usr.sbin/bhyve/block_if.c

   1 /*-
   2  * Copyright (c) 2013  Peter Grehan <grehan@freebsd.org>
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  *
  26  * $FreeBSD$
  27  */
  28
  29 #include <sys/cdefs.h>
  30 __FBSDID("$FreeBSD$");
  31
  32 #include <sys/param.h>
  33 #include <sys/queue.h>
  34 #include <sys/errno.h>
  35 #include <sys/stat.h>
  36 #include <sys/ioctl.h>
  37 #include <sys/disk.h>
  38
  39 #include <assert.h>
  40 #include <fcntl.h>
  41 #include <stdio.h>
  42 #include <stdlib.h>
  43 #include <string.h>
  44 #include <pthread.h>
  45 #include <pthread_np.h>
  46 #include <signal.h>
  47 #include <unistd.h>
  48
  49 #include <machine/atomic.h>
  50
  51 #include "bhyverun.h"
  52 #include "mevent.h"
  53 #include "block_if.h"
  54
  55 #define BLOCKIF_SIG     0xb109b109
  56
  57 #define BLOCKIF_NUMTHR  8
  58 #define BLOCKIF_MAXREQ  (64 + BLOCKIF_NUMTHR)
  59
  60 enum blockop {
  61         BOP_READ,
  62         BOP_WRITE,
  63         BOP_FLUSH,
  64         BOP_DELETE
  65 };
  66
  67 enum blockstat {
  68         BST_FREE,
  69         BST_BLOCK,
  70         BST_PEND,
  71         BST_BUSY,
  72         BST_DONE
  73 };
  74
  75 struct blockif_elem {
  76         TAILQ_ENTRY(blockif_elem) be_link;
  77         struct blockif_req  *be_req;
  78         enum blockop         be_op;
  79         enum blockstat       be_status;
  80         pthread_t            be_tid;
  81         off_t                be_block;
  82 };
  83
  84 struct blockif_ctxt {
  85         int                     bc_magic;
  86         int                     bc_fd;
  87         int                     bc_ischr;
  88         int                     bc_isgeom;
  89         int                     bc_candelete;
  90         int                     bc_rdonly;
  91         off_t                   bc_size;
  92         int                     bc_sectsz;
  93         int                     bc_psectsz;
  94         int                     bc_psectoff;
  95         int                     bc_closing;
  96         pthread_t               bc_btid[BLOCKIF_NUMTHR];
  97         pthread_mutex_t         bc_mtx;
  98         pthread_cond_t          bc_cond;
  99
 100         /* Request elements and free/pending/busy queues */
 101         TAILQ_HEAD(, blockif_elem) bc_freeq;
 102         TAILQ_HEAD(, blockif_elem) bc_pendq;
 103         TAILQ_HEAD(, blockif_elem) bc_busyq;
 104         struct blockif_elem     bc_reqs[BLOCKIF_MAXREQ];
 105 };
 106
 107 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT;
 108
 109 struct blockif_sig_elem {
 110         pthread_mutex_t                 bse_mtx;
 111         pthread_cond_t                  bse_cond;
 112         int                             bse_pending;
 113         struct blockif_sig_elem         *bse_next;
 114 };
 115
 116 static struct blockif_sig_elem *blockif_bse_head;
 117
 118 static int
 119 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq,
 120                 enum blockop op)
 121 {
 122         struct blockif_elem *be, *tbe;
 123         off_t off;
 124         int i;
 125
 126         be = TAILQ_FIRST(&bc->bc_freeq);
 127         assert(be != NULL);
 128         assert(be->be_status == BST_FREE);
 129         TAILQ_REMOVE(&bc->bc_freeq, be, be_link);
 130         be->be_req = breq;
 131         be->be_op = op;
 132         switch (op) {
 133         case BOP_READ:
 134         case BOP_WRITE:
 135         case BOP_DELETE:
 136                 off = breq->br_offset;
 137                 for (i = 0; i < breq->br_iovcnt; i++)
 138                         off += breq->br_iov[i].iov_len;
 139                 break;
 140         default:
 141                 off = OFF_MAX;
 142         }
 143         be->be_block = off;
 144         TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
 145                 if (tbe->be_block == breq->br_offset)
 146                         break;
 147         }
 148         if (tbe == NULL) {
 149                 TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) {
 150                         if (tbe->be_block == breq->br_offset)
 151                                 break;
 152                 }
 153         }
 154         if (tbe == NULL)
 155                 be->be_status = BST_PEND;
 156         else
 157                 be->be_status = BST_BLOCK;
 158         TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link);
 159         return (be->be_status == BST_PEND);
 160 }
 161
 162 static int
 163 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep)
 164 {
 165         struct blockif_elem *be;
 166
 167         TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
 168                 if (be->be_status == BST_PEND)
