blocksize; 992229997Sken 993229997Sken /* 994229997Sken * XXX KDM this is temporary, until we implement chaining of beio 995229997Sken * structures and multiple datamove calls to move all the data in 996229997Sken * or out. 997229997Sken */ 998229997Sken if (io_size_bytes > CTLBLK_MAX_IO_SIZE) { 999229997Sken printf("%s: IO length %ju > max io size %u\n", __func__, 1000229997Sken io_size_bytes, CTLBLK_MAX_IO_SIZE); 1001229997Sken ctl_set_invalid_field(&io->scsiio, 1002229997Sken /*sks_valid*/ 0, 1003229997Sken /*command*/ 1, 1004229997Sken /*field*/ 0, 1005229997Sken /*bit_valid*/ 0, 1006229997Sken /*bit*/ 0); 1007229997Sken ctl_done(io); 1008229997Sken return; 1009229997Sken } 1010229997Sken 1011229997Sken beio = ctl_alloc_beio(softc); 1012229997Sken if (beio == NULL) { 1013229997Sken /* 1014229997Sken * This should not happen. ctl_alloc_beio() will call 1015229997Sken * ctl_grow_beio() with a blocking malloc as needed. 1016229997Sken * A malloc with M_WAITOK should not fail. 1017229997Sken */ 1018229997Sken ctl_set_busy(&io->scsiio); 1019229997Sken ctl_done(io); 1020229997Sken return; 1021229997Sken } 1022229997Sken 1023229997Sken beio->io = io; 1024229997Sken beio->softc = softc; 1025229997Sken beio->lun = be_lun; 1026229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 1027229997Sken 1028229997Sken /* 1029229997Sken * If the I/O came down with an ordered or head of queue tag, set 1030229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 1031229997Sken * pretty much the best we can do. 1032229997Sken * 1033229997Sken * XXX KDM we don't have a great way to easily know about the FUA 1034229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 1035229997Sken * pass that knowledge to the backend), and in any case we would 1036229997Sken * need to determine how to handle it. 1037229997Sken */ 1038229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1039229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1040229997Sken beio->bio_flags = BIO_ORDERED; 1041229997Sken 1042229997Sken switch (io->scsiio.tag_type) { 1043229997Sken case CTL_TAG_ORDERED: 1044229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1045229997Sken break; 1046229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1047229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1048229997Sken break; 1049229997Sken case CTL_TAG_UNTAGGED: 1050229997Sken case CTL_TAG_SIMPLE: 1051229997Sken case CTL_TAG_ACA: 1052229997Sken default: 1053229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1054229997Sken break; 1055229997Sken } 1056229997Sken 1057229997Sken /* 1058229997Sken * This path handles read and write only. The config write path 1059229997Sken * handles flush operations. 1060229997Sken */ 1061229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 1062229997Sken beio->bio_cmd = BIO_READ; 1063229997Sken beio->ds_trans_type = DEVSTAT_READ; 1064229997Sken } else { 1065229997Sken beio->bio_cmd = BIO_WRITE; 1066229997Sken beio->ds_trans_type = DEVSTAT_WRITE; 1067229997Sken } 1068229997Sken 1069229997Sken beio->io_len = lbalen.len * be_lun->blocksize; 1070229997Sken beio->io_offset = lbalen.lba * be_lun->blocksize; 1071229997Sken 1072229997Sken DPRINTF("%s at LBA %jx len %u\n", 1073229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1074229997Sken (uintmax_t)lbalen.lba, lbalen.len); 1075229997Sken 1076229997Sken for (i = 0, len_left = io_size_bytes; i < CTLBLK_MAX_SEGS && 1077229997Sken len_left > 0; i++) { 1078229997Sken 1079229997Sken /* 1080229997Sken * Setup the S/G entry for this chunk. 1081229997Sken */ 1082229997Sken beio->sg_segs[i].len = min(MAXPHYS, len_left); 1083229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1084229997Sken 1085229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1086229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1087229997Sken 1088229997Sken beio->num_segs++; 1089229997Sken len_left -= beio->sg_segs[i].len; 1090229997Sken } 1091229997Sken 1092229997Sken /* 1093229997Sken * For the read case, we need to read the data into our buffers and 1094229997Sken * then we can send it back to the user. For the write case, we 1095229997Sken * need to get the data from the user first. 1096229997Sken */ 1097229997Sken if (beio->bio_cmd == BIO_READ) { 1098229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1099229997Sken be_lun->dispatch(be_lun, beio); 1100229997Sken } else { 1101229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1102229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 1103229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1104229997Sken io->scsiio.kern_data_len = beio->io_len; 1105229997Sken io->scsiio.kern_total_len = beio->io_len; 1106229997Sken io->scsiio.kern_rel_offset = 0; 1107229997Sken io->scsiio.kern_data_resid = 0; 1108229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 1109229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1110229997Sken#ifdef CTL_TIME_IO 1111229997Sken getbintime(&io->io_hdr.dma_start_bt); 1112229997Sken#endif 1113229997Sken ctl_datamove(io); 1114229997Sken } 1115229997Sken} 1116229997Sken 1117229997Skenstatic void 1118229997Skenctl_be_block_worker(void *context, int pending) 1119229997Sken{ 1120229997Sken struct ctl_be_block_lun *be_lun; 1121229997Sken struct ctl_be_block_softc *softc; 1122229997Sken union ctl_io *io; 1123229997Sken 1124229997Sken be_lun = (struct ctl_be_block_lun *)context; 1125229997Sken softc = be_lun->softc; 1126229997Sken 1127229997Sken DPRINTF("entered\n"); 1128229997Sken 1129229997Sken mtx_lock(&be_lun->lock); 1130229997Sken for (;;) { 1131229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1132229997Sken if (io != NULL) { 1133229997Sken struct ctl_be_block_io *beio; 1134229997Sken 1135229997Sken DPRINTF("datamove queue\n"); 1136229997Sken 1137229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1138229997Sken ctl_io_hdr, links); 1139229997Sken 1140229997Sken mtx_unlock(&be_lun->lock); 1141229997Sken 1142229997Sken beio = (struct ctl_be_block_io *) 1143229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1144229997Sken 1145229997Sken be_lun->dispatch(be_lun, beio); 1146229997Sken 1147229997Sken mtx_lock(&be_lun->lock); 1148229997Sken continue; 1149229997Sken } 1150229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1151229997Sken if (io != NULL) { 1152229997Sken 1153229997Sken DPRINTF("config write queue\n"); 1154229997Sken 1155229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1156229997Sken ctl_io_hdr, links); 1157229997Sken 1158229997Sken mtx_unlock(&be_lun->lock); 1159229997Sken 1160229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1161229997Sken 1162229997Sken mtx_lock(&be_lun->lock); 1163229997Sken continue; 1164229997Sken } 1165229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1166229997Sken if (io != NULL) { 1167229997Sken DPRINTF("input queue\n"); 1168229997Sken 1169229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1170229997Sken ctl_io_hdr, links); 1171229997Sken mtx_unlock(&be_lun->lock); 1172229997Sken 1173229997Sken /* 1174229997Sken * We must drop the lock, since this routine and 1175229997Sken * its children may sleep. 