ctl_backend_block.c revision 269622
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 269622 2014-08-06 08:54:31Z mav $"); 44229997Sken 45229997Sken#include <sys/param.h> 46229997Sken#include <sys/systm.h> 47229997Sken#include <sys/kernel.h> 48229997Sken#include <sys/types.h> 49229997Sken#include <sys/kthread.h> 50229997Sken#include <sys/bio.h> 51229997Sken#include <sys/fcntl.h> 52264274Smav#include <sys/limits.h> 53229997Sken#include <sys/lock.h> 54229997Sken#include <sys/mutex.h> 55229997Sken#include <sys/condvar.h> 56229997Sken#include <sys/malloc.h> 57229997Sken#include <sys/conf.h> 58229997Sken#include <sys/ioccom.h> 59229997Sken#include <sys/queue.h> 60229997Sken#include <sys/sbuf.h> 61229997Sken#include <sys/endian.h> 62229997Sken#include <sys/uio.h> 63229997Sken#include <sys/buf.h> 64229997Sken#include <sys/taskqueue.h> 65229997Sken#include <sys/vnode.h> 66229997Sken#include <sys/namei.h> 67229997Sken#include <sys/mount.h> 68229997Sken#include <sys/disk.h> 69229997Sken#include <sys/fcntl.h> 70229997Sken#include <sys/filedesc.h> 71229997Sken#include <sys/proc.h> 72229997Sken#include <sys/pcpu.h> 73229997Sken#include <sys/module.h> 74229997Sken#include <sys/sdt.h> 75229997Sken#include <sys/devicestat.h> 76229997Sken#include <sys/sysctl.h> 77229997Sken 78229997Sken#include <geom/geom.h> 79229997Sken 80229997Sken#include <cam/cam.h> 81229997Sken#include <cam/scsi/scsi_all.h> 82229997Sken#include <cam/scsi/scsi_da.h> 83229997Sken#include <cam/ctl/ctl_io.h> 84229997Sken#include <cam/ctl/ctl.h> 85229997Sken#include <cam/ctl/ctl_backend.h> 86229997Sken#include <cam/ctl/ctl_frontend_internal.h> 87229997Sken#include <cam/ctl/ctl_ioctl.h> 88229997Sken#include <cam/ctl/ctl_scsi_all.h> 89229997Sken#include <cam/ctl/ctl_error.h> 90229997Sken 91229997Sken/* 92264886Smav * The idea here is that we'll allocate enough S/G space to hold a 1MB 93264886Smav * I/O. If we get an I/O larger than that, we'll split it. 94229997Sken */ 95267537Smav#define CTLBLK_HALF_IO_SIZE (512 * 1024) 96267537Smav#define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 97264886Smav#define CTLBLK_MAX_SEG MAXPHYS 98267537Smav#define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 99267537Smav#define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 100229997Sken 101229997Sken#ifdef CTLBLK_DEBUG 102229997Sken#define DPRINTF(fmt, args...) \ 103229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 104229997Sken#else 105229997Sken#define DPRINTF(fmt, args...) do {} while(0) 106229997Sken#endif 107229997Sken 108267519Smav#define PRIV(io) \ 109267519Smav ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 110267537Smav#define ARGS(io) \ 111267537Smav ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 112267519Smav 113229997SkenSDT_PROVIDER_DEFINE(cbb); 114229997Sken 115229997Skentypedef enum { 116229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 117229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 118229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 119229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 120229997Sken} ctl_be_block_lun_flags; 121229997Sken 122229997Skentypedef enum { 123229997Sken CTL_BE_BLOCK_NONE, 124229997Sken CTL_BE_BLOCK_DEV, 125229997Sken CTL_BE_BLOCK_FILE 126229997Sken} ctl_be_block_type; 127229997Sken 128229997Skenstruct ctl_be_block_devdata { 129229997Sken struct cdev *cdev; 130229997Sken struct cdevsw *csw; 131229997Sken int dev_ref; 132229997Sken}; 133229997Sken 134229997Skenstruct ctl_be_block_filedata { 135229997Sken struct ucred *cred; 136229997Sken}; 137229997Sken 138229997Skenunion ctl_be_block_bedata { 139229997Sken struct ctl_be_block_devdata dev; 140229997Sken struct ctl_be_block_filedata file; 141229997Sken}; 142229997Sken 143229997Skenstruct ctl_be_block_io; 144229997Skenstruct ctl_be_block_lun; 145229997Sken 146229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 147229997Sken struct ctl_be_block_io *beio); 148229997Sken 149229997Sken/* 150229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 151229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 152229997Sken */ 153229997Skenstruct ctl_be_block_lun { 154229997Sken struct ctl_block_disk *disk; 155229997Sken char lunname[32]; 156229997Sken char *dev_path; 157229997Sken ctl_be_block_type dev_type; 158229997Sken struct vnode *vn; 159229997Sken union ctl_be_block_bedata backend; 160229997Sken cbb_dispatch_t dispatch; 161229997Sken cbb_dispatch_t lun_flush; 162264274Smav cbb_dispatch_t unmap; 163229997Sken uma_zone_t lun_zone; 164229997Sken uint64_t size_blocks; 165229997Sken uint64_t size_bytes; 166229997Sken uint32_t blocksize; 167229997Sken int blocksize_shift; 168264191Smav uint16_t pblockexp; 169264191Smav uint16_t pblockoff; 170229997Sken struct ctl_be_block_softc *softc; 171229997Sken struct devstat *disk_stats; 172229997Sken ctl_be_block_lun_flags flags; 173229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 174229997Sken struct ctl_be_lun ctl_be_lun; 175229997Sken struct taskqueue *io_taskqueue; 176229997Sken struct task io_task; 177229997Sken int num_threads; 178229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 179229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 180229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 181267877Smav struct mtx_padalign io_lock; 182267877Smav struct mtx_padalign queue_lock; 183229997Sken}; 184229997Sken 185229997Sken/* 186229997Sken * Overall softc structure for the block backend module. 187229997Sken */ 188229997Skenstruct ctl_be_block_softc { 189229997Sken struct mtx lock; 190229997Sken int num_disks; 191229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 192229997Sken int num_luns; 193229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 194229997Sken}; 195229997Sken 196229997Skenstatic struct ctl_be_block_softc backend_block_softc; 197229997Sken 198229997Sken/* 199229997Sken * Per-I/O information. 200229997Sken */ 201229997Skenstruct ctl_be_block_io { 202229997Sken union ctl_io *io; 203229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 204229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 205229997Sken int bio_cmd; 206229997Sken int bio_flags; 207229997Sken int num_segs; 208229997Sken int num_bios_sent; 209229997Sken int num_bios_done; 210229997Sken int send_complete; 211229997Sken int num_errors; 212229997Sken struct bintime ds_t0; 213229997Sken devstat_tag_type ds_tag_type; 214229997Sken devstat_trans_flags ds_trans_type; 215229997Sken uint64_t io_len; 216229997Sken uint64_t io_offset; 217229997Sken struct ctl_be_block_softc *softc; 218229997Sken struct ctl_be_block_lun *lun; 219264274Smav void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 220229997Sken}; 221229997Sken 222229997Skenstatic int cbb_num_threads = 14; 223229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 224229997Sken "CAM Target Layer Block Backend"); 225267992ShselaskySYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 226229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 227229997Sken 228229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 229229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 230229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 231229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 232229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 233229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 234229997Sken struct ctl_be_block_io *beio); 235229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 236229997Sken struct ctl_be_block_io *beio); 237229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 238229997Sken struct ctl_be_block_io *beio); 239264274Smavstatic void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 240264274Smav struct ctl_be_block_io *beio); 241229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 242229997Sken struct ctl_be_block_io *beio); 243229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 244229997Sken union ctl_io *io); 245229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 246229997Sken union ctl_io *io); 247229997Skenstatic void ctl_be_block_worker(void *context, int pending); 248229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 249229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 250229997Sken int flag, struct thread *td); 251229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 252229997Sken struct ctl_lun_req *req); 253229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 254229997Sken struct ctl_lun_req *req); 255229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 256229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 257229997Sken struct ctl_be_block_lun *be_lun, 258229997Sken struct ctl_lun_req *req); 259229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 260229997Sken struct ctl_lun_req *req); 261229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 262229997Sken struct ctl_lun_req *req); 263232604Straszstatic int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 264232604Strasz struct ctl_lun_req *req); 265232604Straszstatic int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 266232604Strasz struct ctl_lun_req *req); 267232604Straszstatic int ctl_be_block_modify(struct ctl_be_block_softc *softc, 268232604Strasz struct ctl_lun_req *req); 269229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 270229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 271229997Sken ctl_lun_config_status status); 272229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 273229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 274229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 275229997Skenint ctl_be_block_init(void); 276229997Sken 277229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 278229997Sken{ 279230334Sken .name = "block", 280230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 281230334Sken .init = ctl_be_block_init, 282230334Sken .data_submit = ctl_be_block_submit, 283230334Sken .data_move_done = ctl_be_block_move_done, 284230334Sken .config_read = ctl_be_block_config_read, 285230334Sken .config_write = ctl_be_block_config_write, 286230334Sken .ioctl = ctl_be_block_ioctl, 287230334Sken .lun_info = ctl_be_block_lun_info 288229997Sken}; 289229997Sken 290229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 291229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 292229997Sken 293264020Straszstatic uma_zone_t beio_zone; 294264020Strasz 295229997Skenstatic struct ctl_be_block_io * 296229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 297229997Sken{ 298229997Sken struct ctl_be_block_io *beio; 299229997Sken 300264020Strasz beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 301264020Strasz beio->softc = softc; 302229997Sken return (beio); 303229997Sken} 304229997Sken 305229997Skenstatic void 306229997Skenctl_free_beio(struct ctl_be_block_io *beio) 307229997Sken{ 308229997Sken int duplicate_free; 309229997Sken int i; 310229997Sken 311229997Sken duplicate_free = 0; 312229997Sken 313229997Sken for (i = 0; i < beio->num_segs; i++) { 314229997Sken if (beio->sg_segs[i].addr == NULL) 315229997Sken duplicate_free++; 316229997Sken 317229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 318229997Sken beio->sg_segs[i].addr = NULL; 319267537Smav 320267537Smav /* For compare we had two equal S/G lists. */ 321267537Smav if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 322267537Smav uma_zfree(beio->lun->lun_zone, 323267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 324267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 325267537Smav } 326229997Sken } 327229997Sken 328229997Sken if (duplicate_free > 0) { 329229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 330229997Sken duplicate_free, beio->num_segs); 331229997Sken } 332229997Sken 333264020Strasz uma_zfree(beio_zone, beio); 334229997Sken} 335229997Sken 336229997Skenstatic void 337229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 338229997Sken{ 339267877Smav union ctl_io *io = beio->io; 340229997Sken 341264274Smav if (beio->beio_cont != NULL) { 342264274Smav beio->beio_cont(beio); 343264274Smav } else { 344264274Smav ctl_free_beio(beio); 345267537Smav ctl_data_submit_done(io); 346264274Smav } 347229997Sken} 348229997Sken 349229997Skenstatic int 350229997Skenctl_be_block_move_done(union ctl_io *io) 351229997Sken{ 352229997Sken struct ctl_be_block_io *beio; 353229997Sken struct ctl_be_block_lun *be_lun; 354267537Smav struct ctl_lba_len_flags *lbalen; 355229997Sken#ifdef CTL_TIME_IO 356229997Sken struct bintime cur_bt; 357267537Smav#endif 358267537Smav int i; 359229997Sken 360267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 361229997Sken be_lun = beio->lun; 362229997Sken 363229997Sken DPRINTF("entered\n"); 364229997Sken 365229997Sken#ifdef CTL_TIME_IO 366229997Sken getbintime(&cur_bt); 367229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 368229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 369229997Sken io->io_hdr.num_dmas++; 370229997Sken#endif 371267537Smav io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 372229997Sken 373229997Sken /* 374229997Sken * We set status at this point for read commands, and write 375229997Sken * commands with errors. 