*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013 by Delphix. All rights reserved.
*/
/* Portions Copyright 2010 Robert Milkowski */
#include <sys/spa.h>
#include <sys/zap.h>
#include <sys/sa.h>
+#include <sys/sa_impl.h>
#include <sys/varargs.h>
#include <sys/policy.h>
#include <sys/atomic.h>
#include <sys/modctl.h>
#include <sys/refstr.h>
#include <sys/zfs_ioctl.h>
+#include <sys/zfs_ctldir.h>
#include <sys/zfs_fuid.h>
#include <sys/bootconf.h>
#include <sys/sunddi.h>
#include <sys/dnlc.h>
#include <sys/dmu_objset.h>
#include <sys/spa_boot.h>
-#include <sys/sa.h>
#include <sys/zpl.h>
#include "zfs_comutil.h"
/*ARGSUSED*/
int
-zfs_sync(zfs_sb_t *zsb, short flag, cred_t *cr)
+zfs_sync(struct super_block *sb, int wait, cred_t *cr)
{
+ zfs_sb_t *zsb = sb->s_fs_info;
+
/*
* Data integrity is job one. We don't want a compromised kernel
* writing to the storage pool, so we never sync during panic.
if (unlikely(oops_in_progress))
return (0);
+ /*
+ * Semantically, the only requirement is that the sync be initiated.
+ * The DMU syncs out txgs frequently, so there's nothing to do.
+ */
+ if (!wait)
+ return (0);
+
if (zsb != NULL) {
/*
* Sync a specific filesystem.
ZFS_ENTER(zsb);
dp = dmu_objset_pool(zsb->z_os);
-#ifdef HAVE_SHUTDOWN
/*
* If the system is shutting down, then skip any
* filesystems which may exist on a suspended pool.
- *
- * XXX: This can be implemented using the Linux reboot
- * notifiers: {un}register_reboot_notifier().
*/
- if (sys_shutdown && spa_suspended(dp->dp_spa)) {
+ if (spa_suspended(dp->dp_spa)) {
ZFS_EXIT(zsb);
return (0);
}
-#endif /* HAVE_SHUTDOWN */
if (zsb->z_log != NULL)
zil_commit(zsb->z_log, 0);
}
EXPORT_SYMBOL(zfs_sync);
+boolean_t
+zfs_is_readonly(zfs_sb_t *zsb)
+{
+ return (!!(zsb->z_sb->s_flags & MS_RDONLY));
+}
+EXPORT_SYMBOL(zfs_is_readonly);
+
static void
atime_changed_cb(void *arg, uint64_t newval)
+{
+ ((zfs_sb_t *)arg)->z_atime = newval;
+}
+
+static void
+xattr_changed_cb(void *arg, uint64_t newval)
{
zfs_sb_t *zsb = arg;
- struct super_block *sb = zsb->z_sb;
- struct vfsmount *vfs = zsb->z_vfs;
- if (newval == TRUE) {
- vfs->mnt_flags &= ~MNT_NOATIME;
- sb->s_flags &= ~MS_NOATIME;
- zsb->z_atime = TRUE;
+ if (newval == ZFS_XATTR_OFF) {
+ zsb->z_flags &= ~ZSB_XATTR;
} else {
- vfs->mnt_flags |= MNT_NOATIME;
- sb->s_flags |= MS_NOATIME;
- zsb->z_atime = FALSE;
+ zsb->z_flags |= ZSB_XATTR;
+
+ if (newval == ZFS_XATTR_SA)
+ zsb->z_xattr_sa = B_TRUE;
+ else
+ zsb->z_xattr_sa = B_FALSE;
}
}
static void
-xattr_changed_cb(void *arg, uint64_t newval)
+acltype_changed_cb(void *arg, uint64_t newval)
{
zfs_sb_t *zsb = arg;
- if (newval == TRUE) {
- zsb->z_flags |= ZSB_XATTR_USER;
- } else {
- zsb->z_flags &= ~ZSB_XATTR_USER;
+ switch (newval) {
+ case ZFS_ACLTYPE_OFF:
+ zsb->z_acl_type = ZFS_ACLTYPE_OFF;
+ zsb->z_sb->s_flags &= ~MS_POSIXACL;
+ break;
+ case ZFS_ACLTYPE_POSIXACL:
+#ifdef CONFIG_FS_POSIX_ACL
+ zsb->z_acl_type = ZFS_ACLTYPE_POSIXACL;
+ zsb->z_sb->s_flags |= MS_POSIXACL;
+#else
+ zsb->z_acl_type = ZFS_ACLTYPE_OFF;
+ zsb->z_sb->s_flags &= ~MS_POSIXACL;
+#endif /* CONFIG_FS_POSIX_ACL */
+ break;
+ default:
+ break;
}
}
{
zfs_sb_t *zsb = arg;
struct super_block *sb = zsb->z_sb;
- struct vfsmount *vfs = zsb->z_vfs;
- if (newval) {
- vfs->mnt_flags |= MNT_READONLY;
+ if (sb == NULL)
+ return;
+
+ if (newval)
sb->s_flags |= MS_RDONLY;
- } else {
