sys/cam/ctl/ctl_backend_block.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2003 Silicon Graphics International Corp.
   5  * Copyright (c) 2009-2011 Spectra Logic Corporation
   6  * Copyright (c) 2012 The FreeBSD Foundation
   7  * Copyright (c) 2014-2015 Alexander Motin <mav@FreeBSD.org>
   8  * All rights reserved.
   9  *
  10  * Portions of this software were developed by Edward Tomasz Napierala
  11  * under sponsorship from the FreeBSD Foundation.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions, and the following disclaimer,
  18  *    without modification.
  19  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  20  *    substantially similar to the "NO WARRANTY" disclaimer below
  21  *    ("Disclaimer") and any redistribution must be conditioned upon
  22  *    including a substantially similar Disclaimer requirement for further
  23  *    binary redistribution.
  24  *
  25  * NO WARRANTY
  26  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  27  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  28  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  29  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  30  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  34  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  35  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  36  * POSSIBILITY OF SUCH DAMAGES.
  37  *
  38  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $
  39  */
  40 /*
  41  * CAM Target Layer driver backend for block devices.
  42  *
  43  * Author: Ken Merry <ken@FreeBSD.org>
  44  */
  45 #include <sys/cdefs.h>
  46 __FBSDID("$FreeBSD$");
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/kernel.h>
  51 #include <sys/types.h>
  52 #include <sys/kthread.h>
  53 #include <sys/bio.h>
  54 #include <sys/fcntl.h>
  55 #include <sys/limits.h>
  56 #include <sys/lock.h>
  57 #include <sys/mutex.h>
  58 #include <sys/condvar.h>
  59 #include <sys/malloc.h>
  60 #include <sys/conf.h>
  61 #include <sys/ioccom.h>
  62 #include <sys/queue.h>
  63 #include <sys/sbuf.h>
  64 #include <sys/endian.h>
  65 #include <sys/uio.h>
  66 #include <sys/buf.h>
  67 #include <sys/taskqueue.h>
  68 #include <sys/vnode.h>
  69 #include <sys/namei.h>
  70 #include <sys/mount.h>
  71 #include <sys/disk.h>
  72 #include <sys/fcntl.h>
  73 #include <sys/filedesc.h>
  74 #include <sys/filio.h>
  75 #include <sys/proc.h>
  76 #include <sys/pcpu.h>
  77 #include <sys/module.h>
  78 #include <sys/sdt.h>
  79 #include <sys/devicestat.h>
  80 #include <sys/sysctl.h>
  81 #include <sys/nv.h>
  82 #include <sys/dnv.h>
  83 #include <sys/sx.h>
  84
  85 #include <geom/geom.h>
  86
  87 #include <cam/cam.h>
  88 #include <cam/scsi/scsi_all.h>
  89 #include <cam/scsi/scsi_da.h>
  90 #include <cam/ctl/ctl_io.h>
  91 #include <cam/ctl/ctl.h>
  92 #include <cam/ctl/ctl_backend.h>
  93 #include <cam/ctl/ctl_ioctl.h>
  94 #include <cam/ctl/ctl_ha.h>
  95 #include <cam/ctl/ctl_scsi_all.h>
  96 #include <cam/ctl/ctl_private.h>
  97 #include <cam/ctl/ctl_error.h>
  98
  99 /*
 100  * The idea here is that we'll allocate enough S/G space to hold a 1MB
 101  * I/O.  If we get an I/O larger than that, we'll split it.
 102  */
 103 #define CTLBLK_HALF_IO_SIZE     (512 * 1024)
 104 #define CTLBLK_MAX_IO_SIZE      (CTLBLK_HALF_IO_SIZE * 2)
 105 #define CTLBLK_MAX_SEG          MAXPHYS
 106 #define CTLBLK_HALF_SEGS        MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1)
 107 #define CTLBLK_MAX_SEGS         (CTLBLK_HALF_SEGS * 2)
 108
 109 #ifdef CTLBLK_DEBUG
 110 #define DPRINTF(fmt, args...) \
 111     printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
 112 #else
 113 #define DPRINTF(fmt, args...) do {} while(0)
 114 #endif
 115
 116 #define PRIV(io)        \
 117     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
 118 #define ARGS(io)        \
 119     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
 120
 121 SDT_PROVIDER_DEFINE(cbb);
 122
 123 typedef enum {
 124         CTL_BE_BLOCK_LUN_UNCONFIGURED   = 0x01,
 125         CTL_BE_BLOCK_LUN_WAITING        = 0x04,
 126 } ctl_be_block_lun_flags;
 127
 128 typedef enum {
 129         CTL_BE_BLOCK_NONE,
 130         CTL_BE_BLOCK_DEV,
 131         CTL_BE_BLOCK_FILE
 132 } ctl_be_block_type;
 133
 134 struct ctl_be_block_filedata {
 135         struct ucred *cred;
 136 };
 137
 138 union ctl_be_block_bedata {
 139         struct ctl_be_block_filedata file;
 140 };
 141
 142 struct ctl_be_block_io;
 143 struct ctl_be_block_lun;
 144
 145 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
 146                                struct ctl_be_block_io *beio);
 147 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
 148                                   const char *attrname);
 149
 150 /*
 151  * Backend LUN structure.  There is a 1:1 mapping between a block device
 152  * and a backend block LUN, and between a backend block LUN and a CTL LUN.
 153  */
 154 struct ctl_be_block_lun {
 155         struct ctl_lun_create_params params;
 156         char *dev_path;
 157         ctl_be_block_type dev_type;
 158         struct vnode *vn;
 159         union ctl_be_block_bedata backend;
 160         cbb_dispatch_t dispatch;
 161         cbb_dispatch_t lun_flush;
 162         cbb_dispatch_t unmap;
 163         cbb_dispatch_t get_lba_status;
 164         cbb_getattr_t getattr;
 165         uint64_t size_blocks;
 166         uint64_t size_bytes;
 167         struct ctl_be_block_softc *softc;
 168         struct devstat *disk_stats;
 169         ctl_be_block_lun_flags flags;
 170         SLIST_ENTRY(ctl_be_block_lun) links;
 171         struct ctl_be_lun cbe_lun;
 172         struct taskqueue *io_taskqueue;
 173         struct task io_task;
 174         int num_threads;
 175         STAILQ_HEAD(, ctl_io_hdr) input_queue;
 176         STAILQ_HEAD(, ctl_io_hdr) config_read_queue;
 177         STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
 178         STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
 179         struct mtx_padalign io_lock;
 180         struct mtx_padalign queue_lock;
 181 };
 182
 183 /*
 184  * Overall softc structure for the block backend module.
 185  */
 186 struct ctl_be_block_softc {
 187         struct sx                        modify_lock;
 188         struct mtx                       lock;
 189         int                              num_luns;
 190         SLIST_HEAD(, ctl_be_block_lun)   lun_list;
 191         uma_zone_t                       beio_zone;
 192         uma_zone_t                       buf_zone;
 193 };
 194
 195 static struct ctl_be_block_softc backend_block_softc;
 196
 197 /*
 198  * Per-I/O information.
 199  */
 200 struct ctl_be_block_io {
 201         union ctl_io                    *io;
 202         struct ctl_sg_entry             sg_segs[CTLBLK_MAX_SEGS];
 203         struct iovec                    xiovecs[CTLBLK_MAX_SEGS];
 204         int                             bio_cmd;
 205         int                             two_sglists;
 206         int                             num_segs;
 207         int                             num_bios_sent;
 208         int                             num_bios_done;
 209         int                             send_complete;
 210         int                             first_error;
 211         uint64_t                        first_error_offset;
 212         struct bintime                  ds_t0;
 213         devstat_tag_type                ds_tag_type;
 214         devstat_trans_flags             ds_trans_type;
 215         uint64_t                        io_len;
 216         uint64_t                        io_offset;
 217         int                             io_arg;
 218         struct ctl_be_block_softc       *softc;
 219         struct ctl_be_block_lun         *lun;
 220         void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
 221 };
 222
 223 extern struct ctl_softc *control_softc;
 224
 225 static int cbb_num_threads = 14;
 226 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 227             "CAM Target Layer Block Backend");
 228 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
 229            &cbb_num_threads, 0, "Number of threads per backing file");
 230
 231 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
 232 static void ctl_free_beio(struct ctl_be_block_io *beio);
 233 static void ctl_complete_beio(struct ctl_be_block_io *beio);
 234 static int ctl_be_block_move_done(union ctl_io *io);
 235 static void ctl_be_block_biodone(struct bio *bio);
 236 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 237                                     struct ctl_be_block_io *beio);
 238 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 239                                        struct ctl_be_block_io *beio);
 240 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
 241                                   struct ctl_be_block_io *beio);
 242 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
 243                                          const char *attrname);
 244 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
 245                                    struct ctl_be_block_io *beio);
 246 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
 247                                    struct ctl_be_block_io *beio);
 248 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
 249                                       struct ctl_be_block_io *beio);
 250 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
 251                                          const char *attrname);
 252 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
 253                                     union ctl_io *io);
 254 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
 255                                     union ctl_io *io);
 256 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
 257                                   union ctl_io *io);
 258 static void ctl_be_block_worker(void *context, int pending);
 259 static int ctl_be_block_submit(union ctl_io *io);
 260 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
 261                                    int flag, struct thread *td);
 262 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
 263                                   struct ctl_lun_req *req);
 264 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
 265                                  struct ctl_lun_req *req);
 266 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
 267 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun,
 268                              struct ctl_lun_req *req);
 269 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
 270                                struct ctl_lun_req *req);
 271 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
 272                            struct ctl_lun_req *req);
 273 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
 274                            struct ctl_lun_req *req);
 275 static void ctl_be_block_lun_shutdown(void *be_lun);
 276 static int ctl_be_block_config_write(union ctl_io *io);
 277 static int ctl_be_block_config_read(union ctl_io *io);
 278 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb);
 279 static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname);
 280 static int ctl_be_block_init(void);
 281 static int ctl_be_block_shutdown(void);
 282
 283 static struct ctl_backend_driver ctl_be_block_driver =
 284 {
 285         .name = "block",
 286         .flags = CTL_BE_FLAG_HAS_CONFIG,
 287         .init = ctl_be_block_init,
 288         .shutdown = ctl_be_block_shutdown,
 289         .data_submit = ctl_be_block_submit,
 290         .data_move_done = ctl_be_block_move_done,
 291         .config_read = ctl_be_block_config_read,
 292         .config_write = ctl_be_block_config_write,
 293         .ioctl = ctl_be_block_ioctl,
 294         .lun_info = ctl_be_block_lun_info,
 295         .lun_attr = ctl_be_block_lun_attr
 296 };
 297
 298 MALLOC_DEFINE(M_CTLBLK, "ctlblock", "Memory used for CTL block backend");
 299 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
 300
 301 static struct ctl_be_block_io *
 302 ctl_alloc_beio(struct ctl_be_block_softc *softc)
 303 {
 304         struct ctl_be_block_io *beio;
 305
 306         beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO);
 307         beio->softc = softc;
 308         return (beio);
 309 }
 310
 311 static void
 312 ctl_free_beio(struct ctl_be_block_io *beio)
 313 {
 314         struct ctl_be_block_softc *softc = beio->softc;
 315         int i;
 316
 317         for (i = 0; i < beio->num_segs; i++) {
 318                 uma_zfree(softc->buf_zone, beio->sg_segs[i].addr);
 319
 320                 /* For compare we had two equal S/G lists. */
 321                 if (beio->two_sglists) {
 322                         uma_zfree(softc->buf_zone,
 323                             beio->sg_segs[i + CTLBLK_HALF_SEGS].addr);
 324                 }
 325         }
 326
 327         uma_zfree(softc->beio_zone, beio);
 328 }
 329
 330 static void
 331 ctl_complete_beio(struct ctl_be_block_io *beio)
 332 {
 333         union ctl_io *io = beio->io;
 334
 335         if (beio->beio_cont != NULL) {
 336                 beio->beio_cont(beio);
 337         } else {
 338                 ctl_free_beio(beio);
 339                 ctl_data_submit_done(io);
 340         }
 341 }
 342
 343 static size_t
 344 cmp(uint8_t *a, uint8_t *b, size_t size)
 345 {
 346         size_t i;
 347
 348         for (i = 0; i < size; i++) {
 349                 if (a[i] != b[i])
 350                         break;
 351         }
 352         return (i);
 353 }
 354
 355 static void
 356 ctl_be_block_compare(union ctl_io *io)
 357 {
 358         struct ctl_be_block_io *beio;
 359         uint64_t off, res;
 360         int i;
 361         uint8_t info[8];
 362
 363         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
 364         off = 0;
 365         for (i = 0; i < beio->num_segs; i++) {
 366                 res = cmp(beio->sg_segs[i].addr,
 367                     beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
 368                     beio->sg_segs[i].len);
 369                 off += res;
 370                 if (res < beio->sg_segs[i].len)
 371                         break;
 372         }
 373         if (i < beio->num_segs) {
 374                 scsi_u64to8b(off, info);
 375                 ctl_set_sense(&io->scsiio, /*current_error*/ 1,
 376                     /*sense_key*/ SSD_KEY_MISCOMPARE,
 377                     /*asc*/ 0x1D, /*ascq*/ 0x00,
 378                     /*type*/ SSD_ELEM_INFO,
 379                     /*size*/ sizeof(info), /*data*/ &info,
 380                     /*type*/ SSD_ELEM_NONE);
 381         } else
 382                 ctl_set_success(&io->scsiio);
 383 }
 384
 385 static int
 386 ctl_be_block_move_done(union ctl_io *io)
 387 {
 388         struct ctl_be_block_io *beio;
 389         struct ctl_be_block_lun *be_lun;
 390         struct ctl_lba_len_flags *lbalen;
 391 #ifdef CTL_TIME_IO
 392         struct bintime cur_bt;
 393 #endif
 394
 395         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
 396         be_lun = beio->lun;
 397
 398         DPRINTF("entered\n");
 399
 400 #ifdef CTL_TIME_IO
 401         getbinuptime(&cur_bt);
 402         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
 403         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
 404 #endif
 405         io->io_hdr.num_dmas++;
 406         io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
 407
 408         /*
 409          * We set status at this point for read commands, and write
 410          * commands with errors.
