/*
* Query domain table for a given domain.
*
* If domain isn't found and addok is set, it is added to AVL trees and
* the zsb->z_fuid_dirty flag will be set to TRUE. It will then be
* necessary for the caller or another thread to detect the dirty table
* and sync out the changes.
*/
int
zfs_fuid_find_by_domain(zfs_sb_t *zsb, const char *domain,
char **retdomain, boolean_t addok)
{
fuid_domain_t searchnode, *findnode;
avl_index_t loc;
krw_t rw = RW_READER;
/*
* If the dummy "nobody" domain then return an index of 0
* to cause the created FUID to be a standard POSIX id
* for the user nobody.
*/
if (domain[0] == '\0') {
if (retdomain)
*retdomain = nulldomain;
return (0);
}
searchnode.f_ksid = ksid_lookupdomain(domain);
if (retdomain)
*retdomain = searchnode.f_ksid->kd_name;
if (!zsb->z_fuid_loaded)
zfs_fuid_init(zsb);
retry:
rw_enter(&zsb->z_fuid_lock, rw);
findnode = avl_find(&zsb->z_fuid_domain, &searchnode, &loc);
if (findnode) {
rw_exit(&zsb->z_fuid_lock);
ksiddomain_rele(searchnode.f_ksid);
return (findnode->f_idx);
} else if (addok) {
fuid_domain_t *domnode;
uint64_t retidx;
if (rw == RW_READER && !rw_tryupgrade(&zsb->z_fuid_lock)) {
rw_exit(&zsb->z_fuid_lock);
rw = RW_WRITER;
goto retry;
}
domnode = kmem_alloc(sizeof (fuid_domain_t), KM_SLEEP);
domnode->f_ksid = searchnode.f_ksid;
retidx = domnode->f_idx = avl_numnodes(&zsb->z_fuid_idx) + 1;
avl_add(&zsb->z_fuid_domain, domnode);
avl_add(&zsb->z_fuid_idx, domnode);
zsb->z_fuid_dirty = B_TRUE;
rw_exit(&zsb->z_fuid_lock);
return (retidx);
} else {
rw_exit(&zsb->z_fuid_lock);
return (-1);
}
}
/*
* Build directory vnodes based on the profile and the global
* dev instance.
*/
void
prof_filldir(sdev_node_t *ddv)
{
sdev_node_t *gdir;
ASSERT(RW_READ_HELD(&ddv->sdev_contents));
if (!prof_dev_needupdate(ddv)) {
ASSERT(RW_READ_HELD(&ddv->sdev_contents));
return;
}
/*
* Upgrade to writer lock
*/
if (rw_tryupgrade(&ddv->sdev_contents) == 0) {
/*
* We need to drop the read lock and re-acquire it as a
* write lock. While we do this the condition may change so we
* need to re-check condition
*/
rw_exit(&ddv->sdev_contents);
rw_enter(&ddv->sdev_contents, RW_WRITER);
if (!prof_dev_needupdate(ddv)) {
/* Downgrade back to the read lock before returning */
rw_downgrade(&ddv->sdev_contents);
return;
}
}
/* At this point we should have a write lock */
ASSERT(RW_WRITE_HELD(&ddv->sdev_contents));
sdcmn_err10(("devtree_gen (%s): %ld -> %ld\n",
ddv->sdev_path, ddv->sdev_devtree_gen, devtree_gen));
gdir = ddv->sdev_origin;
if (gdir != NULL)
sdcmn_err10(("sdev_dir_gen (%s): %ld -> %ld\n",
ddv->sdev_path, ddv->sdev_ldir_gen,
gdir->sdev_gdir_gen));
/* update flags and generation number so next filldir is quick */
if ((ddv->sdev_flags & SDEV_BUILD) == SDEV_BUILD) {
ddv->sdev_flags &= ~SDEV_BUILD;
}
ddv->sdev_devtree_gen = devtree_gen;
if (gdir != NULL)
ddv->sdev_ldir_gen = gdir->sdev_gdir_gen;
prof_make_symlinks(ddv);
prof_make_maps(ddv);
prof_make_names(ddv);
rw_downgrade(&ddv->sdev_contents);
}
static int
zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
{
if (RW_WRITE_HELD(&zap->zap_rwlock))
return (1);
if (rw_tryupgrade(&zap->zap_rwlock)) {
dmu_buf_will_dirty(zap->zap_dbuf, tx);
return (1);
}
return (0);
}
static int
splat_rwlock_test6(struct file *file, void *arg)
{
rw_priv_t *rwp;
int rc = -EINVAL;
rwp = (rw_priv_t *)kmalloc(sizeof(*rwp), GFP_KERNEL);
if (rwp == NULL)
return -ENOMEM;
splat_init_rw_priv(rwp, file);
rw_enter(&rwp->rw_rwlock, RW_READER);
if (!RW_READ_HELD(&rwp->rw_rwlock)) {
splat_vprint(file, SPLAT_RWLOCK_TEST6_NAME,
"rwlock should be read lock: %d\n",
RW_READ_HELD(&rwp->rw_rwlock));
goto out;
}
#if defined(CONFIG_RWSEM_GENERIC_SPINLOCK)
/* With one reader upgrade should never fail */
rc = rw_tryupgrade(&rwp->rw_rwlock);
if (!rc) {
splat_vprint(file, SPLAT_RWLOCK_TEST6_NAME,
"rwlock contended preventing upgrade: %d\n",
RW_READ_HELD(&rwp->rw_rwlock));
goto out;
}
if (RW_READ_HELD(&rwp->rw_rwlock) || !RW_WRITE_HELD(&rwp->rw_rwlock)) {
splat_vprint(file, SPLAT_RWLOCK_TEST6_NAME, "rwlock should "
"have 0 (not %d) reader and 1 (not %d) writer\n",
RW_READ_HELD(&rwp->rw_rwlock),
RW_WRITE_HELD(&rwp->rw_rwlock));
goto out;
}
rc = 0;
splat_vprint(file, SPLAT_RWLOCK_TEST6_NAME, "%s",
"rwlock properly upgraded\n");
#else
rc = 0;
splat_vprint(file, SPLAT_RWLOCK_TEST6_NAME, "%s",
"rw_tryupgrade() is disabled for this arch\n");
#endif
out:
rw_exit(&rwp->rw_rwlock);
rw_destroy(&rwp->rw_rwlock);
kfree(rwp);
return rc;
}
/*
* Clean pts sdev_nodes that are no longer valid.
