#include <dt_object.h>
#include <lustre/lustre_idl.h>
+#include <lustre_lfsck.h>
#include "ofd_internal.h"
dt_obj_version_t curr_version;
LASSERT(ofd_object_exists(fo));
- LASSERT(info->fti_exp);
+
+ if (info->fti_exp)
+ RETURN(0);
curr_version = dt_version_get(info->fti_env, ofd_object_child(fo));
if ((__s64)curr_version == -EOPNOTSUPP)
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)
+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;
-
- info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
+ if (fid_is_sane(pfid))
+ return 0;
- 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,
+ BYPASS_CAPA);
+ 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;
+ return -ENODATA;
+ }
- ENTRY;
+ pfid->f_seq = le64_to_cpu(ff->ff_parent.f_seq);
+ 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. */
+ pfid->f_stripe_idx = le32_to_cpu(ff->ff_parent.f_stripe_idx);
- 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(rc);
+ return 0;
}
void ofd_object_put(const struct lu_env *env, struct ofd_object *fo)
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
+ struct lu_fid *fid = &info->fti_fid;
obd_id tmp;
int rc;
int i;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
+ LASSERT(id != 0);
+
/* prepare objects */
- ostid_set_seq(&info->fti_ostid, ostid_seq(&oseq->os_oi));
+ *fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
for (i = 0; i < nr; i++) {
- ostid_set_id(&info->fti_ostid, id + i);
- rc = ostid_to_fid(&info->fti_fid, &info->fti_ostid, 0);
- if (rc) {
+ rc = fid_set_id(fid, id + i);
+ if (rc != 0) {
if (i == 0)
GOTO(out, rc);
break;
}
- fo = ofd_object_find(env, ofd, &info->fti_fid);
+ fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
th->th_sync |= sync;
- rc = dt_declare_record_write(env, oseq->os_lastid_obj, sizeof(tmp),
+ rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf,
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
- PFID(&info->fti_fid));
+ PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
continue;
}
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
- ofd_name(ofd), PFID(&info->fti_fid), nr);
+ 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,
+ * then it will regard the LAST_ID file crashed. But during the LFSCK
+ * scanning, the OFD may continue to create new OST-objects. Those new
+ * created OST-objects will have larger IDs than the LFSCK known ones.
+ * So from the LFSCK view, it needs to re-load the last_id from disk
+ * file, and if the latest last_id is still smaller than the object's
+ * ID, then the LAST_ID file is real crashed.
+ *
+ * To make above mechanism to work, before OFD pre-create OST-objects,
+ * it needs to update the LAST_ID file firstly, otherwise, the LFSCK
+ * may cannot get latest last_id although new OST-object created. */
+ if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
+ tmp = cpu_to_le64(id + nr - 1);
+ dt_write_lock(env, oseq->os_lastid_obj, 0);
+ rc = dt_record_write(env, oseq->os_lastid_obj,
+ &info->fti_buf, &info->fti_off, th);
+ dt_write_unlock(env, oseq->os_lastid_obj);
+ if (rc != 0)
+ GOTO(trans_stop, rc);
+ }
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
- if (likely(!ofd_object_exists(fo))) {
+ /* 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);
+ }
+
+ if (likely(!ofd_object_exists(fo) &&
+ !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
}
objects = i;
- if (objects > 0) {
+ /* NOT all the wanted objects have been created,
+ * set the LAST_ID as the real created. */
+ if (unlikely(objects < nr)) {
+ int rc1;
+
+ info->fti_off = 0;
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
- rc = dt_record_write(env, oseq->os_lastid_obj,
- &info->fti_buf, &info->fti_off, th);
+ dt_write_lock(env, oseq->os_lastid_obj, 0);
+ rc1 = dt_record_write(env, oseq->os_lastid_obj,
+ &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),
+ ostid_seq(&oseq->os_oi), id + nr - 1,
+ ofd_seq_last_oid(oseq));
}
+
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
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);
- 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);
+ 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);
unlock:
ofd_write_unlock(env, fo);
- RETURN(rc);
+
+ return rc;
}
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);
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);
- 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);
+ 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);
unlock:
ofd_write_unlock(env, fo);
- RETURN(rc);
+
+ return rc;
}
int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo,