sys/kern/subr_diskiocom.c

   1 /*
   2  * Copyright (c) 2012 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34 #include <sys/param.h>
  35 #include <sys/systm.h>
  36 #include <sys/kernel.h>
  37 #include <sys/proc.h>
  38 #include <sys/sysctl.h>
  39 #include <sys/buf.h>
  40 #include <sys/conf.h>
  41 #include <sys/disklabel.h>
  42 #include <sys/disklabel32.h>
  43 #include <sys/disklabel64.h>
  44 #include <sys/diskslice.h>
  45 #include <sys/diskmbr.h>
  46 #include <sys/disk.h>
  47 #include <sys/malloc.h>
  48 #include <sys/device.h>
  49 #include <sys/devfs.h>
  50 #include <sys/thread.h>
  51 #include <sys/queue.h>
  52 #include <sys/lock.h>
  53 #include <sys/stat.h>
  54 #include <sys/uuid.h>
  55
  56 #include <sys/dmsg.h>
  57
  58 #include <sys/buf2.h>
  59 #include <sys/mplock2.h>
  60 #include <sys/msgport2.h>
  61 #include <sys/thread2.h>
  62
  63 struct dios_open {
  64         int     openrd;
  65         int     openwr;
  66 };
  67
  68 struct dios_io {
  69         int     count;
  70         int     eof;
  71 };
  72
  73 static MALLOC_DEFINE(M_DMSG_DISK, "dmsg_disk", "disk dmsg");
  74
  75 static int disk_iocom_reconnect(struct disk *dp, struct file *fp);
  76 static int disk_rcvdmsg(kdmsg_msg_t *msg);
  77
  78 static void disk_blk_open(struct disk *dp, kdmsg_msg_t *msg);
  79 static void disk_blk_read(struct disk *dp, kdmsg_msg_t *msg);
  80 static void disk_blk_write(struct disk *dp, kdmsg_msg_t *msg);
  81 static void disk_blk_flush(struct disk *dp, kdmsg_msg_t *msg);
  82 static void disk_blk_freeblks(struct disk *dp, kdmsg_msg_t *msg);
  83 static void diskiodone(struct bio *bio);
  84
  85 void
  86 disk_iocom_init(struct disk *dp)
  87 {
  88         kdmsg_iocom_init(&dp->d_iocom, dp,
  89                          KDMSG_IOCOMF_AUTOCONN |
  90                          KDMSG_IOCOMF_AUTOSPAN |
  91                          KDMSG_IOCOMF_AUTOCIRC,
  92                          M_DMSG_DISK, disk_rcvdmsg);
  93 }
  94
  95 void
  96 disk_iocom_update(struct disk *dp)
  97 {
  98 }
  99
 100 void
 101 disk_iocom_uninit(struct disk *dp)
 102 {
 103         kdmsg_iocom_uninit(&dp->d_iocom);
 104 }
 105
 106 int
 107 disk_iocom_ioctl(struct disk *dp, int cmd, void *data)
 108 {
 109         struct file *fp;
 110         struct disk_ioc_recluster *recl;
 111         int error;
 112
 113         switch(cmd) {
 114         case DIOCRECLUSTER:
 115                 recl = data;
 116                 fp = holdfp(curproc->p_fd, recl->fd, -1);
 117                 if (fp) {
 118                         error = disk_iocom_reconnect(dp, fp);
 119                 } else {
 120                         error = EINVAL;
 121                 }
 122                 break;
 123         default:
 124                 error = EOPNOTSUPP;
 125                 break;
 126         }
 127         return error;
 128 }
 129
 130 static
 131 int
 132 disk_iocom_reconnect(struct disk *dp, struct file *fp)
 133 {
 134         char devname[64];
 135
 136         ksnprintf(devname, sizeof(devname), "%s%d",
 137                   dev_dname(dp->d_rawdev), dkunit(dp->d_rawdev));
