__u64 llsd_touch_gen;
struct dt_object *llsd_rb_obj;
struct rb_root llsd_rb_root;
- rwlock_t llsd_rb_lock;
+ struct rw_semaphore llsd_rb_rwsem;
unsigned int llsd_rbtree_valid:1;
};
int rc = 0;
ENTRY;
- if (!lad->lad_incomplete)
+ if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
RETURN_EXIT;
/* If the MDT has ever failed to verfiy some OST-objects,
up_read(<ds->ltd_rw_sem);
if (rc == 0 && atomic_read(&count) > 0)
- rc = ptlrpc_set_wait(set);
+ rc = ptlrpc_set_wait(env, set);
ptlrpc_set_destroy(set);
}
static int lfsck_layout_verify_header_v1v3(struct dt_object *obj,
- struct lov_mds_md_v1 *lmm)
+ struct lov_mds_md_v1 *lmm,
+ __u64 start, __u32 comp_id)
{
__u32 magic;
__u32 pattern;
}
pattern = le32_to_cpu(lmm->lmm_pattern);
- /* XXX: currently, we only support LOV_PATTERN_RAID0. */
- if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
+
+#if 0
+ /* XXX: DoM file verification will be supportted via LU-11081. */
+ if (lov_pattern(pattern) == LOV_PATTERN_MDT) {
+ if (start != 0) {
+ CDEBUG(D_LFSCK, "The DoM entry for "DFID" is not "
+ "the first component in the mirror %x/%llu\n",
+ PFID(lfsck_dto2fid(obj)), comp_id, start);
+
+ return -EINVAL;
+ }
+ }
+#endif
+
+ if (!lov_pattern_supported_normal_comp(lov_pattern(pattern))) {
CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u for the file "
- DFID"\n", pattern, PFID(lfsck_dto2fid(obj)));
+ DFID" in the component %x\n",
+ pattern, PFID(lfsck_dto2fid(obj)), comp_id);
return -EOPNOTSUPP;
}
return 0;
}
+static int lfsck_layout_verify_header_foreign(struct dt_object *obj,
+ struct lov_foreign_md *lfm,
+ size_t len)
+{
+ /* magic has been verified already */
+ __u32 value_len = le32_to_cpu(lfm->lfm_length);
+ /* type and flags are not checked for instance */
+
+ CDEBUG(D_INFO, "foreign LOV EA, magic %x, len %u, type %x, flags %x, for file "DFID"\n",
+ le32_to_cpu(lfm->lfm_magic), value_len,
+ le32_to_cpu(lfm->lfm_type), le32_to_cpu(lfm->lfm_flags),
+ PFID(lfsck_dto2fid(obj)));
+
+ if (len != value_len + offsetof(typeof(*lfm), lfm_value))
+ CDEBUG(D_LFSCK, "foreign LOV EA internal size %u does not match EA full size %zu for file "DFID"\n",
+ value_len, len, PFID(lfsck_dto2fid(obj)));
+
+ /* nothing to repair */
+ return -ENODATA;
+}
+
static int lfsck_layout_verify_header(struct dt_object *obj,
- struct lov_mds_md_v1 *lmm)
+ struct lov_mds_md_v1 *lmm, size_t len)
{
int rc = 0;
return -EINVAL;
}
- for (i = 0; i < count; i++) {
+ for (i = 0; i < count && !rc; i++) {
struct lov_comp_md_entry_v1 *lcme =
&lcm->lcm_entries[i];
__u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
}
rc = lfsck_layout_verify_header_v1v3(obj,
- (struct lov_mds_md_v1 *)((char *)lmm +
- le32_to_cpu(lcme->lcme_offset)));
- if (rc)
- return rc;
+ (struct lov_mds_md_v1 *)((char *)lmm +
+ le32_to_cpu(lcme->lcme_offset)), start,
+ comp_id);
}
+ } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_FOREIGN) {
+ rc = lfsck_layout_verify_header_foreign(obj,
+ (struct lov_foreign_md *)lmm,
+ len);
} else {
- rc = lfsck_layout_verify_header_v1v3(obj, lmm);
+ rc = lfsck_layout_verify_header_v1v3(obj, lmm, 1, 0);
}
return rc;
if (rc == -ERANGE) {
rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
if (rc <= 0)
- return rc;
+ return !rc ? -ENODATA : rc;
lu_buf_realloc(buf, rc);
if (buf->lb_buf == NULL)
goto again;
}
- if (rc == -ENODATA)
- rc = 0;
-
if (rc <= 0)
- return rc;
+ return !rc ? -ENODATA : rc;
if (unlikely(buf->lb_buf == NULL)) {
lu_buf_alloc(buf, rc);
goto again;
}
- rc1 = lfsck_layout_verify_header(obj, buf->lb_buf);
+ rc1 = lfsck_layout_verify_header(obj, buf->lb_buf, rc);
return rc1 ? rc1 : rc;
}
lfsck->li_bottom->dd_record_fid_accessed = 0;
/* Invalid the rbtree, then no others will use it. */
- write_lock(&llsd->llsd_rb_lock);
