*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2014, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*
* lustre/ofd/ofd_objects.c
*
- * Author: Alex Zhuravlev <bzzz@whamcloud.com>
- * Author: Mikhail Pershin <tappro@whamcloud.com>
+ * This file contains OSD API methods related to OBD Filter Device (OFD)
+ * object operations.
+ *
+ * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
+ * Author: Mikhail Pershin <mike.pershin@intel.com>
*/
#define DEBUG_SUBSYSTEM S_FILTER
#include "ofd_internal.h"
-int ofd_version_get_check(struct ofd_thread_info *info,
- struct ofd_object *fo)
+/**
+ * Get object version from disk and check it.
+ *
+ * This function checks object version from disk with
+ * ofd_thread_info::fti_pre_version filled from incoming RPC. This is part of
+ * VBR (Version-Based Recovery) and ensures that object has the same version
+ * upon replay as it has during original modification.
+ *
+ * \param[in] info execution thread OFD private data
+ * \param[in] fo OFD object
+ *
+ * \retval 0 if version matches
+ * \retval -EOVERFLOW on version mismatch
+ */
+static int ofd_version_get_check(struct ofd_thread_info *info,
+ struct ofd_object *fo)
{
dt_obj_version_t curr_version;
LASSERT(ofd_object_exists(fo));
- LASSERT(info->fti_exp);
+
+ if (info->fti_exp == NULL)
+ RETURN(0);
curr_version = dt_version_get(info->fti_env, ofd_object_child(fo));
if ((__s64)curr_version == -EOPNOTSUPP)
/* VBR: version is checked always because costs nothing */
if (info->fti_pre_version != 0 &&
info->fti_pre_version != curr_version) {
- CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n",
+ CDEBUG(D_INODE, "Version mismatch %#llx != %#llx\n",
info->fti_pre_version, curr_version);
spin_lock(&info->fti_exp->exp_lock);
info->fti_exp->exp_vbr_failed = 1;
RETURN(0);
}
+/**
+ * Get OFD object by FID.
+ *
+ * This function finds OFD slice of compound object with the given FID.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[in] fid FID of the object
+ *
+ * \retval pointer to the found ofd_object
+ * \retval ERR_PTR(errno) in case of error
+ */
struct ofd_object *ofd_object_find(const struct lu_env *env,
struct ofd_device *ofd,
const struct lu_fid *fid)
RETURN(fo);
}
-struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
- struct ofd_device *ofd,
- const struct lu_fid *fid,
- struct lu_attr *attr)
+/**
+ * Get FID of parent MDT object.
+ *
+ * This function reads extended attribute XATTR_NAME_FID of OFD object which
+ * contains the MDT parent object FID and saves it in ofd_object::ofo_pfid.
+ *
+ * The filter_fid::ff_parent::f_ver field currently holds
+ * the OST-object index in the parent MDT-object's layout EA,
+ * not the actual FID::f_ver of the parent. We therefore access
+ * it via the macro f_stripe_idx.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ *
+ * \retval 0 if successful
+ * \retval -ENODATA if there is no such xattr
+ * \retval negative value on error
+ */
+int ofd_object_ff_load(const struct lu_env *env, struct ofd_object *fo)
{
struct ofd_thread_info *info = ofd_info(env);
- struct lu_object *fo_obj;
- struct dt_object *dto;
+ struct filter_fid_old *ff = &info->fti_mds_fid_old;
+ struct lu_buf *buf = &info->fti_buf;
+ struct lu_fid *pfid = &fo->ofo_pfid;
+ int rc = 0;
- ENTRY;
+ if (fid_is_sane(pfid))
+ return 0;
- info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
-
- dto = dt_find_or_create(env, ofd->ofd_osd, fid, &info->fti_dof, attr);
- if (IS_ERR(dto))
- RETURN(ERR_CAST(dto));
+ buf->lb_buf = ff;
+ buf->lb_len = sizeof(*ff);
+ rc = dt_xattr_get(env, ofd_object_child(fo), buf, XATTR_NAME_FID);
+ if (rc < 0)
+ return rc;
- fo_obj = lu_object_locate(dto->do_lu.lo_header,
- ofd->ofd_dt_dev.dd_lu_dev.ld_type);
- RETURN(ofd_obj(fo_obj));
-}
+ if (rc < sizeof(struct lu_fid)) {
+ fid_zero(pfid);
-int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo)
-{
- int rc = 0;
-
- ENTRY;
-
- if (!fo->ofo_ff_exists) {
- /*
- * This actually means that we don't know whether the object
- * has the "fid" EA or not.
