2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2013 Peter Grehan <grehan@freebsd.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #ifndef WITHOUT_CAPSICUM
36 #include <sys/capsicum.h>
38 #include <sys/queue.h>
39 #include <sys/errno.h>
41 #include <sys/ioctl.h>
45 #ifndef WITHOUT_CAPSICUM
46 #include <capsicum_helpers.h>
54 #include <pthread_np.h>
59 #include <machine/atomic.h>
66 #define BLOCKIF_SIG 0xb109b109
68 #define BLOCKIF_NUMTHR 8
69 #define BLOCKIF_MAXREQ (BLOCKIF_RING_MAX + BLOCKIF_NUMTHR)
87 TAILQ_ENTRY(blockif_elem) be_link;
88 struct blockif_req *be_req;
90 enum blockstat be_status;
107 pthread_t bc_btid[BLOCKIF_NUMTHR];
108 pthread_mutex_t bc_mtx;
109 pthread_cond_t bc_cond;
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];
118 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT;
120 struct blockif_sig_elem {
121 pthread_mutex_t bse_mtx;
122 pthread_cond_t bse_cond;
124 struct blockif_sig_elem *bse_next;
127 static struct blockif_sig_elem *blockif_bse_head;
130 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq,
133 struct blockif_elem *be, *tbe;
137 be = TAILQ_FIRST(&bc->bc_freeq);
139 assert(be->be_status == BST_FREE);
140 TAILQ_REMOVE(&bc->bc_freeq, be, be_link);
147 off = breq->br_offset;
148 for (i = 0; i < breq->br_iovcnt; i++)
149 off += breq->br_iov[i].iov_len;
155 TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
156 if (tbe->be_block == breq->br_offset)
160 TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) {
161 if (tbe->be_block == breq->br_offset)
166 be->be_status = BST_PEND;
168 be->be_status = BST_BLOCK;
169 TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link);
170 return (be->be_status == BST_PEND);
174 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep)
176 struct blockif_elem *be;
178 TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
179 if (be->be_status == BST_PEND)
181 assert(be->be_status == BST_BLOCK);
185 TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
186 be->be_status = BST_BUSY;
188 TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link);
194 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be)
196 struct blockif_elem *tbe;
198 if (be->be_status == BST_DONE || be->be_status == BST_BUSY)
199 TAILQ_REMOVE(&bc->bc_busyq, be, be_link);
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;
207 be->be_status = BST_FREE;
209 TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link);
213 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf)
215 struct blockif_req *br;
217 ssize_t clen, len, off, boff, voff;
221 if (br->br_iovcnt <= 1)
227 if ((len = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt,
236 while (br->br_resid > 0) {
237 len = MIN(br->br_resid, MAXPHYS);
238 if (pread(bc->bc_fd, buf, len, br->br_offset +
245 clen = MIN(len - boff, br->br_iov[i].iov_len -
247 memcpy(br->br_iov[i].iov_base + voff,
249 if (clen < br->br_iov[i].iov_len - voff)
256 } while (boff < len);
267 if ((len = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt,
276 while (br->br_resid > 0) {
277 len = MIN(br->br_resid, MAXPHYS);
280 clen = MIN(len - boff, br->br_iov[i].iov_len -
283 br->br_iov[i].iov_base + voff, clen);
284 if (clen < br->br_iov[i].iov_len - voff)
291 } while (boff < len);
292 if (pwrite(bc->bc_fd, buf, len, br->br_offset +
303 if (ioctl(bc->bc_fd, DIOCGFLUSH))
305 } else if (fsync(bc->bc_fd))
309 if (!bc->bc_candelete)
311 else if (bc->bc_rdonly)
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))
328 be->be_status = BST_DONE;
330 (*br->br_callback)(br, err);
334 blockif_thr(void *arg)
336 struct blockif_ctxt *bc;
337 struct blockif_elem *be;
343 buf = malloc(MAXPHYS);
348 pthread_mutex_lock(&bc->bc_mtx);
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);
356 /* Check ctxt status here to see if exit requested */
359 pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx);
361 pthread_mutex_unlock(&bc->bc_mtx);
370 blockif_sigcont_handler(int signal, enum ev_type type, void *arg)
372 struct blockif_sig_elem *bse;
376 * Process the entire list even if not intended for
380 bse = blockif_bse_head;
383 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
385 (uintptr_t)bse->bse_next));
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);
397 mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL);
398 (void) signal(SIGCONT, SIG_IGN);
401 struct blockif_ctxt *
402 blockif_open(const char *optstr, const char *ident)
404 char tname[MAXCOMLEN + 1];
405 char name[MAXPATHLEN];
406 char *nopt, *xopts, *cp;
407 struct blockif_ctxt *bc;
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
415 cap_ioctl_t cmds[] = { DIOCGFLUSH, DIOCGDELETE };
418 pthread_once(&blockif_once, blockif_init);
427 * The first element in the optstring is always a pathname.