+ down_write(&llsd->llsd_rb_rwsem);
llsd->llsd_rbtree_valid = 0;
- write_unlock(&llsd->llsd_rb_lock);
+ up_write(&llsd->llsd_rb_rwsem);
while (node != NULL) {
next = rb_next(node);
if (!fid_is_idif(fid) && !fid_is_norm(fid))
RETURN_EXIT;
- read_lock(&llsd->llsd_rb_lock);
+ down_read(&llsd->llsd_rb_rwsem);
if (!llsd->llsd_rbtree_valid)
GOTO(unlock, rc = 0);
LASSERT(!insert);
- read_unlock(&llsd->llsd_rb_lock);
+ up_read(&llsd->llsd_rb_rwsem);
tmp = lfsck_rbtree_new(env, fid);
if (IS_ERR(tmp))
GOTO(out, rc = PTR_ERR(tmp));
insert = true;
- write_lock(&llsd->llsd_rb_lock);
+ down_write(&llsd->llsd_rb_rwsem);
if (!llsd->llsd_rbtree_valid) {
lfsck_rbtree_free(tmp);
GOTO(unlock, rc = 0);
unlock:
if (insert)
- write_unlock(&llsd->llsd_rb_lock);
+ up_write(&llsd->llsd_rb_rwsem);
else
- read_unlock(&llsd->llsd_rb_lock);
+ up_read(&llsd->llsd_rb_rwsem);
out:
if (rc != 0 && accessed) {
struct lfsck_layout *lo = com->lc_file_ram;
}
if (lo->ll_bitmap_size == 0) {
- lad->lad_incomplete = 0;
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
CFS_RESET_BITMAP(bitmap);
RETURN(0);
RETURN(rc >= 0 ? -EINVAL : rc);
if (cfs_bitmap_check_empty(bitmap))
- lad->lad_incomplete = 0;
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
else
- lad->lad_incomplete = 1;
+ set_bit(LAD_INCOMPLETE, &lad->lad_flags);
RETURN(0);
}
loa = &lfsck_env_info(env)->lti_loa;
rc = dt_xattr_get(env, obj, lfsck_buf_get(env, loa, sizeof(*loa)),
XATTR_NAME_LMA);
- if (rc >= sizeof(struct lustre_mdt_attrs)) {
+ if (rc >= (int)sizeof(struct lustre_mdt_attrs)) {
lustre_lma_swab(&loa->loa_lma);
return loa->loa_lma.lma_compat & LMAC_FID_ON_OST ? 1 : 0;
if (lfsck->li_master) {
struct lfsck_assistant_data *lad = com->lc_data;
- if (lad->lad_incomplete)
+ if (test_bit(LAD_INCOMPLETE, &lad->lad_flags))
lo->ll_status = LS_PARTIAL;
else
lo->ll_status = LS_COMPLETED;
GOTO(unlock, rc);
rc = dt_insert(env, obj, (const struct dt_rec *)rec,
- (const struct dt_key *)key, th, 1);
+ (const struct dt_key *)key, th);
GOTO(unlock, rc);
}
static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
- struct ost_layout *ol,
- struct dt_object *parent,
- struct lu_buf *buf, __u32 ea_off,
- struct lov_mds_md_v1 **lmm,
- struct lov_ost_data_v1 **objs)
+ struct lu_orphan_rec_v3 *rec,
+ struct dt_object *parent,
+ struct lu_buf *buf, __u32 ea_off,
+ struct lov_mds_md_v1 **lmm,
+ struct lov_ost_data_v1 **objs)
{
+ struct ost_layout *ol = &rec->lor_layout;
struct lov_comp_md_v1 *lcm;
struct lov_comp_md_entry_v1 *lcme;
__u32 pattern = LOV_PATTERN_RAID0;
lcm = buf->lb_buf;
lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
lcm->lcm_size = cpu_to_le32(size);
- lcm->lcm_layout_gen = cpu_to_le32(1);
- lcm->lcm_flags = 0;
+ if (rec->lor_range) {
+ lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
+ rec->lor_range);
+ lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
+ } else if (rec->lor_layout_version) {
+ lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
+ rec->lor_range);
+ lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
+ } else {
+ lcm->lcm_layout_gen = cpu_to_le32(1);
+ lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
+ }
lcm->lcm_entry_count = cpu_to_le16(1);
+ /* Currently, we do not know how many mirrors will be, set it as zero
+ * at the beginning. It will be updated when more mirrors are found. */
+ lcm->lcm_mirror_count = 0;
lcme = &lcm->lcm_entries[0];
lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
lcme->lcme_offset = cpu_to_le32(offset);
lcme->lcme_size = cpu_to_le32(lcme_size);
+ lcme->lcme_layout_gen = lcm->lcm_layout_gen;
if (ol->ol_stripe_count > 1)
pattern |= LOV_PATTERN_F_HOLE;
return size;
}
-static int lfsck_layout_add_comp_comp(const struct lu_env *env,
- struct lfsck_instance *lfsck,
- struct thandle *handle,