*/
static void
devpts_prunedir(struct sdev_node *ddv)
{
struct vnode *vp;
struct sdev_node *dv, *next = NULL;
int (*vtor)(struct sdev_node *) = NULL;
ASSERT(ddv->sdev_flags & SDEV_VTOR);
vtor = (int (*)(struct sdev_node *))sdev_get_vtor(ddv);
ASSERT(vtor);
if (rw_tryupgrade(&ddv->sdev_contents) == NULL) {
rw_exit(&ddv->sdev_contents);
rw_enter(&ddv->sdev_contents, RW_WRITER);
}
for (dv = ddv->sdev_dot; dv; dv = next) {
next = dv->sdev_next;
/* skip stale nodes */
if (dv->sdev_flags & SDEV_STALE)
continue;
/* validate and prune only ready nodes */
if (dv->sdev_state != SDEV_READY)
continue;
switch (vtor(dv)) {
case SDEV_VTOR_VALID:
case SDEV_VTOR_SKIP:
continue;
case SDEV_VTOR_INVALID:
sdcmn_err7(("prunedir: destroy invalid "
"node: %s(%p)\n", dv->sdev_name, (void *)dv));
break;
}
vp = SDEVTOV(dv);
if (vp->v_count > 0)
continue;
SDEV_HOLD(dv);
/* remove the cache node */
(void) sdev_cache_update(ddv, &dv, dv->sdev_name,
SDEV_CACHE_DELETE);
}
rw_downgrade(&ddv->sdev_contents);
}
/*
* This is the predictive prefetch entry point. It associates dnode access
* specified with blkid and nblks arguments with prefetch stream, predicts
* further accesses based on that stats and initiates speculative prefetch.
* fetch_data argument specifies whether actual data blocks should be fetched:
* FALSE -- prefetch only indirect blocks for predicted data blocks;
* TRUE -- prefetch predicted data blocks plus following indirect blocks.
*/
void
dmu_zfetch(zfetch_t *zf, uint64_t blkid, uint64_t nblks, boolean_t fetch_data)
{
zstream_t *zs;
int64_t pf_start, ipf_start, ipf_istart, ipf_iend;
int64_t pf_ahead_blks, max_blks;
int epbs, max_dist_blks, pf_nblks, ipf_nblks;
uint64_t end_of_access_blkid = blkid + nblks;
if (zfs_prefetch_disable)
return;
/*
* As a fast path for small (single-block) files, ignore access
* to the first block.
*/
if (blkid == 0)
return;
rw_enter(&zf->zf_rwlock, RW_READER);
for (zs = list_head(&zf->zf_stream); zs != NULL;
zs = list_next(&zf->zf_stream, zs)) {
if (blkid == zs->zs_blkid) {
mutex_enter(&zs->zs_lock);
/*
* zs_blkid could have changed before we
* acquired zs_lock; re-check them here.
*/
if (blkid != zs->zs_blkid) {
mutex_exit(&zs->zs_lock);
continue;
}
break;
}
}
if (zs == NULL) {
/*
* This access is not part of any existing stream. Create
* a new stream for it.
*/
ZFETCHSTAT_BUMP(zfetchstat_misses);
if (rw_tryupgrade(&zf->zf_rwlock))
dmu_zfetch_stream_create(zf, end_of_access_blkid);
rw_exit(&zf->zf_rwlock);
return;
}
/*
* This access was to a block that we issued a prefetch for on
* behalf of this stream. Issue further prefetches for this stream.
*
* Normally, we start prefetching where we stopped
* prefetching last (zs_pf_blkid). But when we get our first
* hit on this stream, zs_pf_blkid == zs_blkid, we don't
* want to prefetch the block we just accessed. In this case,
* start just after the block we just accessed.
*/
pf_start = MAX(zs->zs_pf_blkid, end_of_access_blkid);
/*
* Double our amount of prefetched data, but don't let the
* prefetch get further ahead than zfetch_max_distance.
*/
if (fetch_data) {
max_dist_blks =
zfetch_max_distance >> zf->zf_dnode->dn_datablkshift;
/*
* Previously, we were (zs_pf_blkid - blkid) ahead. We
* want to now be double that, so read that amount again,
* plus the amount we are catching up by (i.e. the amount
* read just now).
*/
pf_ahead_blks = zs->zs_pf_blkid - blkid + nblks;
max_blks = max_dist_blks - (pf_start - end_of_access_blkid);
pf_nblks = MIN(pf_ahead_blks, max_blks);
} else {
int
rumpuser_rw_tryupgrade(struct rumpuser_rw *rw)
{
return rw_tryupgrade(rw);
}
/*
* Find the policy that matches this device.
*/
static devplcy_t *
match_policy(devplcyent_t *de, dev_t dev, vtype_t spec)
{
char *mname = NULL;
minor_t min = getminor(dev);
for (; de != NULL; de = de->dpe_next) {
if (de->dpe_flags & DPE_ALLMINOR)
break;
if (de->dpe_flags & DPE_EXPANDED) {
if (min >= de->dpe_lomin && min <= de->dpe_himin &&
spec == de->dpe_spec) {
break;
} else {
continue;
}
}
/*
* We now need the minor name to match string or
* simle regexp. Could we use csp->s_dip and not
* allocate a string here?
*/
if (mname == NULL &&
ddi_lyr_get_minor_name(dev, spec, &mname) != DDI_SUCCESS)
/* mname can be set after the function fails */
return (dfltpolicy);
/* Simple wildcard, with only one ``*'' */
if (de->dpe_flags & DPE_WILDC) {
int plen = de->dpe_len - 1;
int slen = strlen(mname);
char *pp = de->dpe_expr;
char *sp = mname;
/* string must be at least as long as pattern w/o '*' */
if (slen < plen - 1)
continue;
/* skip prefix */
while (*pp == *sp && *pp != '\0') {
pp++;
sp++;
}
/* matched single '*' */
if (*pp == '\0')
if (*sp == '\0')
break;
else
continue;
if (*pp != '*')
continue;
pp++;
/*
* skip characters matched by '*': difference of
* length of s and length of pattern sans '*'
*/
sp += slen - (plen - 1);
if (strcmp(pp, sp) == 0) /* match! */
break;
} else if (strcmp(de->dpe_expr, mname) == 0) {
/* Store minor number, if no contention */
if (rw_tryupgrade(&policyrw)) {
de->dpe_lomin = de->dpe_himin = min;
de->dpe_spec = spec;
de->dpe_flags |= DPE_EXPANDED;
}
break;
}
}
if (mname != NULL)
kmem_free(mname, strlen(mname) + 1);
return (de != NULL ? de->dpe_plcy : dfltpolicy);
}
static int
nvpflush_one(nvfd_t *nvfd)
{
int rval = DDI_SUCCESS;
nvlist_t *nvl;
rw_enter(&nvfd->nvf_lock, RW_READER);
if (!NVF_IS_DIRTY(nvfd) || NVF_IS_READONLY(nvfd)) {
rw_exit(&nvfd->nvf_lock);
return (DDI_SUCCESS);
}
if (rw_tryupgrade(&nvfd->nvf_lock) == 0) {
KFIOERR((CE_CONT, "nvpflush: "
"%s rw upgrade failed\n", nvfd->nvf_name));
rw_exit(&nvfd->nvf_lock);
return (DDI_FAILURE);
}
if (((nvfd->nvf_nvp2nvl)(nvfd, &nvl)) != DDI_SUCCESS) {
KFIOERR((CE_CONT, "nvpflush: "
"%s nvlist construction failed\n", nvfd->nvf_name));
rw_exit(&nvfd->nvf_lock);
return (DDI_FAILURE);
}
NVF_CLEAR_DIRTY(nvfd);
nvfd->nvf_flags |= NVF_FLUSHING;
rw_exit(&nvfd->nvf_lock);
rval = e_fwrite_nvlist(nvfd, nvl);
nvlist_free(nvl);
rw_enter(&nvfd->nvf_lock, RW_WRITER);
nvfd->nvf_flags &= ~NVF_FLUSHING;
if (rval == DDI_FAILURE) {
if (NVF_IS_READONLY(nvfd)) {
rval = DDI_SUCCESS;
nvfd->nvf_flags &= ~(NVF_ERROR | NVF_DIRTY);
} else if ((nvfd->nvf_flags & NVF_ERROR) == 0) {
cmn_err(CE_CONT,
"%s: updated failed\n", nvfd->nvf_name);
nvfd->nvf_flags |= NVF_ERROR | NVF_DIRTY;
}
} else {
if (nvfd->nvf_flags & NVF_CREATE_MSG) {
cmn_err(CE_CONT, "!Creating %s\n", nvfd->nvf_name);
nvfd->nvf_flags &= ~NVF_CREATE_MSG;
}
if (nvfd->nvf_flags & NVF_REBUILD_MSG) {
cmn_err(CE_CONT, "!Rebuilding %s\n", nvfd->nvf_name);
nvfd->nvf_flags &= ~NVF_REBUILD_MSG;
}
if (nvfd->nvf_flags & NVF_ERROR) {
cmn_err(CE_CONT,
"%s: update now ok\n", nvfd->nvf_name);
nvfd->nvf_flags &= ~NVF_ERROR;
}
/*
* The file may need to be flushed again if the cached
* data was touched while writing the earlier contents.
*/
if (NVF_IS_DIRTY(nvfd))
rval = DDI_FAILURE;
}
rw_exit(&nvfd->nvf_lock);
return (rval);
}
int
sam_refresh_shared_reader_ino(
sam_node_t *ip, /* Pointer to the inode */
boolean_t writelock, /* Inode WRITER lock held, */
/* otherwise READER lock held. */
cred_t *credp) /* credentials. */
{
sam_id_t id;
struct sam_perm_inode *permip;
buf_t *bp;
int refresh = 0;
int error;
if ((ip->updtime + ip->mp->mt.fi_invalid) > SAM_SECOND()) {
return (0);
}
if (!writelock) {
/*
* Acquire inode lock before buffer lock. Recheck the update
* time.
*/
if (!rw_tryupgrade(&ip->inode_rwl)) {
RW_UNLOCK_OS(&ip->inode_rwl, RW_READER);
RW_LOCK_OS(&ip->inode_rwl, RW_WRITER);
if ((ip->updtime + ip->mp->mt.fi_invalid) >
SAM_SECOND()) {
error = 0;
goto out;
}
}
}
id = ip->di.id;
if ((error = sam_read_ino(ip->mp, id.ino, &bp, &permip))) {
goto out;
}
if ((permip->di.mode != 0) && (permip->di.id.ino == ip->di.id.ino) &&
(permip->di.id.gen == ip->di.id.gen)) {
if ((permip->di.modify_time.tv_sec !=
ip->di.modify_time.tv_sec) ||
(permip->di.modify_time.tv_nsec !=
ip->di.modify_time.tv_nsec)||
(permip->di.change_time.tv_sec !=
ip->di.change_time.tv_sec) ||
(permip->di.change_time.tv_nsec !=
ip->di.change_time.tv_nsec)||
(permip->di.residence_time != ip->di.residence_time) ||
(permip->di.rm.size != ip->di.rm.size) ||
(permip->di.mode != ip->di.mode)) {
refresh = 1;
} else {
ip->di.uid = permip->di.uid;
ip->di.gid = permip->di.gid;
}
} else {
refresh = 1;
error = ENOENT; /* This inode has been removed */
}
if (refresh) {
vnode_t *vp = SAM_ITOV(ip);
/*
* If a refresh is needed on a directory inode,
* invalidate associated dnlc entries.
*/
if (S_ISDIR(ip->di.mode)) {
sam_invalidate_dnlc(vp);
}
/*
* Move shared_writer's inode copy into inode. Set size
* and invalidate pages. Set shared_reader update time.
*/
ip->di = permip->di; /* Move disk ino to incore ino */
ip->di2 = permip->di2;
brelse(bp);
vp->v_type = IFTOVT(S_ISREQ(ip->di.mode) ?