 411          */
 412         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
 413                 ;
 414         } else if ((io->io_hdr.port_status != 0) &&
 415             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
 416              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
 417                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
 418                     /*retry_count*/ io->io_hdr.port_status);
 419         } else if (io->scsiio.kern_data_resid != 0 &&
 420             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
 421             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
 422              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
 423                 ctl_set_invalid_field_ciu(&io->scsiio);
 424         } else if ((io->io_hdr.port_status == 0) &&
 425             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
 426                 lbalen = ARGS(beio->io);
 427                 if (lbalen->flags & CTL_LLF_READ) {
 428                         ctl_set_success(&io->scsiio);
 429                 } else if (lbalen->flags & CTL_LLF_COMPARE) {
 430                         /* We have two data blocks ready for comparison. */
 431                         ctl_be_block_compare(io);
 432                 }
 433         }
 434
 435         /*
 436          * If this is a read, or a write with errors, it is done.
 437          */
 438         if ((beio->bio_cmd == BIO_READ)
 439          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
 440          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
 441                 ctl_complete_beio(beio);
 442                 return (0);
 443         }
 444
 445         /*
 446          * At this point, we have a write and the DMA completed
 447          * successfully.  We now have to queue it to the task queue to
 448          * execute the backend I/O.  That is because we do blocking
 449          * memory allocations, and in the file backing case, blocking I/O.
 450          * This move done routine is generally called in the SIM's
 451          * interrupt context, and therefore we cannot block.
 452          */
 453         mtx_lock(&be_lun->queue_lock);
 454         STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
 455         mtx_unlock(&be_lun->queue_lock);
 456         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
 457
 458         return (0);
 459 }
 460
 461 static void
 462 ctl_be_block_biodone(struct bio *bio)
 463 {
 464         struct ctl_be_block_io *beio;
 465         struct ctl_be_block_lun *be_lun;
 466         union ctl_io *io;
 467         int error;
 468
 469         beio = bio->bio_caller1;
 470         be_lun = beio->lun;
 471         io = beio->io;
 472
 473         DPRINTF("entered\n");
 474
 475         error = bio->bio_error;
 476         mtx_lock(&be_lun->io_lock);
 477         if (error != 0 &&
 478             (beio->first_error == 0 ||
 479              bio->bio_offset < beio->first_error_offset)) {
 480                 beio->first_error = error;
 481                 beio->first_error_offset = bio->bio_offset;
 482         }
 483
 484         beio->num_bios_done++;
 485
 486         /*
 487          * XXX KDM will this cause WITNESS to complain?  Holding a lock
 488          * during the free might cause it to complain.
 489          */
 490         g_destroy_bio(bio);
 491
 492         /*
 493          * If the send complete bit isn't set, or we aren't the last I/O to
 494          * complete, then we're done.
 495          */
 496         if ((beio->send_complete == 0)
 497          || (beio->num_bios_done < beio->num_bios_sent)) {
 498                 mtx_unlock(&be_lun->io_lock);
 499                 return;
 500         }
 501
 502         /*
 503          * At this point, we've verified that we are the last I/O to
 504          * complete, so it's safe to drop the lock.
 505          */
 506         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 507             beio->ds_tag_type, beio->ds_trans_type,
 508             /*now*/ NULL, /*then*/&beio->ds_t0);
 509         mtx_unlock(&be_lun->io_lock);
 510
 511         /*
 512          * If there are any errors from the backing device, we fail the
 513          * entire I/O with a medium error.
 514          */
 515         error = beio->first_error;
 516         if (error != 0) {
 517                 if (error == EOPNOTSUPP) {
 518                         ctl_set_invalid_opcode(&io->scsiio);
 519                 } else if (error == ENOSPC || error == EDQUOT) {
 520                         ctl_set_space_alloc_fail(&io->scsiio);
 521                 } else if (error == EROFS || error == EACCES) {
 522                         ctl_set_hw_write_protected(&io->scsiio);
 523                 } else if (beio->bio_cmd == BIO_FLUSH) {
 524                         /* XXX KDM is there is a better error here? */
 525                         ctl_set_internal_failure(&io->scsiio,
 526                                                  /*sks_valid*/ 1,
 527                                                  /*retry_count*/ 0xbad2);
 528                 } else {
 529                         ctl_set_medium_error(&io->scsiio,
 530                             beio->bio_cmd == BIO_READ);
 531                 }
 532                 ctl_complete_beio(beio);
 533                 return;
 534         }
 535
 536         /*
 537          * If this is a write, a flush, a delete or verify, we're all done.
 538          * If this is a read, we can now send the data to the user.
 539          */
 540         if ((beio->bio_cmd == BIO_WRITE)
 541          || (beio->bio_cmd == BIO_FLUSH)
 542          || (beio->bio_cmd == BIO_DELETE)
 543          || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 544                 ctl_set_success(&io->scsiio);
 545                 ctl_complete_beio(beio);
 546         } else {
 547                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 548                     beio->beio_cont == NULL) {
 549                         ctl_set_success(&io->scsiio);
 550                         ctl_serseq_done(io);
 551                 }
 552 #ifdef CTL_TIME_IO
 553                 getbinuptime(&io->io_hdr.dma_start_bt);
 554 #endif
 555                 ctl_datamove(io);
 556         }
 557 }
 558
 559 static void
 560 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 561                         struct ctl_be_block_io *beio)
 562 {
 563         union ctl_io *io = beio->io;
 564         struct mount *mountpoint;
 565         int error, lock_flags;
 566
 567         DPRINTF("entered\n");
 568
 569         binuptime(&beio->ds_t0);
 570         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 571
 572         (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 573
 574         if (MNT_SHARED_WRITES(mountpoint) ||
 575             ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 576                 lock_flags = LK_SHARED;
 577         else
 578                 lock_flags = LK_EXCLUSIVE;
 579         vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 580         error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT,
 581             curthread);
 582         VOP_UNLOCK(be_lun->vn);
 583
 584         vn_finished_write(mountpoint);
 585
 586         mtx_lock(&be_lun->io_lock);
 587         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 588             beio->ds_tag_type, beio->ds_trans_type,
 589             /*now*/ NULL, /*then*/&beio->ds_t0);
 590         mtx_unlock(&be_lun->io_lock);
 591
 592         if (error == 0)
 593                 ctl_set_success(&io->scsiio);
 594         else {
 595                 /* XXX KDM is there is a better error here? */
 596                 ctl_set_internal_failure(&io->scsiio,
 597                                          /*sks_valid*/ 1,
 598                                          /*retry_count*/ 0xbad1);
 599         }
 600
 601         ctl_complete_beio(beio);
 602 }
 603
 604 SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t");
 605 SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t");
 606 SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t");
 607 SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t");
 608
 609 static void
 610 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 611                            struct ctl_be_block_io *beio)
 612 {
 613         struct ctl_be_block_filedata *file_data;
 614         union ctl_io *io;
 615         struct uio xuio;
 616         struct iovec *xiovec;
 617         size_t s;
 618         int error, flags, i;
 619
 620         DPRINTF("entered\n");
 621
 622         file_data = &be_lun->backend.file;
 623         io = beio->io;
 624         flags = 0;
 625         if (ARGS(io)->flags & CTL_LLF_DPO)
 626                 flags |= IO_DIRECT;
 627         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 628                 flags |= IO_SYNC;
 629
 630         bzero(&xuio, sizeof(xuio));
 631         if (beio->bio_cmd == BIO_READ) {
 632                 SDT_PROBE0(cbb, , read, file_start);
 633                 xuio.uio_rw = UIO_READ;
 634         } else {
 635                 SDT_PROBE0(cbb, , write, file_start);
 636                 xuio.uio_rw = UIO_WRITE;
 637         }
 638         xuio.uio_offset = beio->io_offset;
 639         xuio.uio_resid = beio->io_len;
 640         xuio.uio_segflg = UIO_SYSSPACE;
 641         xuio.uio_iov = beio->xiovecs;
 642         xuio.uio_iovcnt = beio->num_segs;
 643         xuio.uio_td = curthread;
 644
 645         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 646                 xiovec->iov_base = beio->sg_segs[i].addr;
 647                 xiovec->iov_len = beio->sg_segs[i].len;
 648         }
 649
 650         binuptime(&beio->ds_t0);
 651         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 652
 653         if (beio->bio_cmd == BIO_READ) {
 654                 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 655
 656                 /*
 657                  * UFS pays attention to IO_DIRECT for reads.  If the
 658                  * DIRECTIO option is configured into the kernel, it calls
 659                  * ffs_rawread().  But that only works for single-segment
 660                  * uios with user space addresses.  In our case, with a
 661                  * kernel uio, it still reads into the buffer cache, but it
 662                  * will just try to release the buffer from the cache later
 663                  * on in ffs_read().
 664                  *
 665                  * ZFS does not pay attention to IO_DIRECT for reads.
 666                  *
 667                  * UFS does not pay attention to IO_SYNC for reads.
 668                  *
 669                  * ZFS pays attention to IO_SYNC (which translates into the
 670                  * Solaris define FRSYNC for zfs_read()) for reads.  It
 671                  * attempts to sync the file before reading.
 672                  */
 673                 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
 674
 675                 VOP_UNLOCK(be_lun->vn);
 676                 SDT_PROBE0(cbb, , read, file_done);
 677                 if (error == 0 && xuio.uio_resid > 0) {
 678                         /*
 679                          * If we red less then requested (EOF), then
 680                          * we should clean the rest of the buffer.
