/*
* Calculate the logical to physical mapping if not done already,
* then call the device strategy routine.
*
* In order to be able to swap to a file, the ext2_bmaparray() operation may not
* deadlock on memory. See ext2_bmap() for details.
*/
static int
ext2_strategy(struct vop_strategy_args *ap)
{
struct buf *bp = ap->a_bp;
struct vnode *vp = ap->a_vp;
struct inode *ip;
struct bufobj *bo;
int32_t blkno;
int error;
ip = VTOI(vp);
if (vp->v_type == VBLK || vp->v_type == VCHR)
panic("ext2_strategy: spec");
if (bp->b_blkno == bp->b_lblkno) {
error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
bp->b_blkno = blkno;
if (error) {
bp->b_error = error;
bp->b_ioflags |= BIO_ERROR;
bufdone(bp);
return (0);
}
if ((long)bp->b_blkno == -1)
vfs_bio_clrbuf(bp);
}
if ((long)bp->b_blkno == -1) {
bufdone(bp);
return (0);
}
bp->b_iooffset = dbtob(bp->b_blkno);
bo = VFSTOEXT2(vp->v_mount)->um_bo;
BO_STRATEGY(bo, bp);
return (0);
}
/*
struct vnop_strategy_args {
struct vnode *a_vp;
struct buf *a_bp;
};
*/
static int
fuse_vnop_strategy(struct vop_strategy_args *ap)
{
struct vnode *vp = ap->a_vp;
struct buf *bp = ap->a_bp;
fuse_trace_printf_vnop();
if (!vp || fuse_isdeadfs(vp)) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = ENXIO;
bufdone(bp);
return ENXIO;
}
if (bp->b_iocmd == BIO_WRITE)
fuse_vnode_refreshsize(vp, NOCRED);
(void)fuse_io_strategy(vp, bp);
/*
* This is a dangerous function. If returns error, that might mean a
* panic. We prefer pretty much anything over being forced to panic
* by a malicious daemon (a demon?). So we just return 0 anyway. You
* should never mind this: this function has its own error
* propagation mechanism via the argument buffer, so
* not-that-melodramatic residents of the call chain still will be
* able to know what to do.
*/
return 0;
}
static int
_xfs_strategy(
struct vop_strategy_args /* {
struct vnode *a_vp;
struct buf *a_bp;
} */ *ap)
{
daddr_t blkno;
struct buf *bp;;
struct bufobj *bo;
struct vnode *vp;
struct xfs_mount *xmp;
int error;
bp = ap->a_bp;
vp = ap->a_vp;
KASSERT(ap->a_vp == ap->a_bp->b_vp, ("%s(%p != %p)",
__func__, ap->a_vp, ap->a_bp->b_vp));
if (bp->b_blkno == bp->b_lblkno) {
error = VOP_BMAP(vp, bp->b_lblkno, NULL, &blkno, NULL, NULL);
bp->b_blkno = blkno;
bp->b_iooffset = (blkno << BBSHIFT);
if (error) {
bp->b_error = error;
bp->b_ioflags |= BIO_ERROR;
bufdone(bp);
return (0);
}
if ((long)bp->b_blkno == -1)
vfs_bio_clrbuf(bp);
}
if ((long)bp->b_blkno == -1) {
bufdone(bp);
return (0);
}
xmp = XFS_VFSTOM(MNTTOVFS(vp->v_mount));
bo = &xmp->m_ddev_targp->specvp->v_bufobj;
bo->bo_ops->bop_strategy(bo, bp);
return (0);
}
int
fuse_io_strategy(struct vnode *vp, struct buf *bp)
{
struct fuse_filehandle *fufh;
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct ucred *cred;
struct uio *uiop;
struct uio uio;
struct iovec io;
int error = 0;
const int biosize = fuse_iosize(vp);
MPASS(vp->v_type == VREG || vp->v_type == VDIR);
MPASS(bp->b_iocmd == BIO_READ || bp->b_iocmd == BIO_WRITE);
FS_DEBUG("inode=%ju offset=%jd resid=%ld\n",
(uintmax_t)VTOI(vp), (intmax_t)(((off_t)bp->b_blkno) * biosize),
bp->b_bcount);
error = fuse_filehandle_getrw(vp,
(bp->b_iocmd == BIO_READ) ? FUFH_RDONLY : FUFH_WRONLY, &fufh);
if (error) {
printf("FUSE: strategy: filehandles are closed\n");
bp->b_ioflags |= BIO_ERROR;
bp->b_error = error;
return (error);
}
cred = bp->b_iocmd == BIO_READ ? bp->b_rcred : bp->b_wcred;
uiop = &uio;
uiop->uio_iov = &io;
uiop->uio_iovcnt = 1;
uiop->uio_segflg = UIO_SYSSPACE;
uiop->uio_td = curthread;
/*
* clear BIO_ERROR and B_INVAL state prior to initiating the I/O. We
* do this here so we do not have to do it in all the code that
* calls us.