428 * Optional elements follow
430 nopt = xopts = strdup(optstr);
431 while (xopts != NULL) {
432 cp = strsep(&xopts, ",");
433 if (cp == nopt) /* file or device pathname */
435 else if (!strcmp(cp, "nocache"))
437 else if (!strcmp(cp, "sync") || !strcmp(cp, "direct"))
439 else if (!strcmp(cp, "ro"))
441 else if (sscanf(cp, "sectorsize=%d/%d", &ssopt, &pssopt) == 2)
443 else if (sscanf(cp, "sectorsize=%d", &ssopt) == 1)
446 EPRINTLN("Invalid device option \"%s\"", cp);
457 fd = open(nopt, (ro ? O_RDONLY : O_RDWR) | extra);
459 /* Attempt a r/w fail with a r/o open */
460 fd = open(nopt, O_RDONLY | extra);
465 warn("Could not open backing file: %s", nopt);
469 if (fstat(fd, &sbuf) < 0) {
470 warn("Could not stat backing file %s", nopt);
474 #ifndef WITHOUT_CAPSICUM
475 cap_rights_init(&rights, CAP_FSYNC, CAP_IOCTL, CAP_READ, CAP_SEEK,
478 cap_rights_clear(&rights, CAP_FSYNC, CAP_WRITE);
480 if (caph_rights_limit(fd, &rights) == -1)
481 errx(EX_OSERR, "Unable to apply rights for sandbox");
485 * Deal with raw devices
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, §sz)) {
494 perror("Could not fetch dev blk/sector size");
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)
508 psectsz = sbuf.st_blksize;
510 #ifndef WITHOUT_CAPSICUM
511 if (caph_ioctls_limit(fd, cmds, nitems(cmds)) == -1)
512 errx(EX_OSERR, "Unable to apply rights for sandbox");
516 if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 ||
518 EPRINTLN("Invalid sector size %d/%d",
524 * Some backend drivers (e.g. cd0, ada0) require that the I/O
525 * size be a multiple of the device's sector size.
527 * Validate that the emulated sector size complies with this
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",
544 bc = calloc(1, sizeof(struct blockif_ctxt));
550 bc->bc_magic = BLOCKIF_SIG;
552 bc->bc_ischr = S_ISCHR(sbuf.st_mode);
553 bc->bc_isgeom = geom;
554 bc->bc_candelete = candelete;
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);
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);
585 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq,
592 pthread_mutex_lock(&bc->bc_mtx);
593 if (!TAILQ_EMPTY(&bc->bc_freeq)) {
595 * Enqueue and inform the block i/o thread
596 * that there is work available
598 if (blockif_enqueue(bc, breq, op))
599 pthread_cond_signal(&bc->bc_cond);
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
609 pthread_mutex_unlock(&bc->bc_mtx);
615 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq)
618 assert(bc->bc_magic == BLOCKIF_SIG);
619 return (blockif_request(bc, breq, BOP_READ));
623 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq)
626 assert(bc->bc_magic == BLOCKIF_SIG);
627 return (blockif_request(bc, breq, BOP_WRITE));
631 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq)
634 assert(bc->bc_magic == BLOCKIF_SIG);
635 return (blockif_request(bc, breq, BOP_FLUSH));
639 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq)
642 assert(bc->bc_magic == BLOCKIF_SIG);
643 return (blockif_request(bc, breq, BOP_DELETE));
647 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq)
649 struct blockif_elem *be;
651 assert(bc->bc_magic == BLOCKIF_SIG);
653 pthread_mutex_lock(&bc->bc_mtx);
655 * Check pending requests.