 681                          */
 682                         s = beio->io_len - xuio.uio_resid;
 683                         for (i = 0; i < beio->num_segs; i++) {
 684                                 if (s >= beio->sg_segs[i].len) {
 685                                         s -= beio->sg_segs[i].len;
 686                                         continue;
 687                                 }
 688                                 bzero((uint8_t *)beio->sg_segs[i].addr + s,
 689                                     beio->sg_segs[i].len - s);
 690                                 s = 0;
 691                         }
 692                 }
 693         } else {
 694                 struct mount *mountpoint;
 695                 int lock_flags;
 696
 697                 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 698
 699                 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL)
 700                   && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 701                         lock_flags = LK_SHARED;
 702                 else
 703                         lock_flags = LK_EXCLUSIVE;
 704                 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 705
 706                 /*
 707                  * UFS pays attention to IO_DIRECT for writes.  The write
 708                  * is done asynchronously.  (Normally the write would just
 709                  * get put into cache.
 710                  *
 711                  * UFS pays attention to IO_SYNC for writes.  It will
 712                  * attempt to write the buffer out synchronously if that
 713                  * flag is set.
 714                  *
 715                  * ZFS does not pay attention to IO_DIRECT for writes.
 716                  *
 717                  * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
 718                  * for writes.  It will flush the transaction from the
 719                  * cache before returning.
 720                  */
 721                 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
 722                 VOP_UNLOCK(be_lun->vn);
 723
 724                 vn_finished_write(mountpoint);
 725                 SDT_PROBE0(cbb, , write, file_done);
 726         }
 727
 728         mtx_lock(&be_lun->io_lock);
 729         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 730             beio->ds_tag_type, beio->ds_trans_type,
 731             /*now*/ NULL, /*then*/&beio->ds_t0);
 732         mtx_unlock(&be_lun->io_lock);
 733
 734         /*
 735          * If we got an error, set the sense data to "MEDIUM ERROR" and
 736          * return the I/O to the user.
 737          */
 738         if (error != 0) {
 739                 if (error == ENOSPC || error == EDQUOT) {
 740                         ctl_set_space_alloc_fail(&io->scsiio);
 741                 } else if (error == EROFS || error == EACCES) {
 742                         ctl_set_hw_write_protected(&io->scsiio);
 743                 } else {
 744                         ctl_set_medium_error(&io->scsiio,
 745                             beio->bio_cmd == BIO_READ);
 746                 }
 747                 ctl_complete_beio(beio);
 748                 return;
 749         }
 750
 751         /*
 752          * If this is a write or a verify, we're all done.
 753          * If this is a read, we can now send the data to the user.
 754          */
 755         if ((beio->bio_cmd == BIO_WRITE) ||
 756             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 757                 ctl_set_success(&io->scsiio);
 758                 ctl_complete_beio(beio);
 759         } else {
 760                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 761                     beio->beio_cont == NULL) {
 762                         ctl_set_success(&io->scsiio);
 763                         ctl_serseq_done(io);
 764                 }
 765 #ifdef CTL_TIME_IO
 766                 getbinuptime(&io->io_hdr.dma_start_bt);
 767 #endif
 768                 ctl_datamove(io);
 769         }
 770 }
 771
 772 static void
 773 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
 774                         struct ctl_be_block_io *beio)
 775 {
 776         union ctl_io *io = beio->io;
 777         struct ctl_lba_len_flags *lbalen = ARGS(io);
 778         struct scsi_get_lba_status_data *data;
 779         off_t roff, off;
 780         int error, status;
 781
 782         DPRINTF("entered\n");
 783
 784         off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
 785         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 786         error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
 787             0, curthread->td_ucred, curthread);
 788         if (error == 0 && off > roff)
 789                 status = 0;     /* mapped up to off */
 790         else {
 791                 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
 792                     0, curthread->td_ucred, curthread);
 793                 if (error == 0 && off > roff)
 794                         status = 1;     /* deallocated up to off */
 795                 else {
 796                         status = 0;     /* unknown up to the end */
 797                         off = be_lun->size_bytes;
 798                 }
 799         }
 800         VOP_UNLOCK(be_lun->vn);
 801
 802         data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
 803         scsi_u64to8b(lbalen->lba, data->descr[0].addr);
 804         scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
 805             lbalen->lba), data->descr[0].length);
 806         data->descr[0].status = status;
 807
 808         ctl_complete_beio(beio);
 809 }
 810
 811 static uint64_t
 812 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
 813 {
 814         struct vattr            vattr;
 815         struct statfs           statfs;
 816         uint64_t                val;
 817         int                     error;
 818
 819         val = UINT64_MAX;
 820         if (be_lun->vn == NULL)
 821                 return (val);
 822         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 823         if (strcmp(attrname, "blocksused") == 0) {
 824                 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
 825                 if (error == 0)
 826                         val = vattr.va_bytes / be_lun->cbe_lun.blocksize;
 827         }
 828         if (strcmp(attrname, "blocksavail") == 0 &&
 829             !VN_IS_DOOMED(be_lun->vn)) {
 830                 error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
 831                 if (error == 0)
 832                         val = statfs.f_bavail * statfs.f_bsize /
 833                             be_lun->cbe_lun.blocksize;
 834         }
 835         VOP_UNLOCK(be_lun->vn);
 836         return (val);
 837 }
 838
 839 static void
 840 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
 841                            struct ctl_be_block_io *beio)
 842 {
 843         union ctl_io *io;
 844         struct cdevsw *csw;
 845         struct cdev *dev;
 846         struct uio xuio;
 847         struct iovec *xiovec;
 848         int error, flags, i, ref;
 849
 850         DPRINTF("entered\n");
 851
 852         io = beio->io;
 853         flags = 0;
 854         if (ARGS(io)->flags & CTL_LLF_DPO)
 855                 flags |= IO_DIRECT;
 856         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 857                 flags |= IO_SYNC;
 858
 859         bzero(&xuio, sizeof(xuio));
 860         if (beio->bio_cmd == BIO_READ) {
 861                 SDT_PROBE0(cbb, , read, file_start);
 862                 xuio.uio_rw = UIO_READ;
 863         } else {
 864                 SDT_PROBE0(cbb, , write, file_start);
 865                 xuio.uio_rw = UIO_WRITE;
 866         }
 867         xuio.uio_offset = beio->io_offset;
 868         xuio.uio_resid = beio->io_len;
 869         xuio.uio_segflg = UIO_SYSSPACE;
 870         xuio.uio_iov = beio->xiovecs;
 871         xuio.uio_iovcnt = beio->num_segs;
 872         xuio.uio_td = curthread;
 873
 874         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 875                 xiovec->iov_base = beio->sg_segs[i].addr;
 876                 xiovec->iov_len = beio->sg_segs[i].len;
 877         }
 878
 879         binuptime(&beio->ds_t0);
 880         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 881
 882         csw = devvn_refthread(be_lun->vn, &dev, &ref);
 883         if (csw) {
 884                 if (beio->bio_cmd == BIO_READ)
 885                         error = csw->d_read(dev, &xuio, flags);
 886                 else
 887                         error = csw->d_write(dev, &xuio, flags);
 888                 dev_relthread(dev, ref);
 889         } else
 890                 error = ENXIO;
 891
 892         if (beio->bio_cmd == BIO_READ)
 893                 SDT_PROBE0(cbb, , read, file_done);
 894         else
 895                 SDT_PROBE0(cbb, , write, file_done);
 896
 897         mtx_lock(&be_lun->io_lock);
 898         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 899             beio->ds_tag_type, beio->ds_trans_type,
 900             /*now*/ NULL, /*then*/&beio->ds_t0);
 901         mtx_unlock(&be_lun->io_lock);
 902
 903         /*
 904          * If we got an error, set the sense data to "MEDIUM ERROR" and
 905          * return the I/O to the user.
 906          */
 907         if (error != 0) {
 908                 if (error == ENOSPC || error == EDQUOT) {
 909                         ctl_set_space_alloc_fail(&io->scsiio);
 910                 } else if (error == EROFS || error == EACCES) {
 911                         ctl_set_hw_write_protected(&io->scsiio);
 912                 } else {
 913                         ctl_set_medium_error(&io->scsiio,
 914                             beio->bio_cmd == BIO_READ);
 915                 }
 916                 ctl_complete_beio(beio);
 917                 return;
 918         }
 919
 920         /*
 921          * If this is a write or a verify, we're all done.
 922          * If this is a read, we can now send the data to the user.
 923          */
 924         if ((beio->bio_cmd == BIO_WRITE) ||
 925             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 926                 ctl_set_success(&io->scsiio);
 927                 ctl_complete_beio(beio);
 928         } else {
 929                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 930                     beio->beio_cont == NULL) {
 931                         ctl_set_success(&io->scsiio);
 932                         ctl_serseq_done(io);
 933                 }
 934 #ifdef CTL_TIME_IO
 935                 getbinuptime(&io->io_hdr.dma_start_bt);
 936 #endif
 937                 ctl_datamove(io);
 938         }
 939 }
 940
 941 static void
 942 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
 943                         struct ctl_be_block_io *beio)
 944 {
 945         union ctl_io *io = beio->io;
 946         struct cdevsw *csw;
 947         struct cdev *dev;
 948         struct ctl_lba_len_flags *lbalen = ARGS(io);
 949         struct scsi_get_lba_status_data *data;
 950         off_t roff, off;
 951         int error, ref, status;
 952
 953         DPRINTF("entered\n");
 954
 955         csw = devvn_refthread(be_lun->vn, &dev, &ref);
 956         if (csw == NULL) {
 957                 status = 0;     /* unknown up to the end */
 958                 off = be_lun->size_bytes;
 959                 goto done;
 960         }
 961         off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
 962         error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD,
 963             curthread);
 964         if (error == 0 && off > roff)
 965                 status = 0;     /* mapped up to off */
 966         else {
 967                 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD,
 968                     curthread);
 969                 if (error == 0 && off > roff)
 970                         status = 1;     /* deallocated up to off */
 971                 else {
 972                         status = 0;     /* unknown up to the end */
 973                         off = be_lun->size_bytes;
 974                 }
 975         }
 976         dev_relthread(dev, ref);
 977
 978 done:
 979         data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
 980         scsi_u64to8b(lbalen->lba, data->descr[0].addr);
 981         scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
 982             lbalen->lba), data->descr[0].length);
 983         data->descr[0].status = status;
 984
 985         ctl_complete_beio(beio);
 986 }
 987
 988 static void
 989 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
 990                        struct ctl_be_block_io *beio)
 991 {
 992         struct bio *bio;
 993         struct cdevsw *csw;
 994         struct cdev *dev;
 995         int ref;
 996
 997         DPRINTF("entered\n");
 998
 999         /* This can't fail, it's a blocking allocation. */
1000         bio = g_alloc_bio();
1001
1002         bio->bio_cmd        = BIO_FLUSH;
1003         bio->bio_offset     = 0;
1004         bio->bio_data       = 0;
1005         bio->bio_done       = ctl_be_block_biodone;
1006         bio->bio_caller1    = beio;
1007         bio->bio_pblkno     = 0;
1008
1009         /*
1010          * We don't need to acquire the LUN lock here, because we are only
1011          * sending one bio, and so there is no other context to synchronize
1012          * with.
