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_be_lun cbe_lun;              /* Must be first element. */
 156         struct ctl_lun_create_params params;
 157         char *dev_path;
 158         ctl_be_block_type dev_type;
 159         struct vnode *vn;
 160         union ctl_be_block_bedata backend;
 161         cbb_dispatch_t dispatch;
 162         cbb_dispatch_t lun_flush;
 163         cbb_dispatch_t unmap;
 164         cbb_dispatch_t get_lba_status;
 165         cbb_getattr_t getattr;
 166         uint64_t size_blocks;
 167         uint64_t size_bytes;
 168         struct ctl_be_block_softc *softc;
 169         struct devstat *disk_stats;
 170         ctl_be_block_lun_flags flags;
 171         SLIST_ENTRY(ctl_be_block_lun) links;
 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                             refcnt;
 205         int                             bio_cmd;
 206         int                             two_sglists;
 207         int                             num_segs;
 208         int                             num_bios_sent;
 209         int                             num_bios_done;
 210         int                             send_complete;
 211         int                             first_error;
 212         uint64_t                        first_error_offset;
 213         struct bintime                  ds_t0;
 214         devstat_tag_type                ds_tag_type;
 215         devstat_trans_flags             ds_trans_type;
 216         uint64_t                        io_len;
 217         uint64_t                        io_offset;
 218         int                             io_arg;
 219         struct ctl_be_block_softc       *softc;
 220         struct ctl_be_block_lun         *lun;
 221         void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
 222 };
 223
 224 extern struct ctl_softc *control_softc;
 225
 226 static int cbb_num_threads = 14;
 227 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 228             "CAM Target Layer Block Backend");
 229 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
 230            &cbb_num_threads, 0, "Number of threads per backing file");
 231
 232 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
 233 static void ctl_free_beio(struct ctl_be_block_io *beio);
 234 static void ctl_complete_beio(struct ctl_be_block_io *beio);
 235 static int ctl_be_block_move_done(union ctl_io *io);
 236 static void ctl_be_block_biodone(struct bio *bio);
 237 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 238                                     struct ctl_be_block_io *beio);
 239 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 240                                        struct ctl_be_block_io *beio);
 241 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
 242                                   struct ctl_be_block_io *beio);
 243 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
 244                                          const char *attrname);
 245 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
 246                                    struct ctl_be_block_io *beio);
 247 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
 248                                    struct ctl_be_block_io *beio);
 249 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
 250                                       struct ctl_be_block_io *beio);
 251 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
 252                                          const char *attrname);
 253 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
 254                                     union ctl_io *io);
 255 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
 256                                     union ctl_io *io);
 257 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
 258                                   union ctl_io *io);
 259 static void ctl_be_block_worker(void *context, int pending);
 260 static int ctl_be_block_submit(union ctl_io *io);
 261 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
 262                                    int flag, struct thread *td);
 263 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
 264                                   struct ctl_lun_req *req);
 265 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
 266                                  struct ctl_lun_req *req);
 267 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
 268 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun,
 269                              struct ctl_lun_req *req);
 270 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
 271                                struct ctl_lun_req *req);
 272 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
 273                            struct ctl_lun_req *req);
 274 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
 275                            struct ctl_lun_req *req);
 276 static void ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun);
 277 static int ctl_be_block_config_write(union ctl_io *io);
 278 static int ctl_be_block_config_read(union ctl_io *io);
 279 static int ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb);
 280 static uint64_t ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname);
 281 static int ctl_be_block_init(void);
 282 static int ctl_be_block_shutdown(void);
 283
 284 static struct ctl_backend_driver ctl_be_block_driver =
 285 {
 286         .name = "block",
 287         .flags = CTL_BE_FLAG_HAS_CONFIG,
 288         .init = ctl_be_block_init,
 289         .shutdown = ctl_be_block_shutdown,
 290         .data_submit = ctl_be_block_submit,
 291         .data_move_done = ctl_be_block_move_done,
 292         .config_read = ctl_be_block_config_read,
 293         .config_write = ctl_be_block_config_write,
 294         .ioctl = ctl_be_block_ioctl,
 295         .lun_info = ctl_be_block_lun_info,
 296         .lun_attr = ctl_be_block_lun_attr
 297 };
 298
 299 MALLOC_DEFINE(M_CTLBLK, "ctlblock", "Memory used for CTL block backend");
 300 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
 301
 302 static struct ctl_be_block_io *
 303 ctl_alloc_beio(struct ctl_be_block_softc *softc)
 304 {
 305         struct ctl_be_block_io *beio;
 306
 307         beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO);
 308         beio->softc = softc;
 309         beio->refcnt = 1;
 310         return (beio);
 311 }
 312
 313 static void
 314 ctl_real_free_beio(struct ctl_be_block_io *beio)
 315 {
 316         struct ctl_be_block_softc *softc = beio->softc;
 317         int i;
 318
 319         for (i = 0; i < beio->num_segs; i++) {
 320                 uma_zfree(softc->buf_zone, beio->sg_segs[i].addr);
 321
 322                 /* For compare we had two equal S/G lists. */
 323                 if (beio->two_sglists) {
 324                         uma_zfree(softc->buf_zone,
 325                             beio->sg_segs[i + CTLBLK_HALF_SEGS].addr);
 326                 }
 327         }
 328
 329         uma_zfree(softc->beio_zone, beio);
 330 }
 331
 332 static void
 333 ctl_refcnt_beio(void *arg, int diff)
 334 {
 335         struct ctl_be_block_io *beio = arg;
 336
 337         if (atomic_fetchadd_int(&beio->refcnt, diff) + diff == 0)
 338                 ctl_real_free_beio(beio);
 339 }
 340
 341 static void
 342 ctl_free_beio(struct ctl_be_block_io *beio)
 343 {
 344
 345         ctl_refcnt_beio(beio, -1);
 346 }
 347
 348 static void
 349 ctl_complete_beio(struct ctl_be_block_io *beio)
 350 {
 351         union ctl_io *io = beio->io;
 352
 353         if (beio->beio_cont != NULL) {
 354                 beio->beio_cont(beio);
 355         } else {
 356                 ctl_free_beio(beio);
 357                 ctl_data_submit_done(io);
 358         }
 359 }
 360
 361 static size_t
 362 cmp(uint8_t *a, uint8_t *b, size_t size)
 363 {
 364         size_t i;
 365
 366         for (i = 0; i < size; i++) {
 367                 if (a[i] != b[i])
 368                         break;
 369         }
 370         return (i);
 371 }
 372
 373 static void
 374 ctl_be_block_compare(union ctl_io *io)
 375 {
 376         struct ctl_be_block_io *beio;
 377         uint64_t off, res;
 378         int i;
 379         uint8_t info[8];
 380
 381         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
 382         off = 0;
 383         for (i = 0; i < beio->num_segs; i++) {
 384                 res = cmp(beio->sg_segs[i].addr,
 385                     beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
 386                     beio->sg_segs[i].len);
 387                 off += res;
 388                 if (res < beio->sg_segs[i].len)
 389                         break;
 390         }
 391         if (i < beio->num_segs) {
 392                 scsi_u64to8b(off, info);
 393                 ctl_set_sense(&io->scsiio, /*current_error*/ 1,
 394                     /*sense_key*/ SSD_KEY_MISCOMPARE,
 395                     /*asc*/ 0x1D, /*ascq*/ 0x00,
 396                     /*type*/ SSD_ELEM_INFO,
 397                     /*size*/ sizeof(info), /*data*/ &info,
 398                     /*type*/ SSD_ELEM_NONE);
 399         } else
 400                 ctl_set_success(&io->scsiio);
 401 }
 402
 403 static int
 404 ctl_be_block_move_done(union ctl_io *io)
 405 {
 406         struct ctl_be_block_io *beio;
 407         struct ctl_be_block_lun *be_lun;
 408         struct ctl_lba_len_flags *lbalen;
 409 #ifdef CTL_TIME_IO
 410         struct bintime cur_bt;
 411 #endif
 412
 413         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
 414         be_lun = beio->lun;
 415
 416         DPRINTF("entered\n");
 417
 418 #ifdef CTL_TIME_IO
 419         getbinuptime(&cur_bt);
 420         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
 421         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
 422 #endif
 423         io->io_hdr.num_dmas++;
 424         io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
 425
 426         /*
 427          * We set status at this point for read commands, and write
 428          * commands with errors.
 429          */
 430         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
 431                 ;
 432         } else if ((io->io_hdr.port_status != 0) &&
 433             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
 434              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
 435                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
 436                     /*retry_count*/ io->io_hdr.port_status);
 437         } else if (io->scsiio.kern_data_resid != 0 &&
 438             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
 439             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
 440              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
 441                 ctl_set_invalid_field_ciu(&io->scsiio);
 442         } else if ((io->io_hdr.port_status == 0) &&
 443             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
 444                 lbalen = ARGS(beio->io);
 445                 if (lbalen->flags & CTL_LLF_READ) {
 446                         ctl_set_success(&io->scsiio);
 447                 } else if (lbalen->flags & CTL_LLF_COMPARE) {
 448                         /* We have two data blocks ready for comparison. */
 449                         ctl_be_block_compare(io);
 450                 }
 451         }
 452
 453         /*
 454          * If this is a read, or a write with errors, it is done.
 455          */
 456         if ((beio->bio_cmd == BIO_READ)
 457          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
 458          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
 459                 ctl_complete_beio(beio);
 460                 return (0);
 461         }
 462
 463         /*
 464          * At this point, we have a write and the DMA completed
 465          * successfully.  We now have to queue it to the task queue to
 466          * execute the backend I/O.  That is because we do blocking
 467          * memory allocations, and in the file backing case, blocking I/O.
 468          * This move done routine is generally called in the SIM's
 469          * interrupt context, and therefore we cannot block.
 470          */
 471         mtx_lock(&be_lun->queue_lock);
 472         STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
 473         mtx_unlock(&be_lun->queue_lock);
 474         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
 475
 476         return (0);
 477 }
 478
 479 static void
 480 ctl_be_block_biodone(struct bio *bio)
 481 {
 482         struct ctl_be_block_io *beio;
 483         struct ctl_be_block_lun *be_lun;
 484         union ctl_io *io;
 485         int error;
 486
 487         beio = bio->bio_caller1;
 488         be_lun = beio->lun;
 489         io = beio->io;
 490
 491         DPRINTF("entered\n");
 492
 493         error = bio->bio_error;
 494         mtx_lock(&be_lun->io_lock);
 495         if (error != 0 &&
 496             (beio->first_error == 0 ||
 497              bio->bio_offset < beio->first_error_offset)) {
 498                 beio->first_error = error;
 499                 beio->first_error_offset = bio->bio_offset;
 500         }
 501
 502         beio->num_bios_done++;
 503
 504         /*
 505          * XXX KDM will this cause WITNESS to complain?  Holding a lock
 506          * during the free might cause it to complain.
 507          */
 508         g_destroy_bio(bio);
 509
 510         /*
 511          * If the send complete bit isn't set, or we aren't the last I/O to
 512          * complete, then we're done.