 169                         break;
 170                 assert(be->be_status == BST_BLOCK);
 171         }
 172         if (be == NULL)
 173                 return (0);
 174         TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
 175         be->be_status = BST_BUSY;
 176         be->be_tid = t;
 177         TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link);
 178         *bep = be;
 179         return (1);
 180 }
 181
 182 static void
 183 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be)
 184 {
 185         struct blockif_elem *tbe;
 186
 187         if (be->be_status == BST_DONE || be->be_status == BST_BUSY)
 188                 TAILQ_REMOVE(&bc->bc_busyq, be, be_link);
 189         else
 190                 TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
 191         TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
 192                 if (tbe->be_req->br_offset == be->be_block)
 193                         tbe->be_status = BST_PEND;
 194         }
 195         be->be_tid = 0;
 196         be->be_status = BST_FREE;
 197         be->be_req = NULL;
 198         TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link);
 199 }
 200
 201 static void
 202 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf)
 203 {
 204         struct blockif_req *br;
 205         off_t arg[2];
 206         ssize_t clen, len, off, boff, voff;
 207         int i, err;
 208
 209         br = be->be_req;
 210         if (br->br_iovcnt <= 1)
 211                 buf = NULL;
 212         err = 0;
 213         switch (be->be_op) {
 214         case BOP_READ:
 215                 if (buf == NULL) {
 216                         if ((len = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt,
 217                                    br->br_offset)) < 0)
 218                                 err = errno;
 219                         else
 220                                 br->br_resid -= len;
 221                         break;
 222                 }
 223                 i = 0;
 224                 off = voff = 0;
 225                 while (br->br_resid > 0) {
 226                         len = MIN(br->br_resid, MAXPHYS);
 227                         if (pread(bc->bc_fd, buf, len, br->br_offset +
 228                             off) < 0) {
 229                                 err = errno;
 230                                 break;
 231                         }
 232                         boff = 0;
 233                         do {
 234                                 clen = MIN(len - boff, br->br_iov[i].iov_len -
 235                                     voff);
 236                                 memcpy(br->br_iov[i].iov_base + voff,
 237                                     buf + boff, clen);
 238                                 if (clen < br->br_iov[i].iov_len - voff)
 239                                         voff += clen;
 240                                 else {
 241                                         i++;
 242                                         voff = 0;
 243                                 }
 244                                 boff += clen;
 245                         } while (boff < len);
 246                         off += len;
 247                         br->br_resid -= len;
 248                 }
 249                 break;
 250         case BOP_WRITE:
 251                 if (bc->bc_rdonly) {
 252                         err = EROFS;
 253                         break;
 254                 }
 255                 if (buf == NULL) {
 256                         if ((len = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt,
 257                                     br->br_offset)) < 0)
 258                                 err = errno;
 259                         else
 260                                 br->br_resid -= len;
 261                         break;
 262                 }
 263                 i = 0;
 264                 off = voff = 0;
 265                 while (br->br_resid > 0) {
 266                         len = MIN(br->br_resid, MAXPHYS);
 267                         boff = 0;
 268                         do {
 269                                 clen = MIN(len - boff, br->br_iov[i].iov_len -
 270                                     voff);
 271                                 memcpy(buf + boff,
 272                                     br->br_iov[i].iov_base + voff, clen);
 273                                 if (clen < br->br_iov[i].iov_len - voff)
 274                                         voff += clen;
 275                                 else {
 276                                         i++;
 277                                         voff = 0;
 278                                 }