- */
- rc = dt_xattr_get(env, ofd_object_child(fo), &LU_BUF_NULL,
- XATTR_NAME_FID, BYPASS_CAPA);
- if (rc >= 0 || rc == -ENODATA) {
- /*
- * Here we assume that, if the object doesn't have the
- * "fid" EA, the caller will add one, unless a fatal
- * error (e.g., a memory or disk failure) prevents it
- * from doing so.
- */
- fo->ofo_ff_exists = 1;
- }
- if (rc > 0)
- rc = 0;
+ return -ENODATA;
}
- RETURN(rc);
-}
-void ofd_object_put(const struct lu_env *env, struct ofd_object *fo)
-{
- lu_object_put(env, &fo->ofo_obj.do_lu);
+ pfid->f_seq = le64_to_cpu(ff->ff_parent.f_seq);
+ pfid->f_oid = le32_to_cpu(ff->ff_parent.f_oid);
+ pfid->f_stripe_idx = le32_to_cpu(ff->ff_parent.f_stripe_idx);
+
+ return 0;
}
+/**
+ * Precreate the given number \a nr of objects in the given sequence \a oseq.
+ *
+ * This function precreates new OST objects in the given sequence.
+ * The precreation starts from \a id and creates \a nr objects sequentially.
+ *
+ * Notes:
+ * This function may create fewer objects than requested.
+ *
+ * We mark object SUID+SGID to flag it for accepting UID+GID from client on
+ * first write. Currently the permission bits on the OST are never used,
+ * so this is OK.
+ *
+ * Initialize a/c/m time so any client timestamp will always be newer and
+ * update the inode. The ctime = 0 case is also handled specially in
+ * osd_inode_setattr(). See LU-221, LU-1042 for details.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[in] id object ID to start precreation from
+ * \param[in] oseq object sequence
+ * \param[in] nr number of objects to precreate
+ * \param[in] sync synchronous precreation flag
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
- obd_id id, struct ofd_seq *oseq, int nr, int sync)
+ u64 id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct thandle *th;
struct ofd_object **batch;
struct lu_fid *fid = &info->fti_fid;
- obd_id tmp;
- int rc;
- int i;
- int objects = 0;
- int nr_saved = nr;
+ u64 tmp;
+ int rc;
+ int rc2;
+ int i;
+ int objects = 0;
+ int nr_saved = nr;
ENTRY;
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
- /*
- * We mark object SUID+SGID to flag it for accepting UID+GID from
- * client on first write. Currently the permission bits on the OST are
- * never used, so this is OK.
- */
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
- /* Initialize a/c/m time so any client timestamp will always
- * be newer and update the inode. ctime = 0 is also handled
- * specially in osd_inode_setattr(). See LU-221, LU-1042 */
info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
- CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
+ CDEBUG(D_INODE, "object %#llx/%#llx exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
continue;
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
- if (rc) {
+ if (rc < 0) {
+ if (i == 0)
+ GOTO(trans_stop, rc);
+
nr = i;
break;
}
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(fid), nr);
- LASSERT(nr > 0);
-
/* When the LFSCK scanning the whole device to verify the LAST_ID file
* consistency, it will load the last_id into RAM firstly, and compare
* the last_id with each OST-object's ID. If the later one is larger,
/* Only the new created objects need to be recorded. */
if (ofd->ofd_osd->dd_record_fid_accessed) {
- lfsck_pack_rfa(&ofd_info(env)->fti_lr,
- lu_object_fid(&fo->ofo_obj.do_lu));
- lfsck_in_notify(env, ofd->ofd_osd,
- &ofd_info(env)->fti_lr);
+ struct lfsck_request *lr = &ofd_info(env)->fti_lr;
+
+ lfsck_pack_rfa(lr, lu_object_fid(&fo->ofo_obj.do_lu),
+ LE_FID_ACCESSED,
+ LFSCK_TYPE_LAYOUT);
+ lfsck_in_notify(env, ofd->ofd_osd, lr, NULL);
}
if (likely(!ofd_object_exists(fo) &&
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
- if (rc)
+ if (rc < 0) {
+ if (i == 0)
+ GOTO(trans_stop, rc);
+
+ rc = 0;
break;
+ }
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc1 != 0)
- CERROR("%s: fail to reset the LAST_ID for seq ("LPX64
- ") from "LPU64" to "LPU64"\n", ofd_name(ofd),
+ CERROR("%s: fail to reset the LAST_ID for seq (%#llx"
+ ") from %llu to %llu\n", ofd_name(ofd),
ostid_seq(&oseq->os_oi), id + nr - 1,
ofd_seq_last_oid(oseq));
}
trans_stop:
- ofd_trans_stop(env, ofd, th, rc);
+ rc2 = ofd_trans_stop(env, ofd, th, rc);
+ if (rc2)
+ CERROR("%s: failed to stop transaction: rc = %d\n",
+ ofd_name(ofd), rc2);
+ if (!rc)
+ rc = rc2;
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
RETURN(objects > 0 ? objects : rc);
}
-/*
- * If the object still has SUID+SGID bits set (see ofd_precreate_object()) then
- * we will accept the UID+GID if sent by the client for initializing the
+/**
+ * Fix the OFD object ownership.
