* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_OSD
-#include <lustre_ver.h>
#include <libcfs/libcfs.h>
#include <obd_support.h>
#include <lustre_net.h>
#include <obd_class.h>
#include <lustre_disk.h>
#include <lustre_fid.h>
+#include <lustre_linkea.h>
#include "osd_internal.h"
#include <sys/txg.h>
#include <linux/posix_acl_xattr.h>
-
+#include <lustre_scrub.h>
int __osd_xattr_load(struct osd_device *osd, sa_handle_t *hdl, nvlist_t **sa)
{
* \retval 0 on success
* \retval negative negated errno on failure
*/
-int __osd_xattr_get(const struct lu_env *env, struct osd_object *obj,
- struct lu_buf *buf, const char *name, int *sizep)
+int osd_xattr_get_internal(const struct lu_env *env, struct osd_object *obj,
+ struct lu_buf *buf, const char *name, int *sizep)
{
int rc;
buf, name, sizep);
}
+/**
+ * Copy LMA extended attribute into provided buffer
+ *
+ * Note that no locking is done here.
+ *
+ * \param[in] env execution environment
+ * \param[in] obj object for which to retrieve xattr
+ * \param[out] buf buffer to store xattr value in
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on failure
+ */
+int osd_xattr_get_lma(const struct lu_env *env, struct osd_object *obj,
+ struct lu_buf *buf)
+{
+ int size = 0;
+ int rc = -ENOENT;
+
+ if (!buf)
+ return 0;
+
+ if (unlikely(obj->oo_destroyed))
+ goto out_lma;
+
+ /* check SA_ZPL_DXATTR first then fallback to directory xattr */
+ rc = __osd_sa_xattr_get(env, obj, buf, XATTR_NAME_LMA, &size);
+ if (!rc && unlikely(size < sizeof(struct lustre_mdt_attrs)))
+ rc = -EINVAL;
+ if (rc != -ENOENT)
+ goto out_lma;
+
+ rc = __osd_xattr_get_large(env, osd_obj2dev(obj), obj->oo_xattr,
+ buf, XATTR_NAME_LMA, &size);
+ if (!rc && unlikely(size < sizeof(struct lustre_mdt_attrs)))
+ rc = -EINVAL;
+
+out_lma:
+ return rc;
+}
+
+static int osd_get_pfid_from_lma(const struct lu_env *env,
+ struct osd_object *obj,
+ struct lu_buf *buf, int *sizep)
+{
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lustre_ost_attrs *loa =
+ (struct lustre_ost_attrs *)&info->oti_buf;
+ struct lustre_mdt_attrs *lma = &loa->loa_lma;
+ struct filter_fid *ff;
+ struct ost_layout *ol;
+ struct lu_buf tbuf = {
+ .lb_buf = loa,
+ .lb_len = sizeof(info->oti_buf),
+ };
+ int rc;
+ ENTRY;
+
+ BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*loa));
+ rc = osd_xattr_get_internal(env, obj, &tbuf,
+ XATTR_NAME_LMA, sizep);
+ if (rc)
+ RETURN(rc);
+
+ lustre_loa_swab(loa, true);
+ LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
+
+ *sizep = sizeof(*ff);
+ if (buf->lb_len == 0 || !buf->lb_buf)
+ RETURN(0);
+
+ if (buf->lb_len < *sizep)
+ RETURN(-ERANGE);
+
+ ff = buf->lb_buf;
+ ol = &ff->ff_layout;
+ ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
+ PFID_STRIPE_IDX_BITS);
+ ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
+ loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
+ fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
+ if (lma->lma_compat & LMAC_COMP_INFO) {
+ ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
+ ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
+ ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
+ } else {
+ ol->ol_comp_start = 0;
+ ol->ol_comp_end = 0;
+ ol->ol_comp_id = 0;
+ }
+
+ RETURN(0);
+}
+
int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
struct lu_buf *buf, const char *name)
{
RETURN(-EOPNOTSUPP);
down_read(&obj->oo_guard);
- rc = __osd_xattr_get(env, obj, buf, name, &size);
+ if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed)) {
+ up_read(&obj->oo_guard);
+ RETURN(-ENOENT);
+ }
+
+ /* For the OST migrated from ldiskfs, the PFID EA may
+ * be stored in LMA because of ldiskfs inode size. */
+ if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma)
+ rc = osd_get_pfid_from_lma(env, obj, buf, &size);
+ else
+ rc = osd_xattr_get_internal(env, obj, buf, name, &size);
up_read(&obj->oo_guard);
if (rc == -ENOENT)
int vallen, const char *name,
struct osd_thandle *oh)
{
+ struct osd_device *osd = osd_obj2dev(obj);
dmu_tx_t *tx = oh->ot_tx;
int bonuslen;
if (unlikely(obj->oo_destroyed))
return;
+ if (strcmp(name, XATTR_NAME_LINK) == 0 &&
+ osd->od_remote_parent_dir != ZFS_NO_OBJECT) {
+ /* If some name entry resides on remote MDT, then will create
+ * agent entry under remote parent. On the other hand, if the
+ * remote entry will be removed, then related agent entry may
+ * need to be removed from the remote parent. So there may be
+ * kinds of cases, let's declare enough credits. The credits
+ * for create agent entry is enough for remove case. */
+ osd_tx_hold_zap(tx, osd->od_remote_parent_dir,
+ NULL, TRUE, NULL);
+ }
+
if (unlikely(!osd_obj2dev(obj)->od_xattr_in_sa)) {
__osd_xattr_declare_legacy(env, obj, vallen, name, oh);
return;
ENTRY;
LASSERT(handle != NULL);
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
down_read(&obj->oo_guard);
__osd_xattr_declare_set(env, obj, buf->lb_len, name, oh);
RETURN(0);
}
+int __osd_sa_attr_init(const struct lu_env *env, struct osd_object *obj,
+ struct osd_thandle *oh)
+{
+ sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
+ struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
+ struct lu_buf *lb = &osd_oti_get(env)->oti_xattr_lbuf;
+ struct osd_device *osd = osd_obj2dev(obj);
+ uint64_t gen;
+ inode_timespec_t now;
+ size_t size;
+ int rc, cnt;
+
+ obj->oo_late_xattr = 0;
+ obj->oo_late_attr_set = 0;
+
+ gen = dmu_tx_get_txg(oh->ot_tx);
+ gethrestime(&now);
+ ZFS_TIME_ENCODE(&now, osa->btime);
+
+ obj->oo_attr.la_valid |= LA_BTIME;
+ obj->oo_attr.la_btime = osa->btime[0];
+ osa->atime[0] = obj->oo_attr.la_atime;
+ osa->ctime[0] = obj->oo_attr.la_ctime;
+ osa->mtime[0] = obj->oo_attr.la_mtime;
+ osa->mode = obj->oo_attr.la_mode;
+ osa->uid = obj->oo_attr.la_uid;
+ osa->gid = obj->oo_attr.la_gid;
+ osa->rdev = obj->oo_attr.la_rdev;
+ osa->nlink = obj->oo_attr.la_nlink;
+ osa->flags = attrs_fs2zfs(obj->oo_attr.la_flags);
+ osa->size = obj->oo_attr.la_size;
+#ifdef ZFS_PROJINHERIT
+ if (osd->od_projectused_dn) {
+ if (obj->oo_attr.la_valid & LA_PROJID)
+ osa->projid = obj->oo_attr.la_projid;
+ else
+ osa->projid = ZFS_DEFAULT_PROJID;
+ osa->flags |= ZFS_PROJID;
+ obj->oo_with_projid = 1;
+ }
+#endif
+