657 TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
658 if (be->be_req == breq)
665 blockif_complete(bc, be);
666 pthread_mutex_unlock(&bc->bc_mtx);
672 * Check in-flight requests.
674 TAILQ_FOREACH(be, &bc->bc_busyq, be_link) {
675 if (be->be_req == breq)
682 pthread_mutex_unlock(&bc->bc_mtx);
687 * Interrupt the processing thread to force it return
688 * prematurely via it's normal callback path.
690 while (be->be_status == BST_BUSY) {
691 struct blockif_sig_elem bse, *old_head;
693 pthread_mutex_init(&bse.bse_mtx, NULL);
694 pthread_cond_init(&bse.bse_cond, NULL);
699 old_head = blockif_bse_head;
700 bse.bse_next = old_head;
701 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
705 pthread_kill(be->be_tid, SIGCONT);
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);
713 pthread_mutex_unlock(&bc->bc_mtx);
716 * The processing thread has been interrupted. Since it's not
717 * clear if the callback has been invoked yet, return EBUSY.
723 blockif_close(struct blockif_ctxt *bc)
728 assert(bc->bc_magic == BLOCKIF_SIG);
731 * Stop the block i/o thread
733 pthread_mutex_lock(&bc->bc_mtx);
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);
740 /* XXX Cancel queued i/o's ??? */
753 * Return virtual C/H/S values for a given block. Use the algorithm
754 * outlined in the VHD specification to calculate values.
757 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s)
759 off_t sectors; /* total sectors of the block dev */
760 off_t hcyl; /* cylinders times heads */
761 uint16_t secpt; /* sectors per track */
764 assert(bc->bc_magic == BLOCKIF_SIG);
766 sectors = bc->bc_size / bc->bc_sectsz;
768 /* Clamp the size to the largest possible with CHS */
769 if (sectors > 65535UL*16*255)
770 sectors = 65535UL*16*255;
772 if (sectors >= 65536UL*16*63) {
775 hcyl = sectors / secpt;
778 hcyl = sectors / secpt;
779 heads = (hcyl + 1023) / 1024;
784 if (hcyl >= (heads * 1024) || heads > 16) {
787 hcyl = sectors / secpt;
789 if (hcyl >= (heads * 1024)) {
792 hcyl = sectors / secpt;
805 blockif_size(struct blockif_ctxt *bc)
808 assert(bc->bc_magic == BLOCKIF_SIG);
809 return (bc->bc_size);
813 blockif_sectsz(struct blockif_ctxt *bc)
816 assert(bc->bc_magic == BLOCKIF_SIG);
817 return (bc->bc_sectsz);
821 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off)
824 assert(bc->bc_magic == BLOCKIF_SIG);
825 *size = bc->bc_psectsz;
826 *off = bc->bc_psectoff;
830 blockif_queuesz(struct blockif_ctxt *bc)
833 assert(bc->bc_magic == BLOCKIF_SIG);
834 return (BLOCKIF_MAXREQ - 1);
838 blockif_is_ro(struct blockif_ctxt *bc)
841 assert(bc->bc_magic == BLOCKIF_SIG);
842 return (bc->bc_rdonly);
846 blockif_candelete(struct blockif_ctxt *bc)
849 assert(bc->bc_magic == BLOCKIF_SIG);
850 return (bc->bc_candelete);