S_IFREG : ip->di.mode);
sam_set_size(ip);
(void) VOP_PUTPAGE_OS(vp, 0, 0, B_INVAL, credp, NULL);
if (ip->di.status.b.acl) {
(void) sam_acl_inactive(ip);
error = sam_get_acl(ip, &ip->aclp);
}
ip->updtime = SAM_SECOND();
} else {
ip->updtime = SAM_SECOND();
brelse(bp);
}
out:
if (!writelock) {
rw_downgrade(&ip->inode_rwl);
}
return (error);
}
static int
auto_lookup(
vnode_t *dvp,
char *nm,
vnode_t **vpp,
pathname_t *pnp,
int flags,
vnode_t *rdir,
cred_t *cred,
caller_context_t *ct,
int *direntflags,
pathname_t *realpnp)
{
int error = 0;
vnode_t *newvp = NULL;
vfs_t *vfsp;
fninfo_t *dfnip;
fnnode_t *dfnp = NULL;
fnnode_t *fnp = NULL;
char *searchnm;
int operation; /* either AUTOFS_LOOKUP or AUTOFS_MOUNT */
dfnip = vfstofni(dvp->v_vfsp);
AUTOFS_DPRINT((3, "auto_lookup: dvp=%p (%s) name=%s\n",
(void *)dvp, dfnip->fi_map, nm));
if (nm[0] == 0) {
VN_HOLD(dvp);
*vpp = dvp;
return (0);
}
if (error = VOP_ACCESS(dvp, VEXEC, 0, cred, ct))
return (error);
if (nm[0] == '.' && nm[1] == 0) {
VN_HOLD(dvp);
*vpp = dvp;
return (0);
}
if (nm[0] == '.' && nm[1] == '.' && nm[2] == 0) {
fnnode_t *pdfnp;
pdfnp = (vntofn(dvp))->fn_parent;
ASSERT(pdfnp != NULL);
/*
* Since it is legitimate to have the VROOT flag set for the
* subdirectories of the indirect map in autofs filesystem,
* rootfnnodep is checked against fnnode of dvp instead of
* just checking whether VROOT flag is set in dvp
*/
if (pdfnp == pdfnp->fn_globals->fng_rootfnnodep) {
vnode_t *vp;
vfs_rlock_wait(dvp->v_vfsp);
if (dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED) {
vfs_unlock(dvp->v_vfsp);
return (EIO);
}
vp = dvp->v_vfsp->vfs_vnodecovered;
VN_HOLD(vp);
vfs_unlock(dvp->v_vfsp);
error = VOP_LOOKUP(vp, nm, vpp, pnp, flags, rdir, cred,
ct, direntflags, realpnp);
VN_RELE(vp);
return (error);
} else {
*vpp = fntovn(pdfnp);
VN_HOLD(*vpp);
return (0);
}
}
top:
dfnp = vntofn(dvp);
searchnm = nm;
operation = 0;
ASSERT(vn_matchops(dvp, auto_vnodeops));
AUTOFS_DPRINT((3, "auto_lookup: dvp=%p dfnp=%p\n", (void *)dvp,
(void *)dfnp));
/*
* If a lookup or mount of this node is in progress, wait for it
* to finish, and return whatever result it got.
*/
mutex_enter(&dfnp->fn_lock);
if (dfnp->fn_flags & (MF_LOOKUP | MF_INPROG)) {
mutex_exit(&dfnp->fn_lock);
error = auto_wait4mount(dfnp);
if (error == AUTOFS_SHUTDOWN)
error = ENOENT;
if (error == EAGAIN)
goto top;
if (error)
return (error);
} else
mutex_exit(&dfnp->fn_lock);
error = vn_vfsrlock_wait(dvp);
if (error)
return (error);
vfsp = vn_mountedvfs(dvp);
if (vfsp != NULL) {
error = VFS_ROOT(vfsp, &newvp);
vn_vfsunlock(dvp);
if (!error) {
error = VOP_LOOKUP(newvp, nm, vpp, pnp,
flags, rdir, cred, ct, direntflags, realpnp);
VN_RELE(newvp);
}
return (error);
}
vn_vfsunlock(dvp);
rw_enter(&dfnp->fn_rwlock, RW_READER);
error = auto_search(dfnp, nm, &fnp, cred);
if (error) {
if (dfnip->fi_flags & MF_DIRECT) {
/*
* direct map.
*/
if (dfnp->fn_dirents) {
/*
* Mount previously triggered.
* 'nm' not found
*/
error = ENOENT;
} else {
/*
* I need to contact the daemon to trigger
* the mount. 'dfnp' will be the mountpoint.
*/
operation = AUTOFS_MOUNT;
VN_HOLD(fntovn(dfnp));
fnp = dfnp;
error = 0;
}
} else if (dvp == dfnip->fi_rootvp) {
/*
* 'dfnp' is the root of the indirect AUTOFS.
*/
if (rw_tryupgrade(&dfnp->fn_rwlock) == 0) {
/*
* Could not acquire writer lock, release
* reader, and wait until available. We
* need to search for 'nm' again, since we
* had to release the lock before reacquiring
* it.
*/
rw_exit(&dfnp->fn_rwlock);
rw_enter(&dfnp->fn_rwlock, RW_WRITER);
error = auto_search(dfnp, nm, &fnp, cred);
}
ASSERT(RW_WRITE_HELD(&dfnp->fn_rwlock));
if (error) {
/*
* create node being looked-up and request
* mount on it.
*/
error = auto_enter(dfnp, nm, &fnp, kcred);
if (!error)
operation = AUTOFS_LOOKUP;
}
} else if ((dfnp->fn_dirents == NULL) &&
((dvp->v_flag & VROOT) == 0) &&
((fntovn(dfnp->fn_parent))->v_flag & VROOT)) {
/*
* dfnp is the actual 'mountpoint' of indirect map,
* it is the equivalent of a direct mount,
* ie, /home/'user1'
*/
operation = AUTOFS_MOUNT;
VN_HOLD(fntovn(dfnp));
fnp = dfnp;
error = 0;
searchnm = dfnp->fn_name;
}
}
if (error == EAGAIN) {
rw_exit(&dfnp->fn_rwlock);
goto top;
}
if (error) {
rw_exit(&dfnp->fn_rwlock);
return (error);
}
/*
* We now have the actual fnnode we're interested in.
* The 'MF_LOOKUP' indicates another thread is currently
* performing a daemon lookup of this node, therefore we
* wait for its completion.
* The 'MF_INPROG' indicates another thread is currently
* performing a daemon mount of this node, we wait for it
* to be done if we are performing a MOUNT. We don't
* wait for it if we are performing a LOOKUP.
* We can release the reader/writer lock as soon as we acquire
* the mutex, since the state of the lock can only change by
* first acquiring the mutex.
