]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - usr.sbin/bhyve/block_if.c
Merge llvm, clang, compiler-rt, libc++, libunwind, lld, lldb and openmp
[FreeBSD/FreeBSD.git] / usr.sbin / bhyve / block_if.c
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2013  Peter Grehan <grehan@freebsd.org>
5  * All rights reserved.
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  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33
34 #include <sys/param.h>
35 #ifndef WITHOUT_CAPSICUM
36 #include <sys/capsicum.h>
37 #endif
38 #include <sys/queue.h>
39 #include <sys/errno.h>
40 #include <sys/stat.h>
41 #include <sys/ioctl.h>
42 #include <sys/disk.h>
43
44 #include <assert.h>
45 #ifndef WITHOUT_CAPSICUM
46 #include <capsicum_helpers.h>
47 #endif
48 #include <err.h>
49 #include <fcntl.h>
50 #include <stdio.h>
51 #include <stdlib.h>
52 #include <string.h>
53 #include <pthread.h>
54 #include <pthread_np.h>
55 #include <signal.h>
56 #include <sysexits.h>
57 #include <unistd.h>
58
59 #include <machine/atomic.h>
60
61 #include "bhyverun.h"
62 #include "debug.h"
63 #include "mevent.h"
64 #include "block_if.h"
65
66 #define BLOCKIF_SIG     0xb109b109
67
68 #define BLOCKIF_NUMTHR  8
69 #define BLOCKIF_MAXREQ  (BLOCKIF_RING_MAX + BLOCKIF_NUMTHR)
70
71 enum blockop {
72         BOP_READ,
73         BOP_WRITE,
74         BOP_FLUSH,
75         BOP_DELETE
76 };
77
78 enum blockstat {
79         BST_FREE,
80         BST_BLOCK,
81         BST_PEND,
82         BST_BUSY,
83         BST_DONE
84 };
85
86 struct blockif_elem {
87         TAILQ_ENTRY(blockif_elem) be_link;
88         struct blockif_req  *be_req;
89         enum blockop         be_op;
90         enum blockstat       be_status;
91         pthread_t            be_tid;
92         off_t                be_block;
93 };
94
95 struct blockif_ctxt {
96         int                     bc_magic;
97         int                     bc_fd;
98         int                     bc_ischr;
99         int                     bc_isgeom;
100         int                     bc_candelete;
101         int                     bc_rdonly;
102         off_t                   bc_size;
103         int                     bc_sectsz;
104         int                     bc_psectsz;
105         int                     bc_psectoff;
106         int                     bc_closing;
107         pthread_t               bc_btid[BLOCKIF_NUMTHR];
108         pthread_mutex_t         bc_mtx;
109         pthread_cond_t          bc_cond;
110
111         /* Request elements and free/pending/busy queues */
112         TAILQ_HEAD(, blockif_elem) bc_freeq;       
113         TAILQ_HEAD(, blockif_elem) bc_pendq;
114         TAILQ_HEAD(, blockif_elem) bc_busyq;
115         struct blockif_elem     bc_reqs[BLOCKIF_MAXREQ];
116 };
117
118 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT;
119
120 struct blockif_sig_elem {
121         pthread_mutex_t                 bse_mtx;
122         pthread_cond_t                  bse_cond;
123         int                             bse_pending;
124         struct blockif_sig_elem         *bse_next;
125 };
126
127 static struct blockif_sig_elem *blockif_bse_head;
128
129 static int
130 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq,
131                 enum blockop op)
132 {
133         struct blockif_elem *be, *tbe;
134         off_t off;
135         int i;
136
137         be = TAILQ_FIRST(&bc->bc_freeq);
138         assert(be != NULL);
139         assert(be->be_status == BST_FREE);
140         TAILQ_REMOVE(&bc->bc_freeq, be, be_link);
141         be->be_req = breq;
142         be->be_op = op;
143         switch (op) {
144         case BOP_READ:
145         case BOP_WRITE:
146         case BOP_DELETE:
147                 off = breq->br_offset;
148                 for (i = 0; i < breq->br_iovcnt; i++)
149                         off += breq->br_iov[i].iov_len;
150                 break;
151         default:
152                 off = OFF_MAX;
153         }
154         be->be_block = off;
155         TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
156                 if (tbe->be_block == breq->br_offset)
157                         break;
158         }
159         if (tbe == NULL) {
160                 TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) {
161                         if (tbe->be_block == breq->br_offset)
162                                 break;
163                 }
164         }
165         if (tbe == NULL)
166                 be->be_status = BST_PEND;
167         else
168                 be->be_status = BST_BLOCK;
169         TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link);
170         return (be->be_status == BST_PEND);
171 }
172
173 static int
174 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep)