376229997Sken */ 377267537Smav if ((io->io_hdr.port_status == 0) && 378267537Smav ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && 379267537Smav ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 380267537Smav lbalen = ARGS(beio->io); 381267537Smav if (lbalen->flags & CTL_LLF_READ) { 382267537Smav ctl_set_success(&io->scsiio); 383267537Smav } else if (lbalen->flags & CTL_LLF_COMPARE) { 384267537Smav /* We have two data blocks ready for comparison. */ 385267537Smav for (i = 0; i < beio->num_segs; i++) { 386267537Smav if (memcmp(beio->sg_segs[i].addr, 387267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 388267537Smav beio->sg_segs[i].len) != 0) 389267537Smav break; 390267537Smav } 391267537Smav if (i < beio->num_segs) 392267537Smav ctl_set_sense(&io->scsiio, 393267537Smav /*current_error*/ 1, 394267537Smav /*sense_key*/ SSD_KEY_MISCOMPARE, 395267537Smav /*asc*/ 0x1D, 396267537Smav /*ascq*/ 0x00, 397267537Smav SSD_ELEM_NONE); 398267537Smav else 399267537Smav ctl_set_success(&io->scsiio); 400267537Smav } 401267537Smav } 402229997Sken else if ((io->io_hdr.port_status != 0) 403229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 404229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 405229997Sken /* 406229997Sken * For hardware error sense keys, the sense key 407229997Sken * specific value is defined to be a retry count, 408229997Sken * but we use it to pass back an internal FETD 409229997Sken * error code. XXX KDM Hopefully the FETD is only 410229997Sken * using 16 bits for an error code, since that's 411229997Sken * all the space we have in the sks field. 412229997Sken */ 413229997Sken ctl_set_internal_failure(&io->scsiio, 414229997Sken /*sks_valid*/ 1, 415229997Sken /*retry_count*/ 416229997Sken io->io_hdr.port_status); 417229997Sken } 418229997Sken 419229997Sken /* 420229997Sken * If this is a read, or a write with errors, it is done. 421229997Sken */ 422229997Sken if ((beio->bio_cmd == BIO_READ) 423229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 424229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 425229997Sken ctl_complete_beio(beio); 426229997Sken return (0); 427229997Sken } 428229997Sken 429229997Sken /* 430229997Sken * At this point, we have a write and the DMA completed 431229997Sken * successfully. We now have to queue it to the task queue to 432229997Sken * execute the backend I/O. That is because we do blocking 433229997Sken * memory allocations, and in the file backing case, blocking I/O. 434229997Sken * This move done routine is generally called in the SIM's 435229997Sken * interrupt context, and therefore we cannot block. 436229997Sken */ 437267877Smav mtx_lock(&be_lun->queue_lock); 438229997Sken /* 439229997Sken * XXX KDM make sure that links is okay to use at this point. 440229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 441229997Sken * or deal with resource allocation here. 442229997Sken */ 443229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 444267877Smav mtx_unlock(&be_lun->queue_lock); 445229997Sken 446229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 447229997Sken 448229997Sken return (0); 449229997Sken} 450229997Sken 451229997Skenstatic void 452229997Skenctl_be_block_biodone(struct bio *bio) 453229997Sken{ 454229997Sken struct ctl_be_block_io *beio; 455229997Sken struct ctl_be_block_lun *be_lun; 456229997Sken union ctl_io *io; 457261538Smav int error; 458229997Sken 459229997Sken beio = bio->bio_caller1; 460229997Sken be_lun = beio->lun; 461229997Sken io = beio->io; 462229997Sken 463229997Sken DPRINTF("entered\n"); 464229997Sken 465261538Smav error = bio->bio_error; 466267877Smav mtx_lock(&be_lun->io_lock); 467261538Smav if (error != 0) 468229997Sken beio->num_errors++; 469229997Sken 470229997Sken beio->num_bios_done++; 471229997Sken 472229997Sken /* 473229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 474229997Sken * during the free might cause it to complain. 475229997Sken */ 476229997Sken g_destroy_bio(bio); 477229997Sken 478229997Sken /* 479229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 480229997Sken * complete, then we're done. 481229997Sken */ 482229997Sken if ((beio->send_complete == 0) 483229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 484267877Smav mtx_unlock(&be_lun->io_lock); 485229997Sken return; 486229997Sken } 487229997Sken 488229997Sken /* 489229997Sken * At this point, we've verified that we are the last I/O to 490229997Sken * complete, so it's safe to drop the lock. 491229997Sken */ 492267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 493267877Smav beio->ds_tag_type, beio->ds_trans_type, 494267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 495267877Smav mtx_unlock(&be_lun->io_lock); 496229997Sken 497229997Sken /* 498229997Sken * If there are any errors from the backing device, we fail the 499229997Sken * entire I/O with a medium error. 500229997Sken */ 501229997Sken if (beio->num_errors > 0) { 502261538Smav if (error == EOPNOTSUPP) { 503261538Smav ctl_set_invalid_opcode(&io->scsiio); 504261538Smav } else if (beio->bio_cmd == BIO_FLUSH) { 505229997Sken /* XXX KDM is there is a better error here? */ 506229997Sken ctl_set_internal_failure(&io->scsiio, 507229997Sken /*sks_valid*/ 1, 508229997Sken /*retry_count*/ 0xbad2); 509229997Sken } else 510229997Sken ctl_set_medium_error(&io->scsiio); 511229997Sken ctl_complete_beio(beio); 512229997Sken return; 513229997Sken } 514229997Sken 515229997Sken /* 516267537Smav * If this is a write, a flush, a delete or verify, we're all done. 517229997Sken * If this is a read, we can now send the data to the user. 518229997Sken */ 519229997Sken if ((beio->bio_cmd == BIO_WRITE) 520264274Smav || (beio->bio_cmd == BIO_FLUSH) 521267537Smav || (beio->bio_cmd == BIO_DELETE) 522267537Smav || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 523229997Sken ctl_set_success(&io->scsiio); 524229997Sken ctl_complete_beio(beio); 525229997Sken } else { 526229997Sken#ifdef CTL_TIME_IO 527229997Sken getbintime(&io->io_hdr.dma_start_bt); 528229997Sken#endif 529229997Sken ctl_datamove(io); 530229997Sken } 531229997Sken} 532229997Sken 533229997Skenstatic void 534229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 535229997Sken struct ctl_be_block_io *beio) 536229997Sken{ 537267877Smav union ctl_io *io = beio->io; 538229997Sken struct mount *mountpoint; 539241896Skib int error, lock_flags; 540229997Sken 541229997Sken DPRINTF("entered\n"); 542229997Sken 543267877Smav binuptime(&beio->ds_t0); 544267877Smav mtx_lock(&be_lun->io_lock); 545267877Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 546267877Smav mtx_unlock(&be_lun->io_lock); 547229997Sken 548267877Smav (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 549229997Sken 550229997Sken if (MNT_SHARED_WRITES(mountpoint) 551229997Sken || ((mountpoint == NULL) 552229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 553229997Sken lock_flags = LK_SHARED; 554229997Sken else 555229997Sken lock_flags = LK_EXCLUSIVE; 556229997Sken 557229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 558229997Sken 559229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 560229997Sken VOP_UNLOCK(be_lun->vn, 0); 561229997Sken 562229997Sken vn_finished_write(mountpoint); 563229997Sken 564267877Smav mtx_lock(&be_lun->io_lock); 565267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 566267877Smav beio->ds_tag_type, beio->ds_trans_type, 567267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 568267877Smav mtx_unlock(&be_lun->io_lock); 569267877Smav 570229997Sken if (error == 0) 571229997Sken ctl_set_success(&io->scsiio); 572229997Sken else { 573229997Sken /* XXX KDM is there is a better error here? */ 574229997Sken ctl_set_internal_failure(&io->scsiio, 575229997Sken /*sks_valid*/ 1, 576229997Sken /*retry_count*/ 0xbad1); 577229997Sken } 578229997Sken 579229997Sken ctl_complete_beio(beio); 580229997Sken} 581229997Sken 582258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 583258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 584258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 585258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 586229997Sken 587229997Skenstatic void 588229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 589229997Sken struct ctl_be_block_io *beio) 590229997Sken{ 591229997Sken struct ctl_be_block_filedata *file_data; 592229997Sken union ctl_io *io; 593229997Sken struct uio xuio; 594229997Sken struct iovec *xiovec; 595241896Skib int flags; 596229997Sken int error, i; 597229997Sken 598229997Sken DPRINTF("entered\n"); 599229997Sken 600229997Sken file_data = &be_lun->backend.file; 601229997Sken io = beio->io; 602229997Sken flags = beio->bio_flags; 603229997Sken 604267537Smav bzero(&xuio, sizeof(xuio)); 605229997Sken if (beio->bio_cmd == BIO_READ) { 606229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 607267537Smav xuio.uio_rw = UIO_READ; 608229997Sken } else { 609229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 610267537Smav xuio.uio_rw = UIO_WRITE; 611229997Sken } 612229997Sken xuio.uio_offset = beio->io_offset; 613229997Sken xuio.uio_resid = beio->io_len; 614229997Sken xuio.uio_segflg = UIO_SYSSPACE; 615229997Sken xuio.uio_iov = beio->xiovecs; 616229997Sken xuio.uio_iovcnt = beio->num_segs; 617229997Sken xuio.uio_td = curthread; 618229997Sken 619229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 620229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 621229997Sken xiovec->iov_len = beio->sg_segs[i].len; 622229997Sken } 623229997Sken 624267877Smav binuptime(&beio->ds_t0); 625267877Smav mtx_lock(&be_lun->io_lock); 626267877Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 627267877Smav mtx_unlock(&be_lun->io_lock); 628267877Smav 629229997Sken if (beio->bio_cmd == BIO_READ) { 630229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 631229997Sken 632229997Sken /* 633229997Sken * UFS pays attention to IO_DIRECT for reads. If the 634229997Sken * DIRECTIO option is configured into the kernel, it calls 635229997Sken * ffs_rawread(). But that only works for single-segment 636229997Sken * uios with user space addresses. In our case, with a 637229997Sken * kernel uio, it still reads into the buffer cache, but it 638229997Sken * will just try to release the buffer from the cache later 639229997Sken * on in ffs_read(). 640229997Sken * 641229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 642229997Sken * 643229997Sken * UFS does not pay attention to IO_SYNC for reads. 644229997Sken * 645229997Sken * ZFS pays attention to IO_SYNC (which translates into the 646229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 647229997Sken * attempts to sync the file before reading. 648229997Sken * 649229997Sken * So, to attempt to provide some barrier semantics in the 650229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 651229997Sken */ 652229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 653229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 654229997Sken 655229997Sken VOP_UNLOCK(be_lun->vn, 0); 656267537Smav SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 657229997Sken } else { 658229997Sken struct mount *mountpoint; 659229997Sken int lock_flags; 660229997Sken 661229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 662229997Sken 663229997Sken if (MNT_SHARED_WRITES(mountpoint) 664229997Sken || ((mountpoint == NULL) 665229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 666229997Sken lock_flags = LK_SHARED; 667229997Sken else 668229997Sken lock_flags = LK_EXCLUSIVE; 669229997Sken 670229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 671229997Sken 672229997Sken /* 673229997Sken * UFS pays attention to IO_DIRECT for writes. The write 674229997Sken * is done asynchronously. (Normally the write would just 675229997Sken * get put into cache. 