 279                                 boff += clen;
 280                         } while (boff < len);
 281                         if (pwrite(bc->bc_fd, buf, len, br->br_offset +
 282                             off) < 0) {
 283                                 err = errno;
 284                                 break;
 285                         }
 286                         off += len;
 287                         br->br_resid -= len;
 288                 }
 289                 break;
 290         case BOP_FLUSH:
 291                 if (bc->bc_ischr) {
 292                         if (ioctl(bc->bc_fd, DIOCGFLUSH))
 293                                 err = errno;
 294                 } else if (fsync(bc->bc_fd))
 295                         err = errno;
 296                 break;
 297         case BOP_DELETE:
 298                 if (!bc->bc_candelete)
 299                         err = EOPNOTSUPP;
 300                 else if (bc->bc_rdonly)
 301                         err = EROFS;
 302                 else if (bc->bc_ischr) {
 303                         arg[0] = br->br_offset;
 304                         arg[1] = br->br_resid;
 305                         if (ioctl(bc->bc_fd, DIOCGDELETE, arg))
 306                                 err = errno;
 307                         else
 308                                 br->br_resid = 0;
 309                 } else
 310                         err = EOPNOTSUPP;
 311                 break;
 312         default:
 313                 err = EINVAL;
 314                 break;
 315         }
 316
 317         be->be_status = BST_DONE;
 318
 319         (*br->br_callback)(br, err);
 320 }
 321
 322 static void *
 323 blockif_thr(void *arg)
 324 {
 325         struct blockif_ctxt *bc;
 326         struct blockif_elem *be;
 327         pthread_t t;
 328         uint8_t *buf;
 329
 330         bc = arg;
 331         if (bc->bc_isgeom)
 332                 buf = malloc(MAXPHYS);
 333         else
 334                 buf = NULL;
 335         t = pthread_self();
 336
 337         pthread_mutex_lock(&bc->bc_mtx);
 338         for (;;) {
 339                 while (blockif_dequeue(bc, t, &be)) {
 340                         pthread_mutex_unlock(&bc->bc_mtx);
 341                         blockif_proc(bc, be, buf);
 342                         pthread_mutex_lock(&bc->bc_mtx);
 343                         blockif_complete(bc, be);
 344                 }
 345                 /* Check ctxt status here to see if exit requested */
 346                 if (bc->bc_closing)
 347                         break;
 348                 pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx);
 349         }
 350         pthread_mutex_unlock(&bc->bc_mtx);
 351
 352         if (buf)
 353                 free(buf);
 354         pthread_exit(NULL);
 355         return (NULL);
 356 }
 357
 358 static void
 359 blockif_sigcont_handler(int signal, enum ev_type type, void *arg)
 360 {
 361         struct blockif_sig_elem *bse;
 362
 363         for (;;) {
 364                 /*
 365                  * Process the entire list even if not intended for
 366                  * this thread.
 367                  */
 368                 do {
 369                         bse = blockif_bse_head;
 370                         if (bse == NULL)
 371                                 return;
 372                 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
 373                                             (uintptr_t)bse,
 374                                             (uintptr_t)bse->bse_next));
 375
 376                 pthread_mutex_lock(&bse->bse_mtx);
 377                 bse->bse_pending = 0;
 378                 pthread_cond_signal(&bse->bse_cond);
 379                 pthread_mutex_unlock(&bse->bse_mtx);
 380         }
 381 }
 382
 383 static void
 384 blockif_init(void)
 385 {
 386         mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL);
 387         (void) signal(SIGCONT, SIG_IGN);
 388 }
 389
 390 struct blockif_ctxt *
 391 blockif_open(const char *optstr, const char *ident)
 392 {
 393         char tname[MAXCOMLEN + 1];
 394         char name[MAXPATHLEN];
 395         char *nopt, *xopts, *cp;
 396         struct blockif_ctxt *bc;
 397         struct stat sbuf;
 398         struct diocgattr_arg arg;
 399         off_t size, psectsz, psectoff;
 400         int extra, fd, i, sectsz;
 401         int nocache, sync, ro, candelete, geom, ssopt, pssopt;
 402
 403         pthread_once(&blockif_once, blockif_init);
 404
 405         fd = -1;
 406         ssopt = 0;
 407         nocache = 0;
 408         sync = 0;
 409         ro = 0;
 410
 411         /*
 412          * The first element in the optstring is always a pathname.
