FUSE: Respect userspace FS "do-not-cache" of file attributes

The FUSE protocol demands that kernel implementations cache user filesystem
file attributes (vattr data) for a maximum period of time in the range of
[0, ULONG_MAX] seconds.  In practice, typical requests are for 0, 1, or 10
seconds; or "a long time" to represent indefinite caching.

Historically, FreeBSD FUSE has ignored this client directive entirely.  This
works fine for local-only filesystems, but causes consistency issues with
multi-writer network filesystems.

For now, respect 0 second cache TTLs and do not cache such metadata.
Non-zero metadata caching TTLs in the range [0.000000001, ULONG_MAX] seconds
are still cached indefinitely, because it is unclear how a userspace
filesystem could do anything sensible with those semantics even if
implemented.

In the future, as an optimization, we should implement notify_inval_entry,
etc, which provide userspace filesystems a way of evicting the kernel cache.

One potentially bogus access to invalid cached attribute data was left in
fuse_io_strategy.  It is restricted behind the undocumented and non-default
"vfs.fuse.fix_broken_io" sysctl or "brokenio" mount option; maybe these are
deadcode and can be eliminated?

Some minor APIs changed to facilitate this:

1. Attribute cache validity is tracked in FUSE inodes ("fuse_vnode_data").

2. cache_attrs() respects the provided TTL and only caches in the FUSE
inode if TTL > 0.  It also grows an "out" argument, which, if non-NULL,
stores the translated fuse_attr (even if not suitable for caching).

3. FUSE VTOVA(vp) returns NULL if the vnode's cache is invalid, to help
avoid programming mistakes.

4. A VOP_LINK check for potential nlink overflow prior to invoking the FUSE
link op was weakened (only performed when we have a valid attr cache).  The
check is racy in a multi-writer network filesystem anyway -- classic TOCTOU.
We have to trust any userspace filesystem that rejects local caching to
account for it correctly.

PR:		230258 (inspired by; does not fix)
This commit is contained in:
Conrad Meyer 2019-02-15 22:49:15 +00:00
parent 447b492eb8
commit 78a7722fbc
6 changed files with 66 additions and 38 deletions

View File

@ -373,7 +373,6 @@ fuse_internal_readdir_processdata(struct uio *uio,
/* remove */
#define INVALIDATE_CACHED_VATTRS_UPON_UNLINK 1
int
fuse_internal_remove(struct vnode *dvp,
struct vnode *vp,
@ -381,15 +380,11 @@ fuse_internal_remove(struct vnode *dvp,
enum fuse_opcode op)
{
struct fuse_dispatcher fdi;
struct fuse_vnode_data *fvdat;
int err;
struct vattr *vap = VTOVA(vp);
#if INVALIDATE_CACHED_VATTRS_UPON_UNLINK
int need_invalidate = 0;
uint64_t target_nlink = 0;
#endif
int err = 0;
err = 0;
fvdat = VTOFUD(vp);
debug_printf("dvp=%p, cnp=%p, op=%d\n", vp, cnp, op);
@ -399,13 +394,6 @@ fuse_internal_remove(struct vnode *dvp,
memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
((char *)fdi.indata)[cnp->cn_namelen] = '\0';
#if INVALIDATE_CACHED_VATTRS_UPON_UNLINK
if (vap->va_nlink > 1) {
need_invalidate = 1;
target_nlink = vap->va_nlink;
}
#endif
err = fdisp_wait_answ(&fdi);
fdisp_destroy(&fdi);
return err;
@ -489,7 +477,7 @@ fuse_internal_newentry_core(struct vnode *dvp,
feo->nodeid, 1);
return err;
}
cache_attrs(*vpp, feo);
cache_attrs(*vpp, feo, NULL);
return err;
}
@ -563,6 +551,7 @@ fuse_internal_vnode_disappear(struct vnode *vp)
ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
fvdat->flag |= FN_REVOKED;
fvdat->valid_attr_cache = false;
cache_purge(vp);
}