175 {
176         struct blockif_elem *be;
177
178         TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
179                 if (be->be_status == BST_PEND)
180                         break;
181                 assert(be->be_status == BST_BLOCK);
182         }
183         if (be == NULL)
184                 return (0);
185         TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
186         be->be_status = BST_BUSY;
187         be->be_tid = t;
188         TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link);
189         *bep = be;
190         return (1);
191 }
192
193 static void
194 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be)
195 {
196         struct blockif_elem *tbe;
197
198         if (be->be_status == BST_DONE || be->be_status == BST_BUSY)
199                 TAILQ_REMOVE(&bc->bc_busyq, be, be_link);
200         else
201                 TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
202         TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
203                 if (tbe->be_req->br_offset == be->be_block)
204                         tbe->be_status = BST_PEND;
205         }
206         be->be_tid = 0;
207         be->be_status = BST_FREE;
208         be->be_req = NULL;
209         TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link);
210 }
211
212 static void
213 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf)
214 {
215         struct blockif_req *br;
216         off_t arg[2];
217         ssize_t clen, len, off, boff, voff;
218         int i, err;
219
220         br = be->be_req;
221         if (br->br_iovcnt <= 1)
222                 buf = NULL;
223         err = 0;
224         switch (be->be_op) {
225         case BOP_READ:
226                 if (buf == NULL) {
227                         if ((len = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt,
228                                    br->br_offset)) < 0)
229                                 err = errno;
230                         else
231                                 br->br_resid -= len;
232                         break;
233                 }
234                 i = 0;
235                 off = voff = 0;
236                 while (br->br_resid > 0) {
237                         len = MIN(br->br_resid, MAXPHYS);
238                         if (pread(bc->bc_fd, buf, len, br->br_offset +
239                             off) < 0) {
240                                 err = errno;
241                                 break;
242                         }
243                         boff = 0;
244                         do {
245                                 clen = MIN(len - boff, br->br_iov[i].iov_len -
246                                     voff);
247                                 memcpy(br->br_iov[i].iov_base + voff,
248                                     buf + boff, clen);
249                                 if (clen < br->br_iov[i].iov_len - voff)
250                                         voff += clen;
251                                 else {
252                                         i++;
253                                         voff = 0;
254                                 }
255                                 boff += clen;
256                         } while (boff < len);
257                         off += len;
258                         br->br_resid -= len;
259                 }
260                 break;
261         case BOP_WRITE:
262                 if (bc->bc_rdonly) {
263                         err = EROFS;
264                         break;
265                 }
266                 if (buf == NULL) {
267                         if ((len = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt,
268                                     br->br_offset)) < 0)
269                                 err = errno;
270                         else
271                                 br->br_resid -= len;
272                         break;
273                 }
274                 i = 0;
275                 off = voff = 0;
276                 while (br->br_resid > 0) {
277                         len = MIN(br->br_resid, MAXPHYS);
278                         boff = 0;
279                         do {
280                                 clen = MIN(len - boff, br->br_iov[i].iov_len -
281                                     voff);
282                                 memcpy(buf + boff,
283                                     br->br_iov[i].iov_base + voff, clen);
284                                 if (clen < br->br_iov[i].iov_len - voff)
285                                         voff += clen;
286                                 else {
287                                         i++;
288                                         voff = 0;