+ cnt = 0;
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL,
+ &obj->oo_parent, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, osa->btime, 16);
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
+#ifdef ZFS_PROJINHERIT
+ if (osd->od_projectused_dn)
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
+ &osa->projid, 8);
+#endif
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
+ LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
+
+ /* Update the SA for additions, modifications, and removals. */
+ rc = -nvlist_size(obj->oo_sa_xattr, &size, NV_ENCODE_XDR);
+ if (rc)
+ return rc;
+
+ lu_buf_check_and_alloc(lb, size);
+ if (lb->lb_buf == NULL) {
+ CERROR("%s: can't allocate buffer for xattr update\n",
+ osd->od_svname);
+ return -ENOMEM;
+ }
+
+ rc = -nvlist_pack(obj->oo_sa_xattr, (char **)&lb->lb_buf, &size,
+ NV_ENCODE_XDR, KM_SLEEP);
+ if (rc)
+ return rc;
+
+ SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd), NULL, lb->lb_buf, size);
+
+ rc = -sa_replace_all_by_template(obj->oo_sa_hdl, bulk, cnt, oh->ot_tx);
+
+ return rc;
+}
+
+int __osd_sa_xattr_update(const struct lu_env *env, struct osd_object *obj,
+ struct osd_thandle *oh)
+{
+ struct lu_buf *lb = &osd_oti_get(env)->oti_xattr_lbuf;
+ struct osd_device *osd = osd_obj2dev(obj);
+ char *dxattr;
+ size_t size;
+ int rc;
+
+ obj->oo_late_xattr = 0;
+
+ /* Update the SA for additions, modifications, and removals. */
+ rc = -nvlist_size(obj->oo_sa_xattr, &size, NV_ENCODE_XDR);
+ if (rc)
+ return rc;
+
+ lu_buf_check_and_alloc(lb, size);
+ if (lb->lb_buf == NULL) {
+ CERROR("%s: can't allocate buffer for xattr update\n",
+ osd->od_svname);
+ return -ENOMEM;
+ }
+
+ dxattr = lb->lb_buf;
+ rc = -nvlist_pack(obj->oo_sa_xattr, &dxattr, &size,
+ NV_ENCODE_XDR, KM_SLEEP);
+ if (rc)
+ return rc;
+ LASSERT(dxattr == lb->lb_buf);
+
+ sa_update(obj->oo_sa_hdl, SA_ZPL_DXATTR(osd), dxattr, size, oh->ot_tx);
+
+ return 0;
+}
+
/*
* Set an extended attribute.
* This transaction must have called udmu_xattr_declare_set() first.
*
* No locking is done here.
*/
-int __osd_sa_xattr_update(const struct lu_env *env, struct osd_object *obj,
- struct osd_thandle *oh)
+int __osd_sa_xattr_schedule_update(const struct lu_env *env,
+ struct osd_object *obj,
+ struct osd_thandle *oh)
{
- struct osd_device *osd = osd_obj2dev(obj);
- char *dxattr;
- size_t sa_size;
- int rc;
-
ENTRY;
LASSERT(obj->oo_sa_hdl);
LASSERT(obj->oo_sa_xattr);
- /* Update the SA for additions, modifications, and removals. */
- rc = -nvlist_size(obj->oo_sa_xattr, &sa_size, NV_ENCODE_XDR);
- if (rc)
- return rc;
-
- dxattr = osd_zio_buf_alloc(sa_size);
- if (dxattr == NULL)
- RETURN(-ENOMEM);
+ /* schedule batched SA update in osd_object_sa_dirty_rele() */
+ obj->oo_late_xattr = 1;
+ osd_object_sa_dirty_add(obj, oh);
- rc = -nvlist_pack(obj->oo_sa_xattr, &dxattr, &sa_size,
- NV_ENCODE_XDR, KM_SLEEP);
- if (rc)
- GOTO(out_free, rc);
+ RETURN(0);
- rc = osd_object_sa_update(obj, SA_ZPL_DXATTR(osd), dxattr, sa_size, oh);
-out_free:
- osd_zio_buf_free(dxattr, sa_size);
- RETURN(rc);
}
