ctl_backend_block.c revision 264191
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 264191 2014-04-06 10:13:14Z mav $"); 44229997Sken 45229997Sken#include <sys/param.h> 46229997Sken#include <sys/systm.h> 47229997Sken#include <sys/kernel.h> 48229997Sken#include <sys/types.h> 49229997Sken#include <sys/kthread.h> 50229997Sken#include <sys/bio.h> 51229997Sken#include <sys/fcntl.h> 52229997Sken#include <sys/lock.h> 53229997Sken#include <sys/mutex.h> 54229997Sken#include <sys/condvar.h> 55229997Sken#include <sys/malloc.h> 56229997Sken#include <sys/conf.h> 57229997Sken#include <sys/ioccom.h> 58229997Sken#include <sys/queue.h> 59229997Sken#include <sys/sbuf.h> 60229997Sken#include <sys/endian.h> 61229997Sken#include <sys/uio.h> 62229997Sken#include <sys/buf.h> 63229997Sken#include <sys/taskqueue.h> 64229997Sken#include <sys/vnode.h> 65229997Sken#include <sys/namei.h> 66229997Sken#include <sys/mount.h> 67229997Sken#include <sys/disk.h> 68229997Sken#include <sys/fcntl.h> 69229997Sken#include <sys/filedesc.h> 70229997Sken#include <sys/proc.h> 71229997Sken#include <sys/pcpu.h> 72229997Sken#include <sys/module.h> 73229997Sken#include <sys/sdt.h> 74229997Sken#include <sys/devicestat.h> 75229997Sken#include <sys/sysctl.h> 76229997Sken 77229997Sken#include <geom/geom.h> 78229997Sken 79229997Sken#include <cam/cam.h> 80229997Sken#include <cam/scsi/scsi_all.h> 81229997Sken#include <cam/scsi/scsi_da.h> 82229997Sken#include <cam/ctl/ctl_io.h> 83229997Sken#include <cam/ctl/ctl.h> 84229997Sken#include <cam/ctl/ctl_backend.h> 85229997Sken#include <cam/ctl/ctl_frontend_internal.h> 86229997Sken#include <cam/ctl/ctl_ioctl.h> 87229997Sken#include <cam/ctl/ctl_scsi_all.h> 88229997Sken#include <cam/ctl/ctl_error.h> 89229997Sken 90229997Sken/* 91229997Sken * The idea here is that we'll allocate enough S/G space to hold a 16MB 92229997Sken * I/O. If we get an I/O larger than that, we'll reject it. 93229997Sken */ 94229997Sken#define CTLBLK_MAX_IO_SIZE (16 * 1024 * 1024) 95229997Sken#define CTLBLK_MAX_SEGS (CTLBLK_MAX_IO_SIZE / MAXPHYS) + 1 96229997Sken 97229997Sken#ifdef CTLBLK_DEBUG 98229997Sken#define DPRINTF(fmt, args...) \ 99229997Sken printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 100229997Sken#else 101229997Sken#define DPRINTF(fmt, args...) do {} while(0) 102229997Sken#endif 103229997Sken 104229997SkenSDT_PROVIDER_DEFINE(cbb); 105229997Sken 106229997Skentypedef enum { 107229997Sken CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 108229997Sken CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 109229997Sken CTL_BE_BLOCK_LUN_WAITING = 0x04, 110229997Sken CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 111229997Sken} ctl_be_block_lun_flags; 112229997Sken 113229997Skentypedef enum { 114229997Sken CTL_BE_BLOCK_NONE, 115229997Sken CTL_BE_BLOCK_DEV, 116229997Sken CTL_BE_BLOCK_FILE 117229997Sken} ctl_be_block_type; 118229997Sken 119229997Skenstruct ctl_be_block_devdata { 120229997Sken struct cdev *cdev; 121229997Sken struct cdevsw *csw; 122229997Sken int dev_ref; 123229997Sken}; 124229997Sken 125229997Skenstruct ctl_be_block_filedata { 126229997Sken struct ucred *cred; 127229997Sken}; 128229997Sken 129229997Skenunion ctl_be_block_bedata { 130229997Sken struct ctl_be_block_devdata dev; 131229997Sken struct ctl_be_block_filedata file; 132229997Sken}; 133229997Sken 134229997Skenstruct ctl_be_block_io; 135229997Skenstruct ctl_be_block_lun; 136229997Sken 137229997Skentypedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 138229997Sken struct ctl_be_block_io *beio); 139229997Sken 140229997Sken/* 141229997Sken * Backend LUN structure. There is a 1:1 mapping between a block device 142229997Sken * and a backend block LUN, and between a backend block LUN and a CTL LUN. 143229997Sken */ 144229997Skenstruct ctl_be_block_lun { 145229997Sken struct ctl_block_disk *disk; 146229997Sken char lunname[32]; 147229997Sken char *dev_path; 148229997Sken ctl_be_block_type dev_type; 149229997Sken struct vnode *vn; 150229997Sken union ctl_be_block_bedata backend; 151229997Sken cbb_dispatch_t dispatch; 152229997Sken cbb_dispatch_t lun_flush; 153229997Sken struct mtx lock; 154229997Sken uma_zone_t lun_zone; 155229997Sken uint64_t size_blocks; 156229997Sken uint64_t size_bytes; 157229997Sken uint32_t blocksize; 158229997Sken int blocksize_shift; 159264191Smav uint16_t pblockexp; 160264191Smav uint16_t pblockoff; 161229997Sken struct ctl_be_block_softc *softc; 162229997Sken struct devstat *disk_stats; 163229997Sken ctl_be_block_lun_flags flags; 164229997Sken STAILQ_ENTRY(ctl_be_block_lun) links; 165229997Sken struct ctl_be_lun ctl_be_lun; 166229997Sken struct taskqueue *io_taskqueue; 167229997Sken struct task io_task; 168229997Sken int num_threads; 169229997Sken STAILQ_HEAD(, ctl_io_hdr) input_queue; 170229997Sken STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 171229997Sken STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 172229997Sken}; 173229997Sken 174229997Sken/* 175229997Sken * Overall softc structure for the block backend module. 176229997Sken */ 177229997Skenstruct ctl_be_block_softc { 178229997Sken struct mtx lock; 179229997Sken int num_disks; 180229997Sken STAILQ_HEAD(, ctl_block_disk) disk_list; 181229997Sken int num_luns; 182229997Sken STAILQ_HEAD(, ctl_be_block_lun) lun_list; 183229997Sken}; 184229997Sken 185229997Skenstatic struct ctl_be_block_softc backend_block_softc; 186229997Sken 187229997Sken/* 188229997Sken * Per-I/O information. 189229997Sken */ 190229997Skenstruct ctl_be_block_io { 191229997Sken union ctl_io *io; 192229997Sken struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 193229997Sken struct iovec xiovecs[CTLBLK_MAX_SEGS]; 194229997Sken int bio_cmd; 195229997Sken int bio_flags; 196229997Sken int num_segs; 197229997Sken int num_bios_sent; 198229997Sken int num_bios_done; 199229997Sken int send_complete; 200229997Sken int num_errors; 201229997Sken struct bintime ds_t0; 202229997Sken devstat_tag_type ds_tag_type; 203229997Sken devstat_trans_flags ds_trans_type; 204229997Sken uint64_t io_len; 205229997Sken uint64_t io_offset; 206229997Sken struct ctl_be_block_softc *softc; 207229997Sken struct ctl_be_block_lun *lun; 208229997Sken}; 209229997Sken 210229997Skenstatic int cbb_num_threads = 14; 211229997SkenTUNABLE_INT("kern.cam.ctl.block.num_threads", &cbb_num_threads); 212229997SkenSYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 213229997Sken "CAM Target Layer Block Backend"); 214229997SkenSYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RW, 215229997Sken &cbb_num_threads, 0, "Number of threads per backing file"); 216229997Sken 217229997Skenstatic struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 218229997Skenstatic void ctl_free_beio(struct ctl_be_block_io *beio); 219229997Skenstatic void ctl_complete_beio(struct ctl_be_block_io *beio); 220229997Skenstatic int ctl_be_block_move_done(union ctl_io *io); 221229997Skenstatic void ctl_be_block_biodone(struct bio *bio); 222229997Skenstatic void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 223229997Sken struct ctl_be_block_io *beio); 224229997Skenstatic void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 225229997Sken struct ctl_be_block_io *beio); 226229997Skenstatic void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 227229997Sken struct ctl_be_block_io *beio); 228229997Skenstatic void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 229229997Sken struct ctl_be_block_io *beio); 230229997Skenstatic void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 231229997Sken union ctl_io *io); 232229997Skenstatic void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 233229997Sken union ctl_io *io); 234229997Skenstatic void ctl_be_block_worker(void *context, int pending); 235229997Skenstatic int ctl_be_block_submit(union ctl_io *io); 236229997Skenstatic int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 237229997Sken int flag, struct thread *td); 238229997Skenstatic int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 239229997Sken struct ctl_lun_req *req); 240229997Skenstatic int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 241229997Sken struct ctl_lun_req *req); 242229997Skenstatic int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 243229997Skenstatic int ctl_be_block_open(struct ctl_be_block_softc *softc, 244229997Sken struct ctl_be_block_lun *be_lun, 245229997Sken struct ctl_lun_req *req); 246229997Skenstatic int ctl_be_block_create(struct ctl_be_block_softc *softc, 247229997Sken struct ctl_lun_req *req); 248229997Skenstatic int ctl_be_block_rm(struct ctl_be_block_softc *softc, 249229997Sken struct ctl_lun_req *req); 250232604Straszstatic int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 251232604Strasz struct ctl_lun_req *req); 252232604Straszstatic int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 253232604Strasz struct ctl_lun_req *req); 254232604Straszstatic int ctl_be_block_modify(struct ctl_be_block_softc *softc, 255232604Strasz struct ctl_lun_req *req); 256229997Skenstatic void ctl_be_block_lun_shutdown(void *be_lun); 257229997Skenstatic void ctl_be_block_lun_config_status(void *be_lun, 258229997Sken ctl_lun_config_status status); 259229997Skenstatic int ctl_be_block_config_write(union ctl_io *io); 260229997Skenstatic int ctl_be_block_config_read(union ctl_io *io); 261229997Skenstatic int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 262229997Skenint ctl_be_block_init(void); 263229997Sken 264229997Skenstatic struct ctl_backend_driver ctl_be_block_driver = 265229997Sken{ 266230334Sken .name = "block", 267230334Sken .flags = CTL_BE_FLAG_HAS_CONFIG, 268230334Sken .init = ctl_be_block_init, 269230334Sken .data_submit = ctl_be_block_submit, 270230334Sken .data_move_done = ctl_be_block_move_done, 271230334Sken .config_read = ctl_be_block_config_read, 272230334Sken .config_write = ctl_be_block_config_write, 273230334Sken .ioctl = ctl_be_block_ioctl, 274230334Sken .lun_info = ctl_be_block_lun_info 275229997Sken}; 276229997Sken 277229997SkenMALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 278229997SkenCTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 279229997Sken 280264020Straszstatic uma_zone_t beio_zone; 281264020Strasz 282229997Skenstatic struct ctl_be_block_io * 283229997Skenctl_alloc_beio(struct ctl_be_block_softc *softc) 284229997Sken{ 285229997Sken struct ctl_be_block_io *beio; 286229997Sken 287264020Strasz beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 288264020Strasz beio->softc = softc; 289229997Sken return (beio); 290229997Sken} 291229997Sken 292229997Skenstatic void 293229997Skenctl_free_beio(struct ctl_be_block_io *beio) 294229997Sken{ 295229997Sken int duplicate_free; 296229997Sken int i; 297229997Sken 298229997Sken duplicate_free = 0; 299229997Sken 300229997Sken for (i = 0; i < beio->num_segs; i++) { 301229997Sken if (beio->sg_segs[i].addr == NULL) 302229997Sken duplicate_free++; 303229997Sken 304229997Sken uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 305229997Sken beio->sg_segs[i].addr = NULL; 306229997Sken } 307229997Sken 308229997Sken if (duplicate_free > 0) { 309229997Sken printf("%s: %d duplicate frees out of %d segments\n", __func__, 310229997Sken duplicate_free, beio->num_segs); 311229997Sken } 312229997Sken 313264020Strasz uma_zfree(beio_zone, beio); 314229997Sken} 315229997Sken 316229997Skenstatic void 317229997Skenctl_complete_beio(struct ctl_be_block_io *beio) 318229997Sken{ 319229997Sken union ctl_io *io; 320229997Sken int io_len; 321229997Sken 322229997Sken io = beio->io; 323229997Sken 324229997Sken if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) 325229997Sken io_len = beio->io_len; 326229997Sken else 327229997Sken io_len = 0; 328229997Sken 329229997Sken devstat_end_transaction(beio->lun->disk_stats, 330229997Sken /*bytes*/ io_len, 331229997Sken beio->ds_tag_type, 332229997Sken beio->ds_trans_type, 333229997Sken /*now*/ NULL, 334229997Sken /*then*/&beio->ds_t0); 335229997Sken 336229997Sken ctl_free_beio(beio); 337229997Sken ctl_done(io); 338229997Sken} 339229997Sken 340229997Skenstatic int 341229997Skenctl_be_block_move_done(union ctl_io *io) 342229997Sken{ 343229997Sken struct ctl_be_block_io *beio; 344229997Sken struct ctl_be_block_lun *be_lun; 345229997Sken#ifdef CTL_TIME_IO 346229997Sken struct bintime cur_bt; 347229997Sken#endif 348229997Sken 349229997Sken beio = (struct ctl_be_block_io *) 350229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 351229997Sken 352229997Sken be_lun = beio->lun; 353229997Sken 354229997Sken DPRINTF("entered\n"); 355229997Sken 356229997Sken#ifdef CTL_TIME_IO 357229997Sken getbintime(&cur_bt); 358229997Sken bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 359229997Sken bintime_add(&io->io_hdr.dma_bt, &cur_bt); 360229997Sken io->io_hdr.num_dmas++; 361229997Sken#endif 362229997Sken 363229997Sken /* 364229997Sken * We set status at this point for read commands, and write 365229997Sken * commands with errors. 366229997Sken */ 367229997Sken if ((beio->bio_cmd == BIO_READ) 368229997Sken && (io->io_hdr.port_status == 0) 369229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 370229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) 371229997Sken ctl_set_success(&io->scsiio); 372229997Sken else if ((io->io_hdr.port_status != 0) 373229997Sken && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) 374229997Sken && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 375229997Sken /* 376229997Sken * For hardware error sense keys, the sense key 377229997Sken * specific value is defined to be a retry count, 378229997Sken * but we use it to pass back an internal FETD 379229997Sken * error code. XXX KDM Hopefully the FETD is only 380229997Sken * using 16 bits for an error code, since that's 381229997Sken * all the space we have in the sks field. 382229997Sken */ 383229997Sken ctl_set_internal_failure(&io->scsiio, 384229997Sken /*sks_valid*/ 1, 385229997Sken /*retry_count*/ 386229997Sken io->io_hdr.port_status); 387229997Sken } 388229997Sken 389229997Sken /* 390229997Sken * If this is a read, or a write with errors, it is done. 391229997Sken */ 392229997Sken if ((beio->bio_cmd == BIO_READ) 393229997Sken || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 394229997Sken || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 395229997Sken ctl_complete_beio(beio); 396229997Sken return (0); 397229997Sken } 398229997Sken 399229997Sken /* 400229997Sken * At this point, we have a write and the DMA completed 401229997Sken * successfully. We now have to queue it to the task queue to 402229997Sken * execute the backend I/O. That is because we do blocking 403229997Sken * memory allocations, and in the file backing case, blocking I/O. 404229997Sken * This move done routine is generally called in the SIM's 405229997Sken * interrupt context, and therefore we cannot block. 406229997Sken */ 407229997Sken mtx_lock(&be_lun->lock); 408229997Sken /* 409229997Sken * XXX KDM make sure that links is okay to use at this point. 410229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 411229997Sken * or deal with resource allocation here. 412229997Sken */ 413229997Sken STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 414229997Sken mtx_unlock(&be_lun->lock); 415229997Sken 416229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 417229997Sken 418229997Sken return (0); 419229997Sken} 420229997Sken 421229997Skenstatic void 422229997Skenctl_be_block_biodone(struct bio *bio) 423229997Sken{ 424229997Sken struct ctl_be_block_io *beio; 425229997Sken struct ctl_be_block_lun *be_lun; 426229997Sken union ctl_io *io; 427261538Smav int error; 428229997Sken 429229997Sken beio = bio->bio_caller1; 430229997Sken be_lun = beio->lun; 431229997Sken io = beio->io; 432229997Sken 433229997Sken DPRINTF("entered\n"); 434229997Sken 435261538Smav error = bio->bio_error; 436229997Sken mtx_lock(&be_lun->lock); 437261538Smav if (error != 0) 438229997Sken beio->num_errors++; 439229997Sken 440229997Sken beio->num_bios_done++; 441229997Sken 442229997Sken /* 443229997Sken * XXX KDM will this cause WITNESS to complain? Holding a lock 444229997Sken * during the free might cause it to complain. 445229997Sken */ 446229997Sken g_destroy_bio(bio); 447229997Sken 448229997Sken /* 449229997Sken * If the send complete bit isn't set, or we aren't the last I/O to 450229997Sken * complete, then we're done. 451229997Sken */ 452229997Sken if ((beio->send_complete == 0) 453229997Sken || (beio->num_bios_done < beio->num_bios_sent)) { 454229997Sken mtx_unlock(&be_lun->lock); 455229997Sken return; 456229997Sken } 457229997Sken 458229997Sken /* 459229997Sken * At this point, we've verified that we are the last I/O to 460229997Sken * complete, so it's safe to drop the lock. 461229997Sken */ 462229997Sken mtx_unlock(&be_lun->lock); 463229997Sken 464229997Sken /* 465229997Sken * If there are any errors from the backing device, we fail the 466229997Sken * entire I/O with a medium error. 467229997Sken */ 468229997Sken if (beio->num_errors > 0) { 469261538Smav if (error == EOPNOTSUPP) { 470261538Smav ctl_set_invalid_opcode(&io->scsiio); 471261538Smav } else if (beio->bio_cmd == BIO_FLUSH) { 472229997Sken /* XXX KDM is there is a better error here? */ 473229997Sken ctl_set_internal_failure(&io->scsiio, 474229997Sken /*sks_valid*/ 1, 475229997Sken /*retry_count*/ 0xbad2); 476229997Sken } else 477229997Sken ctl_set_medium_error(&io->scsiio); 478229997Sken ctl_complete_beio(beio); 479229997Sken return; 480229997Sken } 481229997Sken 482229997Sken /* 483229997Sken * If this is a write or a flush, we're all done. 484229997Sken * If this is a read, we can now send the data to the user. 485229997Sken */ 486229997Sken if ((beio->bio_cmd == BIO_WRITE) 487229997Sken || (beio->bio_cmd == BIO_FLUSH)) { 488229997Sken ctl_set_success(&io->scsiio); 489229997Sken ctl_complete_beio(beio); 490229997Sken } else { 491229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 492229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 493229997Sken io->scsiio.kern_data_len = beio->io_len; 494229997Sken io->scsiio.kern_total_len = beio->io_len; 495229997Sken io->scsiio.kern_rel_offset = 0; 496229997Sken io->scsiio.kern_data_resid = 0; 497229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 498229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 499229997Sken#ifdef CTL_TIME_IO 500229997Sken getbintime(&io->io_hdr.dma_start_bt); 501229997Sken#endif 502229997Sken ctl_datamove(io); 503229997Sken } 504229997Sken} 505229997Sken 506229997Skenstatic void 507229997Skenctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 508229997Sken struct ctl_be_block_io *beio) 509229997Sken{ 510229997Sken union ctl_io *io; 511229997Sken struct mount *mountpoint; 512241896Skib int error, lock_flags; 513229997Sken 514229997Sken DPRINTF("entered\n"); 515229997Sken 516229997Sken io = beio->io; 517229997Sken 518229997Sken (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 519229997Sken 520229997Sken if (MNT_SHARED_WRITES(mountpoint) 521229997Sken || ((mountpoint == NULL) 522229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 523229997Sken lock_flags = LK_SHARED; 524229997Sken else 525229997Sken lock_flags = LK_EXCLUSIVE; 526229997Sken 527229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 528229997Sken 529229997Sken binuptime(&beio->ds_t0); 530229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 531229997Sken 532229997Sken error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 533229997Sken VOP_UNLOCK(be_lun->vn, 0); 534229997Sken 535229997Sken vn_finished_write(mountpoint); 536229997Sken 537229997Sken if (error == 0) 538229997Sken ctl_set_success(&io->scsiio); 539229997Sken else { 540229997Sken /* XXX KDM is there is a better error here? */ 541229997Sken ctl_set_internal_failure(&io->scsiio, 542229997Sken /*sks_valid*/ 1, 543229997Sken /*retry_count*/ 0xbad1); 544229997Sken } 545229997Sken 546229997Sken ctl_complete_beio(beio); 547229997Sken} 548229997Sken 549258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 550258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 551258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 552258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 553229997Sken 554229997Skenstatic void 555229997Skenctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 556229997Sken struct ctl_be_block_io *beio) 557229997Sken{ 558229997Sken struct ctl_be_block_filedata *file_data; 559229997Sken union ctl_io *io; 560229997Sken struct uio xuio; 561229997Sken struct iovec *xiovec; 562241896Skib int flags; 563229997Sken int error, i; 564229997Sken 565229997Sken DPRINTF("entered\n"); 566229997Sken 567229997Sken file_data = &be_lun->backend.file; 568229997Sken io = beio->io; 569229997Sken flags = beio->bio_flags; 570229997Sken 571229997Sken if (beio->bio_cmd == BIO_READ) { 572229997Sken SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 573229997Sken } else { 574229997Sken SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 575229997Sken } 576229997Sken 577229997Sken bzero(&xuio, sizeof(xuio)); 578229997Sken if (beio->bio_cmd == BIO_READ) 579229997Sken xuio.uio_rw = UIO_READ; 580229997Sken else 581229997Sken xuio.uio_rw = UIO_WRITE; 582229997Sken 583229997Sken xuio.uio_offset = beio->io_offset; 584229997Sken xuio.uio_resid = beio->io_len; 585229997Sken xuio.uio_segflg = UIO_SYSSPACE; 586229997Sken xuio.uio_iov = beio->xiovecs; 587229997Sken xuio.uio_iovcnt = beio->num_segs; 588229997Sken xuio.uio_td = curthread; 589229997Sken 590229997Sken for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 591229997Sken xiovec->iov_base = beio->sg_segs[i].addr; 592229997Sken xiovec->iov_len = beio->sg_segs[i].len; 593229997Sken } 594229997Sken 595229997Sken if (beio->bio_cmd == BIO_READ) { 596229997Sken vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 597229997Sken 598229997Sken binuptime(&beio->ds_t0); 599229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 600229997Sken 601229997Sken /* 602229997Sken * UFS pays attention to IO_DIRECT for reads. If the 603229997Sken * DIRECTIO option is configured into the kernel, it calls 604229997Sken * ffs_rawread(). But that only works for single-segment 605229997Sken * uios with user space addresses. In our case, with a 606229997Sken * kernel uio, it still reads into the buffer cache, but it 607229997Sken * will just try to release the buffer from the cache later 608229997Sken * on in ffs_read(). 609229997Sken * 610229997Sken * ZFS does not pay attention to IO_DIRECT for reads. 611229997Sken * 612229997Sken * UFS does not pay attention to IO_SYNC for reads. 613229997Sken * 614229997Sken * ZFS pays attention to IO_SYNC (which translates into the 615229997Sken * Solaris define FRSYNC for zfs_read()) for reads. It 616229997Sken * attempts to sync the file before reading. 617229997Sken * 618229997Sken * So, to attempt to provide some barrier semantics in the 619229997Sken * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 620229997Sken */ 621229997Sken error = VOP_READ(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 622229997Sken (IO_DIRECT|IO_SYNC) : 0, file_data->cred); 623229997Sken 624229997Sken VOP_UNLOCK(be_lun->vn, 0); 625229997Sken } else { 626229997Sken struct mount *mountpoint; 627229997Sken int lock_flags; 628229997Sken 629229997Sken (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 630229997Sken 631229997Sken if (MNT_SHARED_WRITES(mountpoint) 632229997Sken || ((mountpoint == NULL) 633229997Sken && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 634229997Sken lock_flags = LK_SHARED; 635229997Sken else 636229997Sken lock_flags = LK_EXCLUSIVE; 637229997Sken 638229997Sken vn_lock(be_lun->vn, lock_flags | LK_RETRY); 639229997Sken 640229997Sken binuptime(&beio->ds_t0); 641229997Sken devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 642229997Sken 643229997Sken /* 644229997Sken * UFS pays attention to IO_DIRECT for writes. The write 645229997Sken * is done asynchronously. (Normally the write would just 646229997Sken * get put into cache. 647229997Sken * 648229997Sken * UFS pays attention to IO_SYNC for writes. It will 649229997Sken * attempt to write the buffer out synchronously if that 650229997Sken * flag is set. 651229997Sken * 652229997Sken * ZFS does not pay attention to IO_DIRECT for writes. 653229997Sken * 654229997Sken * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 655229997Sken * for writes. It will flush the transaction from the 656229997Sken * cache before returning. 657229997Sken * 658229997Sken * So if we've got the BIO_ORDERED flag set, we want 659229997Sken * IO_SYNC in either the UFS or ZFS case. 660229997Sken */ 661229997Sken error = VOP_WRITE(be_lun->vn, &xuio, (flags & BIO_ORDERED) ? 662229997Sken IO_SYNC : 0, file_data->cred); 663229997Sken VOP_UNLOCK(be_lun->vn, 0); 664229997Sken 665229997Sken vn_finished_write(mountpoint); 666229997Sken } 667229997Sken 668229997Sken /* 669229997Sken * If we got an error, set the sense data to "MEDIUM ERROR" and 670229997Sken * return the I/O to the user. 671229997Sken */ 672229997Sken if (error != 0) { 673229997Sken char path_str[32]; 674229997Sken 675229997Sken ctl_scsi_path_string(io, path_str, sizeof(path_str)); 676229997Sken /* 677229997Sken * XXX KDM ZFS returns ENOSPC when the underlying 678229997Sken * filesystem fills up. What kind of SCSI error should we 679229997Sken * return for that? 680229997Sken */ 681229997Sken printf("%s%s command returned errno %d\n", path_str, 682229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 683229997Sken ctl_set_medium_error(&io->scsiio); 684229997Sken ctl_complete_beio(beio); 685229997Sken return; 686229997Sken } 687229997Sken 688229997Sken /* 689229997Sken * If this is a write, we're all done. 690229997Sken * If this is a read, we can now send the data to the user. 691229997Sken */ 692229997Sken if (beio->bio_cmd == BIO_WRITE) { 693229997Sken ctl_set_success(&io->scsiio); 694229997Sken SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 695229997Sken ctl_complete_beio(beio); 696229997Sken } else { 697229997Sken SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 698229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 699229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 700229997Sken io->scsiio.kern_data_len = beio->io_len; 701229997Sken io->scsiio.kern_total_len = beio->io_len; 702229997Sken io->scsiio.kern_rel_offset = 0; 703229997Sken io->scsiio.kern_data_resid = 0; 704229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 705229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 706229997Sken#ifdef CTL_TIME_IO 707229997Sken getbintime(&io->io_hdr.dma_start_bt); 708229997Sken#endif 709229997Sken ctl_datamove(io); 710229997Sken } 711229997Sken} 712229997Sken 713229997Skenstatic void 714229997Skenctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 715229997Sken struct ctl_be_block_io *beio) 716229997Sken{ 717229997Sken struct bio *bio; 718229997Sken union ctl_io *io; 719229997Sken struct ctl_be_block_devdata *dev_data; 720229997Sken 721229997Sken dev_data = &be_lun->backend.dev; 722229997Sken io = beio->io; 723229997Sken 724229997Sken DPRINTF("entered\n"); 725229997Sken 726229997Sken /* This can't fail, it's a blocking allocation. */ 727229997Sken bio = g_alloc_bio(); 728229997Sken 729229997Sken bio->bio_cmd = BIO_FLUSH; 730229997Sken bio->bio_flags |= BIO_ORDERED; 731229997Sken bio->bio_dev = dev_data->cdev; 732229997Sken bio->bio_offset = 0; 733229997Sken bio->bio_data = 0; 734229997Sken bio->bio_done = ctl_be_block_biodone; 735229997Sken bio->bio_caller1 = beio; 736229997Sken bio->bio_pblkno = 0; 737229997Sken 738229997Sken /* 739229997Sken * We don't need to acquire the LUN lock here, because we are only 740229997Sken * sending one bio, and so there is no other context to synchronize 741229997Sken * with. 742229997Sken */ 743229997Sken beio->num_bios_sent = 1; 744229997Sken beio->send_complete = 1; 745229997Sken 746229997Sken binuptime(&beio->ds_t0); 747229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 748229997Sken 749229997Sken (*dev_data->csw->d_strategy)(bio); 750229997Sken} 751229997Sken 752229997Skenstatic void 753229997Skenctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 754229997Sken struct ctl_be_block_io *beio) 755229997Sken{ 756229997Sken int i; 757229997Sken struct bio *bio; 758229997Sken struct ctl_be_block_devdata *dev_data; 759229997Sken off_t cur_offset; 760229997Sken int max_iosize; 761229997Sken 762229997Sken DPRINTF("entered\n"); 763229997Sken 764229997Sken dev_data = &be_lun->backend.dev; 765229997Sken 766229997Sken /* 767229997Sken * We have to limit our I/O size to the maximum supported by the 768229997Sken * backend device. Hopefully it is MAXPHYS. If the driver doesn't 769229997Sken * set it properly, use DFLTPHYS. 770229997Sken */ 771229997Sken max_iosize = dev_data->cdev->si_iosize_max; 772229997Sken if (max_iosize < PAGE_SIZE) 773229997Sken max_iosize = DFLTPHYS; 774229997Sken 775229997Sken cur_offset = beio->io_offset; 776229997Sken 777229997Sken /* 778229997Sken * XXX KDM need to accurately reflect the number of I/Os outstanding 779229997Sken * to a device. 780229997Sken */ 781229997Sken binuptime(&beio->ds_t0); 782229997Sken devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 783229997Sken 784229997Sken for (i = 0; i < beio->num_segs; i++) { 785229997Sken size_t cur_size; 786229997Sken uint8_t *cur_ptr; 787229997Sken 788229997Sken cur_size = beio->sg_segs[i].len; 789229997Sken cur_ptr = beio->sg_segs[i].addr; 790229997Sken 791229997Sken while (cur_size > 0) { 792229997Sken /* This can't fail, it's a blocking allocation. */ 793229997Sken bio = g_alloc_bio(); 794229997Sken 795229997Sken KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 796229997Sken 797229997Sken bio->bio_cmd = beio->bio_cmd; 798229997Sken bio->bio_flags |= beio->bio_flags; 799229997Sken bio->bio_dev = dev_data->cdev; 800229997Sken bio->bio_caller1 = beio; 801229997Sken bio->bio_length = min(cur_size, max_iosize); 802229997Sken bio->bio_offset = cur_offset; 803229997Sken bio->bio_data = cur_ptr; 804229997Sken bio->bio_done = ctl_be_block_biodone; 805229997Sken bio->bio_pblkno = cur_offset / be_lun->blocksize; 806229997Sken 807229997Sken cur_offset += bio->bio_length; 808229997Sken cur_ptr += bio->bio_length; 809229997Sken cur_size -= bio->bio_length; 810229997Sken 811229997Sken /* 812229997Sken * Make sure we set the complete bit just before we 813229997Sken * issue the last bio so we don't wind up with a 814229997Sken * race. 815229997Sken * 816229997Sken * Use the LUN mutex here instead of a combination 817229997Sken * of atomic variables for simplicity. 818229997Sken * 819229997Sken * XXX KDM we could have a per-IO lock, but that 820229997Sken * would cause additional per-IO setup and teardown 821229997Sken * overhead. Hopefully there won't be too much 822229997Sken * contention on the LUN lock. 823229997Sken */ 824229997Sken mtx_lock(&be_lun->lock); 825229997Sken 826229997Sken beio->num_bios_sent++; 827229997Sken 828229997Sken if ((i == beio->num_segs - 1) 829229997Sken && (cur_size == 0)) 830229997Sken beio->send_complete = 1; 831229997Sken 832229997Sken mtx_unlock(&be_lun->lock); 833229997Sken 834229997Sken (*dev_data->csw->d_strategy)(bio); 835229997Sken } 836229997Sken } 837229997Sken} 838229997Sken 839229997Skenstatic void 840229997Skenctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 841229997Sken union ctl_io *io) 842229997Sken{ 843229997Sken struct ctl_be_block_io *beio; 844229997Sken struct ctl_be_block_softc *softc; 845229997Sken 846229997Sken DPRINTF("entered\n"); 847229997Sken 848229997Sken softc = be_lun->softc; 849229997Sken beio = ctl_alloc_beio(softc); 850264020Strasz KASSERT(beio != NULL, ("ctl_alloc_beio() failed")); 851229997Sken 852229997Sken beio->io = io; 853229997Sken beio->softc = softc; 854229997Sken beio->lun = be_lun; 855229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 856229997Sken 857229997Sken switch (io->scsiio.cdb[0]) { 858229997Sken case SYNCHRONIZE_CACHE: 859229997Sken case SYNCHRONIZE_CACHE_16: 860249194Strasz beio->bio_cmd = BIO_FLUSH; 861229997Sken beio->ds_trans_type = DEVSTAT_NO_DATA; 862229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 863229997Sken beio->io_len = 0; 864229997Sken be_lun->lun_flush(be_lun, beio); 865229997Sken break; 866229997Sken default: 867229997Sken panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 868229997Sken break; 869229997Sken } 870229997Sken} 871229997Sken 872258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 873258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 