View File

@ -200,15 +200,47 @@ fuse_internal_access(struct vnode *vp,
/* attributes */
/*
* Cache FUSE attributes 'fat', with nominal expiration
* 'attr_valid'.'attr_valid_nsec', in attr cache associated with vnode 'vp'.
* Optionally, if argument 'vap' is not NULL, store a copy of the converted
* attributes there as well.
*
* If the nominal attribute cache TTL is zero, do not cache on the 'vp' (but do
* return the result to the caller).
*/
static __inline
void
fuse_internal_attr_fat2vat(struct mount *mp,
fuse_internal_attr_fat2vat(struct vnode *vp,
struct fuse_attr *fat,
uint64_t attr_valid,
uint32_t attr_valid_nsec,
struct vattr *vap)
{
struct mount *mp;
struct fuse_vnode_data *fvdat;
struct vattr *vp_cache_at;
mp = vnode_mount(vp);
fvdat = VTOFUD(vp);
DEBUGX(FUSE_DEBUG_INTERNAL,
"node #%ju, mode 0%o\n", (uintmax_t)fat->ino, fat->mode);
/* Honor explicit do-not-cache requests from user filesystems. */
if (attr_valid == 0 && attr_valid_nsec == 0)
fvdat->valid_attr_cache = false;
else
fvdat->valid_attr_cache = true;
vp_cache_at = VTOVA(vp);
if (vap == NULL && vp_cache_at == NULL)
return;
if (vap == NULL)
vap = vp_cache_at;
vattr_null(vap);
vap->va_fsid = mp->mnt_stat.f_fsid.val[0];
@ -227,21 +259,17 @@ fuse_internal_attr_fat2vat(struct mount *mp,
vap->va_ctime.tv_nsec = fat->ctimensec;
vap->va_blocksize = PAGE_SIZE;
vap->va_type = IFTOVT(fat->mode);
#if (S_BLKSIZE == 512)
/* Optimize this case */
vap->va_bytes = fat->blocks << 9;
#else
vap->va_bytes = fat->blocks * S_BLKSIZE;
#endif
vap->va_flags = 0;
if (vap != vp_cache_at && vp_cache_at != NULL)
memcpy(vp_cache_at, vap, sizeof(*vap));
}
#define cache_attrs(vp, fuse_out) \
fuse_internal_attr_fat2vat(vnode_mount(vp), &(fuse_out)->attr, \
VTOVA(vp));
#define cache_attrs(vp, fuse_out, vap_out) \
fuse_internal_attr_fat2vat((vp), &(fuse_out)->attr, \
(fuse_out)->attr_valid, (fuse_out)->attr_valid_nsec, (vap_out))
/* fsync */

View File

@ -655,6 +655,7 @@ fuse_io_strategy(struct vnode *vp, struct buf *bp)
uiop->uio_offset = ((off_t)bp->b_blkno) * biosize;
error = fuse_read_directbackend(vp, uiop, cred, fufh);
/* XXXCEM: Potentially invalid access to cached_attrs here */
if ((!error && uiop->uio_resid) ||
(fsess_opt_brokenio(vnode_mount(vp)) && error == EIO &&
uiop->uio_offset < fvdat->filesize && fvdat->filesize > 0 &&

View File

@ -147,6 +147,7 @@ fuse_vnode_init(struct vnode *vp, struct fuse_vnode_data *fvdat,
int i;
fvdat->nid = nodeid;
vattr_null(&fvdat->cached_attrs);
if (nodeid == FUSE_ROOT_ID) {
vp->v_vflag |= VV_ROOT;
}

View File

@ -86,6 +86,7 @@ struct fuse_vnode_data {
uint32_t flag;
/** meta **/
bool valid_attr_cache;
struct vattr cached_attrs;
off_t filesize;
uint64_t nlookup;
@ -95,7 +96,9 @@ struct fuse_vnode_data {
#define VTOFUD(vp) \
((struct fuse_vnode_data *)((vp)->v_data))
#define VTOI(vp) (VTOFUD(vp)->nid)
#define VTOVA(vp) (&(VTOFUD(vp)->cached_attrs))
#define VTOVA(vp) \
(VTOFUD(vp)->valid_attr_cache ? \
&(VTOFUD(vp)->cached_attrs) : NULL)
#define VTOILLU(vp) ((uint64_t)(VTOFUD(vp) ? VTOI(vp) : 0))
#define FUSE_NULL_ID 0

View File

@ -518,10 +518,8 @@ fuse_vnop_getattr(struct vop_getattr_args *ap)
}
goto out;
}
cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
if (vap != VTOVA(vp)) {
memcpy(vap, VTOVA(vp), sizeof(*vap));
}
cache_attrs(vp, (struct fuse_attr_out *)fdi.answ, vap);
if (vap->va_type != vnode_vtype(vp)) {
fuse_internal_vnode_disappear(vp);
err = ENOENT;
@ -628,9 +626,15 @@ fuse_vnop_link(struct vop_link_args *ap)
if (vnode_mount(tdvp) != vnode_mount(vp)) {
return EXDEV;
}
if (vap->va_nlink >= FUSE_LINK_MAX) {
/*
* This is a seatbelt check to protect naive userspace filesystems from
* themselves and the limitations of the FUSE IPC protocol. If a
* filesystem does not allow attribute caching, assume it is capable of
* validating that nlink does not overflow.
*/
if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
return EMLINK;
}
fli.oldnodeid = VTOI(vp);
fdisp_init(&fdi, 0);
@ -966,9 +970,11 @@ fuse_vnop_lookup(struct vop_lookup_args *ap)
}
if (op == FUSE_GETATTR) {
cache_attrs(*vpp, (struct fuse_attr_out *)fdi.answ);
cache_attrs(*vpp, (struct fuse_attr_out *)fdi.answ,
NULL);
} else {
cache_attrs(*vpp, (struct fuse_entry_out *)fdi.answ);
cache_attrs(*vpp, (struct fuse_entry_out *)fdi.answ,
NULL);
}
/* Insert name into cache if appropriate. */
@ -1644,7 +1650,7 @@ fuse_vnop_setattr(struct vop_setattr_args *ap)
}
}
if (!err && !sizechanged) {
cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
cache_attrs(vp, (struct fuse_attr_out *)fdi.answ, NULL);
}
out:
fdisp_destroy(&fdi);