- vfs->mnt_flags &= ~MNT_READONLY;
+ else
sb->s_flags &= ~MS_RDONLY;
- }
}
static void
devices_changed_cb(void *arg, uint64_t newval)
{
- zfs_sb_t *zsb = arg;
- struct super_block *sb = zsb->z_sb;
- struct vfsmount *vfs = zsb->z_vfs;
-
- if (newval == FALSE) {
- vfs->mnt_flags |= MNT_NODEV;
- sb->s_flags |= MS_NODEV;
- } else {
- vfs->mnt_flags &= ~MNT_NODEV;
- sb->s_flags &= ~MS_NODEV;
- }
}
static void
setuid_changed_cb(void *arg, uint64_t newval)
{
- zfs_sb_t *zsb = arg;
- struct super_block *sb = zsb->z_sb;
- struct vfsmount *vfs = zsb->z_vfs;
-
- if (newval == FALSE) {
- vfs->mnt_flags |= MNT_NOSUID;
- sb->s_flags |= MS_NOSUID;
- } else {
- vfs->mnt_flags &= ~MNT_NOSUID;
- sb->s_flags &= ~MS_NOSUID;
- }
}
static void
exec_changed_cb(void *arg, uint64_t newval)
{
- zfs_sb_t *zsb = arg;
- struct super_block *sb = zsb->z_sb;
- struct vfsmount *vfs = zsb->z_vfs;
-
- if (newval == FALSE) {
- vfs->mnt_flags |= MNT_NOEXEC;
- sb->s_flags |= MS_NOEXEC;
- } else {
- vfs->mnt_flags &= ~MNT_NOEXEC;
- sb->s_flags &= ~MS_NOEXEC;
- }
}
-/*
- * The nbmand mount option can be changed at mount time.
- * We can't allow it to be toggled on live file systems or incorrect
- * behavior may be seen from cifs clients
- *
- * This property isn't registered via dsl_prop_register(), but this callback
- * will be called when a file system is first mounted
- */
static void
nbmand_changed_cb(void *arg, uint64_t newval)
{
zfs_sb_t *zsb = arg;
struct super_block *sb = zsb->z_sb;
- if (newval == TRUE) {
+ if (sb == NULL)
+ return;
+
+ if (newval == TRUE)
sb->s_flags |= MS_MANDLOCK;
- } else {
+ else
sb->s_flags &= ~MS_MANDLOCK;
- }
}
static void
int
zfs_register_callbacks(zfs_sb_t *zsb)
{
- struct vfsmount *vfsp = zsb->z_vfs;
struct dsl_dataset *ds = NULL;
objset_t *os = zsb->z_os;
- uint64_t nbmand;
- boolean_t readonly = B_FALSE;
- boolean_t setuid = B_TRUE;
- boolean_t exec = B_TRUE;
- boolean_t devices = B_TRUE;
- boolean_t xattr = B_TRUE;
- boolean_t atime = B_TRUE;
- char osname[MAXNAMELEN];
+ boolean_t do_readonly = B_FALSE;
int error = 0;
- /*
- * While Linux allows multiple vfs mounts per super block we have
- * limited it artificially to one in zfs_fill_super. Thus it is
- * safe for us to modify the vfs mount fails through the callbacks.
- */
- if ((vfsp->mnt_flags & MNT_READONLY) ||
- !spa_writeable(dmu_objset_spa(os)))
- readonly = B_TRUE;
-
- if (vfsp->mnt_flags & MNT_NOSUID) {
- devices = B_FALSE;
- setuid = B_FALSE;
- } else {
- if (vfsp->mnt_flags & MNT_NODEV)
- devices = B_FALSE;
- }
-
- if (vfsp->mnt_flags & MNT_NOEXEC)
- exec = B_FALSE;
-
- if (vfsp->mnt_flags & MNT_NOATIME)
- atime = B_FALSE;
-
- /*
- * nbmand is a special property which may only be changed at
- * mount time. Unfortunately, Linux does not have a VFS mount
- * flag instead this is a super block flag. So setting this
- * option at mount time will have to wait until we can parse
- * the mount option string. For now we rely on the nbmand
- * value stored with the object set. Additional mount option
- * string to be handled:
- *
- * case: sensitive|insensitive|mixed
- * zerocopy: on|off
- */
-
- dmu_objset_name(os, osname);
- if ((error = dsl_prop_get_integer(osname, "nbmand", &nbmand, NULL)))
- return (error);
+ if (zfs_is_readonly(zsb) || !spa_writeable(dmu_objset_spa(os)))
+ do_readonly = B_TRUE;
/*
* Register property callbacks.
* overboard...