- struct ost_layout *ol,
- struct dt_object *parent,
- const struct lu_fid *cfid,
- struct lu_buf *buf, __u32 ost_idx,
- __u32 ea_off, int pos)
+static void lfsck_layout_update_lcm(struct lov_comp_md_v1 *lcm,
+ struct lov_comp_md_entry_v1 *lcme,
+ __u32 version, __u32 range)
{
+ struct lov_comp_md_entry_v1 *tmp;
+ __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
+ __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
+ __u32 gen = version + range;
+ __u32 tmp_gen;
+ int i;
+ __u16 count = le16_to_cpu(lcm->lcm_entry_count);
+ __u16 flags = le16_to_cpu(lcm->lcm_flags);
+
+ if (!gen)
+ gen = 1;
+ lcme->lcme_layout_gen = cpu_to_le32(gen);
+ if (le32_to_cpu(lcm->lcm_layout_gen) < gen)
+ lcm->lcm_layout_gen = cpu_to_le32(gen);
+
+ if (range)
+ lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
+ else if (flags == LCM_FL_NONE && le16_to_cpu(lcm->lcm_mirror_count) > 0)
+ lcm->lcm_flags = cpu_to_le16(LCM_FL_RDONLY);
+
+ for (i = 0; i < count; i++) {
+ tmp = &lcm->lcm_entries[i];
+ if (le64_to_cpu(tmp->lcme_extent.e_end) <= start)
+ continue;
+
+ if (le64_to_cpu(tmp->lcme_extent.e_start) >= end)
+ continue;
+
+ if (le32_to_cpu(tmp->lcme_flags) & LCME_FL_STALE)
+ continue;
+
+ tmp_gen = le32_to_cpu(tmp->lcme_layout_gen);
+ /* "lcme_layout_gen == 0" but without LCME_FL_STALE flag,
+ * then it should be the latest version of all mirrors. */
+ if (tmp_gen == 0 || tmp_gen > gen) {
+ lcme->lcme_flags = cpu_to_le32(
+ le32_to_cpu(lcme->lcme_flags) | LCME_FL_STALE);
+ break;
+ }
+
+ if (tmp_gen < gen)
+ tmp->lcme_flags = cpu_to_le32(
+ le32_to_cpu(tmp->lcme_flags) | LCME_FL_STALE);
+ }
+}
+
+static int lfsck_layout_add_comp(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct thandle *handle,
+ struct lu_orphan_rec_v3 *rec,
+ struct dt_object *parent,
+ const struct lu_fid *cfid,
+ struct lu_buf *buf, __u32 ost_idx,
+ __u32 ea_off, int pos, bool new_mirror)
+{
+ struct ost_layout *ol = &rec->lor_layout;
struct lov_comp_md_v1 *lcm = buf->lb_buf;
struct lov_comp_md_entry_v1 *lcme;
struct lov_mds_md_v1 *lmm;
* have reallocated the buf. */
lcm = buf->lb_buf;
lcm->lcm_size = cpu_to_le32(size);
- le32_add_cpu(&lcm->lcm_layout_gen, 1);
lcm->lcm_entry_count = cpu_to_le16(count + 1);
+ if (new_mirror)
+ le16_add_cpu(&lcm->lcm_mirror_count, 1);
/* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
* with distance of 'added'. */
ol->ol_stripe_size, ea_off,
pattern, ol->ol_stripe_count);
+ /* 6. Update mirror related flags and version. */
+ lfsck_layout_update_lcm(lcm, lcme, rec->lor_layout_version,
+ rec->lor_range);
+
rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
lmm, objs, LU_XATTR_REPLACE, ost_idx,
le32_to_cpu(lcm->lcm_size));
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
DFID": parent "DFID", OST-index %u, stripe-index %u, "
"stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
- "comp_end %llu, %s LOV EA hole: rc = %d\n",
+ "comp_end %llu, layout version %u, range %u, "
+ "%s LOV EA hole: rc = %d\n",
lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
+ rec->lor_layout_version, rec->lor_range,
le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
"with" : "without", rc);
static int lfsck_layout_update_lovea(const struct lu_env *env,
struct lfsck_instance *lfsck,
struct thandle *handle,
- struct ost_layout *ol,
+ struct lu_orphan_rec_v3 *rec,
struct dt_object *parent,
const struct lu_fid *cfid,
struct lu_buf *buf, int fl,
__u32 ost_idx, __u32 ea_off)
{
+ struct ost_layout *ol = &rec->lor_layout;
struct lov_mds_md_v1 *lmm = NULL;
struct lov_ost_data_v1 *objs = NULL;
int rc = 0;
ENTRY;
if (ol->ol_comp_id != 0)
- rc = lfsck_layout_new_comp_lovea(env, ol, parent, buf, ea_off,
- &lmm, &objs);
+ rc = lfsck_layout_new_comp_lovea(env, rec, parent, buf, ea_off,