int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
return rc;
LASSERT(obj->oo_sa_xattr);
- /* Limited to 32k to keep nvpair memory allocations small */
- if (buf->lb_len > DXATTR_MAX_ENTRY_SIZE) {
+ if (buf->lb_len > OBD_MAX_EA_SIZE) {
too_big = 1;
} else {
/* Prevent the DXATTR SA from consuming the entire SA
DATA_TYPE_BYTE_ARRAY);
if (rc < 0)
return rc;
- rc = __osd_sa_xattr_update(env, obj, oh);
+ rc = __osd_sa_xattr_schedule_update(env, obj, oh);
return rc == 0 ? -EFBIG : rc;
}
} else if (rc == -ENOENT) {
if (rc)
return rc;
- rc = __osd_sa_xattr_update(env, obj, oh);
+ /* batch updates only for just created dnodes where we
+ * used to set number of EAs in a single transaction */
+ if (obj->oo_dn->dn_allocated_txg == oh->ot_tx->tx_txg)
+ rc = __osd_sa_xattr_schedule_update(env, obj, oh);
+ else
+ rc = __osd_sa_xattr_update(env, obj, oh);
+
return rc;
}
la->la_valid = LA_MODE;
la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
- rc = __osd_zap_create(env, osd, &xa_zap_dn, tx, la, 0);
+ rc = __osd_zap_create(env, osd, &xa_zap_dn, tx, la, 0, 0);
if (rc)
return rc;
la->la_valid = LA_MODE;
la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
- rc = __osd_object_create(env, obj, &xa_data_dn, tx, la);
+ rc = __osd_object_create(env, osd, obj,
+ lu_object_fid(&obj->oo_dt.do_lu),
+ &xa_data_dn, tx, la);
if (rc)
goto out;
xa_data_obj = xa_data_dn->dn_object;
return rc;
}
+static int osd_xattr_split_pfid(const struct lu_env *env,
+ struct osd_object *obj, struct osd_thandle *oh)
+{
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lustre_ost_attrs *loa =
+ (struct lustre_ost_attrs *)&info->oti_buf;
+ struct lustre_mdt_attrs *lma = &loa->loa_lma;
+ struct lu_buf buf = {
+ .lb_buf = loa,
+ .lb_len = sizeof(info->oti_buf),
+ };
+ int size;
+ int rc;
+ ENTRY;
+
+ BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*loa));
+ rc = osd_xattr_get_internal(env, obj, &buf, XATTR_NAME_LMA, &size);
+ if (rc)
+ RETURN(rc);
+
+ lustre_loa_swab(loa, true);
+ LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
+
+ lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
+ lustre_lma_swab(lma);
+ buf.lb_buf = lma;
+ buf.lb_len = sizeof(*lma);
+ rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
+ LU_XATTR_REPLACE, oh);
+ if (!rc)
+ obj->oo_pfid_in_lma = 0;
+
+ RETURN(rc);
+}
+
+/*
+ * In DNE environment, the object (in spite of regular file or directory)
+ * and its name entry may reside on different MDTs. Under such case, we will
+ * create an agent entry on the MDT where the object resides. The agent entry
+ * references the object locally, that makes the object to be visible to the
+ * userspace when mounted as 'zfs' directly. Then the userspace tools, such
+ * as 'tar' can handle the object properly.
+ *
+ * We handle the agent entry during set linkEA that is the common interface
+ * for both regular file and directroy, can handle kinds of cases, such as
+ * create/link/unlink/rename, and so on.
+ *
+ * NOTE: we need to do that for both directory and regular file, so we can NOT
+ * do that when ea_{insert,delete} that are directory based operations.