 513          */
 514         if ((beio->send_complete == 0)
 515          || (beio->num_bios_done < beio->num_bios_sent)) {
 516                 mtx_unlock(&be_lun->io_lock);
 517                 return;
 518         }
 519
 520         /*
 521          * At this point, we've verified that we are the last I/O to
 522          * complete, so it's safe to drop the lock.
 523          */
 524         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 525             beio->ds_tag_type, beio->ds_trans_type,
 526             /*now*/ NULL, /*then*/&beio->ds_t0);
 527         mtx_unlock(&be_lun->io_lock);
 528
 529         /*
 530          * If there are any errors from the backing device, we fail the
 531          * entire I/O with a medium error.
 532          */
 533         error = beio->first_error;
 534         if (error != 0) {
 535                 if (error == EOPNOTSUPP) {
 536                         ctl_set_invalid_opcode(&io->scsiio);
 537                 } else if (error == ENOSPC || error == EDQUOT) {
 538                         ctl_set_space_alloc_fail(&io->scsiio);
 539                 } else if (error == EROFS || error == EACCES) {
 540                         ctl_set_hw_write_protected(&io->scsiio);
 541                 } else if (beio->bio_cmd == BIO_FLUSH) {
 542                         /* XXX KDM is there is a better error here? */
 543                         ctl_set_internal_failure(&io->scsiio,
 544                                                  /*sks_valid*/ 1,
 545                                                  /*retry_count*/ 0xbad2);
 546                 } else {
 547                         ctl_set_medium_error(&io->scsiio,
 548                             beio->bio_cmd == BIO_READ);
 549                 }
 550                 ctl_complete_beio(beio);
 551                 return;
 552         }
 553
 554         /*
 555          * If this is a write, a flush, a delete or verify, we're all done.
 556          * If this is a read, we can now send the data to the user.
 557          */
 558         if ((beio->bio_cmd == BIO_WRITE)
 559          || (beio->bio_cmd == BIO_FLUSH)
 560          || (beio->bio_cmd == BIO_DELETE)
 561          || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 562                 ctl_set_success(&io->scsiio);
 563                 ctl_complete_beio(beio);
 564         } else {
 565                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 566                     beio->beio_cont == NULL) {
 567                         ctl_set_success(&io->scsiio);
 568                         ctl_serseq_done(io);
 569                 }
 570 #ifdef CTL_TIME_IO
 571                 getbinuptime(&io->io_hdr.dma_start_bt);
 572 #endif
 573                 ctl_datamove(io);
 574         }
 575 }
 576
 577 static void
 578 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 579                         struct ctl_be_block_io *beio)
 580 {
 581         union ctl_io *io = beio->io;
 582         struct mount *mountpoint;
 583         int error, lock_flags;
 584
 585         DPRINTF("entered\n");
 586
 587         binuptime(&beio->ds_t0);
 588         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 589
 590         (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 591
 592         if (MNT_SHARED_WRITES(mountpoint) ||
 593             ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 594                 lock_flags = LK_SHARED;
 595         else
 596                 lock_flags = LK_EXCLUSIVE;
 597         vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 598         error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT,
 599             curthread);
 600         VOP_UNLOCK(be_lun->vn);
 601
 602         vn_finished_write(mountpoint);
 603
 604         mtx_lock(&be_lun->io_lock);
 605         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 606             beio->ds_tag_type, beio->ds_trans_type,
 607             /*now*/ NULL, /*then*/&beio->ds_t0);
 608         mtx_unlock(&be_lun->io_lock);
 609
 610         if (error == 0)
 611                 ctl_set_success(&io->scsiio);
 612         else {
 613                 /* XXX KDM is there is a better error here? */
 614                 ctl_set_internal_failure(&io->scsiio,
 615                                          /*sks_valid*/ 1,
 616                                          /*retry_count*/ 0xbad1);
 617         }
 618
 619         ctl_complete_beio(beio);
 620 }
 621
 622 SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t");
 623 SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t");
 624 SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t");
 625 SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t");
 626
 627 static void
 628 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 629                            struct ctl_be_block_io *beio)
 630 {
 631         struct ctl_be_block_filedata *file_data;
 632         union ctl_io *io;
 633         struct uio xuio;
 634         struct iovec *xiovec;
 635         size_t s;
 636         int error, flags, i;
 637
 638         DPRINTF("entered\n");
 639
 640         file_data = &be_lun->backend.file;
 641         io = beio->io;
 642         flags = 0;
 643         if (ARGS(io)->flags & CTL_LLF_DPO)
 644                 flags |= IO_DIRECT;
 645         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 646                 flags |= IO_SYNC;
 647
 648         bzero(&xuio, sizeof(xuio));
 649         if (beio->bio_cmd == BIO_READ) {
 650                 SDT_PROBE0(cbb, , read, file_start);
 651                 xuio.uio_rw = UIO_READ;
 652         } else {
 653                 SDT_PROBE0(cbb, , write, file_start);
 654                 xuio.uio_rw = UIO_WRITE;
 655         }
 656         xuio.uio_offset = beio->io_offset;
 657         xuio.uio_resid = beio->io_len;
 658         xuio.uio_segflg = UIO_SYSSPACE;
 659         xuio.uio_iov = beio->xiovecs;
 660         xuio.uio_iovcnt = beio->num_segs;
 661         xuio.uio_td = curthread;
 662
 663         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 664                 xiovec->iov_base = beio->sg_segs[i].addr;
 665                 xiovec->iov_len = beio->sg_segs[i].len;
 666         }
 667
 668         binuptime(&beio->ds_t0);
 669         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 670
 671         if (beio->bio_cmd == BIO_READ) {
 672                 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 673
 674                 /*
 675                  * UFS pays attention to IO_DIRECT for reads.  If the
 676                  * DIRECTIO option is configured into the kernel, it calls
 677                  * ffs_rawread().  But that only works for single-segment
 678                  * uios with user space addresses.  In our case, with a
 679                  * kernel uio, it still reads into the buffer cache, but it
 680                  * will just try to release the buffer from the cache later
 681                  * on in ffs_read().
 682                  *
 683                  * ZFS does not pay attention to IO_DIRECT for reads.
 684                  *
 685                  * UFS does not pay attention to IO_SYNC for reads.
 686                  *
 687                  * ZFS pays attention to IO_SYNC (which translates into the
 688                  * Solaris define FRSYNC for zfs_read()) for reads.  It
 689                  * attempts to sync the file before reading.
 690                  */
 691                 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
 692
 693                 VOP_UNLOCK(be_lun->vn);
 694                 SDT_PROBE0(cbb, , read, file_done);
 695                 if (error == 0 && xuio.uio_resid > 0) {
 696                         /*
 697                          * If we red less then requested (EOF), then
 698                          * we should clean the rest of the buffer.
 699                          */
 700                         s = beio->io_len - xuio.uio_resid;
 701                         for (i = 0; i < beio->num_segs; i++) {
 702                                 if (s >= beio->sg_segs[i].len) {
 703                                         s -= beio->sg_segs[i].len;
 704                                         continue;
 705                                 }
 706                                 bzero((uint8_t *)beio->sg_segs[i].addr + s,
 707                                     beio->sg_segs[i].len - s);
 708                                 s = 0;
 709                         }
 710                 }
 711         } else {
 712                 struct mount *mountpoint;
 713                 int lock_flags;
 714
 715                 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 716
 717                 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL)
 718                   && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 719                         lock_flags = LK_SHARED;
 720                 else
 721                         lock_flags = LK_EXCLUSIVE;
 722                 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 723
 724                 /*
 725                  * UFS pays attention to IO_DIRECT for writes.  The write
 726                  * is done asynchronously.  (Normally the write would just
 727                  * get put into cache.
 728                  *
 729                  * UFS pays attention to IO_SYNC for writes.  It will
 730                  * attempt to write the buffer out synchronously if that
 731                  * flag is set.
 732                  *
 733                  * ZFS does not pay attention to IO_DIRECT for writes.
 734                  *
 735                  * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
 736                  * for writes.  It will flush the transaction from the
 737                  * cache before returning.
 738                  */
 739                 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
 740                 VOP_UNLOCK(be_lun->vn);
 741
 742                 vn_finished_write(mountpoint);
 743                 SDT_PROBE0(cbb, , write, file_done);
 744         }
 745
 746         mtx_lock(&be_lun->io_lock);
 747         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 748             beio->ds_tag_type, beio->ds_trans_type,
 749             /*now*/ NULL, /*then*/&beio->ds_t0);
 750         mtx_unlock(&be_lun->io_lock);
 751
 752         /*
 753          * If we got an error, set the sense data to "MEDIUM ERROR" and
 754          * return the I/O to the user.
 755          */
 756         if (error != 0) {
 757                 if (error == ENOSPC || error == EDQUOT) {
 758                         ctl_set_space_alloc_fail(&io->scsiio);
 759                 } else if (error == EROFS || error == EACCES) {
 760                         ctl_set_hw_write_protected(&io->scsiio);
 761                 } else {
 762                         ctl_set_medium_error(&io->scsiio,
 763                             beio->bio_cmd == BIO_READ);
 764                 }
 765                 ctl_complete_beio(beio);
 766                 return;
 767         }
 768
 769         /*
 770          * If this is a write or a verify, we're all done.
 771          * If this is a read, we can now send the data to the user.