289                                 }
290                                 boff += clen;
291                         } while (boff < len);
292                         if (pwrite(bc->bc_fd, buf, len, br->br_offset +
293                             off) < 0) {
294                                 err = errno;
295                                 break;
296                         }
297                         off += len;
298                         br->br_resid -= len;
299                 }
300                 break;
301         case BOP_FLUSH:
302                 if (bc->bc_ischr) {
303                         if (ioctl(bc->bc_fd, DIOCGFLUSH))
304                                 err = errno;
305                 } else if (fsync(bc->bc_fd))
306                         err = errno;
307                 break;
308         case BOP_DELETE:
309                 if (!bc->bc_candelete)
310                         err = EOPNOTSUPP;
311                 else if (bc->bc_rdonly)
312                         err = EROFS;
313                 else if (bc->bc_ischr) {
314                         arg[0] = br->br_offset;
315                         arg[1] = br->br_resid;
316                         if (ioctl(bc->bc_fd, DIOCGDELETE, arg))
317                                 err = errno;
318                         else
319                                 br->br_resid = 0;
320                 } else
321                         err = EOPNOTSUPP;
322                 break;
323         default:
324                 err = EINVAL;
325                 break;
326         }
327
328         be->be_status = BST_DONE;
329
330         (*br->br_callback)(br, err);
331 }
332
333 static void *
334 blockif_thr(void *arg)
335 {
336         struct blockif_ctxt *bc;
337         struct blockif_elem *be;
338         pthread_t t;
339         uint8_t *buf;
340
341         bc = arg;
342         if (bc->bc_isgeom)
343                 buf = malloc(MAXPHYS);
344         else
345                 buf = NULL;
346         t = pthread_self();
347
348         pthread_mutex_lock(&bc->bc_mtx);
349         for (;;) {
350                 while (blockif_dequeue(bc, t, &be)) {
351                         pthread_mutex_unlock(&bc->bc_mtx);
352                         blockif_proc(bc, be, buf);
353                         pthread_mutex_lock(&bc->bc_mtx);
354                         blockif_complete(bc, be);
355                 }
356                 /* Check ctxt status here to see if exit requested */
357                 if (bc->bc_closing)
358                         break;
359                 pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx);
360         }
361         pthread_mutex_unlock(&bc->bc_mtx);
362
363         if (buf)
364                 free(buf);
365         pthread_exit(NULL);
366         return (NULL);
367 }
368
369 static void
370 blockif_sigcont_handler(int signal, enum ev_type type, void *arg)
371 {
372         struct blockif_sig_elem *bse;
373
374         for (;;) {
375                 /*
376                  * Process the entire list even if not intended for
377                  * this thread.
378                  */
379                 do {
380                         bse = blockif_bse_head;
381                         if (bse == NULL)
382                                 return;
383                 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
384                                             (uintptr_t)bse,
385                                             (uintptr_t)bse->bse_next));
386
387                 pthread_mutex_lock(&bse->bse_mtx);
388                 bse->bse_pending = 0;
389                 pthread_cond_signal(&bse->bse_cond);
390                 pthread_mutex_unlock(&bse->bse_mtx);
391         }
392 }
393
394 static void
395 blockif_init(void)
396 {
397         mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL);
398         (void) signal(SIGCONT, SIG_IGN);
399 }
400
401 struct blockif_ctxt *
402 blockif_open(const char *optstr, const char *ident)
403 {
404         char tname[MAXCOMLEN + 1];
405         char name[MAXPATHLEN];
406         char *nopt, *xopts, *cp;
407         struct blockif_ctxt *bc;
408         struct stat sbuf;
409         struct diocgattr_arg arg;
410         off_t size, psectsz, psectoff;
411         int extra, fd, i, sectsz;
412         int nocache, sync, ro, candelete, geom, ssopt, pssopt;
413 #ifndef WITHOUT_CAPSICUM
414         cap_rights_t rights;
415         cap_ioctl_t cmds[] = { DIOCGFLUSH, DIOCGDELETE };
416 #endif
417
418         pthread_once(&blockif_once, blockif_init);
419
420         fd = -1;
421         ssopt = 0;
422         nocache = 0;
423         sync = 0;
424         ro = 0;
425
426         /*
427          * The first element in the optstring is always a pathname.