1176229997Sken */ 1177229997Sken ctl_be_block_dispatch(be_lun, io); 1178229997Sken 1179229997Sken mtx_lock(&be_lun->lock); 1180229997Sken continue; 1181229997Sken } 1182229997Sken 1183229997Sken /* 1184229997Sken * If we get here, there is no work left in the queues, so 1185229997Sken * just break out and let the task queue go to sleep. 1186229997Sken */ 1187229997Sken break; 1188229997Sken } 1189229997Sken mtx_unlock(&be_lun->lock); 1190229997Sken} 1191229997Sken 1192229997Sken/* 1193229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1194229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1195229997Sken * thread. 1196229997Sken */ 1197229997Skenstatic int 1198229997Skenctl_be_block_submit(union ctl_io *io) 1199229997Sken{ 1200229997Sken struct ctl_be_block_lun *be_lun; 1201229997Sken struct ctl_be_lun *ctl_be_lun; 1202229997Sken int retval; 1203229997Sken 1204229997Sken DPRINTF("entered\n"); 1205229997Sken 1206229997Sken retval = CTL_RETVAL_COMPLETE; 1207229997Sken 1208229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1209229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1210229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1211229997Sken 1212229997Sken /* 1213229997Sken * Make sure we only get SCSI I/O. 1214229997Sken */ 1215229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1216229997Sken "%#x) encountered", io->io_hdr.io_type)); 1217229997Sken 1218229997Sken mtx_lock(&be_lun->lock); 1219229997Sken /* 1220229997Sken * XXX KDM make sure that links is okay to use at this point. 1221229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1222229997Sken * or deal with resource allocation here. 1223229997Sken */ 1224229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1225229997Sken mtx_unlock(&be_lun->lock); 1226229997Sken 1227229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1228229997Sken 1229229997Sken return (retval); 1230229997Sken} 1231229997Sken 1232229997Skenstatic int 1233229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1234229997Sken int flag, struct thread *td) 1235229997Sken{ 1236229997Sken struct ctl_be_block_softc *softc; 1237229997Sken int error; 1238229997Sken 1239229997Sken softc = &backend_block_softc; 1240229997Sken 1241229997Sken error = 0; 1242229997Sken 1243229997Sken switch (cmd) { 1244229997Sken case CTL_LUN_REQ: { 1245229997Sken struct ctl_lun_req *lun_req; 1246229997Sken 1247229997Sken lun_req = (struct ctl_lun_req *)addr; 1248229997Sken 1249229997Sken switch (lun_req->reqtype) { 1250229997Sken case CTL_LUNREQ_CREATE: 1251229997Sken error = ctl_be_block_create(softc, lun_req); 1252229997Sken break; 1253229997Sken case CTL_LUNREQ_RM: 1254229997Sken error = ctl_be_block_rm(softc, lun_req); 1255229997Sken break; 1256232604Strasz case CTL_LUNREQ_MODIFY: 1257232604Strasz error = ctl_be_block_modify(softc, lun_req); 1258232604Strasz break; 1259229997Sken default: 1260229997Sken lun_req->status = CTL_LUN_ERROR; 1261229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1262229997Sken "%s: invalid LUN request type %d", __func__, 1263229997Sken lun_req->reqtype); 1264229997Sken break; 1265229997Sken } 1266229997Sken break; 1267229997Sken } 1268229997Sken default: 1269229997Sken error = ENOTTY; 1270229997Sken break; 1271229997Sken } 1272229997Sken 1273229997Sken return (error); 1274229997Sken} 1275229997Sken 1276229997Skenstatic int 1277229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1278229997Sken{ 1279229997Sken struct ctl_be_block_filedata *file_data; 1280229997Sken struct ctl_lun_create_params *params; 1281229997Sken struct vattr vattr; 1282229997Sken int error; 1283229997Sken 1284229997Sken error = 0; 1285229997Sken file_data = &be_lun->backend.file; 1286229997Sken params = &req->reqdata.create; 1287229997Sken 1288229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1289229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1290229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1291229997Sken 1292229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1293229997Sken if (error != 0) { 1294229997Sken snprintf(req->error_str, sizeof(req->error_str), 1295229997Sken "error calling VOP_GETATTR() for file %s", 1296229997Sken be_lun->dev_path); 1297229997Sken return (error); 1298229997Sken } 1299229997Sken 1300229997Sken /* 1301229997Sken * Verify that we have the ability to upgrade to exclusive 1302229997Sken * access on this file so we can trap errors at open instead 1303229997Sken * of reporting them during first access. 1304229997Sken */ 1305229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1306229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1307229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1308229997Sken error = EBADF; 1309229997Sken snprintf(req->error_str, sizeof(req->error_str), 1310229997Sken "error locking file %s", be_lun->dev_path); 1311229997Sken return (error); 1312229997Sken } 1313229997Sken } 1314229997Sken 1315229997Sken 1316229997Sken file_data->cred = crhold(curthread->td_ucred); 1317232604Strasz if (params->lun_size_bytes != 0) 1318232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1319232604Strasz else 1320232604Strasz be_lun->size_bytes = vattr.va_size; 1321229997Sken /* 1322229997Sken * We set the multi thread flag for file operations because all 1323229997Sken * filesystems (in theory) are capable of allowing multiple readers 1324229997Sken * of a file at once. So we want to get the maximum possible 1325229997Sken * concurrency. 