 772          */
 773         if ((beio->bio_cmd == BIO_WRITE) ||
 774             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 775                 ctl_set_success(&io->scsiio);
 776                 ctl_complete_beio(beio);
 777         } else {
 778                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 779                     beio->beio_cont == NULL) {
 780                         ctl_set_success(&io->scsiio);
 781                         ctl_serseq_done(io);
 782                 }
 783 #ifdef CTL_TIME_IO
 784                 getbinuptime(&io->io_hdr.dma_start_bt);
 785 #endif
 786                 ctl_datamove(io);
 787         }
 788 }
 789
 790 static void
 791 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
 792                         struct ctl_be_block_io *beio)
 793 {
 794         union ctl_io *io = beio->io;
 795         struct ctl_lba_len_flags *lbalen = ARGS(io);
 796         struct scsi_get_lba_status_data *data;
 797         off_t roff, off;
 798         int error, status;
 799
 800         DPRINTF("entered\n");
 801
 802         off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
 803         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 804         error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
 805             0, curthread->td_ucred, curthread);
 806         if (error == 0 && off > roff)
 807                 status = 0;     /* mapped up to off */
 808         else {
 809                 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
 810                     0, curthread->td_ucred, curthread);
 811                 if (error == 0 && off > roff)
 812                         status = 1;     /* deallocated up to off */
 813                 else {
 814                         status = 0;     /* unknown up to the end */
 815                         off = be_lun->size_bytes;
 816                 }
 817         }
 818         VOP_UNLOCK(be_lun->vn);
 819
 820         data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
 821         scsi_u64to8b(lbalen->lba, data->descr[0].addr);
 822         scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
 823             lbalen->lba), data->descr[0].length);
 824         data->descr[0].status = status;
 825
 826         ctl_complete_beio(beio);
 827 }
 828
 829 static uint64_t
 830 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
 831 {
 832         struct vattr            vattr;
 833         struct statfs           statfs;
 834         uint64_t                val;
 835         int                     error;
 836
 837         val = UINT64_MAX;
 838         if (be_lun->vn == NULL)
 839                 return (val);
 840         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 841         if (strcmp(attrname, "blocksused") == 0) {
 842                 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
 843                 if (error == 0)
 844                         val = vattr.va_bytes / be_lun->cbe_lun.blocksize;
 845         }
 846         if (strcmp(attrname, "blocksavail") == 0 &&
 847             !VN_IS_DOOMED(be_lun->vn)) {
 848                 error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
 849                 if (error == 0)
 850                         val = statfs.f_bavail * statfs.f_bsize /
 851                             be_lun->cbe_lun.blocksize;
 852         }
 853         VOP_UNLOCK(be_lun->vn);
 854         return (val);
 855 }
 856
 857 static void
 858 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
 859                            struct ctl_be_block_io *beio)
 860 {
 861         union ctl_io *io;
 862         struct cdevsw *csw;
 863         struct cdev *dev;
 864         struct uio xuio;
 865         struct iovec *xiovec;
 866         int error, flags, i, ref;
 867
 868         DPRINTF("entered\n");
 869
 870         io = beio->io;
 871         flags = 0;
 872         if (ARGS(io)->flags & CTL_LLF_DPO)
 873                 flags |= IO_DIRECT;
 874         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 875                 flags |= IO_SYNC;
 876
 877         bzero(&xuio, sizeof(xuio));
 878         if (beio->bio_cmd == BIO_READ) {
 879                 SDT_PROBE0(cbb, , read, file_start);
 880                 xuio.uio_rw = UIO_READ;
 881         } else {
 882                 SDT_PROBE0(cbb, , write, file_start);
 883                 xuio.uio_rw = UIO_WRITE;
 884         }
 885         xuio.uio_offset = beio->io_offset;
 886         xuio.uio_resid = beio->io_len;
 887         xuio.uio_segflg = UIO_SYSSPACE;
 888         xuio.uio_iov = beio->xiovecs;
 889         xuio.uio_iovcnt = beio->num_segs;
 890         xuio.uio_td = curthread;
 891
 892         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 893                 xiovec->iov_base = beio->sg_segs[i].addr;
 894                 xiovec->iov_len = beio->sg_segs[i].len;
 895         }
 896
 897         binuptime(&beio->ds_t0);
 898         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 899
 900         csw = devvn_refthread(be_lun->vn, &dev, &ref);
 901         if (csw) {
 902                 if (beio->bio_cmd == BIO_READ)
 903                         error = csw->d_read(dev, &xuio, flags);
 904                 else
 905                         error = csw->d_write(dev, &xuio, flags);
 906                 dev_relthread(dev, ref);
 907         } else
 908                 error = ENXIO;
 909
 910         if (beio->bio_cmd == BIO_READ)
 911                 SDT_PROBE0(cbb, , read, file_done);
 912         else
 913                 SDT_PROBE0(cbb, , write, file_done);
 914
 915         mtx_lock(&be_lun->io_lock);
 916         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 917             beio->ds_tag_type, beio->ds_trans_type,
 918             /*now*/ NULL, /*then*/&beio->ds_t0);
 919         mtx_unlock(&be_lun->io_lock);
 920
 921         /*
 922          * If we got an error, set the sense data to "MEDIUM ERROR" and
 923          * return the I/O to the user.
 924          */
 925         if (error != 0) {
 926                 if (error == ENOSPC || error == EDQUOT) {
 927                         ctl_set_space_alloc_fail(&io->scsiio);
 928                 } else if (error == EROFS || error == EACCES) {
 929                         ctl_set_hw_write_protected(&io->scsiio);
 930                 } else {
 931                         ctl_set_medium_error(&io->scsiio,
 932                             beio->bio_cmd == BIO_READ);
 933                 }
 934                 ctl_complete_beio(beio);
 935                 return;
 936         }
 937
 938         /*
 939          * If this is a write or a verify, we're all done.
 940          * If this is a read, we can now send the data to the user.
 941          */
 942         if ((beio->bio_cmd == BIO_WRITE) ||
 943             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 944                 ctl_set_success(&io->scsiio);
 945                 ctl_complete_beio(beio);
 946         } else {
 947                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 948                     beio->beio_cont == NULL) {
 949                         ctl_set_success(&io->scsiio);
 950                         ctl_serseq_done(io);
 951                 }
 952 #ifdef CTL_TIME_IO
 953                 getbinuptime(&io->io_hdr.dma_start_bt);
 954 #endif
 955                 ctl_datamove(io);
 956         }
 957 }
 958
 959 static void
 960 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
 961                         struct ctl_be_block_io *beio)
 962 {
 963         union ctl_io *io = beio->io;
 964         struct cdevsw *csw;
 965         struct cdev *dev;
 966         struct ctl_lba_len_flags *lbalen = ARGS(io);
 967         struct scsi_get_lba_status_data *data;
 968         off_t roff, off;
 969         int error, ref, status;
 970
 971         DPRINTF("entered\n");
 972
 973         csw = devvn_refthread(be_lun->vn, &dev, &ref);
 974         if (csw == NULL) {
 975                 status = 0;     /* unknown up to the end */
 976                 off = be_lun->size_bytes;
 977                 goto done;
 978         }
 979         off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
 980         error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD,
 981             curthread);
 982         if (error == 0 && off > roff)
 983                 status = 0;     /* mapped up to off */
 984         else {
 985                 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD,
 986                     curthread);
 987                 if (error == 0 && off > roff)
 988                         status = 1;     /* deallocated up to off */
 989                 else {
 990                         status = 0;     /* unknown up to the end */
 991                         off = be_lun->size_bytes;
 992                 }
 993         }
 994         dev_relthread(dev, ref);
 995
 996 done:
 997         data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
 998         scsi_u64to8b(lbalen->lba, data->descr[0].addr);
 999         scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
1000             lbalen->lba), data->descr[0].length);
1001         data->descr[0].status = status;
1002
1003         ctl_complete_beio(beio);
1004 }
1005
1006 static void
1007 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
1008                        struct ctl_be_block_io *beio)
1009 {
1010         struct bio *bio;
1011         struct cdevsw *csw;
1012         struct cdev *dev;
1013         int ref;
1014
1015         DPRINTF("entered\n");
1016
1017         /* This can't fail, it's a blocking allocation. */
1018         bio = g_alloc_bio();
1019
1020         bio->bio_cmd        = BIO_FLUSH;
1021         bio->bio_offset     = 0;
1022         bio->bio_data       = 0;
1023         bio->bio_done       = ctl_be_block_biodone;
1024         bio->bio_caller1    = beio;
1025         bio->bio_pblkno     = 0;
1026
1027         /*
1028          * We don't need to acquire the LUN lock here, because we are only
1029          * sending one bio, and so there is no other context to synchronize
1030          * with.
1031          */
1032         beio->num_bios_sent = 1;
1033         beio->send_complete = 1;
1034
1035         binuptime(&beio->ds_t0);
1036         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1037
1038         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1039         if (csw) {
1040                 bio->bio_dev = dev;
1041                 csw->d_strategy(bio);
1042                 dev_relthread(dev, ref);
1043         } else {
1044                 bio->bio_error = ENXIO;
1045                 ctl_be_block_biodone(bio);
1046         }
1047 }
1048
1049 static void
1050 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1051                        struct ctl_be_block_io *beio,
1052                        uint64_t off, uint64_t len, int last)
1053 {
1054         struct bio *bio;
1055         uint64_t maxlen;
1056         struct cdevsw *csw;
1057         struct cdev *dev;
1058         int ref;
1059
1060         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1061         maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize);
1062         while (len > 0) {
1063                 bio = g_alloc_bio();
1064                 bio->bio_cmd        = BIO_DELETE;
1065                 bio->bio_dev        = dev;
1066                 bio->bio_offset     = off;
1067                 bio->bio_length     = MIN(len, maxlen);
1068                 bio->bio_data       = 0;
1069                 bio->bio_done       = ctl_be_block_biodone;
1070                 bio->bio_caller1    = beio;
1071                 bio->bio_pblkno     = off / be_lun->cbe_lun.blocksize;
1072
1073                 off += bio->bio_length;
1074                 len -= bio->bio_length;
1075
1076                 mtx_lock(&be_lun->io_lock);
1077                 beio->num_bios_sent++;
1078                 if (last && len == 0)
1079                         beio->send_complete = 1;
1080                 mtx_unlock(&be_lun->io_lock);
1081
1082                 if (csw) {
1083                         csw->d_strategy(bio);
1084                 } else {
1085                         bio->bio_error = ENXIO;
1086                         ctl_be_block_biodone(bio);
1087                 }
1088         }
1089         if (csw)
1090                 dev_relthread(dev, ref);
1091 }
1092
1093 static void
1094 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1095                        struct ctl_be_block_io *beio)
1096 {
1097         union ctl_io *io;
1098         struct ctl_ptr_len_flags *ptrlen;
1099         struct scsi_unmap_desc *buf, *end;
1100         uint64_t len;
1101
1102         io = beio->io;
1103
1104         DPRINTF("entered\n");
1105
1106         binuptime(&beio->ds_t0);
1107         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1108
1109         if (beio->io_offset == -1) {
1110                 beio->io_len = 0;
1111                 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1112                 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1113                 end = buf + ptrlen->len / sizeof(*buf);
1114                 for (; buf < end; buf++) {
1115                         len = (uint64_t)scsi_4btoul(buf->length) *
1116                             be_lun->cbe_lun.blocksize;
1117                         beio->io_len += len;
1118                         ctl_be_block_unmap_dev_range(be_lun, beio,
1119                             scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize,
1120                             len, (end - buf < 2) ? TRUE : FALSE);
1121                 }
1122         } else
1123                 ctl_be_block_unmap_dev_range(be_lun, beio,
1124                     beio->io_offset, beio->io_len, TRUE);
1125 }
1126
1127 static void
1128 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1129                           struct ctl_be_block_io *beio)
1130 {
1131         TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1132         struct bio *bio;
1133         struct cdevsw *csw;
1134         struct cdev *dev;
1135         off_t cur_offset;
1136         int i, max_iosize, ref;
1137
1138         DPRINTF("entered\n");
1139         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1140
1141         /*
1142          * We have to limit our I/O size to the maximum supported by the
1143          * backend device.  Hopefully it is MAXPHYS.  If the driver doesn't
1144          * set it properly, use DFLTPHYS.