1013          */
1014         beio->num_bios_sent = 1;
1015         beio->send_complete = 1;
1016
1017         binuptime(&beio->ds_t0);
1018         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1019
1020         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1021         if (csw) {
1022                 bio->bio_dev = dev;
1023                 csw->d_strategy(bio);
1024                 dev_relthread(dev, ref);
1025         } else {
1026                 bio->bio_error = ENXIO;
1027                 ctl_be_block_biodone(bio);
1028         }
1029 }
1030
1031 static void
1032 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1033                        struct ctl_be_block_io *beio,
1034                        uint64_t off, uint64_t len, int last)
1035 {
1036         struct bio *bio;
1037         uint64_t maxlen;
1038         struct cdevsw *csw;
1039         struct cdev *dev;
1040         int ref;
1041
1042         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1043         maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize);
1044         while (len > 0) {
1045                 bio = g_alloc_bio();
1046                 bio->bio_cmd        = BIO_DELETE;
1047                 bio->bio_dev        = dev;
1048                 bio->bio_offset     = off;
1049                 bio->bio_length     = MIN(len, maxlen);
1050                 bio->bio_data       = 0;
1051                 bio->bio_done       = ctl_be_block_biodone;
1052                 bio->bio_caller1    = beio;
1053                 bio->bio_pblkno     = off / be_lun->cbe_lun.blocksize;
1054
1055                 off += bio->bio_length;
1056                 len -= bio->bio_length;
1057
1058                 mtx_lock(&be_lun->io_lock);
1059                 beio->num_bios_sent++;
1060                 if (last && len == 0)
1061                         beio->send_complete = 1;
1062                 mtx_unlock(&be_lun->io_lock);
1063
1064                 if (csw) {
1065                         csw->d_strategy(bio);
1066                 } else {
1067                         bio->bio_error = ENXIO;
1068                         ctl_be_block_biodone(bio);
1069                 }
1070         }
1071         if (csw)
1072                 dev_relthread(dev, ref);
1073 }
1074
1075 static void
1076 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1077                        struct ctl_be_block_io *beio)
1078 {
1079         union ctl_io *io;
1080         struct ctl_ptr_len_flags *ptrlen;
1081         struct scsi_unmap_desc *buf, *end;
1082         uint64_t len;
1083
1084         io = beio->io;
1085
1086         DPRINTF("entered\n");
1087
1088         binuptime(&beio->ds_t0);
1089         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1090
1091         if (beio->io_offset == -1) {
1092                 beio->io_len = 0;
1093                 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1094                 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1095                 end = buf + ptrlen->len / sizeof(*buf);
1096                 for (; buf < end; buf++) {
1097                         len = (uint64_t)scsi_4btoul(buf->length) *
1098                             be_lun->cbe_lun.blocksize;
1099                         beio->io_len += len;
1100                         ctl_be_block_unmap_dev_range(be_lun, beio,
1101                             scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize,
1102                             len, (end - buf < 2) ? TRUE : FALSE);
1103                 }
1104         } else
1105                 ctl_be_block_unmap_dev_range(be_lun, beio,
1106                     beio->io_offset, beio->io_len, TRUE);
1107 }
1108
1109 static void
1110 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1111                           struct ctl_be_block_io *beio)
1112 {
1113         TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1114         struct bio *bio;
1115         struct cdevsw *csw;
1116         struct cdev *dev;
1117         off_t cur_offset;
1118         int i, max_iosize, ref;
1119
1120         DPRINTF("entered\n");
1121         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1122
1123         /*
1124          * We have to limit our I/O size to the maximum supported by the
1125          * backend device.  Hopefully it is MAXPHYS.  If the driver doesn't
1126          * set it properly, use DFLTPHYS.
1127          */
1128         if (csw) {
1129                 max_iosize = dev->si_iosize_max;
1130                 if (max_iosize < PAGE_SIZE)
1131                         max_iosize = DFLTPHYS;
1132         } else
1133                 max_iosize = DFLTPHYS;
1134
1135         cur_offset = beio->io_offset;
1136         for (i = 0; i < beio->num_segs; i++) {
1137                 size_t cur_size;
1138                 uint8_t *cur_ptr;
1139
1140                 cur_size = beio->sg_segs[i].len;
1141                 cur_ptr = beio->sg_segs[i].addr;
1142
1143                 while (cur_size > 0) {
1144                         /* This can't fail, it's a blocking allocation. */
1145                         bio = g_alloc_bio();
1146
1147                         KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1148
1149                         bio->bio_cmd = beio->bio_cmd;
1150                         bio->bio_dev = dev;
1151                         bio->bio_caller1 = beio;
1152                         bio->bio_length = min(cur_size, max_iosize);
1153                         bio->bio_offset = cur_offset;
1154                         bio->bio_data = cur_ptr;
1155                         bio->bio_done = ctl_be_block_biodone;
1156                         bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize;
1157
1158                         cur_offset += bio->bio_length;
1159                         cur_ptr += bio->bio_length;
1160                         cur_size -= bio->bio_length;
1161
1162                         TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1163                         beio->num_bios_sent++;
1164                 }
1165         }
1166         beio->send_complete = 1;
1167         binuptime(&beio->ds_t0);
1168         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1169
1170         /*
1171          * Fire off all allocated requests!
1172          */
1173         while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1174                 TAILQ_REMOVE(&queue, bio, bio_queue);
1175                 if (csw)
1176                         csw->d_strategy(bio);
1177                 else {
1178                         bio->bio_error = ENXIO;
1179                         ctl_be_block_biodone(bio);
1180                 }
1181         }
1182         if (csw)
1183                 dev_relthread(dev, ref);
1184 }
1185
1186 static uint64_t
1187 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1188 {
1189         struct diocgattr_arg    arg;
1190         struct cdevsw *csw;
1191         struct cdev *dev;
1192         int error, ref;
1193
1194         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1195         if (csw == NULL)
1196                 return (UINT64_MAX);
1197         strlcpy(arg.name, attrname, sizeof(arg.name));
1198         arg.len = sizeof(arg.value.off);
1199         if (csw->d_ioctl) {
1200                 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
1201                     curthread);
1202         } else
1203                 error = ENODEV;
1204         dev_relthread(dev, ref);
1205         if (error != 0)
1206                 return (UINT64_MAX);
1207         return (arg.value.off);
1208 }
1209
1210 static void
1211 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun,
1212                             union ctl_io *io)
1213 {
1214         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1215         struct ctl_be_block_io *beio;
1216         struct ctl_lba_len_flags *lbalen;
1217
1218         DPRINTF("entered\n");
1219         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1220         lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1221
1222         beio->io_len = lbalen->len * cbe_lun->blocksize;
1223         beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1224         beio->io_arg = (lbalen->flags & SSC_IMMED) != 0;
1225         beio->bio_cmd = BIO_FLUSH;
1226         beio->ds_trans_type = DEVSTAT_NO_DATA;
1227         DPRINTF("SYNC\n");
1228         be_lun->lun_flush(be_lun, beio);
1229 }
1230
1231 static void
1232 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1233 {
1234         union ctl_io *io;
1235
1236         io = beio->io;
1237         ctl_free_beio(beio);
1238         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1239             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1240              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1241                 ctl_config_write_done(io);
1242                 return;
1243         }
1244
1245         ctl_be_block_config_write(io);
1246 }
1247
1248 static void
1249 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1250                             union ctl_io *io)
1251 {
1252         struct ctl_be_block_softc *softc = be_lun->softc;
1253         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1254         struct ctl_be_block_io *beio;
1255         struct ctl_lba_len_flags *lbalen;
1256         uint64_t len_left, lba;
1257         uint32_t pb, pbo, adj;
1258         int i, seglen;
1259         uint8_t *buf, *end;
1260
1261         DPRINTF("entered\n");
1262
1263         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1264         lbalen = ARGS(beio->io);
1265
1266         if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1267             (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1268                 ctl_free_beio(beio);
1269                 ctl_set_invalid_field(&io->scsiio,
1270                                       /*sks_valid*/ 1,
1271                                       /*command*/ 1,
1272                                       /*field*/ 1,
1273                                       /*bit_valid*/ 0,
1274                                       /*bit*/ 0);
1275                 ctl_config_write_done(io);
1276                 return;
1277         }
1278
1279         if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1280                 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1281                 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1282                 beio->bio_cmd = BIO_DELETE;
1283                 beio->ds_trans_type = DEVSTAT_FREE;
1284
1285                 be_lun->unmap(be_lun, beio);
1286                 return;
1287         }
1288
1289         beio->bio_cmd = BIO_WRITE;
1290         beio->ds_trans_type = DEVSTAT_WRITE;
1291
1292         DPRINTF("WRITE SAME at LBA %jx len %u\n",
1293                (uintmax_t)lbalen->lba, lbalen->len);
1294
1295         pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1296         if (be_lun->cbe_lun.pblockoff > 0)
1297                 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1298         else
1299                 pbo = 0;
1300         len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1301         for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1302
1303                 /*
1304                  * Setup the S/G entry for this chunk.
1305                  */
1306                 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1307                 if (pb > cbe_lun->blocksize) {
1308                         adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1309                             seglen - pbo) % pb;
1310                         if (seglen > adj)
1311                                 seglen -= adj;
1312                         else
1313                                 seglen -= seglen % cbe_lun->blocksize;
1314                 } else
1315                         seglen -= seglen % cbe_lun->blocksize;
1316                 beio->sg_segs[i].len = seglen;
1317                 beio->sg_segs[i].addr = uma_zalloc(softc->buf_zone, M_WAITOK);
1318
1319                 DPRINTF("segment %d addr %p len %zd\n", i,
1320                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1321
1322                 beio->num_segs++;
1323                 len_left -= seglen;
1324
1325                 buf = beio->sg_segs[i].addr;
1326                 end = buf + seglen;
1327                 for (; buf < end; buf += cbe_lun->blocksize) {
1328                         if (lbalen->flags & SWS_NDOB) {
1329                                 memset(buf, 0, cbe_lun->blocksize);
1330                         } else {
1331                                 memcpy(buf, io->scsiio.kern_data_ptr,
1332                                     cbe_lun->blocksize);
1333                         }
1334                         if (lbalen->flags & SWS_LBDATA)
1335                                 scsi_ulto4b(lbalen->lba + lba, buf);
1336                         lba++;
1337                 }
1338         }
1339
1340         beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1341         beio->io_len = lba * cbe_lun->blocksize;
1342
1343         /* We can not do all in one run. Correct and schedule rerun. */
1344         if (len_left > 0) {
1345                 lbalen->lba += lba;
1346                 lbalen->len -= lba;
1347                 beio->beio_cont = ctl_be_block_cw_done_ws;
1348         }
1349
1350         be_lun->dispatch(be_lun, beio);
1351 }
1352
1353 static void
1354 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1355                             union ctl_io *io)
1356 {
1357         struct ctl_be_block_io *beio;
1358         struct ctl_ptr_len_flags *ptrlen;
1359
1360         DPRINTF("entered\n");
1361
1362         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1363         ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1364
1365         if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1366                 ctl_free_beio(beio);
1367                 ctl_set_invalid_field(&io->scsiio,
1368                                       /*sks_valid*/ 0,
1369                                       /*command*/ 1,
1370                                       /*field*/ 0,
1371                                       /*bit_valid*/ 0,
1372                                       /*bit*/ 0);
1373                 ctl_config_write_done(io);
1374                 return;
1375         }
1376