 413          * Optional elements follow
 414          */
 415         nopt = xopts = strdup(optstr);
 416         while (xopts != NULL) {
 417                 cp = strsep(&xopts, ",");
 418                 if (cp == nopt)         /* file or device pathname */
 419                         continue;
 420                 else if (!strcmp(cp, "nocache"))
 421                         nocache = 1;
 422                 else if (!strcmp(cp, "sync") || !strcmp(cp, "direct"))
 423                         sync = 1;
 424                 else if (!strcmp(cp, "ro"))
 425                         ro = 1;
 426                 else if (sscanf(cp, "sectorsize=%d/%d", &ssopt, &pssopt) == 2)
 427                         ;
 428                 else if (sscanf(cp, "sectorsize=%d", &ssopt) == 1)
 429                         pssopt = ssopt;
 430                 else {
 431                         fprintf(stderr, "Invalid device option \"%s\"\n", cp);
 432                         goto err;
 433                 }
 434         }
 435
 436         extra = 0;
 437         if (nocache)
 438                 extra |= O_DIRECT;
 439         if (sync)
 440                 extra |= O_SYNC;
 441
 442         fd = open(nopt, (ro ? O_RDONLY : O_RDWR) | extra);
 443         if (fd < 0 && !ro) {
 444                 /* Attempt a r/w fail with a r/o open */
 445                 fd = open(nopt, O_RDONLY | extra);
 446                 ro = 1;
 447         }
 448
 449         if (fd < 0) {
 450                 perror("Could not open backing file");
 451                 goto err;
 452         }
 453
 454         if (fstat(fd, &sbuf) < 0) {
 455                 perror("Could not stat backing file");
 456                 goto err;
 457         }
 458
 459         /*
 460          * Deal with raw devices
 461          */
 462         size = sbuf.st_size;
 463         sectsz = DEV_BSIZE;
 464         psectsz = psectoff = 0;
 465         candelete = geom = 0;
 466         if (S_ISCHR(sbuf.st_mode)) {
 467                 if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
 468                     ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
 469                         perror("Could not fetch dev blk/sector size");
 470                         goto err;
 471                 }
 472                 assert(size != 0);
 473                 assert(sectsz != 0);
 474                 if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0)
 475                         ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff);
 476                 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
 477                 arg.len = sizeof(arg.value.i);
 478                 if (ioctl(fd, DIOCGATTR, &arg) == 0)
 479                         candelete = arg.value.i;
 480                 if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0)
 481                         geom = 1;
 482         } else
 483                 psectsz = sbuf.st_blksize;
 484
 485         if (ssopt != 0) {
 486                 if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 ||
 487                     ssopt > pssopt) {
 488                         fprintf(stderr, "Invalid sector size %d/%d\n",
 489                             ssopt, pssopt);
 490                         goto err;
 491                 }
 492
 493                 /*
 494                  * Some backend drivers (e.g. cd0, ada0) require that the I/O
 495                  * size be a multiple of the device's sector size.
 496                  *
 497                  * Validate that the emulated sector size complies with this
 498                  * requirement.
 499                  */
 500                 if (S_ISCHR(sbuf.st_mode)) {
 501                         if (ssopt < sectsz || (ssopt % sectsz) != 0) {
 502                                 fprintf(stderr, "Sector size %d incompatible "
 503                                     "with underlying device sector size %d\n",
 504                                     ssopt, sectsz);
 505                                 goto err;
 506                         }
 507                 }
 508
 509                 sectsz = ssopt;
 510                 psectsz = pssopt;
 511                 psectoff = 0;
 512         }
 513
 514         bc = calloc(1, sizeof(struct blockif_ctxt));
 515         if (bc == NULL) {
 516                 perror("calloc");
 517                 goto err;
 518         }
 519
 520         bc->bc_magic = BLOCKIF_SIG;
 521         bc->bc_fd = fd;
 522         bc->bc_ischr = S_ISCHR(sbuf.st_mode);
 523         bc->bc_isgeom = geom;
 524         bc->bc_candelete = candelete;
 525         bc->bc_rdonly = ro;
 526         bc->bc_size = size;
 527         bc->bc_sectsz = sectsz;
 528         bc->bc_psectsz = psectsz;
 529         bc->bc_psectoff = psectoff;
 530         pthread_mutex_init(&bc->bc_mtx, NULL);
 531         pthread_cond_init(&bc->bc_cond, NULL);
 532         TAILQ_INIT(&bc->bc_freeq);
 533         TAILQ_INIT(&bc->bc_pendq);
 534         TAILQ_INIT(&bc->bc_busyq);
 535         for (i = 0; i < BLOCKIF_MAXREQ; i++) {
 536                 bc->bc_reqs[i].be_status = BST_FREE;
 537                 TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link);
 538         }
 539
 540         for (i = 0; i < BLOCKIF_NUMTHR; i++) {
 541                 pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc);
 542                 snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i);
 543                 pthread_set_name_np(bc->bc_btid[i], tname);
 544         }
 545
 546         return (bc);
 547 err:
 548         if (fd >= 0)
 549                 close(fd);
 550         return (NULL);
 551 }
 552
 553 static int
 554 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq,
 555                 enum blockop op)
 556 {
 557         int err;
 558
 559         err = 0;
 560
 561         pthread_mutex_lock(&bc->bc_mtx);
 562         if (!TAILQ_EMPTY(&bc->bc_freeq)) {
 563                 /*
 564                  * Enqueue and inform the block i/o thread
 565                  * that there is work available
 566                  */
 567                 if (blockif_enqueue(bc, breq, op))
 568                         pthread_cond_signal(&bc->bc_cond);
 569         } else {
 570                 /*
 571                  * Callers are not allowed to enqueue more than
 572                  * the specified blockif queue limit. Return an
 573                  * error to indicate that the queue length has been
 574                  * exceeded.