1145          */
1146         if (csw) {
1147                 max_iosize = dev->si_iosize_max;
1148                 if (max_iosize < PAGE_SIZE)
1149                         max_iosize = DFLTPHYS;
1150         } else
1151                 max_iosize = DFLTPHYS;
1152
1153         cur_offset = beio->io_offset;
1154         for (i = 0; i < beio->num_segs; i++) {
1155                 size_t cur_size;
1156                 uint8_t *cur_ptr;
1157
1158                 cur_size = beio->sg_segs[i].len;
1159                 cur_ptr = beio->sg_segs[i].addr;
1160
1161                 while (cur_size > 0) {
1162                         /* This can't fail, it's a blocking allocation. */
1163                         bio = g_alloc_bio();
1164
1165                         KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1166
1167                         bio->bio_cmd = beio->bio_cmd;
1168                         bio->bio_dev = dev;
1169                         bio->bio_caller1 = beio;
1170                         bio->bio_length = min(cur_size, max_iosize);
1171                         bio->bio_offset = cur_offset;
1172                         bio->bio_data = cur_ptr;
1173                         bio->bio_done = ctl_be_block_biodone;
1174                         bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize;
1175
1176                         cur_offset += bio->bio_length;
1177                         cur_ptr += bio->bio_length;
1178                         cur_size -= bio->bio_length;
1179
1180                         TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1181                         beio->num_bios_sent++;
1182                 }
1183         }
1184         beio->send_complete = 1;
1185         binuptime(&beio->ds_t0);
1186         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1187
1188         /*
1189          * Fire off all allocated requests!
1190          */
1191         while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1192                 TAILQ_REMOVE(&queue, bio, bio_queue);
1193                 if (csw)
1194                         csw->d_strategy(bio);
1195                 else {
1196                         bio->bio_error = ENXIO;
1197                         ctl_be_block_biodone(bio);
1198                 }
1199         }
1200         if (csw)
1201                 dev_relthread(dev, ref);
1202 }
1203
1204 static uint64_t
1205 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1206 {
1207         struct diocgattr_arg    arg;
1208         struct cdevsw *csw;
1209         struct cdev *dev;
1210         int error, ref;
1211
1212         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1213         if (csw == NULL)
1214                 return (UINT64_MAX);
1215         strlcpy(arg.name, attrname, sizeof(arg.name));
1216         arg.len = sizeof(arg.value.off);
1217         if (csw->d_ioctl) {
1218                 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
1219                     curthread);
1220         } else
1221                 error = ENODEV;
1222         dev_relthread(dev, ref);
1223         if (error != 0)
1224                 return (UINT64_MAX);
1225         return (arg.value.off);
1226 }
1227
1228 static void
1229 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun,
1230                             union ctl_io *io)
1231 {
1232         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1233         struct ctl_be_block_io *beio;
1234         struct ctl_lba_len_flags *lbalen;
1235
1236         DPRINTF("entered\n");
1237         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1238         lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1239
1240         beio->io_len = lbalen->len * cbe_lun->blocksize;
1241         beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1242         beio->io_arg = (lbalen->flags & SSC_IMMED) != 0;
1243         beio->bio_cmd = BIO_FLUSH;
1244         beio->ds_trans_type = DEVSTAT_NO_DATA;
1245         DPRINTF("SYNC\n");
1246         be_lun->lun_flush(be_lun, beio);
1247 }
1248
1249 static void
1250 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1251 {
1252         union ctl_io *io;
1253
1254         io = beio->io;
1255         ctl_free_beio(beio);
1256         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1257             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1258              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1259                 ctl_config_write_done(io);
1260                 return;
1261         }
1262
1263         ctl_be_block_config_write(io);
1264 }
1265
1266 static void
1267 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1268                             union ctl_io *io)
1269 {
1270         struct ctl_be_block_softc *softc = be_lun->softc;
1271         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1272         struct ctl_be_block_io *beio;
1273         struct ctl_lba_len_flags *lbalen;
1274         uint64_t len_left, lba;
1275         uint32_t pb, pbo, adj;
1276         int i, seglen;
1277         uint8_t *buf, *end;
1278
1279         DPRINTF("entered\n");
1280
1281         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1282         lbalen = ARGS(beio->io);
1283
1284         if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1285             (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1286                 ctl_free_beio(beio);
1287                 ctl_set_invalid_field(&io->scsiio,
1288                                       /*sks_valid*/ 1,
1289                                       /*command*/ 1,
1290                                       /*field*/ 1,
1291                                       /*bit_valid*/ 0,
1292                                       /*bit*/ 0);
1293                 ctl_config_write_done(io);
1294                 return;
1295         }
1296
1297         if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1298                 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1299                 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1300                 beio->bio_cmd = BIO_DELETE;
1301                 beio->ds_trans_type = DEVSTAT_FREE;
1302
1303                 be_lun->unmap(be_lun, beio);
1304                 return;
1305         }
1306
1307         beio->bio_cmd = BIO_WRITE;
1308         beio->ds_trans_type = DEVSTAT_WRITE;
1309
1310         DPRINTF("WRITE SAME at LBA %jx len %u\n",
1311                (uintmax_t)lbalen->lba, lbalen->len);
1312
1313         pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1314         if (be_lun->cbe_lun.pblockoff > 0)
1315                 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1316         else
1317                 pbo = 0;
1318         len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1319         for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1320                 /*
1321                  * Setup the S/G entry for this chunk.
1322                  */
1323                 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1324                 if (pb > cbe_lun->blocksize) {
1325                         adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1326                             seglen - pbo) % pb;
1327                         if (seglen > adj)
1328                                 seglen -= adj;
1329                         else
1330                                 seglen -= seglen % cbe_lun->blocksize;
1331                 } else
1332                         seglen -= seglen % cbe_lun->blocksize;
1333                 beio->sg_segs[i].len = seglen;
1334                 beio->sg_segs[i].addr = uma_zalloc(softc->buf_zone, M_WAITOK);
1335
1336                 DPRINTF("segment %d addr %p len %zd\n", i,
1337                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1338
1339                 beio->num_segs++;
1340                 len_left -= seglen;
1341
1342                 buf = beio->sg_segs[i].addr;
1343                 end = buf + seglen;
1344                 for (; buf < end; buf += cbe_lun->blocksize) {
1345                         if (lbalen->flags & SWS_NDOB) {
1346                                 memset(buf, 0, cbe_lun->blocksize);
1347                         } else {
1348                                 memcpy(buf, io->scsiio.kern_data_ptr,
1349                                     cbe_lun->blocksize);
1350                         }
1351                         if (lbalen->flags & SWS_LBDATA)
1352                                 scsi_ulto4b(lbalen->lba + lba, buf);
1353                         lba++;
1354                 }
1355         }
1356
1357         beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1358         beio->io_len = lba * cbe_lun->blocksize;
1359
1360         /* We can not do all in one run. Correct and schedule rerun. */
1361         if (len_left > 0) {
1362                 lbalen->lba += lba;
1363                 lbalen->len -= lba;
1364                 beio->beio_cont = ctl_be_block_cw_done_ws;
1365         }
1366
1367         be_lun->dispatch(be_lun, beio);
1368 }
1369
1370 static void
1371 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1372                             union ctl_io *io)
1373 {
1374         struct ctl_be_block_io *beio;
1375         struct ctl_ptr_len_flags *ptrlen;
1376
1377         DPRINTF("entered\n");
1378
1379         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1380         ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1381
1382         if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1383                 ctl_free_beio(beio);
1384                 ctl_set_invalid_field(&io->scsiio,
1385                                       /*sks_valid*/ 0,
1386                                       /*command*/ 1,
1387                                       /*field*/ 0,
1388                                       /*bit_valid*/ 0,
1389                                       /*bit*/ 0);
1390                 ctl_config_write_done(io);
1391                 return;
1392         }
1393
1394         beio->io_len = 0;
1395         beio->io_offset = -1;
1396         beio->bio_cmd = BIO_DELETE;
1397         beio->ds_trans_type = DEVSTAT_FREE;
1398         DPRINTF("UNMAP\n");
1399         be_lun->unmap(be_lun, beio);
1400 }
1401
1402 static void
1403 ctl_be_block_cr_done(struct ctl_be_block_io *beio)
1404 {
1405         union ctl_io *io;
1406
1407         io = beio->io;
1408         ctl_free_beio(beio);
1409         ctl_config_read_done(io);
1410 }
1411
1412 static void
1413 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1414                          union ctl_io *io)
1415 {
1416         struct ctl_be_block_io *beio;
1417         struct ctl_be_block_softc *softc;
1418
1419         DPRINTF("entered\n");
1420
1421         softc = be_lun->softc;
1422         beio = ctl_alloc_beio(softc);
1423         beio->io = io;
1424         beio->lun = be_lun;
1425         beio->beio_cont = ctl_be_block_cr_done;
1426         PRIV(io)->ptr = (void *)beio;
1427
1428         switch (io->scsiio.cdb[0]) {
1429         case SERVICE_ACTION_IN:         /* GET LBA STATUS */
1430                 beio->bio_cmd = -1;
1431                 beio->ds_trans_type = DEVSTAT_NO_DATA;
1432                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1433                 beio->io_len = 0;
1434                 if (be_lun->get_lba_status)
1435                         be_lun->get_lba_status(be_lun, beio);
1436                 else
1437                         ctl_be_block_cr_done(beio);
1438                 break;
1439         default:
1440                 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1441                 break;
1442         }
1443 }
1444
1445 static void
1446 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1447 {
1448         union ctl_io *io;
1449
1450         io = beio->io;
1451         ctl_free_beio(beio);
1452         ctl_config_write_done(io);
1453 }
1454
1455 static void
1456 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1457                          union ctl_io *io)
1458 {
1459         struct ctl_be_block_io *beio;
1460         struct ctl_be_block_softc *softc;
1461
1462         DPRINTF("entered\n");
1463
1464         softc = be_lun->softc;
1465         beio = ctl_alloc_beio(softc);
1466         beio->io = io;
1467         beio->lun = be_lun;
1468         beio->beio_cont = ctl_be_block_cw_done;
1469         switch (io->scsiio.tag_type) {
1470         case CTL_TAG_ORDERED:
1471                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1472                 break;
1473         case CTL_TAG_HEAD_OF_QUEUE:
1474                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1475                 break;
1476         case CTL_TAG_UNTAGGED:
1477         case CTL_TAG_SIMPLE:
1478         case CTL_TAG_ACA:
1479         default:
1480                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1481                 break;
1482         }
1483         PRIV(io)->ptr = (void *)beio;
1484
1485         switch (io->scsiio.cdb[0]) {
1486         case SYNCHRONIZE_CACHE:
1487         case SYNCHRONIZE_CACHE_16:
1488                 ctl_be_block_cw_dispatch_sync(be_lun, io);
1489                 break;
1490         case WRITE_SAME_10:
1491         case WRITE_SAME_16:
1492                 ctl_be_block_cw_dispatch_ws(be_lun, io);
1493                 break;
1494         case UNMAP:
1495                 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1496                 break;
1497         default:
1498                 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1499                 break;
1500         }
1501 }
1502
1503 SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t");
1504 SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t");
1505 SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t");
1506 SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t");
1507
1508 static void
1509 ctl_be_block_next(struct ctl_be_block_io *beio)
1510 {
1511         struct ctl_be_block_lun *be_lun;
1512         union ctl_io *io;
1513
1514         io = beio->io;
1515         be_lun = beio->lun;
1516         ctl_free_beio(beio);
1517         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1518             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1519              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1520                 ctl_data_submit_done(io);
1521                 return;
1522         }
1523
1524         io->io_hdr.status &= ~CTL_STATUS_MASK;
1525         io->io_hdr.status |= CTL_STATUS_NONE;
1526
1527         mtx_lock(&be_lun->queue_lock);
1528         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1529         mtx_unlock(&be_lun->queue_lock);
1530         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1531 }
1532
1533 static void
1534 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1535                            union ctl_io *io)
1536 {
1537         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1538         struct ctl_be_block_io *beio;
1539         struct ctl_be_block_softc *softc;
1540         struct ctl_lba_len_flags *lbalen;
1541         struct ctl_ptr_len_flags *bptrlen;
1542         uint64_t len_left, lbas;
1543         int i;
1544
1545         softc = be_lun->softc;
1546
1547         DPRINTF("entered\n");
1548
1549         lbalen = ARGS(io);
1550         if (lbalen->flags & CTL_LLF_WRITE) {
1551                 SDT_PROBE0(cbb, , write, start);
1552         } else {
1553                 SDT_PROBE0(cbb, , read, start);
1554         }
1555
1556         beio = ctl_alloc_beio(softc);
1557         beio->io = io;
1558         beio->lun = be_lun;
1559         bptrlen = PRIV(io);
1560         bptrlen->ptr = (void *)beio;
1561
1562         switch (io->scsiio.tag_type) {
1563         case CTL_TAG_ORDERED:
1564                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1565                 break;
1566         case CTL_TAG_HEAD_OF_QUEUE:
1567                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1568                 break;
1569         case CTL_TAG_UNTAGGED:
1570         case CTL_TAG_SIMPLE:
1571         case CTL_TAG_ACA:
1572         default:
1573                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1574                 break;
1575         }
1576
1577         if (lbalen->flags & CTL_LLF_WRITE) {
1578                 beio->bio_cmd = BIO_WRITE;
1579                 beio->ds_trans_type = DEVSTAT_WRITE;
1580         } else {
1581                 beio->bio_cmd = BIO_READ;
1582                 beio->ds_trans_type = DEVSTAT_READ;
1583         }
1584
1585         DPRINTF("%s at LBA %jx len %u @%ju\n",
1586                (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1587                (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1588         if (lbalen->flags & CTL_LLF_COMPARE) {
1589                 beio->two_sglists = 1;
1590                 lbas = CTLBLK_HALF_IO_SIZE;
1591         } else {
1592                 lbas = CTLBLK_MAX_IO_SIZE;
1593         }
1594         lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize);
1595         beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize;
1596         beio->io_len = lbas * cbe_lun->blocksize;
1597         bptrlen->len += lbas;
1598
1599         for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1600                 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1601                     i, CTLBLK_MAX_SEGS));
1602
1603                 /*
1604                  * Setup the S/G entry for this chunk.
1605                  */
1606                 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1607                 beio->sg_segs[i].addr = uma_zalloc(softc->buf_zone, M_WAITOK);
1608
1609                 DPRINTF("segment %d addr %p len %zd\n", i,
1610                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1611
1612                 /* Set up second segment for compare operation. */
1613                 if (beio->two_sglists) {
1614                         beio->sg_segs[i + CTLBLK_HALF_SEGS].len =
1615                             beio->sg_segs[i].len;
1616                         beio->sg_segs[i + CTLBLK_HALF_SEGS].addr =
1617                             uma_zalloc(softc->buf_zone, M_WAITOK);
1618                 }
1619
1620                 beio->num_segs++;
1621                 len_left -= beio->sg_segs[i].len;
1622         }
1623         if (bptrlen->len < lbalen->len)
1624                 beio->beio_cont = ctl_be_block_next;
1625         io->scsiio.be_move_done = ctl_be_block_move_done;
1626         /* For compare we have separate S/G lists for read and datamove. */
1627         if (beio->two_sglists)
1628                 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1629         else
1630                 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1631         io->scsiio.kern_data_len = beio->io_len;
1632         io->scsiio.kern_sg_entries = beio->num_segs;
1633         io->scsiio.kern_data_ref = ctl_refcnt_beio;
1634         io->scsiio.kern_data_arg = beio;
1635         io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
1636
1637         /*
1638          * For the read case, we need to read the data into our buffers and
1639          * then we can send it back to the user.  For the write case, we
1640          * need to get the data from the user first.
1641          */
1642         if (beio->bio_cmd == BIO_READ) {
1643                 SDT_PROBE0(cbb, , read, alloc_done);
1644                 be_lun->dispatch(be_lun, beio);
1645         } else {
1646                 SDT_PROBE0(cbb, , write, alloc_done);
1647 #ifdef CTL_TIME_IO
1648                 getbinuptime(&io->io_hdr.dma_start_bt);
1649 #endif
1650                 ctl_datamove(io);
1651         }
1652 }
1653
1654 static void
1655 ctl_be_block_worker(void *context, int pending)
1656 {
1657         struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context;
1658         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1659         union ctl_io *io;
1660         struct ctl_be_block_io *beio;
1661
1662         DPRINTF("entered\n");
1663         /*
1664          * Fetch and process I/Os from all queues.  If we detect LUN
1665          * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race,
1666          * so make response maximally opaque to not confuse initiator.
1667          */
1668         for (;;) {
1669                 mtx_lock(&be_lun->queue_lock);
1670                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1671                 if (io != NULL) {
1672                         DPRINTF("datamove queue\n");
1673                         STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1674                                       ctl_io_hdr, links);
1675                         mtx_unlock(&be_lun->queue_lock);
1676                         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1677                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1678                                 ctl_set_busy(&io->scsiio);
1679                                 ctl_complete_beio(beio);
1680                                 return;
1681                         }
1682                         be_lun->dispatch(be_lun, beio);
1683                         continue;
1684                 }
1685                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1686                 if (io != NULL) {
1687                         DPRINTF("config write queue\n");
1688                         STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1689                                       ctl_io_hdr, links);
1690                         mtx_unlock(&be_lun->queue_lock);
1691                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1692                                 ctl_set_busy(&io->scsiio);
1693                                 ctl_config_write_done(io);
1694                                 return;
1695                         }
1696                         ctl_be_block_cw_dispatch(be_lun, io);
1697                         continue;
1698                 }
1699                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
1700                 if (io != NULL) {
1701                         DPRINTF("config read queue\n");
1702                         STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr,
1703                                       ctl_io_hdr, links);
1704                         mtx_unlock(&be_lun->queue_lock);
1705                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1706                                 ctl_set_busy(&io->scsiio);
1707                                 ctl_config_read_done(io);
1708                                 return;
1709                         }
1710                         ctl_be_block_cr_dispatch(be_lun, io);
1711                         continue;
1712                 }
1713                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1714                 if (io != NULL) {
1715                         DPRINTF("input queue\n");
1716                         STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1717                                       ctl_io_hdr, links);
1718                         mtx_unlock(&be_lun->queue_lock);
1719                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1720                                 ctl_set_busy(&io->scsiio);
1721                                 ctl_data_submit_done(io);
1722                                 return;
1723                         }
1724                         ctl_be_block_dispatch(be_lun, io);
1725                         continue;
1726                 }
1727
1728                 /*
1729                  * If we get here, there is no work left in the queues, so
1730                  * just break out and let the task queue go to sleep.
1731                  */
1732                 mtx_unlock(&be_lun->queue_lock);
1733                 break;
1734         }
1735 }
1736
1737 /*
1738  * Entry point from CTL to the backend for I/O.  We queue everything to a
1739  * work thread, so this just puts the I/O on a queue and wakes up the
1740  * thread.
1741  */
1742 static int
1743 ctl_be_block_submit(union ctl_io *io)
1744 {
1745         struct ctl_be_block_lun *be_lun;
1746
1747         DPRINTF("entered\n");
1748
1749         be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
1750
1751         /*
1752          * Make sure we only get SCSI I/O.
1753          */
1754         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1755                 "%#x) encountered", io->io_hdr.io_type));
1756
1757         PRIV(io)->len = 0;
1758
1759         mtx_lock(&be_lun->queue_lock);
1760         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1761         mtx_unlock(&be_lun->queue_lock);
1762         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1763
1764         return (CTL_RETVAL_COMPLETE);
1765 }
1766
1767 static int
1768 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1769                         int flag, struct thread *td)
1770 {
1771         struct ctl_be_block_softc *softc = &backend_block_softc;
1772         int error;
1773
1774         error = 0;
1775         switch (cmd) {
1776         case CTL_LUN_REQ: {
1777                 struct ctl_lun_req *lun_req;
1778
1779                 lun_req = (struct ctl_lun_req *)addr;
1780
1781                 switch (lun_req->reqtype) {
1782                 case CTL_LUNREQ_CREATE:
1783                         error = ctl_be_block_create(softc, lun_req);
1784                         break;
1785                 case CTL_LUNREQ_RM:
1786                         error = ctl_be_block_rm(softc, lun_req);
1787                         break;
1788                 case CTL_LUNREQ_MODIFY:
1789                         error = ctl_be_block_modify(softc, lun_req);
1790                         break;
1791                 default:
1792                         lun_req->status = CTL_LUN_ERROR;
1793                         snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1794                                  "invalid LUN request type %d",
1795                                  lun_req->reqtype);
1796                         break;
1797                 }
1798                 break;
1799         }
1800         default:
1801                 error = ENOTTY;
1802                 break;
1803         }
1804
1805         return (error);
1806 }
1807
1808 static int
1809 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1810 {
1811         struct ctl_be_lun *cbe_lun;
1812         struct ctl_be_block_filedata *file_data;
1813         struct ctl_lun_create_params *params;
1814         const char                   *value;
1815         struct vattr                  vattr;
1816         off_t                         ps, pss, po, pos, us, uss, uo, uos;
1817         int                           error;
1818
1819         cbe_lun = &be_lun->cbe_lun;
1820         file_data = &be_lun->backend.file;
1821         params = &be_lun->params;
1822
1823         be_lun->dev_type = CTL_BE_BLOCK_FILE;
1824         be_lun->dispatch = ctl_be_block_dispatch_file;
1825         be_lun->lun_flush = ctl_be_block_flush_file;
1826         be_lun->get_lba_status = ctl_be_block_gls_file;
1827         be_lun->getattr = ctl_be_block_getattr_file;
1828         be_lun->unmap = NULL;
1829         cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
1830
1831         error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1832         if (error != 0) {
1833                 snprintf(req->error_str, sizeof(req->error_str),
1834                          "error calling VOP_GETATTR() for file %s",
1835                          be_lun->dev_path);
1836                 return (error);
1837         }
1838
1839         file_data->cred = crhold(curthread->td_ucred);
1840         if (params->lun_size_bytes != 0)
1841                 be_lun->size_bytes = params->lun_size_bytes;
1842         else
1843                 be_lun->size_bytes = vattr.va_size;
1844
1845         /*
1846          * For files we can use any logical block size.  Prefer 512 bytes
1847          * for compatibility reasons.  If file's vattr.va_blocksize
1848          * (preferred I/O block size) is bigger and multiple to chosen
1849          * logical block size -- report it as physical block size.