*/
ds = dmu_objset_ds(os);
+ dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
error = dsl_prop_register(ds,
- "atime", atime_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zsb);
+ error = error ? error : dsl_prop_register(ds,
+ zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "xattr", xattr_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "recordsize", blksz_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "readonly", readonly_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "devices", devices_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "setuid", setuid_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "exec", exec_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "snapdir", snapdir_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "aclinherit", acl_inherit_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
- "vscan", vscan_changed_cb, zsb);
+ zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zsb);
+ error = error ? error : dsl_prop_register(ds,
+ zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zsb);
+ dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
if (error)
goto unregister;
- /*
- * Invoke our callbacks to set required flags.
- */
- readonly_changed_cb(zsb, readonly);
- setuid_changed_cb(zsb, setuid);
- exec_changed_cb(zsb, exec);
- devices_changed_cb(zsb, devices);
- xattr_changed_cb(zsb, xattr);
- atime_changed_cb(zsb, atime);
- nbmand_changed_cb(zsb, nbmand);
+ if (do_readonly)
+ readonly_changed_cb(zsb, B_TRUE);
return (0);
* registered, but this is OK; it will simply return ENOMSG,
* which we will ignore.
*/
- (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zsb);
- (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
- zsb);
- (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
+ atime_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
+ xattr_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
+ blksz_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
+ readonly_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
+ devices_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
+ setuid_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
+ exec_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
+ snapdir_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLTYPE),
+ acltype_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
+ acl_inherit_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
+ vscan_changed_cb, zsb);
+ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_NBMAND),
+ nbmand_changed_cb, zsb);
return (error);
}
zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
uint64_t *userp, uint64_t *groupp)
{
- znode_phys_t *znp = data;
- int error = 0;
-
/*
* Is it a valid type of object to track?
*/
if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
- return (ENOENT);
+ return (SET_ERROR(ENOENT));
/*
* If we have a NULL data pointer
* use the same ids
*/
if (data == NULL)
- return (EEXIST);
+ return (SET_ERROR(EEXIST));
if (bonustype == DMU_OT_ZNODE) {
+ znode_phys_t *znp = data;
*userp = znp->zp_uid;
*groupp = znp->zp_gid;
} else {
int hdrsize;
+ sa_hdr_phys_t *sap = data;
+ sa_hdr_phys_t sa = *sap;
+ boolean_t swap = B_FALSE;
ASSERT(bonustype == DMU_OT_SA);
- hdrsize = sa_hdrsize(data);
- if (hdrsize != 0) {
- *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
- SA_UID_OFFSET));
- *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
- SA_GID_OFFSET));
- } else {
+ if (sa.sa_magic == 0) {
/*
* This should only happen for newly created
* files that haven't had the znode data filled
*/
*userp = 0;
*groupp = 0;
+ return (0);
+ }
+ if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
+ sa.sa_magic = SA_MAGIC;
+ sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
+ swap = B_TRUE;
+ } else {
+ VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
+ }
+
+ hdrsize = sa_hdrsize(&sa);
+ VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
+ *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
+ SA_UID_OFFSET));
+ *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
+ SA_GID_OFFSET));
+ if (swap) {
+ *userp = BSWAP_64(*userp);
+ *groupp = BSWAP_64(*groupp);
}
}
- return (error);
+ return (0);
}
static void
case ZFS_PROP_GROUPQUOTA:
return (zsb->z_groupquota_obj);
default:
- return (ENOTSUP);
+ return (SET_ERROR(ENOTSUP));
}
return (0);
}
uint64_t obj;
if (!dmu_objset_userspace_present(zsb->z_os))
- return (ENOTSUP);
+ return (SET_ERROR(ENOTSUP));
obj = zfs_userquota_prop_to_obj(zsb, type);
if (obj == 0) {
if (domain && domain[0]) {
domainid = zfs_fuid_find_by_domain(zsb, domain, NULL, addok);
if (domainid == -1)
- return (ENOENT);
+ return (SET_ERROR(ENOENT));
}
fuid = FUID_ENCODE(domainid, rid);
(void) sprintf(buf, "%llx", (longlong_t)fuid);
*valp = 0;
if (!dmu_objset_userspace_present(zsb->z_os))
- return (ENOTSUP);
+ return (SET_ERROR(ENOTSUP));
obj = zfs_userquota_prop_to_obj(zsb, type);
if (obj == 0)
boolean_t fuid_dirtied;
if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
- return (EINVAL);
+ return (SET_ERROR(EINVAL));
if (zsb->z_version < ZPL_VERSION_USERSPACE)
- return (ENOTSUP);
+ return (SET_ERROR(ENOTSUP));
objp = (type == ZFS_PROP_USERQUOTA) ? &zsb->z_userquota_obj :
&zsb->z_groupquota_obj;
int i, error;
uint64_t sa_obj;
- zsb = kmem_zalloc(sizeof (zfs_sb_t), KM_SLEEP);
+ zsb = kmem_zalloc(sizeof (zfs_sb_t), KM_SLEEP | KM_NODEBUG);
/*
* We claim to always be readonly so we can open snapshots;
* Should probably make this a kmem cache, shuffle fields,
* and just bzero up to z_hold_mtx[].