1326229997Sken */ 1327229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1328229997Sken 1329229997Sken /* 1330229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1331229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1332229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1333229997Sken * may not work with other OSes as well. So just export a sector 1334229997Sken * size of 512 bytes, which should work with any OS or 1335229997Sken * application. Since our backing is a file, any block size will 1336229997Sken * work fine for the backing store. 1337229997Sken */ 1338229997Sken#if 0 1339229997Sken be_lun->blocksize= vattr.va_blocksize; 1340229997Sken#endif 1341229997Sken if (params->blocksize_bytes != 0) 1342229997Sken be_lun->blocksize = params->blocksize_bytes; 1343229997Sken else 1344229997Sken be_lun->blocksize = 512; 1345229997Sken 1346229997Sken /* 1347229997Sken * Sanity check. The media size has to be at least one 1348229997Sken * sector long. 1349229997Sken */ 1350229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1351229997Sken error = EINVAL; 1352229997Sken snprintf(req->error_str, sizeof(req->error_str), 1353229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1354229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1355229997Sken } 1356229997Sken return (error); 1357229997Sken} 1358229997Sken 1359229997Skenstatic int 1360229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1361229997Sken{ 1362229997Sken struct ctl_lun_create_params *params; 1363229997Sken struct vattr vattr; 1364229997Sken struct cdev *dev; 1365229997Sken struct cdevsw *devsw; 1366229997Sken int error; 1367229997Sken 1368229997Sken params = &req->reqdata.create; 1369229997Sken 1370229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1371229997Sken be_lun->dispatch = ctl_be_block_dispatch_dev; 1372229997Sken be_lun->lun_flush = ctl_be_block_flush_dev; 1373229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1374229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1375229997Sken &be_lun->backend.dev.dev_ref); 1376229997Sken if (be_lun->backend.dev.csw == NULL) 1377229997Sken panic("Unable to retrieve device switch"); 1378229997Sken 1379229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1380229997Sken if (error) { 1381229997Sken snprintf(req->error_str, sizeof(req->error_str), 1382229997Sken "%s: error getting vnode attributes for device %s", 1383229997Sken __func__, be_lun->dev_path); 1384229997Sken return (error); 1385229997Sken } 1386229997Sken 1387229997Sken dev = be_lun->vn->v_rdev; 1388229997Sken devsw = dev->si_devsw; 1389229997Sken if (!devsw->d_ioctl) { 1390229997Sken snprintf(req->error_str, sizeof(req->error_str), 1391229997Sken "%s: no d_ioctl for device %s!", __func__, 1392229997Sken be_lun->dev_path); 1393229997Sken return (ENODEV); 1394229997Sken } 1395229997Sken 1396229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1397229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1398229997Sken curthread); 1399229997Sken if (error) { 1400229997Sken snprintf(req->error_str, sizeof(req->error_str), 1401229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1402229997Sken "on %s!", __func__, error, be_lun->dev_path); 1403229997Sken return (error); 1404229997Sken } 1405229997Sken 1406229997Sken /* 1407229997Sken * If the user has asked for a blocksize that is greater than the 1408229997Sken * backing device's blocksize, we can do it only if the blocksize 1409229997Sken * the user is asking for is an even multiple of the underlying 1410229997Sken * device's blocksize. 1411229997Sken */ 1412229997Sken if ((params->blocksize_bytes != 0) 1413229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1414229997Sken uint32_t bs_multiple, tmp_blocksize; 1415229997Sken 1416229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1417229997Sken 1418229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1419229997Sken 1420229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1421229997Sken be_lun->blocksize = params->blocksize_bytes; 1422229997Sken } else { 1423229997Sken snprintf(req->error_str, sizeof(req->error_str), 1424229997Sken "%s: requested blocksize %u is not an even " 1425229997Sken "multiple of backing device blocksize %u", 1426229997Sken __func__, params->blocksize_bytes, 1427229997Sken be_lun->blocksize); 1428229997Sken return (EINVAL); 1429229997Sken 1430229997Sken } 1431229997Sken } else if ((params->blocksize_bytes != 0) 1432229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1433229997Sken snprintf(req->error_str, sizeof(req->error_str), 1434229997Sken "%s: requested blocksize %u < backing device " 1435229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1436229997Sken be_lun->blocksize); 1437229997Sken return (EINVAL); 1438229997Sken } 1439229997Sken 1440229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1441229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1442229997Sken curthread); 1443229997Sken if (error) { 1444229997Sken snprintf(req->error_str, sizeof(req->error_str), 1445232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1446232604Strasz " ioctl on %s!", __func__, error, 1447232604Strasz be_lun->dev_path); 1448229997Sken return (error); 1449229997Sken } 1450229997Sken 1451232604Strasz if (params->lun_size_bytes != 0) { 1452232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1453232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1454232604Strasz "%s: requested LUN size %ju > backing device " 1455232604Strasz "size %ju", __func__, 1456232604Strasz (uintmax_t)params->lun_size_bytes, 1457232604Strasz (uintmax_t)be_lun->size_bytes); 1458232604Strasz return (EINVAL); 1459232604Strasz } 1460232604Strasz 1461232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1462232604Strasz } 1463232604Strasz 1464229997Sken return (0); 1465229997Sken} 1466229997Sken 1467229997Skenstatic int 1468229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1469229997Sken{ 1470229997Sken DROP_GIANT(); 1471229997Sken if (be_lun->vn) { 1472229997Sken int flags = FREAD | FWRITE; 1473229997Sken 1474229997Sken switch (be_lun->dev_type) { 1475229997Sken case CTL_BE_BLOCK_DEV: 1476229997Sken if (be_lun->backend.dev.csw) { 1477229997Sken dev_relthread(be_lun->backend.dev.cdev, 1478229997Sken be_lun->backend.dev.dev_ref); 1479229997Sken be_lun->backend.dev.csw = NULL; 1480229997Sken be_lun->backend.dev.cdev = NULL; 1481229997Sken } 1482229997Sken break; 1483229997Sken case CTL_BE_BLOCK_FILE: 1484229997Sken break; 1485229997Sken case CTL_BE_BLOCK_NONE: 1486258871Strasz break; 1487229997Sken default: 1488229997Sken panic("Unexpected backend type."); 1489229997Sken break; 1490229997Sken } 1491229997Sken 