void
zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
{
dmu_buf_t *db = sa_get_db(hdl);
znode_t *zp = sa_get_userdata(hdl);
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
sa_bulk_attr_t bulk[20];
int count = 0;
sa_bulk_attr_t sa_attrs[20] = { { 0 } };
zfs_acl_locator_cb_t locate = { 0 };
uint64_t uid, gid, mode, rdev, xattr, parent;
uint64_t crtime[2], mtime[2], ctime[2];
zfs_acl_phys_t znode_acl;
char scanstamp[AV_SCANSTAMP_SZ];
boolean_t drop_lock = B_FALSE;
/*
* No upgrade if ACL isn't cached
* since we won't know which locks are held
* and ready the ACL would require special "locked"
* interfaces that would be messy
*/
if (zp->z_acl_cached == NULL || vnode_islnk(ZTOV(zp)))
return;
/*
* If the z_lock is held and we aren't the owner
* the just return since we don't want to deadlock
* trying to update the status of z_is_sa. This
* file can then be upgraded at a later time.
*
* Otherwise, we know we are doing the
* sa_update() that caused us to enter this function.
*/
if (mutex_owner(&zp->z_lock) != curthread) {
if (mutex_tryenter(&zp->z_lock) == 0)
return;
else
drop_lock = B_TRUE;
}
/* First do a bulk query of the attributes that aren't cached */
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
&znode_acl, 88);
if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
goto done;
/*
* While the order here doesn't matter its best to try and organize
* it is such a way to pick up an already existing layout number
*/
count = 0;
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
&zp->z_size, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
NULL, &zp->z_gen, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
NULL, &parent, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
&zp->z_pflags, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
zp->z_atime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
&mtime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
&ctime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
&crtime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
&zp->z_links, 8);
if (vnode_isblk(zp->z_vnode) || vnode_islnk(zp->z_vnode))
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
&rdev, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
&zp->z_acl_cached->z_acl_count, 8);
if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
zfs_acl_xform(zp, zp->z_acl_cached, CRED());
locate.cb_aclp = zp->z_acl_cached;
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
if (xattr)
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zfsvfs),
NULL, &xattr, 8);
/* if scanstamp then add scanstamp */
if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
scanstamp, AV_SCANSTAMP_SZ);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
NULL, scanstamp, AV_SCANSTAMP_SZ);
zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
}
VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
count, tx) == 0);
if (znode_acl.z_acl_extern_obj)
VERIFY(0 == dmu_object_free(zfsvfs->z_os,
znode_acl.z_acl_extern_obj, tx));
zp->z_is_sa = B_TRUE;
done:
if (drop_lock)
mutex_exit(&zp->z_lock);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd = vd->vdev_tsd;
vnode_t *devvp = NULLVP;
vfs_context_t context = NULL;
uint64_t blkcnt;
uint32_t blksize;
int fmode = 0;
int error = 0;
int isssd;
/*
* 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 (SET_ERROR(EINVAL));
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
if (dvd->vd_offline) {
/*
* If we are opening a device in its offline notify
* context, the LDI handle was just closed. Clean
* up the LDI event callbacks and free vd->vdev_tsd.
*/
vdev_disk_free(vd);
} else {
ASSERT(vd->vdev_reopening);
devvp = dvd->vd_devvp;
goto skip_open;
}
}
/*
* Create vd->vdev_tsd.
*/
vdev_disk_alloc(vd);
dvd = vd->vdev_tsd;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
/* ### APPLE TODO ### */
#ifdef illumos
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
}
#endif
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
context = vfs_context_create( spl_vfs_context_kernel() );
/* Obtain an opened/referenced vnode for the device. */
if ((error = vnode_open(vd->vdev_path, spa_mode(spa), 0, 0,
&devvp, context))) {
goto out;
}
if (!vnode_isblk(devvp)) {
error = ENOTBLK;
goto out;
}
/*
* ### APPLE TODO ###
* vnode_authorize devvp for KAUTH_VNODE_READ_DATA and
* KAUTH_VNODE_WRITE_DATA
*/
/*
* Disallow opening of a device that is currently in use.
* Flush out any old buffers remaining from a previous use.
*/
if ((error = vfs_mountedon(devvp))) {
goto out;
}
if (VNOP_FSYNC(devvp, MNT_WAIT, context) != 0) {
error = ENOTBLK;
goto out;
}
if ((error = buf_invalidateblks(devvp, BUF_WRITE_DATA, 0, 0))) {
goto out;
}
} else {
goto out;
}
int len = MAXPATHLEN;
if (vn_getpath(devvp, dvd->vd_readlinkname, &len) == 0) {
dprintf("ZFS: '%s' resolved name is '%s'\n",
vd->vdev_path, dvd->vd_readlinkname);
} else {
dvd->vd_readlinkname[0] = 0;
}
skip_open:
/*
* Determine the actual size of the device.