874258622SavgSDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 875258622SavgSDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 876229997Sken 877229997Skenstatic void 878229997Skenctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 879229997Sken union ctl_io *io) 880229997Sken{ 881229997Sken struct ctl_be_block_io *beio; 882229997Sken struct ctl_be_block_softc *softc; 883229997Sken struct ctl_lba_len lbalen; 884229997Sken uint64_t len_left, io_size_bytes; 885229997Sken int i; 886229997Sken 887229997Sken softc = be_lun->softc; 888229997Sken 889229997Sken DPRINTF("entered\n"); 890229997Sken 891229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 892229997Sken SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 893229997Sken } else { 894229997Sken SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 895229997Sken } 896229997Sken 897229997Sken memcpy(&lbalen, io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, 898229997Sken sizeof(lbalen)); 899229997Sken 900229997Sken io_size_bytes = lbalen.len * be_lun->blocksize; 901229997Sken 902229997Sken /* 903229997Sken * XXX KDM this is temporary, until we implement chaining of beio 904229997Sken * structures and multiple datamove calls to move all the data in 905229997Sken * or out. 906229997Sken */ 907229997Sken if (io_size_bytes > CTLBLK_MAX_IO_SIZE) { 908229997Sken printf("%s: IO length %ju > max io size %u\n", __func__, 909229997Sken io_size_bytes, CTLBLK_MAX_IO_SIZE); 910229997Sken ctl_set_invalid_field(&io->scsiio, 911229997Sken /*sks_valid*/ 0, 912229997Sken /*command*/ 1, 913229997Sken /*field*/ 0, 914229997Sken /*bit_valid*/ 0, 915229997Sken /*bit*/ 0); 916229997Sken ctl_done(io); 917229997Sken return; 918229997Sken } 919229997Sken 920229997Sken beio = ctl_alloc_beio(softc); 921264020Strasz KASSERT(beio != NULL, ("ctl_alloc_beio() failed")); 922229997Sken 923229997Sken beio->io = io; 924229997Sken beio->softc = softc; 925229997Sken beio->lun = be_lun; 926229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr = beio; 927229997Sken 928229997Sken /* 929229997Sken * If the I/O came down with an ordered or head of queue tag, set 930229997Sken * the BIO_ORDERED attribute. For head of queue tags, that's 931229997Sken * pretty much the best we can do. 932229997Sken * 933229997Sken * XXX KDM we don't have a great way to easily know about the FUA 934229997Sken * bit right now (it is decoded in ctl_read_write(), but we don't 935229997Sken * pass that knowledge to the backend), and in any case we would 936229997Sken * need to determine how to handle it. 937229997Sken */ 938229997Sken if ((io->scsiio.tag_type == CTL_TAG_ORDERED) 939229997Sken || (io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)) 940229997Sken beio->bio_flags = BIO_ORDERED; 941229997Sken 942229997Sken switch (io->scsiio.tag_type) { 943229997Sken case CTL_TAG_ORDERED: 944229997Sken beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 945229997Sken break; 946229997Sken case CTL_TAG_HEAD_OF_QUEUE: 947229997Sken beio->ds_tag_type = DEVSTAT_TAG_HEAD; 948229997Sken break; 949229997Sken case CTL_TAG_UNTAGGED: 950229997Sken case CTL_TAG_SIMPLE: 951229997Sken case CTL_TAG_ACA: 952229997Sken default: 953229997Sken beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 954229997Sken break; 955229997Sken } 956229997Sken 957229997Sken /* 958229997Sken * This path handles read and write only. The config write path 959229997Sken * handles flush operations. 960229997Sken */ 961229997Sken if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { 962229997Sken beio->bio_cmd = BIO_READ; 963229997Sken beio->ds_trans_type = DEVSTAT_READ; 964229997Sken } else { 965229997Sken beio->bio_cmd = BIO_WRITE; 966229997Sken beio->ds_trans_type = DEVSTAT_WRITE; 967229997Sken } 968229997Sken 969229997Sken beio->io_len = lbalen.len * be_lun->blocksize; 970229997Sken beio->io_offset = lbalen.lba * be_lun->blocksize; 971229997Sken 972229997Sken DPRINTF("%s at LBA %jx len %u\n", 973229997Sken (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 974229997Sken (uintmax_t)lbalen.lba, lbalen.len); 975229997Sken 976229997Sken for (i = 0, len_left = io_size_bytes; i < CTLBLK_MAX_SEGS && 977229997Sken len_left > 0; i++) { 978229997Sken 979229997Sken /* 980229997Sken * Setup the S/G entry for this chunk. 981229997Sken */ 982229997Sken beio->sg_segs[i].len = min(MAXPHYS, len_left); 983229997Sken beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 984229997Sken 985229997Sken DPRINTF("segment %d addr %p len %zd\n", i, 986229997Sken beio->sg_segs[i].addr, beio->sg_segs[i].len); 987229997Sken 988229997Sken beio->num_segs++; 989229997Sken len_left -= beio->sg_segs[i].len; 990229997Sken } 991229997Sken 992229997Sken /* 993229997Sken * For the read case, we need to read the data into our buffers and 994229997Sken * then we can send it back to the user. For the write case, we 995229997Sken * need to get the data from the user first. 996229997Sken */ 997229997Sken if (beio->bio_cmd == BIO_READ) { 998229997Sken SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 999229997Sken be_lun->dispatch(be_lun, beio); 1000229997Sken } else { 1001229997Sken SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1002229997Sken io->scsiio.be_move_done = ctl_be_block_move_done; 1003229997Sken io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1004229997Sken io->scsiio.kern_data_len = beio->io_len; 1005229997Sken io->scsiio.kern_total_len = beio->io_len; 1006229997Sken io->scsiio.kern_rel_offset = 0; 1007229997Sken io->scsiio.kern_data_resid = 0; 1008229997Sken io->scsiio.kern_sg_entries = beio->num_segs; 1009229997Sken io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1010229997Sken#ifdef CTL_TIME_IO 1011229997Sken getbintime(&io->io_hdr.dma_start_bt); 1012229997Sken#endif 1013229997Sken ctl_datamove(io); 1014229997Sken } 1015229997Sken} 1016229997Sken 1017229997Skenstatic void 1018229997Skenctl_be_block_worker(void *context, int pending) 1019229997Sken{ 1020229997Sken struct ctl_be_block_lun *be_lun; 1021229997Sken struct ctl_be_block_softc *softc; 1022229997Sken union ctl_io *io; 1023229997Sken 1024229997Sken be_lun = (struct ctl_be_block_lun *)context; 1025229997Sken softc = be_lun->softc; 1026229997Sken 1027229997Sken DPRINTF("entered\n"); 1028229997Sken 1029229997Sken mtx_lock(&be_lun->lock); 1030229997Sken for (;;) { 1031229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1032229997Sken if (io != NULL) { 1033229997Sken struct ctl_be_block_io *beio; 1034229997Sken 1035229997Sken DPRINTF("datamove queue\n"); 1036229997Sken 1037229997Sken STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1038229997Sken ctl_io_hdr, links); 1039229997Sken 1040229997Sken mtx_unlock(&be_lun->lock); 1041229997Sken 1042229997Sken beio = (struct ctl_be_block_io *) 1043229997Sken io->io_hdr.ctl_private[CTL_PRIV_BACKEND].ptr; 1044229997Sken 1045229997Sken be_lun->dispatch(be_lun, beio); 1046229997Sken 1047229997Sken mtx_lock(&be_lun->lock); 1048229997Sken continue; 1049229997Sken } 1050229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1051229997Sken if (io != NULL) { 1052229997Sken 1053229997Sken DPRINTF("config write queue\n"); 1054229997Sken 1055229997Sken STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1056229997Sken ctl_io_hdr, links); 1057229997Sken 1058229997Sken mtx_unlock(&be_lun->lock); 1059229997Sken 1060229997Sken ctl_be_block_cw_dispatch(be_lun, io); 1061229997Sken 1062229997Sken mtx_lock(&be_lun->lock); 1063229997Sken continue; 1064229997Sken } 1065229997Sken io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1066229997Sken if (io != NULL) { 1067229997Sken DPRINTF("input queue\n"); 1068229997Sken 1069229997Sken STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1070229997Sken ctl_io_hdr, links); 1071229997Sken mtx_unlock(&be_lun->lock); 1072229997Sken 1073229997Sken /* 1074229997Sken * We must drop the lock, since this routine and 1075229997Sken * its children may sleep. 1076229997Sken */ 1077229997Sken ctl_be_block_dispatch(be_lun, io); 1078229997Sken 1079229997Sken mtx_lock(&be_lun->lock); 1080229997Sken continue; 1081229997Sken } 1082229997Sken 1083229997Sken /* 1084229997Sken * If we get here, there is no work left in the queues, so 1085229997Sken * just break out and let the task queue go to sleep. 1086229997Sken */ 1087229997Sken break; 1088229997Sken } 1089229997Sken mtx_unlock(&be_lun->lock); 1090229997Sken} 1091229997Sken 1092229997Sken/* 1093229997Sken * Entry point from CTL to the backend for I/O. We queue everything to a 1094229997Sken * work thread, so this just puts the I/O on a queue and wakes up the 1095229997Sken * thread. 1096229997Sken */ 1097229997Skenstatic int 1098229997Skenctl_be_block_submit(union ctl_io *io) 1099229997Sken{ 1100229997Sken struct ctl_be_block_lun *be_lun; 1101229997Sken struct ctl_be_lun *ctl_be_lun; 1102229997Sken int retval; 1103229997Sken 1104229997Sken DPRINTF("entered\n"); 1105229997Sken 1106229997Sken retval = CTL_RETVAL_COMPLETE; 1107229997Sken 1108229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1109229997Sken CTL_PRIV_BACKEND_LUN].ptr; 1110229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1111229997Sken 1112229997Sken /* 1113229997Sken * Make sure we only get SCSI I/O. 1114229997Sken */ 1115229997Sken KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1116229997Sken "%#x) encountered", io->io_hdr.io_type)); 1117229997Sken 1118229997Sken mtx_lock(&be_lun->lock); 1119229997Sken /* 1120229997Sken * XXX KDM make sure that links is okay to use at this point. 