1492229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1493229997Sken be_lun->vn = NULL; 1494229997Sken 1495229997Sken switch (be_lun->dev_type) { 1496229997Sken case CTL_BE_BLOCK_DEV: 1497229997Sken break; 1498229997Sken case CTL_BE_BLOCK_FILE: 1499229997Sken if (be_lun->backend.file.cred != NULL) { 1500229997Sken crfree(be_lun->backend.file.cred); 1501229997Sken be_lun->backend.file.cred = NULL; 1502229997Sken } 1503229997Sken break; 1504229997Sken case CTL_BE_BLOCK_NONE: 1505258871Strasz break; 1506229997Sken default: 1507229997Sken panic("Unexpected backend type."); 1508229997Sken break; 1509229997Sken } 1510229997Sken } 1511229997Sken PICKUP_GIANT(); 1512229997Sken 1513229997Sken return (0); 1514229997Sken} 1515229997Sken 1516229997Skenstatic int 1517229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1518229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1519229997Sken{ 1520229997Sken struct nameidata nd; 1521229997Sken int flags; 1522229997Sken int error; 1523229997Sken 1524229997Sken /* 1525229997Sken * XXX KDM allow a read-only option? 1526229997Sken */ 1527229997Sken flags = FREAD | FWRITE; 1528229997Sken error = 0; 1529229997Sken 1530229997Sken if (rootvnode == NULL) { 1531229997Sken snprintf(req->error_str, sizeof(req->error_str), 1532229997Sken "%s: Root filesystem is not mounted", __func__); 1533229997Sken return (1); 1534229997Sken } 1535229997Sken 1536229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1537229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1538229997Sken VREF(rootvnode); 1539229997Sken } 1540229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1541229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1542229997Sken VREF(rootvnode); 1543229997Sken } 1544229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1545229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1546229997Sken VREF(rootvnode); 1547229997Sken } 1548229997Sken 1549229997Sken again: 1550229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1551229997Sken error = vn_open(&nd, &flags, 0, NULL); 1552229997Sken if (error) { 1553229997Sken /* 1554229997Sken * This is the only reasonable guess we can make as far as 1555229997Sken * path if the user doesn't give us a fully qualified path. 1556229997Sken * If they want to specify a file, they need to specify the 1557229997Sken * full path. 1558229997Sken */ 1559229997Sken if (be_lun->dev_path[0] != '/') { 1560229997Sken char *dev_path = "/dev/"; 1561229997Sken char *dev_name; 1562229997Sken 1563229997Sken /* Try adding device path at beginning of name */ 1564229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1565229997Sken + strlen(dev_path) + 1, 1566229997Sken M_CTLBLK, M_WAITOK); 1567229997Sken if (dev_name) { 1568229997Sken sprintf(dev_name, "%s%s", dev_path, 1569229997Sken be_lun->dev_path); 1570229997Sken free(be_lun->dev_path, M_CTLBLK); 1571229997Sken be_lun->dev_path = dev_name; 1572229997Sken goto again; 1573229997Sken } 1574229997Sken } 1575229997Sken snprintf(req->error_str, sizeof(req->error_str), 1576229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1577229997Sken return (error); 1578229997Sken } 1579229997Sken 1580229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1581229997Sken 1582229997Sken be_lun->vn = nd.ni_vp; 1583229997Sken 1584229997Sken /* We only support disks and files. */ 1585229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1586229997Sken error = ctl_be_block_open_dev(be_lun, req); 1587229997Sken } else if (be_lun->vn->v_type == VREG) { 1588229997Sken error = ctl_be_block_open_file(be_lun, req); 1589229997Sken } else { 1590229997Sken error = EINVAL; 1591229997Sken snprintf(req->error_str, sizeof(req->error_str), 1592258871Strasz "%s is not a disk or plain file", be_lun->dev_path); 1593229997Sken } 1594229997Sken VOP_UNLOCK(be_lun->vn, 0); 1595229997Sken 1596229997Sken if (error != 0) { 1597229997Sken ctl_be_block_close(be_lun); 1598229997Sken return (error); 1599229997Sken } 1600229997Sken 1601229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1602229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1603229997Sken 1604229997Sken return (0); 1605229997Sken} 1606229997Sken 1607229997Skenstatic int 1608229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1609229997Sken{ 1610229997Sken struct ctl_be_block_lun *be_lun; 1611229997Sken struct ctl_lun_create_params *params; 1612229997Sken struct ctl_be_arg *file_arg; 1613229997Sken char tmpstr[32]; 1614229997Sken int retval, num_threads; 1615229997Sken int i; 1616229997Sken 1617229997Sken params = &req->reqdata.create; 1618229997Sken retval = 0; 1619229997Sken 1620229997Sken num_threads = cbb_num_threads; 1621229997Sken 1622229997Sken file_arg = NULL; 1623229997Sken 1624229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1625229997Sken 1626229997Sken be_lun->softc = softc; 1627229997Sken STAILQ_INIT(&be_lun->input_queue); 1628229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1629229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1630254759Strasz STAILQ_INIT(&be_lun->ctl_be_lun.options); 1631229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1632229997Sken mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF); 1633229997Sken 1634229997Sken be_lun->lun_zone = uma_zcreate(be_lun->lunname, MAXPHYS, 1635256995Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1636229997Sken 1637229997Sken if (be_lun->lun_zone == NULL) { 1638229997Sken snprintf(req->error_str, sizeof(req->error_str), 1639229997Sken "%s: error allocating UMA zone", __func__); 1640229997Sken goto bailout_error; 1641229997Sken } 1642229997Sken 1643229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1644229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1645229997Sken else 1646229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1647229997Sken 1648229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1649229997Sken for (i = 0; i < req->num_be_args; i++) { 1650249026Strasz if (strcmp(req->kern_be_args[i].kname, "file") == 0) { 1651229997Sken file_arg = &req->kern_be_args[i]; 1652229997Sken break; 1653229997Sken } 1654229997Sken } 1655229997Sken 1656229997Sken if (file_arg == NULL) { 1657229997Sken snprintf(req->error_str, sizeof(req->error_str), 1658229997Sken "%s: no file argument specified", __func__); 1659229997Sken goto bailout_error; 1660229997Sken } 1661229997Sken 1662229997Sken be_lun->dev_path = malloc(file_arg->vallen, M_CTLBLK, 1663229997Sken M_WAITOK | M_ZERO); 1664229997Sken 1665249026Strasz strlcpy(be_lun->dev_path, (char *)file_arg->kvalue, 1666229997Sken file_arg->vallen); 1667229997Sken 1668229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1669229997Sken if (retval != 0) { 1670229997Sken retval = 0; 1671229997Sken goto bailout_error; 1672229997Sken } 1673229997Sken 1674229997Sken /* 1675229997Sken * Tell the user the size of the file/device. 