 138
 139         kdmsg_iocom_reconnect(&dp->d_iocom, fp, devname);
 140
 141         dp->d_iocom.auto_lnk_conn.pfs_type = DMSG_PFSTYPE_SERVER;
 142         dp->d_iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
 143         dp->d_iocom.auto_lnk_conn.peer_type = DMSG_PEER_BLOCK;
 144         dp->d_iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_BLOCK;
 145         dp->d_iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
 146         ksnprintf(dp->d_iocom.auto_lnk_conn.cl_label,
 147                   sizeof(dp->d_iocom.auto_lnk_conn.cl_label),
 148                   "%s/%s", hostname, devname);
 149         if (dp->d_info.d_serialno) {
 150                 ksnprintf(dp->d_iocom.auto_lnk_conn.fs_label,
 151                           sizeof(dp->d_iocom.auto_lnk_conn.fs_label),
 152                           "%s", dp->d_info.d_serialno);
 153         }
 154
 155         dp->d_iocom.auto_lnk_span.pfs_type = DMSG_PFSTYPE_SERVER;
 156         dp->d_iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
 157         dp->d_iocom.auto_lnk_span.peer_type = DMSG_PEER_BLOCK;
 158         dp->d_iocom.auto_lnk_span.media.block.bytes =
 159                                                 dp->d_info.d_media_size;
 160         dp->d_iocom.auto_lnk_span.media.block.blksize =
 161                                                 dp->d_info.d_media_blksize;
 162         ksnprintf(dp->d_iocom.auto_lnk_span.cl_label,
 163                   sizeof(dp->d_iocom.auto_lnk_span.cl_label),
 164                   "%s/%s", hostname, devname);
 165         if (dp->d_info.d_serialno) {
 166                 ksnprintf(dp->d_iocom.auto_lnk_span.fs_label,
 167                           sizeof(dp->d_iocom.auto_lnk_span.fs_label),
 168                           "%s", dp->d_info.d_serialno);
 169         }
 170
 171         kdmsg_iocom_autoinitiate(&dp->d_iocom, NULL);
 172
 173         return (0);
 174 }
 175
 176 int
 177 disk_rcvdmsg(kdmsg_msg_t *msg)
 178 {
 179         struct disk *dp = msg->iocom->handle;
 180
 181         /*
 182          * Handle debug messages (these might not be in transactions)
 183          */
 184         switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
 185         case DMSG_DBG_SHELL:
 186                 /*
 187                  * Execute shell command (not supported atm)
 188                  */
 189                 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
 190                 return(0);
 191         case DMSG_DBG_SHELL | DMSGF_REPLY:
 192                 if (msg->aux_data) {
 193                         msg->aux_data[msg->aux_size - 1] = 0;
 194                         kprintf("diskiocom: DEBUGMSG: %s\n", msg->aux_data);
 195                 }
 196                 return(0);
 197         }
 198
 199         /*
 200          * All remaining messages must be in a transaction
 201          *
 202          * NOTE!  We are switching on the first message's command.  The
 203          *        actual message command within the transaction may be
 204          *        different (if streaming within a transaction).