+ */
+static int osd_xattr_handle_linkea(const struct lu_env *env,
+ struct osd_device *osd,
+ struct osd_object *obj,
+ const struct lu_buf *buf,
+ struct osd_thandle *oh)
+{
+ const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
+ struct lu_fid *tfid = &osd_oti_get(env)->oti_fid;
+ struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
+ struct lu_name tmpname;
+ int rc;
+ bool remote = false;
+ ENTRY;
+
+ rc = linkea_init_with_rec(&ldata);
+ if (!rc) {
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL && !remote) {
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+ &tmpname, tfid);
+ if (osd_remote_fid(env, osd, tfid) > 0)
+ remote = true;
+ else
+ linkea_next_entry(&ldata);
+ }
+ } else if (rc == -ENODATA) {
+ rc = 0;
+ } else {
+ RETURN(rc);
+ }
+
+ if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
+ rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
+ if (rc)
+ CERROR("%s: failed to remove agent entry for "DFID
+ ": rc = %d\n", osd_name(osd), PFID(fid), rc);
+ } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
+ rc = osd_add_to_remote_parent(env, osd, obj, oh);
+ if (rc)
+ CWARN("%s: failed to create agent entry for "DFID
+ ": rc = %d\n", osd_name(osd), PFID(fid), rc);
+ }
+
+ RETURN(rc);
+}
+
int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, const char *name, int fl,
struct thandle *handle)
{
- struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
int rc = 0;
ENTRY;
strcmp(name, XATTR_NAME_POSIX_ACL_DEFAULT) == 0))
RETURN(-EOPNOTSUPP);
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
down_write(&obj->oo_guard);
CDEBUG(D_INODE, "Setting xattr %s with size %d\n",
name, (int)buf->lb_len);
- rc = osd_xattr_set_internal(env, obj, buf, name, fl, oh);
+ /* For the OST migrated from ldiskfs, the PFID EA may
+ * be stored in LMA because of ldiskfs inode size. */
+ if (unlikely(strcmp(name, XATTR_NAME_FID) == 0 &&
+ obj->oo_pfid_in_lma)) {
+ rc = osd_xattr_split_pfid(env, obj, oh);
+ if (!rc)
+ fl = LU_XATTR_CREATE;
+ } else if (strcmp(name, XATTR_NAME_LINK) == 0 &&
+ osd->od_remote_parent_dir != ZFS_NO_OBJECT) {
+ rc = osd_xattr_handle_linkea(env, osd, obj, buf, oh);
+ }
+
+ if (!rc)
+ rc = osd_xattr_set_internal(env, obj, buf, name, fl, oh);
up_write(&obj->oo_guard);
RETURN(rc);
LASSERT(handle != NULL);
LASSERT(osd_invariant(obj));
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
LASSERT(obj->oo_dn != NULL);
return rc;
rc = -nvlist_remove(obj->oo_sa_xattr, name, DATA_TYPE_BYTE_ARRAY);
- if (rc == 0)
+ if (rc)
+ return rc;
+
+ /*
+ * only migrate delete LMV, and it needs to be done immediately, because
+ * it's used in deleting sub stripes, and if this is delayed, later when
+ * destroying the master object, it will delete sub stripes again.
+ */
+ if (!strcmp(name, XATTR_NAME_LMV))
rc = __osd_sa_xattr_update(env, obj, oh);
+ else
+ rc = __osd_sa_xattr_schedule_update(env, obj, oh);
return rc;
}
LASSERT(obj->oo_dn != NULL);
LASSERT(osd_invariant(obj));
LASSERT(dt_object_exists(dt));
- oh = container_of0(handle, struct osd_thandle, ot_super);
+ oh = container_of(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
if (!osd_obj2dev(obj)->od_posix_acl &&
RETURN(-EOPNOTSUPP);
down_write(&obj->oo_guard);
- rc = __osd_xattr_del(env, obj, name, oh);
+ /* For the OST migrated from ldiskfs, the PFID EA may
+ * be stored in LMA because of ldiskfs inode size. */
+ if (unlikely(strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma))
+ rc = osd_xattr_split_pfid(env, obj, oh);
+ else
+ rc = __osd_xattr_del(env, obj, name, oh);
up_write(&obj->oo_guard);
RETURN(rc);