1676229997Sken */ 1677229997Sken params->lun_size_bytes = be_lun->size_bytes; 1678229997Sken 1679229997Sken /* 1680229997Sken * The maximum LBA is the size - 1. 1681229997Sken */ 1682229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1683229997Sken } else { 1684229997Sken /* 1685229997Sken * For processor devices, we don't have any size. 1686229997Sken */ 1687229997Sken be_lun->blocksize = 0; 1688229997Sken be_lun->size_blocks = 0; 1689229997Sken be_lun->size_bytes = 0; 1690229997Sken be_lun->ctl_be_lun.maxlba = 0; 1691229997Sken params->lun_size_bytes = 0; 1692229997Sken 1693229997Sken /* 1694229997Sken * Default to just 1 thread for processor devices. 1695229997Sken */ 1696229997Sken num_threads = 1; 1697229997Sken } 1698229997Sken 1699229997Sken /* 1700229997Sken * XXX This searching loop might be refactored to be combined with 1701229997Sken * the loop above, 1702229997Sken */ 1703229997Sken for (i = 0; i < req->num_be_args; i++) { 1704249026Strasz if (strcmp(req->kern_be_args[i].kname, "num_threads") == 0) { 1705229997Sken struct ctl_be_arg *thread_arg; 1706229997Sken char num_thread_str[16]; 1707229997Sken int tmp_num_threads; 1708229997Sken 1709229997Sken 1710229997Sken thread_arg = &req->kern_be_args[i]; 1711229997Sken 1712249026Strasz strlcpy(num_thread_str, (char *)thread_arg->kvalue, 1713229997Sken min(thread_arg->vallen, 1714229997Sken sizeof(num_thread_str))); 1715229997Sken 1716229997Sken tmp_num_threads = strtol(num_thread_str, NULL, 0); 1717229997Sken 1718229997Sken /* 1719229997Sken * We don't let the user specify less than one 1720229997Sken * thread, but hope he's clueful enough not to 1721229997Sken * specify 1000 threads. 1722229997Sken */ 1723229997Sken if (tmp_num_threads < 1) { 1724229997Sken snprintf(req->error_str, sizeof(req->error_str), 1725229997Sken "%s: invalid number of threads %s", 1726229997Sken __func__, num_thread_str); 1727229997Sken goto bailout_error; 1728229997Sken } 1729229997Sken 1730229997Sken num_threads = tmp_num_threads; 1731254759Strasz } else if (strcmp(req->kern_be_args[i].kname, "file") != 0 && 1732254759Strasz strcmp(req->kern_be_args[i].kname, "dev") != 0) { 1733254759Strasz struct ctl_be_lun_option *opt; 1734254759Strasz 1735254759Strasz opt = malloc(sizeof(*opt), M_CTLBLK, M_WAITOK); 1736254759Strasz opt->name = malloc(strlen(req->kern_be_args[i].kname) + 1, M_CTLBLK, M_WAITOK); 1737254759Strasz strcpy(opt->name, req->kern_be_args[i].kname); 1738254759Strasz opt->value = malloc(strlen(req->kern_be_args[i].kvalue) + 1, M_CTLBLK, M_WAITOK); 1739254759Strasz strcpy(opt->value, req->kern_be_args[i].kvalue); 1740254759Strasz STAILQ_INSERT_TAIL(&be_lun->ctl_be_lun.options, opt, links); 1741229997Sken } 1742229997Sken } 1743229997Sken 1744229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1745229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1746229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 1747229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1748229997Sken /* Tell the user the blocksize we ended up using */ 1749229997Sken params->blocksize_bytes = be_lun->blocksize; 1750229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1751229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1752229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1753229997Sken } else 1754229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 1755229997Sken 1756229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1757229997Sken be_lun->ctl_be_lun.lun_config_status = 1758229997Sken ctl_be_block_lun_config_status; 1759229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1760229997Sken 1761229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1762229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1763229997Sken softc->num_luns); 1764229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1765229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1766229997Sken sizeof(tmpstr))); 1767229997Sken 1768229997Sken /* Tell the user what we used for a serial number */ 1769229997Sken strncpy((char *)params->serial_num, tmpstr, 1770229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1771229997Sken } else { 1772229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 1773229997Sken params->serial_num, 1774229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1775229997Sken sizeof(params->serial_num))); 1776229997Sken } 1777229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1778229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1779229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1780229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1781229997Sken sizeof(tmpstr))); 1782229997Sken 1783229997Sken /* Tell the user what we used for a device ID */ 1784229997Sken strncpy((char *)params->device_id, tmpstr, 1785229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1786229997Sken } else { 1787229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 1788229997Sken params->device_id, 1789229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1790229997Sken sizeof(params->device_id))); 1791229997Sken } 1792229997Sken 1793229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 1794229997Sken 1795229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 1796229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 1797229997Sken 1798229997Sken if (be_lun->io_taskqueue == NULL) { 1799229997Sken snprintf(req->error_str, sizeof(req->error_str), 1800229997Sken "%s: Unable to create taskqueue", __func__); 1801229997Sken goto bailout_error; 1802229997Sken } 1803229997Sken 1804229997Sken /* 1805229997Sken * Note that we start the same number of threads by default for 1806229997Sken * both the file case and the block device case. For the file 1807229997Sken * case, we need multiple threads to allow concurrency, because the 1808229997Sken * vnode interface is designed to be a blocking interface. For the 1809229997Sken * block device case, ZFS zvols at least will block the caller's 1810229997Sken * context in many instances, and so we need multiple threads to 1811229997Sken * overcome that problem. Other block devices don't need as many 1812229997Sken * threads, but they shouldn't cause too many problems. 