1850          */
1851         if (params->blocksize_bytes != 0)
1852                 cbe_lun->blocksize = params->blocksize_bytes;
1853         else if (cbe_lun->lun_type == T_CDROM)
1854                 cbe_lun->blocksize = 2048;
1855         else
1856                 cbe_lun->blocksize = 512;
1857         be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
1858         cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
1859             0 : (be_lun->size_blocks - 1);
1860
1861         us = ps = vattr.va_blocksize;
1862         uo = po = 0;
1863
1864         value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
1865         if (value != NULL)
1866                 ctl_expand_number(value, &ps);
1867         value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
1868         if (value != NULL)
1869                 ctl_expand_number(value, &po);
1870         pss = ps / cbe_lun->blocksize;
1871         pos = po / cbe_lun->blocksize;
1872         if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
1873             ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
1874                 cbe_lun->pblockexp = fls(pss) - 1;
1875                 cbe_lun->pblockoff = (pss - pos) % pss;
1876         }
1877
1878         value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
1879         if (value != NULL)
1880                 ctl_expand_number(value, &us);
1881         value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
1882         if (value != NULL)
1883                 ctl_expand_number(value, &uo);
1884         uss = us / cbe_lun->blocksize;
1885         uos = uo / cbe_lun->blocksize;
1886         if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
1887             ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
1888                 cbe_lun->ublockexp = fls(uss) - 1;
1889                 cbe_lun->ublockoff = (uss - uos) % uss;
1890         }
1891
1892         /*
1893          * Sanity check.  The media size has to be at least one
1894          * sector long.
1895          */
1896         if (be_lun->size_bytes < cbe_lun->blocksize) {
1897                 error = EINVAL;
1898                 snprintf(req->error_str, sizeof(req->error_str),
1899                          "file %s size %ju < block size %u", be_lun->dev_path,
1900                          (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize);
1901         }
1902
1903         cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize;
1904         return (error);
1905 }
1906
1907 static int
1908 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1909 {
1910         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1911         struct ctl_lun_create_params *params;
1912         struct cdevsw                *csw;
1913         struct cdev                  *dev;
1914         const char                   *value;
1915         int                           error, atomic, maxio, ref, unmap, tmp;
1916         off_t                         ps, pss, po, pos, us, uss, uo, uos, otmp;
1917
1918         params = &be_lun->params;
1919
1920         be_lun->dev_type = CTL_BE_BLOCK_DEV;
1921         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1922         if (csw == NULL)
1923                 return (ENXIO);
1924         if (strcmp(csw->d_name, "zvol") == 0) {
1925                 be_lun->dispatch = ctl_be_block_dispatch_zvol;
1926                 be_lun->get_lba_status = ctl_be_block_gls_zvol;
1927                 atomic = maxio = CTLBLK_MAX_IO_SIZE;
1928         } else {
1929                 be_lun->dispatch = ctl_be_block_dispatch_dev;
1930                 be_lun->get_lba_status = NULL;
1931                 atomic = 0;
1932                 maxio = dev->si_iosize_max;
1933                 if (maxio <= 0)
1934                         maxio = DFLTPHYS;
1935                 if (maxio > CTLBLK_MAX_IO_SIZE)
1936                         maxio = CTLBLK_MAX_IO_SIZE;
1937         }
1938         be_lun->lun_flush = ctl_be_block_flush_dev;
1939         be_lun->getattr = ctl_be_block_getattr_dev;
1940         be_lun->unmap = ctl_be_block_unmap_dev;
1941
1942         if (!csw->d_ioctl) {
1943                 dev_relthread(dev, ref);
1944                 snprintf(req->error_str, sizeof(req->error_str),
1945                          "no d_ioctl for device %s!", be_lun->dev_path);
1946                 return (ENODEV);
1947         }
1948
1949         error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD,
1950                                curthread);
1951         if (error) {
1952                 dev_relthread(dev, ref);
1953                 snprintf(req->error_str, sizeof(req->error_str),
1954                          "error %d returned for DIOCGSECTORSIZE ioctl "
1955                          "on %s!", error, be_lun->dev_path);
1956                 return (error);
1957         }
1958
1959         /*
1960          * If the user has asked for a blocksize that is greater than the
1961          * backing device's blocksize, we can do it only if the blocksize
1962          * the user is asking for is an even multiple of the underlying
1963          * device's blocksize.
1964          */
1965         if ((params->blocksize_bytes != 0) &&
1966             (params->blocksize_bytes >= tmp)) {
1967                 if (params->blocksize_bytes % tmp == 0) {
1968                         cbe_lun->blocksize = params->blocksize_bytes;
1969                 } else {
1970                         dev_relthread(dev, ref);
1971                         snprintf(req->error_str, sizeof(req->error_str),
1972                                  "requested blocksize %u is not an even "
1973                                  "multiple of backing device blocksize %u",
1974                                  params->blocksize_bytes, tmp);
1975                         return (EINVAL);
1976                 }
1977         } else if (params->blocksize_bytes != 0) {
1978                 dev_relthread(dev, ref);
1979                 snprintf(req->error_str, sizeof(req->error_str),
1980                          "requested blocksize %u < backing device "
1981                          "blocksize %u", params->blocksize_bytes, tmp);
1982                 return (EINVAL);
1983         } else if (cbe_lun->lun_type == T_CDROM)
1984                 cbe_lun->blocksize = MAX(tmp, 2048);
1985         else
1986                 cbe_lun->blocksize = tmp;
1987
1988         error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD,
1989                              curthread);
1990         if (error) {
1991                 dev_relthread(dev, ref);
1992                 snprintf(req->error_str, sizeof(req->error_str),
1993                          "error %d returned for DIOCGMEDIASIZE "
1994                          " ioctl on %s!", error,
1995                          be_lun->dev_path);
1996                 return (error);
1997         }
1998
1999         if (params->lun_size_bytes != 0) {
2000                 if (params->lun_size_bytes > otmp) {
2001                         dev_relthread(dev, ref);
2002                         snprintf(req->error_str, sizeof(req->error_str),
2003                                  "requested LUN size %ju > backing device "
2004                                  "size %ju",
2005                                  (uintmax_t)params->lun_size_bytes,
2006                                  (uintmax_t)otmp);
2007                         return (EINVAL);
2008                 }
2009
2010                 be_lun->size_bytes = params->lun_size_bytes;
2011         } else
2012                 be_lun->size_bytes = otmp;
2013         be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2014         cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2015             0 : (be_lun->size_blocks - 1);
2016
2017         error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD,
2018             curthread);
2019         if (error)
2020                 ps = po = 0;
2021         else {
2022                 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po,
2023                     FREAD, curthread);
2024                 if (error)
2025                         po = 0;
2026         }
2027         us = ps;
2028         uo = po;
2029
2030         value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
2031         if (value != NULL)
2032                 ctl_expand_number(value, &ps);
2033         value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
2034         if (value != NULL)
2035                 ctl_expand_number(value, &po);
2036         pss = ps / cbe_lun->blocksize;
2037         pos = po / cbe_lun->blocksize;
2038         if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
2039             ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
2040                 cbe_lun->pblockexp = fls(pss) - 1;
2041                 cbe_lun->pblockoff = (pss - pos) % pss;
2042         }
2043
2044         value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
2045         if (value != NULL)
2046                 ctl_expand_number(value, &us);
2047         value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
2048         if (value != NULL)
2049                 ctl_expand_number(value, &uo);
2050         uss = us / cbe_lun->blocksize;
2051         uos = uo / cbe_lun->blocksize;
2052         if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
2053             ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
2054                 cbe_lun->ublockexp = fls(uss) - 1;
2055                 cbe_lun->ublockoff = (uss - uos) % uss;
2056         }
2057
2058         cbe_lun->atomicblock = atomic / cbe_lun->blocksize;
2059         cbe_lun->opttxferlen = maxio / cbe_lun->blocksize;
2060
2061         if (be_lun->dispatch == ctl_be_block_dispatch_zvol) {
2062                 unmap = 1;
2063         } else {
2064                 struct diocgattr_arg    arg;
2065
2066                 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
2067                 arg.len = sizeof(arg.value.i);
2068                 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
2069                     curthread);
2070                 unmap = (error == 0) ? arg.value.i : 0;
2071         }
2072         value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
2073         if (value != NULL)
2074                 unmap = (strcmp(value, "on") == 0);
2075         if (unmap)
2076                 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
2077         else
2078                 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
2079
2080         dev_relthread(dev, ref);
2081         return (0);
2082 }
2083
2084 static int
2085 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
2086 {
2087         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2088         int flags;
2089
2090         if (be_lun->vn) {
2091                 flags = FREAD;
2092                 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0)
2093                         flags |= FWRITE;
2094                 (void)vn_close(be_lun->vn, flags, NOCRED, curthread);
2095                 be_lun->vn = NULL;
2096
2097                 switch (be_lun->dev_type) {
2098                 case CTL_BE_BLOCK_DEV:
2099                         break;
2100                 case CTL_BE_BLOCK_FILE:
2101                         if (be_lun->backend.file.cred != NULL) {
2102                                 crfree(be_lun->backend.file.cred);
2103                                 be_lun->backend.file.cred = NULL;
2104                         }
2105                         break;
2106                 case CTL_BE_BLOCK_NONE:
2107                         break;
2108                 default:
2109                         panic("Unexpected backend type %d", be_lun->dev_type);
2110                         break;
2111                 }
2112                 be_lun->dev_type = CTL_BE_BLOCK_NONE;
2113         }
2114         return (0);
2115 }
2116
2117 static int
2118 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2119 {
2120         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2121         struct nameidata nd;
2122         const char      *value;
2123         int              error, flags;
2124
2125         error = 0;
2126         if (rootvnode == NULL) {
2127                 snprintf(req->error_str, sizeof(req->error_str),
2128                          "Root filesystem is not mounted");
2129                 return (1);
2130         }
2131         pwd_ensure_dirs();
2132
2133         value = dnvlist_get_string(cbe_lun->options, "file", NULL);
2134         if (value == NULL) {
2135                 snprintf(req->error_str, sizeof(req->error_str),
2136                          "no file argument specified");
2137                 return (1);
2138         }
2139         free(be_lun->dev_path, M_CTLBLK);
2140         be_lun->dev_path = strdup(value, M_CTLBLK);
2141
2142         flags = FREAD;
2143         value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
2144         if (value != NULL) {
2145                 if (strcmp(value, "on") != 0)
2146                         flags |= FWRITE;
2147         } else if (cbe_lun->lun_type == T_DIRECT)
2148                 flags |= FWRITE;
2149
2150 again:
2151         NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread);
2152         error = vn_open(&nd, &flags, 0, NULL);
2153         if ((error == EROFS || error == EACCES) && (flags & FWRITE)) {
2154                 flags &= ~FWRITE;
2155                 goto again;
2156         }
2157         if (error) {
2158                 /*
2159                  * This is the only reasonable guess we can make as far as
2160                  * path if the user doesn't give us a fully qualified path.
2161                  * If they want to specify a file, they need to specify the
2162                  * full path.