1377         beio->io_len = 0;
1378         beio->io_offset = -1;
1379         beio->bio_cmd = BIO_DELETE;
1380         beio->ds_trans_type = DEVSTAT_FREE;
1381         DPRINTF("UNMAP\n");
1382         be_lun->unmap(be_lun, beio);
1383 }
1384
1385 static void
1386 ctl_be_block_cr_done(struct ctl_be_block_io *beio)
1387 {
1388         union ctl_io *io;
1389
1390         io = beio->io;
1391         ctl_free_beio(beio);
1392         ctl_config_read_done(io);
1393 }
1394
1395 static void
1396 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1397                          union ctl_io *io)
1398 {
1399         struct ctl_be_block_io *beio;
1400         struct ctl_be_block_softc *softc;
1401
1402         DPRINTF("entered\n");
1403
1404         softc = be_lun->softc;
1405         beio = ctl_alloc_beio(softc);
1406         beio->io = io;
1407         beio->lun = be_lun;
1408         beio->beio_cont = ctl_be_block_cr_done;
1409         PRIV(io)->ptr = (void *)beio;
1410
1411         switch (io->scsiio.cdb[0]) {
1412         case SERVICE_ACTION_IN:         /* GET LBA STATUS */
1413                 beio->bio_cmd = -1;
1414                 beio->ds_trans_type = DEVSTAT_NO_DATA;
1415                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1416                 beio->io_len = 0;
1417                 if (be_lun->get_lba_status)
1418                         be_lun->get_lba_status(be_lun, beio);
1419                 else
1420                         ctl_be_block_cr_done(beio);
1421                 break;
1422         default:
1423                 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1424                 break;
1425         }
1426 }
1427
1428 static void
1429 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1430 {
1431         union ctl_io *io;
1432
1433         io = beio->io;
1434         ctl_free_beio(beio);
1435         ctl_config_write_done(io);
1436 }
1437
1438 static void
1439 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1440                          union ctl_io *io)
1441 {
1442         struct ctl_be_block_io *beio;
1443         struct ctl_be_block_softc *softc;
1444
1445         DPRINTF("entered\n");
1446
1447         softc = be_lun->softc;
1448         beio = ctl_alloc_beio(softc);
1449         beio->io = io;
1450         beio->lun = be_lun;
1451         beio->beio_cont = ctl_be_block_cw_done;
1452         switch (io->scsiio.tag_type) {
1453         case CTL_TAG_ORDERED:
1454                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1455                 break;
1456         case CTL_TAG_HEAD_OF_QUEUE:
1457                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1458                 break;
1459         case CTL_TAG_UNTAGGED:
1460         case CTL_TAG_SIMPLE:
1461         case CTL_TAG_ACA:
1462         default:
1463                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1464                 break;
1465         }
1466         PRIV(io)->ptr = (void *)beio;
1467
1468         switch (io->scsiio.cdb[0]) {
1469         case SYNCHRONIZE_CACHE:
1470         case SYNCHRONIZE_CACHE_16:
1471                 ctl_be_block_cw_dispatch_sync(be_lun, io);
1472                 break;
1473         case WRITE_SAME_10:
1474         case WRITE_SAME_16:
1475                 ctl_be_block_cw_dispatch_ws(be_lun, io);
1476                 break;
1477         case UNMAP:
1478                 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1479                 break;
1480         default:
1481                 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1482                 break;
1483         }
1484 }
1485
1486 SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t");
1487 SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t");
1488 SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t");
1489 SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t");
1490
1491 static void
1492 ctl_be_block_next(struct ctl_be_block_io *beio)
1493 {
1494         struct ctl_be_block_lun *be_lun;
1495         union ctl_io *io;
1496
1497         io = beio->io;
1498         be_lun = beio->lun;
1499         ctl_free_beio(beio);
1500         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1501             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1502              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1503                 ctl_data_submit_done(io);
1504                 return;
1505         }
1506
1507         io->io_hdr.status &= ~CTL_STATUS_MASK;
1508         io->io_hdr.status |= CTL_STATUS_NONE;
1509
1510         mtx_lock(&be_lun->queue_lock);
1511         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1512         mtx_unlock(&be_lun->queue_lock);
1513         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1514 }
1515
1516 static void
1517 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1518                            union ctl_io *io)
1519 {
1520         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1521         struct ctl_be_block_io *beio;
1522         struct ctl_be_block_softc *softc;
1523         struct ctl_lba_len_flags *lbalen;
1524         struct ctl_ptr_len_flags *bptrlen;
1525         uint64_t len_left, lbas;
1526         int i;
1527
1528         softc = be_lun->softc;
1529
1530         DPRINTF("entered\n");
1531
1532         lbalen = ARGS(io);
1533         if (lbalen->flags & CTL_LLF_WRITE) {
1534                 SDT_PROBE0(cbb, , write, start);
1535         } else {
1536                 SDT_PROBE0(cbb, , read, start);
1537         }
1538
1539         beio = ctl_alloc_beio(softc);
1540         beio->io = io;
1541         beio->lun = be_lun;
1542         bptrlen = PRIV(io);
1543         bptrlen->ptr = (void *)beio;
1544
1545         switch (io->scsiio.tag_type) {
1546         case CTL_TAG_ORDERED:
1547                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1548                 break;
1549         case CTL_TAG_HEAD_OF_QUEUE:
1550                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1551                 break;
1552         case CTL_TAG_UNTAGGED:
1553         case CTL_TAG_SIMPLE:
1554         case CTL_TAG_ACA:
1555         default:
1556                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1557                 break;
1558         }
1559
1560         if (lbalen->flags & CTL_LLF_WRITE) {
1561                 beio->bio_cmd = BIO_WRITE;
1562                 beio->ds_trans_type = DEVSTAT_WRITE;
1563         } else {
1564                 beio->bio_cmd = BIO_READ;
1565                 beio->ds_trans_type = DEVSTAT_READ;
1566         }
1567
1568         DPRINTF("%s at LBA %jx len %u @%ju\n",
1569                (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1570                (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1571         if (lbalen->flags & CTL_LLF_COMPARE) {
1572                 beio->two_sglists = 1;
1573                 lbas = CTLBLK_HALF_IO_SIZE;
1574         } else {
1575                 lbas = CTLBLK_MAX_IO_SIZE;
1576         }
1577         lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize);
1578         beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize;
1579         beio->io_len = lbas * cbe_lun->blocksize;
1580         bptrlen->len += lbas;
1581
1582         for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1583                 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1584                     i, CTLBLK_MAX_SEGS));
1585
1586                 /*
1587                  * Setup the S/G entry for this chunk.
1588                  */
1589                 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1590                 beio->sg_segs[i].addr = uma_zalloc(softc->buf_zone, M_WAITOK);
1591
1592                 DPRINTF("segment %d addr %p len %zd\n", i,
1593                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1594
1595                 /* Set up second segment for compare operation. */
1596                 if (beio->two_sglists) {
1597                         beio->sg_segs[i + CTLBLK_HALF_SEGS].len =
1598                             beio->sg_segs[i].len;
1599                         beio->sg_segs[i + CTLBLK_HALF_SEGS].addr =
1600                             uma_zalloc(softc->buf_zone, M_WAITOK);
1601                 }
1602
1603                 beio->num_segs++;
1604                 len_left -= beio->sg_segs[i].len;
1605         }
1606         if (bptrlen->len < lbalen->len)
1607                 beio->beio_cont = ctl_be_block_next;
1608         io->scsiio.be_move_done = ctl_be_block_move_done;
1609         /* For compare we have separate S/G lists for read and datamove. */
1610         if (beio->two_sglists)
1611                 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1612         else
1613                 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1614         io->scsiio.kern_data_len = beio->io_len;
1615         io->scsiio.kern_sg_entries = beio->num_segs;
1616         io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
1617
1618         /*
1619          * For the read case, we need to read the data into our buffers and
1620          * then we can send it back to the user.  For the write case, we
1621          * need to get the data from the user first.
1622          */
1623         if (beio->bio_cmd == BIO_READ) {
1624                 SDT_PROBE0(cbb, , read, alloc_done);
1625                 be_lun->dispatch(be_lun, beio);
1626         } else {
1627                 SDT_PROBE0(cbb, , write, alloc_done);
1628 #ifdef CTL_TIME_IO
1629                 getbinuptime(&io->io_hdr.dma_start_bt);
1630 #endif
1631                 ctl_datamove(io);
1632         }
1633 }
1634
1635 static void
1636 ctl_be_block_worker(void *context, int pending)
1637 {
1638         struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context;
1639         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1640         union ctl_io *io;
1641         struct ctl_be_block_io *beio;
1642
1643         DPRINTF("entered\n");
1644         /*
1645          * Fetch and process I/Os from all queues.  If we detect LUN
1646          * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race,
1647          * so make response maximally opaque to not confuse initiator.
1648          */
1649         for (;;) {
1650                 mtx_lock(&be_lun->queue_lock);
1651                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1652                 if (io != NULL) {
1653                         DPRINTF("datamove queue\n");
1654                         STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1655                                       ctl_io_hdr, links);
1656                         mtx_unlock(&be_lun->queue_lock);
1657                         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1658                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1659                                 ctl_set_busy(&io->scsiio);
1660                                 ctl_complete_beio(beio);
1661                                 return;
1662                         }
1663                         be_lun->dispatch(be_lun, beio);
1664                         continue;
1665                 }
1666                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1667                 if (io != NULL) {
1668                         DPRINTF("config write queue\n");
1669                         STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1670                                       ctl_io_hdr, links);
1671                         mtx_unlock(&be_lun->queue_lock);
1672                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1673                                 ctl_set_busy(&io->scsiio);
1674                                 ctl_config_write_done(io);
1675                                 return;
1676                         }
1677                         ctl_be_block_cw_dispatch(be_lun, io);
1678                         continue;
1679                 }
1680                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
1681                 if (io != NULL) {
1682                         DPRINTF("config read queue\n");
1683                         STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr,
1684                                       ctl_io_hdr, links);
1685                         mtx_unlock(&be_lun->queue_lock);
1686                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1687                                 ctl_set_busy(&io->scsiio);
1688                                 ctl_config_read_done(io);
1689                                 return;
1690                         }
1691                         ctl_be_block_cr_dispatch(be_lun, io);
1692                         continue;
1693                 }
1694                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1695                 if (io != NULL) {
1696                         DPRINTF("input queue\n");
1697                         STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1698                                       ctl_io_hdr, links);
1699                         mtx_unlock(&be_lun->queue_lock);
1700                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1701                                 ctl_set_busy(&io->scsiio);
1702                                 ctl_data_submit_done(io);
1703                                 return;
1704                         }
1705                         ctl_be_block_dispatch(be_lun, io);
1706                         continue;
1707                 }
1708
1709                 /*
1710                  * If we get here, there is no work left in the queues, so
1711                  * just break out and let the task queue go to sleep.
1712                  */
1713                 mtx_unlock(&be_lun->queue_lock);
1714                 break;
1715         }
1716 }
1717
1718 /*
1719  * Entry point from CTL to the backend for I/O.  We queue everything to a
1720  * work thread, so this just puts the I/O on a queue and wakes up the
1721  * thread.
1722  */
1723 static int
1724 ctl_be_block_submit(union ctl_io *io)
1725 {
1726         struct ctl_be_block_lun *be_lun;
1727         struct ctl_be_lun *cbe_lun;
1728
1729         DPRINTF("entered\n");
1730
1731         cbe_lun = CTL_BACKEND_LUN(io);
1732         be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun;
1733
1734         /*
1735          * Make sure we only get SCSI I/O.