1813229997Sken * 1814229997Sken * If the user wants to just have a single thread for a block 1815229997Sken * device, he can specify that when the LUN is created, or change 1816229997Sken * the tunable/sysctl to alter the default number of threads. 1817229997Sken */ 1818229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 1819229997Sken /*num threads*/num_threads, 1820229997Sken /*priority*/PWAIT, 1821229997Sken /*thread name*/ 1822229997Sken "%s taskq", be_lun->lunname); 1823229997Sken 1824229997Sken if (retval != 0) 1825229997Sken goto bailout_error; 1826229997Sken 1827229997Sken be_lun->num_threads = num_threads; 1828229997Sken 1829229997Sken mtx_lock(&softc->lock); 1830229997Sken softc->num_luns++; 1831229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 1832229997Sken 1833229997Sken mtx_unlock(&softc->lock); 1834229997Sken 1835229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 1836229997Sken if (retval != 0) { 1837229997Sken mtx_lock(&softc->lock); 1838229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1839229997Sken links); 1840229997Sken softc->num_luns--; 1841229997Sken mtx_unlock(&softc->lock); 1842229997Sken snprintf(req->error_str, sizeof(req->error_str), 1843229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 1844229997Sken "details", __func__, retval); 1845229997Sken retval = 0; 1846229997Sken goto bailout_error; 1847229997Sken } 1848229997Sken 1849229997Sken mtx_lock(&softc->lock); 1850229997Sken 1851229997Sken /* 1852229997Sken * Tell the config_status routine that we're waiting so it won't 1853229997Sken * clean up the LUN in the event of an error. 1854229997Sken */ 1855229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1856229997Sken 1857229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 1858229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1859229997Sken if (retval == EINTR) 1860229997Sken break; 1861229997Sken } 1862229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1863229997Sken 1864229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 1865229997Sken snprintf(req->error_str, sizeof(req->error_str), 1866229997Sken "%s: LUN configuration error, see dmesg for details", 1867229997Sken __func__); 1868229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1869229997Sken links); 1870229997Sken softc->num_luns--; 1871229997Sken mtx_unlock(&softc->lock); 1872229997Sken goto bailout_error; 1873229997Sken } else { 1874229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 1875229997Sken } 1876229997Sken 1877229997Sken mtx_unlock(&softc->lock); 1878229997Sken 1879229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 1880229997Sken be_lun->blocksize, 1881229997Sken DEVSTAT_ALL_SUPPORTED, 1882229997Sken be_lun->ctl_be_lun.lun_type 1883229997Sken | DEVSTAT_TYPE_IF_OTHER, 1884229997Sken DEVSTAT_PRIORITY_OTHER); 1885229997Sken 1886229997Sken 1887229997Sken req->status = CTL_LUN_OK; 1888229997Sken 1889229997Sken return (retval); 1890229997Sken 1891229997Skenbailout_error: 1892229997Sken req->status = CTL_LUN_ERROR; 1893229997Sken 1894229997Sken ctl_be_block_close(be_lun); 1895229997Sken 1896229997Sken free(be_lun->dev_path, M_CTLBLK); 1897229997Sken free(be_lun, M_CTLBLK); 1898229997Sken 1899229997Sken return (retval); 1900229997Sken} 1901229997Sken 1902229997Skenstatic int 1903229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1904229997Sken{ 1905229997Sken struct ctl_lun_rm_params *params; 1906229997Sken struct ctl_be_block_lun *be_lun; 1907229997Sken int retval; 1908229997Sken 1909229997Sken params = &req->reqdata.rm; 1910229997Sken 1911229997Sken mtx_lock(&softc->lock); 1912229997Sken 1913229997Sken be_lun = NULL; 1914229997Sken 1915229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 1916229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 1917229997Sken break; 1918229997Sken } 1919229997Sken mtx_unlock(&softc->lock); 1920229997Sken 1921229997Sken if (be_lun == NULL) { 1922229997Sken snprintf(req->error_str, sizeof(req->error_str), 1923229997Sken "%s: LUN %u is not managed by the block backend", 1924229997Sken __func__, params->lun_id); 1925229997Sken goto bailout_error; 1926229997Sken } 1927229997Sken 1928229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 1929229997Sken 1930229997Sken if (retval != 0) { 1931229997Sken snprintf(req->error_str, sizeof(req->error_str), 1932229997Sken "%s: error %d returned from ctl_disable_lun() for " 1933229997Sken "LUN %d", __func__, retval, params->lun_id); 1934229997Sken goto bailout_error; 1935229997Sken 1936229997Sken } 1937229997Sken 1938229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 1939229997Sken if (retval != 0) { 1940229997Sken snprintf(req->error_str, sizeof(req->error_str), 1941229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 1942229997Sken "LUN %d", __func__, retval, params->lun_id); 1943229997Sken goto bailout_error; 1944229997Sken } 1945229997Sken 1946229997Sken mtx_lock(&softc->lock); 1947229997Sken 1948229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1949229997Sken 1950229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1951229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1952229997Sken if (retval == EINTR) 1953229997Sken break; 1954229997Sken } 1955229997Sken 1956229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1957229997Sken 1958229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1959229997Sken snprintf(req->error_str, sizeof(req->error_str), 1960229997Sken "%s: interrupted waiting for LUN to be freed", 1961229997Sken __func__); 1962229997Sken mtx_unlock(&softc->lock); 1963229997Sken goto bailout_error; 1964229997Sken } 1965229997Sken 1966229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 1967229997Sken 1968229997Sken softc->num_luns--; 1969229997Sken mtx_unlock(&softc->lock); 1970229997Sken 1971229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 1972229997Sken 1973229997Sken taskqueue_free(be_lun->io_taskqueue); 1974229997Sken 1975229997Sken ctl_be_block_close(be_lun); 1976229997Sken 1977229997Sken if (be_lun->disk_stats != NULL) 1978229997Sken devstat_remove_entry(be_lun->disk_stats); 1979229997Sken 1980229997Sken uma_zdestroy(be_lun->lun_zone); 1981229997Sken 1982229997Sken free(be_lun->dev_path, M_CTLBLK); 1983229997Sken 