2163                  */
2164                 if (be_lun->dev_path[0] != '/') {
2165                         char *dev_name;
2166
2167                         asprintf(&dev_name, M_CTLBLK, "/dev/%s",
2168                                 be_lun->dev_path);
2169                         free(be_lun->dev_path, M_CTLBLK);
2170                         be_lun->dev_path = dev_name;
2171                         goto again;
2172                 }
2173                 snprintf(req->error_str, sizeof(req->error_str),
2174                     "error opening %s: %d", be_lun->dev_path, error);
2175                 return (error);
2176         }
2177         if (flags & FWRITE)
2178                 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY;
2179         else
2180                 cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
2181
2182         NDFREE(&nd, NDF_ONLY_PNBUF);
2183         be_lun->vn = nd.ni_vp;
2184
2185         /* We only support disks and files. */
2186         if (vn_isdisk_error(be_lun->vn, &error)) {
2187                 error = ctl_be_block_open_dev(be_lun, req);
2188         } else if (be_lun->vn->v_type == VREG) {
2189                 error = ctl_be_block_open_file(be_lun, req);
2190         } else {
2191                 error = EINVAL;
2192                 snprintf(req->error_str, sizeof(req->error_str),
2193                          "%s is not a disk or plain file", be_lun->dev_path);
2194         }
2195         VOP_UNLOCK(be_lun->vn);
2196
2197         if (error != 0)
2198                 ctl_be_block_close(be_lun);
2199         cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2200         if (be_lun->dispatch != ctl_be_block_dispatch_dev)
2201                 cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2202         value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
2203         if (value != NULL && strcmp(value, "on") == 0)
2204                 cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
2205         else if (value != NULL && strcmp(value, "read") == 0)
2206                 cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2207         else if (value != NULL && strcmp(value, "off") == 0)
2208                 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2209         return (0);
2210 }
2211
2212 static int
2213 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2214 {
2215         struct ctl_be_lun *cbe_lun;
2216         struct ctl_be_block_lun *be_lun;
2217         struct ctl_lun_create_params *params;
2218         char num_thread_str[16];
2219         char tmpstr[32];
2220         const char *value;
2221         int retval, num_threads;
2222         int tmp_num_threads;
2223
2224         params = &req->reqdata.create;
2225         retval = 0;
2226         req->status = CTL_LUN_OK;
2227
2228         be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
2229         cbe_lun = &be_lun->cbe_lun;
2230         be_lun->params = req->reqdata.create;
2231         be_lun->softc = softc;
2232         STAILQ_INIT(&be_lun->input_queue);
2233         STAILQ_INIT(&be_lun->config_read_queue);
2234         STAILQ_INIT(&be_lun->config_write_queue);
2235         STAILQ_INIT(&be_lun->datamove_queue);
2236         mtx_init(&be_lun->io_lock, "ctlblock io", NULL, MTX_DEF);
2237         mtx_init(&be_lun->queue_lock, "ctlblock queue", NULL, MTX_DEF);
2238         cbe_lun->options = nvlist_clone(req->args_nvl);
2239
2240         if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
2241                 cbe_lun->lun_type = params->device_type;
2242         else
2243                 cbe_lun->lun_type = T_DIRECT;
2244         be_lun->flags = 0;
2245         cbe_lun->flags = 0;
2246         value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2247         if (value != NULL) {
2248                 if (strcmp(value, "primary") == 0)
2249                         cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2250         } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2251                 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2252
2253         if (cbe_lun->lun_type == T_DIRECT ||
2254             cbe_lun->lun_type == T_CDROM) {
2255                 be_lun->size_bytes = params->lun_size_bytes;
2256                 if (params->blocksize_bytes != 0)
2257                         cbe_lun->blocksize = params->blocksize_bytes;
2258                 else if (cbe_lun->lun_type == T_CDROM)
2259                         cbe_lun->blocksize = 2048;
2260                 else
2261                         cbe_lun->blocksize = 512;
2262                 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2263                 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2264                     0 : (be_lun->size_blocks - 1);
2265
2266                 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2267                     control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2268                         retval = ctl_be_block_open(be_lun, req);
2269                         if (retval != 0) {
2270                                 retval = 0;
2271                                 req->status = CTL_LUN_WARNING;
2272                         }
2273                 }
2274                 num_threads = cbb_num_threads;
2275         } else {
2276                 num_threads = 1;
2277         }
2278
2279         value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL);
2280         if (value != NULL) {
2281                 tmp_num_threads = strtol(value, NULL, 0);
2282
2283                 /*
2284                  * We don't let the user specify less than one
2285                  * thread, but hope he's clueful enough not to
2286                  * specify 1000 threads.
2287                  */
2288                 if (tmp_num_threads < 1) {
2289                         snprintf(req->error_str, sizeof(req->error_str),
2290                                  "invalid number of threads %s",
2291                                  num_thread_str);
2292                         goto bailout_error;
2293                 }
2294                 num_threads = tmp_num_threads;
2295         }
2296
2297         if (be_lun->vn == NULL)
2298                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2299         /* Tell the user the blocksize we ended up using */
2300         params->lun_size_bytes = be_lun->size_bytes;
2301         params->blocksize_bytes = cbe_lun->blocksize;
2302         if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2303                 cbe_lun->req_lun_id = params->req_lun_id;
2304                 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
2305         } else
2306                 cbe_lun->req_lun_id = 0;
2307
2308         cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown;
2309         cbe_lun->be = &ctl_be_block_driver;
2310
2311         if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2312                 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
2313                          softc->num_luns);
2314                 strncpy((char *)cbe_lun->serial_num, tmpstr,
2315                         MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
2316
2317                 /* Tell the user what we used for a serial number */
2318                 strncpy((char *)params->serial_num, tmpstr,
2319                         MIN(sizeof(params->serial_num), sizeof(tmpstr)));
2320         } else {
2321                 strncpy((char *)cbe_lun->serial_num, params->serial_num,
2322                         MIN(sizeof(cbe_lun->serial_num),
2323                         sizeof(params->serial_num)));
2324         }
2325         if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2326                 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
2327                 strncpy((char *)cbe_lun->device_id, tmpstr,
2328                         MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
2329
2330                 /* Tell the user what we used for a device ID */
2331                 strncpy((char *)params->device_id, tmpstr,
2332                         MIN(sizeof(params->device_id), sizeof(tmpstr)));
2333         } else {
2334                 strncpy((char *)cbe_lun->device_id, params->device_id,
2335                         MIN(sizeof(cbe_lun->device_id),
2336                             sizeof(params->device_id)));
2337         }
2338
2339         TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2340
2341         be_lun->io_taskqueue = taskqueue_create("ctlblocktq", M_WAITOK,
2342             taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2343
2344         if (be_lun->io_taskqueue == NULL) {
2345                 snprintf(req->error_str, sizeof(req->error_str),
2346                          "unable to create taskqueue");
2347                 goto bailout_error;
2348         }
2349
2350         /*
2351          * Note that we start the same number of threads by default for
2352          * both the file case and the block device case.  For the file
2353          * case, we need multiple threads to allow concurrency, because the
2354          * vnode interface is designed to be a blocking interface.  For the
2355          * block device case, ZFS zvols at least will block the caller's
2356          * context in many instances, and so we need multiple threads to
2357          * overcome that problem.  Other block devices don't need as many
2358          * threads, but they shouldn't cause too many problems.
2359          *
2360          * If the user wants to just have a single thread for a block
2361          * device, he can specify that when the LUN is created, or change
2362          * the tunable/sysctl to alter the default number of threads.
2363          */
2364         retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
2365                                          /*num threads*/num_threads,
2366                                          /*priority*/PUSER,
2367                                          /*proc*/control_softc->ctl_proc,
2368                                          /*thread name*/"block");
2369
2370         if (retval != 0)
2371                 goto bailout_error;
2372
2373         be_lun->num_threads = num_threads;
2374
2375         retval = ctl_add_lun(&be_lun->cbe_lun);
2376         if (retval != 0) {
2377                 snprintf(req->error_str, sizeof(req->error_str),
2378                          "ctl_add_lun() returned error %d, see dmesg for "
2379                          "details", retval);
2380                 retval = 0;
2381                 goto bailout_error;
2382         }
2383
2384         be_lun->disk_stats = devstat_new_entry("cbb", cbe_lun->lun_id,
2385                                                cbe_lun->blocksize,
2386                                                DEVSTAT_ALL_SUPPORTED,
2387                                                cbe_lun->lun_type
2388                                                | DEVSTAT_TYPE_IF_OTHER,
2389                                                DEVSTAT_PRIORITY_OTHER);
2390
2391         mtx_lock(&softc->lock);
2392         softc->num_luns++;
2393         SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
2394         mtx_unlock(&softc->lock);
2395
2396         params->req_lun_id = cbe_lun->lun_id;
2397
2398         return (retval);
2399
2400 bailout_error:
2401         req->status = CTL_LUN_ERROR;
2402
2403         if (be_lun->io_taskqueue != NULL)
2404                 taskqueue_free(be_lun->io_taskqueue);
2405         ctl_be_block_close(be_lun);
2406         if (be_lun->dev_path != NULL)
2407                 free(be_lun->dev_path, M_CTLBLK);
2408         nvlist_destroy(cbe_lun->options);
2409         mtx_destroy(&be_lun->queue_lock);
2410         mtx_destroy(&be_lun->io_lock);
2411         free(be_lun, M_CTLBLK);
2412
2413         return (retval);
2414 }
2415
2416 static int
2417 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2418 {
2419         struct ctl_lun_rm_params *params;
2420         struct ctl_be_block_lun *be_lun;
2421         struct ctl_be_lun *cbe_lun;
2422         int retval;
2423
2424         params = &req->reqdata.rm;
2425
2426         sx_xlock(&softc->modify_lock);
2427         mtx_lock(&softc->lock);
2428         SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2429                 if (be_lun->cbe_lun.lun_id == params->lun_id) {
2430                         SLIST_REMOVE(&softc->lun_list, be_lun,
2431                             ctl_be_block_lun, links);
2432                         softc->num_luns--;
2433                         break;
2434                 }
2435         }
2436         mtx_unlock(&softc->lock);
2437         sx_xunlock(&softc->modify_lock);
2438         if (be_lun == NULL) {
2439                 snprintf(req->error_str, sizeof(req->error_str),
2440                          "LUN %u is not managed by the block backend",
2441                          params->lun_id);
2442                 goto bailout_error;
2443         }
2444         cbe_lun = &be_lun->cbe_lun;
2445
2446         if (be_lun->vn != NULL) {
2447                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2448                 ctl_lun_no_media(cbe_lun);
2449                 taskqueue_drain_all(be_lun->io_taskqueue);
2450                 ctl_be_block_close(be_lun);
2451         }
2452
2453         mtx_lock(&softc->lock);
2454         be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2455         mtx_unlock(&softc->lock);
2456
2457         retval = ctl_remove_lun(cbe_lun);
2458         if (retval != 0) {
2459                 snprintf(req->error_str, sizeof(req->error_str),
2460                          "error %d returned from ctl_remove_lun() for "
2461                          "LUN %d", retval, params->lun_id);
2462                 mtx_lock(&softc->lock);
2463                 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2464                 mtx_unlock(&softc->lock);
2465                 goto bailout_error;
2466         }
2467
2468         mtx_lock(&softc->lock);
2469         while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2470                 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblockrm", 0);