1736          */
1737         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1738                 "%#x) encountered", io->io_hdr.io_type));
1739
1740         PRIV(io)->len = 0;
1741
1742         mtx_lock(&be_lun->queue_lock);
1743         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1744         mtx_unlock(&be_lun->queue_lock);
1745         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1746
1747         return (CTL_RETVAL_COMPLETE);
1748 }
1749
1750 static int
1751 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1752                         int flag, struct thread *td)
1753 {
1754         struct ctl_be_block_softc *softc = &backend_block_softc;
1755         int error;
1756
1757         error = 0;
1758         switch (cmd) {
1759         case CTL_LUN_REQ: {
1760                 struct ctl_lun_req *lun_req;
1761
1762                 lun_req = (struct ctl_lun_req *)addr;
1763
1764                 switch (lun_req->reqtype) {
1765                 case CTL_LUNREQ_CREATE:
1766                         error = ctl_be_block_create(softc, lun_req);
1767                         break;
1768                 case CTL_LUNREQ_RM:
1769                         error = ctl_be_block_rm(softc, lun_req);
1770                         break;
1771                 case CTL_LUNREQ_MODIFY:
1772                         error = ctl_be_block_modify(softc, lun_req);
1773                         break;
1774                 default:
1775                         lun_req->status = CTL_LUN_ERROR;
1776                         snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1777                                  "invalid LUN request type %d",
1778                                  lun_req->reqtype);
1779                         break;
1780                 }
1781                 break;
1782         }
1783         default:
1784                 error = ENOTTY;
1785                 break;
1786         }
1787
1788         return (error);
1789 }
1790
1791 static int
1792 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1793 {
1794         struct ctl_be_lun *cbe_lun;
1795         struct ctl_be_block_filedata *file_data;
1796         struct ctl_lun_create_params *params;
1797         const char                   *value;
1798         struct vattr                  vattr;
1799         off_t                         ps, pss, po, pos, us, uss, uo, uos;
1800         int                           error;
1801
1802         cbe_lun = &be_lun->cbe_lun;
1803         file_data = &be_lun->backend.file;
1804         params = &be_lun->params;
1805
1806         be_lun->dev_type = CTL_BE_BLOCK_FILE;
1807         be_lun->dispatch = ctl_be_block_dispatch_file;
1808         be_lun->lun_flush = ctl_be_block_flush_file;
1809         be_lun->get_lba_status = ctl_be_block_gls_file;
1810         be_lun->getattr = ctl_be_block_getattr_file;
1811         be_lun->unmap = NULL;
1812         cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
1813
1814         error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1815         if (error != 0) {
1816                 snprintf(req->error_str, sizeof(req->error_str),
1817                          "error calling VOP_GETATTR() for file %s",
1818                          be_lun->dev_path);
1819                 return (error);
1820         }
1821
1822         file_data->cred = crhold(curthread->td_ucred);
1823         if (params->lun_size_bytes != 0)
1824                 be_lun->size_bytes = params->lun_size_bytes;
1825         else
1826                 be_lun->size_bytes = vattr.va_size;
1827
1828         /*
1829          * For files we can use any logical block size.  Prefer 512 bytes
1830          * for compatibility reasons.  If file's vattr.va_blocksize
1831          * (preferred I/O block size) is bigger and multiple to chosen
1832          * logical block size -- report it as physical block size.
1833          */
1834         if (params->blocksize_bytes != 0)
1835                 cbe_lun->blocksize = params->blocksize_bytes;
1836         else if (cbe_lun->lun_type == T_CDROM)
1837                 cbe_lun->blocksize = 2048;
1838         else
1839                 cbe_lun->blocksize = 512;
1840         be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
1841         cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
1842             0 : (be_lun->size_blocks - 1);
1843
1844         us = ps = vattr.va_blocksize;
1845         uo = po = 0;
1846
1847         value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
1848         if (value != NULL)
1849                 ctl_expand_number(value, &ps);
1850         value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
1851         if (value != NULL)
1852                 ctl_expand_number(value, &po);
1853         pss = ps / cbe_lun->blocksize;
1854         pos = po / cbe_lun->blocksize;
1855         if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
1856             ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
1857                 cbe_lun->pblockexp = fls(pss) - 1;
1858                 cbe_lun->pblockoff = (pss - pos) % pss;
1859         }
1860
1861         value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
1862         if (value != NULL)
1863                 ctl_expand_number(value, &us);
1864         value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
1865         if (value != NULL)
1866                 ctl_expand_number(value, &uo);
1867         uss = us / cbe_lun->blocksize;
1868         uos = uo / cbe_lun->blocksize;
1869         if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
1870             ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
1871                 cbe_lun->ublockexp = fls(uss) - 1;
1872                 cbe_lun->ublockoff = (uss - uos) % uss;
1873         }
1874
1875         /*
1876          * Sanity check.  The media size has to be at least one
1877          * sector long.
1878          */
1879         if (be_lun->size_bytes < cbe_lun->blocksize) {
1880                 error = EINVAL;
1881                 snprintf(req->error_str, sizeof(req->error_str),
1882                          "file %s size %ju < block size %u", be_lun->dev_path,
1883                          (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize);
1884         }
1885
1886         cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize;
1887         return (error);
1888 }
1889
1890 static int
1891 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1892 {
1893         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1894         struct ctl_lun_create_params *params;
1895         struct cdevsw                *csw;
1896         struct cdev                  *dev;
1897         const char                   *value;
1898         int                           error, atomic, maxio, ref, unmap, tmp;
1899         off_t                         ps, pss, po, pos, us, uss, uo, uos, otmp;
1900
1901         params = &be_lun->params;
1902
1903         be_lun->dev_type = CTL_BE_BLOCK_DEV;
1904         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1905         if (csw == NULL)
1906                 return (ENXIO);
1907         if (strcmp(csw->d_name, "zvol") == 0) {
1908                 be_lun->dispatch = ctl_be_block_dispatch_zvol;
1909                 be_lun->get_lba_status = ctl_be_block_gls_zvol;
1910                 atomic = maxio = CTLBLK_MAX_IO_SIZE;
1911         } else {
1912                 be_lun->dispatch = ctl_be_block_dispatch_dev;
1913                 be_lun->get_lba_status = NULL;
1914                 atomic = 0;
1915                 maxio = dev->si_iosize_max;
1916                 if (maxio <= 0)
1917                         maxio = DFLTPHYS;
1918                 if (maxio > CTLBLK_MAX_IO_SIZE)
1919                         maxio = CTLBLK_MAX_IO_SIZE;
1920         }
1921         be_lun->lun_flush = ctl_be_block_flush_dev;
1922         be_lun->getattr = ctl_be_block_getattr_dev;
1923         be_lun->unmap = ctl_be_block_unmap_dev;
1924
1925         if (!csw->d_ioctl) {
1926                 dev_relthread(dev, ref);
1927                 snprintf(req->error_str, sizeof(req->error_str),
1928                          "no d_ioctl for device %s!", be_lun->dev_path);
1929                 return (ENODEV);
1930         }
1931
1932         error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD,
1933                                curthread);
1934         if (error) {
1935                 dev_relthread(dev, ref);
1936                 snprintf(req->error_str, sizeof(req->error_str),
1937                          "error %d returned for DIOCGSECTORSIZE ioctl "
1938                          "on %s!", error, be_lun->dev_path);
1939                 return (error);
1940         }
1941
1942         /*
1943          * If the user has asked for a blocksize that is greater than the
1944          * backing device's blocksize, we can do it only if the blocksize
1945          * the user is asking for is an even multiple of the underlying
1946          * device's blocksize.
1947          */
1948         if ((params->blocksize_bytes != 0) &&
1949             (params->blocksize_bytes >= tmp)) {
1950                 if (params->blocksize_bytes % tmp == 0) {
1951                         cbe_lun->blocksize = params->blocksize_bytes;
1952                 } else {
1953                         dev_relthread(dev, ref);
1954                         snprintf(req->error_str, sizeof(req->error_str),
1955                                  "requested blocksize %u is not an even "
1956                                  "multiple of backing device blocksize %u",
1957                                  params->blocksize_bytes, tmp);
1958                         return (EINVAL);
1959                 }
1960         } else if (params->blocksize_bytes != 0) {
1961                 dev_relthread(dev, ref);
1962                 snprintf(req->error_str, sizeof(req->error_str),
1963                          "requested blocksize %u < backing device "
1964                          "blocksize %u", params->blocksize_bytes, tmp);
1965                 return (EINVAL);
1966         } else if (cbe_lun->lun_type == T_CDROM)
1967                 cbe_lun->blocksize = MAX(tmp, 2048);
1968         else
1969                 cbe_lun->blocksize = tmp;
1970
1971         error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD,
1972                              curthread);
1973         if (error) {
1974                 dev_relthread(dev, ref);
1975                 snprintf(req->error_str, sizeof(req->error_str),
1976                          "error %d returned for DIOCGMEDIASIZE "
1977                          " ioctl on %s!", error,
1978                          be_lun->dev_path);
1979                 return (error);
1980         }
1981
1982         if (params->lun_size_bytes != 0) {
1983                 if (params->lun_size_bytes > otmp) {
1984                         dev_relthread(dev, ref);
1985                         snprintf(req->error_str, sizeof(req->error_str),
1986                                  "requested LUN size %ju > backing device "
1987                                  "size %ju",
1988                                  (uintmax_t)params->lun_size_bytes,
1989                                  (uintmax_t)otmp);
1990                         return (EINVAL);
1991                 }
1992
1993                 be_lun->size_bytes = params->lun_size_bytes;
1994         } else
1995                 be_lun->size_bytes = otmp;
1996         be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
1997         cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
1998             0 : (be_lun->size_blocks - 1);
1999
2000         error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD,
2001             curthread);
2002         if (error)
2003                 ps = po = 0;
2004         else {
2005                 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po,
2006                     FREAD, curthread);
2007                 if (error)
2008                         po = 0;
2009         }
2010         us = ps;
2011         uo = po;
2012
2013         value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
2014         if (value != NULL)
2015                 ctl_expand_number(value, &ps);
2016         value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
2017         if (value != NULL)
2018                 ctl_expand_number(value, &po);
2019         pss = ps / cbe_lun->blocksize;
2020         pos = po / cbe_lun->blocksize;
2021         if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
2022             ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
2023                 cbe_lun->pblockexp = fls(pss) - 1;
2024                 cbe_lun->pblockoff = (pss - pos) % pss;
2025         }
2026
2027         value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
2028         if (value != NULL)
2029                 ctl_expand_number(value, &us);
2030         value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
2031         if (value != NULL)
2032                 ctl_expand_number(value, &uo);
2033         uss = us / cbe_lun->blocksize;
2034         uos = uo / cbe_lun->blocksize;
2035         if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
2036             ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
2037                 cbe_lun->ublockexp = fls(uss) - 1;
2038                 cbe_lun->ublockoff = (uss - uos) % uss;
2039         }
2040
2041         cbe_lun->atomicblock = atomic / cbe_lun->blocksize;
2042         cbe_lun->opttxferlen = maxio / cbe_lun->blocksize;
2043
2044         if (be_lun->dispatch == ctl_be_block_dispatch_zvol) {
2045                 unmap = 1;
2046         } else {
2047                 struct diocgattr_arg    arg;
2048
2049                 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
2050                 arg.len = sizeof(arg.value.i);
2051                 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
2052                     curthread);
2053                 unmap = (error == 0) ? arg.value.i : 0;
2054         }
2055         value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
2056         if (value != NULL)
2057                 unmap = (strcmp(value, "on") == 0);
2058         if (unmap)
2059                 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
2060         else
2061                 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
2062
2063         dev_relthread(dev, ref);
2064         return (0);
2065 }
2066
2067 static int
2068 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
2069 {
2070         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2071         int flags;
2072
2073         if (be_lun->vn) {
2074                 flags = FREAD;
2075                 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0)
2076                         flags |= FWRITE;
2077                 (void)vn_close(be_lun->vn, flags, NOCRED, curthread);
2078                 be_lun->vn = NULL;
2079
2080                 switch (be_lun->dev_type) {
2081                 case CTL_BE_BLOCK_DEV:
2082                         break;
2083                 case CTL_BE_BLOCK_FILE:
2084                         if (be_lun->backend.file.cred != NULL) {
2085                                 crfree(be_lun->backend.file.cred);
2086                                 be_lun->backend.file.cred = NULL;
2087                         }
2088                         break;
2089                 case CTL_BE_BLOCK_NONE:
2090                         break;
2091                 default:
2092                         panic("Unexpected backend type %d", be_lun->dev_type);
2093                         break;
2094                 }
2095                 be_lun->dev_type = CTL_BE_BLOCK_NONE;
2096         }
2097         return (0);
2098 }
2099
2100 static int
2101 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2102 {
2103         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2104         struct nameidata nd;
2105         const char      *value;
2106         int              error, flags;
2107
2108         error = 0;
2109         if (rootvnode == NULL) {
2110                 snprintf(req->error_str, sizeof(req->error_str),
2111                          "Root filesystem is not mounted");
2112                 return (1);
2113         }
2114         pwd_ensure_dirs();
2115
2116         value = dnvlist_get_string(cbe_lun->options, "file", NULL);
2117         if (value == NULL) {
2118                 snprintf(req->error_str, sizeof(req->error_str),
2119                          "no file argument specified");
2120                 return (1);
2121         }
2122         free(be_lun->dev_path, M_CTLBLK);
2123         be_lun->dev_path = strdup(value, M_CTLBLK);
2124
2125         flags = FREAD;
2126         value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
2127         if (value != NULL) {
2128                 if (strcmp(value, "on") != 0)
2129                         flags |= FWRITE;
2130         } else if (cbe_lun->lun_type == T_DIRECT)
2131                 flags |= FWRITE;
2132
2133 again:
2134         NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread);
2135         error = vn_open(&nd, &flags, 0, NULL);
2136         if ((error == EROFS || error == EACCES) && (flags & FWRITE)) {
2137                 flags &= ~FWRITE;
2138                 goto again;
2139         }
2140         if (error) {
2141                 /*
2142                  * This is the only reasonable guess we can make as far as
2143                  * path if the user doesn't give us a fully qualified path.