 205          */
 206         if (msg->state == NULL) {
 207                 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
 208                 return(0);
 209         }
 210
 211         switch(msg->state->rxcmd & DMSGF_CMDSWMASK) {
 212         case DMSG_BLK_OPEN:
 213         case DMSG_BLK_CLOSE:
 214                 disk_blk_open(dp, msg);
 215                 break;
 216         case DMSG_BLK_READ:
 217                 disk_blk_read(dp, msg);
 218                 break;
 219         case DMSG_BLK_WRITE:
 220                 disk_blk_write(dp, msg);
 221                 break;
 222         case DMSG_BLK_FLUSH:
 223                 disk_blk_flush(dp, msg);
 224                 break;
 225         case DMSG_BLK_FREEBLKS:
 226                 disk_blk_freeblks(dp, msg);
 227                 break;
 228         default:
 229                 if ((msg->any.head.cmd & DMSGF_REPLY) == 0) {
 230                         if (msg->any.head.cmd & DMSGF_DELETE)
 231                                 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
 232                         else
 233                                 kdmsg_msg_result(msg, DMSG_ERR_NOSUPP);
 234                 }
 235                 break;
 236         }
 237         return (0);
 238 }
 239
 240 static
 241 void
 242 disk_blk_open(struct disk *dp, kdmsg_msg_t *msg)
 243 {
 244         struct dios_open *openst;
 245         int error = DMSG_ERR_NOSUPP;
 246         int fflags;
 247
 248         openst = msg->state->any.any;
 249         if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_OPEN) {
 250                 if (openst == NULL) {
 251                         openst = kmalloc(sizeof(*openst), M_DEVBUF,
 252                                                 M_WAITOK | M_ZERO);
 253                         msg->state->any.any = openst;
 254                 }
 255                 fflags = 0;
 256                 if (msg->any.blk_open.modes & DMSG_BLKOPEN_RD)
 257                         fflags = FREAD;
 258                 if (msg->any.blk_open.modes & DMSG_BLKOPEN_WR)
 259                         fflags |= FWRITE;
 260                 error = dev_dopen(dp->d_rawdev, fflags, S_IFCHR, proc0.p_ucred);
 261                 if (error) {
 262                         error = DMSG_ERR_IO;
 263                 } else {
 264                         if (msg->any.blk_open.modes & DMSG_BLKOPEN_RD)
 265                                 ++openst->openrd;
 266                         if (msg->any.blk_open.modes & DMSG_BLKOPEN_WR)
 267                                 ++openst->openwr;
 268                 }
 269         }
 270         if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_CLOSE &&
 271             openst) {
 272                 fflags = 0;
 273                 if ((msg->any.blk_open.modes & DMSG_BLKOPEN_RD) &&
 274                     openst->openrd) {
 275                         fflags = FREAD;
 276                 }
 277                 if ((msg->any.blk_open.modes & DMSG_BLKOPEN_WR) &&
 278                     openst->openwr) {
 279                         fflags |= FWRITE;
 280                 }
 281                 error = dev_dclose(dp->d_rawdev, fflags, S_IFCHR);
 282                 if (error) {
 283                         error = DMSG_ERR_IO;
 284                 } else {
 285                         if (msg->any.blk_open.modes & DMSG_BLKOPEN_RD)
 286                                 --openst->openrd;
 287                         if (msg->any.blk_open.modes & DMSG_BLKOPEN_WR)
 288                                 --openst->openwr;
 289                 }
 290         }
 291         if (msg->any.head.cmd & DMSGF_DELETE) {
 292                 if (openst) {
 293                         while (openst->openrd && openst->openwr) {
 294                                 --openst->openrd;
 295                                 --openst->openwr;
 296                                 dev_dclose(dp->d_rawdev, FREAD|FWRITE, S_IFCHR);
 297                         }
 298                         while (openst->openrd) {
 299                                 --openst->openrd;
 300                                 dev_dclose(dp->d_rawdev, FREAD, S_IFCHR);
 301                         }
 302                         while (openst->openwr) {
 303                                 --openst->openwr;
 304                                 dev_dclose(dp->d_rawdev, FWRITE, S_IFCHR);
 305                         }
 306                         kfree(openst, M_DEVBUF);
 307                         msg->state->any.any = NULL;
 308                 }
 309                 kdmsg_msg_reply(msg, error);
 310         } else {
 311                 kdmsg_msg_result(msg, error);
 312         }
 313 }
 314
 315 static
 316 void
 317 disk_blk_read(struct disk *dp, kdmsg_msg_t *msg)
 318 {
 319         struct dios_io *iost;
 320         struct buf *bp;
 321         struct bio *bio;
 322         int error = DMSG_ERR_NOSUPP;
 323         int reterr = 1;
 324
 325         /*
 326          * Only DMSG_BLK_READ commands imply read ops.
