static int
fuse_write_directbackend(struct vnode *vp, struct uio *uio,
struct ucred *cred, struct fuse_filehandle *fufh, int ioflag)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct fuse_write_in *fwi;
struct fuse_dispatcher fdi;
size_t chunksize;
int diff;
int err = 0;
if (uio->uio_resid == 0)
return (0);
if (ioflag & IO_APPEND)
uio_setoffset(uio, fvdat->filesize);
fdisp_init(&fdi, 0);
while (uio->uio_resid > 0) {
chunksize = MIN(uio->uio_resid,
fuse_get_mpdata(vp->v_mount)->max_write);
fdi.iosize = sizeof(*fwi) + chunksize;
fdisp_make_vp(&fdi, FUSE_WRITE, vp, uio->uio_td, cred);
fwi = fdi.indata;
fwi->fh = fufh->fh_id;
fwi->offset = uio->uio_offset;
fwi->size = chunksize;
if ((err = uiomove((char *)fdi.indata + sizeof(*fwi),
chunksize, uio)))
break;
if ((err = fdisp_wait_answ(&fdi)))
break;
diff = chunksize - ((struct fuse_write_out *)fdi.answ)->size;
if (diff < 0) {
err = EINVAL;
break;
}
uio->uio_resid += diff;
uio->uio_offset -= diff;
if (uio->uio_offset > fvdat->filesize)
fuse_vnode_setsize(vp, cred, uio->uio_offset);
}
fdisp_destroy(&fdi);
return (err);
}
static int
fuse_write_biobackend(struct vnode *vp, struct uio *uio,
struct ucred *cred, struct fuse_filehandle *fufh, int ioflag)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct buf *bp;
daddr_t lbn;
int bcount;
int n, on, err = 0;
const int biosize = fuse_iosize(vp);
KASSERT(uio->uio_rw == UIO_WRITE, ("ncl_write mode"));
FS_DEBUG("resid=%zx offset=%jx fsize=%jx\n",
uio->uio_resid, uio->uio_offset, fvdat->filesize);
if (vp->v_type != VREG)
return (EIO);
if (uio->uio_offset < 0)
return (EINVAL);
if (uio->uio_resid == 0)
return (0);
if (ioflag & IO_APPEND)
uio_setoffset(uio, fvdat->filesize);
/*
* Find all of this file's B_NEEDCOMMIT buffers. If our writes
* would exceed the local maximum per-file write commit size when
* combined with those, we must decide whether to flush,
* go synchronous, or return err. We don't bother checking
* IO_UNIT -- we just make all writes atomic anyway, as there's
* no point optimizing for something that really won't ever happen.
*/
do {
if (fuse_isdeadfs(vp)) {
err = ENXIO;
break;
}
lbn = uio->uio_offset / biosize;
on = uio->uio_offset & (biosize - 1);
n = MIN((unsigned)(biosize - on), uio->uio_resid);
FS_DEBUG2G("lbn %ju, on %d, n %d, uio offset %ju, uio resid %zd\n",
(uintmax_t)lbn, on, n,
(uintmax_t)uio->uio_offset, uio->uio_resid);
again:
/*
* Handle direct append and file extension cases, calculate
* unaligned buffer size.
*/
if (uio->uio_offset == fvdat->filesize && n) {
/*
* Get the buffer (in its pre-append state to maintain
* B_CACHE if it was previously set). Resize the
* nfsnode after we have locked the buffer to prevent
* readers from reading garbage.
*/
bcount = on;
FS_DEBUG("getting block from OS, bcount %d\n", bcount);
bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
if (bp != NULL) {
long save;
err = fuse_vnode_setsize(vp, cred,
uio->uio_offset + n);
if (err) {
brelse(bp);
break;
}
save = bp->b_flags & B_CACHE;
bcount += n;
allocbuf(bp, bcount);
bp->b_flags |= save;
}
} else {
/*
* Obtain the locked cache block first, and then
* adjust the file's size as appropriate.
*/
bcount = on + n;
if ((off_t)lbn * biosize + bcount < fvdat->filesize) {
if ((off_t)(lbn + 1) * biosize < fvdat->filesize)
bcount = biosize;
else
bcount = fvdat->filesize -
(off_t)lbn *biosize;
}
FS_DEBUG("getting block from OS, bcount %d\n", bcount);
bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
if (bp && uio->uio_offset + n > fvdat->filesize) {
err = fuse_vnode_setsize(vp, cred,
uio->uio_offset + n);
if (err) {
brelse(bp);
break;
}
}
}
if (!bp) {
err = EINTR;
break;
}
/*
* Issue a READ if B_CACHE is not set. In special-append
* mode, B_CACHE is based on the buffer prior to the write
* op and is typically set, avoiding the read. If a read
* is required in special append mode, the server will
* probably send us a short-read since we extended the file
* on our end, resulting in b_resid == 0 and, thusly,
* B_CACHE getting set.