428          * Optional elements follow
429          */
430         nopt = xopts = strdup(optstr);
431         while (xopts != NULL) {
432                 cp = strsep(&xopts, ",");
433                 if (cp == nopt)         /* file or device pathname */
434                         continue;
435                 else if (!strcmp(cp, "nocache"))
436                         nocache = 1;
437                 else if (!strcmp(cp, "sync") || !strcmp(cp, "direct"))
438                         sync = 1;
439                 else if (!strcmp(cp, "ro"))
440                         ro = 1;
441                 else if (sscanf(cp, "sectorsize=%d/%d", &ssopt, &pssopt) == 2)
442                         ;
443                 else if (sscanf(cp, "sectorsize=%d", &ssopt) == 1)
444                         pssopt = ssopt;
445                 else {
446                         EPRINTLN("Invalid device option \"%s\"", cp);
447                         goto err;
448                 }
449         }
450
451         extra = 0;
452         if (nocache)
453                 extra |= O_DIRECT;
454         if (sync)
455                 extra |= O_SYNC;
456
457         fd = open(nopt, (ro ? O_RDONLY : O_RDWR) | extra);
458         if (fd < 0 && !ro) {
459                 /* Attempt a r/w fail with a r/o open */
460                 fd = open(nopt, O_RDONLY | extra);
461                 ro = 1;
462         }
463
464         if (fd < 0) {
465                 warn("Could not open backing file: %s", nopt);
466                 goto err;
467         }
468
469         if (fstat(fd, &sbuf) < 0) {
470                 warn("Could not stat backing file %s", nopt);
471                 goto err;
472         }
473
474 #ifndef WITHOUT_CAPSICUM
475         cap_rights_init(&rights, CAP_FSYNC, CAP_IOCTL, CAP_READ, CAP_SEEK,
476             CAP_WRITE);
477         if (ro)
478                 cap_rights_clear(&rights, CAP_FSYNC, CAP_WRITE);
479
480         if (caph_rights_limit(fd, &rights) == -1)
481                 errx(EX_OSERR, "Unable to apply rights for sandbox");
482 #endif
483
484         /*
485          * Deal with raw devices
486          */
487         size = sbuf.st_size;
488         sectsz = DEV_BSIZE;
489         psectsz = psectoff = 0;
490         candelete = geom = 0;
491         if (S_ISCHR(sbuf.st_mode)) {
492                 if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
493                     ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
494                         perror("Could not fetch dev blk/sector size");
495                         goto err;
496                 }
497                 assert(size != 0);
498                 assert(sectsz != 0);
499                 if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0)
500                         ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff);
501                 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
502                 arg.len = sizeof(arg.value.i);
503                 if (ioctl(fd, DIOCGATTR, &arg) == 0)
504                         candelete = arg.value.i;
505                 if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0)
506                         geom = 1;
507         } else
508                 psectsz = sbuf.st_blksize;
509
510 #ifndef WITHOUT_CAPSICUM
511         if (caph_ioctls_limit(fd, cmds, nitems(cmds)) == -1)
512                 errx(EX_OSERR, "Unable to apply rights for sandbox");
513 #endif
514
515         if (ssopt != 0) {
516                 if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 ||
517                     ssopt > pssopt) {
518                         EPRINTLN("Invalid sector size %d/%d",
519                             ssopt, pssopt);
520                         goto err;
521                 }
522
523                 /*
524                  * Some backend drivers (e.g. cd0, ada0) require that the I/O
525                  * size be a multiple of the device's sector size.
526                  *
527                  * Validate that the emulated sector size complies with this
528                  * requirement.