 327          */
 328         iost = msg->state->any.any;
 329         if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_READ) {
 330                 if (msg->any.blk_read.bytes < DEV_BSIZE ||
 331                     msg->any.blk_read.bytes > MAXPHYS) {
 332                         error = DMSG_ERR_PARAM;
 333                         goto done;
 334                 }
 335                 if (iost == NULL) {
 336                         iost = kmalloc(sizeof(*iost), M_DEVBUF,
 337                                        M_WAITOK | M_ZERO);
 338                         msg->state->any.any = iost;
 339                 }
 340                 reterr = 0;
 341                 bp = geteblk(msg->any.blk_read.bytes);
 342                 bio = &bp->b_bio1;
 343                 bp->b_cmd = BUF_CMD_READ;
 344                 bp->b_bcount = msg->any.blk_read.bytes;
 345                 bp->b_resid = bp->b_bcount;
 346                 bio->bio_offset = msg->any.blk_read.offset;
 347                 bio->bio_caller_info1.ptr = msg->state;
 348                 bio->bio_done = diskiodone;
 349                 /* kdmsg_state_hold(msg->state); */
 350
 351                 atomic_add_int(&iost->count, 1);
 352                 if (msg->any.head.cmd & DMSGF_DELETE)
 353                         iost->eof = 1;
 354                 BUF_KERNPROC(bp);
 355                 dev_dstrategy(dp->d_rawdev, bio);
 356         }
 357 done:
 358         if (reterr) {
 359                 if (msg->any.head.cmd & DMSGF_DELETE) {
 360                         if (iost && iost->count == 0) {
 361                                 kfree(iost, M_DEVBUF);
 362                                 msg->state->any.any = NULL;
 363                         }
 364                         kdmsg_msg_reply(msg, error);
 365                 } else {
 366                         kdmsg_msg_result(msg, error);
 367                 }
 368         }
 369 }
 370
 371 static
 372 void
 373 disk_blk_write(struct disk *dp, kdmsg_msg_t *msg)
 374 {
 375         struct dios_io *iost;
 376         struct buf *bp;
 377         struct bio *bio;
 378         int error = DMSG_ERR_NOSUPP;
 379         int reterr = 1;
 380
 381         /*
 382          * Only DMSG_BLK_WRITE commands imply read ops.
 383          */
 384         iost = msg->state->any.any;
 385         if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_WRITE) {
 386                 if (msg->any.blk_write.bytes < DEV_BSIZE ||
 387                     msg->any.blk_write.bytes > MAXPHYS) {
 388                         error = DMSG_ERR_PARAM;
 389                         goto done;
 390                 }
 391                 if (iost == NULL) {
 392                         iost = kmalloc(sizeof(*iost), M_DEVBUF,
 393                                        M_WAITOK | M_ZERO);
 394                         msg->state->any.any = iost;
 395                 }
 396
 397                 /*
 398                  * Issue WRITE.  Short data implies zeros.  Try to optimize
 399                  * the buffer cache buffer for the case where we can just
 400                  * use the message's data pointer.
 401                  */
 402                 reterr = 0;
 403                 if (msg->aux_size >= msg->any.blk_write.bytes)
 404                         bp = getpbuf(NULL);
 405                 else
 406                         bp = geteblk(msg->any.blk_write.bytes);
 407                 bio = &bp->b_bio1;
 408                 bp->b_cmd = BUF_CMD_WRITE;
 409                 bp->b_bcount = msg->any.blk_write.bytes;
 410                 bp->b_resid = bp->b_bcount;
 411                 if (msg->aux_size >= msg->any.blk_write.bytes) {
 412                         bp->b_data = msg->aux_data;
 413                 } else {
 414                         bcopy(msg->aux_data, bp->b_data, msg->aux_size);
 415                         bzero(bp->b_data + msg->aux_size,
 416                               msg->any.blk_write.bytes - msg->aux_size);
 417                 }
 418                 bio->bio_offset = msg->any.blk_write.offset;
 419                 bio->bio_caller_info1.ptr = msg->state;
 420                 bio->bio_done = diskiodone;
 421                 /* kdmsg_state_hold(msg->state); */
 422
 423                 atomic_add_int(&iost->count, 1);
 424                 if (msg->any.head.cmd & DMSGF_DELETE)
 425                         iost->eof = 1;
 426                 BUF_KERNPROC(bp);
 427                 dev_dstrategy(dp->d_rawdev, bio);
 428         }
 429 done:
 430         if (reterr) {
 431                 if (msg->any.head.cmd & DMSGF_DELETE) {
 432                         if (iost && iost->count == 0) {
 433                                 kfree(iost, M_DEVBUF);
 434                                 msg->state->any.any = NULL;
 435                         }
 436                         kdmsg_msg_reply(msg, error);
 437                 } else {
 438                         kdmsg_msg_result(msg, error);
 439                 }
 440         }
 441 }
 442
 443 static
 444 void
 445 disk_blk_flush(struct disk *dp, kdmsg_msg_t *msg)
 446 {
 447         struct dios_io *iost;
 448         struct buf *bp;
 449         struct bio *bio;
 450         int error = DMSG_ERR_NOSUPP;
 451         int reterr = 1;
 452
 453         /*
 454          * Only DMSG_BLK_FLUSH commands imply read ops.