1121229997Sken * Otherwise, we either need to add another field to ctl_io_hdr, 1122229997Sken * or deal with resource allocation here. 1123229997Sken */ 1124229997Sken STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1125229997Sken mtx_unlock(&be_lun->lock); 1126229997Sken 1127229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1128229997Sken 1129229997Sken return (retval); 1130229997Sken} 1131229997Sken 1132229997Skenstatic int 1133229997Skenctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1134229997Sken int flag, struct thread *td) 1135229997Sken{ 1136229997Sken struct ctl_be_block_softc *softc; 1137229997Sken int error; 1138229997Sken 1139229997Sken softc = &backend_block_softc; 1140229997Sken 1141229997Sken error = 0; 1142229997Sken 1143229997Sken switch (cmd) { 1144229997Sken case CTL_LUN_REQ: { 1145229997Sken struct ctl_lun_req *lun_req; 1146229997Sken 1147229997Sken lun_req = (struct ctl_lun_req *)addr; 1148229997Sken 1149229997Sken switch (lun_req->reqtype) { 1150229997Sken case CTL_LUNREQ_CREATE: 1151229997Sken error = ctl_be_block_create(softc, lun_req); 1152229997Sken break; 1153229997Sken case CTL_LUNREQ_RM: 1154229997Sken error = ctl_be_block_rm(softc, lun_req); 1155229997Sken break; 1156232604Strasz case CTL_LUNREQ_MODIFY: 1157232604Strasz error = ctl_be_block_modify(softc, lun_req); 1158232604Strasz break; 1159229997Sken default: 1160229997Sken lun_req->status = CTL_LUN_ERROR; 1161229997Sken snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1162229997Sken "%s: invalid LUN request type %d", __func__, 1163229997Sken lun_req->reqtype); 1164229997Sken break; 1165229997Sken } 1166229997Sken break; 1167229997Sken } 1168229997Sken default: 1169229997Sken error = ENOTTY; 1170229997Sken break; 1171229997Sken } 1172229997Sken 1173229997Sken return (error); 1174229997Sken} 1175229997Sken 1176229997Skenstatic int 1177229997Skenctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1178229997Sken{ 1179229997Sken struct ctl_be_block_filedata *file_data; 1180229997Sken struct ctl_lun_create_params *params; 1181229997Sken struct vattr vattr; 1182229997Sken int error; 1183229997Sken 1184229997Sken error = 0; 1185229997Sken file_data = &be_lun->backend.file; 1186229997Sken params = &req->reqdata.create; 1187229997Sken 1188229997Sken be_lun->dev_type = CTL_BE_BLOCK_FILE; 1189229997Sken be_lun->dispatch = ctl_be_block_dispatch_file; 1190229997Sken be_lun->lun_flush = ctl_be_block_flush_file; 1191229997Sken 1192229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1193229997Sken if (error != 0) { 1194229997Sken snprintf(req->error_str, sizeof(req->error_str), 1195229997Sken "error calling VOP_GETATTR() for file %s", 1196229997Sken be_lun->dev_path); 1197229997Sken return (error); 1198229997Sken } 1199229997Sken 1200229997Sken /* 1201229997Sken * Verify that we have the ability to upgrade to exclusive 1202229997Sken * access on this file so we can trap errors at open instead 1203229997Sken * of reporting them during first access. 1204229997Sken */ 1205229997Sken if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1206229997Sken vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1207229997Sken if (be_lun->vn->v_iflag & VI_DOOMED) { 1208229997Sken error = EBADF; 1209229997Sken snprintf(req->error_str, sizeof(req->error_str), 1210229997Sken "error locking file %s", be_lun->dev_path); 1211229997Sken return (error); 1212229997Sken } 1213229997Sken } 1214229997Sken 1215229997Sken 1216229997Sken file_data->cred = crhold(curthread->td_ucred); 1217232604Strasz if (params->lun_size_bytes != 0) 1218232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1219232604Strasz else 1220232604Strasz be_lun->size_bytes = vattr.va_size; 1221229997Sken /* 1222229997Sken * We set the multi thread flag for file operations because all 1223229997Sken * filesystems (in theory) are capable of allowing multiple readers 1224229997Sken * of a file at once. So we want to get the maximum possible 1225229997Sken * concurrency. 1226229997Sken */ 1227229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1228229997Sken 1229229997Sken /* 1230229997Sken * XXX KDM vattr.va_blocksize may be larger than 512 bytes here. 1231229997Sken * With ZFS, it is 131072 bytes. Block sizes that large don't work 1232229997Sken * with disklabel and UFS on FreeBSD at least. Large block sizes 1233229997Sken * may not work with other OSes as well. So just export a sector 1234229997Sken * size of 512 bytes, which should work with any OS or 1235229997Sken * application. Since our backing is a file, any block size will 1236229997Sken * work fine for the backing store. 1237229997Sken */ 1238229997Sken#if 0 1239229997Sken be_lun->blocksize= vattr.va_blocksize; 1240229997Sken#endif 1241229997Sken if (params->blocksize_bytes != 0) 1242229997Sken be_lun->blocksize = params->blocksize_bytes; 1243229997Sken else 1244229997Sken be_lun->blocksize = 512; 1245229997Sken 1246229997Sken /* 1247229997Sken * Sanity check. The media size has to be at least one 1248229997Sken * sector long. 1249229997Sken */ 1250229997Sken if (be_lun->size_bytes < be_lun->blocksize) { 1251229997Sken error = EINVAL; 1252229997Sken snprintf(req->error_str, sizeof(req->error_str), 1253229997Sken "file %s size %ju < block size %u", be_lun->dev_path, 1254229997Sken (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1255229997Sken } 1256229997Sken return (error); 1257229997Sken} 1258229997Sken 1259229997Skenstatic int 1260229997Skenctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1261229997Sken{ 1262229997Sken struct ctl_lun_create_params *params; 1263229997Sken struct vattr vattr; 1264229997Sken struct cdev *dev; 1265229997Sken struct cdevsw *devsw; 1266229997Sken int error; 1267264191Smav off_t ps, pss, po, pos; 1268229997Sken 1269229997Sken params = &req->reqdata.create; 1270229997Sken 1271229997Sken be_lun->dev_type = CTL_BE_BLOCK_DEV; 1272229997Sken be_lun->dispatch = ctl_be_block_dispatch_dev; 1273229997Sken be_lun->lun_flush = ctl_be_block_flush_dev; 1274229997Sken be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1275229997Sken be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1276229997Sken &be_lun->backend.dev.dev_ref); 1277229997Sken if (be_lun->backend.dev.csw == NULL) 1278229997Sken panic("Unable to retrieve device switch"); 1279229997Sken 1280229997Sken error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1281229997Sken if (error) { 1282229997Sken snprintf(req->error_str, sizeof(req->error_str), 1283229997Sken "%s: error getting vnode attributes for device %s", 1284229997Sken __func__, be_lun->dev_path); 1285229997Sken return (error); 1286229997Sken } 1287229997Sken 1288229997Sken dev = be_lun->vn->v_rdev; 1289229997Sken devsw = dev->si_devsw; 1290229997Sken if (!devsw->d_ioctl) { 1291229997Sken snprintf(req->error_str, sizeof(req->error_str), 1292229997Sken "%s: no d_ioctl for device %s!", __func__, 1293229997Sken be_lun->dev_path); 1294229997Sken return (ENODEV); 1295229997Sken } 1296229997Sken 1297229997Sken error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1298229997Sken (caddr_t)&be_lun->blocksize, FREAD, 1299229997Sken curthread); 1300229997Sken if (error) { 1301229997Sken snprintf(req->error_str, sizeof(req->error_str), 1302229997Sken "%s: error %d returned for DIOCGSECTORSIZE ioctl " 1303229997Sken "on %s!", __func__, error, be_lun->dev_path); 1304229997Sken return (error); 1305229997Sken } 1306229997Sken 1307229997Sken /* 1308229997Sken * If the user has asked for a blocksize that is greater than the 1309229997Sken * backing device's blocksize, we can do it only if the blocksize 1310229997Sken * the user is asking for is an even multiple of the underlying 1311229997Sken * device's blocksize. 1312229997Sken */ 1313229997Sken if ((params->blocksize_bytes != 0) 1314229997Sken && (params->blocksize_bytes > be_lun->blocksize)) { 1315229997Sken uint32_t bs_multiple, tmp_blocksize; 1316229997Sken 1317229997Sken bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1318229997Sken 1319229997Sken tmp_blocksize = bs_multiple * be_lun->blocksize; 1320229997Sken 1321229997Sken if (tmp_blocksize == params->blocksize_bytes) { 1322229997Sken be_lun->blocksize = params->blocksize_bytes; 1323229997Sken } else { 1324229997Sken snprintf(req->error_str, sizeof(req->error_str), 1325229997Sken "%s: requested blocksize %u is not an even " 1326229997Sken "multiple of backing device blocksize %u", 1327229997Sken __func__, params->blocksize_bytes, 1328229997Sken be_lun->blocksize); 1329229997Sken return (EINVAL); 1330229997Sken 1331229997Sken } 1332229997Sken } else if ((params->blocksize_bytes != 0) 1333229997Sken && (params->blocksize_bytes != be_lun->blocksize)) { 1334229997Sken snprintf(req->error_str, sizeof(req->error_str), 1335229997Sken "%s: requested blocksize %u < backing device " 1336229997Sken "blocksize %u", __func__, params->blocksize_bytes, 1337229997Sken be_lun->blocksize); 1338229997Sken return (EINVAL); 1339229997Sken } 1340229997Sken 1341229997Sken error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1342229997Sken (caddr_t)&be_lun->size_bytes, FREAD, 1343229997Sken curthread); 1344229997Sken if (error) { 1345229997Sken snprintf(req->error_str, sizeof(req->error_str), 1346232604Strasz "%s: error %d returned for DIOCGMEDIASIZE " 1347232604Strasz " ioctl on %s!", __func__, error, 1348232604Strasz be_lun->dev_path); 1349229997Sken return (error); 1350229997Sken } 1351229997Sken 1352232604Strasz if (params->lun_size_bytes != 0) { 1353232604Strasz if (params->lun_size_bytes > be_lun->size_bytes) { 1354232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1355232604Strasz "%s: requested LUN size %ju > backing device " 1356232604Strasz "size %ju", __func__, 1357232604Strasz (uintmax_t)params->lun_size_bytes, 1358232604Strasz (uintmax_t)be_lun->size_bytes); 1359232604Strasz return (EINVAL); 1360232604Strasz } 1361232604Strasz 1362232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1363232604Strasz } 1364232604Strasz 1365264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1366264191Smav (caddr_t)&ps, FREAD, curthread); 1367264191Smav if (error) 1368264191Smav ps = po = 0; 1369264191Smav else { 1370264191Smav error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1371264191Smav (caddr_t)&po, FREAD, curthread); 1372264191Smav if (error) 1373264191Smav po = 0; 1374264191Smav } 1375264191Smav pss = ps / be_lun->blocksize; 1376264191Smav pos = po / be_lun->blocksize; 1377264191Smav if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1378264191Smav ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1379264191Smav be_lun->pblockexp = fls(pss) - 1; 1380264191Smav be_lun->pblockoff = (pss - pos) % pss; 1381264191Smav } 1382264191Smav 1383229997Sken return (0); 1384229997Sken} 1385229997Sken 1386229997Skenstatic int 1387229997Skenctl_be_block_close(struct ctl_be_block_lun *be_lun) 1388229997Sken{ 1389229997Sken DROP_GIANT(); 1390229997Sken if (be_lun->vn) { 1391229997Sken int flags = FREAD | FWRITE; 1392229997Sken 1393229997Sken switch (be_lun->dev_type) { 1394229997Sken case CTL_BE_BLOCK_DEV: 1395229997Sken if (be_lun->backend.dev.csw) { 1396229997Sken dev_relthread(be_lun->backend.dev.cdev, 1397229997Sken be_lun->backend.dev.dev_ref); 1398229997Sken be_lun->backend.dev.csw = NULL; 1399229997Sken be_lun->backend.dev.cdev = NULL; 1400229997Sken } 1401229997Sken break; 1402229997Sken case CTL_BE_BLOCK_FILE: 1403229997Sken break; 1404229997Sken case CTL_BE_BLOCK_NONE: 1405258871Strasz break; 1406229997Sken default: 1407229997Sken panic("Unexpected backend type."); 1408229997Sken break; 1409229997Sken } 1410229997Sken 1411229997Sken (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1412229997Sken be_lun->vn = NULL; 1413229997Sken 1414229997Sken switch (be_lun->dev_type) { 1415229997Sken case CTL_BE_BLOCK_DEV: 1416229997Sken break; 1417229997Sken case CTL_BE_BLOCK_FILE: 1418229997Sken if (be_lun->backend.file.cred != NULL) { 1419229997Sken crfree(be_lun->backend.file.cred); 1420229997Sken be_lun->backend.file.cred = NULL; 1421229997Sken } 1422229997Sken break; 1423229997Sken case CTL_BE_BLOCK_NONE: 1424258871Strasz break; 1425229997Sken default: 1426229997Sken panic("Unexpected backend type."); 1427229997Sken break; 1428229997Sken } 1429229997Sken } 1430229997Sken PICKUP_GIANT(); 1431229997Sken 1432229997Sken return (0); 1433229997Sken} 1434229997Sken 1435229997Skenstatic int 1436229997Skenctl_be_block_open(struct ctl_be_block_softc *softc, 1437229997Sken struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1438229997Sken{ 1439229997Sken struct nameidata nd; 1440229997Sken int flags; 1441229997Sken int error; 1442229997Sken 1443229997Sken /* 1444229997Sken * XXX KDM allow a read-only option? 1445229997Sken */ 1446229997Sken flags = FREAD | FWRITE; 1447229997Sken error = 0; 1448229997Sken 1449229997Sken if (rootvnode == NULL) { 1450229997Sken snprintf(req->error_str, sizeof(req->error_str), 1451229997Sken "%s: Root filesystem is not mounted", __func__); 1452229997Sken return (1); 1453229997Sken } 1454229997Sken 1455229997Sken if (!curthread->td_proc->p_fd->fd_cdir) { 1456229997Sken curthread->td_proc->p_fd->fd_cdir = rootvnode; 1457229997Sken VREF(rootvnode); 1458229997Sken } 1459229997Sken if (!curthread->td_proc->p_fd->fd_rdir) { 1460229997Sken curthread->td_proc->p_fd->fd_rdir = rootvnode; 1461229997Sken VREF(rootvnode); 1462229997Sken } 1463229997Sken if (!curthread->td_proc->p_fd->fd_jdir) { 1464229997Sken curthread->td_proc->p_fd->fd_jdir = rootvnode; 1465229997Sken VREF(rootvnode); 1466229997Sken } 1467229997Sken 1468229997Sken again: 1469229997Sken NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 1470229997Sken error = vn_open(&nd, &flags, 0, NULL); 1471229997Sken if (error) { 1472229997Sken /* 1473229997Sken * This is the only reasonable guess we can make as far as 1474229997Sken * path if the user doesn't give us a fully qualified path. 1475229997Sken * If they want to specify a file, they need to specify the 1476229997Sken * full path. 1477229997Sken */ 1478229997Sken if (be_lun->dev_path[0] != '/') { 1479229997Sken char *dev_path = "/dev/"; 1480229997Sken char *dev_name; 1481229997Sken 1482229997Sken /* Try adding device path at beginning of name */ 1483229997Sken dev_name = malloc(strlen(be_lun->dev_path) 1484229997Sken + strlen(dev_path) + 1, 1485229997Sken M_CTLBLK, M_WAITOK); 1486229997Sken if (dev_name) { 1487229997Sken sprintf(dev_name, "%s%s", dev_path, 1488229997Sken be_lun->dev_path); 1489229997Sken free(be_lun->dev_path, M_CTLBLK); 1490229997Sken be_lun->dev_path = dev_name; 1491229997Sken goto again; 1492229997Sken } 1493229997Sken } 1494229997Sken snprintf(req->error_str, sizeof(req->error_str), 1495229997Sken "%s: error opening %s", __func__, be_lun->dev_path); 1496229997Sken return (error); 1497229997Sken } 1498229997Sken 1499229997Sken NDFREE(&nd, NDF_ONLY_PNBUF); 1500229997Sken 1501229997Sken be_lun->vn = nd.ni_vp; 1502229997Sken 1503229997Sken /* We only support disks and files. */ 1504229997Sken if (vn_isdisk(be_lun->vn, &error)) { 1505229997Sken error = ctl_be_block_open_dev(be_lun, req); 1506229997Sken } else if (be_lun->vn->v_type == VREG) { 1507229997Sken error = ctl_be_block_open_file(be_lun, req); 1508229997Sken } else { 1509229997Sken error = EINVAL; 1510229997Sken snprintf(req->error_str, sizeof(req->error_str), 1511258871Strasz "%s is not a disk or plain file", be_lun->dev_path); 1512229997Sken } 1513229997Sken VOP_UNLOCK(be_lun->vn, 0); 1514229997Sken 1515229997Sken if (error != 0) { 1516229997Sken ctl_be_block_close(be_lun); 1517229997Sken return (error); 1518229997Sken } 1519229997Sken 1520229997Sken be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 1521229997Sken be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 1522229997Sken 1523229997Sken return (0); 1524229997Sken} 1525229997Sken 1526229997Skenstatic int 1527229997Skenctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1528229997Sken{ 1529229997Sken struct ctl_be_block_lun *be_lun; 1530229997Sken struct ctl_lun_create_params *params; 1531229997Sken struct ctl_be_arg *file_arg; 1532229997Sken char tmpstr[32]; 1533229997Sken int retval, num_threads; 1534229997Sken int i; 1535229997Sken 1536229997Sken params = &req->reqdata.create; 1537229997Sken retval = 0; 1538229997Sken 1539229997Sken num_threads = cbb_num_threads; 1540229997Sken 1541229997Sken file_arg = NULL; 1542229997Sken 1543229997Sken be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 1544229997Sken 1545229997Sken be_lun->softc = softc; 1546229997Sken STAILQ_INIT(&be_lun->input_queue); 1547229997Sken STAILQ_INIT(&be_lun->config_write_queue); 1548229997Sken STAILQ_INIT(&be_lun->datamove_queue); 1549254759Strasz STAILQ_INIT(&be_lun->ctl_be_lun.options); 1550229997Sken sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 1551229997Sken mtx_init(&be_lun->lock, be_lun->lunname, NULL, MTX_DEF); 1552229997Sken 1553229997Sken be_lun->lun_zone = uma_zcreate(be_lun->lunname, MAXPHYS, 1554256995Smav NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 1555229997Sken 1556229997Sken if (be_lun->lun_zone == NULL) { 1557229997Sken snprintf(req->error_str, sizeof(req->error_str), 1558229997Sken "%s: error allocating UMA zone", __func__); 1559229997Sken goto bailout_error; 1560229997Sken } 1561229997Sken 1562229997Sken if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 1563229997Sken be_lun->ctl_be_lun.lun_type = params->device_type; 1564229997Sken else 1565229997Sken be_lun->ctl_be_lun.lun_type = T_DIRECT; 1566229997Sken 1567229997Sken if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 1568229997Sken for (i = 0; i < req->num_be_args; i++) { 1569249026Strasz if (strcmp(req->kern_be_args[i].kname, "file") == 0) { 1570229997Sken file_arg = &req->kern_be_args[i]; 1571229997Sken break; 1572229997Sken } 1573229997Sken } 1574229997Sken 1575229997Sken if (file_arg == NULL) { 1576229997Sken snprintf(req->error_str, sizeof(req->error_str), 1577229997Sken "%s: no file argument specified", __func__); 1578229997Sken goto bailout_error; 1579229997Sken } 1580229997Sken 1581229997Sken be_lun->dev_path = malloc(file_arg->vallen, M_CTLBLK, 1582229997Sken M_WAITOK | M_ZERO); 1583229997Sken 1584249026Strasz strlcpy(be_lun->dev_path, (char *)file_arg->kvalue, 1585229997Sken file_arg->vallen); 1586229997Sken 1587229997Sken retval = ctl_be_block_open(softc, be_lun, req); 1588229997Sken if (retval != 0) { 1589229997Sken retval = 0; 1590229997Sken goto bailout_error; 1591229997Sken } 1592229997Sken 1593229997Sken /* 1594229997Sken * Tell the user the size of the file/device. 1595229997Sken */ 1596229997Sken params->lun_size_bytes = be_lun->size_bytes; 1597229997Sken 1598229997Sken /* 1599229997Sken * The maximum LBA is the size - 1. 1600229997Sken */ 1601229997Sken be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 1602229997Sken } else { 1603229997Sken /* 1604229997Sken * For processor devices, we don't have any size. 1605229997Sken */ 1606229997Sken be_lun->blocksize = 0; 1607264191Smav be_lun->pblockexp = 0; 1608264191Smav be_lun->pblockoff = 0; 1609229997Sken be_lun->size_blocks = 0; 1610229997Sken be_lun->size_bytes = 0; 1611229997Sken be_lun->ctl_be_lun.maxlba = 0; 1612229997Sken params->lun_size_bytes = 0; 1613229997Sken 1614229997Sken /* 1615229997Sken * Default to just 1 thread for processor devices. 1616229997Sken */ 1617229997Sken num_threads = 1; 1618229997Sken } 1619229997Sken 1620229997Sken /* 1621229997Sken * XXX This searching loop might be refactored to be combined with 1622229997Sken * the loop above, 1623229997Sken */ 1624229997Sken for (i = 0; i < req->num_be_args; i++) { 1625249026Strasz if (strcmp(req->kern_be_args[i].kname, "num_threads") == 0) { 1626229997Sken struct ctl_be_arg *thread_arg; 1627229997Sken char num_thread_str[16]; 1628229997Sken int tmp_num_threads; 1629229997Sken 1630229997Sken 1631229997Sken thread_arg = &req->kern_be_args[i]; 1632229997Sken 1633249026Strasz strlcpy(num_thread_str, (char *)thread_arg->kvalue, 1634229997Sken min(thread_arg->vallen, 1635229997Sken sizeof(num_thread_str))); 1636229997Sken 1637229997Sken tmp_num_threads = strtol(num_thread_str, NULL, 0); 1638229997Sken 1639229997Sken /* 1640229997Sken * We don't let the user specify less than one 1641229997Sken * thread, but hope he's clueful enough not to 1642229997Sken * specify 1000 threads. 1643229997Sken */ 1644229997Sken if (tmp_num_threads < 1) { 1645229997Sken snprintf(req->error_str, sizeof(req->error_str), 1646229997Sken "%s: invalid number of threads %s", 1647229997Sken __func__, num_thread_str); 1648229997Sken goto bailout_error; 1649229997Sken } 1650229997Sken 1651229997Sken num_threads = tmp_num_threads; 1652254759Strasz } else if (strcmp(req->kern_be_args[i].kname, "file") != 0 && 1653254759Strasz strcmp(req->kern_be_args[i].kname, "dev") != 0) { 1654254759Strasz struct ctl_be_lun_option *opt; 1655254759Strasz 1656254759Strasz opt = malloc(sizeof(*opt), M_CTLBLK, M_WAITOK); 1657254759Strasz opt->name = malloc(strlen(req->kern_be_args[i].kname) + 1, M_CTLBLK, M_WAITOK); 1658254759Strasz strcpy(opt->name, req->kern_be_args[i].kname); 1659254759Strasz opt->value = malloc(strlen(req->kern_be_args[i].kvalue) + 1, M_CTLBLK, M_WAITOK); 1660254759Strasz strcpy(opt->value, req->kern_be_args[i].kvalue); 1661254759Strasz STAILQ_INSERT_TAIL(&be_lun->ctl_be_lun.options, opt, links); 1662229997Sken } 1663229997Sken } 1664229997Sken 1665229997Sken be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 1666229997Sken be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 1667229997Sken be_lun->ctl_be_lun.be_lun = be_lun; 1668229997Sken be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 1669264191Smav be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 1670264191Smav be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 1671229997Sken /* Tell the user the blocksize we ended up using */ 1672229997Sken params->blocksize_bytes = be_lun->blocksize; 1673229997Sken if (params->flags & CTL_LUN_FLAG_ID_REQ) { 1674229997Sken be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 1675229997Sken be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 1676229997Sken } else 1677229997Sken be_lun->ctl_be_lun.req_lun_id = 0; 1678229997Sken 1679229997Sken be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 1680229997Sken be_lun->ctl_be_lun.lun_config_status = 1681229997Sken ctl_be_block_lun_config_status; 1682229997Sken be_lun->ctl_be_lun.be = &ctl_be_block_driver; 1683229997Sken 1684229997Sken if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 1685229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 1686229997Sken softc->num_luns); 1687229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 1688229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1689229997Sken sizeof(tmpstr))); 1690229997Sken 1691229997Sken /* Tell the user what we used for a serial number */ 1692229997Sken strncpy((char *)params->serial_num, tmpstr, 1693229997Sken ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 