*/
- zsb->z_vfs = NULL;
+ zsb->z_sb = NULL;
zsb->z_parent = zsb;
zsb->z_max_blksz = SPA_MAXBLOCKSIZE;
zsb->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
"on a version %lld pool\n. Pool must be upgraded to mount "
"this file system.", (u_longlong_t)zsb->z_version,
(u_longlong_t)spa_version(dmu_objset_spa(os)));
- error = ENOTSUP;
+ error = SET_ERROR(ENOTSUP);
goto out;
}
if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
goto out;
zsb->z_case = (uint_t)zval;
+ if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &zval)) != 0)
+ goto out;
+ zsb->z_acl_type = (uint_t)zval;
+
/*
* Fold case on file systems that are always or sometimes case
* insensitive.
error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
&sa_obj);
if (error)
- return (error);
+ goto out;
+
+ error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &zval);
+ if ((error == 0) && (zval == ZFS_XATTR_SA))
+ zsb->z_xattr_sa = B_TRUE;
} else {
/*
* Pre SA versions file systems should never touch
mutex_init(&zsb->z_lock, NULL, MUTEX_DEFAULT, NULL);
list_create(&zsb->z_all_znodes, sizeof (znode_t),
offsetof(znode_t, z_link_node));
- rrw_init(&zsb->z_teardown_lock);
+ rrw_init(&zsb->z_teardown_lock, B_FALSE);
rw_init(&zsb->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
rw_init(&zsb->z_fuid_lock, NULL, RW_DEFAULT, NULL);
for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
mutex_init(&zsb->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
+ avl_create(&zsb->z_ctldir_snaps, snapentry_compare,
+ sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
+ mutex_init(&zsb->z_ctldir_lock, NULL, MUTEX_DEFAULT, NULL);
+
*zsbp = zsb;
return (0);
kmem_free(zsb, sizeof (zfs_sb_t));
return (error);
}
+EXPORT_SYMBOL(zfs_sb_create);
-static int
+int
zfs_sb_setup(zfs_sb_t *zsb, boolean_t mounting)
{
int error;
* During replay we remove the read only flag to
* allow replays to succeed.
*/
- readonly = zsb->z_vfs->mnt_flags & MNT_READONLY;
+ readonly = zfs_is_readonly(zsb);
if (readonly != 0)
- zsb->z_vfs->mnt_flags &= ~MNT_READONLY;
+ readonly_changed_cb(zsb, B_FALSE);
else
zfs_unlinked_drain(zsb);
zsb->z_replay = B_FALSE;
}
}
- zsb->z_vfs->mnt_flags |= readonly; /* restore readonly bit */
+
+ /* restore readonly bit */
+ if (readonly != 0)
+ readonly_changed_cb(zsb, B_TRUE);
}
return (0);
}
+EXPORT_SYMBOL(zfs_sb_setup);
void
zfs_sb_free(zfs_sb_t *zsb)
rw_destroy(&zsb->z_fuid_lock);
for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
mutex_destroy(&zsb->z_hold_mtx[i]);
+ mutex_destroy(&zsb->z_ctldir_lock);
+ avl_destroy(&zsb->z_ctldir_snaps);
kmem_free(zsb, sizeof (zfs_sb_t));
}
+EXPORT_SYMBOL(zfs_sb_free);
static void
zfs_set_fuid_feature(zfs_sb_t *zsb)
VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
zsb) == 0);
+ VERIFY(dsl_prop_unregister(ds, "acltype", acltype_changed_cb,
+ zsb) == 0);
+
VERIFY(dsl_prop_unregister(ds, "aclinherit",
acl_inherit_changed_cb, zsb) == 0);
VERIFY(dsl_prop_unregister(ds, "vscan",
vscan_changed_cb, zsb) == 0);
+
+ VERIFY(dsl_prop_unregister(ds, "nbmand",
+ nbmand_changed_cb, zsb) == 0);
}
}
EXPORT_SYMBOL(zfs_unregister_callbacks);
#ifdef HAVE_MLSLABEL
/*
- * zfs_check_global_label:
- * Check that the hex label string is appropriate for the dataset
- * being mounted into the global_zone proper.
+ * Check that the hex label string is appropriate for the dataset being
+ * mounted into the global_zone proper.
*
- * Return an error if the hex label string is not default or
- * admin_low/admin_high. For admin_low labels, the corresponding
- * dataset must be readonly.
+ * Return an error if the hex label string is not default or
+ * admin_low/admin_high. For admin_low labels, the corresponding
+ * dataset must be readonly.
*/
int
zfs_check_global_label(const char *dsname, const char *hexsl)
if (dsl_prop_get_integer(dsname,
zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
- return (EACCES);
+ return (SET_ERROR(EACCES));
return (rdonly ? 0 : EACCES);
}
- return (EACCES);
+ return (SET_ERROR(EACCES));
}
+EXPORT_SYMBOL(zfs_check_global_label);
#endif /* HAVE_MLSLABEL */
int
{
zfs_sb_t *zsb = dentry->d_sb->s_fs_info;
uint64_t refdbytes, availbytes, usedobjs, availobjs;
+ uint64_t fsid;
uint32_t bshift;
ZFS_ENTER(zsb);
dmu_objset_space(zsb->z_os,
&refdbytes, &availbytes, &usedobjs, &availobjs);
+ fsid = dmu_objset_fsid_guid(zsb->z_os);
/*
- * The underlying storage pool actually uses multiple block sizes.