1984229997Sken free(be_lun, M_CTLBLK); 1985229997Sken 1986229997Sken req->status = CTL_LUN_OK; 1987229997Sken 1988229997Sken return (0); 1989229997Sken 1990229997Skenbailout_error: 1991229997Sken 1992229997Sken req->status = CTL_LUN_ERROR; 1993229997Sken 1994229997Sken return (0); 1995229997Sken} 1996229997Sken 1997232604Straszstatic int 1998232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 1999232604Strasz struct ctl_lun_req *req) 2000232604Strasz{ 2001232604Strasz struct vattr vattr; 2002232604Strasz int error; 2003232604Strasz struct ctl_lun_modify_params *params; 2004232604Strasz 2005232604Strasz params = &req->reqdata.modify; 2006232604Strasz 2007232604Strasz if (params->lun_size_bytes != 0) { 2008232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2009232604Strasz } else { 2010232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2011232604Strasz if (error != 0) { 2012232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2013232604Strasz "error calling VOP_GETATTR() for file %s", 2014232604Strasz be_lun->dev_path); 2015232604Strasz return (error); 2016232604Strasz } 2017232604Strasz 2018232604Strasz be_lun->size_bytes = vattr.va_size; 2019232604Strasz } 2020232604Strasz 2021232604Strasz return (0); 2022232604Strasz} 2023232604Strasz 2024232604Straszstatic int 2025232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2026232604Strasz struct ctl_lun_req *req) 2027232604Strasz{ 2028232604Strasz struct cdev *dev; 2029232604Strasz struct cdevsw *devsw; 2030232604Strasz int error; 2031232604Strasz struct ctl_lun_modify_params *params; 2032232604Strasz uint64_t size_bytes; 2033232604Strasz 2034232604Strasz params = &req->reqdata.modify; 2035232604Strasz 2036232604Strasz dev = be_lun->vn->v_rdev; 2037232604Strasz devsw = dev->si_devsw; 2038232604Strasz if (!devsw->d_ioctl) { 2039232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2040232604Strasz "%s: no d_ioctl for device %s!", __func__, 2041232604Strasz be_lun->dev_path); 2042232604Strasz return (ENODEV); 2043232604Strasz } 2044232604Strasz 2045232604Strasz error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2046232604Strasz (caddr_t)&size_bytes, FREAD, 2047232604Strasz curthread); 2048232604Strasz if (error) { 2049232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2050232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2051232604Strasz "on %s!", __func__, error, be_lun->dev_path); 2052232604Strasz return (error); 2053232604Strasz } 2054232604Strasz 2055232604Strasz if (params->lun_size_bytes != 0) { 2056232604Strasz if (params->lun_size_bytes > size_bytes) { 2057232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2058232604Strasz "%s: requested LUN size %ju > backing device " 2059232604Strasz "size %ju", __func__, 2060232604Strasz (uintmax_t)params->lun_size_bytes, 2061232604Strasz (uintmax_t)size_bytes); 2062232604Strasz return (EINVAL); 2063232604Strasz } 2064232604Strasz 2065232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2066232604Strasz } else { 2067232604Strasz be_lun->size_bytes = size_bytes; 2068232604Strasz } 2069232604Strasz 2070232604Strasz return (0); 2071232604Strasz} 2072232604Strasz 2073232604Straszstatic int 2074232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2075232604Strasz{ 2076232604Strasz struct ctl_lun_modify_params *params; 2077232604Strasz struct ctl_be_block_lun *be_lun; 2078241896Skib int error; 2079232604Strasz 2080232604Strasz params = &req->reqdata.modify; 2081232604Strasz 2082232604Strasz mtx_lock(&softc->lock); 2083232604Strasz 2084232604Strasz be_lun = NULL; 2085232604Strasz 2086232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2087232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2088232604Strasz break; 2089232604Strasz } 2090232604Strasz mtx_unlock(&softc->lock); 2091232604Strasz 2092232604Strasz if (be_lun == NULL) { 2093232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2094232604Strasz "%s: LUN %u is not managed by the block backend", 2095232604Strasz __func__, params->lun_id); 2096232604Strasz goto bailout_error; 2097232604Strasz } 2098232604Strasz 2099232604Strasz if (params->lun_size_bytes != 0) { 2100232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2101232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2102232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2103232604Strasz params->lun_size_bytes, be_lun->blocksize); 2104232604Strasz goto bailout_error; 2105232604Strasz } 2106232604Strasz } 2107232604Strasz 2108232604Strasz vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2109232604Strasz 2110232604Strasz if (be_lun->vn->v_type == VREG) 2111232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2112232604Strasz else 2113232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2114232604Strasz 2115232604Strasz VOP_UNLOCK(be_lun->vn, 0); 2116232604Strasz 2117232604Strasz if (error != 0) 2118232604Strasz goto bailout_error; 2119232604Strasz 2120232604Strasz be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2121232604Strasz 2122232604Strasz /* 2123232604Strasz * The maximum LBA is the size - 1. 2124232604Strasz * 2125232604Strasz * XXX: Note that this field is being updated without locking, 2126232604Strasz * which might cause problems on 32-bit architectures. 2127232604Strasz */ 2128232604Strasz be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2129232604Strasz ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2130232604Strasz 2131232604Strasz /* Tell the user the exact size we ended up using */ 2132232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2133232604Strasz 2134232604Strasz req->status = CTL_LUN_OK; 2135232604Strasz 2136232604Strasz return (0); 2137232604Strasz 2138232604Straszbailout_error: 2139232604Strasz req->status = CTL_LUN_ERROR; 2140232604Strasz 2141232604Strasz return (0); 2142232604Strasz} 2143232604Strasz 2144229997Skenstatic void 2145229997Skenctl_be_block_lun_shutdown(void *be_lun) 2146229997Sken{ 2147229997Sken struct ctl_be_block_lun *lun; 2148229997Sken struct ctl_be_block_softc *softc; 2149229997Sken 2150229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2151229997Sken 2152229997Sken softc = lun->softc; 2153229997Sken 2154229997Sken mtx_lock(&softc->lock); 2155229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2156229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2157229997Sken wakeup(lun); 2158229997Sken mtx_unlock(&softc->lock); 2159229997Sken 2160229997Sken} 