*/
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
KASSERT(!(bp->b_flags & B_DONE),
("fuse_io_strategy: bp %p already marked done", bp));
if (bp->b_iocmd == BIO_READ) {
io.iov_len = uiop->uio_resid = bp->b_bcount;
io.iov_base = bp->b_data;
uiop->uio_rw = UIO_READ;
uiop->uio_offset = ((off_t)bp->b_blkno) * biosize;
error = fuse_read_directbackend(vp, uiop, cred, fufh);
if ((!error && uiop->uio_resid) ||
(fsess_opt_brokenio(vnode_mount(vp)) && error == EIO &&
uiop->uio_offset < fvdat->filesize && fvdat->filesize > 0 &&
uiop->uio_offset >= fvdat->cached_attrs.va_size)) {
/*
* If we had a short read with no error, we must have
* hit a file hole. We should zero-fill the remainder.
* This can also occur if the server hits the file EOF.
*
* Holes used to be able to occur due to pending
* writes, but that is not possible any longer.
*/
int nread = bp->b_bcount - uiop->uio_resid;
int left = uiop->uio_resid;
if (error != 0) {
printf("FUSE: Fix broken io: offset %ju, "
" resid %zd, file size %ju/%ju\n",
(uintmax_t)uiop->uio_offset,
uiop->uio_resid, fvdat->filesize,
fvdat->cached_attrs.va_size);
error = 0;
}
if (left > 0)
bzero((char *)bp->b_data + nread, left);
uiop->uio_resid = 0;
}
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = error;
}
} else {
/*
* If we only need to commit, try to commit
*/
if (bp->b_flags & B_NEEDCOMMIT) {
FS_DEBUG("write: B_NEEDCOMMIT flags set\n");
}
/*
* Setup for actual write
*/
if ((off_t)bp->b_blkno * biosize + bp->b_dirtyend >
fvdat->filesize)
bp->b_dirtyend = fvdat->filesize -
(off_t)bp->b_blkno * biosize;
if (bp->b_dirtyend > bp->b_dirtyoff) {
io.iov_len = uiop->uio_resid = bp->b_dirtyend
- bp->b_dirtyoff;
uiop->uio_offset = (off_t)bp->b_blkno * biosize
+ bp->b_dirtyoff;
io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
uiop->uio_rw = UIO_WRITE;
error = fuse_write_directbackend(vp, uiop, cred, fufh, 0);
if (error == EINTR || error == ETIMEDOUT
|| (!error && (bp->b_flags & B_NEEDCOMMIT))) {
bp->b_flags &= ~(B_INVAL | B_NOCACHE);
if ((bp->b_flags & B_PAGING) == 0) {
bdirty(bp);
bp->b_flags &= ~B_DONE;
}
if ((error == EINTR || error == ETIMEDOUT) &&
(bp->b_flags & B_ASYNC) == 0)
bp->b_flags |= B_EINTR;
} else {
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_flags |= B_INVAL;
bp->b_error = error;
}
bp->b_dirtyoff = bp->b_dirtyend = 0;
}
} else {
bp->b_resid = 0;
bufdone(bp);
return (0);
}
}
bp->b_resid = uiop->uio_resid;
bufdone(bp);
return (error);
}
/*
* Do an I/O operation to/from a cache block.