+ *
+ * If the object still has SUID+SGID bits set, meaning that it was precreated
+ * by the MDT before it was assigned to any file, (see ofd_precreate_objects())
+ * then we will accept the UID+GID if sent by the client for initializing the
* ownership of this object. We only allow this to happen once (so clear these
* bits) and later only allow setattr.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] la object attributes
+ * \param[in] is_setattr was this function called from setattr or not
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
*/
int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la, int is_setattr)
if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID))
RETURN(0);
- rc = dt_attr_get(env, ofd_object_child(fo), ln, BYPASS_CAPA);
+ rc = dt_attr_get(env, ofd_object_child(fo), ln);
if (rc != 0)
RETURN(rc);
RETURN(0);
}
+/**
+ * Set OFD object attributes.
+ *
+ * This function sets OFD object attributes taken from incoming request.
+ * It sets not only regular attributes but also XATTR_NAME_FID extended
+ * attribute if needed. The "fid" xattr allows the object's MDT parent inode
+ * to be found and verified by LFSCK and other tools in case of inconsistency.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] la object attributes
+ * \param[in] ff filter_fid structure, contains additional attributes
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la, struct filter_fid *ff)
{
struct ofd_device *ofd = ofd_obj2dev(fo);
struct thandle *th;
struct ofd_mod_data *fmd;
- int ff_needed = 0;
- int rc;
+ int ff_needed = 0;
+ int rc;
+ int rc2;
ENTRY;
ofd_write_lock(env, fo);
GOTO(unlock, rc);
if (ff != NULL) {
- rc = ofd_object_ff_check(env, fo);
+ rc = ofd_object_ff_load(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
if (rc)
GOTO(stop, rc);
- rc = dt_attr_set(env, ofd_object_child(fo), la, th,
- ofd_object_capa(env, fo));
+ rc = dt_attr_set(env, ofd_object_child(fo), la, th);
if (rc)
GOTO(stop, rc);
- if (ff_needed)
+ if (ff_needed) {
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
- XATTR_NAME_FID, 0, th, BYPASS_CAPA);
+ XATTR_NAME_FID, 0, th);
+ if (rc == 0) {
+ fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
+ fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
+ /* Currently, the filter_fid::ff_parent::f_ver is not
+ * the real parent MDT-object's FID::f_ver, instead it
+ * is the OST-object index in its parent MDT-object's
+ * layout EA. */
+ fo->ofo_pfid.f_stripe_idx =
+ le32_to_cpu(ff->ff_parent.f_stripe_idx);
+ }
+ }
+
+ GOTO(stop, rc);
stop:
- ofd_trans_stop(env, ofd, th, rc);
+ rc2 = ofd_trans_stop(env, ofd, th, rc);
+ if (rc2)
+ CERROR("%s: failed to stop transaction: rc = %d\n",
+ ofd_name(ofd), rc2);
+ if (!rc)
+ rc = rc2;
+
unlock:
ofd_write_unlock(env, fo);
- RETURN(rc);
+
+ return rc;
}
+/**
+ * Truncate/punch OFD object.
+ *
+ * This function frees all of the allocated object's space from the \a start
+ * offset to the \a end offset. For truncate() operations the \a end offset
+ * is OBD_OBJECT_EOF. The functionality to punch holes in an object via
+ * fallocate(FALLOC_FL_PUNCH_HOLE) is not yet implemented (see LU-3606).