1694229997Sken } else { 1695229997Sken strncpy((char *)be_lun->ctl_be_lun.serial_num, 1696229997Sken params->serial_num, 1697229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 1698229997Sken sizeof(params->serial_num))); 1699229997Sken } 1700229997Sken if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 1701229997Sken snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 1702229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 1703229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1704229997Sken sizeof(tmpstr))); 1705229997Sken 1706229997Sken /* Tell the user what we used for a device ID */ 1707229997Sken strncpy((char *)params->device_id, tmpstr, 1708229997Sken ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 1709229997Sken } else { 1710229997Sken strncpy((char *)be_lun->ctl_be_lun.device_id, 1711229997Sken params->device_id, 1712229997Sken ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 1713229997Sken sizeof(params->device_id))); 1714229997Sken } 1715229997Sken 1716229997Sken TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 1717229997Sken 1718229997Sken be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 1719229997Sken taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 1720229997Sken 1721229997Sken if (be_lun->io_taskqueue == NULL) { 1722229997Sken snprintf(req->error_str, sizeof(req->error_str), 1723229997Sken "%s: Unable to create taskqueue", __func__); 1724229997Sken goto bailout_error; 1725229997Sken } 1726229997Sken 1727229997Sken /* 1728229997Sken * Note that we start the same number of threads by default for 1729229997Sken * both the file case and the block device case. For the file 1730229997Sken * case, we need multiple threads to allow concurrency, because the 1731229997Sken * vnode interface is designed to be a blocking interface. For the 1732229997Sken * block device case, ZFS zvols at least will block the caller's 1733229997Sken * context in many instances, and so we need multiple threads to 1734229997Sken * overcome that problem. Other block devices don't need as many 1735229997Sken * threads, but they shouldn't cause too many problems. 1736229997Sken * 1737229997Sken * If the user wants to just have a single thread for a block 1738229997Sken * device, he can specify that when the LUN is created, or change 1739229997Sken * the tunable/sysctl to alter the default number of threads. 1740229997Sken */ 1741229997Sken retval = taskqueue_start_threads(&be_lun->io_taskqueue, 1742229997Sken /*num threads*/num_threads, 1743229997Sken /*priority*/PWAIT, 1744229997Sken /*thread name*/ 1745229997Sken "%s taskq", be_lun->lunname); 1746229997Sken 1747229997Sken if (retval != 0) 1748229997Sken goto bailout_error; 1749229997Sken 1750229997Sken be_lun->num_threads = num_threads; 1751229997Sken 1752229997Sken mtx_lock(&softc->lock); 1753229997Sken softc->num_luns++; 1754229997Sken STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 1755229997Sken 1756229997Sken mtx_unlock(&softc->lock); 1757229997Sken 1758229997Sken retval = ctl_add_lun(&be_lun->ctl_be_lun); 1759229997Sken if (retval != 0) { 1760229997Sken mtx_lock(&softc->lock); 1761229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1762229997Sken links); 1763229997Sken softc->num_luns--; 1764229997Sken mtx_unlock(&softc->lock); 1765229997Sken snprintf(req->error_str, sizeof(req->error_str), 1766229997Sken "%s: ctl_add_lun() returned error %d, see dmesg for " 1767229997Sken "details", __func__, retval); 1768229997Sken retval = 0; 1769229997Sken goto bailout_error; 1770229997Sken } 1771229997Sken 1772229997Sken mtx_lock(&softc->lock); 1773229997Sken 1774229997Sken /* 1775229997Sken * Tell the config_status routine that we're waiting so it won't 1776229997Sken * clean up the LUN in the event of an error. 1777229997Sken */ 1778229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1779229997Sken 1780229997Sken while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 1781229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1782229997Sken if (retval == EINTR) 1783229997Sken break; 1784229997Sken } 1785229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1786229997Sken 1787229997Sken if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 1788229997Sken snprintf(req->error_str, sizeof(req->error_str), 1789229997Sken "%s: LUN configuration error, see dmesg for details", 1790229997Sken __func__); 1791229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 1792229997Sken links); 1793229997Sken softc->num_luns--; 1794229997Sken mtx_unlock(&softc->lock); 1795229997Sken goto bailout_error; 1796229997Sken } else { 1797229997Sken params->req_lun_id = be_lun->ctl_be_lun.lun_id; 1798229997Sken } 1799229997Sken 1800229997Sken mtx_unlock(&softc->lock); 1801229997Sken 1802229997Sken be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 1803229997Sken be_lun->blocksize, 1804229997Sken DEVSTAT_ALL_SUPPORTED, 1805229997Sken be_lun->ctl_be_lun.lun_type 1806229997Sken | DEVSTAT_TYPE_IF_OTHER, 1807229997Sken DEVSTAT_PRIORITY_OTHER); 1808229997Sken 1809229997Sken 1810229997Sken req->status = CTL_LUN_OK; 1811229997Sken 1812229997Sken return (retval); 1813229997Sken 1814229997Skenbailout_error: 1815229997Sken req->status = CTL_LUN_ERROR; 1816229997Sken 1817229997Sken ctl_be_block_close(be_lun); 1818229997Sken 1819229997Sken free(be_lun->dev_path, M_CTLBLK); 1820229997Sken free(be_lun, M_CTLBLK); 1821229997Sken 1822229997Sken return (retval); 1823229997Sken} 1824229997Sken 1825229997Skenstatic int 1826229997Skenctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1827229997Sken{ 1828229997Sken struct ctl_lun_rm_params *params; 1829229997Sken struct ctl_be_block_lun *be_lun; 1830229997Sken int retval; 1831229997Sken 1832229997Sken params = &req->reqdata.rm; 1833229997Sken 1834229997Sken mtx_lock(&softc->lock); 1835229997Sken 1836229997Sken be_lun = NULL; 1837229997Sken 1838229997Sken STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 1839229997Sken if (be_lun->ctl_be_lun.lun_id == params->lun_id) 1840229997Sken break; 1841229997Sken } 1842229997Sken mtx_unlock(&softc->lock); 1843229997Sken 1844229997Sken if (be_lun == NULL) { 1845229997Sken snprintf(req->error_str, sizeof(req->error_str), 1846229997Sken "%s: LUN %u is not managed by the block backend", 1847229997Sken __func__, params->lun_id); 1848229997Sken goto bailout_error; 1849229997Sken } 1850229997Sken 1851229997Sken retval = ctl_disable_lun(&be_lun->ctl_be_lun); 1852229997Sken 1853229997Sken if (retval != 0) { 1854229997Sken snprintf(req->error_str, sizeof(req->error_str), 1855229997Sken "%s: error %d returned from ctl_disable_lun() for " 1856229997Sken "LUN %d", __func__, retval, params->lun_id); 1857229997Sken goto bailout_error; 1858229997Sken 1859229997Sken } 1860229997Sken 1861229997Sken retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 1862229997Sken if (retval != 0) { 1863229997Sken snprintf(req->error_str, sizeof(req->error_str), 1864229997Sken "%s: error %d returned from ctl_invalidate_lun() for " 1865229997Sken "LUN %d", __func__, retval, params->lun_id); 1866229997Sken goto bailout_error; 1867229997Sken } 1868229997Sken 1869229997Sken mtx_lock(&softc->lock); 1870229997Sken 1871229997Sken be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 1872229997Sken 1873229997Sken while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1874229997Sken retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 1875229997Sken if (retval == EINTR) 1876229997Sken break; 1877229997Sken } 1878229997Sken 1879229997Sken be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 1880229997Sken 1881229997Sken if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 1882229997Sken snprintf(req->error_str, sizeof(req->error_str), 1883229997Sken "%s: interrupted waiting for LUN to be freed", 1884229997Sken __func__); 1885229997Sken mtx_unlock(&softc->lock); 1886229997Sken goto bailout_error; 1887229997Sken } 1888229997Sken 1889229997Sken STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 1890229997Sken 1891229997Sken softc->num_luns--; 1892229997Sken mtx_unlock(&softc->lock); 1893229997Sken 1894229997Sken taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 1895229997Sken 1896229997Sken taskqueue_free(be_lun->io_taskqueue); 1897229997Sken 1898229997Sken ctl_be_block_close(be_lun); 1899229997Sken 1900229997Sken if (be_lun->disk_stats != NULL) 1901229997Sken devstat_remove_entry(be_lun->disk_stats); 1902229997Sken 1903229997Sken uma_zdestroy(be_lun->lun_zone); 1904229997Sken 1905229997Sken free(be_lun->dev_path, M_CTLBLK); 1906229997Sken 1907229997Sken free(be_lun, M_CTLBLK); 1908229997Sken 1909229997Sken req->status = CTL_LUN_OK; 1910229997Sken 1911229997Sken return (0); 1912229997Sken 1913229997Skenbailout_error: 1914229997Sken 1915229997Sken req->status = CTL_LUN_ERROR; 1916229997Sken 1917229997Sken return (0); 1918229997Sken} 1919229997Sken 1920232604Straszstatic int 1921232604Straszctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 1922232604Strasz struct ctl_lun_req *req) 1923232604Strasz{ 1924232604Strasz struct vattr vattr; 1925232604Strasz int error; 1926232604Strasz struct ctl_lun_modify_params *params; 1927232604Strasz 1928232604Strasz params = &req->reqdata.modify; 1929232604Strasz 1930232604Strasz if (params->lun_size_bytes != 0) { 1931232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1932232604Strasz } else { 1933232604Strasz error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1934232604Strasz if (error != 0) { 1935232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1936232604Strasz "error calling VOP_GETATTR() for file %s", 1937232604Strasz be_lun->dev_path); 1938232604Strasz return (error); 1939232604Strasz } 1940232604Strasz 1941232604Strasz be_lun->size_bytes = vattr.va_size; 1942232604Strasz } 1943232604Strasz 1944232604Strasz return (0); 1945232604Strasz} 1946232604Strasz 1947232604Straszstatic int 1948232604Straszctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 1949232604Strasz struct ctl_lun_req *req) 1950232604Strasz{ 1951232604Strasz struct cdev *dev; 1952232604Strasz struct cdevsw *devsw; 1953232604Strasz int error; 1954232604Strasz struct ctl_lun_modify_params *params; 1955232604Strasz uint64_t size_bytes; 1956232604Strasz 1957232604Strasz params = &req->reqdata.modify; 1958232604Strasz 1959232604Strasz dev = be_lun->vn->v_rdev; 1960232604Strasz devsw = dev->si_devsw; 1961232604Strasz if (!devsw->d_ioctl) { 1962232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1963232604Strasz "%s: no d_ioctl for device %s!", __func__, 1964232604Strasz be_lun->dev_path); 1965232604Strasz return (ENODEV); 1966232604Strasz } 1967232604Strasz 1968232604Strasz error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1969232604Strasz (caddr_t)&size_bytes, FREAD, 1970232604Strasz curthread); 1971232604Strasz if (error) { 