- * We report the fragsize as the smallest block size we support,
- * and we report our blocksize as the filesystem's maximum blocksize.
+ * The underlying storage pool actually uses multiple block
+ * size. Under Solaris frsize (fragment size) is reported as
+ * the smallest block size we support, and bsize (block size)
+ * as the filesystem's maximum block size. Unfortunately,
+ * under Linux the fragment size and block size are often used
+ * interchangeably. Thus we are forced to report both of them
+ * as the filesystem's maximum block size.
*/
- statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
+ statp->f_frsize = zsb->z_max_blksz;
statp->f_bsize = zsb->z_max_blksz;
bshift = fls(statp->f_bsize) - 1;
* For f_ffree, report the smaller of the number of object available
* and the number of blocks (each object will take at least a block).
*/
- statp->f_ffree = MIN(availobjs, statp->f_bfree);
+ statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
statp->f_files = statp->f_ffree + usedobjs;
- statp->f_fsid.val[0] = 0; /* XXX: Map up some unique ID */
- statp->f_fsid.val[1] = 0;
+ statp->f_fsid.val[0] = (uint32_t)fsid;
+ statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
statp->f_type = ZFS_SUPER_MAGIC;
statp->f_namelen = ZFS_MAXNAMELEN;
}
EXPORT_SYMBOL(zfs_root);
+#ifdef HAVE_SHRINK
+int
+zfs_sb_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
+{
+ zfs_sb_t *zsb = sb->s_fs_info;
+ struct shrinker *shrinker = &sb->s_shrink;
+ struct shrink_control sc = {
+ .nr_to_scan = nr_to_scan,
+ .gfp_mask = GFP_KERNEL,
+ };
+
+ ZFS_ENTER(zsb);
+ *objects = (*shrinker->shrink)(shrinker, &sc);
+ ZFS_EXIT(zsb);
+
+ return (0);
+}
+EXPORT_SYMBOL(zfs_sb_prune);
+#endif /* HAVE_SHRINK */
+
/*
- * Teardown the zfs_sb_t::z_os.
+ * Teardown the zfs_sb_t.
*
* Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
* and 'z_teardown_inactive_lock' held.
*/
int
-zfsvfs_teardown(zfs_sb_t *zsb, boolean_t unmounting)
+zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting)
{
znode_t *zp;
* for non-snapshots.
*/
shrink_dcache_sb(zsb->z_parent->z_sb);
- invalidate_inodes(zsb->z_parent->z_sb);
}
+ /*
+ * If someone has not already unmounted this file system,
+ * drain the iput_taskq to ensure all active references to the
+ * zfs_sb_t have been handled only then can it be safely destroyed.
+ */
+ if (zsb->z_os)
+ taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb->z_os)));
+
/*
* Close the zil. NB: Can't close the zil while zfs_inactive
* threads are blocked as zil_close can call zfs_inactive.
if (!unmounting && (zsb->z_unmounted || zsb->z_os == NULL)) {
rw_exit(&zsb->z_teardown_inactive_lock);
rrw_exit(&zsb->z_teardown_lock, FTAG);
- return (EIO);
+ return (SET_ERROR(EIO));
}
/*
- * At this point there are no vops active, and any new vops will
- * fail with EIO since we have z_teardown_lock for writer (only
- * relavent for forced unmount).
+ * At this point there are no VFS ops active, and any new VFS ops
+ * will fail with EIO since we have z_teardown_lock for writer (only
+ * relevant for forced unmount).
*
* Release all holds on dbufs.
*/
mutex_enter(&zsb->z_znodes_lock);
for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
- zp = list_next(&zsb->z_all_znodes, zp))
- if (zp->z_sa_hdl) {
- ASSERT(atomic_read(&ZTOI(zp)->i_count) > 0);
+ zp = list_next(&zsb->z_all_znodes, zp)) {
+ if (zp->z_sa_hdl)
zfs_znode_dmu_fini(zp);
- }
+ }
mutex_exit(&zsb->z_znodes_lock);
/*
- * If we are unmounting, set the unmounted flag and let new vops
+ * If we are unmounting, set the unmounted flag and let new VFS ops
* unblock. zfs_inactive will have the unmounted behavior, and all
- * other vops will fail with EIO.
+ * other VFS ops will fail with EIO.