+ &lmm, &objs);
else
- rc = lfsck_layout_new_v1_lovea(env, lfsck, ol, parent, buf,
- ea_off, &lmm, &objs);
-
+ rc = lfsck_layout_new_v1_lovea(env, lfsck, &rec->lor_layout,
+ parent, buf, ea_off, &lmm,
+ &objs);
if (rc > 0)
rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
buf, lmm, objs, fl, ost_idx, rc);
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
DFID": parent "DFID", OST-index %u, stripe-index %u, "
"stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
- "comp_end %llu, fl %d, %s LOV EA hole: rc = %d\n",
+ "comp_end %llu, layout version %u, range %u, fl %d, "
+ "%s LOV EA hole: rc = %d\n",
lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
- ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end, fl,
+ ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
+ rec->lor_layout_version, rec->lor_range, fl,
le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
"with" : "without", rc);
static int __lfsck_layout_update_pfid(const struct lu_env *env,
struct dt_object *child,
const struct lu_fid *pfid,
- const struct ost_layout *ol, __u32 offset)
+ const struct ost_layout *ol, __u32 offset,
+ __u32 version, __u32 range)
{
struct dt_device *dev = lfsck_obj2dev(child);
struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
* parent MDT-object's layout EA. */
ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
ost_layout_cpu_to_le(&ff->ff_layout, ol);
+ ff->ff_layout_version = cpu_to_le32(version);
+ ff->ff_range = cpu_to_le32(range);
lfsck_buf_init(&buf, ff, sizeof(*ff));
handle = dt_trans_create(env, dev);
struct dt_object *parent,
struct lu_fid *cfid,
struct dt_device *cdev,
- struct ost_layout *ol, __u32 ea_off)
+ struct lu_orphan_rec_v3 *rec, __u32 ea_off)
{
struct dt_object *child;
int rc = 0;
rc = __lfsck_layout_update_pfid(env, child,
lu_object_fid(&parent->do_lu),
- ol, ea_off);
+ &rec->lor_layout, ea_off,
+ rec->lor_layout_version,
+ rec->lor_range);
lfsck_object_put(env, child);
RETURN(rc == 0 ? 1 : rc);
static int lfsck_layout_recreate_parent(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_tgt_desc *ltd,
- struct lu_orphan_rec_v2 *rec,
+ struct lu_orphan_rec_v3 *rec,
struct lu_fid *cfid,
const char *infix,
const char *type,
struct lu_attr *la = &info->lti_la2;
struct dt_object_format *dof = &info->lti_dof;
struct lfsck_instance *lfsck = com->lc_lfsck;
- struct ost_layout *ol = &rec->lor_layout;
struct lu_fid *pfid = &rec->lor_rec.lor_fid;
struct lu_fid *tfid = &info->lti_fid3;
struct dt_device *dev = lfsck->li_bottom;
* the stripe(s). The LFSCK will specify the LOV EA via
* lfsck_layout_update_lovea(). */
- size = lfsck_lovea_size(ol, ea_off);
+ size = lfsck_lovea_size(&rec->lor_layout, ea_off);
if (ea_buf->lb_len < size) {
lu_buf_realloc(ea_buf, size);
if (ea_buf->lb_buf == NULL)
dt_write_lock(env, pobj, 0);
rc = dt_create(env, pobj, la, NULL, dof, th);
if (rc == 0)
- rc = lfsck_layout_update_lovea(env, lfsck, th, ol, pobj, cfid,
+ rc = lfsck_layout_update_lovea(env, lfsck, th, rec, pobj, cfid,
&lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
dt_write_unlock(env, pobj);
if (rc < 0)
GOTO(stop, rc);
rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)name, th);
if (rc != 0)
GOTO(stop, rc);
th = NULL;
/* The 2nd transaction. */
- rc = __lfsck_layout_update_pfid(env, cobj, pfid, ol, ea_off);
+ rc = __lfsck_layout_update_pfid(env, cobj, pfid,
+ &rec->lor_layout, ea_off,
+ rec->lor_layout_version,
+ rec->lor_range);
}
GOTO(stop, rc);
memset(policy, 0, sizeof(*policy));
policy->l_extent.end = OBD_OBJECT_EOF;
ost_fid_build_resid(fid, resid);
- rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
- policy, LCK_EX, &flags, ldlm_blocking_ast,
- ldlm_completion_ast, NULL, NULL, 0,
- LVB_T_NONE, NULL, &lh);
+ rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