2161229997Sken 2162229997Skenstatic void 2163229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2164229997Sken{ 2165229997Sken struct ctl_be_block_lun *lun; 2166229997Sken struct ctl_be_block_softc *softc; 2167229997Sken 2168229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2169229997Sken softc = lun->softc; 2170229997Sken 2171229997Sken if (status == CTL_LUN_CONFIG_OK) { 2172229997Sken mtx_lock(&softc->lock); 2173229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2174229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2175229997Sken wakeup(lun); 2176229997Sken mtx_unlock(&softc->lock); 2177229997Sken 2178229997Sken /* 2179229997Sken * We successfully added the LUN, attempt to enable it. 2180229997Sken */ 2181229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2182229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2183229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2184229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2185229997Sken __func__); 2186229997Sken } 2187229997Sken } 2188229997Sken 2189229997Sken return; 2190229997Sken } 2191229997Sken 2192229997Sken 2193229997Sken mtx_lock(&softc->lock); 2194229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2195229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2196229997Sken wakeup(lun); 2197229997Sken mtx_unlock(&softc->lock); 2198229997Sken} 2199229997Sken 2200229997Sken 2201229997Skenstatic int 2202229997Skenctl_be_block_config_write(union ctl_io *io) 2203229997Sken{ 2204229997Sken struct ctl_be_block_lun *be_lun; 2205229997Sken struct ctl_be_lun *ctl_be_lun; 2206229997Sken int retval; 2207229997Sken 2208229997Sken retval = 0; 2209229997Sken 2210229997Sken DPRINTF("entered\n"); 2211229997Sken 2212229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2213229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2214229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2215229997Sken 2216229997Sken switch (io->scsiio.cdb[0]) { 2217229997Sken case SYNCHRONIZE_CACHE: 2218229997Sken case SYNCHRONIZE_CACHE_16: 2219229997Sken /* 2220229997Sken * The upper level CTL code will filter out any CDBs with 2221229997Sken * the immediate bit set and return the proper error. 2222229997Sken * 2223229997Sken * We don't really need to worry about what LBA range the 2224229997Sken * user asked to be synced out. When they issue a sync 2225229997Sken * cache command, we'll sync out the whole thing. 2226229997Sken */ 2227229997Sken mtx_lock(&be_lun->lock); 2228229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2229229997Sken links); 2230229997Sken mtx_unlock(&be_lun->lock); 2231229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2232229997Sken break; 2233229997Sken case START_STOP_UNIT: { 2234229997Sken struct scsi_start_stop_unit *cdb; 2235229997Sken 2236229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2237229997Sken 2238229997Sken if (cdb->how & SSS_START) 2239229997Sken retval = ctl_start_lun(ctl_be_lun); 2240229997Sken else { 2241229997Sken retval = ctl_stop_lun(ctl_be_lun); 2242229997Sken /* 2243229997Sken * XXX KDM Copan-specific offline behavior. 2244229997Sken * Figure out a reasonable way to port this? 2245229997Sken */ 2246229997Sken#ifdef NEEDTOPORT 2247229997Sken if ((retval == 0) 2248229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2249229997Sken retval = ctl_lun_offline(ctl_be_lun); 2250229997Sken#endif 2251229997Sken } 2252229997Sken 2253229997Sken /* 2254229997Sken * In general, the above routines should not fail. They 2255229997Sken * just set state for the LUN. So we've got something 2256229997Sken * pretty wrong here if we can't start or stop the LUN. 2257229997Sken */ 2258229997Sken if (retval != 0) { 2259229997Sken ctl_set_internal_failure(&io->scsiio, 2260229997Sken /*sks_valid*/ 1, 2261229997Sken /*retry_count*/ 0xf051); 2262229997Sken retval = CTL_RETVAL_COMPLETE; 2263229997Sken } else { 2264229997Sken ctl_set_success(&io->scsiio); 2265229997Sken } 2266229997Sken ctl_config_write_done(io); 2267229997Sken break; 2268229997Sken } 2269229997Sken default: 2270229997Sken ctl_set_invalid_opcode(&io->scsiio); 2271229997Sken ctl_config_write_done(io); 2272229997Sken retval = CTL_RETVAL_COMPLETE; 2273229997Sken break; 2274229997Sken } 2275229997Sken 2276229997Sken return (retval); 2277229997Sken 2278229997Sken} 2279229997Sken 2280229997Skenstatic int 2281229997Skenctl_be_block_config_read(union ctl_io *io) 2282229997Sken{ 2283229997Sken return (0); 2284229997Sken} 2285229997Sken 2286229997Skenstatic int 2287229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2288229997Sken{ 2289229997Sken struct ctl_be_block_lun *lun; 2290229997Sken int retval; 2291229997Sken 2292229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2293229997Sken retval = 0; 2294229997Sken 2295229997Sken retval = sbuf_printf(sb, "<num_threads>"); 2296229997Sken 2297229997Sken if (retval != 0) 2298229997Sken goto bailout; 2299229997Sken 2300229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2301229997Sken 2302229997Sken if (retval != 0) 2303229997Sken goto bailout; 2304229997Sken 2305229997Sken retval = sbuf_printf(sb, "</num_threads>"); 2306229997Sken 2307229997Sken /* 2308229997Sken * For processor devices, we don't have a path variable. 2309229997Sken */ 2310229997Sken if ((retval != 0) 2311229997Sken || (lun->dev_path == NULL)) 2312229997Sken goto bailout; 2313229997Sken 2314229997Sken retval = sbuf_printf(sb, "<file>"); 2315229997Sken 2316229997Sken if (retval != 0) 2317229997Sken goto bailout; 2318229997Sken 2319229997Sken retval = ctl_sbuf_printf_esc(sb, lun->dev_path); 2320229997Sken 2321229997Sken if (retval != 0) 2322229997Sken goto bailout; 2323229997Sken 2324229997Sken retval = sbuf_printf(sb, "</file>\n"); 2325229997Sken 2326229997Skenbailout: 2327229997Sken 2328229997Sken return (retval); 2329229997Sken} 2330229997Sken 2331229997Skenint 2332229997Skenctl_be_block_init(void) 2333229997Sken{ 2334229997Sken struct ctl_be_block_softc *softc; 2335229997Sken int retval; 2336229997Sken 2337229997Sken softc = &backend_block_softc; 2338229997Sken retval = 0; 2339229997Sken 2340229997Sken mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF); 2341229997Sken STAILQ_INIT(&softc->beio_free_queue); 2342229997Sken STAILQ_INIT(&softc->disk_list); 2343229997Sken STAILQ_INIT(&softc->lun_list); 2344229997Sken ctl_grow_beio(softc, 200); 2345229997Sken 2346229997Sken return (retval); 2347229997Sken} 2348