*/
if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0,
context) != 0 ||
VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt, 0,
context) != 0) {
error = EINVAL;
goto out;
}
*psize = blkcnt * (uint64_t)blksize;
*max_psize = *psize;
dvd->vd_ashift = highbit(blksize) - 1;
dprintf("vdev_disk: Device %p ashift set to %d\n", devvp,
dvd->vd_ashift);
*ashift = highbit(MAX(blksize, SPA_MINBLOCKSIZE)) - 1;
/*
* ### APPLE TODO ###
*/
#ifdef illumos
if (vd->vdev_wholedisk == 1) {
int wce = 1;
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize = capacity * blksz;
}
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
#endif
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
/* Inform the ZIO pipeline that we are non-rotational */
vd->vdev_nonrot = B_FALSE;
if (VNOP_IOCTL(devvp, DKIOCISSOLIDSTATE, (caddr_t)&isssd, 0,
context) == 0) {
if (isssd)
vd->vdev_nonrot = B_TRUE;
}
dprintf("ZFS: vdev_disk(%s) isSSD %d\n", vd->vdev_path ? vd->vdev_path : "",
isssd);
dvd->vd_devvp = devvp;
out:
if (error) {
if (devvp) {
vnode_close(devvp, fmode, context);
dvd->vd_devvp = NULL;
}
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
}
if (context)
(void) vfs_context_rele(context);
if (error) printf("ZFS: vdev_disk_open('%s') failed error %d\n",
vd->vdev_path ? vd->vdev_path : "", error);
return (error);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *size, uint64_t *max_size, uint64_t *ashift)
{
vdev_disk_t *dvd = NULL;
vnode_t *devvp = NULLVP;
vfs_context_t context = NULL;
uint64_t blkcnt;
uint32_t blksize;
int fmode = 0;
int error = 0;
/*
* 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);
}
dvd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
if (dvd == NULL)
return ENOMEM;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the save device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*
*/
/* ### APPLE TODO ### */
/* ddi_devid_str_decode */
context = vfs_context_create((vfs_context_t)0);
/* Obtain an opened/referenced vnode for the device. */
error = vnode_open(vd->vdev_path, spa_mode(vd->vdev_spa), 0, 0, &devvp, context);
if (error) {
goto out;
}
if (!vnode_isblk(devvp)) {
error = ENOTBLK;
goto out;
}
/* ### APPLE TODO ### */
/* vnode_authorize devvp for KAUTH_VNODE_READ_DATA and
* KAUTH_VNODE_WRITE_DATA
*/
/*
* Disallow opening of a device that is currently in use.
* Flush out any old buffers remaining from a previous use.
*/
if ((error = vfs_mountedon(devvp))) {
goto out;
}
if (VNOP_FSYNC(devvp, MNT_WAIT, context) != 0) {
error = ENOTBLK;
goto out;
}
if ((error = buf_invalidateblks(devvp, BUF_WRITE_DATA, 0, 0))) {
goto out;
}
/*
* Determine the actual size of the device.
*/
if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0, context)
!= 0 || VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt,
0, context) != 0) {
error = EINVAL;
goto out;
}
*size = blkcnt * (uint64_t)blksize;
/*
* ### APPLE TODO ###
* If we own the whole disk, try to enable disk write caching.
*/
/*
* Take the device's minimum transfer size into account.
*/
*ashift = highbit(MAX(blksize, SPA_MINBLOCKSIZE)) - 1;
/*
* Setting the vdev_ashift did in fact break the pool for import
* on ZEVO. This puts the logic into question. It appears that vdev_top
* will also then change. It then panics in space_map from metaslab_alloc
*/
//vd->vdev_ashift = *ashift;
dvd->vd_ashift = *ashift;
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
vd->vdev_tsd = dvd;
dvd->vd_devvp = devvp;
out:
if (error) {
if (devvp)
vnode_close(devvp, fmode, context);
if (dvd)
kmem_free(dvd, sizeof (vdev_disk_t));
/*
* Since the open has failed, vd->vdev_tsd should
* be NULL when we get here, signaling to the
* rest of the spa not to try and reopen or close this device
*/
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
}
if (context) {
(void) vfs_context_rele(context);
}
return (error);
}