676229997Sken * 677229997Sken * UFS pays attention to IO_SYNC for writes. It will 678229997Sken * attempt to write the buffer out synchronously if that 679229997Sken * flag is set. 680229997Sken * 681229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 682229997Sken * 683229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 684229997Sken * for writes. It will flush the transaction from the 685229997Sken * cache before returning. 686229997Sken * 687229997Sken * So if we've got the BIO_ORDERED flag set, we want 688229997Sken * IO_SYNC in either the UFS or ZFS case. 689229997Sken */ 690229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 691229997Sken IO_SYNC : 0, file_data->cred); 692229997Sken VOP_UNLOCK(be_lun->vn, 0); 693229997Sken 694229997Sken vn_finished_write(mountpoint); 695267537Smav SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 696229997Sken } 697229997Sken 698267877Smav mtx_lock(&be_lun->io_lock); 699267877Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 700267877Smav beio->ds_tag_type, beio->ds_trans_type, 701267877Smav /*now*/ NULL, /*then*/&beio->ds_t0); 702267877Smav mtx_unlock(&be_lun->io_lock); 703267877Smav 704229997Sken /* 705229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 706229997Sken * return the I/O to the user. 707229997Sken */ 708229997Sken if (error != 0) { 709229997Sken char path_str[32]; 710229997Sken 711229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 712229997Sken /* 713229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 714229997Sken * filesystem fills up. What kind of SCSI error should we 715229997Sken * return for that? 716229997Sken */ 717229997Sken printf("%s%s command returned errno %d\n", path_str, 718229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 719229997Sken ctl_set_medium_error(&io->scsiio); 720229997Sken ctl_complete_beio(beio); 721229997Sken return; 722229997Sken } 723229997Sken 724229997Sken /* 725269122Smav * If this is a write or a verify, we're all done. 726229997Sken * If this is a read, we can now send the data to the user. 727229997Sken */ 728269122Smav if ((beio->bio_cmd == BIO_WRITE) || 729269122Smav (ARGS(io)->flags & CTL_LLF_VERIFY)) { 730229997Sken ctl_set_success(&io->scsiio); 731229997Sken ctl_complete_beio(beio); 732229997Sken } else { 733229997Sken#ifdef CTL_TIME_IO 734229997Sken getbintime(&io->io_hdr.dma_start_bt); 735229997Sken#endif 736229997Sken ctl_datamove(io); 737229997Sken } 738229997Sken} 739229997Sken 740229997Skenstatic void 741269123Smavctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 742269123Smav struct ctl_be_block_io *beio) 743269123Smav{ 744269123Smav struct ctl_be_block_devdata *dev_data; 745269123Smav union ctl_io *io; 746269123Smav struct uio xuio; 747269123Smav struct iovec *xiovec; 748269123Smav int flags; 749269123Smav int error, i; 750269123Smav 751269123Smav DPRINTF("entered\n"); 752269123Smav 753269123Smav dev_data = &be_lun->backend.dev; 754269123Smav io = beio->io; 755269123Smav flags = beio->bio_flags; 756269123Smav 757269123Smav bzero(&xuio, sizeof(xuio)); 758269123Smav if (beio->bio_cmd == BIO_READ) { 759269123Smav SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 760269123Smav xuio.uio_rw = UIO_READ; 761269123Smav } else { 762269123Smav SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 763269123Smav xuio.uio_rw = UIO_WRITE; 764269123Smav } 765269123Smav xuio.uio_offset = beio->io_offset; 766269123Smav xuio.uio_resid = beio->io_len; 767269123Smav xuio.uio_segflg = UIO_SYSSPACE; 768269123Smav xuio.uio_iov = beio->xiovecs; 769269123Smav xuio.uio_iovcnt = beio->num_segs; 770269123Smav xuio.uio_td = curthread; 771269123Smav 772269123Smav for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 773269123Smav xiovec->iov_base = beio->sg_segs[i].addr; 774269123Smav xiovec->iov_len = beio->sg_segs[i].len; 775269123Smav } 776269123Smav 777269123Smav binuptime(&beio->ds_t0); 778269123Smav mtx_lock(&be_lun->io_lock); 779269123Smav devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 780269123Smav mtx_unlock(&be_lun->io_lock); 781269123Smav 782269123Smav if (beio->bio_cmd == BIO_READ) { 783269123Smav error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, 0); 784269123Smav SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 785269123Smav } else { 786269123Smav error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, 0); 787269123Smav SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 788269123Smav } 789269123Smav 790269123Smav mtx_lock(&be_lun->io_lock); 791269123Smav devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 792269123Smav beio->ds_tag_type, beio->ds_trans_type, 793269123Smav /*now*/ NULL, /*then*/&beio->ds_t0); 794269123Smav mtx_unlock(&be_lun->io_lock); 795269123Smav 796269123Smav /* 797269123Smav * If we got an error, set the sense data to "MEDIUM ERROR" and 798269123Smav * return the I/O to the user. 799269123Smav */ 800269123Smav if (error != 0) { 801269123Smav ctl_set_medium_error(&io->scsiio); 802269123Smav ctl_complete_beio(beio); 803269123Smav return; 804269123Smav } 805269123Smav 806269123Smav /* 807269123Smav * If this is a write or a verify, we're all done. 808269123Smav * If this is a read, we can now send the data to the user. 809269123Smav */ 810269123Smav if ((beio->bio_cmd == BIO_WRITE) || 811269123Smav (ARGS(io)->flags & CTL_LLF_VERIFY)) { 812269123Smav ctl_set_success(&io->scsiio); 813269123Smav ctl_complete_beio(beio); 814269123Smav } else { 815269123Smav#ifdef CTL_TIME_IO 816269123Smav getbintime(&io->io_hdr.dma_start_bt); 817269123Smav#endif 818269123Smav ctl_datamove(io); 819269123Smav } 820269123Smav} 821269123Smav 822269123Smavstatic void 823229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 824229997Sken struct ctl_be_block_io *beio) 825229997Sken{ 826229997Sken struct bio *bio; 827229997Sken union ctl_io *io; 828229997Sken struct ctl_be_block_devdata *dev_data; 829229997Sken 830229997Sken dev_data = &be_lun->backend.dev; 831229997Sken io = beio->io; 832229997Sken 833229997Sken DPRINTF("entered\n"); 834229997Sken 835229997Sken /* This can't fail, it's a blocking allocation. */ 836229997Sken bio = g_alloc_bio(); 837229997Sken 838229997Sken bio->bio_cmd = BIO_FLUSH; 839229997Sken bio->bio_flags |= BIO_ORDERED; 840229997Sken bio->bio_dev = dev_data->cdev; 841229997Sken bio->bio_offset = 0; 842229997Sken bio->bio_data = 0; 843229997Sken bio->bio_done = ctl_be_block_biodone; 844229997Sken bio->bio_caller1 = beio; 845229997Sken bio->bio_pblkno = 0; 846229997Sken 847229997Sken /* 848229997Sken * We don't need to acquire the LUN lock here, because we are only 849229997Sken * sending one bio, and so there is no other context to synchronize 850229997Sken * with. 851229997Sken */ 852229997Sken beio->num_bios_sent = 1; 853229997Sken beio->send_complete = 1; 854229997Sken 855229997Sken binuptime(&beio->ds_t0); 856267877Smav mtx_lock(&be_lun->io_lock); 857229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 858267877Smav mtx_unlock(&be_lun->io_lock); 859229997Sken 860229997Sken (*dev_data->csw->d_strategy)(bio); 861229997Sken} 862229997Sken 863229997Skenstatic void 864264274Smavctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 865264274Smav struct ctl_be_block_io *beio, 866264274Smav uint64_t off, uint64_t len, int last) 867264274Smav{ 868264274Smav struct bio *bio; 869264274Smav struct ctl_be_block_devdata *dev_data; 870264296Smav uint64_t maxlen; 871264274Smav 872264274Smav dev_data = &be_lun->backend.dev; 873264296Smav maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 874264274Smav while (len > 0) { 875264274Smav bio = g_alloc_bio(); 876264274Smav bio->bio_cmd = BIO_DELETE; 877264274Smav bio->bio_flags |= beio->bio_flags; 878264274Smav bio->bio_dev = dev_data->cdev; 879264274Smav bio->bio_offset = off; 880264296Smav bio->bio_length = MIN(len, maxlen); 881264274Smav bio->bio_data = 0; 882264274Smav bio->bio_done = ctl_be_block_biodone; 883264274Smav bio->bio_caller1 = beio; 884264296Smav bio->bio_pblkno = off / be_lun->blocksize; 885264274Smav 886264274Smav off += bio->bio_length; 887264274Smav len -= bio->bio_length; 888264274Smav 889267877Smav mtx_lock(&be_lun->io_lock); 890264274Smav beio->num_bios_sent++; 891264274Smav if (last && len == 0) 892264274Smav beio->send_complete = 1; 893267877Smav mtx_unlock(&be_lun->io_lock); 894264274Smav 895264274Smav (*dev_data->csw->d_strategy)(bio); 896264274Smav } 897264274Smav} 898264274Smav 899264274Smavstatic void 900264274Smavctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 901264274Smav struct ctl_be_block_io *beio) 902264274Smav{ 903264274Smav union ctl_io *io; 904264274Smav struct ctl_be_block_devdata *dev_data; 905267515Smav struct ctl_ptr_len_flags *ptrlen; 906264274Smav struct scsi_unmap_desc *buf, *end; 907264274Smav uint64_t len; 908264274Smav 909264274Smav dev_data = &be_lun->backend.dev; 910264274Smav io = beio->io; 911264274Smav 912264274Smav DPRINTF("entered\n"); 913264274Smav 914264274Smav binuptime(&beio->ds_t0); 915267877Smav mtx_lock(&be_lun->io_lock); 916264274Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 917267877Smav mtx_unlock(&be_lun->io_lock); 918264274Smav 919264274Smav if (beio->io_offset == -1) { 920264274Smav beio->io_len = 0; 921267515Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 922267515Smav buf = (struct scsi_unmap_desc *)ptrlen->ptr; 923267515Smav end = buf + ptrlen->len / sizeof(*buf); 924264274Smav for (; buf < end; buf++) { 925264274Smav len = (uint64_t)scsi_4btoul(buf->length) * 926264274Smav be_lun->blocksize; 927264274Smav beio->io_len += len; 928264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 929264274Smav scsi_8btou64(buf->lba) * be_lun->blocksize, len, 930264283Smav (end - buf < 2) ? TRUE : FALSE); 931264274Smav } 932264274Smav } else 933264274Smav ctl_be_block_unmap_dev_range(be_lun, beio, 934264274Smav beio->io_offset, beio->io_len, TRUE); 935264274Smav} 936264274Smav 937264274Smavstatic void 938229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 939229997Sken struct ctl_be_block_io *beio) 940229997Sken{ 941267877Smav TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 942229997Sken int i; 943229997Sken struct bio *bio; 944229997Sken struct ctl_be_block_devdata *dev_data; 945229997Sken off_t cur_offset; 946229997Sken int max_iosize; 947229997Sken 948229997Sken DPRINTF("entered\n"); 949229997Sken 950229997Sken dev_data = &be_lun->backend.dev; 951229997Sken 952229997Sken /* 953229997Sken * We have to limit our I/O size to the maximum supported by the 954229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 955229997Sken * set it properly, use DFLTPHYS. 956229997Sken */ 957229997Sken max_iosize = dev_data->cdev->si_iosize_max; 958229997Sken if (max_iosize < PAGE_SIZE) 959229997Sken max_iosize = DFLTPHYS; 960229997Sken 961229997Sken cur_offset = beio->io_offset; 962229997Sken for (i = 0; i < beio->num_segs; i++) { 963229997Sken size_t cur_size; 964229997Sken uint8_t *cur_ptr; 965229997Sken 966229997Sken cur_size = beio->sg_segs[i].len; 967229997Sken cur_ptr = beio->sg_segs[i].addr; 968229997Sken 969229997Sken while (cur_size > 0) { 970229997Sken /* This can't fail, it's a blocking allocation. */ 971229997Sken bio = g_alloc_bio(); 972229997Sken 973229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 974229997Sken 975229997Sken bio->bio_cmd = beio->bio_cmd; 976229997Sken bio->bio_flags |= beio->bio_flags; 977229997Sken bio->bio_dev = dev_data->cdev; 978229997Sken bio->bio_caller1 = beio; 979229997Sken bio->bio_length = min(cur_size, max_iosize); 980229997Sken bio->bio_offset = cur_offset; 981229997Sken bio->bio_data = cur_ptr; 982229997Sken bio->bio_done = ctl_be_block_biodone; 983229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 984229997Sken 985229997Sken cur_offset += bio->bio_length; 986229997Sken cur_ptr += bio->bio_length; 987229997Sken cur_size -= bio->bio_length; 988229997Sken 989267877Smav TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 990229997Sken beio->num_bios_sent++; 991229997Sken } 992229997Sken } 993267877Smav binuptime(&beio->ds_t0); 994267877Smav mtx_lock(&be_lun->io_lock); 995267877Smav devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 996267877Smav beio->send_complete = 1; 997267877Smav mtx_unlock(&be_lun->io_lock); 998267877Smav 999267877Smav /* 1000267877Smav * Fire off all allocated requests! 