1972232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1973232604Strasz "%s: error %d returned for DIOCGMEDIASIZE ioctl " 1974232604Strasz "on %s!", __func__, error, be_lun->dev_path); 1975232604Strasz return (error); 1976232604Strasz } 1977232604Strasz 1978232604Strasz if (params->lun_size_bytes != 0) { 1979232604Strasz if (params->lun_size_bytes > size_bytes) { 1980232604Strasz snprintf(req->error_str, sizeof(req->error_str), 1981232604Strasz "%s: requested LUN size %ju > backing device " 1982232604Strasz "size %ju", __func__, 1983232604Strasz (uintmax_t)params->lun_size_bytes, 1984232604Strasz (uintmax_t)size_bytes); 1985232604Strasz return (EINVAL); 1986232604Strasz } 1987232604Strasz 1988232604Strasz be_lun->size_bytes = params->lun_size_bytes; 1989232604Strasz } else { 1990232604Strasz be_lun->size_bytes = size_bytes; 1991232604Strasz } 1992232604Strasz 1993232604Strasz return (0); 1994232604Strasz} 1995232604Strasz 1996232604Straszstatic int 1997232604Straszctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 1998232604Strasz{ 1999232604Strasz struct ctl_lun_modify_params *params; 2000232604Strasz struct ctl_be_block_lun *be_lun; 2001241896Skib int error; 2002232604Strasz 2003232604Strasz params = &req->reqdata.modify; 2004232604Strasz 2005232604Strasz mtx_lock(&softc->lock); 2006232604Strasz 2007232604Strasz be_lun = NULL; 2008232604Strasz 2009232604Strasz STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2010232604Strasz if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2011232604Strasz break; 2012232604Strasz } 2013232604Strasz mtx_unlock(&softc->lock); 2014232604Strasz 2015232604Strasz if (be_lun == NULL) { 2016232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2017232604Strasz "%s: LUN %u is not managed by the block backend", 2018232604Strasz __func__, params->lun_id); 2019232604Strasz goto bailout_error; 2020232604Strasz } 2021232604Strasz 2022232604Strasz if (params->lun_size_bytes != 0) { 2023232604Strasz if (params->lun_size_bytes < be_lun->blocksize) { 2024232604Strasz snprintf(req->error_str, sizeof(req->error_str), 2025232604Strasz "%s: LUN size %ju < blocksize %u", __func__, 2026232604Strasz params->lun_size_bytes, be_lun->blocksize); 2027232604Strasz goto bailout_error; 2028232604Strasz } 2029232604Strasz } 2030232604Strasz 2031232604Strasz vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2032232604Strasz 2033232604Strasz if (be_lun->vn->v_type == VREG) 2034232604Strasz error = ctl_be_block_modify_file(be_lun, req); 2035232604Strasz else 2036232604Strasz error = ctl_be_block_modify_dev(be_lun, req); 2037232604Strasz 2038232604Strasz VOP_UNLOCK(be_lun->vn, 0); 2039232604Strasz 2040232604Strasz if (error != 0) 2041232604Strasz goto bailout_error; 2042232604Strasz 2043232604Strasz be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2044232604Strasz 2045232604Strasz /* 2046232604Strasz * The maximum LBA is the size - 1. 2047232604Strasz * 2048232604Strasz * XXX: Note that this field is being updated without locking, 2049232604Strasz * which might cause problems on 32-bit architectures. 2050232604Strasz */ 2051232604Strasz be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; 2052232604Strasz ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2053232604Strasz 2054232604Strasz /* Tell the user the exact size we ended up using */ 2055232604Strasz params->lun_size_bytes = be_lun->size_bytes; 2056232604Strasz 2057232604Strasz req->status = CTL_LUN_OK; 2058232604Strasz 2059232604Strasz return (0); 2060232604Strasz 2061232604Straszbailout_error: 2062232604Strasz req->status = CTL_LUN_ERROR; 2063232604Strasz 2064232604Strasz return (0); 2065232604Strasz} 2066232604Strasz 2067229997Skenstatic void 2068229997Skenctl_be_block_lun_shutdown(void *be_lun) 2069229997Sken{ 2070229997Sken struct ctl_be_block_lun *lun; 2071229997Sken struct ctl_be_block_softc *softc; 2072229997Sken 2073229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2074229997Sken 2075229997Sken softc = lun->softc; 2076229997Sken 2077229997Sken mtx_lock(&softc->lock); 2078229997Sken lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2079229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2080229997Sken wakeup(lun); 2081229997Sken mtx_unlock(&softc->lock); 2082229997Sken 2083229997Sken} 2084229997Sken 2085229997Skenstatic void 2086229997Skenctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2087229997Sken{ 2088229997Sken struct ctl_be_block_lun *lun; 2089229997Sken struct ctl_be_block_softc *softc; 2090229997Sken 2091229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2092229997Sken softc = lun->softc; 2093229997Sken 2094229997Sken if (status == CTL_LUN_CONFIG_OK) { 2095229997Sken mtx_lock(&softc->lock); 2096229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2097229997Sken if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2098229997Sken wakeup(lun); 2099229997Sken mtx_unlock(&softc->lock); 2100229997Sken 2101229997Sken /* 2102229997Sken * We successfully added the LUN, attempt to enable it. 2103229997Sken */ 2104229997Sken if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2105229997Sken printf("%s: ctl_enable_lun() failed!\n", __func__); 2106229997Sken if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2107229997Sken printf("%s: ctl_invalidate_lun() failed!\n", 2108229997Sken __func__); 2109229997Sken } 2110229997Sken } 2111229997Sken 2112229997Sken return; 2113229997Sken } 2114229997Sken 2115229997Sken 2116229997Sken mtx_lock(&softc->lock); 2117229997Sken lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2118229997Sken lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2119229997Sken wakeup(lun); 2120229997Sken mtx_unlock(&softc->lock); 2121229997Sken} 2122229997Sken 2123229997Sken 2124229997Skenstatic int 2125229997Skenctl_be_block_config_write(union ctl_io *io) 2126229997Sken{ 2127229997Sken struct ctl_be_block_lun *be_lun; 2128229997Sken struct ctl_be_lun *ctl_be_lun; 2129229997Sken int retval; 2130229997Sken 2131229997Sken retval = 0; 2132229997Sken 2133229997Sken DPRINTF("entered\n"); 2134229997Sken 2135229997Sken ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2136229997Sken CTL_PRIV_BACKEND_LUN].ptr; 2137229997Sken be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2138229997Sken 2139229997Sken switch (io->scsiio.cdb[0]) { 2140229997Sken case SYNCHRONIZE_CACHE: 2141229997Sken case SYNCHRONIZE_CACHE_16: 2142229997Sken /* 2143229997Sken * The upper level CTL code will filter out any CDBs with 2144229997Sken * the immediate bit set and return the proper error. 2145229997Sken * 2146229997Sken * We don't really need to worry about what LBA range the 2147229997Sken * user asked to be synced out. When they issue a sync 2148229997Sken * cache command, we'll sync out the whole thing. 2149229997Sken */ 2150229997Sken mtx_lock(&be_lun->lock); 2151229997Sken STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2152229997Sken links); 2153229997Sken mtx_unlock(&be_lun->lock); 2154229997Sken taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2155229997Sken break; 2156229997Sken case START_STOP_UNIT: { 2157229997Sken struct scsi_start_stop_unit *cdb; 2158229997Sken 2159229997Sken cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2160229997Sken 2161229997Sken if (cdb->how & SSS_START) 2162229997Sken retval = ctl_start_lun(ctl_be_lun); 2163229997Sken else { 2164229997Sken retval = ctl_stop_lun(ctl_be_lun); 2165229997Sken /* 2166229997Sken * XXX KDM Copan-specific offline behavior. 2167229997Sken * Figure out a reasonable way to port this? 2168229997Sken */ 2169229997Sken#ifdef NEEDTOPORT 2170229997Sken if ((retval == 0) 2171229997Sken && (cdb->byte2 & SSS_ONOFFLINE)) 2172229997Sken retval = ctl_lun_offline(ctl_be_lun); 2173229997Sken#endif 2174229997Sken } 2175229997Sken 2176229997Sken /* 2177229997Sken * In general, the above routines should not fail. They 2178229997Sken * just set state for the LUN. So we've got something 2179229997Sken * pretty wrong here if we can't start or stop the LUN. 2180229997Sken */ 2181229997Sken if (retval != 0) { 2182229997Sken ctl_set_internal_failure(&io->scsiio, 2183229997Sken /*sks_valid*/ 1, 2184229997Sken /*retry_count*/ 0xf051); 2185229997Sken retval = CTL_RETVAL_COMPLETE; 2186229997Sken } else { 2187229997Sken ctl_set_success(&io->scsiio); 2188229997Sken } 2189229997Sken ctl_config_write_done(io); 2190229997Sken break; 2191229997Sken } 2192229997Sken default: 2193229997Sken ctl_set_invalid_opcode(&io->scsiio); 2194229997Sken ctl_config_write_done(io); 2195229997Sken retval = CTL_RETVAL_COMPLETE; 2196229997Sken break; 2197229997Sken } 2198229997Sken 2199229997Sken return (retval); 2200229997Sken 2201229997Sken} 2202229997Sken 2203229997Skenstatic int 2204229997Skenctl_be_block_config_read(union ctl_io *io) 2205229997Sken{ 2206229997Sken return (0); 2207229997Sken} 2208229997Sken 2209229997Skenstatic int 2210229997Skenctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2211229997Sken{ 2212229997Sken struct ctl_be_block_lun *lun; 2213229997Sken int retval; 2214229997Sken 2215229997Sken lun = (struct ctl_be_block_lun *)be_lun; 2216229997Sken retval = 0; 2217229997Sken 2218229997Sken retval = sbuf_printf(sb, "<num_threads>"); 2219229997Sken 2220229997Sken if (retval != 0) 2221229997Sken goto bailout; 2222229997Sken 2223229997Sken retval = sbuf_printf(sb, "%d", lun->num_threads); 2224229997Sken 2225229997Sken if (retval != 0) 2226229997Sken goto bailout; 2227229997Sken 2228229997Sken retval = sbuf_printf(sb, "</num_threads>"); 2229229997Sken 2230229997Sken /* 2231229997Sken * For processor devices, we don't have a path variable. 2232229997Sken */ 2233229997Sken if ((retval != 0) 2234229997Sken || (lun->dev_path == NULL)) 2235229997Sken goto bailout; 2236229997Sken 2237229997Sken retval = sbuf_printf(sb, "<file>"); 2238229997Sken 2239229997Sken if (retval != 0) 2240229997Sken goto bailout; 2241229997Sken 2242229997Sken retval = ctl_sbuf_printf_esc(sb, lun->dev_path); 2243229997Sken 2244229997Sken if (retval != 0) 2245229997Sken goto bailout; 2246229997Sken 2247229997Sken retval = sbuf_printf(sb, "</file>\n"); 2248229997Sken 2249229997Skenbailout: 2250229997Sken 2251229997Sken return (retval); 2252229997Sken} 2253229997Sken 2254229997Skenint 2255229997Skenctl_be_block_init(void) 2256229997Sken{ 2257229997Sken struct ctl_be_block_softc *softc; 2258229997Sken int retval; 2259229997Sken 2260229997Sken softc = &backend_block_softc; 2261229997Sken retval = 0; 2262229997Sken 2263229997Sken mtx_init(&softc->lock, "ctlblk", NULL, MTX_DEF); 2264264020Strasz beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2265264020Strasz NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2266229997Sken STAILQ_INIT(&softc->disk_list); 2267229997Sken STAILQ_INIT(&softc->lun_list); 2268229997Sken 2269229997Sken return (retval); 2270229997Sken} 2271