 575                  */
 576                 err = E2BIG;
 577         }
 578         pthread_mutex_unlock(&bc->bc_mtx);
 579
 580         return (err);
 581 }
 582
 583 int
 584 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq)
 585 {
 586
 587         assert(bc->bc_magic == BLOCKIF_SIG);
 588         return (blockif_request(bc, breq, BOP_READ));
 589 }
 590
 591 int
 592 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq)
 593 {
 594
 595         assert(bc->bc_magic == BLOCKIF_SIG);
 596         return (blockif_request(bc, breq, BOP_WRITE));
 597 }
 598
 599 int
 600 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq)
 601 {
 602
 603         assert(bc->bc_magic == BLOCKIF_SIG);
 604         return (blockif_request(bc, breq, BOP_FLUSH));
 605 }
 606
 607 int
 608 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq)
 609 {
 610
 611         assert(bc->bc_magic == BLOCKIF_SIG);
 612         return (blockif_request(bc, breq, BOP_DELETE));
 613 }
 614
 615 int
 616 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq)
 617 {
 618         struct blockif_elem *be;
 619
 620         assert(bc->bc_magic == BLOCKIF_SIG);
 621
 622         pthread_mutex_lock(&bc->bc_mtx);
 623         /*
 624          * Check pending requests.
 625          */
 626         TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
 627                 if (be->be_req == breq)
 628                         break;
 629         }
 630         if (be != NULL) {
 631                 /*
 632                  * Found it.
 633                  */
 634                 blockif_complete(bc, be);
 635                 pthread_mutex_unlock(&bc->bc_mtx);
 636
 637                 return (0);
 638         }
 639
 640         /*
 641          * Check in-flight requests.
 642          */
 643         TAILQ_FOREACH(be, &bc->bc_busyq, be_link) {
 644                 if (be->be_req == breq)
 645                         break;
 646         }
 647         if (be == NULL) {
 648                 /*
 649                  * Didn't find it.
 650                  */
 651                 pthread_mutex_unlock(&bc->bc_mtx);
 652                 return (EINVAL);
 653         }
 654
 655         /*
 656          * Interrupt the processing thread to force it return
 657          * prematurely via it's normal callback path.
 658          */
 659         while (be->be_status == BST_BUSY) {
 660                 struct blockif_sig_elem bse, *old_head;
 661
 662                 pthread_mutex_init(&bse.bse_mtx, NULL);
 663                 pthread_cond_init(&bse.bse_cond, NULL);
 664
 665                 bse.bse_pending = 1;
 666
 667                 do {
 668                         old_head = blockif_bse_head;
 669                         bse.bse_next = old_head;
 670                 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
 671                                             (uintptr_t)old_head,
 672                                             (uintptr_t)&bse));
 673
 674                 pthread_kill(be->be_tid, SIGCONT);
 675
 676                 pthread_mutex_lock(&bse.bse_mtx);
 677                 while (bse.bse_pending)
 678                         pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx);
 679                 pthread_mutex_unlock(&bse.bse_mtx);
 680         }
 681
 682         pthread_mutex_unlock(&bc->bc_mtx);
 683
 684         /*
 685          * The processing thread has been interrupted.  Since it's not
 686          * clear if the callback has been invoked yet, return EBUSY.