1001267877Smav */ 1002267877Smav while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1003267877Smav TAILQ_REMOVE(&queue, bio, bio_queue); 1004267877Smav (*dev_data->csw->d_strategy)(bio); 1005267877Smav } 1006229997Sken} 1007229997Sken 1008229997Skenstatic void 1009264274Smavctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1010264274Smav{ 1011264274Smav union ctl_io *io; 1012264274Smav 1013264274Smav io = beio->io; 1014264274Smav ctl_free_beio(beio); 1015267641Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1016267641Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1017267641Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1018264274Smav ctl_config_write_done(io); 1019264274Smav return; 1020264274Smav } 1021264274Smav 1022264274Smav ctl_be_block_config_write(io); 1023264274Smav} 1024264274Smav 1025264274Smavstatic void 1026264274Smavctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1027264274Smav union ctl_io *io) 1028264274Smav{ 1029264274Smav struct ctl_be_block_io *beio; 1030264274Smav struct ctl_be_block_softc *softc; 1031267515Smav struct ctl_lba_len_flags *lbalen; 1032264274Smav uint64_t len_left, lba; 1033264274Smav int i, seglen; 1034264274Smav uint8_t *buf, *end; 1035264274Smav 1036264274Smav DPRINTF("entered\n"); 1037264274Smav 1038267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1039264274Smav softc = be_lun->softc; 1040267537Smav lbalen = ARGS(beio->io); 1041264274Smav 1042269622Smav if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR) || 1043269622Smav (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1044264274Smav ctl_free_beio(beio); 1045264274Smav ctl_set_invalid_field(&io->scsiio, 1046264274Smav /*sks_valid*/ 1, 1047264274Smav /*command*/ 1, 1048264274Smav /*field*/ 1, 1049264274Smav /*bit_valid*/ 0, 1050264274Smav /*bit*/ 0); 1051264274Smav ctl_config_write_done(io); 1052264274Smav return; 1053264274Smav } 1054264274Smav 1055264274Smav /* 1056264274Smav * If the I/O came down with an ordered or head of queue tag, set 1057264274Smav * the BIO_ORDERED attribute. For head of queue tags, that's 1058264274Smav * pretty much the best we can do. 1059264274Smav */ 1060264274Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1061264274Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1062264274Smav beio->bio_flags = BIO_ORDERED; 1063264274Smav 1064264274Smav switch (io->scsiio.tag_type) { 1065264274Smav case CTL_TAG_ORDERED: 1066264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1067264274Smav break; 1068264274Smav case CTL_TAG_HEAD_OF_QUEUE: 1069264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1070264274Smav break; 1071264274Smav case CTL_TAG_UNTAGGED: 1072264274Smav case CTL_TAG_SIMPLE: 1073264274Smav case CTL_TAG_ACA: 1074264274Smav default: 1075264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1076264274Smav break; 1077264274Smav } 1078264274Smav 1079269622Smav if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1080267515Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 1081267515Smav beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1082264274Smav beio->bio_cmd = BIO_DELETE; 1083264274Smav beio->ds_trans_type = DEVSTAT_FREE; 1084264274Smav 1085264274Smav be_lun->unmap(be_lun, beio); 1086264274Smav return; 1087264274Smav } 1088264274Smav 1089264274Smav beio->bio_cmd = BIO_WRITE; 1090264274Smav beio->ds_trans_type = DEVSTAT_WRITE; 1091264274Smav 1092264274Smav DPRINTF("WRITE SAME at LBA %jx len %u\n", 1093267515Smav (uintmax_t)lbalen->lba, lbalen->len); 1094264274Smav 1095267515Smav len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1096264274Smav for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1097264274Smav 1098264274Smav /* 1099264274Smav * Setup the S/G entry for this chunk. 1100264274Smav */ 1101264886Smav seglen = MIN(CTLBLK_MAX_SEG, len_left); 1102264274Smav seglen -= seglen % be_lun->blocksize; 1103264274Smav beio->sg_segs[i].len = seglen; 1104264274Smav beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1105264274Smav 1106264274Smav DPRINTF("segment %d addr %p len %zd\n", i, 1107264274Smav beio->sg_segs[i].addr, beio->sg_segs[i].len); 1108264274Smav 1109264274Smav beio->num_segs++; 1110264274Smav len_left -= seglen; 1111264274Smav 1112264274Smav buf = beio->sg_segs[i].addr; 1113264274Smav end = buf + seglen; 1114264274Smav for (; buf < end; buf += be_lun->blocksize) { 1115264274Smav memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1116267515Smav if (lbalen->flags & SWS_LBDATA) 1117267515Smav scsi_ulto4b(lbalen->lba + lba, buf); 1118264274Smav lba++; 1119264274Smav } 1120264274Smav } 1121264274Smav 1122267515Smav beio->io_offset = lbalen->lba * be_lun->blocksize; 1123264274Smav beio->io_len = lba * be_lun->blocksize; 1124264274Smav 1125264274Smav /* We can not do all in one run. Correct and schedule rerun. */ 1126264274Smav if (len_left > 0) { 1127267515Smav lbalen->lba += lba; 1128267515Smav lbalen->len -= lba; 1129264274Smav beio->beio_cont = ctl_be_block_cw_done_ws; 1130264274Smav } 1131264274Smav 1132264274Smav be_lun->dispatch(be_lun, beio); 1133264274Smav} 1134264274Smav 1135264274Smavstatic void 1136264274Smavctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1137264274Smav union ctl_io *io) 1138264274Smav{ 1139264274Smav struct ctl_be_block_io *beio; 1140264274Smav struct ctl_be_block_softc *softc; 1141267515Smav struct ctl_ptr_len_flags *ptrlen; 1142264274Smav 1143264274Smav DPRINTF("entered\n"); 1144264274Smav 1145267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1146264274Smav softc = be_lun->softc; 1147267515Smav ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1148264274Smav 1149269622Smav if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1150264274Smav ctl_free_beio(beio); 1151264274Smav ctl_set_invalid_field(&io->scsiio, 1152264274Smav /*sks_valid*/ 0, 1153264274Smav /*command*/ 1, 1154264274Smav /*field*/ 0, 1155264274Smav /*bit_valid*/ 0, 1156264274Smav /*bit*/ 0); 1157264274Smav ctl_config_write_done(io); 1158264274Smav return; 1159264274Smav } 1160264274Smav 1161264274Smav /* 1162264274Smav * If the I/O came down with an ordered or head of queue tag, set 1163264274Smav * the BIO_ORDERED attribute. For head of queue tags, that's 1164264274Smav * pretty much the best we can do. 1165264274Smav */ 1166264274Smav if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1167264274Smav || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1168264274Smav beio->bio_flags = BIO_ORDERED; 1169264274Smav 1170264274Smav switch (io->scsiio.tag_type) { 1171264274Smav case CTL_TAG_ORDERED: 1172264274Smav beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1173264274Smav break; 1174264274Smav case CTL_TAG_HEAD_OF_QUEUE: 1175264274Smav beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1176264274Smav break; 1177264274Smav case CTL_TAG_UNTAGGED: 1178264274Smav case CTL_TAG_SIMPLE: 1179264274Smav case CTL_TAG_ACA: 1180264274Smav default: 1181264274Smav beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1182264274Smav break; 1183264274Smav } 1184264274Smav 1185264274Smav beio->io_len = 0; 1186264274Smav beio->io_offset = -1; 1187264274Smav 1188264274Smav beio->bio_cmd = BIO_DELETE; 1189264274Smav beio->ds_trans_type = DEVSTAT_FREE; 1190264274Smav 1191267515Smav DPRINTF("UNMAP\n"); 1192264274Smav 1193264274Smav be_lun->unmap(be_lun, beio); 1194264274Smav} 1195264274Smav 1196264274Smavstatic void 1197264274Smavctl_be_block_cw_done(struct ctl_be_block_io *beio) 1198264274Smav{ 1199264274Smav union ctl_io *io; 1200264274Smav 1201264274Smav io = beio->io; 1202264274Smav ctl_free_beio(beio); 1203264274Smav ctl_config_write_done(io); 1204264274Smav} 1205264274Smav 1206264274Smavstatic void 1207229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1208229997Sken union ctl_io *io) 1209229997Sken{ 1210229997Sken struct ctl_be_block_io *beio; 1211229997Sken struct ctl_be_block_softc *softc; 1212229997Sken 1213229997Sken DPRINTF("entered\n"); 1214229997Sken 1215229997Sken softc = be_lun->softc; 1216229997Sken beio = ctl_alloc_beio(softc); 1217229997Sken beio->io = io; 1218229997Sken beio->lun = be_lun; 1219264274Smav beio->beio_cont = ctl_be_block_cw_done; 1220267519Smav PRIV(io)->ptr = (void *)beio; 1221229997Sken 1222229997Sken switch (io->scsiio.cdb[0]) { 1223229997Sken case SYNCHRONIZE_CACHE: 1224229997Sken case SYNCHRONIZE_CACHE_16: 1225249194Strasz beio->bio_cmd = BIO_FLUSH; 1226229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 1227229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1228229997Sken beio->io_len = 0; 1229229997Sken be_lun->lun_flush(be_lun, beio); 1230229997Sken break; 1231264274Smav case WRITE_SAME_10: 1232264274Smav case WRITE_SAME_16: 1233264274Smav ctl_be_block_cw_dispatch_ws(be_lun, io); 1234264274Smav break; 1235264274Smav case UNMAP: 1236264274Smav ctl_be_block_cw_dispatch_unmap(be_lun, io); 1237264274Smav break; 1238229997Sken default: 1239229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1240229997Sken break; 1241229997Sken } 1242229997Sken} 1243229997Sken 1244258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1245258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1246258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1247258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1248229997Sken 1249229997Skenstatic void 1250264886Smavctl_be_block_next(struct ctl_be_block_io *beio) 1251264886Smav{ 1252264886Smav struct ctl_be_block_lun *be_lun; 1253264886Smav union ctl_io *io; 1254264886Smav 1255264886Smav io = beio->io; 1256264886Smav be_lun = beio->lun; 1257264886Smav ctl_free_beio(beio); 1258267641Smav if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1259267641Smav ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1260267641Smav (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1261267537Smav ctl_data_submit_done(io); 1262264886Smav return; 1263264886Smav } 1264264886Smav 1265264886Smav io->io_hdr.status &= ~CTL_STATUS_MASK; 1266264886Smav io->io_hdr.status |= CTL_STATUS_NONE; 1267264886Smav 1268267877Smav mtx_lock(&be_lun->queue_lock); 1269264886Smav /* 1270264886Smav * XXX KDM make sure that links is okay to use at this point. 1271264886Smav * Otherwise, we either need to add another field to ctl_io_hdr, 1272264886Smav * or deal with resource allocation here. 1273264886Smav */ 1274264886Smav STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1275267877Smav mtx_unlock(&be_lun->queue_lock); 1276264886Smav 1277264886Smav taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1278264886Smav} 1279264886Smav 1280264886Smavstatic void 1281229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1282229997Sken union ctl_io *io) 1283229997Sken{ 1284229997Sken struct ctl_be_block_io *beio; 1285229997Sken struct ctl_be_block_softc *softc; 1286267537Smav struct ctl_lba_len_flags *lbalen; 1287267519Smav struct ctl_ptr_len_flags *bptrlen; 1288267519Smav uint64_t len_left, lbas; 1289229997Sken int i; 1290229997Sken 1291229997Sken softc = be_lun->softc; 1292229997Sken 1293229997Sken DPRINTF("entered\n"); 1294229997Sken 1295267537Smav lbalen = ARGS(io); 1296267537Smav if (lbalen->flags & CTL_LLF_WRITE) { 1297267537Smav SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1298267537Smav } else { 1299229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1300229997Sken } 1301229997Sken 1302229997Sken beio = ctl_alloc_beio(softc); 1303229997Sken beio->io = io; 1304229997Sken beio->lun = be_lun; 1305267519Smav bptrlen = PRIV(io); 1306267519Smav bptrlen->ptr = (void *)beio; 1307229997Sken 1308229997Sken /* 1309229997Sken * If the I/O came down with an ordered or head of queue tag, set 1310229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 1311229997Sken * pretty much the best we can do. 1312229997Sken * 1313229997Sken * XXX KDM we don't have a great way to easily know about the FUA 1314229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 1315229997Sken * pass that knowledge to the backend), and in any case we would 1316229997Sken * need to determine how to handle it. 1317229997Sken */ 1318229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 1319229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 1320229997Sken beio->bio_flags = BIO_ORDERED; 1321229997Sken 1322229997Sken switch (io->scsiio.tag_type) { 1323229997Sken case CTL_TAG_ORDERED: 1324229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1325229997Sken break; 1326229997Sken case CTL_TAG_HEAD_OF_QUEUE: 1327229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1328229997Sken break; 1329229997Sken case CTL_TAG_UNTAGGED: 1330229997Sken case CTL_TAG_SIMPLE: 1331229997Sken case CTL_TAG_ACA: 1332229997Sken default: 1333229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1334229997Sken break; 1335229997Sken } 1336229997Sken 1337267537Smav if (lbalen->flags & CTL_LLF_WRITE) { 1338267537Smav beio->bio_cmd = BIO_WRITE; 1339267537Smav beio->ds_trans_type = DEVSTAT_WRITE; 1340267537Smav } else { 1341229997Sken beio->bio_cmd = BIO_READ; 1342229997Sken beio->ds_trans_type = DEVSTAT_READ; 1343229997Sken } 1344229997Sken 1345264886Smav DPRINTF("%s at LBA %jx len %u @%ju\n", 1346229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1347267519Smav (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1348267537Smav if (lbalen->flags & CTL_LLF_COMPARE) 1349267537Smav lbas = CTLBLK_HALF_IO_SIZE; 1350267537Smav else 1351267537Smav lbas = CTLBLK_MAX_IO_SIZE; 1352267537Smav lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1353267519Smav beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1354267519Smav beio->io_len = lbas * be_lun->blocksize; 1355267519Smav bptrlen->len += lbas; 1356229997Sken 1357264886Smav for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1358264886Smav KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1359264886Smav i, CTLBLK_MAX_SEGS)); 1360229997Sken 1361229997Sken /* 1362229997Sken * Setup the S/G entry for this chunk. 