/*
* purge a vnode from the cache
* At this point the vnode is guaranteed to have no references (vn_count == 0)
* The caller has to make sure that there are no ways someone could
* get a handle (via vn_get) on the vnode (usually done via a mount/vfs lock).
*/
void
vn_purge(
struct vnode *vp,
vmap_t *vmap)
{
vn_trace_entry(vp, "vn_purge", (inst_t *)__return_address);
again:
/*
* Check whether vp has already been reclaimed since our caller
* sampled its version while holding a filesystem cache lock that
* its VOP_RECLAIM function acquires.
*/
VN_LOCK(vp);
if (vp->v_number != vmap->v_number) {
VN_UNLOCK(vp, 0);
return;
}
/*
* If vp is being reclaimed or inactivated, wait until it is inert,
* then proceed. Can't assume that vnode is actually reclaimed
* just because the reclaimed flag is asserted -- a vn_alloc
* reclaim can fail.
*/
if (vp->v_flag & (VINACT | VRECLM)) {
ASSERT(vn_count(vp) == 0);
vp->v_flag |= VWAIT;
sv_wait(vptosync(vp), PINOD, &vp->v_lock, 0);
goto again;
}
/*
* Another process could have raced in and gotten this vnode...
*/
if (vn_count(vp) > 0) {
VN_UNLOCK(vp, 0);
return;
}
XFS_STATS_DEC(vn_active);
vp->v_flag |= VRECLM;
VN_UNLOCK(vp, 0);
/*
* Call VOP_RECLAIM and clean vp. The FSYNC_INVAL flag tells
* vp's filesystem to flush and invalidate all cached resources.
* When vn_reclaim returns, vp should have no private data,
* either in a system cache or attached to v_data.
*/
if (vn_reclaim(vp) != 0)
panic("vn_purge: cannot reclaim");
/*
* Wakeup anyone waiting for vp to be reclaimed.
*/
vn_wakeup(vp);
}
/*
* Clean a vnode of filesystem-specific data and prepare it for reuse.
*/
STATIC int
vn_reclaim(
struct vnode *vp)
{
int error;
XFS_STATS_INC(vn_reclaim);
vn_trace_entry(vp, "vn_reclaim", (inst_t *)__return_address);
/*
* Only make the VOP_RECLAIM call if there are behaviors
* to call.
*/
if (vp->v_fbhv) {
VOP_RECLAIM(vp, error);
if (error)
return -error;
}
ASSERT(vp->v_fbhv == NULL);
VN_LOCK(vp);
vp->v_flag &= (VRECLM|VWAIT);
VN_UNLOCK(vp, 0);
vp->v_type = VNON;
vp->v_fbhv = NULL;
#ifdef XFS_VNODE_TRACE
ktrace_free(vp->v_trace);
vp->v_trace = NULL;
#endif
return 0;
}
/*
* Call VOP_INACTIVE on last reference.
*/
void
vn_rele(
struct vnode *vp)
{
int vcnt;
int cache;
XFS_STATS_INC(vn_rele);
VN_LOCK(vp);
vn_trace_entry(vp, "vn_rele", (inst_t *)__return_address);
vcnt = vn_count(vp);
/*
* Since we always get called from put_inode we know
* that i_count won't be decremented after we
* return.
*/
if (!vcnt) {
/*
* As soon as we turn this on, noone can find us in vn_get
* until we turn off VINACT or VRECLM
*/
vp->v_flag |= VINACT;
VN_UNLOCK(vp, 0);
/*
* Do not make the VOP_INACTIVE call if there
* are no behaviors attached to the vnode to call.
*/
if (vp->v_fbhv)
VOP_INACTIVE(vp, NULL, cache);
VN_LOCK(vp);
if (vp->v_flag & VWAIT)
sv_broadcast(vptosync(vp));
vp->v_flag &= ~(VINACT|VWAIT|VRECLM|VMODIFIED);
}
VN_UNLOCK(vp, 0);
vn_trace_exit(vp, "vn_rele", (inst_t *)__return_address);
}
STATIC void
vn_wakeup(
struct vnode *vp)
{
VN_LOCK(vp);
if (vp->v_flag & VWAIT)
sv_broadcast(vptosync(vp));
vp->v_flag &= ~(VRECLM|VWAIT|VMODIFIED);
VN_UNLOCK(vp, 0);
}
int
vn_wait(
struct vnode *vp)
{
VN_LOCK(vp);
if (vp->v_flag & (VINACT | VRECLM)) {
vp->v_flag |= VWAIT;
sv_wait(vptosync(vp), PINOD, &vp->v_lock, 0);
return 1;
}
VN_UNLOCK(vp, 0);
return 0;
}
/*
* Add a reference to a referenced vnode.
*/
struct vnode *
vn_hold(
struct vnode *vp)
{
struct inode *inode;
XFS_STATS_INC(vn_hold);
VN_LOCK(vp);
inode = igrab(LINVFS_GET_IP(vp));
ASSERT(inode);
VN_UNLOCK(vp, 0);
return vp;
}
/* Try to discard pages, in order to recycle a vcache entry.
*
* We also make some sanity checks: ref count, open count, held locks.
*
* We also do some non-VM-related chores, such as releasing the cred pointer
* (for AIX and Solaris) and releasing the gnode (for AIX).
*
* Locking: afs_xvcache lock is held. It must not be dropped.
*/
int
osi_VM_FlushVCache(struct vcache *avc)
{
int s, code;
vnode_t *vp = &avc->v;
if (avc->vrefCount != 0)
return EBUSY;
if (avc->opens != 0)
return EBUSY;
/*
* Just in case someone is still referring to the vnode we give up
* trying to get rid of this guy.