*/
mutex_enter(&fnp->fn_lock);
rw_exit(&dfnp->fn_rwlock);
if ((fnp->fn_flags & MF_LOOKUP) ||
((operation == AUTOFS_MOUNT) && (fnp->fn_flags & MF_INPROG))) {
mutex_exit(&fnp->fn_lock);
error = auto_wait4mount(fnp);
VN_RELE(fntovn(fnp));
if (error == AUTOFS_SHUTDOWN)
error = ENOENT;
if (error && error != EAGAIN)
return (error);
goto top;
}
if (operation == 0) {
/*
* got the fnnode, check for any errors
* on the previous operation on that node.
*/
error = fnp->fn_error;
if ((error == EINTR) || (error == EAGAIN)) {
/*
* previous operation on this node was
* not completed, do a lookup now.
*/
operation = AUTOFS_LOOKUP;
} else {
/*
* previous operation completed. Return
* a pointer to the node only if there was
* no error.
*/
mutex_exit(&fnp->fn_lock);
if (!error)
*vpp = fntovn(fnp);
else
VN_RELE(fntovn(fnp));
return (error);
}
}
/*
* Since I got to this point, it means I'm the one
* responsible for triggering the mount/look-up of this node.
*/
switch (operation) {
case AUTOFS_LOOKUP:
AUTOFS_BLOCK_OTHERS(fnp, MF_LOOKUP);
fnp->fn_error = 0;
mutex_exit(&fnp->fn_lock);
error = auto_lookup_aux(fnp, searchnm, cred);
if (!error) {
/*
* Return this vnode
*/
*vpp = fntovn(fnp);
} else {
/*
* release our reference to this vnode
* and return error
*/
VN_RELE(fntovn(fnp));
}
break;
case AUTOFS_MOUNT:
AUTOFS_BLOCK_OTHERS(fnp, MF_INPROG);
fnp->fn_error = 0;
mutex_exit(&fnp->fn_lock);
/*
* auto_new_mount_thread fires up a new thread which
* calls automountd finishing up the work
*/
auto_new_mount_thread(fnp, searchnm, cred);
/*
* At this point, we are simply another thread
* waiting for the mount to complete
*/
error = auto_wait4mount(fnp);
if (error == AUTOFS_SHUTDOWN)
error = ENOENT;
/*
* now release our reference to this vnode
*/
VN_RELE(fntovn(fnp));
if (!error)
goto top;
break;
default:
auto_log(dfnp->fn_globals->fng_verbose,
dfnp->fn_globals->fng_zoneid, CE_WARN,
"auto_lookup: unknown operation %d",
operation);
}
AUTOFS_DPRINT((5, "auto_lookup: name=%s *vpp=%p return=%d\n",
nm, (void *)*vpp, error));
return (error);
}
/*ARGSUSED*/
int
ufs_rdwr_data(
vnode_t *vnodep,
u_offset_t offset,
size_t len,
fdbuffer_t *fdbp,
int flags,
cred_t *credp)
{
struct inode *ip = VTOI(vnodep);
struct fs *fs;
struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
struct buf *bp;
krw_t rwtype = RW_READER;
u_offset_t offset1 = offset; /* Initial offset */
size_t iolen;
int curlen = 0;
int pplen;
daddr_t bn;
int contig = 0;
int error = 0;
int nbytes; /* Number bytes this IO */
int offsetn; /* Start point this IO */
int iswrite = flags & B_WRITE;
int io_started = 0; /* No IO started */
struct ulockfs *ulp;
uint_t protp = PROT_ALL;
error = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, segkmap, !iswrite,
&protp);
if (error) {
if (flags & B_ASYNC) {
fdb_ioerrdone(fdbp, error);
}
return (error);
}
fs = ufsvfsp->vfs_fs;
iolen = len;
DEBUGF((CE_CONT, "?ufs_rdwr: %s vp: %p pages:%p off %llx len %lx"
" isize: %llx fdb: %p\n",
flags & B_READ ? "READ" : "WRITE", (void *)vnodep,
(void *)vnodep->v_pages, offset1, iolen, ip->i_size, (void *)fdbp));
rw_enter(&ip->i_ufsvfs->vfs_dqrwlock, RW_READER);
rw_enter(&ip->i_contents, rwtype);
ASSERT(offset1 < ip->i_size);
if ((offset1 + iolen) > ip->i_size) {
iolen = ip->i_size - offset1;
}
while (!error && curlen < iolen) {
contig = 0;
if ((error = bmap_read(ip, offset1, &bn, &contig)) != 0) {
break;
}
ASSERT(!(bn == UFS_HOLE && iswrite));
if (bn == UFS_HOLE) {
/*
* If the above assertion is true,
* then the following if statement can never be true.
*/
if (iswrite && (rwtype == RW_READER)) {
rwtype = RW_WRITER;
if (!rw_tryupgrade(&ip->i_contents)) {
rw_exit(&ip->i_contents);
rw_enter(&ip->i_contents, rwtype);
continue;
}
}
offsetn = blkoff(fs, offset1);
pplen = P2ROUNDUP(len, PAGESIZE);
nbytes = MIN((pplen - curlen),
(fs->fs_bsize - offsetn));
ASSERT(nbytes > 0);
/*
* We may be reading or writing.