1363229997Sken */ 1364264886Smav beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1365229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1366229997Sken 1367229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 1368229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 1369229997Sken 1370267537Smav /* Set up second segment for compare operation. */ 1371267537Smav if (lbalen->flags & CTL_LLF_COMPARE) { 1372267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1373267537Smav beio->sg_segs[i].len; 1374267537Smav beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1375267537Smav uma_zalloc(be_lun->lun_zone, M_WAITOK); 1376267537Smav } 1377267537Smav 1378229997Sken beio->num_segs++; 1379229997Sken len_left -= beio->sg_segs[i].len; 1380229997Sken } 1381267519Smav if (bptrlen->len < lbalen->len) 1382264886Smav beio->beio_cont = ctl_be_block_next; 1383264886Smav io->scsiio.be_move_done = ctl_be_block_move_done; 1384267537Smav /* For compare we have separate S/G lists for read and datamove. */ 1385267537Smav if (lbalen->flags & CTL_LLF_COMPARE) 1386267537Smav io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1387267537Smav else 1388267537Smav io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1389264886Smav io->scsiio.kern_data_len = beio->io_len; 1390264886Smav io->scsiio.kern_data_resid = 0; 1391264886Smav io->scsiio.kern_sg_entries = beio->num_segs; 1392264886Smav io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1393229997Sken 1394229997Sken /* 1395229997Sken * For the read case, we need to read the data into our buffers and 1396229997Sken * then we can send it back to the user. For the write case, we 1397229997Sken * need to get the data from the user first. 1398229997Sken */ 1399229997Sken if (beio->bio_cmd == BIO_READ) { 1400229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1401229997Sken be_lun->dispatch(be_lun, beio); 1402229997Sken } else { 1403229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1404229997Sken#ifdef CTL_TIME_IO 1405229997Sken getbintime(&io->io_hdr.dma_start_bt); 1406229997Sken#endif 1407229997Sken ctl_datamove(io); 1408229997Sken } 1409229997Sken} 1410229997Sken 1411229997Skenstatic void 1412229997Skenctl_be_block_worker(void *context, int pending) 1413229997Sken{ 1414229997Sken struct ctl_be_block_lun *be_lun; 1415229997Sken struct ctl_be_block_softc *softc; 1416229997Sken union ctl_io *io; 1417229997Sken 1418229997Sken be_lun = (struct ctl_be_block_lun *)context; 1419229997Sken softc = be_lun->softc; 1420229997Sken 1421229997Sken DPRINTF("entered\n"); 1422229997Sken 1423267877Smav mtx_lock(&be_lun->queue_lock); 1424229997Sken for (;;) { 1425229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1426229997Sken if (io != NULL) { 1427229997Sken struct ctl_be_block_io *beio; 1428229997Sken 1429229997Sken DPRINTF("datamove queue\n"); 1430229997Sken 1431229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1432229997Sken ctl_io_hdr, links); 1433229997Sken 1434267877Smav mtx_unlock(&be_lun->queue_lock); 1435229997Sken 1436267519Smav beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1437229997Sken 1438229997Sken be_lun->dispatch(be_lun, beio); 1439229997Sken 1440267877Smav mtx_lock(&be_lun->queue_lock); 1441229997Sken continue; 1442229997Sken } 1443229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1444229997Sken if (io != NULL) { 1445229997Sken 1446229997Sken DPRINTF("config write queue\n"); 1447229997Sken 1448229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1449229997Sken ctl_io_hdr, links); 1450229997Sken 1451267877Smav mtx_unlock(&be_lun->queue_lock); 1452229997Sken 1453229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1454229997Sken 1455267877Smav mtx_lock(&be_lun->queue_lock); 1456229997Sken continue; 1457229997Sken } 1458229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1459229997Sken if (io != NULL) { 1460229997Sken DPRINTF("input queue\n"); 1461229997Sken 1462229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1463229997Sken ctl_io_hdr, links); 1464267877Smav mtx_unlock(&be_lun->queue_lock); 1465229997Sken 1466229997Sken /* 1467229997Sken * We must drop the lock, since this routine and 1468229997Sken * its children may sleep. 1469229997Sken */ 1470229997Sken ctl_be_block_dispatch(be_lun, io); 1471229997Sken 1472267877Smav mtx_lock(&be_lun->queue_lock); 1473229997Sken continue; 1474229997Sken } 1475229997Sken 1476229997Sken /* 1477229997Sken * If we get here, there is no work left in the queues, so 1478229997Sken * just break out and let the task queue go to sleep. 1479229997Sken */ 1480229997Sken break; 1481229997Sken } 1482267877Smav mtx_unlock(&be_lun->queue_lock); 1483229997Sken} 1484229997Sken 1485229997Sken/* 1486229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1487229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1488229997Sken * thread. 1489229997Sken */ 1490229997Skenstatic int 1491229997Skenctl_be_block_submit(union ctl_io *io) 1492229997Sken{ 1493229997Sken struct ctl_be_block_lun *be_lun; 1494229997Sken struct ctl_be_lun *ctl_be_lun; 1495229997Sken 1496229997Sken DPRINTF("entered\n"); 1497229997Sken 1498229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1499229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1500229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1501229997Sken 1502229997Sken /* 1503229997Sken * Make sure we only get SCSI I/O. 1504229997Sken */ 1505229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1506229997Sken "%#x) encountered", io->io_hdr.io_type)); 1507229997Sken 1508267519Smav PRIV(io)->len = 0; 1509267519Smav 1510267877Smav mtx_lock(&be_lun->queue_lock); 1511229997Sken /* 1512229997Sken * XXX KDM make sure that links is okay to use at this point. 1513229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1514229997Sken * or deal with resource allocation here. 1515229997Sken */ 1516229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1517267877Smav mtx_unlock(&be_lun->queue_lock); 1518229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1519229997Sken 1520267514Smav return (CTL_RETVAL_COMPLETE); 1521229997Sken} 1522229997Sken 1523229997Skenstatic int 1524229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1525229997Sken int flag, struct thread *td) 1526229997Sken{ 1527229997Sken struct ctl_be_block_softc *softc; 1528229997Sken int error; 1529229997Sken 1530229997Sken softc = &backend_block_softc; 1531229997Sken 1532229997Sken error = 0; 1533229997Sken 1534229997Sken switch (cmd) { 1535229997Sken case CTL_LUN_REQ: { 1536229997Sken struct ctl_lun_req *lun_req; 1537229997Sken 1538229997Sken lun_req = (struct ctl_lun_req *)addr; 1539229997Sken 1540229997Sken switch (lun_req->reqtype) { 1541229997Sken case CTL_LUNREQ_CREATE: 1542229997Sken error = ctl_be_block_create(softc, lun_req); 1543229997Sken break; 1544229997Sken case CTL_LUNREQ_RM: 1545229997Sken error = ctl_be_block_rm(softc, lun_req); 1546229997Sken break; 1547232604Strasz case CTL_LUNREQ_MODIFY: 1548232604Strasz error = ctl_be_block_modify(softc, lun_req); 1549232604Strasz break; 1550229997Sken default: 1551229997Sken lun_req->status = CTL_LUN_ERROR; 1552229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1553229997Sken "%s: invalid LUN request type %d", __func__, 1554229997Sken lun_req->reqtype); 1555229997Sken break; 1556229997Sken } 1557229997Sken break; 1558229997Sken } 1559229997Sken default: 1560229997Sken error = ENOTTY; 1561229997Sken break; 1562229997Sken } 1563229997Sken 1564229997Sken return (error); 1565229997Sken} 1566229997Sken 1567229997Skenstatic int 1568229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1569229997Sken{ 1570229997Sken struct ctl_be_block_filedata *file_data; 1571229997Sken struct ctl_lun_create_params *params; 1572229997Sken struct vattr vattr; 1573229997Sken int error; 1574229997Sken 1575229997Sken error = 0; 1576229997Sken file_data = &be_lun->backend.file; 1577229997Sken params = &req->reqdata.create; 1578229997Sken 1579229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1580229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1581229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1582229997Sken 1583229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1584229997Sken if (error != 0) { 1585229997Sken snprintf(req->error_str, sizeof(req->error_str), 1586229997Sken "error calling VOP_GETATTR() for file %s", 1587229997Sken be_lun->dev_path); 1588229997Sken return (error); 1589229997Sken } 1590229997Sken 1591229997Sken /* 1592229997Sken * Verify that we have the ability to upgrade to exclusive 1593229997Sken * access on this file so we can trap errors at open instead 1594229997Sken * of reporting them during first access. 1595229997Sken */ 1596229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1597229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1598229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1599229997Sken error = EBADF; 1600229997Sken snprintf(req->error_str, sizeof(req->error_str), 1601229997Sken "error locking file %s", be_lun->dev_path); 1602229997Sken return (error); 1603229997Sken } 1604229997Sken } 1605229997Sken 1606229997Sken 1607229997Sken file_data->cred = crhold(curthread->td_ucred); 1608232604Strasz if (params->lun_size_bytes != 0) 1609232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1610232604Strasz else 1611232604Strasz be_lun->size_bytes = vattr.va_size; 1612229997Sken /* 1613229997Sken * We set the multi thread flag for file operations because all 1614229997Sken * filesystems (in theory) are capable of allowing multiple readers 1615229997Sken * of a file at once. So we want to get the maximum possible 1616229997Sken * concurrency. 1617229997Sken */ 1618229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1619229997Sken 1620229997Sken /* 1621229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1622229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1623229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1624229997Sken * may not work with other OSes as well. So just export a sector 1625229997Sken * size of 512 bytes, which should work with any OS or 1626229997Sken * application. Since our backing is a file, any block size will 1627229997Sken * work fine for the backing store. 