*/
if (unmounting) {
zsb->z_unmounted = B_TRUE;
/*
* Evict cached data
*/
- if (dmu_objset_is_dirty_anywhere(zsb->z_os))
- if (!(zsb->z_vfs->mnt_flags & MNT_READONLY))
- txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
- (void) dmu_objset_evict_dbufs(zsb->z_os);
+ if (dsl_dataset_is_dirty(dmu_objset_ds(zsb->z_os)) &&
+ !zfs_is_readonly(zsb))
+ txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
+ dmu_objset_evict_dbufs(zsb->z_os);
return (0);
}
+EXPORT_SYMBOL(zfs_sb_teardown);
+
+#if defined(HAVE_BDI) && !defined(HAVE_BDI_SETUP_AND_REGISTER)
+atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
+#endif /* HAVE_BDI && !HAVE_BDI_SETUP_AND_REGISTER */
int
zfs_domount(struct super_block *sb, void *data, int silent)
uint64_t recordsize;
int error;
- /*
- * Linux allows multiple vfs mounts per super block. However, the
- * zfs_sb_t only contains a pointer for a single vfs mount. This
- * back reference in the long term could be extended to a list of
- * vfs mounts if a hook were added to the kernel to notify us when
- * a vfsmount is destroyed. Until then we must limit the number
- * of mounts per super block to one.
- */
- if (atomic_read(&sb->s_active) > 1)
- return (EBUSY);
-
error = zfs_sb_create(osname, &zsb);
if (error)
return (error);
goto out;
zsb->z_sb = sb;
- zsb->z_vfs = zmd->z_vfs;
sb->s_fs_info = zsb;
sb->s_magic = ZFS_SUPER_MAGIC;
sb->s_maxbytes = MAX_LFS_FILESIZE;
sb->s_blocksize = recordsize;
sb->s_blocksize_bits = ilog2(recordsize);
+#ifdef HAVE_BDI
+ /*
+ * 2.6.32 API change,
+ * Added backing_device_info (BDI) per super block interfaces. A BDI
+ * must be configured when using a non-device backed filesystem for
+ * proper writeback. This is not required for older pdflush kernels.
+ *
+ * NOTE: Linux read-ahead is disabled in favor of zfs read-ahead.
+ */
+ zsb->z_bdi.ra_pages = 0;
+ sb->s_bdi = &zsb->z_bdi;
+
+ error = -bdi_setup_and_register(&zsb->z_bdi, "zfs", BDI_CAP_MAP_COPY);
+ if (error)
+ goto out;
+#endif /* HAVE_BDI */
+
/* Set callback operations for the file system. */
sb->s_op = &zpl_super_operations;
sb->s_xattr = zpl_xattr_handlers;
-#ifdef HAVE_EXPORTS
- sb->s_export_op = &zpl_export_operations;
-#endif /* HAVE_EXPORTS */
+ sb->s_export_op = &zpl_export_operations;
+#ifdef HAVE_S_D_OP
+ sb->s_d_op = &zpl_dentry_operations;
+#endif /* HAVE_S_D_OP */
/* Set features for file system. */
zfs_set_fuid_feature(zsb);
atime_changed_cb(zsb, B_FALSE);
readonly_changed_cb(zsb, B_TRUE);
- if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
+ if ((error = dsl_prop_get_integer(osname,
+ "xattr", &pval, NULL)))
goto out;
xattr_changed_cb(zsb, pval);
+ if ((error = dsl_prop_get_integer(osname,
+ "acltype", &pval, NULL)))
+ goto out;
+ acltype_changed_cb(zsb, pval);
zsb->z_issnap = B_TRUE;
zsb->z_os->os_sync = ZFS_SYNC_DISABLED;
mutex_exit(&zsb->z_os->os_user_ptr_lock);
} else {
error = zfs_sb_setup(zsb, B_TRUE);
-#ifdef HAVE_SNAPSHOT
- (void) zfs_snap_create(zsb);
-#endif /* HAVE_SNAPSHOT */
}
/* Allocate a root inode for the filesystem. */
}
/* Allocate a root dentry for the filesystem */
- sb->s_root = d_alloc_root(root_inode);
+ sb->s_root = d_make_root(root_inode);
if (sb->s_root == NULL) {
(void) zfs_umount(sb);
- error = ENOMEM;
+ error = SET_ERROR(ENOMEM);
goto out;
}
+
+ if (!zsb->z_issnap)
+ zfsctl_create(zsb);
out:
if (error) {
dmu_objset_disown(zsb->z_os, zsb);
}
EXPORT_SYMBOL(zfs_domount);
+/*
+ * Called when an unmount is requested and certain sanity checks have
+ * already passed. At this point no dentries or inodes have been reclaimed
+ * from their respective caches. We drop the extra reference on the .zfs
+ * control directory to allow everything to be reclaimed. All snapshots
+ * must already have been unmounted to reach this point.
+ */
+void
+zfs_preumount(struct super_block *sb)
+{
+ zfs_sb_t *zsb = sb->s_fs_info;
+
+ if (zsb != NULL && zsb->z_ctldir != NULL)
+ zfsctl_destroy(zsb);
+}
+EXPORT_SYMBOL(zfs_preumount);
+
+/*
+ * Called once all other unmount released tear down has occurred.
+ * It is our responsibility to release any remaining infrastructure.