*/
DEBUGF((CE_CONT, "?ufs_rdwr_data: hole %llx - %lx\n",
offset1, (iolen - curlen)));
if (iswrite) {
printf("**WARNING: ignoring hole in write\n");
error = ENOSPC;
} else {
fdb_add_hole(fdbp, offset1 - offset, nbytes);
}
offset1 += nbytes;
curlen += nbytes;
continue;
}
ASSERT(contig > 0);
pplen = P2ROUNDUP(len, PAGESIZE);
contig = MIN(contig, len - curlen);
contig = P2ROUNDUP(contig, DEV_BSIZE);
bp = fdb_iosetup(fdbp, offset1 - offset, contig, vnodep, flags);
bp->b_edev = ip->i_dev;
bp->b_dev = cmpdev(ip->i_dev);
bp->b_blkno = bn;
bp->b_file = ip->i_vnode;
bp->b_offset = (offset_t)offset1;
if (ufsvfsp->vfs_snapshot) {
fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
} else {
(void) bdev_strategy(bp);
}
io_started = 1;
offset1 += contig;
curlen += contig;
if (iswrite)
lwp_stat_update(LWP_STAT_OUBLK, 1);
else
lwp_stat_update(LWP_STAT_INBLK, 1);
if ((flags & B_ASYNC) == 0) {
error = biowait(bp);
fdb_iodone(bp);
}
DEBUGF((CE_CONT, "?loop ufs_rdwr_data.. off %llx len %lx\n",
offset1, (iolen - curlen)));
}
DEBUGF((CE_CONT, "?ufs_rdwr_data: off %llx len %lx pages: %p ------\n",
offset1, (iolen - curlen), (void *)vnodep->v_pages));
rw_exit(&ip->i_contents);
rw_exit(&ip->i_ufsvfs->vfs_dqrwlock);
if (flags & B_ASYNC) {
/*
* Show that no more asynchronous IO will be added
*/
fdb_ioerrdone(fdbp, error);
}
if (ulp) {
ufs_lockfs_end(ulp);
}
if (io_started && flags & B_ASYNC) {
return (0);
} else {
return (error);
}
}
int
priv_getbyname(const char *name, uint_t flag)
{
int i;
int wheld = 0;
int len;
char *p;
if (flag != 0 && flag != PRIV_ALLOC)
return (-EINVAL);
if (strncasecmp(name, "priv_", 5) == 0)
name += 5;
rw_enter(&privinfo_lock, RW_READER);
rescan:
for (i = 0; i < nprivs; i++)
if (strcasecmp(priv_names[i], name) == 0) {
rw_exit(&privinfo_lock);
return (i);
}
if (!wheld) {
if (!(flag & PRIV_ALLOC)) {
rw_exit(&privinfo_lock);
return (-EINVAL);
}
/* check length, validity and available space */
len = strlen(name) + 1;
if (len > PRIVNAME_MAX) {
rw_exit(&privinfo_lock);
return (-ENAMETOOLONG);
}
for (p = (char *)name; *p != '\0'; p++) {
char c = *p;
if (!((c >= 'A' && c <= 'Z') ||
(c >= 'a' && c <= 'z') ||
(c >= '0' && c <= '9') ||
c == '_')) {
rw_exit(&privinfo_lock);
return (-EINVAL);
}
}
if (!rw_tryupgrade(&privinfo_lock)) {
rw_exit(&privinfo_lock);
rw_enter(&privinfo_lock, RW_WRITER);
wheld = 1;
/* Someone may have added our privilege */
goto rescan;
}
}
if (nprivs == MAX_PRIVILEGE || len + privbytes > maxprivbytes) {
rw_exit(&privinfo_lock);
return (-ENOMEM);
}
priv_names[i] = p = priv_str + privbytes;
bcopy(name, p, len);
/* make the priv_names[i] and privilege name globally visible */
membar_producer();
/* adjust priv count and bytes count */
priv_ninfo->cnt = priv_info->priv_max = ++nprivs;
privbytes += len;
rw_exit(&privinfo_lock);
return (i);
}
/*
* Query domain table for a given domain.
*
* If domain isn't found it is added to AVL trees and
* the results are pushed out to disk.
*/
int
zfs_fuid_find_by_domain(zfsvfs_t *zfsvfs, const char *domain, char **retdomain,
dmu_tx_t *tx)
{
fuid_domain_t searchnode, *findnode;
avl_index_t loc;
krw_t rw = RW_READER;
/*
* If the dummy "nobody" domain then return an index of 0
* to cause the created FUID to be a standard POSIX id
* for the user nobody.
*/
if (domain[0] == '\0') {
*retdomain = nulldomain;
return (0);
}
searchnode.f_ksid = ksid_lookupdomain(domain);
if (retdomain) {
*retdomain = searchnode.f_ksid->kd_name;
}
if (!zfsvfs->z_fuid_loaded)
zfs_fuid_init(zfsvfs, tx);
retry:
rw_enter(&zfsvfs->z_fuid_lock, rw);
findnode = avl_find(&zfsvfs->z_fuid_domain, &searchnode, &loc);
if (findnode) {
rw_exit(&zfsvfs->z_fuid_lock);
ksiddomain_rele(searchnode.f_ksid);
return (findnode->f_idx);
} else {
fuid_domain_t *domnode;
nvlist_t *nvp;
nvlist_t **fuids;
uint64_t retidx;
size_t nvsize = 0;
char *packed;
dmu_buf_t *db;
int i = 0;
if (rw == RW_READER && !rw_tryupgrade(&zfsvfs->z_fuid_lock)) {
rw_exit(&zfsvfs->z_fuid_lock);
rw = RW_WRITER;
goto retry;
}
domnode = kmem_alloc(sizeof (fuid_domain_t), KM_SLEEP);
domnode->f_ksid = searchnode.f_ksid;
retidx = domnode->f_idx = avl_numnodes(&zfsvfs->z_fuid_idx) + 1;
avl_add(&zfsvfs->z_fuid_domain, domnode);
avl_add(&zfsvfs->z_fuid_idx, domnode);
/*
* Now resync the on-disk nvlist.