529                  */
530                 if (S_ISCHR(sbuf.st_mode)) {
531                         if (ssopt < sectsz || (ssopt % sectsz) != 0) {
532                                 EPRINTLN("Sector size %d incompatible "
533                                     "with underlying device sector size %d",
534                                     ssopt, sectsz);
535                                 goto err;
536                         }
537                 }
538
539                 sectsz = ssopt;
540                 psectsz = pssopt;
541                 psectoff = 0;
542         }
543
544         bc = calloc(1, sizeof(struct blockif_ctxt));
545         if (bc == NULL) {
546                 perror("calloc");
547                 goto err;
548         }
549
550         bc->bc_magic = BLOCKIF_SIG;
551         bc->bc_fd = fd;
552         bc->bc_ischr = S_ISCHR(sbuf.st_mode);
553         bc->bc_isgeom = geom;
554         bc->bc_candelete = candelete;
555         bc->bc_rdonly = ro;
556         bc->bc_size = size;
557         bc->bc_sectsz = sectsz;
558         bc->bc_psectsz = psectsz;
559         bc->bc_psectoff = psectoff;
560         pthread_mutex_init(&bc->bc_mtx, NULL);
561         pthread_cond_init(&bc->bc_cond, NULL);
562         TAILQ_INIT(&bc->bc_freeq);
563         TAILQ_INIT(&bc->bc_pendq);
564         TAILQ_INIT(&bc->bc_busyq);
565         for (i = 0; i < BLOCKIF_MAXREQ; i++) {
566                 bc->bc_reqs[i].be_status = BST_FREE;
567                 TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link);
568         }
569
570         for (i = 0; i < BLOCKIF_NUMTHR; i++) {
571                 pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc);
572                 snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i);
573                 pthread_set_name_np(bc->bc_btid[i], tname);
574         }
575
576         return (bc);
577 err:
578         if (fd >= 0)
579                 close(fd);
580         free(nopt);
581         return (NULL);
582 }
583
584 static int
585 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq,
586                 enum blockop op)
587 {
588         int err;
589
590         err = 0;
591
592         pthread_mutex_lock(&bc->bc_mtx);
593         if (!TAILQ_EMPTY(&bc->bc_freeq)) {
594                 /*
595                  * Enqueue and inform the block i/o thread
596                  * that there is work available
597                  */
598                 if (blockif_enqueue(bc, breq, op))
599                         pthread_cond_signal(&bc->bc_cond);
600         } else {
601                 /*
602                  * Callers are not allowed to enqueue more than
603                  * the specified blockif queue limit. Return an
604                  * error to indicate that the queue length has been
605                  * exceeded.
606                  */
607                 err = E2BIG;
608         }
609         pthread_mutex_unlock(&bc->bc_mtx);
610
611         return (err);
612 }
613
614 int
615 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq)
616 {
617
618         assert(bc->bc_magic == BLOCKIF_SIG);
619         return (blockif_request(bc, breq, BOP_READ));
620 }
621
622 int
623 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq)
624 {
625
626         assert(bc->bc_magic == BLOCKIF_SIG);
627         return (blockif_request(bc, breq, BOP_WRITE));
628 }
629
630 int
631 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq)
632 {
633
634         assert(bc->bc_magic == BLOCKIF_SIG);
635         return (blockif_request(bc, breq, BOP_FLUSH));
636 }
637
638 int
639 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq)
640 {
641
642         assert(bc->bc_magic == BLOCKIF_SIG);
643         return (blockif_request(bc, breq, BOP_DELETE));
644 }
645
646 int
647 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq)
648 {
649         struct blockif_elem *be;
650
651         assert(bc->bc_magic == BLOCKIF_SIG);
652
653         pthread_mutex_lock(&bc->bc_mtx);
654         /*
655          * Check pending requests.
656          */
657         TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
658                 if (be->be_req == breq)
659                         break;
660         }
661         if (be != NULL) {
662                 /*
663                  * Found it.
664                  */
665                 blockif_complete(bc, be);
666                 pthread_mutex_unlock(&bc->bc_mtx);
667
668                 return (0);
669         }
670
671         /*
672          * Check in-flight requests.
673          */
674         TAILQ_FOREACH(be, &bc->bc_busyq, be_link) {
675                 if (be->be_req == breq)
676                         break;
677         }
678         if (be == NULL) {
679                 /*
680                  * Didn't find it.
681                  */
682                 pthread_mutex_unlock(&bc->bc_mtx);
683                 return (EINVAL);
684         }
685
686         /*
687          * Interrupt the processing thread to force it return
688          * prematurely via it's normal callback path.
689          */
690         while (be->be_status == BST_BUSY) {
691                 struct blockif_sig_elem bse, *old_head;
692
693                 pthread_mutex_init(&bse.bse_mtx, NULL);
694                 pthread_cond_init(&bse.bse_cond, NULL);
695
696                 bse.bse_pending = 1;
697
698                 do {
699                         old_head = blockif_bse_head;
700                         bse.bse_next = old_head;
701                 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
702                                             (uintptr_t)old_head,
703                                             (uintptr_t)&bse));
704
705                 pthread_kill(be->be_tid, SIGCONT);
706
707                 pthread_mutex_lock(&bse.bse_mtx);
708                 while (bse.bse_pending)
709                         pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx);
710                 pthread_mutex_unlock(&bse.bse_mtx);
711         }
712
713         pthread_mutex_unlock(&bc->bc_mtx);
714
715         /*
716          * The processing thread has been interrupted.  Since it's not
717          * clear if the callback has been invoked yet, return EBUSY.
718          */
719         return (EBUSY);
720 }
721
722 int
723 blockif_close(struct blockif_ctxt *bc)
724 {
725         void *jval;
726         int i;
727
728         assert(bc->bc_magic == BLOCKIF_SIG);
729
730         /*
731          * Stop the block i/o thread
732          */
733         pthread_mutex_lock(&bc->bc_mtx);
734         bc->bc_closing = 1;
735         pthread_mutex_unlock(&bc->bc_mtx);
736         pthread_cond_broadcast(&bc->bc_cond);
737         for (i = 0; i < BLOCKIF_NUMTHR; i++)
738                 pthread_join(bc->bc_btid[i], &jval);
739
740         /* XXX Cancel queued i/o's ??? */
741
742         /*
743          * Release resources
744          */
745         bc->bc_magic = 0;
746         close(bc->bc_fd);
747         free(bc);
748
749         return (0);
750 }
751
752 /*
753  * Return virtual C/H/S values for a given block. Use the algorithm
754  * outlined in the VHD specification to calculate values.
755  */
756 void
757 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s)
758 {
759         off_t sectors;          /* total sectors of the block dev */
760         off_t hcyl;             /* cylinders times heads */
761         uint16_t secpt;         /* sectors per track */
762         uint8_t heads;
763
764         assert(bc->bc_magic == BLOCKIF_SIG);
765
766         sectors = bc->bc_size / bc->bc_sectsz;
767
768         /* Clamp the size to the largest possible with CHS */
769         if (sectors > 65535UL*16*255)
770                 sectors = 65535UL*16*255;
771
772         if (sectors >= 65536UL*16*63) {
773                 secpt = 255;
774                 heads = 16;
775                 hcyl = sectors / secpt;
776         } else {
777                 secpt = 17;
778                 hcyl = sectors / secpt;
779                 heads = (hcyl + 1023) / 1024;
780
781                 if (heads < 4)
782                         heads = 4;
783
784                 if (hcyl >= (heads * 1024) || heads > 16) {
785                         secpt = 31;
786                         heads = 16;
787                         hcyl = sectors / secpt;
788                 }
789                 if (hcyl >= (heads * 1024)) {
790                         secpt = 63;
791                         heads = 16;
792                         hcyl = sectors / secpt;
793                 }
794         }
795
796         *c = hcyl / heads;
797         *h = heads;
798         *s = secpt;
799 }
800
801 /*
802  * Accessors
803  */
804 off_t
805 blockif_size(struct blockif_ctxt *bc)
806 {
807
808         assert(bc->bc_magic == BLOCKIF_SIG);
809         return (bc->bc_size);
810 }
811
812 int
813 blockif_sectsz(struct blockif_ctxt *bc)
814 {
815
816         assert(bc->bc_magic == BLOCKIF_SIG);
817         return (bc->bc_sectsz);
818 }
819
820 void
821 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off)
822 {
823
824         assert(bc->bc_magic == BLOCKIF_SIG);
825         *size = bc->bc_psectsz;
826         *off = bc->bc_psectoff;
827 }
828
829 int
830 blockif_queuesz(struct blockif_ctxt *bc)
831 {
832
833         assert(bc->bc_magic == BLOCKIF_SIG);
834         return (BLOCKIF_MAXREQ - 1);
835 }
836
837 int
838 blockif_is_ro(struct blockif_ctxt *bc)
839 {
840
841         assert(bc->bc_magic == BLOCKIF_SIG);
842         return (bc->bc_rdonly);
843 }
844
845 int
846 blockif_candelete(struct blockif_ctxt *bc)
847 {
848
849         assert(bc->bc_magic == BLOCKIF_SIG);
850         return (bc->bc_candelete);
851 }