*/
if (CheckLock(&avc->lock) || LockWaiters(&avc->lock))
return EBUSY;
s = VN_LOCK(vp);
/*
* we just need to avoid the race
* in vn_rele between the ref count going to 0 and VOP_INACTIVE
* finishing up.
* Note that although we checked vcount above, we didn't have the lock
*/
if (vp->v_count > 0 || (vp->v_flag & VINACT)) {
VN_UNLOCK(vp, s);
return EBUSY;
}
VN_UNLOCK(vp, s);
/*
* Since we store on last close and on VOP_INACTIVE
* there should be NO dirty pages
* Note that we hold the xvcache lock the entire time.
*/
AFS_GUNLOCK();
PTOSSVP(vp, (off_t) 0, (off_t) MAXLONG);
AFS_GLOCK();
/* afs_chkpgoob will drop and re-acquire the global lock. */
afs_chkpgoob(vp, 0);
osi_Assert(!VN_GET_PGCNT(vp));
osi_Assert(!AFS_VN_MAPPED(vp));
osi_Assert(!AFS_VN_DIRTY(&avc->v));
#if defined(AFS_SGI65_ENV)
if (vp->v_filocks)
cleanlocks(vp, IGN_PID, 0);
mutex_destroy(&vp->v_filocksem);
#else /* AFS_SGI65_ENV */
if (vp->v_filocksem) {
if (vp->v_filocks)
#ifdef AFS_SGI64_ENV
cleanlocks(vp, &curprocp->p_flid);
#else
cleanlocks(vp, IGN_PID, 0);
#endif
osi_Assert(vp->v_filocks == NULL);
mutex_destroy(vp->v_filocksem);
kmem_free(vp->v_filocksem, sizeof *vp->v_filocksem);
vp->v_filocksem = NULL;
}
#endif /* AFS_SGI65_ENV */
if (avc->vrefCount)
osi_Panic("flushVcache: vm race");
#ifdef AFS_SGI64_ENV
AFS_GUNLOCK();
vnode_pcache_reclaim(vp); /* this can sleep */
vnode_pcache_free(vp);
if (vp->v_op != &Afs_vnodeops) {
VOP_RECLAIM(vp, FSYNC_WAIT, code);
}
AFS_GLOCK();
#ifdef AFS_SGI65_ENV
#ifdef VNODE_TRACING
ktrace_free(vp->v_trace);
#endif /* VNODE_TRACING */
vn_bhv_remove(VN_BHV_HEAD(vp), &(avc->vc_bhv_desc));
vn_bhv_head_destroy(&(vp->v_bh));
destroy_bitlock(&vp->v_pcacheflag);
mutex_destroy(&vp->v_buf_lock);
#else
bhv_remove(VN_BHV_HEAD(vp), &(avc->vc_bhv_desc));
bhv_head_destroy(&(vp->v_bh));
#endif
vp->v_flag = 0; /* debug */
#if defined(DEBUG) && defined(VNODE_INIT_BITLOCK)
destroy_bitlock(&vp->v_flag);
#endif
#ifdef INTR_KTHREADS
AFS_VN_DESTROY_BUF_LOCK(vp);
#endif
#endif /* AFS_SGI64_ENV */
return 0;
}
static int
vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
static vattr_t vattr;
vdev_file_t *vf;
struct vnode *vp;
int error = 0;
struct vnode *rootdir;
dprintf("vdev_file_open %p\n", vd->vdev_tsd);
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
#ifdef _KERNEL
if (vd->vdev_tsd != NULL) {
ASSERT(vd->vdev_reopening);
vf = vd->vdev_tsd;
vnode_getwithvid(vf->vf_vnode, vf->vf_vid);
dprintf("skip to open\n");
goto skip_open;
}
#endif
vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_PUSHPAGE);
/*
* We always open the files from the root of the global zone, even if
* we're in a local zone. If the user has gotten to this point, the
* administrator has already decided that the pool should be available
* to local zone users, so the underlying devices should be as well.
*/
ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
/*
vn_openat(char *pnamep,
enum uio_seg seg,
int filemode,
int createmode,
struct vnode **vpp,
enum create crwhy,
mode_t umask,
struct vnode *startvp)
extern int vn_openat(char *pnamep, enum uio_seg seg, int filemode,
int createmode, struct vnode **vpp, enum create crwhy,
mode_t umask, struct vnode *startvp);
*/
rootdir = getrootdir();
error = vn_openat(vd->vdev_path + 1,
UIO_SYSSPACE,
spa_mode(vd->vdev_spa) | FOFFMAX,
0,
&vp,
0,
0,
rootdir
);
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (error);
}
vf->vf_vnode = vp;
#ifdef _KERNEL
vf->vf_vid = vnode_vid(vp);
dprintf("assigning vid %d\n", vf->vf_vid);
/*
* Make sure it's a regular file.
*/
if (!vnode_isreg(vp)) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
VN_RELE(vf->vf_vnode);
return (ENODEV);
}
#endif
#if _KERNEL
skip_open:
/*
* Determine the physical size of the file.
*/
vattr.va_mask = AT_SIZE;
vn_lock(vf->vf_vnode, LK_SHARED | LK_RETRY);
error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
VN_UNLOCK(vf->vf_vnode);
#endif
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
VN_RELE(vf->vf_vnode);
return (error);
}
*max_psize = *psize = vattr.va_size;
*ashift = SPA_MINBLOCKSHIFT;
VN_RELE(vf->vf_vnode);
return (0);
}