 455          */
 456         iost = msg->state->any.any;
 457         if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_FLUSH) {
 458                 if (iost == NULL) {
 459                         iost = kmalloc(sizeof(*iost), M_DEVBUF,
 460                                        M_WAITOK | M_ZERO);
 461                         msg->state->any.any = iost;
 462                 }
 463                 reterr = 0;
 464                 bp = getpbuf(NULL);
 465                 bio = &bp->b_bio1;
 466                 bp->b_cmd = BUF_CMD_FLUSH;
 467                 bp->b_bcount = msg->any.blk_flush.bytes;
 468                 bp->b_resid = 0;
 469                 bio->bio_offset = msg->any.blk_flush.offset;
 470                 bio->bio_caller_info1.ptr = msg->state;
 471                 bio->bio_done = diskiodone;
 472                 /* kdmsg_state_hold(msg->state); */
 473
 474                 atomic_add_int(&iost->count, 1);
 475                 if (msg->any.head.cmd & DMSGF_DELETE)
 476                         iost->eof = 1;
 477                 BUF_KERNPROC(bp);
 478                 dev_dstrategy(dp->d_rawdev, bio);
 479         }
 480         if (reterr) {
 481                 if (msg->any.head.cmd & DMSGF_DELETE) {
 482                         if (iost && iost->count == 0) {
 483                                 kfree(iost, M_DEVBUF);
 484                                 msg->state->any.any = NULL;
 485                         }
 486                         kdmsg_msg_reply(msg, error);
 487                 } else {
 488                         kdmsg_msg_result(msg, error);
 489                 }
 490         }
 491 }
 492
 493 static
 494 void
 495 disk_blk_freeblks(struct disk *dp, kdmsg_msg_t *msg)
 496 {
 497         struct dios_io *iost;
 498         struct buf *bp;
 499         struct bio *bio;
 500         int error = DMSG_ERR_NOSUPP;
 501         int reterr = 1;
 502
 503         /*
 504          * Only DMSG_BLK_FREEBLKS commands imply read ops.
 505          */
 506         iost = msg->state->any.any;
 507         if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_FREEBLKS) {
 508                 if (iost == NULL) {
 509                         iost = kmalloc(sizeof(*iost), M_DEVBUF,
 510                                        M_WAITOK | M_ZERO);
 511                         msg->state->any.any = iost;
 512                 }
 513                 reterr = 0;
 514                 bp = getpbuf(NULL);
 515                 bio = &bp->b_bio1;
 516                 bp->b_cmd = BUF_CMD_FREEBLKS;
 517                 bp->b_bcount = msg->any.blk_freeblks.bytes;
 518                 bp->b_resid = 0;
 519                 bio->bio_offset = msg->any.blk_freeblks.offset;
 520                 bio->bio_caller_info1.ptr = msg->state;
 521                 bio->bio_done = diskiodone;
 522                 /* kdmsg_state_hold(msg->state); */
 523
 524                 atomic_add_int(&iost->count, 1);
 525                 if (msg->any.head.cmd & DMSGF_DELETE)
 526                         iost->eof = 1;
 527                 BUF_KERNPROC(bp);
 528                 dev_dstrategy(dp->d_rawdev, bio);
 529         }
 530         if (reterr) {
 531                 if (msg->any.head.cmd & DMSGF_DELETE) {
 532                         if (iost && iost->count == 0) {
 533                                 kfree(iost, M_DEVBUF);
 534                                 msg->state->any.any = NULL;
 535                         }
 536                         kdmsg_msg_reply(msg, error);
 537                 } else {
 538                         kdmsg_msg_result(msg, error);