*
* We can also avoid issuing the read if the write covers
* the entire buffer. We have to make sure the buffer state
* is reasonable in this case since we will not be initiating
* I/O. See the comments in kern/vfs_bio.c's getblk() for
* more information.
*
* B_CACHE may also be set due to the buffer being cached
* normally.
*/
if (on == 0 && n == bcount) {
bp->b_flags |= B_CACHE;
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
}
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_iocmd = BIO_READ;
vfs_busy_pages(bp, 0);
fuse_io_strategy(vp, bp);
if ((err = bp->b_error)) {
brelse(bp);
break;
}
}
if (bp->b_wcred == NOCRED)
bp->b_wcred = crhold(cred);
/*
* If dirtyend exceeds file size, chop it down. This should
* not normally occur but there is an append race where it
* might occur XXX, so we log it.
*
* If the chopping creates a reverse-indexed or degenerate
* situation with dirtyoff/end, we 0 both of them.
*/
if (bp->b_dirtyend > bcount) {
FS_DEBUG("FUSE append race @%lx:%d\n",
(long)bp->b_blkno * biosize,
bp->b_dirtyend - bcount);
bp->b_dirtyend = bcount;
}
if (bp->b_dirtyoff >= bp->b_dirtyend)
bp->b_dirtyoff = bp->b_dirtyend = 0;
/*
* If the new write will leave a contiguous dirty
* area, just update the b_dirtyoff and b_dirtyend,
* otherwise force a write rpc of the old dirty area.
*
* While it is possible to merge discontiguous writes due to
* our having a B_CACHE buffer ( and thus valid read data
* for the hole), we don't because it could lead to
* significant cache coherency problems with multiple clients,
* especially if locking is implemented later on.
*
* as an optimization we could theoretically maintain
* a linked list of discontinuous areas, but we would still
* have to commit them separately so there isn't much
* advantage to it except perhaps a bit of asynchronization.
*/
if (bp->b_dirtyend > 0 &&
(on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
/*
* Yes, we mean it. Write out everything to "storage"
* immediately, without hesitation. (Apart from other
* reasons: the only way to know if a write is valid
* if its actually written out.)
*/
bwrite(bp);
if (bp->b_error == EINTR) {
err = EINTR;
break;
}
goto again;
}
err = uiomove((char *)bp->b_data + on, n, uio);
/*
* Since this block is being modified, it must be written
* again and not just committed. Since write clustering does
* not work for the stage 1 data write, only the stage 2
* commit rpc, we have to clear B_CLUSTEROK as well.
*/
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
if (err) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = err;
brelse(bp);
break;
}
/*
* Only update dirtyoff/dirtyend if not a degenerate
* condition.
*/
if (n) {
if (bp->b_dirtyend > 0) {
bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
bp->b_dirtyend = MAX((on + n), bp->b_dirtyend);
} else {
bp->b_dirtyoff = on;
bp->b_dirtyend = on + n;
}
vfs_bio_set_valid(bp, on, n);
}
err = bwrite(bp);
if (err)
break;
} while (uio->uio_resid > 0 && n > 0);
if (fuse_sync_resize && (fvdat->flag & FN_SIZECHANGE) != 0)
fuse_vnode_savesize(vp, cred);
return (err);
}
/*
struct vnop_getattr_args {
struct vnode *a_vp;
struct vattr *a_vap;
struct ucred *a_cred;
struct thread *a_td;
};
*/
static int
fuse_vnop_getattr(struct vop_getattr_args *ap)
{
struct vnode *vp = ap->a_vp;
struct vattr *vap = ap->a_vap;
struct ucred *cred = ap->a_cred;
struct thread *td = curthread;
struct fuse_vnode_data *fvdat = VTOFUD(vp);
int err = 0;
int dataflags;
struct fuse_dispatcher fdi;
FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
/* Note that we are not bailing out on a dead file system just yet. */
if (!(dataflags & FSESS_INITED)) {
if (!vnode_isvroot(vp)) {
fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
err = ENOTCONN;
debug_printf("fuse_getattr b: returning ENOTCONN\n");
return err;
} else {
goto fake;
}
}
fdisp_init(&fdi, 0);
if ((err = fdisp_simple_putget_vp(&fdi, FUSE_GETATTR, vp, td, cred))) {
if ((err == ENOTCONN) && vnode_isvroot(vp)) {
/* see comment at similar place in fuse_statfs() */
fdisp_destroy(&fdi);
goto fake;
}
if (err == ENOENT) {
fuse_internal_vnode_disappear(vp);
}
goto out;
}
cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
if (vap != VTOVA(vp)) {
memcpy(vap, VTOVA(vp), sizeof(*vap));
}
if (vap->va_type != vnode_vtype(vp)) {
fuse_internal_vnode_disappear(vp);
err = ENOENT;
goto out;
}
if ((fvdat->flag & FN_SIZECHANGE) != 0)
vap->va_size = fvdat->filesize;
if (vnode_isreg(vp) && (fvdat->flag & FN_SIZECHANGE) == 0) {
/*
* This is for those cases when the file size changed without us
* knowing, and we want to catch up.