+ LDLM_EXTENT, policy, LCK_EX, &flags,
+ ldlm_blocking_ast, ldlm_completion_ast,
+ NULL, NULL, 0, LVB_T_NONE, NULL, &lh);
if (rc != ELDLM_OK)
GOTO(put, rc = -EIO);
static int lfsck_layout_conflict_create(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_tgt_desc *ltd,
- struct lu_orphan_rec_v2 *rec,
+ struct lu_orphan_rec_v3 *rec,
struct dt_object *parent,
struct lu_fid *cfid,
struct lu_buf *ea_buf,
static int lfsck_layout_recreate_lovea(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_tgt_desc *ltd,
- struct lu_orphan_rec_v2 *rec,
+ struct lu_orphan_rec_v3 *rec,
struct dt_object *parent,
struct lu_fid *cfid,
__u32 ost_idx, __u32 ea_off)
int rc = 0;
int rc1;
int i;
- __u16 count;
- bool locked = false;
+ int pos = 0;
+ __u16 count;
+ bool locked = false;
+ bool new_mirror = true;
ENTRY;
rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
"LOV EA for "DFID": parent "DFID", OST-index %u, "
"stripe-index %u, comp_id %u, comp_start %llu, "
- "comp_end %llu: rc = %d\n",
+ "comp_end %llu, layout version %u, range %u: rc = %d\n",
lfsck_lfsck2name(lfsck), PFID(cfid),
PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
ol->ol_comp_id, ol->ol_comp_start,
- ol->ol_comp_end, rc);
+ ol->ol_comp_end, rec->lor_layout_version,
+ rec->lor_range, rc);
RETURN(rc);
}
LASSERT(buf->lb_len >= lovea_size);
- rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
+ rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
cfid, buf, fl, ost_idx, ea_off);
GOTO(unlock_parent, rc);
}
lmm = buf->lb_buf;
- rc1 = lfsck_layout_verify_header(parent, lmm);
+ rc1 = lfsck_layout_verify_header(parent, lmm, lovea_size);
/* If the LOV EA crashed, the rebuild it. */
if (rc1 == -EINVAL) {
LASSERT(buf->lb_len >= lovea_size);
- rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
+ rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
cfid, buf, fl, ost_idx, ea_off);
GOTO(unlock_parent, rc);
}
/* For other unknown magic/pattern, keep the current LOV EA. */
- if (rc1 != 0)
+ if (rc1 == -EOPNOTSUPP)
+ GOTO(unlock_parent, rc1 = 0);
+
+ if (rc1)
GOTO(unlock_parent, rc = rc1);
magic = le32_to_cpu(lmm->lmm_magic);
if (magic == LOV_MAGIC_COMP_V1) {
__u64 start;
__u64 end;
+ __u16 mirror_id0 = mirror_id_of(ol->ol_comp_id);
+ __u16 mirror_id1;
+
+ if (bk->lb_param & LPF_DRYRUN)
+ GOTO(unlock_parent, rc = 1);
lcm = buf->lb_buf;
count = le16_to_cpu(lcm->lcm_entry_count);
- for (i = 0; i < count; i++) {
+ for (i = 0; i < count; pos = ++i) {
lcme = &lcm->lcm_entries[i];
start = le64_to_cpu(lcme->lcme_extent.e_start);
end = le64_to_cpu(lcme->lcme_extent.e_end);
+ mirror_id1 = mirror_id_of(le32_to_cpu(lcme->lcme_id));
+
+ if (mirror_id0 > mirror_id1)
+ continue;
+
+ if (mirror_id0 < mirror_id1)
+ break;
+ new_mirror = false;
if (end <= ol->ol_comp_start)
continue;
goto further;
}
- rc = lfsck_layout_add_comp_comp(env, lfsck, handle, ol, parent,
- cfid, buf, ost_idx, ea_off, i);
+ rc = lfsck_layout_add_comp(env, lfsck, handle, rec, parent,
+ cfid, buf, ost_idx, ea_off, pos, new_mirror);
GOTO(unlock_parent, rc);
}
goto again;
}
- if (lcme && !(flags & LCME_FL_INIT))
+ if (lcm) {
+ LASSERT(lcme);
+
lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
+ lfsck_layout_update_lcm(lcm, lcme,
+ rec->lor_layout_version,
+ rec->lor_range);
+ }
rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
parent, cfid, buf, ost_idx, ea_off);
GOTO(unlock_parent, rc = -EINVAL);
}
- le32_add_cpu(&lcm->lcm_layout_gen, 1);
lovea_size = le32_to_cpu(lcm->lcm_size);
- if (!(flags & LCME_FL_INIT))
- lcme->lcme_flags = cpu_to_le32(flags |
- LCME_FL_INIT);
+ lcme->lcme_flags = cpu_to_le32(flags |
+ LCME_FL_INIT);
+ lfsck_layout_update_lcm(lcm, lcme,
+ rec->lor_layout_version,
+ rec->lor_range);
}
LASSERTF(buf->lb_len >= lovea_size,