+ */
/*ARGSUSED*/
int
zfs_umount(struct super_block *sb)
zfs_sb_t *zsb = sb->s_fs_info;
objset_t *os;
- VERIFY(zfsvfs_teardown(zsb, B_TRUE) == 0);
+ VERIFY(zfs_sb_teardown(zsb, B_TRUE) == 0);
os = zsb->z_os;
+#ifdef HAVE_BDI
+ bdi_destroy(sb->s_bdi);
+#endif /* HAVE_BDI */
+
/*
* z_os will be NULL if there was an error in
* attempting to reopen zsb.
EXPORT_SYMBOL(zfs_umount);
int
-zfs_vget(struct vfsmount *vfsp, struct inode **ipp, fid_t *fidp)
+zfs_remount(struct super_block *sb, int *flags, char *data)
+{
+ /*
+ * All namespace flags (MNT_*) and super block flags (MS_*) will
+ * be handled by the Linux VFS. Only handle custom options here.
+ */
+ return (0);
+}
+EXPORT_SYMBOL(zfs_remount);
+
+int
+zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
{
- zfs_sb_t *zsb = VTOZSB(vfsp);
+ zfs_sb_t *zsb = sb->s_fs_info;
znode_t *zp;
uint64_t object = 0;
uint64_t fid_gen = 0;
ZFS_EXIT(zsb);
-#ifdef HAVE_SNAPSHOT
- err = zfsctl_lookup_objset(vfsp, objsetid, &zsb);
+ err = zfsctl_lookup_objset(sb, objsetid, &zsb);
if (err)
- return (EINVAL);
-#endif /* HAVE_SNAPSHOT */
+ return (SET_ERROR(EINVAL));
+
ZFS_ENTER(zsb);
}
fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
} else {
ZFS_EXIT(zsb);
- return (EINVAL);
+ return (SET_ERROR(EINVAL));
}
-#ifdef HAVE_SNAPSHOT
/* A zero fid_gen means we are in the .zfs control directories */
if (fid_gen == 0 &&
(object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
*ipp = zsb->z_ctldir;
ASSERT(*ipp != NULL);
if (object == ZFSCTL_INO_SNAPDIR) {
- VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp, NULL,
- 0, NULL, NULL, NULL, NULL, NULL) == 0);
+ VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
+ 0, kcred, NULL, NULL) == 0);
} else {
igrab(*ipp);
}
ZFS_EXIT(zsb);
return (0);
}
-#endif /* HAVE_SNAPSHOT */
gen_mask = -1ULL >> (64 - 8 * i);
dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
iput(ZTOI(zp));
ZFS_EXIT(zsb);
- return (EINVAL);
+ return (SET_ERROR(EINVAL));
}
*ipp = ZTOI(zp);
EXPORT_SYMBOL(zfs_vget);
/*
- * Block out VOPs and close zfs_sb_t::z_os
+ * Block out VFS ops and close zfs_sb_t
*
* Note, if successful, then we return with the 'z_teardown_lock' and
- * 'z_teardown_inactive_lock' write held.
+ * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
+ * dataset and objset intact so that they can be atomically handed off during
+ * a subsequent rollback or recv operation and the resume thereafter.
*/
int
zfs_suspend_fs(zfs_sb_t *zsb)
{
int error;
- if ((error = zfsvfs_teardown(zsb, B_FALSE)) != 0)
+ if ((error = zfs_sb_teardown(zsb, B_FALSE)) != 0)
return (error);
- dmu_objset_disown(zsb->z_os, zsb);
return (0);
}
EXPORT_SYMBOL(zfs_suspend_fs);
/*
- * Reopen zfs_sb_t::z_os and release VOPs.
+ * Reopen zfs_sb_t and release VFS ops.
*/
int
zfs_resume_fs(zfs_sb_t *zsb, const char *osname)
{
int err, err2;
+ znode_t *zp;
+ uint64_t sa_obj = 0;
ASSERT(RRW_WRITE_HELD(&zsb->z_teardown_lock));
ASSERT(RW_WRITE_HELD(&zsb->z_teardown_inactive_lock));
- err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zsb, &zsb->z_os);
- if (err) {
- zsb->z_os = NULL;
- } else {
- znode_t *zp;
- uint64_t sa_obj = 0;
+ /*
+ * We already own this, so just hold and rele it to update the
+ * objset_t, as the one we had before may have been evicted.