*/
VERIFY(nvlist_alloc(&nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
domnode = avl_first(&zfsvfs->z_fuid_domain);
fuids = kmem_alloc(retidx * sizeof (void *), KM_SLEEP);
while (domnode) {
VERIFY(nvlist_alloc(&fuids[i],
NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(fuids[i], FUID_IDX,
domnode->f_idx) == 0);
VERIFY(nvlist_add_uint64(fuids[i],
FUID_OFFSET, 0) == 0);
VERIFY(nvlist_add_string(fuids[i++], FUID_DOMAIN,
domnode->f_ksid->kd_name) == 0);
domnode = AVL_NEXT(&zfsvfs->z_fuid_domain, domnode);
}
VERIFY(nvlist_add_nvlist_array(nvp, FUID_NVP_ARRAY,
fuids, retidx) == 0);
for (i = 0; i != retidx; i++)
nvlist_free(fuids[i]);
kmem_free(fuids, retidx * sizeof (void *));
VERIFY(nvlist_size(nvp, &nvsize, NV_ENCODE_XDR) == 0);
packed = kmem_alloc(nvsize, KM_SLEEP);
VERIFY(nvlist_pack(nvp, &packed, &nvsize,
NV_ENCODE_XDR, KM_SLEEP) == 0);
nvlist_free(nvp);
zfsvfs->z_fuid_size = nvsize;
dmu_write(zfsvfs->z_os, zfsvfs->z_fuid_obj, 0,
zfsvfs->z_fuid_size, packed, tx);
kmem_free(packed, zfsvfs->z_fuid_size);
VERIFY(0 == dmu_bonus_hold(zfsvfs->z_os, zfsvfs->z_fuid_obj,
FTAG, &db));
dmu_buf_will_dirty(db, tx);
*(uint64_t *)db->db_data = zfsvfs->z_fuid_size;
dmu_buf_rele(db, FTAG);
rw_exit(&zfsvfs->z_fuid_lock);
return (retidx);
}
}
static void
devvt_cleandir(struct vnode *dvp, struct cred *cred)
{
struct sdev_node *sdvp = VTOSDEV(dvp);
struct sdev_node *dv, *next = NULL;
int min, cnt;
char found = 0;
mutex_enter(&vc_lock);
cnt = VC_INSTANCES_COUNT;
mutex_exit(&vc_lock);
/* We have to fool warlock this way, otherwise it will complain */
#ifndef __lock_lint
if (rw_tryupgrade(&sdvp->sdev_contents) == NULL) {
rw_exit(&sdvp->sdev_contents);
rw_enter(&sdvp->sdev_contents, RW_WRITER);
}
#else
rw_enter(&sdvp->sdev_contents, RW_WRITER);
#endif
/* 1. prune invalid nodes and rebuild stale symlinks */
devvt_prunedir(sdvp);
/* 2. create missing nodes */
for (min = 0; min < cnt; min++) {
char nm[16];
if (vt_minor_valid(min) == B_FALSE)
continue;
(void) snprintf(nm, sizeof (nm), "%d", min);
found = 0;
for (dv = SDEV_FIRST_ENTRY(sdvp); dv; dv = next) {
next = SDEV_NEXT_ENTRY(sdvp, dv);
/* validate only ready nodes */
if (dv->sdev_state != SDEV_READY)
continue;
if (strcmp(nm, dv->sdev_name) == 0) {
found = 1;
break;
}
}
if (!found) {
devvt_create_snode(sdvp, nm, cred, SDEV_VATTR);
}
}
/* 3. create active link node and console user link node */
found = 0;
for (dv = SDEV_FIRST_ENTRY(sdvp); dv; dv = next) {
next = SDEV_NEXT_ENTRY(sdvp, dv);
/* validate only ready nodes */
if (dv->sdev_state != SDEV_READY)
continue;
if ((strcmp(dv->sdev_name, DEVVT_ACTIVE_NAME) == NULL))
found |= 0x01;
if ((strcmp(dv->sdev_name, DEVVT_CONSUSER_NAME) == NULL))
found |= 0x02;
if ((found & 0x01) && (found & 0x02))
break;
}
if (!(found & 0x01))
devvt_create_snode(sdvp, DEVVT_ACTIVE_NAME, cred, SDEV_VLINK);
if (!(found & 0x02))
devvt_create_snode(sdvp, DEVVT_CONSUSER_NAME, cred, SDEV_VLINK);
#ifndef __lock_lint
rw_downgrade(&sdvp->sdev_contents);
#else
rw_exit(&sdvp->sdev_contents);
#endif
}
/*
* This is the predictive prefetch entry point. It associates dnode access
* specified with blkid and nblks arguments with prefetch stream, predicts
* further accesses based on that stats and initiates speculative prefetch.
* fetch_data argument specifies whether actual data blocks should be fetched:
* FALSE -- prefetch only indirect blocks for predicted data blocks;
* TRUE -- prefetch predicted data blocks plus following indirect blocks.
*/
void
dmu_zfetch(zfetch_t *zf, uint64_t blkid, uint64_t nblks, boolean_t fetch_data)
{
zstream_t *zs;
int64_t pf_start, ipf_start, ipf_istart, ipf_iend;
int64_t pf_ahead_blks, max_blks;
int epbs, max_dist_blks, pf_nblks, ipf_nblks;
uint64_t end_of_access_blkid = blkid + nblks;
spa_t *spa = zf->zf_dnode->dn_objset->os_spa;
if (zfs_prefetch_disable)
return;
/*
* If we haven't yet loaded the indirect vdevs' mappings, we
* can only read from blocks that we carefully ensure are on
* concrete vdevs (or previously-loaded indirect vdevs). So we
* can't allow the predictive prefetcher to attempt reads of other
* blocks (e.g. of the MOS's dnode obejct).
*/
if (!spa_indirect_vdevs_loaded(spa))
return;
/*
* As a fast path for small (single-block) files, ignore access
* to the first block.
*/
if (blkid == 0)
return;
rw_enter(&zf->zf_rwlock, RW_READER);
/*
* Find matching prefetch stream. Depending on whether the accesses
* are block-aligned, first block of the new access may either follow
* the last block of the previous access, or be equal to it.
*/
for (zs = list_head(&zf->zf_stream); zs != NULL;
zs = list_next(&zf->zf_stream, zs)) {
if (blkid == zs->zs_blkid || blkid + 1 == zs->zs_blkid) {
mutex_enter(&zs->zs_lock);
/*
* zs_blkid could have changed before we
* acquired zs_lock; re-check them here.
*/
if (blkid == zs->zs_blkid) {
break;
} else if (blkid + 1 == zs->zs_blkid) {
blkid++;
nblks--;
if (nblks == 0) {
/* Already prefetched this before. */
mutex_exit(&zs->zs_lock);
rw_exit(&zf->zf_rwlock);
return;
}
break;
}
mutex_exit(&zs->zs_lock);
}
}
if (zs == NULL) {
/*
* This access is not part of any existing stream. Create
* a new stream for it.
*/
ZFETCHSTAT_BUMP(zfetchstat_misses);
if (rw_tryupgrade(&zf->zf_rwlock))
dmu_zfetch_stream_create(zf, end_of_access_blkid);
rw_exit(&zf->zf_rwlock);
return;
}
/*
* This access was to a block that we issued a prefetch for on
* behalf of this stream. Issue further prefetches for this stream.
*
* Normally, we start prefetching where we stopped
* prefetching last (zs_pf_blkid). But when we get our first
* hit on this stream, zs_pf_blkid == zs_blkid, we don't
* want to prefetch the block we just accessed. In this case,
* start just after the block we just accessed.