*/
off_t new_filesize = ((struct fuse_attr_out *)
fdi.answ)->attr.size;
if (fvdat->filesize != new_filesize) {
fuse_vnode_setsize(vp, cred, new_filesize);
}
}
debug_printf("fuse_getattr e: returning 0\n");
out:
fdisp_destroy(&fdi);
return err;
fake:
bzero(vap, sizeof(*vap));
vap->va_type = vnode_vtype(vp);
return 0;
}
/*
struct vnop_setattr_args {
struct vnode *a_vp;
struct vattr *a_vap;
struct ucred *a_cred;
struct thread *a_td;
};
*/
static int
fuse_vnop_setattr(struct vop_setattr_args *ap)
{
struct vnode *vp = ap->a_vp;
struct vattr *vap = ap->a_vap;
struct ucred *cred = ap->a_cred;
struct thread *td = curthread;
struct fuse_dispatcher fdi;
struct fuse_setattr_in *fsai;
struct fuse_access_param facp;
int err = 0;
enum vtype vtyp;
int sizechanged = 0;
uint64_t newsize = 0;
FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
if (fuse_isdeadfs(vp)) {
return ENXIO;
}
fdisp_init(&fdi, sizeof(*fsai));
fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
fsai = fdi.indata;
fsai->valid = 0;
bzero(&facp, sizeof(facp));
facp.xuid = vap->va_uid;
facp.xgid = vap->va_gid;
if (vap->va_uid != (uid_t)VNOVAL) {
facp.facc_flags |= FACCESS_CHOWN;
fsai->uid = vap->va_uid;
fsai->valid |= FATTR_UID;
}
if (vap->va_gid != (gid_t)VNOVAL) {
facp.facc_flags |= FACCESS_CHOWN;
fsai->gid = vap->va_gid;
fsai->valid |= FATTR_GID;
}
if (vap->va_size != VNOVAL) {
struct fuse_filehandle *fufh = NULL;
/*Truncate to a new value. */
fsai->size = vap->va_size;
sizechanged = 1;
newsize = vap->va_size;
fsai->valid |= FATTR_SIZE;
fuse_filehandle_getrw(vp, FUFH_WRONLY, &fufh);
if (fufh) {
fsai->fh = fufh->fh_id;
fsai->valid |= FATTR_FH;
}
}
if (vap->va_atime.tv_sec != VNOVAL) {
fsai->atime = vap->va_atime.tv_sec;
fsai->atimensec = vap->va_atime.tv_nsec;
fsai->valid |= FATTR_ATIME;
}
if (vap->va_mtime.tv_sec != VNOVAL) {
fsai->mtime = vap->va_mtime.tv_sec;
fsai->mtimensec = vap->va_mtime.tv_nsec;
fsai->valid |= FATTR_MTIME;
}
if (vap->va_mode != (mode_t)VNOVAL) {
fsai->mode = vap->va_mode & ALLPERMS;
fsai->valid |= FATTR_MODE;
}
if (!fsai->valid) {
goto out;
}
vtyp = vnode_vtype(vp);
if (fsai->valid & FATTR_SIZE && vtyp == VDIR) {
err = EISDIR;
goto out;
}
if (vfs_isrdonly(vnode_mount(vp)) && (fsai->valid & ~FATTR_SIZE || vtyp == VREG)) {
err = EROFS;
goto out;
}
if (fsai->valid & ~FATTR_SIZE) {
/*err = fuse_internal_access(vp, VADMIN, context, &facp); */
/*XXX */
err = 0;
}
facp.facc_flags &= ~FACCESS_XQUERIES;
if (err && !(fsai->valid & ~(FATTR_ATIME | FATTR_MTIME)) &&
vap->va_vaflags & VA_UTIMES_NULL) {
err = fuse_internal_access(vp, VWRITE, &facp, td, cred);
}
if (err)
goto out;
if ((err = fdisp_wait_answ(&fdi)))
goto out;
vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
if (vnode_vtype(vp) != vtyp) {
if (vnode_vtype(vp) == VNON && vtyp != VNON) {
debug_printf("FUSE: Dang! vnode_vtype is VNON and vtype isn't.\n");
} else {
/*
* STALE vnode, ditch
*
* The vnode has changed its type "behind our back". There's
* nothing really we can do, so let us just force an internal
* revocation and tell the caller to try again, if interested.
*/
fuse_internal_vnode_disappear(vp);
err = EAGAIN;
}
}
if (!err && !sizechanged) {
cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
}
out:
fdisp_destroy(&fdi);
if (!err && sizechanged) {
fuse_vnode_setsize(vp, cred, newsize);
VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
}
return err;
}