2144                  * If they want to specify a file, they need to specify the
2145                  * full path.
2146                  */
2147                 if (be_lun->dev_path[0] != '/') {
2148                         char *dev_name;
2149
2150                         asprintf(&dev_name, M_CTLBLK, "/dev/%s",
2151                                 be_lun->dev_path);
2152                         free(be_lun->dev_path, M_CTLBLK);
2153                         be_lun->dev_path = dev_name;
2154                         goto again;
2155                 }
2156                 snprintf(req->error_str, sizeof(req->error_str),
2157                     "error opening %s: %d", be_lun->dev_path, error);
2158                 return (error);
2159         }
2160         if (flags & FWRITE)
2161                 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY;
2162         else
2163                 cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
2164
2165         NDFREE(&nd, NDF_ONLY_PNBUF);
2166         be_lun->vn = nd.ni_vp;
2167
2168         /* We only support disks and files. */
2169         if (vn_isdisk(be_lun->vn, &error)) {
2170                 error = ctl_be_block_open_dev(be_lun, req);
2171         } else if (be_lun->vn->v_type == VREG) {
2172                 error = ctl_be_block_open_file(be_lun, req);
2173         } else {
2174                 error = EINVAL;
2175                 snprintf(req->error_str, sizeof(req->error_str),
2176                          "%s is not a disk or plain file", be_lun->dev_path);
2177         }
2178         VOP_UNLOCK(be_lun->vn);
2179
2180         if (error != 0)
2181                 ctl_be_block_close(be_lun);
2182         cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2183         if (be_lun->dispatch != ctl_be_block_dispatch_dev)
2184                 cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2185         value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
2186         if (value != NULL && strcmp(value, "on") == 0)
2187                 cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
2188         else if (value != NULL && strcmp(value, "read") == 0)
2189                 cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2190         else if (value != NULL && strcmp(value, "off") == 0)
2191                 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2192         return (0);
2193 }
2194
2195 static int
2196 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2197 {
2198         struct ctl_be_lun *cbe_lun;
2199         struct ctl_be_block_lun *be_lun;
2200         struct ctl_lun_create_params *params;
2201         char num_thread_str[16];
2202         char tmpstr[32];
2203         const char *value;
2204         int retval, num_threads;
2205         int tmp_num_threads;
2206
2207         params = &req->reqdata.create;
2208         retval = 0;
2209         req->status = CTL_LUN_OK;
2210
2211         be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
2212         cbe_lun = &be_lun->cbe_lun;
2213         cbe_lun->be_lun = be_lun;
2214         be_lun->params = req->reqdata.create;
2215         be_lun->softc = softc;
2216         STAILQ_INIT(&be_lun->input_queue);
2217         STAILQ_INIT(&be_lun->config_read_queue);
2218         STAILQ_INIT(&be_lun->config_write_queue);
2219         STAILQ_INIT(&be_lun->datamove_queue);
2220         mtx_init(&be_lun->io_lock, "ctlblock io", NULL, MTX_DEF);
2221         mtx_init(&be_lun->queue_lock, "ctlblock queue", NULL, MTX_DEF);
2222         cbe_lun->options = nvlist_clone(req->args_nvl);
2223
2224         if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
2225                 cbe_lun->lun_type = params->device_type;
2226         else
2227                 cbe_lun->lun_type = T_DIRECT;
2228         be_lun->flags = 0;
2229         cbe_lun->flags = 0;
2230         value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2231         if (value != NULL) {
2232                 if (strcmp(value, "primary") == 0)
2233                         cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2234         } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2235                 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2236
2237         if (cbe_lun->lun_type == T_DIRECT ||
2238             cbe_lun->lun_type == T_CDROM) {
2239                 be_lun->size_bytes = params->lun_size_bytes;
2240                 if (params->blocksize_bytes != 0)
2241                         cbe_lun->blocksize = params->blocksize_bytes;
2242                 else if (cbe_lun->lun_type == T_CDROM)
2243                         cbe_lun->blocksize = 2048;
2244                 else
2245                         cbe_lun->blocksize = 512;
2246                 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2247                 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2248                     0 : (be_lun->size_blocks - 1);
2249
2250                 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2251                     control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2252                         retval = ctl_be_block_open(be_lun, req);
2253                         if (retval != 0) {
2254                                 retval = 0;
2255                                 req->status = CTL_LUN_WARNING;
2256                         }
2257                 }
2258                 num_threads = cbb_num_threads;
2259         } else {
2260                 num_threads = 1;
2261         }
2262
2263         value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL);
2264         if (value != NULL) {
2265                 tmp_num_threads = strtol(value, NULL, 0);
2266
2267                 /*
2268                  * We don't let the user specify less than one
2269                  * thread, but hope he's clueful enough not to
2270                  * specify 1000 threads.
2271                  */
2272                 if (tmp_num_threads < 1) {
2273                         snprintf(req->error_str, sizeof(req->error_str),
2274                                  "invalid number of threads %s",
2275                                  num_thread_str);
2276                         goto bailout_error;
2277                 }
2278                 num_threads = tmp_num_threads;
2279         }
2280
2281         if (be_lun->vn == NULL)
2282                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2283         /* Tell the user the blocksize we ended up using */
2284         params->lun_size_bytes = be_lun->size_bytes;
2285         params->blocksize_bytes = cbe_lun->blocksize;
2286         if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2287                 cbe_lun->req_lun_id = params->req_lun_id;
2288                 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
2289         } else
2290                 cbe_lun->req_lun_id = 0;
2291
2292         cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown;
2293         cbe_lun->be = &ctl_be_block_driver;
2294
2295         if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2296                 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
2297                          softc->num_luns);
2298                 strncpy((char *)cbe_lun->serial_num, tmpstr,
2299                         MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
2300
2301                 /* Tell the user what we used for a serial number */
2302                 strncpy((char *)params->serial_num, tmpstr,
2303                         MIN(sizeof(params->serial_num), sizeof(tmpstr)));
2304         } else {
2305                 strncpy((char *)cbe_lun->serial_num, params->serial_num,
2306                         MIN(sizeof(cbe_lun->serial_num),
2307                         sizeof(params->serial_num)));
2308         }
2309         if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2310                 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
2311                 strncpy((char *)cbe_lun->device_id, tmpstr,
2312                         MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
2313
2314                 /* Tell the user what we used for a device ID */
2315                 strncpy((char *)params->device_id, tmpstr,
2316                         MIN(sizeof(params->device_id), sizeof(tmpstr)));
2317         } else {
2318                 strncpy((char *)cbe_lun->device_id, params->device_id,
2319                         MIN(sizeof(cbe_lun->device_id),
2320                             sizeof(params->device_id)));
2321         }
2322
2323         TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2324
2325         be_lun->io_taskqueue = taskqueue_create("ctlblocktq", M_WAITOK,
2326             taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2327
2328         if (be_lun->io_taskqueue == NULL) {
2329                 snprintf(req->error_str, sizeof(req->error_str),
2330                          "unable to create taskqueue");
2331                 goto bailout_error;
2332         }
2333
2334         /*
2335          * Note that we start the same number of threads by default for
2336          * both the file case and the block device case.  For the file
2337          * case, we need multiple threads to allow concurrency, because the
2338          * vnode interface is designed to be a blocking interface.  For the
2339          * block device case, ZFS zvols at least will block the caller's
2340          * context in many instances, and so we need multiple threads to
2341          * overcome that problem.  Other block devices don't need as many
2342          * threads, but they shouldn't cause too many problems.
2343          *
2344          * If the user wants to just have a single thread for a block
2345          * device, he can specify that when the LUN is created, or change
2346          * the tunable/sysctl to alter the default number of threads.
2347          */
2348         retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
2349                                          /*num threads*/num_threads,
2350                                          /*priority*/PUSER,
2351                                          /*proc*/control_softc->ctl_proc,
2352                                          /*thread name*/"block");
2353
2354         if (retval != 0)
2355                 goto bailout_error;
2356
2357         be_lun->num_threads = num_threads;
2358
2359         retval = ctl_add_lun(&be_lun->cbe_lun);
2360         if (retval != 0) {
2361                 snprintf(req->error_str, sizeof(req->error_str),
2362                          "ctl_add_lun() returned error %d, see dmesg for "
2363                          "details", retval);
2364                 retval = 0;
2365                 goto bailout_error;
2366         }
2367
2368         be_lun->disk_stats = devstat_new_entry("cbb", cbe_lun->lun_id,
2369                                                cbe_lun->blocksize,
2370                                                DEVSTAT_ALL_SUPPORTED,
2371                                                cbe_lun->lun_type
2372                                                | DEVSTAT_TYPE_IF_OTHER,
2373                                                DEVSTAT_PRIORITY_OTHER);
2374
2375         mtx_lock(&softc->lock);
2376         softc->num_luns++;
2377         SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
2378         mtx_unlock(&softc->lock);
2379
2380         params->req_lun_id = cbe_lun->lun_id;
2381
2382         return (retval);
2383
2384 bailout_error:
2385         req->status = CTL_LUN_ERROR;
2386
2387         if (be_lun->io_taskqueue != NULL)
2388                 taskqueue_free(be_lun->io_taskqueue);
2389         ctl_be_block_close(be_lun);
2390         if (be_lun->dev_path != NULL)
2391                 free(be_lun->dev_path, M_CTLBLK);
2392         nvlist_destroy(cbe_lun->options);
2393         mtx_destroy(&be_lun->queue_lock);
2394         mtx_destroy(&be_lun->io_lock);
2395         free(be_lun, M_CTLBLK);
2396
2397         return (retval);
2398 }
2399
2400 static int
2401 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2402 {
2403         struct ctl_lun_rm_params *params;
2404         struct ctl_be_block_lun *be_lun;
2405         struct ctl_be_lun *cbe_lun;
2406         int retval;
2407
2408         params = &req->reqdata.rm;
2409
2410         sx_xlock(&softc->modify_lock);
2411         mtx_lock(&softc->lock);
2412         SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2413                 if (be_lun->cbe_lun.lun_id == params->lun_id) {
2414                         SLIST_REMOVE(&softc->lun_list, be_lun,
2415                             ctl_be_block_lun, links);
2416                         softc->num_luns--;
2417                         break;
2418                 }
2419         }
2420         mtx_unlock(&softc->lock);
2421         sx_xunlock(&softc->modify_lock);
2422         if (be_lun == NULL) {
2423                 snprintf(req->error_str, sizeof(req->error_str),
2424                          "LUN %u is not managed by the block backend",
2425                          params->lun_id);
2426                 goto bailout_error;
2427         }
2428         cbe_lun = &be_lun->cbe_lun;
2429
2430         if (be_lun->vn != NULL) {
2431                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2432                 ctl_lun_no_media(cbe_lun);
2433                 taskqueue_drain_all(be_lun->io_taskqueue);
2434                 ctl_be_block_close(be_lun);
2435         }
2436
2437         mtx_lock(&softc->lock);
2438         be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2439         mtx_unlock(&softc->lock);
2440
2441         retval = ctl_remove_lun(cbe_lun);
2442         if (retval != 0) {
2443                 snprintf(req->error_str, sizeof(req->error_str),
2444                          "error %d returned from ctl_remove_lun() for "
2445                          "LUN %d", retval, params->lun_id);
2446                 mtx_lock(&softc->lock);
2447                 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2448                 mtx_unlock(&softc->lock);
2449                 goto bailout_error;
2450         }
2451
2452         mtx_lock(&softc->lock);
2453         while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2454                 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblockrm", 0);