lfsck_ibits_unlock(&lh, LCK_EX);
rc = lfsck_layout_update_pfid(env, com, parent,
cfid, ltd->ltd_tgt,
- ol, i);
+ rec, i);
CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
"updated OST-object's pfid for "DFID
static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_tgt_desc *ltd,
- struct lu_orphan_rec_v2 *rec,
+ struct lu_orphan_rec_v3 *rec,
struct lu_fid *cfid)
{
struct lfsck_layout *lo = com->lc_file_ram;
do {
struct dt_key *key;
- struct lu_orphan_rec_v2 *rec = &info->lti_rec;
+ struct lu_orphan_rec_v3 *rec = &info->lti_rec;
if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
unlikely(!thread_is_running(&lfsck->li_thread)))
return rc > 0 ? 0 : rc;
}
-static int lfsck_lmm2layout(struct lov_mds_md_v1 *lmm, struct ost_layout *ol,
+static int lfsck_lov2layout(struct lov_mds_md_v1 *lmm, struct filter_fid *ff,
__u32 comp_id)
{
+ struct ost_layout *ol = &ff->ff_layout;
__u32 magic = le32_to_cpu(lmm->lmm_magic);
int rc = 0;
ENTRY;
ol->ol_comp_start = 0;
ol->ol_comp_end = 0;
ol->ol_comp_id = 0;
+ ff->ff_layout_version = 0;
+ ff->ff_range = 0;
} else if (magic == LOV_MAGIC_COMP_V1) {
struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
struct lov_comp_md_entry_v1 *lcme = NULL;
ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
+ ff->ff_layout_version = le32_to_cpu(lcme->lcme_layout_gen);
+ ff->ff_range = 0;
} else {
GOTO(out, rc = -EINVAL);
}
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct filter_fid *ff = &info->lti_ff;
- struct ost_layout *ol = &ff->ff_layout;
struct dt_object_format *dof = &info->lti_dof;
struct lu_attr *la = &info->lti_la;
struct lfsck_instance *lfsck = com->lc_lfsck;
ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
rc = lfsck_layout_get_lovea(env, parent, tbuf);
- if (rc < 0)
+ if (unlikely(rc == -ENODATA))
+ rc = 0;
+ if (rc <= 0)
GOTO(unlock1, rc);
- rc = lfsck_lmm2layout(tbuf->lb_buf, ol, comp_id);
+ rc = lfsck_lov2layout(tbuf->lb_buf, ff, comp_id);
if (rc)
GOTO(unlock1, rc);
int idx2;
rc = lfsck_layout_get_lovea(env, parent, lovea);
+ if (unlikely(rc == -ENODATA))
+ rc = 0;
if (rc <= 0)
GOTO(unlock2, rc);
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct filter_fid *ff = &info->lti_ff;
- struct ost_layout *ol = &ff->ff_layout;
struct dt_object *child = llr->llr_child;
struct dt_device *dev = lfsck_obj2dev(child);
const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
rc = lfsck_layout_get_lovea(env, parent, tbuf);
- if (rc < 0)
+ if (unlikely(rc == -ENODATA))
+ rc = 0;
+ if (rc <= 0)
GOTO(unlock1, rc);
- rc = lfsck_lmm2layout(tbuf->lb_buf, ol, llr->llr_comp_id);
+ rc = lfsck_lov2layout(tbuf->lb_buf, ff, llr->llr_comp_id);
if (rc)
GOTO(unlock1, rc);
GOTO(unlock, rc = 0);
rc = lfsck_layout_get_lovea(env, parent, buf);
- if (unlikely(!rc || rc == -ENODATA))
- GOTO(unlock, rc = 0);
+ if (unlikely(rc == -ENODATA))
+ rc = 0;
+ if (rc <= 0)
+ GOTO(unlock, rc);
lmm = buf->lb_buf;
magic = le32_to_cpu(lmm->lmm_magic);
struct dt_device *dev = lfsck_obj2dev(child);
struct thandle *handle;
int rc;
+ dt_obj_version_t version;
ENTRY;
tla->la_uid = pla->la_uid;
if (unlikely(lfsck_is_dead_obj(parent)))
GOTO(unlock, rc = 1);
+ version = dt_version_get(env, child);
+ if (version == -EOPNOTSUPP)
+ version = 0;
+
/* Get the latest parent's owner. */
rc = dt_attr_get(env, parent, pla);
if (rc != 0)
GOTO(unlock, rc);
/* Some others chown/chgrp during the LFSCK, needs to do nothing. */
- if (unlikely(tla->la_uid != pla->la_uid ||
- tla->la_gid != pla->la_gid))
+ if (unlikely((!version && tla->la_ctime == 0) ||
+ tla->la_uid != pla->la_uid || tla->la_gid != pla->la_gid))
rc = 1;
else
rc = dt_attr_set(env, child, tla, handle);
* is in such layout. If yes, it is multiple referenced, otherwise it
* is unmatched referenced case. */