1628229997Sken */ 1629229997Sken#if 0 1630229997Sken be_lun->blocksize= vattr.va_blocksize; 1631229997Sken#endif 1632229997Sken if (params->blocksize_bytes != 0) 1633229997Sken be_lun->blocksize = params->blocksize_bytes; 1634229997Sken else 1635229997Sken be_lun->blocksize = 512; 1636229997Sken 1637229997Sken /* 1638229997Sken * Sanity check. The media size has to be at least one 1639229997Sken * sector long. 1640229997Sken */ 1641229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1642229997Sken error = EINVAL; 1643229997Sken snprintf(req->error_str, sizeof(req->error_str), 1644229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1645229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1646229997Sken } 1647229997Sken return (error); 1648229997Sken} 1649229997Sken 1650229997Skenstatic int 1651229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1652229997Sken{ 1653229997Sken struct ctl_lun_create_params *params; 1654229997Sken struct vattr vattr; 1655229997Sken struct cdev *dev; 1656229997Sken struct cdevsw *devsw; 1657229997Sken int error; 1658264191Smav off_t ps, pss, po, pos; 1659229997Sken 1660229997Sken params = &req->reqdata.create; 1661229997Sken 1662229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1663229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1664229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1665229997Sken &be_lun->backend.dev.dev_ref); 1666229997Sken if (be_lun->backend.dev.csw == NULL) 1667229997Sken panic("Unable to retrieve device switch"); 1668269123Smav if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) 1669269123Smav be_lun->dispatch = ctl_be_block_dispatch_zvol; 1670269123Smav else 1671269123Smav be_lun->dispatch = ctl_be_block_dispatch_dev; 1672269123Smav be_lun->lun_flush = ctl_be_block_flush_dev; 1673269123Smav be_lun->unmap = ctl_be_block_unmap_dev; 1674229997Sken 1675229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1676229997Sken if (error) { 1677229997Sken snprintf(req->error_str, sizeof(req->error_str), 1678229997Sken "%s: error getting vnode attributes for device %s", 1679229997Sken __func__, be_lun->dev_path); 1680229997Sken return (error); 1681229997Sken } 1682229997Sken 1683229997Sken dev = be_lun->vn->v_rdev; 1684229997Sken devsw = dev->si_devsw; 1685229997Sken if (!devsw->d_ioctl) { 1686229997Sken snprintf(req->error_str, sizeof(req->error_str), 1687229997Sken "%s: no d_ioctl for device %s!", __func__, 1688229997Sken be_lun->dev_path); 1689229997Sken return (ENODEV); 1690229997Sken } 1691229997Sken 1692229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1693229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1694229997Sken curthread); 1695229997Sken if (error) { 1696229997Sken snprintf(req->error_str, sizeof(req->error_str), 1697229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1698229997Sken "on %s!", __func__, error, be_lun->dev_path); 1699229997Sken return (error); 1700229997Sken } 1701229997Sken 1702229997Sken /* 1703229997Sken * If the user has asked for a blocksize that is greater than the 1704229997Sken * backing device's blocksize, we can do it only if the blocksize 1705229997Sken * the user is asking for is an even multiple of the underlying 1706229997Sken * device's blocksize. 1707229997Sken */ 1708229997Sken if ((params->blocksize_bytes != 0) 1709229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1710229997Sken uint32_t bs_multiple, tmp_blocksize; 1711229997Sken 1712229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1713229997Sken 1714229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1715229997Sken 1716229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1717229997Sken be_lun->blocksize = params->blocksize_bytes; 1718229997Sken } else { 1719229997Sken snprintf(req->error_str, sizeof(req->error_str), 1720229997Sken "%s: requested blocksize %u is not an even " 1721229997Sken "multiple of backing device blocksize %u", 1722229997Sken __func__, params->blocksize_bytes, 1723229997Sken be_lun->blocksize); 1724229997Sken return (EINVAL); 1725229997Sken 1726229997Sken } 1727229997Sken } else if ((params->blocksize_bytes != 0) 1728229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1729229997Sken snprintf(req->error_str, sizeof(req->error_str), 1730229997Sken "%s: requested blocksize %u < backing device " 1731229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1732229997Sken be_lun->blocksize); 1733229997Sken return (EINVAL); 1734229997Sken } 1735229997Sken 1736229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1737229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1738229997Sken curthread); 1739229997Sken if (error) { 1740229997Sken snprintf(req->error_str, sizeof(req->error_str), 1741232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1742232604Strasz " ioctl on %s!", __func__, error, 1743232604Strasz be_lun->dev_path); 1744229997Sken return (error); 1745229997Sken } 1746229997Sken 1747232604Strasz if (params->lun_size_bytes != 0) { 1748232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1749232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1750232604Strasz "%s: requested LUN size %ju > backing device " 1751232604Strasz "size %ju", __func__, 1752232604Strasz (uintmax_t)params->lun_size_bytes, 1753232604Strasz (uintmax_t)be_lun->size_bytes); 1754232604Strasz return (EINVAL); 1755232604Strasz } 1756232604Strasz 1757232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1758232604Strasz } 1759232604Strasz 1760264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1761264191Smav (caddr_t)&ps, FREAD, curthread); 1762264191Smav if (error) 1763264191Smav ps = po = 0; 1764264191Smav else { 1765264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1766264191Smav (caddr_t)&po, FREAD, curthread); 1767264191Smav if (error) 1768264191Smav po = 0; 1769264191Smav } 1770264191Smav pss = ps / be_lun->blocksize; 1771264191Smav pos = po / be_lun->blocksize; 1772264191Smav if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1773264191Smav ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1774264191Smav be_lun->pblockexp = fls(pss) - 1; 1775264191Smav be_lun->pblockoff = (pss - pos) % pss; 1776264191Smav } 1777264191Smav 1778229997Sken return (0); 1779229997Sken} 1780229997Sken 1781229997Skenstatic int 1782229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1783229997Sken{ 1784229997Sken DROP_GIANT(); 1785229997Sken if (be_lun->vn) { 1786229997Sken int flags = FREAD | FWRITE; 1787229997Sken 1788229997Sken switch (be_lun->dev_type) { 1789229997Sken case CTL_BE_BLOCK_DEV: 1790229997Sken if (be_lun->backend.dev.csw) { 1791229997Sken dev_relthread(be_lun->backend.dev.cdev, 1792229997Sken be_lun->backend.dev.dev_ref); 1793229997Sken be_lun->backend.dev.csw = NULL; 1794229997Sken be_lun->backend.dev.cdev = NULL; 1795229997Sken } 1796229997Sken break; 1797229997Sken case CTL_BE_BLOCK_FILE: 1798229997Sken break; 1799229997Sken case CTL_BE_BLOCK_NONE: 1800258871Strasz break; 1801229997Sken default: 1802229997Sken panic("Unexpected backend type."); 1803229997Sken break; 1804229997Sken } 1805229997Sken 1806229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1807229997Sken be_lun->vn = NULL; 1808229997Sken 1809229997Sken switch (be_lun->dev_type) { 1810229997Sken case CTL_BE_BLOCK_DEV: 1811229997Sken break; 1812229997Sken case CTL_BE_BLOCK_FILE: 1813229997Sken if (be_lun->backend.file.cred != NULL) { 1814229997Sken crfree(be_lun->backend.file.cred); 1815229997Sken be_lun->backend.file.cred = NULL; 1816229997Sken } 1817229997Sken break; 1818229997Sken case CTL_BE_BLOCK_NONE: 1819258871Strasz break; 1820229997Sken default: 1821229997Sken panic("Unexpected backend type."); 1822229997Sken break; 1823229997Sken } 1824229997Sken } 1825229997Sken PICKUP_GIANT(); 1826229997Sken 1827229997Sken return (0); 1828229997Sken} 1829229997Sken 1830229997Skenstatic int 1831229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1832229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1833229997Sken{ 1834229997Sken struct nameidata nd; 1835229997Sken int flags; 1836229997Sken int error; 1837229997Sken 1838229997Sken /* 1839229997Sken * XXX KDM allow a read-only option? 1840229997Sken */ 1841229997Sken flags = FREAD | FWRITE; 1842229997Sken error = 0; 1843229997Sken 1844229997Sken if (rootvnode == NULL) { 1845229997Sken snprintf(req->error_str, sizeof(req->error_str), 1846229997Sken "%s: Root filesystem is not mounted", __func__); 1847229997Sken return (1); 1848229997Sken } 1849229997Sken 1850229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1851229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1852229997Sken VREF(rootvnode); 1853229997Sken } 1854229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1855229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1856229997Sken VREF(rootvnode); 1857229997Sken } 1858229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1859229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1860229997Sken VREF(rootvnode); 1861229997Sken } 1862229997Sken 1863229997Sken again: 1864229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1865229997Sken error = vn_open(&nd, &flags, 0, NULL); 1866229997Sken if (error) { 1867229997Sken /* 1868229997Sken * This is the only reasonable guess we can make as far as 1869229997Sken * path if the user doesn't give us a fully qualified path. 1870229997Sken * If they want to specify a file, they need to specify the 1871229997Sken * full path. 1872229997Sken */ 1873229997Sken if (be_lun->dev_path[0] != '/') { 1874229997Sken char *dev_path = "/dev/"; 1875229997Sken char *dev_name; 1876229997Sken 1877229997Sken /* Try adding device path at beginning of name */ 1878229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1879229997Sken + strlen(dev_path) + 1, 1880229997Sken M_CTLBLK, M_WAITOK); 1881229997Sken if (dev_name) { 1882229997Sken sprintf(dev_name, "%s%s", dev_path, 1883229997Sken be_lun->dev_path); 1884229997Sken free(be_lun->dev_path, M_CTLBLK); 1885229997Sken be_lun->dev_path = dev_name; 1886229997Sken goto again; 1887229997Sken } 1888229997Sken } 1889229997Sken snprintf(req->error_str, sizeof(req->error_str), 1890229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1891229997Sken return (error); 1892229997Sken } 1893229997Sken 1894229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1895229997Sken 1896229997Sken be_lun->vn = nd.ni_vp; 1897229997Sken 1898229997Sken /* We only support disks and files. */ 1899229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1900229997Sken error = ctl_be_block_open_dev(be_lun, req); 1901229997Sken } else if (be_lun->vn->v_type == VREG) { 1902229997Sken error = ctl_be_block_open_file(be_lun, req); 1903229997Sken } else { 1904229997Sken error = EINVAL; 1905229997Sken snprintf(req->error_str, sizeof(req->error_str), 1906258871Strasz "%s is not a disk or plain file", be_lun->dev_path); 1907229997Sken } 1908229997Sken VOP_UNLOCK(be_lun->vn, 0); 1909229997Sken 1910229997Sken if (error != 0) { 1911229997Sken ctl_be_block_close(be_lun); 1912229997Sken return (error); 1913229997Sken } 1914229997Sken 1915229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1916229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1917229997Sken 1918229997Sken return (0); 1919229997Sken} 1920229997Sken 1921229997Skenstatic int 1922229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1923229997Sken{ 1924229997Sken struct ctl_be_block_lun *be_lun; 1925229997Sken struct ctl_lun_create_params *params; 1926267481Smav char num_thread_str[16]; 1927229997Sken char tmpstr[32]; 1928267481Smav char *value; 1929264274Smav int retval, num_threads, unmap; 1930267481Smav int tmp_num_threads; 1931229997Sken 1932229997Sken params = &req->reqdata.create; 1933229997Sken retval = 0; 1934229997Sken 1935229997Sken num_threads = cbb_num_threads; 1936229997Sken 1937229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1938229997Sken 1939229997Sken be_lun->softc = softc; 1940229997Sken STAILQ_INIT(&be_lun->input_queue); 1941229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1942229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1943229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1944267877Smav mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 1945267877Smav mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 1946268280Smav ctl_init_opts(&be_lun->ctl_be_lun.options, 1947268280Smav req->num_be_args, req->kern_be_args); 1948229997Sken 1949264886Smav be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 1950256995Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1951229997Sken 1952229997Sken if (be_lun->lun_zone == NULL) { 1953229997Sken snprintf(req->error_str, sizeof(req->error_str), 1954229997Sken "%s: error allocating UMA zone", __func__); 1955229997Sken goto bailout_error; 1956229997Sken } 1957229997Sken 1958229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1959229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1960229997Sken else 1961229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1962229997Sken 1963229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1964268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 1965267499Smav if (value == NULL) { 1966229997Sken snprintf(req->error_str, sizeof(req->error_str), 1967229997Sken "%s: no file argument specified", __func__); 1968229997Sken goto bailout_error; 1969229997Sken } 1970267499Smav be_lun->dev_path = strdup(value, M_CTLBLK); 1971229997Sken 1972229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1973229997Sken if (retval != 0) { 1974229997Sken retval = 0; 1975229997Sken goto bailout_error; 1976229997Sken } 1977229997Sken 1978229997Sken /* 1979229997Sken * Tell the user the size of the file/device. 