2455                 if (retval == EINTR)
2456                         break;
2457         }
2458         be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2459         if (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2460                 mtx_unlock(&softc->lock);
2461                 free(be_lun, M_CTLBLK);
2462         } else {
2463                 mtx_unlock(&softc->lock);
2464                 return (EINTR);
2465         }
2466
2467         req->status = CTL_LUN_OK;
2468         return (0);
2469
2470 bailout_error:
2471         req->status = CTL_LUN_ERROR;
2472         return (0);
2473 }
2474
2475 static int
2476 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2477 {
2478         struct ctl_lun_modify_params *params;
2479         struct ctl_be_block_lun *be_lun;
2480         struct ctl_be_lun *cbe_lun;
2481         const char *value;
2482         uint64_t oldsize;
2483         int error, wasprim;
2484
2485         params = &req->reqdata.modify;
2486
2487         sx_xlock(&softc->modify_lock);
2488         mtx_lock(&softc->lock);
2489         SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2490                 if (be_lun->cbe_lun.lun_id == params->lun_id)
2491                         break;
2492         }
2493         mtx_unlock(&softc->lock);
2494         if (be_lun == NULL) {
2495                 snprintf(req->error_str, sizeof(req->error_str),
2496                          "LUN %u is not managed by the block backend",
2497                          params->lun_id);
2498                 goto bailout_error;
2499         }
2500         cbe_lun = &be_lun->cbe_lun;
2501
2502         if (params->lun_size_bytes != 0)
2503                 be_lun->params.lun_size_bytes = params->lun_size_bytes;
2504
2505         if (req->args_nvl != NULL) {
2506                 nvlist_destroy(cbe_lun->options);
2507                 cbe_lun->options = nvlist_clone(req->args_nvl);
2508         }
2509
2510         wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
2511         value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2512         if (value != NULL) {
2513                 if (strcmp(value, "primary") == 0)
2514                         cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2515                 else
2516                         cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2517         } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2518                 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2519         else
2520                 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2521         if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
2522                 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
2523                         ctl_lun_primary(cbe_lun);
2524                 else
2525                         ctl_lun_secondary(cbe_lun);
2526         }
2527
2528         oldsize = be_lun->size_blocks;
2529         if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2530             control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2531                 if (be_lun->vn == NULL)
2532                         error = ctl_be_block_open(be_lun, req);
2533                 else if (vn_isdisk(be_lun->vn, &error))
2534                         error = ctl_be_block_open_dev(be_lun, req);
2535                 else if (be_lun->vn->v_type == VREG) {
2536                         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2537                         error = ctl_be_block_open_file(be_lun, req);
2538                         VOP_UNLOCK(be_lun->vn);
2539                 } else
2540                         error = EINVAL;
2541                 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) &&
2542                     be_lun->vn != NULL) {
2543                         cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2544                         ctl_lun_has_media(cbe_lun);
2545                 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 &&
2546                     be_lun->vn == NULL) {
2547                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2548                         ctl_lun_no_media(cbe_lun);
2549                 }
2550                 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2551         } else {
2552                 if (be_lun->vn != NULL) {
2553                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2554                         ctl_lun_no_media(cbe_lun);
2555                         taskqueue_drain_all(be_lun->io_taskqueue);
2556                         error = ctl_be_block_close(be_lun);
2557                 } else
2558                         error = 0;
2559         }
2560         if (be_lun->size_blocks != oldsize)
2561                 ctl_lun_capacity_changed(cbe_lun);
2562
2563         /* Tell the user the exact size we ended up using */
2564         params->lun_size_bytes = be_lun->size_bytes;
2565
2566         sx_xunlock(&softc->modify_lock);
2567         req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2568         return (0);
2569
2570 bailout_error:
2571         sx_xunlock(&softc->modify_lock);
2572         req->status = CTL_LUN_ERROR;
2573         return (0);
2574 }
2575
2576 static void
2577 ctl_be_block_lun_shutdown(void *lun)
2578 {
2579         struct ctl_be_block_lun *be_lun = lun;
2580         struct ctl_be_block_softc *softc = be_lun->softc;
2581
2582         taskqueue_drain_all(be_lun->io_taskqueue);
2583         taskqueue_free(be_lun->io_taskqueue);
2584         if (be_lun->disk_stats != NULL)
2585                 devstat_remove_entry(be_lun->disk_stats);
2586         nvlist_destroy(be_lun->cbe_lun.options);
2587         free(be_lun->dev_path, M_CTLBLK);
2588         mtx_destroy(&be_lun->queue_lock);
2589         mtx_destroy(&be_lun->io_lock);
2590
2591         mtx_lock(&softc->lock);
2592         be_lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2593         if (be_lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2594                 wakeup(be_lun);
2595         else
2596                 free(be_lun, M_CTLBLK);
2597         mtx_unlock(&softc->lock);
2598 }
2599
2600 static int
2601 ctl_be_block_config_write(union ctl_io *io)
2602 {
2603         struct ctl_be_block_lun *be_lun;
2604         struct ctl_be_lun *cbe_lun;
2605         int retval;
2606
2607         DPRINTF("entered\n");
2608
2609         cbe_lun = CTL_BACKEND_LUN(io);
2610         be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun;
2611
2612         retval = 0;
2613         switch (io->scsiio.cdb[0]) {
2614         case SYNCHRONIZE_CACHE:
2615         case SYNCHRONIZE_CACHE_16:
2616         case WRITE_SAME_10:
2617         case WRITE_SAME_16:
2618         case UNMAP:
2619                 /*
2620                  * The upper level CTL code will filter out any CDBs with
2621                  * the immediate bit set and return the proper error.
2622                  *
2623                  * We don't really need to worry about what LBA range the
2624                  * user asked to be synced out.  When they issue a sync
2625                  * cache command, we'll sync out the whole thing.
2626                  */
2627                 mtx_lock(&be_lun->queue_lock);
2628                 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2629                                    links);
2630                 mtx_unlock(&be_lun->queue_lock);
2631                 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2632                 break;
2633         case START_STOP_UNIT: {
2634                 struct scsi_start_stop_unit *cdb;
2635                 struct ctl_lun_req req;
2636
2637                 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2638                 if ((cdb->how & SSS_PC_MASK) != 0) {
2639                         ctl_set_success(&io->scsiio);
2640                         ctl_config_write_done(io);
2641                         break;
2642                 }
2643                 if (cdb->how & SSS_START) {
2644                         if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) {
2645                                 retval = ctl_be_block_open(be_lun, &req);
2646                                 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2647                                 if (retval == 0) {
2648                                         cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2649                                         ctl_lun_has_media(cbe_lun);
2650                                 } else {
2651                                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2652                                         ctl_lun_no_media(cbe_lun);
2653                                 }
2654                         }
2655                         ctl_start_lun(cbe_lun);
2656                 } else {
2657                         ctl_stop_lun(cbe_lun);
2658                         if (cdb->how & SSS_LOEJ) {
2659                                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2660                                 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED;
2661                                 ctl_lun_ejected(cbe_lun);
2662                                 if (be_lun->vn != NULL)
2663                                         ctl_be_block_close(be_lun);
2664                         }
2665                 }
2666
2667                 ctl_set_success(&io->scsiio);
2668                 ctl_config_write_done(io);
2669                 break;
2670         }
2671         case PREVENT_ALLOW:
2672                 ctl_set_success(&io->scsiio);
2673                 ctl_config_write_done(io);
2674                 break;
2675         default:
2676                 ctl_set_invalid_opcode(&io->scsiio);
2677                 ctl_config_write_done(io);
2678                 retval = CTL_RETVAL_COMPLETE;
2679                 break;
2680         }
2681
2682         return (retval);
2683 }
2684
2685 static int
2686 ctl_be_block_config_read(union ctl_io *io)
2687 {
2688         struct ctl_be_block_lun *be_lun;
2689         struct ctl_be_lun *cbe_lun;
2690         int retval = 0;
2691
2692         DPRINTF("entered\n");
2693
2694         cbe_lun = CTL_BACKEND_LUN(io);
2695         be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun;
2696
2697         switch (io->scsiio.cdb[0]) {
2698         case SERVICE_ACTION_IN:
2699                 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
2700                         mtx_lock(&be_lun->queue_lock);
2701                         STAILQ_INSERT_TAIL(&be_lun->config_read_queue,
2702                             &io->io_hdr, links);
2703                         mtx_unlock(&be_lun->queue_lock);
2704                         taskqueue_enqueue(be_lun->io_taskqueue,
2705                             &be_lun->io_task);
2706                         retval = CTL_RETVAL_QUEUED;
2707                         break;
2708                 }
2709                 ctl_set_invalid_field(&io->scsiio,
2710                                       /*sks_valid*/ 1,
2711                                       /*command*/ 1,
2712                                       /*field*/ 1,
2713                                       /*bit_valid*/ 1,
2714                                       /*bit*/ 4);
2715                 ctl_config_read_done(io);
2716                 retval = CTL_RETVAL_COMPLETE;
2717                 break;
2718         default:
2719                 ctl_set_invalid_opcode(&io->scsiio);
2720                 ctl_config_read_done(io);
2721                 retval = CTL_RETVAL_COMPLETE;
2722                 break;
2723         }
2724
2725         return (retval);
2726 }
2727
2728 static int
2729 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb)
2730 {
2731         struct ctl_be_block_lun *lun;
2732         int retval;
2733
2734         lun = (struct ctl_be_block_lun *)be_lun;
2735
2736         retval = sbuf_printf(sb, "\t<num_threads>");
2737         if (retval != 0)
2738                 goto bailout;
2739         retval = sbuf_printf(sb, "%d", lun->num_threads);
2740         if (retval != 0)
2741                 goto bailout;
2742         retval = sbuf_printf(sb, "</num_threads>\n");
2743
2744 bailout:
2745         return (retval);
2746 }
2747
2748 static uint64_t
2749 ctl_be_block_lun_attr(void *be_lun, const char *attrname)
2750 {
2751         struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun;
2752
2753         if (lun->getattr == NULL)
2754                 return (UINT64_MAX);
2755         return (lun->getattr(lun, attrname));
2756 }
2757
2758 static int
2759 ctl_be_block_init(void)
2760 {
2761         struct ctl_be_block_softc *softc = &backend_block_softc;
2762
2763         sx_init(&softc->modify_lock, "ctlblock modify");
2764         mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
2765         softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2766             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2767         softc->buf_zone = uma_zcreate("ctlblock", CTLBLK_MAX_SEG,
2768             NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
2769         SLIST_INIT(&softc->lun_list);
2770         return (0);
2771 }
2772
2773
2774 static int
2775 ctl_be_block_shutdown(void)
2776 {
2777         struct ctl_be_block_softc *softc = &backend_block_softc;
2778         struct ctl_be_block_lun *lun;
2779
2780         mtx_lock(&softc->lock);
2781         while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
2782                 SLIST_REMOVE_HEAD(&softc->lun_list, links);
2783                 softc->num_luns--;
2784                 /*
2785                  * Drop our lock here.  Since ctl_remove_lun() can call
2786                  * back into us, this could potentially lead to a recursive
2787                  * lock of the same mutex, which would cause a hang.
2788                  */
2789                 mtx_unlock(&softc->lock);
2790                 ctl_remove_lun(&lun->cbe_lun);
2791                 mtx_lock(&softc->lock);
2792         }
2793         mtx_unlock(&softc->lock);
2794         uma_zdestroy(softc->buf_zone);
2795         uma_zdestroy(softc->beio_zone);
2796         mtx_destroy(&softc->lock);
2797         sx_destroy(&softc->modify_lock);
2798         return (0);
2799 }