 539                 }
 540         }
 541 }
 542
 543 static
 544 void
 545 diskiodone(struct bio *bio)
 546 {
 547         struct buf *bp = bio->bio_buf;
 548         kdmsg_state_t *state = bio->bio_caller_info1.ptr;
 549         kdmsg_msg_t *rmsg;
 550         struct dios_io *iost = state->any.any;
 551         int error;
 552         int resid = 0;
 553         int bytes;
 554         uint32_t cmd;
 555         void *data;
 556
 557         cmd = DMSG_LNK_ERROR;
 558         data = NULL;
 559         bytes = 0;
 560
 561         switch(bp->b_cmd) {
 562         case BUF_CMD_READ:
 563                 cmd = DMSG_LNK_ERROR;
 564                 data = bp->b_data;
 565                 bytes = bp->b_bcount;
 566                 /* fall through */
 567         case BUF_CMD_WRITE:
 568                 if (bp->b_flags & B_ERROR) {
 569                         error = bp->b_error;
 570                 } else {
 571                         error = 0;
 572                         resid = bp->b_resid;
 573                 }
 574                 break;
 575         case BUF_CMD_FLUSH:
 576         case BUF_CMD_FREEBLKS:
 577                 if (bp->b_flags & B_ERROR)
 578                         error = bp->b_error;
 579                 else
 580                         error = 0;
 581                 break;
 582         default:
 583                 panic("diskiodone: Unknown bio cmd = %d\n",
 584                       bio->bio_buf->b_cmd);
 585                 error = 0;      /* avoid compiler warning */
 586                 break;          /* NOT REACHED */
 587         }
 588
 589         /*
 590          * Convert error to DMSG_ERR_* code.
 591          */
 592         if (error)
 593                 error = DMSG_ERR_IO;
 594
 595         /*
 596          * Convert LNK_ERROR or BLK_ERROR if non-zero resid.  READS will
 597          * have already converted cmd to BLK_ERROR and set up data to return.
 598          */
 599         if (resid && cmd == DMSG_LNK_ERROR)
 600                 cmd = DMSG_BLK_ERROR;
 601         /* XXX txcmd is delayed so this won't work for streaming */
 602         if ((state->txcmd & DMSGF_CREATE) == 0) /* assume serialized */
 603                 cmd |= DMSGF_CREATE;
 604         if (iost->eof) {
 605                 if (atomic_fetchadd_int(&iost->count, -1) == 1)
 606                         cmd |= DMSGF_DELETE;
 607         } else {
 608                 atomic_add_int(&iost->count, -1);
 609         }
 610         cmd |= DMSGF_REPLY;
 611
 612         /*
 613          * Allocate a basic or extended reply.  Be careful not to populate
 614          * extended header fields unless we allocated an extended reply.
 615          */
 616         rmsg = kdmsg_msg_alloc_state(state, cmd, NULL, 0);
 617         if (data) {
 618                 rmsg->aux_data = kmalloc(bytes, state->iocom->mmsg, M_INTWAIT);
 619                 rmsg->aux_size = bytes;
 620                 rmsg->flags |= KDMSG_FLAG_AUXALLOC;
 621                 bcopy(data, rmsg->aux_data, bytes);
 622         }
 623         rmsg->any.blk_error.head.error = error;
 624         if ((cmd & DMSGF_BASECMDMASK) == DMSG_BLK_ERROR)
 625                 rmsg->any.blk_error.resid = resid;
 626         bio->bio_caller_info1.ptr = NULL;
 627         /* kdmsg_state_drop(state); */
 628         kdmsg_msg_write(rmsg);
 629         if (bp->b_flags & B_PAGING) {
 630                 relpbuf(bio->bio_buf, NULL);
 631         } else {
 632                 bp->b_flags |= B_INVAL | B_AGE;
 633                 brelse(bp);
 634         }
 635 }