1980229997Sken */ 1981229997Sken params->lun_size_bytes = be_lun->size_bytes; 1982229997Sken 1983229997Sken /* 1984229997Sken * The maximum LBA is the size - 1. 1985229997Sken */ 1986229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1987229997Sken } else { 1988229997Sken /* 1989229997Sken * For processor devices, we don't have any size. 1990229997Sken */ 1991229997Sken be_lun->blocksize = 0; 1992264191Smav be_lun->pblockexp = 0; 1993264191Smav be_lun->pblockoff = 0; 1994229997Sken be_lun->size_blocks = 0; 1995229997Sken be_lun->size_bytes = 0; 1996229997Sken be_lun->ctl_be_lun.maxlba = 0; 1997229997Sken params->lun_size_bytes = 0; 1998229997Sken 1999229997Sken /* 2000229997Sken * Default to just 1 thread for processor devices. 2001229997Sken */ 2002229997Sken num_threads = 1; 2003229997Sken } 2004229997Sken 2005229997Sken /* 2006229997Sken * XXX This searching loop might be refactored to be combined with 2007229997Sken * the loop above, 2008229997Sken */ 2009268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2010267481Smav if (value != NULL) { 2011267481Smav tmp_num_threads = strtol(value, NULL, 0); 2012229997Sken 2013267481Smav /* 2014267481Smav * We don't let the user specify less than one 2015267481Smav * thread, but hope he's clueful enough not to 2016267481Smav * specify 1000 threads. 2017267481Smav */ 2018267481Smav if (tmp_num_threads < 1) { 2019267481Smav snprintf(req->error_str, sizeof(req->error_str), 2020267481Smav "%s: invalid number of threads %s", 2021267481Smav __func__, num_thread_str); 2022267481Smav goto bailout_error; 2023229997Sken } 2024267481Smav num_threads = tmp_num_threads; 2025229997Sken } 2026267481Smav unmap = 0; 2027268280Smav value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2028267481Smav if (value != NULL && strcmp(value, "on") == 0) 2029267481Smav unmap = 1; 2030229997Sken 2031229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2032229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2033264274Smav if (unmap) 2034264274Smav be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2035229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 2036229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2037264191Smav be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2038264191Smav be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2039229997Sken /* Tell the user the blocksize we ended up using */ 2040229997Sken params->blocksize_bytes = be_lun->blocksize; 2041229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2042229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2043229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2044229997Sken } else 2045229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 2046229997Sken 2047229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2048229997Sken be_lun->ctl_be_lun.lun_config_status = 2049229997Sken ctl_be_block_lun_config_status; 2050229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2051229997Sken 2052229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2053229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2054229997Sken softc->num_luns); 2055229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2056229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2057229997Sken sizeof(tmpstr))); 2058229997Sken 2059229997Sken /* Tell the user what we used for a serial number */ 2060229997Sken strncpy((char *)params->serial_num, tmpstr, 2061229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 2062229997Sken } else { 2063229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 2064229997Sken params->serial_num, 2065229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2066229997Sken sizeof(params->serial_num))); 2067229997Sken } 2068229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2069229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2070229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2071229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2072229997Sken sizeof(tmpstr))); 2073229997Sken 2074229997Sken /* Tell the user what we used for a device ID */ 2075229997Sken strncpy((char *)params->device_id, tmpstr, 2076229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 2077229997Sken } else { 2078229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 2079229997Sken params->device_id, 2080229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2081229997Sken sizeof(params->device_id))); 2082229997Sken } 2083229997Sken 2084229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2085229997Sken 2086229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2087229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2088229997Sken 2089229997Sken if (be_lun->io_taskqueue == NULL) { 2090229997Sken snprintf(req->error_str, sizeof(req->error_str), 2091229997Sken "%s: Unable to create taskqueue", __func__); 2092229997Sken goto bailout_error; 2093229997Sken } 2094229997Sken 2095229997Sken /* 2096229997Sken * Note that we start the same number of threads by default for 2097229997Sken * both the file case and the block device case. For the file 2098229997Sken * case, we need multiple threads to allow concurrency, because the 2099229997Sken * vnode interface is designed to be a blocking interface. For the 2100229997Sken * block device case, ZFS zvols at least will block the caller's 2101229997Sken * context in many instances, and so we need multiple threads to 2102229997Sken * overcome that problem. Other block devices don't need as many 2103229997Sken * threads, but they shouldn't cause too many problems. 2104229997Sken * 2105229997Sken * If the user wants to just have a single thread for a block 2106229997Sken * device, he can specify that when the LUN is created, or change 2107229997Sken * the tunable/sysctl to alter the default number of threads. 2108229997Sken */ 2109229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2110229997Sken /*num threads*/num_threads, 2111229997Sken /*priority*/PWAIT, 2112229997Sken /*thread name*/ 2113229997Sken "%s taskq", be_lun->lunname); 2114229997Sken 2115229997Sken if (retval != 0) 2116229997Sken goto bailout_error; 2117229997Sken 2118229997Sken be_lun->num_threads = num_threads; 2119229997Sken 2120229997Sken mtx_lock(&softc->lock); 2121229997Sken softc->num_luns++; 2122229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2123229997Sken 2124229997Sken mtx_unlock(&softc->lock); 2125229997Sken 2126229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 2127229997Sken if (retval != 0) { 2128229997Sken mtx_lock(&softc->lock); 2129229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2130229997Sken links); 2131229997Sken softc->num_luns--; 2132229997Sken mtx_unlock(&softc->lock); 2133229997Sken snprintf(req->error_str, sizeof(req->error_str), 2134229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 2135229997Sken "details", __func__, retval); 2136229997Sken retval = 0; 2137229997Sken goto bailout_error; 2138229997Sken } 2139229997Sken 2140229997Sken mtx_lock(&softc->lock); 2141229997Sken 2142229997Sken /* 2143229997Sken * Tell the config_status routine that we're waiting so it won't 2144229997Sken * clean up the LUN in the event of an error. 2145229997Sken */ 2146229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2147229997Sken 2148229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2149229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2150229997Sken if (retval == EINTR) 2151229997Sken break; 2152229997Sken } 2153229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2154229997Sken 2155229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2156229997Sken snprintf(req->error_str, sizeof(req->error_str), 2157229997Sken "%s: LUN configuration error, see dmesg for details", 2158229997Sken __func__); 2159229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2160229997Sken links); 2161229997Sken softc->num_luns--; 2162229997Sken mtx_unlock(&softc->lock); 2163229997Sken goto bailout_error; 2164229997Sken } else { 2165229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2166229997Sken } 2167229997Sken 2168229997Sken mtx_unlock(&softc->lock); 2169229997Sken 2170229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2171229997Sken be_lun->blocksize, 2172229997Sken DEVSTAT_ALL_SUPPORTED, 2173229997Sken be_lun->ctl_be_lun.lun_type 2174229997Sken | DEVSTAT_TYPE_IF_OTHER, 2175229997Sken DEVSTAT_PRIORITY_OTHER); 2176229997Sken 2177229997Sken 2178229997Sken req->status = CTL_LUN_OK; 2179229997Sken 2180229997Sken return (retval); 2181229997Sken 2182229997Skenbailout_error: 2183229997Sken req->status = CTL_LUN_ERROR; 2184229997Sken 2185267429Smav if (be_lun->io_taskqueue != NULL) 2186267429Smav taskqueue_free(be_lun->io_taskqueue); 2187229997Sken ctl_be_block_close(be_lun); 2188267429Smav if (be_lun->dev_path != NULL) 2189267429Smav free(be_lun->dev_path, M_CTLBLK); 2190267429Smav if (be_lun->lun_zone != NULL) 2191267429Smav uma_zdestroy(be_lun->lun_zone); 2192268280Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2193267877Smav mtx_destroy(&be_lun->queue_lock); 2194267877Smav mtx_destroy(&be_lun->io_lock); 2195229997Sken free(be_lun, M_CTLBLK); 2196229997Sken 2197229997Sken return (retval); 2198229997Sken} 2199229997Sken 2200229997Skenstatic int 2201229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2202229997Sken{ 2203229997Sken struct ctl_lun_rm_params *params; 2204229997Sken struct ctl_be_block_lun *be_lun; 2205229997Sken int retval; 2206229997Sken 2207229997Sken params = &req->reqdata.rm; 2208229997Sken 2209229997Sken mtx_lock(&softc->lock); 2210229997Sken 2211229997Sken be_lun = NULL; 2212229997Sken 2213229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2214229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2215229997Sken break; 2216229997Sken } 2217229997Sken mtx_unlock(&softc->lock); 2218229997Sken 2219229997Sken if (be_lun == NULL) { 2220229997Sken snprintf(req->error_str, sizeof(req->error_str), 2221229997Sken "%s: LUN %u is not managed by the block backend", 2222229997Sken __func__, params->lun_id); 2223229997Sken goto bailout_error; 2224229997Sken } 2225229997Sken 2226229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2227229997Sken 2228229997Sken if (retval != 0) { 2229229997Sken snprintf(req->error_str, sizeof(req->error_str), 2230229997Sken "%s: error %d returned from ctl_disable_lun() for " 2231229997Sken "LUN %d", __func__, retval, params->lun_id); 2232229997Sken goto bailout_error; 2233229997Sken 2234229997Sken } 2235229997Sken 2236229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2237229997Sken if (retval != 0) { 2238229997Sken snprintf(req->error_str, sizeof(req->error_str), 2239229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 2240229997Sken "LUN %d", __func__, retval, params->lun_id); 2241229997Sken goto bailout_error; 2242229997Sken } 2243229997Sken 2244229997Sken mtx_lock(&softc->lock); 2245229997Sken 2246229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2247229997Sken 2248229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2249229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2250229997Sken if (retval == EINTR) 2251229997Sken break; 2252229997Sken } 2253229997Sken 2254229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2255229997Sken 2256229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2257229997Sken snprintf(req->error_str, sizeof(req->error_str), 2258229997Sken "%s: interrupted waiting for LUN to be freed", 2259229997Sken __func__); 2260229997Sken mtx_unlock(&softc->lock); 2261229997Sken goto bailout_error; 2262229997Sken } 2263229997Sken 2264229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2265229997Sken 2266229997Sken softc->num_luns--; 2267229997Sken mtx_unlock(&softc->lock); 2268229997Sken 2269229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2270229997Sken 2271229997Sken taskqueue_free(be_lun->io_taskqueue); 2272229997Sken 2273229997Sken ctl_be_block_close(be_lun); 2274229997Sken 2275229997Sken if (be_lun->disk_stats != NULL) 2276229997Sken devstat_remove_entry(be_lun->disk_stats); 2277229997Sken 2278229997Sken uma_zdestroy(be_lun->lun_zone); 2279229997Sken 2280268280Smav ctl_free_opts(&be_lun->ctl_be_lun.options); 2281229997Sken free(be_lun->dev_path, M_CTLBLK); 2282267877Smav mtx_destroy(&be_lun->queue_lock); 