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] start start offset to punch from
+ * \param[in] end end of punch
+ * \param[in] la object attributes
+ * \param[in] ff filter_fid structure
+ * \param[in] oa obdo struct from incoming request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
__u64 start, __u64 end, struct lu_attr *la,
- struct filter_fid *ff)
+ struct filter_fid *ff, struct obdo *oa)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_device *ofd = ofd_obj2dev(fo);
struct ofd_mod_data *fmd;
struct dt_object *dob = ofd_object_child(fo);
struct thandle *th;
- int ff_needed = 0;
- int rc;
+ int ff_needed = 0;
+ int rc;
+ int rc2;
ENTRY;
/* we support truncate, not punch yet */
LASSERT(end == OBD_OBJECT_EOF);
+ ofd_write_lock(env, fo);
fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
fmd->fmd_mactime_xid = info->fti_xid;
ofd_fmd_put(info->fti_exp, fmd);
- ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
+ if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
+ rc = ofd_verify_ff(env, fo, oa);
+ if (rc != 0)
+ GOTO(unlock, rc);
+ }
+
/* VBR: version recovery check */
rc = ofd_version_get_check(info, fo);
if (rc)
GOTO(unlock, rc);
if (ff != NULL) {
- rc = ofd_object_ff_check(env, fo);
+ rc = ofd_object_ff_load(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
if (rc)
GOTO(stop, rc);
- rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th,
- ofd_object_capa(env, fo));
+ rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th);
if (rc)
GOTO(stop, rc);
- rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo));
+ rc = dt_attr_set(env, dob, la, th);
if (rc)
GOTO(stop, rc);
- if (ff_needed)
+ if (ff_needed) {
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
- XATTR_NAME_FID, 0, th, BYPASS_CAPA);
+ XATTR_NAME_FID, 0, th);
+ if (rc == 0) {
+ fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
+ fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
+ /* Currently, the filter_fid::ff_parent::f_ver is not
+ * the real parent MDT-object's FID::f_ver, instead it
+ * is the OST-object index in its parent MDT-object's
+ * layout EA. */
+ fo->ofo_pfid.f_stripe_idx =
+ le32_to_cpu(ff->ff_parent.f_stripe_idx);
+ }
+ }
+
+ GOTO(stop, rc);
stop:
- ofd_trans_stop(env, ofd, th, rc);
+ rc2 = ofd_trans_stop(env, ofd, th, rc);
+ if (rc2 != 0)
+ CERROR("%s: failed to stop transaction: rc = %d\n",
+ ofd_name(ofd), rc2);
+ if (!rc)
+ rc = rc2;
unlock:
ofd_write_unlock(env, fo);
- RETURN(rc);
+
+ return rc;
}
+/**
+ * Destroy OFD object.
+ *
+ * This function destroys OFD object. If object wasn't used at all (orphan)
+ * then local transaction is used, which means the transaction data is not
+ * returned back in reply.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] orphan flag to indicate that object is orphaned
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo,
int orphan)
{
struct ofd_device *ofd = ofd_obj2dev(fo);
struct thandle *th;
- int rc = 0;
+ int rc = 0;
+ int rc2;
ENTRY;
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
- dt_declare_ref_del(env, ofd_object_child(fo), th);
- dt_declare_destroy(env, ofd_object_child(fo), th);
+ rc = dt_declare_ref_del(env, ofd_object_child(fo), th);
+ if (rc < 0)
+ GOTO(stop, rc);
+
+ rc = dt_declare_destroy(env, ofd_object_child(fo), th);
+ if (rc < 0)
+ GOTO(stop, rc);
+
if (orphan)
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
else
dt_ref_del(env, ofd_object_child(fo), th);
dt_destroy(env, ofd_object_child(fo), th);
stop:
- ofd_trans_stop(env, ofd, th, rc);
+ rc2 = ofd_trans_stop(env, ofd, th, rc);
+ if (rc2)
+ CERROR("%s failed to stop transaction: %d\n",
+ ofd_name(ofd), rc2);
+ if (!rc)
+ rc = rc2;
unlock:
ofd_write_unlock(env, fo);
RETURN(rc);
}
+/**
+ * Get OFD object attributes.
+ *
+ * This function gets OFD object regular attributes. It is used to serve
+ * incoming request as well as for local OFD purposes.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] la object attributes
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la)
{
ENTRY;
if (ofd_object_exists(fo)) {
- rc = dt_attr_get(env, ofd_object_child(fo), la,
- ofd_object_capa(env, fo));
-
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
- /* Try to correct for a bug in 2.1.0 (LU-221) that caused
- * negative timestamps to appear to be in the far future,
- * due old timestamp being stored on disk as an unsigned value.
- * This fixes up any bad values stored on disk before
- * returning them to the client, and ensures any timestamp
- * updates are correct. LU-1042 */
- if (unlikely(la->la_atime == LU221_BAD_TIME))
- la->la_atime = 0;
- if (unlikely(la->la_mtime == LU221_BAD_TIME))
- la->la_mtime = 0;
- if (unlikely(la->la_ctime == LU221_BAD_TIME))
- la->la_ctime = 0;
-#else
-#warning "remove old LU-221/LU-1042 workaround code"
-#endif
+ rc = dt_attr_get(env, ofd_object_child(fo), la);
} else {
rc = -ENOENT;
}