 687          */
 688         return (EBUSY);
 689 }
 690
 691 int
 692 blockif_close(struct blockif_ctxt *bc)
 693 {
 694         void *jval;
 695         int err, i;
 696
 697         err = 0;
 698
 699         assert(bc->bc_magic == BLOCKIF_SIG);
 700
 701         /*
 702          * Stop the block i/o thread
 703          */
 704         pthread_mutex_lock(&bc->bc_mtx);
 705         bc->bc_closing = 1;
 706         pthread_mutex_unlock(&bc->bc_mtx);
 707         pthread_cond_broadcast(&bc->bc_cond);
 708         for (i = 0; i < BLOCKIF_NUMTHR; i++)
 709                 pthread_join(bc->bc_btid[i], &jval);
 710
 711         /* XXX Cancel queued i/o's ??? */
 712
 713         /*
 714          * Release resources
 715          */
 716         bc->bc_magic = 0;
 717         close(bc->bc_fd);
 718         free(bc);
 719
 720         return (0);
 721 }
 722
 723 /*
 724  * Return virtual C/H/S values for a given block. Use the algorithm
 725  * outlined in the VHD specification to calculate values.
 726  */
 727 void
 728 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s)
 729 {
 730         off_t sectors;          /* total sectors of the block dev */
 731         off_t hcyl;             /* cylinders times heads */
 732         uint16_t secpt;         /* sectors per track */
 733         uint8_t heads;
 734
 735         assert(bc->bc_magic == BLOCKIF_SIG);
 736
 737         sectors = bc->bc_size / bc->bc_sectsz;
 738
 739         /* Clamp the size to the largest possible with CHS */
 740         if (sectors > 65535UL*16*255)
 741                 sectors = 65535UL*16*255;
 742
 743         if (sectors >= 65536UL*16*63) {
 744                 secpt = 255;
 745                 heads = 16;
 746                 hcyl = sectors / secpt;
 747         } else {
 748                 secpt = 17;
 749                 hcyl = sectors / secpt;
 750                 heads = (hcyl + 1023) / 1024;
 751
 752                 if (heads < 4)
 753                         heads = 4;
 754
 755                 if (hcyl >= (heads * 1024) || heads > 16) {
 756                         secpt = 31;
 757                         heads = 16;
 758                         hcyl = sectors / secpt;
 759                 }
 760                 if (hcyl >= (heads * 1024)) {
 761                         secpt = 63;
 762                         heads = 16;
 763                         hcyl = sectors / secpt;
 764                 }
 765         }
 766
 767         *c = hcyl / heads;
 768         *h = heads;
 769         *s = secpt;
 770 }
 771
 772 /*
 773  * Accessors
 774  */
 775 off_t
 776 blockif_size(struct blockif_ctxt *bc)
 777 {
 778
 779         assert(bc->bc_magic == BLOCKIF_SIG);
 780         return (bc->bc_size);
 781 }
 782
 783 int
 784 blockif_sectsz(struct blockif_ctxt *bc)
 785 {
 786
 787         assert(bc->bc_magic == BLOCKIF_SIG);
 788         return (bc->bc_sectsz);
 789 }
 790
 791 void
 792 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off)
 793 {
 794
 795         assert(bc->bc_magic == BLOCKIF_SIG);
 796         *size = bc->bc_psectsz;
 797         *off = bc->bc_psectoff;
 798 }
 799
 800 int
 801 blockif_queuesz(struct blockif_ctxt *bc)
 802 {
 803
 804         assert(bc->bc_magic == BLOCKIF_SIG);
 805         return (BLOCKIF_MAXREQ - 1);
 806 }
 807
 808 int
 809 blockif_is_ro(struct blockif_ctxt *bc)
 810 {
 811
 812         assert(bc->bc_magic == BLOCKIF_SIG);
 813         return (bc->bc_rdonly);
 814 }
 815
 816 int
 817 blockif_candelete(struct blockif_ctxt *bc)
 818 {
 819
 820         assert(bc->bc_magic == BLOCKIF_SIG);
 821         return (bc->bc_candelete);
 822 }