diff options
author | mav <mav@FreeBSD.org> | 2015-04-18 20:10:19 +0000 |
---|---|---|
committer | mav <mav@FreeBSD.org> | 2015-04-18 20:10:19 +0000 |
commit | 22312966effcdb0de8b734f30aa3438a628ba497 (patch) | |
tree | 8ca6046216bdf0cb81ec044a06c118576214e084 /usr.sbin | |
parent | ec156468498da600ba8b0fde444b783708e7246d (diff) | |
download | FreeBSD-src-22312966effcdb0de8b734f30aa3438a628ba497.zip FreeBSD-src-22312966effcdb0de8b734f30aa3438a628ba497.tar.gz |
Workaround bhyve virtual disks operation on top of GEOM providers.
GEOM does not support scatter/gather lists in its I/Os. Such requests
are cut in pieces by physio(), that may be problematic, if those pieces
are not multiple of provider's sector size. If such case is detected,
move the data through temporary sequential buffer.
MFC after: 2 weeks
Diffstat (limited to 'usr.sbin')
-rw-r--r-- | usr.sbin/bhyve/block_if.c | 108 | ||||
-rw-r--r-- | usr.sbin/bhyve/block_if.h | 1 | ||||
-rw-r--r-- | usr.sbin/bhyve/pci_ahci.c | 4 | ||||
-rw-r--r-- | usr.sbin/bhyve/pci_virtio_block.c | 5 |
4 files changed, 100 insertions, 18 deletions
diff --git a/usr.sbin/bhyve/block_if.c b/usr.sbin/bhyve/block_if.c index b8b27a6..bcb1617 100644 --- a/usr.sbin/bhyve/block_if.c +++ b/usr.sbin/bhyve/block_if.c @@ -85,6 +85,7 @@ struct blockif_ctxt { int bc_magic; int bc_fd; int bc_ischr; + int bc_isgeom; int bc_candelete; int bc_rdonly; off_t bc_size; @@ -198,27 +199,93 @@ blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be) } static void -blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be) +blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf) { struct blockif_req *br; off_t arg[2]; - int err; + ssize_t clen, len, off, boff, voff; + int i, err; br = be->be_req; + if (br->br_iovcnt <= 1) + buf = NULL; err = 0; - switch (be->be_op) { case BOP_READ: - if (preadv(bc->bc_fd, br->br_iov, br->br_iovcnt, - br->br_offset) < 0) - err = errno; + if (buf == NULL) { + if ((len = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt, + br->br_offset)) < 0) + err = errno; + else + br->br_resid -= len; + break; + } + i = 0; + off = voff = 0; + while (br->br_resid > 0) { + len = MIN(br->br_resid, MAXPHYS); + if (pread(bc->bc_fd, buf, len, br->br_offset + + off) < 0) { + err = errno; + break; + } + boff = 0; + do { + clen = MIN(len - boff, br->br_iov[i].iov_len - + voff); + memcpy(br->br_iov[i].iov_base + voff, + buf + boff, clen); + if (clen < br->br_iov[i].iov_len - voff) + voff += clen; + else { + i++; + voff = 0; + } + boff += clen; + } while (boff < len); + off += len; + br->br_resid -= len; + } break; case BOP_WRITE: - if (bc->bc_rdonly) + if (bc->bc_rdonly) { err = EROFS; - else if (pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt, - br->br_offset) < 0) - err = errno; + break; + } + if (buf == NULL) { + if ((len = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt, + br->br_offset)) < 0) + err = errno; + else + br->br_resid -= len; + break; + } + i = 0; + off = voff = 0; + while (br->br_resid > 0) { + len = MIN(br->br_resid, MAXPHYS); + boff = 0; + do { + clen = MIN(len - boff, br->br_iov[i].iov_len - + voff); + memcpy(buf + boff, + br->br_iov[i].iov_base + voff, clen); + if (clen < br->br_iov[i].iov_len - voff) + voff += clen; + else { + i++; + voff = 0; + } + boff += clen; + } while (boff < len); + if (pwrite(bc->bc_fd, buf, len, br->br_offset + + off) < 0) { + err = errno; + break; + } + off += len; + br->br_resid -= len; + } break; case BOP_FLUSH: if (bc->bc_ischr) { @@ -234,9 +301,11 @@ blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be) err = EROFS; else if (bc->bc_ischr) { arg[0] = br->br_offset; - arg[1] = br->br_iov[0].iov_len; + arg[1] = br->br_resid; if (ioctl(bc->bc_fd, DIOCGDELETE, arg)) err = errno; + else + br->br_resid = 0; } else err = EOPNOTSUPP; break; @@ -256,15 +325,20 @@ blockif_thr(void *arg) struct blockif_ctxt *bc; struct blockif_elem *be; pthread_t t; + uint8_t *buf; bc = arg; + if (bc->bc_isgeom) + buf = malloc(MAXPHYS); + else + buf = NULL; t = pthread_self(); pthread_mutex_lock(&bc->bc_mtx); for (;;) { while (blockif_dequeue(bc, t, &be)) { pthread_mutex_unlock(&bc->bc_mtx); - blockif_proc(bc, be); + blockif_proc(bc, be, buf); pthread_mutex_lock(&bc->bc_mtx); blockif_complete(bc, be); } @@ -275,6 +349,8 @@ blockif_thr(void *arg) } pthread_mutex_unlock(&bc->bc_mtx); + if (buf) + free(buf); pthread_exit(NULL); return (NULL); } @@ -315,13 +391,14 @@ struct blockif_ctxt * blockif_open(const char *optstr, const char *ident) { char tname[MAXCOMLEN + 1]; + char name[MAXPATHLEN]; char *nopt, *xopts; struct blockif_ctxt *bc; struct stat sbuf; struct diocgattr_arg arg; off_t size, psectsz, psectoff; int extra, fd, i, sectsz; - int nocache, sync, ro, candelete; + int nocache, sync, ro, candelete, geom; pthread_once(&blockif_once, blockif_init); @@ -375,7 +452,7 @@ blockif_open(const char *optstr, const char *ident) size = sbuf.st_size; sectsz = DEV_BSIZE; psectsz = psectoff = 0; - candelete = 0; + candelete = geom = 0; if (S_ISCHR(sbuf.st_mode)) { if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 || ioctl(fd, DIOCGSECTORSIZE, §sz)) { @@ -391,6 +468,8 @@ blockif_open(const char *optstr, const char *ident) arg.len = sizeof(arg.value.i); if (ioctl(fd, DIOCGATTR, &arg) == 0) candelete = arg.value.i; + if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0) + geom = 1; } else psectsz = sbuf.st_blksize; @@ -403,6 +482,7 @@ blockif_open(const char *optstr, const char *ident) bc->bc_magic = BLOCKIF_SIG; bc->bc_fd = fd; bc->bc_ischr = S_ISCHR(sbuf.st_mode); + bc->bc_isgeom = geom; bc->bc_candelete = candelete; bc->bc_rdonly = ro; bc->bc_size = size; diff --git a/usr.sbin/bhyve/block_if.h b/usr.sbin/bhyve/block_if.h index 1cb7c80..8e63407 100644 --- a/usr.sbin/bhyve/block_if.h +++ b/usr.sbin/bhyve/block_if.h @@ -45,6 +45,7 @@ struct blockif_req { struct iovec br_iov[BLOCKIF_IOV_MAX]; int br_iovcnt; off_t br_offset; + ssize_t br_resid; void (*br_callback)(struct blockif_req *req, int err); void *br_param; }; diff --git a/usr.sbin/bhyve/pci_ahci.c b/usr.sbin/bhyve/pci_ahci.c index 87a1cea..95cfdd9 100644 --- a/usr.sbin/bhyve/pci_ahci.c +++ b/usr.sbin/bhyve/pci_ahci.c @@ -571,6 +571,7 @@ ahci_build_iov(struct ahci_port *p, struct ahci_ioreq *aior, } breq->br_iovcnt = j; + breq->br_resid = todo; aior->done += todo; aior->more = (aior->done < aior->len && i < prdtl); } @@ -776,8 +777,7 @@ next: breq = &aior->io_req; breq->br_offset = elba * blockif_sectsz(p->bctx); - breq->br_iovcnt = 1; - breq->br_iov[0].iov_len = elen * blockif_sectsz(p->bctx); + breq->br_resid = elen * blockif_sectsz(p->bctx); /* * Mark this command in-flight. diff --git a/usr.sbin/bhyve/pci_virtio_block.c b/usr.sbin/bhyve/pci_virtio_block.c index 967b528..35daa1f 100644 --- a/usr.sbin/bhyve/pci_virtio_block.c +++ b/usr.sbin/bhyve/pci_virtio_block.c @@ -198,7 +198,7 @@ pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq) struct pci_vtblk_ioreq *io; int i, n; int err; - int iolen; + ssize_t iolen; int writeop, type; off_t offset; struct iovec iov[BLOCKIF_IOV_MAX + 2]; @@ -246,8 +246,9 @@ pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq) assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop); iolen += iov[i].iov_len; } + io->io_req.br_resid = iolen; - DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r", + DPRINTF(("virtio-block: %s op, %zd bytes, %d segs, offset %ld\n\r", writeop ? "write" : "read/ident", iolen, i - 1, offset)); switch (type) { |