2283267877Smav mtx_destroy(&be_lun->io_lock); 2284229997Sken free(be_lun, M_CTLBLK); 2285229997Sken 2286229997Sken req->status = CTL_LUN_OK; 2287229997Sken 2288229997Sken return (0); 2289229997Sken 2290229997Skenbailout_error: 2291229997Sken 2292229997Sken req->status = CTL_LUN_ERROR; 2293229997Sken 2294229997Sken return (0); 2295229997Sken} 2296229997Sken 2297232604Straszstatic int 2298232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2299232604Strasz struct ctl_lun_req *req) 2300232604Strasz{ 2301232604Strasz struct vattr vattr; 2302232604Strasz int error; 2303232604Strasz struct ctl_lun_modify_params *params; 2304232604Strasz 2305232604Strasz params = &req->reqdata.modify; 2306232604Strasz 2307232604Strasz if (params->lun_size_bytes != 0) { 2308232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2309232604Strasz } else { 2310232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2311232604Strasz if (error != 0) { 2312232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2313232604Strasz "error calling VOP_GETATTR() for file %s", 2314232604Strasz be_lun->dev_path); 2315232604Strasz return (error); 2316232604Strasz } 2317232604Strasz 2318232604Strasz be_lun->size_bytes = vattr.va_size; 2319232604Strasz } 2320232604Strasz 2321232604Strasz return (0); 2322232604Strasz} 2323232604Strasz 2324232604Straszstatic int 2325232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2326232604Strasz struct ctl_lun_req *req) 2327232604Strasz{ 2328232604Strasz struct cdev *dev; 2329232604Strasz struct cdevsw *devsw; 2330232604Strasz int error; 2331232604Strasz struct ctl_lun_modify_params *params; 2332232604Strasz uint64_t size_bytes; 2333232604Strasz 2334232604Strasz params = &req->reqdata.modify; 2335232604Strasz 2336232604Strasz dev = be_lun->vn->v_rdev; 2337232604Strasz devsw = dev->si_devsw; 2338232604Strasz if (!devsw->d_ioctl) { 2339232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2340232604Strasz "%s: no d_ioctl for device %s!", __func__, 2341232604Strasz be_lun->dev_path); 2342232604Strasz return (ENODEV); 2343232604Strasz } 2344232604Strasz 2345232604Strasz error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 2346232604Strasz (caddr_t)&size_bytes, FREAD, 2347232604Strasz curthread); 2348232604Strasz if (error) { 2349232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2350232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 2351232604Strasz "on %s!", __func__, error, be_lun->dev_path); 2352232604Strasz return (error); 2353232604Strasz } 2354232604Strasz 2355232604Strasz if (params->lun_size_bytes != 0) { 2356232604Strasz if (params->lun_size_bytes > size_bytes) { 2357232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2358232604Strasz "%s: requested LUN size %ju > backing device " 2359232604Strasz "size %ju", __func__, 2360232604Strasz (uintmax_t)params->lun_size_bytes, 2361232604Strasz (uintmax_t)size_bytes); 2362232604Strasz return (EINVAL); 2363232604Strasz } 2364232604Strasz 2365232604Strasz be_lun->size_bytes = params->lun_size_bytes; 2366232604Strasz } else { 2367232604Strasz be_lun->size_bytes = size_bytes; 2368232604Strasz } 2369232604Strasz 2370232604Strasz return (0); 2371232604Strasz} 2372232604Strasz 2373232604Straszstatic int 2374232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2375232604Strasz{ 2376232604Strasz struct ctl_lun_modify_params *params; 2377232604Strasz struct ctl_be_block_lun *be_lun; 2378241896Skib int error; 2379232604Strasz 2380232604Strasz params = &req->reqdata.modify; 2381232604Strasz 2382232604Strasz mtx_lock(&softc->lock); 2383232604Strasz 2384232604Strasz be_lun = NULL; 2385232604Strasz 2386232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2387232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2388232604Strasz break; 2389232604Strasz } 2390232604Strasz mtx_unlock(&softc->lock); 2391232604Strasz 2392232604Strasz if (be_lun == NULL) { 2393232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2394232604Strasz "%s: LUN %u is not managed by the block backend", 2395232604Strasz __func__, params->lun_id); 2396232604Strasz goto bailout_error; 2397232604Strasz } 2398232604Strasz 2399232604Strasz if (params->lun_size_bytes != 0) { 2400232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2401232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2402232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2403232604Strasz params->lun_size_bytes, be_lun->blocksize); 2404232604Strasz goto bailout_error; 2405232604Strasz } 2406232604Strasz } 2407232604Strasz 2408232604Strasz vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2409232604Strasz 2410232604Strasz if (be_lun->vn->v_type == VREG) 2411232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2412232604Strasz else 2413232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2414232604Strasz 2415232604Strasz VOP_UNLOCK(be_lun->vn, 0); 2416232604Strasz 2417232604Strasz if (error != 0) 2418232604Strasz goto bailout_error; 2419232604Strasz 2420232604Strasz be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2421232604Strasz 2422232604Strasz /* 2423232604Strasz * The maximum LBA is the size - 1. 2424232604Strasz * 2425232604Strasz * XXX: Note that this field is being updated without locking, 2426232604Strasz * which might cause problems on 32-bit architectures. 2427232604Strasz */ 2428232604Strasz be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2429232604Strasz ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2430232604Strasz 2431232604Strasz /* Tell the user the exact size we ended up using */ 2432232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2433232604Strasz 2434232604Strasz req->status = CTL_LUN_OK; 2435232604Strasz 2436232604Strasz return (0); 2437232604Strasz 2438232604Straszbailout_error: 2439232604Strasz req->status = CTL_LUN_ERROR; 2440232604Strasz 2441232604Strasz return (0); 2442232604Strasz} 2443232604Strasz 2444229997Skenstatic void 2445229997Skenctl_be_block_lun_shutdown(void *be_lun) 2446229997Sken{ 2447229997Sken struct ctl_be_block_lun *lun; 2448229997Sken struct ctl_be_block_softc *softc; 2449229997Sken 2450229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2451229997Sken 2452229997Sken softc = lun->softc; 2453229997Sken 2454229997Sken mtx_lock(&softc->lock); 2455229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2456229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2457229997Sken wakeup(lun); 2458229997Sken mtx_unlock(&softc->lock); 2459229997Sken 2460229997Sken} 2461229997Sken 2462229997Skenstatic void 2463229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2464229997Sken{ 2465229997Sken struct ctl_be_block_lun *lun; 2466229997Sken struct ctl_be_block_softc *softc; 2467229997Sken 2468229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2469229997Sken softc = lun->softc; 2470229997Sken 2471229997Sken if (status == CTL_LUN_CONFIG_OK) { 2472229997Sken mtx_lock(&softc->lock); 2473229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2474229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2475229997Sken wakeup(lun); 2476229997Sken mtx_unlock(&softc->lock); 2477229997Sken 2478229997Sken /* 2479229997Sken * We successfully added the LUN, attempt to enable it. 2480229997Sken */ 2481229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2482229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2483229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2484229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2485229997Sken __func__); 2486229997Sken } 2487229997Sken } 2488229997Sken 2489229997Sken return; 2490229997Sken } 2491229997Sken 2492229997Sken 2493229997Sken mtx_lock(&softc->lock); 2494229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2495229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2496229997Sken wakeup(lun); 2497229997Sken mtx_unlock(&softc->lock); 2498229997Sken} 2499229997Sken 2500229997Sken 2501229997Skenstatic int 2502229997Skenctl_be_block_config_write(union ctl_io *io) 2503229997Sken{ 2504229997Sken struct ctl_be_block_lun *be_lun; 2505229997Sken struct ctl_be_lun *ctl_be_lun; 2506229997Sken int retval; 2507229997Sken 2508229997Sken retval = 0; 2509229997Sken 2510229997Sken DPRINTF("entered\n"); 2511229997Sken 2512229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2513229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2514229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2515229997Sken 2516229997Sken switch (io->scsiio.cdb[0]) { 2517229997Sken case SYNCHRONIZE_CACHE: 2518229997Sken case SYNCHRONIZE_CACHE_16: 2519264274Smav case WRITE_SAME_10: 2520264274Smav case WRITE_SAME_16: 2521264274Smav case UNMAP: 2522229997Sken /* 2523229997Sken * The upper level CTL code will filter out any CDBs with 2524229997Sken * the immediate bit set and return the proper error. 2525229997Sken * 2526229997Sken * We don't really need to worry about what LBA range the 2527229997Sken * user asked to be synced out. When they issue a sync 2528229997Sken * cache command, we'll sync out the whole thing. 2529229997Sken */ 2530267877Smav mtx_lock(&be_lun->queue_lock); 2531229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2532229997Sken links); 2533267877Smav mtx_unlock(&be_lun->queue_lock); 2534229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2535229997Sken break; 2536229997Sken case START_STOP_UNIT: { 2537229997Sken struct scsi_start_stop_unit *cdb; 2538229997Sken 2539229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2540229997Sken 2541229997Sken if (cdb->how & SSS_START) 2542229997Sken retval = ctl_start_lun(ctl_be_lun); 2543229997Sken else { 2544229997Sken retval = ctl_stop_lun(ctl_be_lun); 2545229997Sken /* 2546229997Sken * XXX KDM Copan-specific offline behavior. 2547229997Sken * Figure out a reasonable way to port this? 2548229997Sken */ 2549229997Sken#ifdef NEEDTOPORT 2550229997Sken if ((retval == 0) 2551229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2552229997Sken retval = ctl_lun_offline(ctl_be_lun); 2553229997Sken#endif 2554229997Sken } 2555229997Sken 2556229997Sken /* 2557229997Sken * In general, the above routines should not fail. They 2558229997Sken * just set state for the LUN. So we've got something 2559229997Sken * pretty wrong here if we can't start or stop the LUN. 2560229997Sken */ 2561229997Sken if (retval != 0) { 2562229997Sken ctl_set_internal_failure(&io->scsiio, 2563229997Sken /*sks_valid*/ 1, 2564229997Sken /*retry_count*/ 0xf051); 2565229997Sken retval = CTL_RETVAL_COMPLETE; 2566229997Sken } else { 2567229997Sken ctl_set_success(&io->scsiio); 2568229997Sken } 2569229997Sken ctl_config_write_done(io); 2570229997Sken break; 2571229997Sken } 2572229997Sken default: 2573229997Sken ctl_set_invalid_opcode(&io->scsiio); 2574229997Sken ctl_config_write_done(io); 2575229997Sken retval = CTL_RETVAL_COMPLETE; 2576229997Sken break; 2577229997Sken } 2578229997Sken 2579229997Sken return (retval); 2580229997Sken 2581229997Sken} 2582229997Sken 2583229997Skenstatic int 2584229997Skenctl_be_block_config_read(union ctl_io *io) 2585229997Sken{ 2586229997Sken return (0); 2587229997Sken} 2588229997Sken 2589229997Skenstatic int 2590229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2591229997Sken{ 2592229997Sken struct ctl_be_block_lun *lun; 2593229997Sken int retval; 2594229997Sken 2595229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2596229997Sken retval = 0; 2597229997Sken 2598268283Smav retval = sbuf_printf(sb, "\t<num_threads>"); 2599229997Sken 2600229997Sken if (retval != 0) 2601229997Sken goto bailout; 2602229997Sken 2603229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2604229997Sken 2605229997Sken if (retval != 0) 2606229997Sken goto bailout; 2607229997Sken 2608268283Smav retval = sbuf_printf(sb, "</num_threads>\n"); 2609229997Sken 2610229997Skenbailout: 2611229997Sken 2612229997Sken return (retval); 2613229997Sken} 2614229997Sken 2615229997Skenint 2616229997Skenctl_be_block_init(void) 2617229997Sken{ 2618229997Sken struct ctl_be_block_softc *softc; 2619229997Sken int retval; 2620229997Sken 2621229997Sken softc = &backend_block_softc; 2622229997Sken retval = 0; 2623229997Sken 2624267877Smav mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2625264020Strasz beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2626264020Strasz NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2627229997Sken STAILQ_INIT(&softc->disk_list); 2628229997Sken STAILQ_INIT(&softc->lun_list); 2629229997Sken 2630229997Sken return (retval); 2631229997Sken} 2632