+ */
+ VERIFY0(dmu_objset_hold(osname, zsb, &zsb->z_os));
+ VERIFY3P(zsb->z_os->os_dsl_dataset->ds_owner, ==, zsb);
+ VERIFY(dsl_dataset_long_held(zsb->z_os->os_dsl_dataset));
+ dmu_objset_rele(zsb->z_os, zsb);
- err2 = zap_lookup(zsb->z_os, MASTER_NODE_OBJ,
- ZFS_SA_ATTRS, 8, 1, &sa_obj);
+ /*
+ * Make sure version hasn't changed
+ */
- if ((err || err2) && zsb->z_version >= ZPL_VERSION_SA)
- goto bail;
+ err = zfs_get_zplprop(zsb->z_os, ZFS_PROP_VERSION,
+ &zsb->z_version);
+ if (err)
+ goto bail;
- if ((err = sa_setup(zsb->z_os, sa_obj,
- zfs_attr_table, ZPL_END, &zsb->z_attr_table)) != 0)
- goto bail;
+ err = zap_lookup(zsb->z_os, MASTER_NODE_OBJ,
+ ZFS_SA_ATTRS, 8, 1, &sa_obj);
- VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0);
+ if (err && zsb->z_version >= ZPL_VERSION_SA)
+ goto bail;
- /*
- * Attempt to re-establish all the active znodes with
- * their dbufs. If a zfs_rezget() fails, then we'll let
- * any potential callers discover that via ZFS_ENTER_VERIFY_VP
- * when they try to use their znode.
- */
- mutex_enter(&zsb->z_znodes_lock);
- for (zp = list_head(&zsb->z_all_znodes); zp;
- zp = list_next(&zsb->z_all_znodes, zp)) {
- (void) zfs_rezget(zp);
- }
- mutex_exit(&zsb->z_znodes_lock);
+ if ((err = sa_setup(zsb->z_os, sa_obj,
+ zfs_attr_table, ZPL_END, &zsb->z_attr_table)) != 0)
+ goto bail;
+
+ if (zsb->z_version >= ZPL_VERSION_SA)
+ sa_register_update_callback(zsb->z_os,
+ zfs_sa_upgrade);
+
+ VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0);
+ zfs_set_fuid_feature(zsb);
+ zsb->z_rollback_time = jiffies;
+
+ /*
+ * Attempt to re-establish all the active inodes with their
+ * dbufs. If a zfs_rezget() fails, then we unhash the inode
+ * and mark it stale. This prevents a collision if a new
+ * inode/object is created which must use the same inode
+ * number. The stale inode will be be released when the
+ * VFS prunes the dentry holding the remaining references
+ * on the stale inode.
+ */
+ mutex_enter(&zsb->z_znodes_lock);
+ for (zp = list_head(&zsb->z_all_znodes); zp;
+ zp = list_next(&zsb->z_all_znodes, zp)) {
+ err2 = zfs_rezget(zp);
+ if (err2) {
+ remove_inode_hash(ZTOI(zp));
+ zp->z_is_stale = B_TRUE;
+ }
}
+ mutex_exit(&zsb->z_znodes_lock);
bail:
- /* release the VOPs */
+ /* release the VFS ops */
rw_exit(&zsb->z_teardown_inactive_lock);
rrw_exit(&zsb->z_teardown_lock, FTAG);
if (err) {
/*
- * Since we couldn't reopen zfs_sb_t::z_os, force
+ * Since we couldn't setup the sa framework, try to force
* unmount this file system.
*/
- (void) zfs_umount(zsb->z_sb);
+ if (zsb->z_os)
+ (void) zfs_umount(zsb->z_sb);
}
return (err);
}
dmu_tx_t *tx;
if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
- return (EINVAL);
+ return (SET_ERROR(EINVAL));
if (newvers < zsb->z_version)
- return (EINVAL);
+ return (SET_ERROR(EINVAL));
if (zfs_spa_version_map(newvers) >
spa_version(dmu_objset_spa(zsb->z_os)))
- return (ENOTSUP);
+ return (SET_ERROR(ENOTSUP));
tx = dmu_tx_create(os);
dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
error = zap_add(os, MASTER_NODE_OBJ,
ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
- ASSERT3U(error, ==, 0);
+ ASSERT0(error);
VERIFY(0 == sa_set_sa_object(os, sa_obj));
sa_register_update_callback(os, zfs_sa_upgrade);
}
- spa_history_log_internal(LOG_DS_UPGRADE,
- dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
- zsb->z_version, newvers, dmu_objset_id(os));
+ spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
+ "from %llu to %llu", zsb->z_version, newvers);
dmu_tx_commit(tx);
zsb->z_version = newvers;
- if (zsb->z_version >= ZPL_VERSION_FUID)
- zfs_set_fuid_feature(zsb);
+ zfs_set_fuid_feature(zsb);
return (0);
}
zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
{
const char *pname;
- int error = ENOENT;
+ int error = SET_ERROR(ENOENT);
/*
* Look up the file system's value for the property. For the
case ZFS_PROP_CASE:
*value = ZFS_CASE_SENSITIVE;
break;
+ case ZFS_PROP_ACLTYPE:
+ *value = ZFS_ACLTYPE_OFF;
+ break;
default:
return (error);
}
}
return (error);
}
+EXPORT_SYMBOL(zfs_get_zplprop);
void
zfs_init(void)
{
+ zfsctl_init();
zfs_znode_init();
dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
register_filesystem(&zpl_fs_type);
+ (void) arc_add_prune_callback(zpl_prune_sbs, NULL);
}
void
zfs_fini(void)
{
+ taskq_wait(system_taskq);
unregister_filesystem(&zpl_fs_type);
zfs_znode_fini();
+ zfsctl_fini();
}