2471                 if (retval == EINTR)
2472                         break;
2473         }
2474         be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2475         if (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2476                 mtx_unlock(&softc->lock);
2477                 free(be_lun, M_CTLBLK);
2478         } else {
2479                 mtx_unlock(&softc->lock);
2480                 return (EINTR);
2481         }
2482
2483         req->status = CTL_LUN_OK;
2484         return (0);
2485
2486 bailout_error:
2487         req->status = CTL_LUN_ERROR;
2488         return (0);
2489 }
2490
2491 static int
2492 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2493 {
2494         struct ctl_lun_modify_params *params;
2495         struct ctl_be_block_lun *be_lun;
2496         struct ctl_be_lun *cbe_lun;
2497         const char *value;
2498         uint64_t oldsize;
2499         int error, wasprim;
2500
2501         params = &req->reqdata.modify;
2502
2503         sx_xlock(&softc->modify_lock);
2504         mtx_lock(&softc->lock);
2505         SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2506                 if (be_lun->cbe_lun.lun_id == params->lun_id)
2507                         break;
2508         }
2509         mtx_unlock(&softc->lock);
2510         if (be_lun == NULL) {
2511                 snprintf(req->error_str, sizeof(req->error_str),
2512                          "LUN %u is not managed by the block backend",
2513                          params->lun_id);
2514                 goto bailout_error;
2515         }
2516         cbe_lun = &be_lun->cbe_lun;
2517
2518         if (params->lun_size_bytes != 0)
2519                 be_lun->params.lun_size_bytes = params->lun_size_bytes;
2520
2521         if (req->args_nvl != NULL) {
2522                 nvlist_destroy(cbe_lun->options);
2523                 cbe_lun->options = nvlist_clone(req->args_nvl);
2524         }
2525
2526         wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
2527         value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2528         if (value != NULL) {
2529                 if (strcmp(value, "primary") == 0)
2530                         cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2531                 else
2532                         cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2533         } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2534                 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2535         else
2536                 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2537         if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
2538                 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
2539                         ctl_lun_primary(cbe_lun);
2540                 else
2541                         ctl_lun_secondary(cbe_lun);
2542         }
2543
2544         oldsize = be_lun->size_blocks;
2545         if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2546             control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2547                 if (be_lun->vn == NULL)
2548                         error = ctl_be_block_open(be_lun, req);
2549                 else if (vn_isdisk_error(be_lun->vn, &error))
2550                         error = ctl_be_block_open_dev(be_lun, req);
2551                 else if (be_lun->vn->v_type == VREG) {
2552                         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2553                         error = ctl_be_block_open_file(be_lun, req);
2554                         VOP_UNLOCK(be_lun->vn);
2555                 } else
2556                         error = EINVAL;
2557                 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) &&
2558                     be_lun->vn != NULL) {
2559                         cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2560                         ctl_lun_has_media(cbe_lun);
2561                 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 &&
2562                     be_lun->vn == NULL) {
2563                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2564                         ctl_lun_no_media(cbe_lun);
2565                 }
2566                 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2567         } else {
2568                 if (be_lun->vn != NULL) {
2569                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2570                         ctl_lun_no_media(cbe_lun);
2571                         taskqueue_drain_all(be_lun->io_taskqueue);
2572                         error = ctl_be_block_close(be_lun);
2573                 } else
2574                         error = 0;
2575         }
2576         if (be_lun->size_blocks != oldsize)
2577                 ctl_lun_capacity_changed(cbe_lun);
2578
2579         /* Tell the user the exact size we ended up using */
2580         params->lun_size_bytes = be_lun->size_bytes;
2581
2582         sx_xunlock(&softc->modify_lock);
2583         req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2584         return (0);
2585
2586 bailout_error:
2587         sx_xunlock(&softc->modify_lock);
2588         req->status = CTL_LUN_ERROR;
2589         return (0);
2590 }
2591
2592 static void
2593 ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun)
2594 {
2595         struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)cbe_lun;
2596         struct ctl_be_block_softc *softc = be_lun->softc;
2597
2598         taskqueue_drain_all(be_lun->io_taskqueue);
2599         taskqueue_free(be_lun->io_taskqueue);
2600         if (be_lun->disk_stats != NULL)
2601                 devstat_remove_entry(be_lun->disk_stats);
2602         nvlist_destroy(be_lun->cbe_lun.options);
2603         free(be_lun->dev_path, M_CTLBLK);
2604         mtx_destroy(&be_lun->queue_lock);
2605         mtx_destroy(&be_lun->io_lock);
2606
2607         mtx_lock(&softc->lock);
2608         be_lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2609         if (be_lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2610                 wakeup(be_lun);
2611         else
2612                 free(be_lun, M_CTLBLK);
2613         mtx_unlock(&softc->lock);
2614 }
2615
2616 static int
2617 ctl_be_block_config_write(union ctl_io *io)
2618 {
2619         struct ctl_be_block_lun *be_lun;
2620         struct ctl_be_lun *cbe_lun;
2621         int retval;
2622
2623         DPRINTF("entered\n");
2624
2625         cbe_lun = CTL_BACKEND_LUN(io);
2626         be_lun = (struct ctl_be_block_lun *)cbe_lun;
2627
2628         retval = 0;
2629         switch (io->scsiio.cdb[0]) {
2630         case SYNCHRONIZE_CACHE:
2631         case SYNCHRONIZE_CACHE_16:
2632         case WRITE_SAME_10:
2633         case WRITE_SAME_16:
2634         case UNMAP:
2635                 /*
2636                  * The upper level CTL code will filter out any CDBs with
2637                  * the immediate bit set and return the proper error.
2638                  *
2639                  * We don't really need to worry about what LBA range the
2640                  * user asked to be synced out.  When they issue a sync
2641                  * cache command, we'll sync out the whole thing.
2642                  */
2643                 mtx_lock(&be_lun->queue_lock);
2644                 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2645                                    links);
2646                 mtx_unlock(&be_lun->queue_lock);
2647                 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2648                 break;
2649         case START_STOP_UNIT: {
2650                 struct scsi_start_stop_unit *cdb;
2651                 struct ctl_lun_req req;
2652
2653                 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2654                 if ((cdb->how & SSS_PC_MASK) != 0) {
2655                         ctl_set_success(&io->scsiio);
2656                         ctl_config_write_done(io);
2657                         break;
2658                 }
2659                 if (cdb->how & SSS_START) {
2660                         if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) {
2661                                 retval = ctl_be_block_open(be_lun, &req);
2662                                 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2663                                 if (retval == 0) {
2664                                         cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2665                                         ctl_lun_has_media(cbe_lun);
2666                                 } else {
2667                                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2668                                         ctl_lun_no_media(cbe_lun);
2669                                 }
2670                         }
2671                         ctl_start_lun(cbe_lun);
2672                 } else {
2673                         ctl_stop_lun(cbe_lun);
2674                         if (cdb->how & SSS_LOEJ) {
2675                                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2676                                 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED;
2677                                 ctl_lun_ejected(cbe_lun);
2678                                 if (be_lun->vn != NULL)
2679                                         ctl_be_block_close(be_lun);
2680                         }
2681                 }
2682
2683                 ctl_set_success(&io->scsiio);
2684                 ctl_config_write_done(io);
2685                 break;
2686         }
2687         case PREVENT_ALLOW:
2688                 ctl_set_success(&io->scsiio);
2689                 ctl_config_write_done(io);
2690                 break;
2691         default:
2692                 ctl_set_invalid_opcode(&io->scsiio);
2693                 ctl_config_write_done(io);
2694                 retval = CTL_RETVAL_COMPLETE;
2695                 break;
2696         }
2697
2698         return (retval);
2699 }
2700
2701 static int
2702 ctl_be_block_config_read(union ctl_io *io)
2703 {
2704         struct ctl_be_block_lun *be_lun;
2705         int retval = 0;
2706
2707         DPRINTF("entered\n");
2708
2709         be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
2710
2711         switch (io->scsiio.cdb[0]) {
2712         case SERVICE_ACTION_IN:
2713                 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
2714                         mtx_lock(&be_lun->queue_lock);
2715                         STAILQ_INSERT_TAIL(&be_lun->config_read_queue,
2716                             &io->io_hdr, links);
2717                         mtx_unlock(&be_lun->queue_lock);
2718                         taskqueue_enqueue(be_lun->io_taskqueue,
2719                             &be_lun->io_task);
2720                         retval = CTL_RETVAL_QUEUED;
2721                         break;
2722                 }
2723                 ctl_set_invalid_field(&io->scsiio,
2724                                       /*sks_valid*/ 1,
2725                                       /*command*/ 1,
2726                                       /*field*/ 1,
2727                                       /*bit_valid*/ 1,
2728                                       /*bit*/ 4);
2729                 ctl_config_read_done(io);
2730                 retval = CTL_RETVAL_COMPLETE;
2731                 break;
2732         default:
2733                 ctl_set_invalid_opcode(&io->scsiio);
2734                 ctl_config_read_done(io);
2735                 retval = CTL_RETVAL_COMPLETE;
2736                 break;
2737         }
2738
2739         return (retval);
2740 }
2741
2742 static int
2743 ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb)
2744 {
2745         struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
2746         int retval;
2747
2748         retval = sbuf_printf(sb, "\t<num_threads>");
2749         if (retval != 0)
2750                 goto bailout;
2751         retval = sbuf_printf(sb, "%d", lun->num_threads);
2752         if (retval != 0)
2753                 goto bailout;
2754         retval = sbuf_printf(sb, "</num_threads>\n");
2755
2756 bailout:
2757         return (retval);
2758 }
2759
2760 static uint64_t
2761 ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
2762 {
2763         struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
2764
2765         if (lun->getattr == NULL)
2766                 return (UINT64_MAX);
2767         return (lun->getattr(lun, attrname));
2768 }
2769
2770 static int
2771 ctl_be_block_init(void)
2772 {
2773         struct ctl_be_block_softc *softc = &backend_block_softc;
2774
2775         sx_init(&softc->modify_lock, "ctlblock modify");
2776         mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
2777         softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2778             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2779         softc->buf_zone = uma_zcreate("ctlblock", CTLBLK_MAX_SEG,
2780             NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
2781         SLIST_INIT(&softc->lun_list);
2782         return (0);
2783 }
2784
2785 static int
2786 ctl_be_block_shutdown(void)
2787 {
2788         struct ctl_be_block_softc *softc = &backend_block_softc;
2789         struct ctl_be_block_lun *lun;
2790
2791         mtx_lock(&softc->lock);
2792         while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
2793                 SLIST_REMOVE_HEAD(&softc->lun_list, links);
2794                 softc->num_luns--;
2795                 /*
2796                  * Drop our lock here.  Since ctl_remove_lun() can call
2797                  * back into us, this could potentially lead to a recursive
2798                  * lock of the same mutex, which would cause a hang.
2799                  */
2800                 mtx_unlock(&softc->lock);
2801                 ctl_remove_lun(&lun->cbe_lun);
2802                 mtx_lock(&softc->lock);
2803         }
2804         mtx_unlock(&softc->lock);
2805         uma_zdestroy(softc->buf_zone);
2806         uma_zdestroy(softc->beio_zone);
2807         mtx_destroy(&softc->lock);
2808         sx_destroy(&softc->modify_lock);
2809         return (0);
2810 }