*/
int
smbfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
{
struct smbmount *smp = VFSTOSMBFS(vp->v_mount);
struct smbnode *np = VTOSMB(vp);
struct uio *uiop;
struct iovec io;
struct smb_cred *scred;
int error = 0;
uiop = malloc(sizeof(struct uio), M_SMBFSDATA, M_WAITOK);
uiop->uio_iov = &io;
uiop->uio_iovcnt = 1;
uiop->uio_segflg = UIO_SYSSPACE;
uiop->uio_td = td;
scred = smbfs_malloc_scred();
smb_makescred(scred, td, cr);
if (bp->b_iocmd == BIO_READ) {
io.iov_len = uiop->uio_resid = bp->b_bcount;
io.iov_base = bp->b_data;
uiop->uio_rw = UIO_READ;
switch (vp->v_type) {
case VREG:
uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE;
error = smb_read(smp->sm_share, np->n_fid, uiop, scred);
if (error)
break;
if (uiop->uio_resid) {
int left = uiop->uio_resid;
int nread = bp->b_bcount - left;
if (left > 0)
bzero((char *)bp->b_data + nread, left);
}
break;
default:
printf("smbfs_doio: type %x unexpected\n",vp->v_type);
break;
};
if (error) {
bp->b_error = error;
bp->b_ioflags |= BIO_ERROR;
}
} else { /* write */
if (((bp->b_blkno * DEV_BSIZE) + bp->b_dirtyend) > np->n_size)
bp->b_dirtyend = np->n_size - (bp->b_blkno * DEV_BSIZE);
if (bp->b_dirtyend > bp->b_dirtyoff) {
io.iov_len = uiop->uio_resid = bp->b_dirtyend - bp->b_dirtyoff;
uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff;
io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
uiop->uio_rw = UIO_WRITE;
error = smb_write(smp->sm_share, np->n_fid, uiop, scred);
/*
* For an interrupted write, the buffer is still valid
* and the write hasn't been pushed to the server yet,
* so we can't set BIO_ERROR and report the interruption
* by setting B_EINTR. For the B_ASYNC case, B_EINTR
* is not relevant, so the rpc attempt is essentially
* a noop. For the case of a V3 write rpc not being
* committed to stable storage, the block is still
* dirty and requires either a commit rpc or another
* write rpc with iomode == NFSV3WRITE_FILESYNC before
* the block is reused. This is indicated by setting
* the B_DELWRI and B_NEEDCOMMIT flags.
*/
if (error == EINTR
|| (!error && (bp->b_flags & B_NEEDCOMMIT))) {
int s;
s = splbio();
bp->b_flags &= ~(B_INVAL|B_NOCACHE);
if ((bp->b_flags & B_ASYNC) == 0)
bp->b_flags |= B_EINTR;
if ((bp->b_flags & B_PAGING) == 0) {
bdirty(bp);
bp->b_flags &= ~B_DONE;
}
if ((bp->b_flags & B_ASYNC) == 0)
bp->b_flags |= B_EINTR;
splx(s);
} else {
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = error;
}
bp->b_dirtyoff = bp->b_dirtyend = 0;
}
} else {
bp->b_resid = 0;
bufdone(bp);
free(uiop, M_SMBFSDATA);
smbfs_free_scred(scred);
return 0;
}
}
bp->b_resid = uiop->uio_resid;
bufdone(bp);
free(uiop, M_SMBFSDATA);
smbfs_free_scred(scred);
return error;
}