rc = lfsck_layout_get_lovea(env, tobj, buf);
- if (rc == 0 || rc == -ENOENT)
+ if (rc == 0 || rc == -ENODATA || rc == -ENOENT)
GOTO(out, rc = LLIT_UNMATCHED_PAIR);
+ if (unlikely(rc == -EOPNOTSUPP))
+ GOTO(out, rc = LLIT_NONE);
+
if (rc < 0)
GOTO(out, rc);
if (lso->lso_dead)
RETURN(0);
- CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
+ CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ENGINE_DELAY, cfs_fail_val);
rc = dt_attr_get(env, child, cla);
if (rc == -ENOENT) {
if (rc < 0) {
struct lfsck_assistant_data *lad = com->lc_data;
- if (unlikely(lad->lad_exit)) {
+ if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags))) {
rc = 0;
} else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
if (rc != 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
- if (unlikely(lad->lad_exit ||
+ if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags) ||
!thread_is_running(&lfsck->li_thread)))
RETURN(0);
spin_lock(<ds->ltd_lock);
void *args, int rc)
{
struct lfsck_layout_slave_async_args *llsaa = args;
- struct obd_export *exp = llsaa->llsaa_exp;
- struct lfsck_component *com = llsaa->llsaa_com;
- struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
- struct lfsck_layout_slave_data *llsd = com->lc_data;
- struct lfsck_reply *lr = NULL;
- bool done = false;
+ struct obd_export *exp = llsaa->llsaa_exp;
+ struct lfsck_component *com = llsaa->llsaa_com;
+ struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
+ struct lfsck_layout_slave_data *llsd = com->lc_data;
+ struct lfsck_reply *lr = NULL;
+ bool done = false;
if (rc != 0) {
/* It is probably caused by network trouble, or target crash,
llsaa->llsaa_llst = llst;
req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
req->rq_allow_intr = 1;
+ req->rq_no_delay = 1;
ptlrpc_set_add_req(set, req);
RETURN(0);
*tmp = *lr;
ptlrpc_request_set_replen(req);
req->rq_allow_intr = 1;
+ req->rq_no_delay = 1;
ptlrpc_set_add_req(set, req);
RETURN(0);
}
spin_unlock(&llsd->llsd_lock);
- rc = ptlrpc_set_wait(set);
+ rc = ptlrpc_set_wait(env, set);
ptlrpc_set_destroy(set);
GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
}
spin_unlock(&llsd->llsd_lock);
- ptlrpc_set_wait(set);
+ ptlrpc_set_wait(env, set);
ptlrpc_set_destroy(set);
RETURN_EXIT;
if (rc < 0)
GOTO(unlock, rc);
- if (rc == 0)
- GOTO(unlock, rc = -ENODATA);
-
lmm = buf->lb_buf;
magic = le32_to_cpu(lmm->lmm_magic);
if (magic == LOV_MAGIC_COMP_V1) {
rc = __lfsck_layout_update_pfid(env, obj, &lrl->lrl_ff_client.ff_parent,
&lrl->lrl_ff_client.ff_layout,
+ lrl->lrl_ff_client.ff_layout_version,
+ lrl->lrl_ff_client.ff_range,
lrl->lrl_ff_client.ff_parent.f_ver);
GOTO(unlock, rc);
if (com->lc_lfsck->li_master) {
struct lfsck_assistant_data *lad = com->lc_data;
- lad->lad_incomplete = 0;
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
CFS_RESET_BITMAP(lad->lad_bitmap);
}
if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
LASSERT(!llsd->llsd_rbtree_valid);
- write_lock(&llsd->llsd_rb_lock);
+ down_write(&llsd->llsd_rb_rwsem);
rc = lfsck_rbtree_setup(env, com);
- write_unlock(&llsd->llsd_rb_lock);
+ up_write(&llsd->llsd_rb_rwsem);
}
CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
goto next;
}
- if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
- rc = dt_declare_attr_get(env, cobj);
- if (rc != 0)
- goto next;
+ rc = dt_declare_attr_get(env, cobj);
+ if (rc)
+ goto next;
- rc = dt_declare_xattr_get(env, cobj, &buf,
- XATTR_NAME_FID);
- if (rc != 0)
- goto next;
- }
+ rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID);
+ if (rc)
+ goto next;
if (lso == NULL) {
struct lu_attr *attr = &info->lti_la;
GOTO(out, rc = 0);
rc = lfsck_layout_get_lovea(env, obj, buf);
- if (rc <= 0)
+ if (rc == -EINVAL || rc == -ENODATA || rc == -EOPNOTSUPP)