*/
pf_start = MAX(zs->zs_pf_blkid, end_of_access_blkid);
/*
* Double our amount of prefetched data, but don't let the
* prefetch get further ahead than zfetch_max_distance.
*/
if (fetch_data) {
max_dist_blks =
zfetch_max_distance >> zf->zf_dnode->dn_datablkshift;
/*
* Previously, we were (zs_pf_blkid - blkid) ahead. We
* want to now be double that, so read that amount again,
* plus the amount we are catching up by (i.e. the amount
* read just now).
*/
pf_ahead_blks = zs->zs_pf_blkid - blkid + nblks;
max_blks = max_dist_blks - (pf_start - end_of_access_blkid);
pf_nblks = MIN(pf_ahead_blks, max_blks);
} else {
/*
* If DV_BUILD is set, we call into nexus driver to do a BUS_CONFIG_ALL.
* Otherwise, simply return cached dv_node's. Hotplug code always call
* devfs_clean() to invalid the dv_node cache.
*/
static int
devfs_readdir(struct vnode *dvp, struct uio *uiop, struct cred *cred, int *eofp)
{
struct dv_node *ddv, *dv;
struct dirent64 *de, *bufp;
offset_t diroff;
offset_t soff;
size_t reclen, movesz;
int error;
struct vattr va;
size_t bufsz;
ddv = VTODV(dvp);
dcmn_err2(("devfs_readdir %s: offset %lld len %ld\n",
ddv->dv_name, uiop->uio_loffset, uiop->uio_iov->iov_len));
ASSERT(ddv->dv_attr || ddv->dv_attrvp);
ASSERT(RW_READ_HELD(&ddv->dv_contents));
if (uiop->uio_loffset >= MAXOFF_T) {
if (eofp)
*eofp = 1;
return (0);
}
if (uiop->uio_iovcnt != 1)
return (EINVAL);
if (dvp->v_type != VDIR)
return (ENOTDIR);
/* Load the initial contents */
if (ddv->dv_flags & DV_BUILD) {
if (!rw_tryupgrade(&ddv->dv_contents)) {
rw_exit(&ddv->dv_contents);
rw_enter(&ddv->dv_contents, RW_WRITER);
}
/* recheck and fill */
if (ddv->dv_flags & DV_BUILD)
dv_filldir(ddv);
rw_downgrade(&ddv->dv_contents);
}
soff = uiop->uio_offset;
bufsz = uiop->uio_iov->iov_len;
de = bufp = kmem_alloc(bufsz, KM_SLEEP);
movesz = 0;
dv = (struct dv_node *)-1;
/*
* Move as many entries into the uio structure as it will take.
* Special case "." and "..".
*/
diroff = 0;
if (soff == 0) { /* . */
reclen = DIRENT64_RECLEN(strlen("."));
if ((movesz + reclen) > bufsz)
goto full;
de->d_ino = (ino64_t)ddv->dv_ino;
de->d_off = (off64_t)diroff + 1;
de->d_reclen = (ushort_t)reclen;
/* use strncpy(9f) to zero out uninitialized bytes */
(void) strncpy(de->d_name, ".", DIRENT64_NAMELEN(reclen));
movesz += reclen;
de = (dirent64_t *)((char *)de + reclen);
dcmn_err3(("devfs_readdir: A: diroff %lld, soff %lld: '%s' "
"reclen %lu\n", diroff, soff, ".", reclen));
}
diroff++;
if (soff <= 1) { /* .. */
reclen = DIRENT64_RECLEN(strlen(".."));
if ((movesz + reclen) > bufsz)
goto full;
de->d_ino = (ino64_t)ddv->dv_dotdot->dv_ino;
de->d_off = (off64_t)diroff + 1;
de->d_reclen = (ushort_t)reclen;
/* use strncpy(9f) to zero out uninitialized bytes */
(void) strncpy(de->d_name, "..", DIRENT64_NAMELEN(reclen));
movesz += reclen;
de = (dirent64_t *)((char *)de + reclen);
dcmn_err3(("devfs_readdir: B: diroff %lld, soff %lld: '%s' "
"reclen %lu\n", diroff, soff, "..", reclen));
}
diroff++;
for (dv = ddv->dv_dot; dv; dv = dv->dv_next, diroff++) {
/*
* although DDM_INTERNAL_PATH minor nodes are skipped for
* readdirs outside the kernel, they still occupy directory
* offsets
*/
if (diroff < soff ||
((dv->dv_flags & DV_INTERNAL) && (cred != kcred)))
continue;
reclen = DIRENT64_RECLEN(strlen(dv->dv_name));
if ((movesz + reclen) > bufsz) {
dcmn_err3(("devfs_readdir: C: diroff "
"%lld, soff %lld: '%s' reclen %lu\n",
diroff, soff, dv->dv_name, reclen));
goto full;
}
de->d_ino = (ino64_t)dv->dv_ino;
de->d_off = (off64_t)diroff + 1;
de->d_reclen = (ushort_t)reclen;
/* use strncpy(9f) to zero out uninitialized bytes */
ASSERT(strlen(dv->dv_name) + 1 <=
DIRENT64_NAMELEN(reclen));
(void) strncpy(de->d_name, dv->dv_name,
DIRENT64_NAMELEN(reclen));
movesz += reclen;
de = (dirent64_t *)((char *)de + reclen);
dcmn_err4(("devfs_readdir: D: diroff "
"%lld, soff %lld: '%s' reclen %lu\n", diroff, soff,
dv->dv_name, reclen));
}
/* the buffer is full, or we exhausted everything */
full: dcmn_err3(("devfs_readdir: moving %lu bytes: "
"diroff %lld, soff %lld, dv %p\n",
movesz, diroff, soff, (void *)dv));
if ((movesz == 0) && dv)
error = EINVAL; /* cannot be represented */
else {
error = uiomove(bufp, movesz, UIO_READ, uiop);
if (error == 0) {
if (eofp)
*eofp = dv ? 0 : 1;
uiop->uio_offset = diroff;
}
va.va_mask = AT_ATIME;
gethrestime(&va.va_atime);
rw_exit(&ddv->dv_contents);
(void) devfs_setattr(dvp, &va, 0, cred, NULL);
rw_enter(&ddv->dv_contents, RW_READER);
}
kmem_free(bufp, bufsz);
return (error);
}