/* Skip bad lov EA during the 1st cycle scanning, and
* try to recover it via orphan in the 2nd scanning. */
- GOTO(out, rc = (rc == -EINVAL ? 0 : rc));
+ rc = 0;
+ if (rc <= 0)
+ GOTO(out, rc);
size = rc;
lmm = buf->lb_buf;
if (IS_ERR(handle))
GOTO(out, rc = PTR_ERR(handle));
- lfsck_buf_init(&ea_buf, lmm, size);
+ lfsck_buf_init(&ea_buf, buf->lb_buf, size);
rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
LU_XATTR_REPLACE, handle);
if (rc != 0)
INIT_LIST_HEAD(&llsd->llsd_master_list);
spin_lock_init(&llsd->llsd_lock);
llsd->llsd_rb_root = RB_ROOT;
- rwlock_init(&llsd->llsd_rb_lock);
+ init_rwsem(&llsd->llsd_rb_rwsem);
com->lc_data = llsd;
}
com->lc_file_size = sizeof(*lo);
com->lc_obj = obj;
rc = lfsck_layout_load(env, com);
- if (rc > 0)
+ if (rc > 0) {
rc = lfsck_layout_reset(env, com, true);
- else if (rc == -ENOENT)
+ } else if (rc == -ENOENT) {
rc = lfsck_layout_init(env, com);
- else if (lfsck->li_master)
+ } else if (lfsck->li_master) {
rc = lfsck_load_sub_trace_files(env, com,
&dt_lfsck_layout_dangling_features,
LFSCK_LAYOUT, false);
+ if (rc)
+ rc = lfsck_layout_reset(env, com, true);
+ }
if (rc != 0)
GOTO(out, rc);
struct lfsck_rbtree_node *loi_lrn;
struct lfsck_layout_slave_target *loi_llst;
struct lu_fid loi_key;
- struct lu_orphan_rec_v2 loi_rec;
+ struct lu_orphan_rec_v3 loi_rec;
__u64 loi_hash;
unsigned int loi_over:1;
};
struct dt_object *dt,
const struct dt_rec *rec,
const struct dt_key *key,
- struct thandle *handle,
- int ignore_quota)
+ struct thandle *handle)
{
return -EOPNOTSUPP;
}
if (dev->dd_record_fid_accessed) {
/* The first iteration against the rbtree, scan the whole rbtree
* to remove the nodes which do NOT need to be handled. */
- write_lock(&llsd->llsd_rb_lock);
+ down_write(&llsd->llsd_rb_rwsem);
if (dev->dd_record_fid_accessed) {
struct rb_node *node;
struct rb_node *next;
node = next;
}
}
- write_unlock(&llsd->llsd_rb_lock);
+ up_write(&llsd->llsd_rb_rwsem);
}
/* read lock the rbtree when init, and unlock when fini */
- read_lock(&llsd->llsd_rb_lock);
+ down_read(&llsd->llsd_rb_rwsem);
it->loi_com = com;
com = NULL;
lfsck_lfsck2name(com->lc_lfsck));
llsd = com->lc_data;
- read_unlock(&llsd->llsd_rb_lock);
+ up_read(&llsd->llsd_rb_rwsem);
llst = it->loi_llst;
LASSERT(llst != NULL);
struct lu_attr *la = &info->lti_la;
struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
struct lu_fid *key = &it->loi_key;
- struct lu_orphan_rec_v2 *rec = &it->loi_rec;
+ struct lu_orphan_rec_v3 *rec = &it->loi_rec;
struct ost_layout *ol = &rec->lor_layout;
struct lfsck_component *com = it->loi_com;
struct lfsck_instance *lfsck = com->lc_lfsck;
rec->lor_rec.lor_uid = la->la_uid;
rec->lor_rec.lor_gid = la->la_gid;
memset(ol, 0, sizeof(*ol));
+ rec->lor_layout_version = 0;
+ rec->lor_range = 0;
GOTO(out, rc = 0);
}
rec->lor_rec.lor_uid = la->la_uid;
rec->lor_rec.lor_gid = la->la_gid;
ost_layout_le_to_cpu(ol, &ff->ff_layout);
+ rec->lor_layout_version =
+ le32_to_cpu(ff->ff_layout_version & ~LU_LAYOUT_RESYNC);
+ rec->lor_range = le32_to_cpu(ff->ff_range);
CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
"stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
- "COMP end %llu\n", lfsck_lfsck2name(com->lc_lfsck), PFID(key),
+ "COMP end %llu, layout version %u, range %u\n",
+ lfsck_lfsck2name(com->lc_lfsck), PFID(key),
PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
- ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end);
+ ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
+ rec->lor_layout_version, rec->lor_range);
GOTO(out, rc = 0);
{
struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
- *(struct lu_orphan_rec_v2 *)rec = it->loi_rec;
+ *(struct lu_orphan_rec_v3 *)rec = it->loi_rec;
return 0;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff