* GPL HEADER END
*/
/*
- * Copyright (c) 2013, 2015, Intel Corporation.
+ * Copyright (c) 2013, 2017, Intel Corporation.
*/
/*
* lustre/lfsck/lfsck_namespace.c
#define DEBUG_SUBSYSTEM S_LFSCK
-#include <lustre/lustre_idl.h>
#include <lu_object.h>
#include <dt_object.h>
#include <md_object.h>
#include <lustre_fid.h>
#include <lustre_lib.h>
#include <lustre_net.h>
-#include <lustre/lustre_user.h>
#include "lfsck_internal.h"
#define LFSCK_NAMESPACE_MAGIC_V1 0xA0629D03
#define LFSCK_NAMESPACE_MAGIC_V2 0xA0621A0B
+#define LFSCK_NAMESPACE_MAGIC_V3 0xA06249FF
/* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
* as the trace file magic. When downgrade to such old release, the old LFSCK
* will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
* then it will reset the whole LFSCK, and will not cause start failure. The
* similar case will happen when upgrade from such old release. */
-#define LFSCK_NAMESPACE_MAGIC LFSCK_NAMESPACE_MAGIC_V2
+#define LFSCK_NAMESPACE_MAGIC LFSCK_NAMESPACE_MAGIC_V3
enum lfsck_nameentry_check {
LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
dst->ln_status = le32_to_cpu(src->ln_status);
dst->ln_flags = le32_to_cpu(src->ln_flags);
dst->ln_success_count = le32_to_cpu(src->ln_success_count);
- dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
- dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
+ dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
+ dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
dst->ln_time_last_checkpoint =
dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
+ dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
+ dst->ln_linkea_overflow_cleared =
+ le64_to_cpu(src->ln_linkea_overflow_cleared);
+ dst->ln_agent_entries_repaired =
+ le64_to_cpu(src->ln_agent_entries_repaired);
}
static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
dst->ln_status = cpu_to_le32(src->ln_status);
dst->ln_flags = cpu_to_le32(src->ln_flags);
dst->ln_success_count = cpu_to_le32(src->ln_success_count);
- dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
- dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
+ dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
+ dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
dst->ln_time_last_checkpoint =
dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
+ dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
+ dst->ln_linkea_overflow_cleared =
+ cpu_to_le64(src->ln_linkea_overflow_cleared);
+ dst->ln_agent_entries_repaired =
+ cpu_to_le64(src->ln_agent_entries_repaired);
}
static void lfsck_namespace_record_failure(const struct lu_env *env,
}
if (ns->ln_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);
}
return rc;
}
-static struct dt_object *
-lfsck_namespace_load_one_trace_file(const struct lu_env *env,
- struct lfsck_component *com,
- struct dt_object *parent,
- const char *name, bool reset)
-{
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct dt_object *obj;
- int rc;
-
- if (reset) {
- rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
- if (rc != 0 && rc != -ENOENT)
- return ERR_PTR(rc);
- }
-
- obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
- S_IFREG | S_IRUGO | S_IWUSR,
- &dt_lfsck_features);
-
- return obj;
-}
-
-static int lfsck_namespace_load_sub_trace_files(const struct lu_env *env,
- struct lfsck_component *com,
- bool reset)
-{
- char *name = lfsck_env_info(env)->lti_key;
- struct lfsck_sub_trace_obj *lsto;
- struct dt_object *obj;
- int rc;
- int i;
-
- for (i = 0, lsto = &com->lc_sub_trace_objs[0];
- i < LFSCK_STF_COUNT; i++, lsto++) {
- snprintf(name, NAME_MAX, "%s_%02d", LFSCK_NAMESPACE, i);
- if (lsto->lsto_obj != NULL) {
- if (!reset)
- continue;
-
- lfsck_object_put(env, lsto->lsto_obj);
- lsto->lsto_obj = NULL;
- }
-
- obj = lfsck_namespace_load_one_trace_file(env, com,
- com->lc_lfsck->li_lfsck_dir, name, reset);
- if (IS_ERR(obj))
- return PTR_ERR(obj);
-
- lsto->lsto_obj = obj;
- rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
- if (rc != 0)
- return rc;
- }
-
- return 0;
-}
-
static int lfsck_namespace_init(const struct lu_env *env,
struct lfsck_component *com)
{
memset(ns, 0, sizeof(*ns));
ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
ns->ln_status = LS_INIT;
+ ns->ln_time_latest_reset = ktime_get_real_seconds();
down_write(&com->lc_sem);
rc = lfsck_namespace_store(env, com);
- up_write(&com->lc_sem);
if (rc == 0)
- rc = lfsck_namespace_load_sub_trace_files(env, com, true);
+ rc = lfsck_load_sub_trace_files(env, com,
+ &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
+ up_write(&com->lc_sem);
return rc;
}
RETURN(0);
idx = lfsck_sub_trace_file_fid2idx(fid);
+ mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
obj = com->lc_sub_trace_objs[idx].lsto_obj;
+ if (unlikely(obj == NULL)) {
+ mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
+ RETURN(0);
+ }
+
+ lfsck_object_get(obj);
dev = lfsck_obj2dev(obj);
- mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
fid_cpu_to_be(key, fid);
rc = dt_lookup(env, obj, (struct dt_rec *)&old,
(const struct dt_key *)key);
if (new != 0) {
rc = dt_insert(env, obj, (const struct dt_rec *)&new,
- (const struct dt_key *)key, th, 1);
+ (const struct dt_key *)key, th);
if (rc != 0)
GOTO(log, rc);
}
unlock:
mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
+ lfsck_object_put(env, obj);
return rc;
}
{
int rc;
+ /* For remote updating LINKEA, there may be further LFSCK action
+ * on remote MDT after the updating, so update the LINKEA ASAP. */
+ if (dt_object_remote(obj))
+ handle->th_sync = 1;
+
/* For destroying all invalid linkEA entries. */
rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
if (rc == 0)
/* For insert new linkEA entry. */
rc = dt_declare_xattr_set(env, obj,
- lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
+ lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
XATTR_NAME_LINK, 0, handle);
return rc;
}
int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
- struct linkea_data *ldata)
+ struct linkea_data *ldata, bool with_rec)
{
int rc;
if (unlikely(rc == 0))
return -ENODATA;
- if (rc > 0)
- rc = linkea_init(ldata);
+ if (rc > 0) {
+ if (with_rec)
+ rc = linkea_init_with_rec(ldata);
+ else
+ rc = linkea_init(ldata);
+ }
return rc;
}
static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
struct linkea_data *ldata, struct thandle *handle)
{
- const struct lu_buf *buf = lfsck_buf_get_const(env,
- ldata->ld_buf->lb_buf,
- ldata->ld_leh->leh_len);
+ struct lu_buf buf;
+ int rc;
+
+ lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
- return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle);
+again:
+ rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
+ if (unlikely(rc == -ENOSPC)) {
+ rc = linkea_overflow_shrink(ldata);
+ if (likely(rc > 0)) {
+ buf.lb_len = rc;
+ goto again;
+ }
+ }
+
+ return rc;
+}
+
+static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
+{
+ if (ldata->ld_reclen <= 0)
+ return false;
+
+ if ((char *)ldata->ld_lee + ldata->ld_reclen >
+ (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
+ return false;
+
+ return true;
+}
+
+static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
+ const struct lu_name *cname,
+ const struct lu_fid *pfid)
+{
+ if (!linkea_reclen_is_valid(ldata))
+ return false;
+
+ if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
+ return false;
+
+ if (!fid_is_sane(pfid))
+ return false;
+
+ return true;
}
static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
char *buf, const int buflen)
{
linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
- if (unlikely(ldata->ld_reclen <= 0 ||
- ldata->ld_reclen + sizeof(struct link_ea_header) >
- ldata->ld_leh->leh_len ||
- cname->ln_namelen <= 0 ||
- cname->ln_namelen > NAME_MAX ||
- cname->ln_namelen >= buflen ||
- !fid_is_sane(pfid)))
+ if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
return -EINVAL;
/* To guarantee the 'name' is terminated with '0'. */
/* If current record is corrupted, all the subsequent
* records will be dropped. */
- if (unlikely(ldata->ld_reclen <= 0 ||
- ldata->ld_reclen + sizeof(struct link_ea_header) >
- ldata->ld_leh->leh_len)) {
+ if (unlikely(!linkea_reclen_is_valid(ldata))) {
void *ptr = ldata->ld_lee;
ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
while (ldata->ld_lee != NULL) {
ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
ldata->ld_lee->lee_reclen[1];
- if (unlikely(ldata->ld_reclen == oldlen &&
+ if (unlikely(!linkea_reclen_is_valid(ldata))) {
+ lfsck_linkea_del_buf(ldata, NULL);
+ LASSERT(ldata->ld_lee == NULL);
+ } else if (unlikely(ldata->ld_reclen == oldlen &&
memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
repeated++;
if (!remove)
cname->ln_name = info->lti_key;
cname->ln_namelen = namelen;
- rc = linkea_data_new(&ldata2, &info->lti_linkea_buf2);
- if (rc != 0)
- GOTO(log, rc);
-
- rc = linkea_add_buf(&ldata2, cname, pfid);
+ rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
+ cname, pfid);
if (rc != 0)
GOTO(log, rc);
}
memset(la, 0, sizeof(*la));
- la->la_ctime = cfs_time_current_sec();
+ la->la_ctime = ktime_get_real_seconds();
la->la_valid = LA_CTIME;
rc = dt_declare_attr_set(env, orphan, la, th);
if (rc != 0)
GOTO(stop, rc);
dt_write_lock(env, orphan, 0);
- rc = lfsck_links_read2(env, orphan, &ldata2);
- if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
- (rc == 0 && ldata2.ld_leh != NULL &&
- ldata2.ld_leh->leh_reccount == 0))) {
+ rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
+ if (likely(rc == -ENODATA || rc == -EINVAL)) {
if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
GOTO(unlock, rc = 1);
rec->rec_type = S_IFDIR;
rec->rec_fid = pfid;
rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock, rc);
}
rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
rec->rec_fid = cfid;
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)cname->ln_name, th, 1);
+ (const struct dt_key *)cname->ln_name, th);
if (rc == 0 && S_ISDIR(rec->rec_type)) {
dt_write_lock(env, parent, 0);
rc = dt_ref_add(env, parent, th);
return rc;
}
+static int lfsck_lmv_set(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj,
+ struct lmv_mds_md_v1 *lmv)
+{
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ struct lu_buf buf = { lmv, sizeof(*lmv) };
+ int rc;
+
+ ENTRY;
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
+
+ rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV".set", 0, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV".set", 0, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ EXIT;
+stop:
+ dt_trans_stop(env, dev, th);
+
+ return rc;
+}
+
+static int lfsck_lmv_delete(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj)
+{
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ int rc;
+
+ ENTRY;
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
+
+ rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ EXIT;
+stop:
+ dt_trans_stop(env, dev, th);
+
+ return rc;
+}
+
+static inline int lfsck_object_is_shard(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj,
+ const struct lu_name *lname)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
+ int rc;
+
+ rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
+ lfsck_object_type(obj),
+ lfsck_dto2fid(obj));
+ if (rc < 0)
+ return 0;
+
+ rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
+ if (rc == -ENODATA)
+ return 0;
+
+ if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
+ return 1;
+
+ return rc;
+}
+
/**
* Add the specified name entry back to namespace.
*
* it is quite possible that the name entry is lost. Then the LFSCK
* should add the name entry back to the namespace.
*
+ * If \a child is shard, which means \a parent is a striped directory,
+ * if \a parent has LMV, we need to delete it before insertion because
+ * now parent's striping is broken and can't be parsed correctly.
+ *
* \param[in] env pointer to the thread context
* \param[in] com pointer to the lfsck component
* \param[in] parent pointer to the directory under which the name entry
* will be inserted into
* \param[in] child pointer to the object referenced by the name entry
* that to be inserted into the parent
- * \param[in] name the name for the child in the parent directory
+ * \param[in] lname the name for the child in the parent directory
*
* \retval positive number for repaired cases
* \retval 0 if nothing to be repaired
struct lfsck_component *com,
struct dt_object *parent,
struct dt_object *child,
- const char *name)
+ const struct lu_name *lname)
{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lu_attr *la = &info->lti_la;
- struct dt_insert_rec *rec = &info->lti_dt_rec;
- struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
/* The child and its name may be on different MDTs. */
- const struct lu_fid *pfid = lfsck_dto2fid(parent);
- const struct lu_fid *cfid = lfsck_dto2fid(child);
- struct dt_device *dev = lfsck->li_next;
- struct thandle *th = NULL;
- struct lfsck_lock_handle *llh = &info->lti_llh;
- int rc = 0;
+ const struct lu_fid *pfid = lfsck_dto2fid(parent);
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ struct lfsck_lock_handle *llh = &info->lti_llh;
+ struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
+ struct lu_buf buf = { lmv, sizeof(*lmv) };
+ /* whether parent's LMV is deleted before insertion */
+ bool parent_lmv_deleted = false;
+ /* whether parent's LMV is missing */
+ bool parent_lmv_lost = false;
+ int rc = 0;
+
ENTRY;
/* @parent/@child may be based on lfsck->li_bottom,
if (IS_ERR(parent))
GOTO(log, rc = PTR_ERR(parent));
- if (unlikely(!dt_try_as_dir(env, parent)))
- GOTO(log, rc = -ENOTDIR);
-
child = lfsck_object_locate(dev, child);
if (IS_ERR(child))
GOTO(log, rc = PTR_ERR(child));
if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
GOTO(log, rc = 1);
- rc = lfsck_lock(env, lfsck, parent, name, llh,
- MDS_INODELOCK_UPDATE, LCK_PW);
- if (rc != 0)
+ rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
+ MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc)
GOTO(log, rc);
+ rc = lfsck_object_is_shard(env, lfsck, child, lname);
+ if (rc < 0)
+ GOTO(unlock, rc);
+
+ if (rc == 1) {
+ rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
+ if (!rc) {
+ /*
+ * To add a shard, we need to convert parent to a
+ * plain directory by deleting its LMV, and after
+ * insertion set it back.
+ */
+ rc = lfsck_lmv_delete(env, lfsck, parent);
+ if (rc)
+ GOTO(unlock, rc);
+ parent_lmv_deleted = true;
+ lmv->lmv_layout_version++;
+ lfsck_lmv_header_cpu_to_le(lmv, lmv);
+ } else if (rc == -ENODATA) {
+ struct lu_seq_range *range = &info->lti_range;
+ struct seq_server_site *ss = lfsck_dev_site(lfsck);
+
+ rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
+ if (rc)
+ GOTO(unlock, rc);
+
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(lfsck_dto2fid(parent)), range);
+ if (rc)
+ GOTO(unlock, rc);
+
+ parent_lmv_lost = true;
+ lmv->lmv_magic = LMV_MAGIC;
+ lmv->lmv_master_mdt_index = range->lsr_index;
+ lmv->lmv_layout_version++;
+ lfsck_lmv_header_cpu_to_le(lmv, lmv);
+ } else {
+ GOTO(unlock, rc);
+ }
+ }
+
+ if (unlikely(!dt_try_as_dir(env, parent)))
+ GOTO(unlock, rc = -ENOTDIR);
+
th = dt_trans_create(env, dev);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rec->rec_type = lfsck_object_type(child) & S_IFMT;
rec->rec_fid = cfid;
rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th);
+ (const struct dt_key *)lname->ln_name, th);
if (rc != 0)
GOTO(stop, rc);
GOTO(stop, rc);
}
- memset(la, 0, sizeof(*la));
- la->la_ctime = cfs_time_current_sec();
+ if (parent_lmv_lost) {
+ rc = dt_declare_xattr_set(env, parent, &buf,
+ XATTR_NAME_LMV".set", 0, th);
+ if (rc)
+ GOTO(stop, rc);
+ }
+
+ la->la_ctime = ktime_get_real_seconds();
la->la_valid = LA_CTIME;
rc = dt_declare_attr_set(env, parent, la, th);
if (rc != 0)
GOTO(stop, rc);
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)lname->ln_name, th);
if (rc != 0)
GOTO(stop, rc);
GOTO(stop, rc);
}
- la->la_ctime = cfs_time_current_sec();
+ if (parent_lmv_lost) {
+ rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV".set", 0,
+ th);
+ if (rc)
+ GOTO(stop, rc);
+ }
+
rc = dt_attr_set(env, parent, la, th);
if (rc != 0)
GOTO(stop, rc);
dt_trans_stop(env, dev, th);
unlock:
+ if (parent_lmv_deleted)
+ lfsck_lmv_set(env, lfsck, parent, lmv);
+
lfsck_unlock(llh);
log:
CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
"the name %s and type %o to the parent "DFID": rc = %d\n",
- lfsck_lfsck2name(lfsck), PFID(cfid), name,
+ lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
if (rc != 0) {
memset(dof, 0, sizeof(*dof));
dof->dof_type = dt_mode_to_dft(S_IFDIR);
- rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
- if (rc != 0)
- GOTO(unlock1, rc);
-
- rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
+ rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
+ cname, lfsck_dto2fid(parent));
if (rc != 0)
GOTO(unlock1, rc);
rec->rec_fid = cfid;
rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
- (const struct dt_key *)dot, th, 1);
+ (const struct dt_key *)dot, th);
if (rc != 0)
GOTO(unlock2, rc);
rec->rec_fid = lfsck_dto2fid(parent);
rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock2, rc);
rec->rec_fid = cfid;
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)name, th);
if (rc == 0) {
dt_write_lock(env, parent, 0);
rc = dt_ref_add(env, parent, th);
else
lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
true);
- if (ldata->ld_leh->leh_reccount > 0) {
+ if (ldata->ld_leh->leh_reccount > 0 ||
+ unlikely(ldata->ld_leh->leh_overflow_time)) {
lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
ldata->ld_leh->leh_len);
buflen = linkea_buf.lb_len;
if (unlikely(lfsck_is_dead_obj(obj)))
GOTO(unlock2, rc = -ENOENT);
- rc = lfsck_links_read2(env, obj, &ldata_new);
- if (rc != 0)
- GOTO(unlock2,
- rc = (rc == -ENODATA ? 0 : rc));
+ rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
+ if (rc)
+ GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
/* The specified linkEA entry has been removed by race. */
rc = linkea_links_find(&ldata_new, cname, pfid);
lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
true);
- if (buflen < ldata_new.ld_leh->leh_len) {
+ /*
+ * linkea may change because it doesn't take lock in the first read, if
+ * it becomes larger, restart from beginning.
+ */
+ if ((ldata_new.ld_leh->leh_reccount > 0 ||
+ unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
+ buflen < ldata_new.ld_leh->leh_len) {
dt_write_unlock(env, obj);
dt_trans_stop(env, dev, th);
lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
ldata_new.ld_leh->leh_len);
+ buflen = linkea_buf.lb_len;
goto again;
}
- if (ldata_new.ld_leh->leh_reccount > 0) {
- lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
- ldata_new.ld_leh->leh_len);
- rc = dt_xattr_set(env, obj, &linkea_buf,
- XATTR_NAME_LINK, 0, th);
- } else {
+ if (buflen)
+ rc = lfsck_links_write(env, obj, &ldata_new, th);
+ else
rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
- }
GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
replace:
dt_read_lock(env, child, 0);
- rc = lfsck_links_read2(env, child, &ldata);
+ rc = lfsck_links_read2_with_rec(env, child, &ldata);
dt_read_unlock(env, child);
/* Someone changed the child, no need to replace. */
GOTO(stop, rc);
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)name, th);
GOTO(stop, rc = (rc == 0 ? 1 : rc));
struct lfsck_thread_info *info = lfsck_env_info(env);
struct dt_insert_rec *rec = &info->lti_dt_rec;
const struct lu_fid *pfid = lfsck_dto2fid(parent);
- const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct lu_fid cfid = {0};
struct lu_fid tfid;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct dt_device *dev = lfsck->li_next;
int rc = 0;
ENTRY;
+ if (child)
+ cfid = *lfsck_dto2fid(child);
parent = lfsck_object_locate(dev, parent);
if (IS_ERR(parent))
GOTO(log, rc = PTR_ERR(parent));
if (update) {
rec->rec_type = lfsck_object_type(child) & S_IFMT;
- rec->rec_fid = cfid;
+ LASSERT(!fid_is_zero(&cfid));
+ rec->rec_fid = &cfid;
rc = dt_declare_insert(env, parent,
(const struct dt_rec *)rec,
(const struct dt_key *)name2, th);
/* Someone has removed the bad name entry and reused it for other
* object by race. */
- if (!lu_fid_eq(&tfid, cfid))
+ if (!lu_fid_eq(&tfid, &cfid))
GOTO(unlock2, rc = 0);
if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
if (update) {
rc = dt_insert(env, parent,
(const struct dt_rec *)rec,
- (const struct dt_key *)name2, th, 1);
+ (const struct dt_key *)name2, th);
if (rc != 0)
GOTO(unlock2, rc);
}
/* We are not sure whether the child will become orphan or not.
* Record it in the LFSCK trace file for further checking in
* the second-stage scanning. */
- if (!update && !dec && rc == 0)
- lfsck_namespace_trace_update(env, com, cfid,
+ if (!update && !dec && child && rc == 0)
+ lfsck_namespace_trace_update(env, com, &cfid,
LNTF_CHECK_LINKEA, true);
unlock1:
"entry for: parent "DFID", child "DFID", name %s, type "
"in name entry %o, type claimed by child %o. repair it "
"by %s with new name2 %s: rc = %d\n",
- lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
+ lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
name, type, update ? lfsck_object_type(child) : 0,
update ? "updating" : "removing", name2, rc);
LASSERT(!dt_object_remote(obj));
LASSERT(S_ISDIR(lfsck_object_type(obj)));
- rc = linkea_data_new(&ldata, &info->lti_big_buf);
- if (rc != 0)
- GOTO(log, rc);
-
- rc = linkea_add_buf(&ldata, cname, pfid);
+ rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
if (rc != 0)
GOTO(log, rc);
dt_delete(env, obj, (const struct dt_key *)dotdot, th);
rc = dt_insert(env, obj, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock, rc);
- rc = dt_xattr_set(env, obj, &linkea_buf,
- XATTR_NAME_LINK, 0, th);
+ rc = lfsck_links_write(env, obj, &ldata, th);
GOTO(unlock, rc = (rc == 0 ? 1 : rc));
lost_parent:
lmv = &info->lti_lmv;
- rc = lfsck_read_stripe_lmv(env, child, lmv);
+ rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
if (rc != 0 && rc != -ENODATA)
GOTO(out, rc);
if (rc >= 0) {
/* Add the missing name entry to the parent. */
rc = lfsck_namespace_insert_normal(env, com, parent,
- child, cname->ln_name);
+ child, cname);
if (unlikely(rc == -EEXIST)) {
/* Unfortunately, someone reused the name
* under the parent by race. So we have
}
lfsck_ibits_unlock(lh, LCK_EX);
- rc = lfsck_namespace_check_name(env, parent, child, cname);
+ rc = lfsck_namespace_check_name(env, lfsck, parent, child,
+ cname);
if (rc == -ENOENT)
goto lost_parent;
/* Add the missing name entry back to the namespace. */
rc = lfsck_namespace_insert_normal(env, com, parent, child,
- cname->ln_name);
+ cname);
if (unlikely(rc == -ESTALE))
/* It may happen when the remote object has been
* removed, but the local MDT is not aware of that. */
* When the LFSCK runs again, if the dangling name is still
* there, the LFSCK should move the orphan directory object
* back to the normal namespace. */
- if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
+ if (!lpf && !fid_is_zero(pfid) &&
+ !lu_fid_eq(pfid, &tfid) && once) {
linkea_next_entry(ldata);
continue;
}
* directory contains the specified child, but such
* parent does not match the dotdot name entry, then
* trust the linkEA. */
- if (!lu_fid_eq(pfid, pfid2)) {
+ if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
*type = LNIT_UNMATCHED_PAIRS;
rc = lfsck_namespace_repair_unmatched_pairs(env,
com, child, pfid2, cname);
/* It is the most common case that we find the
* name entry corresponding to the linkEA entry
* that matches the ".." name entry. */
- rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
- if (rc != 0)
- RETURN(rc);
-
- rc = linkea_add_buf(&ldata_new, cname, pfid2);
+ rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
+ cname, pfid2);
if (rc != 0)
RETURN(rc);
lfsck_linkea_del_buf(ldata, cname);
} /* while (ldata->ld_lee != NULL) */
+ /* If there is still linkEA overflow, return. */
+ if (unlikely(ldata->ld_leh->leh_overflow_time))
+ RETURN(0);
+
linkea_first_entry(ldata);
if (ldata->ld_leh->leh_reccount == 1) {
rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
*
* If all the known name entries have been verified, then the object's hard
* link attribute should match the object's linkEA entries count unless the
- * object's has too much hard link to be recorded in the linkEA. Such cases
+ * object's has too many hard link to be recorded in the linkEA. Such cases
* should have been marked in the LFSCK trace file. Otherwise, trust the
* linkEA to update the object's nlink attribute.
*
struct dt_object *obj,
struct lu_attr *la)
{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lu_fid *tfid = &info->lti_fid3;
struct lfsck_namespace *ns = com->lc_file_ram;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct dt_device *dev = lfsck_obj2dev(obj);
struct linkea_data ldata = { NULL };
struct lustre_handle lh = { 0 };
__u32 old = la->la_nlink;
- int idx;
int rc = 0;
- __u8 flags;
ENTRY;
LASSERT(!dt_object_remote(obj));
if (ns->ln_flags & LF_INCOMPLETE)
GOTO(unlock, rc = 0);
- fid_cpu_to_be(tfid, cfid);
- idx = lfsck_sub_trace_file_fid2idx(cfid);
- rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj,
- (struct dt_rec *)&flags, (const struct dt_key *)tfid);
- if (rc != 0)
- GOTO(unlock, rc);
-
- if (flags & LNTF_SKIP_NLINK)
- GOTO(unlock, rc = 0);
-
rc = dt_attr_get(env, obj, la);
if (rc != 0)
GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
- rc = lfsck_links_read2(env, obj, &ldata);
- if (rc != 0)
+ rc = lfsck_links_read2_with_rec(env, obj, &ldata);
+ if (rc)
GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
- if (la->la_nlink == ldata.ld_leh->leh_reccount ||
- unlikely(la->la_nlink == 0))
+ /* XXX: Currently, we only update the nlink attribute if the known
+ * linkEA entries is larger than the nlink attribute. That is
+ * safe action. */
+ if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
+ unlikely(la->la_nlink == 0 ||
+ ldata.ld_leh->leh_overflow_time))
GOTO(unlock, rc = 0);
la->la_nlink = ldata.ld_leh->leh_reccount;
GOTO(out, rc);
} /* rc != 0 */
- linkea_first_entry(&ldata);
- /* This is the most common case: the object has unique linkEA entry. */
- if (ldata.ld_leh->leh_reccount == 1) {
- rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
- &lh, &type, &retry, &unknown);
- if (retry) {
- LASSERT(!lustre_handle_is_used(&lh));
+ linkea_first_entry(&ldata);
+ /* This is the most common case: the object has unique linkEA entry. */
+ if (ldata.ld_leh->leh_reccount == 1) {
+ rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
+ &lh, &type, &retry, &unknown);
+ if (retry) {
+ LASSERT(!lustre_handle_is_used(&lh));
+
+ retry = false;
+ goto lock;
+ }
+
+ GOTO(out, rc);
+ }
+
+ if (!lustre_handle_is_used(&lh))
+ goto lock;
+
+ if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
+ rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
+ &type);
+
+ GOTO(out, rc);
+ }
+
+ /* When we come here, the cases usually like that:
+ * 1) The directory object has a corrupted linkEA entry. During the
+ * first-stage scanning, the LFSCK cannot know such corruption,
+ * then it appends the right linkEA entry according to the found
+ * name entry after the bad one.
+ *
+ * 2) The directory object has a right linkEA entry. During the
+ * first-stage scanning, the LFSCK finds some bad name entry,
+ * but the LFSCK cannot aware that at that time, then it adds
+ * the bad linkEA entry for further processing. */
+ rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
+ &lh, &type, lpf, &unknown);
+
+ GOTO(out, rc);
+
+out:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ if (rc > 0) {
+ switch (type) {
+ case LNIT_BAD_LINKEA:
+ ns->ln_linkea_repaired++;
+ break;
+ case LNIT_UNMATCHED_PAIRS:
+ ns->ln_unmatched_pairs_repaired++;
+ break;
+ case LNIT_MUL_REF:
+ ns->ln_mul_ref_repaired++;
+ break;
+ default:
+ break;
+ }
+ }
+
+ if (unknown)
+ ns->ln_unknown_inconsistency++;
+
+ return rc;
+}
+
+static inline bool
+lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
+ struct lfsck_namespace *ns)
+{
+ /* Both the leh_overflow_time and ln_time_latest_reset are
+ * local time based, so need NOT to care about clock drift
+ * among the servers. */
+ return ldata->ld_leh->leh_overflow_time &&
+ ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
+}
+
+/**
+ * Clear the object's linkEA overflow timestamp.
+ *
+ * If the MDT-object has too many hard links as to the linkEA cannot hold
+ * all of them, then overflow timestamp will be set in the linkEA header.
+ * If some hard links are removed after that, then it is possible to hold
+ * other missed linkEA entries. If the namespace LFSCK have added all the
+ * related linkEA entries, then it will remove the overflow timestamp.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] ldata pointer to the linkEA data for the given @obj
+ * \param[in] obj pointer to the dt_object to be handled
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct linkea_data *ldata,
+ struct dt_object *obj)
+{
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct thandle *th = NULL;
+ struct lustre_handle lh = { 0 };
+ struct lu_buf linkea_buf;
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(!dt_object_remote(obj));
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE, LCK_PW);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
+
+ rc = dt_declare_xattr_set(env, obj,
+ lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ rc = lfsck_links_read(env, obj, ldata);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
+ GOTO(unlock, rc = 0);
+
+ ldata->ld_leh->leh_overflow_time = 0;
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(unlock, rc = 1);
+
+ /* If all known entries are in the linkEA, then the 'leh_reccount'
+ * should NOT be zero. */
+ LASSERT(ldata->ld_leh->leh_reccount > 0);
+
+ lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
+ ldata->ld_leh->leh_len);
+ rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
+ if (unlikely(rc == -ENOSPC))
+ rc = 0;
+ else if (!rc)
+ rc = 1;
+
+ GOTO(unlock, rc);
+
+unlock:
+ dt_write_unlock(env, obj);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+log:
+ lfsck_ibits_unlock(&lh, LCK_PW);
+ CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
+ DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
+
+ return rc;
+}
+
+/**
+ * Verify the object's agent entry.
+ *
+ * If the object claims to have agent entry but the linkEA does not contain
+ * remote parent, then remove the agent entry. Otherwise, if the object has
+ * no agent entry but its linkEA contains remote parent, then will generate
+ * agent entry for it.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj)
+{
+ struct linkea_data ldata = { NULL };
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lu_fid *pfid = &info->lti_fid2;
+ struct lu_name *cname = &info->lti_name;
+ struct lu_seq_range *range = &info->lti_range;
+ struct seq_server_site *ss = lfsck_dev_site(lfsck);
+ __u32 idx = lfsck_dev_idx(lfsck);
+ int rc;
+ bool remote = false;
+ ENTRY;
+
+ if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
+ RETURN(0);
+
+ rc = lfsck_links_read_with_rec(env, obj, &ldata);
+ if (rc == -ENOENT || rc == -ENODATA)
+ RETURN(0);
+
+ if (rc && rc != -EINVAL)
+ GOTO(out, rc);
+
+ /* We check the agent entry again after verifying the linkEA
+ * successfully. So invalid linkEA should be dryrun mode. */
+ if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
+ RETURN(0);
+
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL && !remote) {
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+ cname, pfid);
+ if (!linkea_entry_is_valid(&ldata, cname, pfid))
+ GOTO(out, rc = 0);
+
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(pfid), range);
+ if (rc)
+ GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
+
+ if (range->lsr_index != idx)
+ remote = true;
+ else
+ linkea_next_entry(&ldata);
+ }
+
+ if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+ (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct linkea_data ldata2 = { NULL };
+ struct lustre_handle lh = { 0 };
+ struct lu_buf linkea_buf;
+ struct thandle *handle;
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(out, rc = 1);
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc)
+ GOTO(out, rc);
+
+ handle = dt_trans_create(env, dev);
+ if (IS_ERR(handle))
+ GOTO(unlock, rc = PTR_ERR(handle));
+
+ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len);
+ rc = dt_declare_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
+ if (rc)
+ GOTO(stop, rc);
- retry = false;
- goto lock;
+ rc = dt_trans_start_local(env, dev, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
+ if (rc) {
+ if (rc == -ENOENT || rc == -ENODATA)
+ rc = 0;
+ GOTO(unlock2, rc);
}
- GOTO(out, rc);
- }
+ /* If someone changed linkEA by race, then the agent
+ * entry will be updated by lower layer automatically. */
+ if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
+ memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len) != 0)
+ GOTO(unlock2, rc = 0);
- if (!lustre_handle_is_used(&lh))
- goto lock;
+ rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
+ LU_XATTR_REPLACE, handle);
+ if (!rc)
+ rc = 1;
- if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
- rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
- &type);
+ GOTO(unlock2, rc);
- GOTO(out, rc);
+unlock2:
+ dt_write_unlock(env, obj);
+stop:
+ dt_trans_stop(env, dev, handle);
+unlock:
+ lfsck_ibits_unlock(&lh, LCK_EX);
}
- /* When we come here, the cases usually like that:
- * 1) The directory object has a corrupted linkEA entry. During the
- * first-stage scanning, the LFSCK cannot know such corruption,
- * then it appends the right linkEA entry according to the found
- * name entry after the bad one.
- *
- * 2) The directory object has a right linkEA entry. During the
- * first-stage scanning, the LFSCK finds some bad name entry,
- * but the LFSCK cannot aware that at that time, then it adds
- * the bad linkEA entry for further processing. */
- rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
- &lh, &type, lpf, &unknown);
-
GOTO(out, rc);
out:
- lfsck_ibits_unlock(&lh, LCK_EX);
- if (rc > 0) {
- switch (type) {
- case LNIT_BAD_LINKEA:
- ns->ln_linkea_repaired++;
- break;
- case LNIT_UNMATCHED_PAIRS:
- ns->ln_unmatched_pairs_repaired++;
- break;
- case LNIT_MUL_REF:
- ns->ln_mul_ref_repaired++;
- break;
- default:
- break;
- }
- }
-
- if (unknown)
- ns->ln_unknown_inconsistency++;
-
+ if (rc > 0)
+ ns->ln_agent_entries_repaired++;
+ if (rc)
+ CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
return rc;
}
if (S_ISDIR(lfsck_object_type(child))) {
dt_read_unlock(env, child);
rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
+ if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+ rc = lfsck_namespace_check_agent_entry(env, com, child);
RETURN(rc);
}
lfsck_ibits_unlock(&lh, LCK_EX);
}
- GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
+ GOTO(out, rc);
}
if (rc != 0)
GOTO(out, rc);
+ if (!(ns->ln_flags & LF_INCOMPLETE) &&
+ unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
+ rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
+ child);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ ns->ln_linkea_overflow_cleared++;
+ }
+
linkea_first_entry(&ldata);
while (ldata.ld_lee != NULL) {
rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
/* Add the missing name entry to the parent. */
rc = lfsck_namespace_insert_normal(env, com,
- parent, child, cname->ln_name);
+ parent, child, cname);
if (unlikely(rc == -EEXIST))
/* Unfortunately, someone reused the
* name under the parent by race. So we
GOTO(out, rc);
/* If there is no name entry in the parent dir and the object
- * link count is less than the linkea entries count, then the
+ * link count is fewer than the linkea entries count, then the
* linkea entry should be removed. */
if (ldata.ld_leh->leh_reccount > la->la_nlink) {
rc = lfsck_namespace_shrink_linkea_cond(env, com,
GOTO(out, rc = 0);
}
- rc = lfsck_namespace_check_name(env, parent, child, cname);
+ rc = lfsck_namespace_check_name(env, lfsck, parent, child,
+ cname);
if (rc == -ENOENT)
goto lost_parent;
/* Add the missing name entry back to the namespace. */
rc = lfsck_namespace_insert_normal(env, com, parent, child,
- cname->ln_name);
+ cname);
if (unlikely(rc == -ESTALE))
/* It may happen when the remote object has been
* removed, but the local MDT is not aware of that. */
* other MDT that references this object with another name,
* so we cannot know whether this linkEA is valid or not.
* So keep it there and maybe resolved when next LFSCK run. */
- if (!(ns->ln_flags & LF_INCOMPLETE)) {
+ if (!(ns->ln_flags & LF_INCOMPLETE) &&
+ (ldata.ld_leh == NULL ||
+ !ldata.ld_leh->leh_overflow_time)) {
/* If the child becomes orphan, then insert it into
* the global .lustre/lost+found/MDTxxxx directory. */
rc = lfsck_namespace_insert_orphan(env, com, child,
PFID(lfsck_dto2fid(child)),
la->la_nlink, count,
lfsck_object_type(child));
- } else {
+ } else if (la->la_nlink < count &&
+ likely(!ldata.ld_leh->leh_overflow_time)) {
rc = lfsck_namespace_repair_nlink(env, com,
child, la);
if (rc > 0) {
rc = 1;
}
+ if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+ rc = lfsck_namespace_check_agent_entry(env, com, child);
+
return rc;
}
struct lfsck_namespace *ns,
__u64 checked_phase1,
__u64 checked_phase2,
- __u32 time_phase1,
- __u32 time_phase2)
+ time64_t time_phase1,
+ time64_t time_phase2, bool dryrun)
{
+ const char *postfix = dryrun ? "inconsistent" : "repaired";
+
seq_printf(m, "checked_phase1: %llu\n"
"checked_phase2: %llu\n"
- "updated_phase1: %llu\n"
- "updated_phase2: %llu\n"
+ "%s_phase1: %llu\n"
+ "%s_phase2: %llu\n"
"failed_phase1: %llu\n"
"failed_phase2: %llu\n"
"directories: %llu\n"
- "dirent_repaired: %llu\n"
- "linkea_repaired: %llu\n"
- "nlinks_repaired: %llu\n"
+ "dirent_%s: %llu\n"
+ "linkea_%s: %llu\n"
+ "nlinks_%s: %llu\n"
"multiple_linked_checked: %llu\n"
- "multiple_linked_repaired: %llu\n"
+ "multiple_linked_%s: %llu\n"
"unknown_inconsistency: %llu\n"
- "unmatched_pairs_repaired: %llu\n"
- "dangling_repaired: %llu\n"
- "multiple_referenced_repaired: %llu\n"
- "bad_file_type_repaired: %llu\n"
- "lost_dirent_repaired: %llu\n"
+ "unmatched_pairs_%s: %llu\n"
+ "dangling_%s: %llu\n"
+ "multiple_referenced_%s: %llu\n"
+ "bad_file_type_%s: %llu\n"
+ "lost_dirent_%s: %llu\n"
"local_lost_found_scanned: %llu\n"
"local_lost_found_moved: %llu\n"
"local_lost_found_skipped: %llu\n"
"local_lost_found_failed: %llu\n"
"striped_dirs_scanned: %llu\n"
- "striped_dirs_repaired: %llu\n"
+ "striped_dirs_%s: %llu\n"
"striped_dirs_failed: %llu\n"
"striped_dirs_disabled: %llu\n"
"striped_dirs_skipped: %llu\n"
"striped_shards_scanned: %llu\n"
- "striped_shards_repaired: %llu\n"
+ "striped_shards_%s: %llu\n"
"striped_shards_failed: %llu\n"
"striped_shards_skipped: %llu\n"
- "name_hash_repaired: %llu\n"
+ "name_hash_%s: %llu\n"
+ "linkea_overflow_%s: %llu\n"
+ "agent_entries_%s: %llu\n"
"success_count: %u\n"
- "run_time_phase1: %u seconds\n"
- "run_time_phase2: %u seconds\n",
+ "run_time_phase1: %lld seconds\n"
+ "run_time_phase2: %lld seconds\n",
checked_phase1,
checked_phase2,
+ dryrun ? "inconsistent" : "updated",
ns->ln_items_repaired,
+ dryrun ? "inconsistent" : "updated",
ns->ln_objs_repaired_phase2,
ns->ln_items_failed,
ns->ln_objs_failed_phase2,
ns->ln_dirs_checked,
- ns->ln_dirent_repaired,
- ns->ln_linkea_repaired,
- ns->ln_objs_nlink_repaired,
+ postfix, ns->ln_dirent_repaired,
+ postfix, ns->ln_linkea_repaired,
+ postfix, ns->ln_objs_nlink_repaired,
ns->ln_mul_linked_checked,
- ns->ln_mul_linked_repaired,
+ postfix, ns->ln_mul_linked_repaired,
ns->ln_unknown_inconsistency,
- ns->ln_unmatched_pairs_repaired,
- ns->ln_dangling_repaired,
- ns->ln_mul_ref_repaired,
- ns->ln_bad_type_repaired,
- ns->ln_lost_dirent_repaired,
+ postfix, ns->ln_unmatched_pairs_repaired,
+ postfix, ns->ln_dangling_repaired,
+ postfix, ns->ln_mul_ref_repaired,
+ postfix, ns->ln_bad_type_repaired,
+ postfix, ns->ln_lost_dirent_repaired,
ns->ln_local_lpf_scanned,
ns->ln_local_lpf_moved,
ns->ln_local_lpf_skipped,
ns->ln_local_lpf_failed,
ns->ln_striped_dirs_scanned,
- ns->ln_striped_dirs_repaired,
+ postfix, ns->ln_striped_dirs_repaired,
ns->ln_striped_dirs_failed,
ns->ln_striped_dirs_disabled,
ns->ln_striped_dirs_skipped,
ns->ln_striped_shards_scanned,
- ns->ln_striped_shards_repaired,
+ postfix, ns->ln_striped_shards_repaired,
ns->ln_striped_shards_failed,
ns->ln_striped_shards_skipped,
- ns->ln_name_hash_repaired,
+ postfix, ns->ln_name_hash_repaired,
+ dryrun ? "inconsistent" : "cleared",
+ ns->ln_linkea_overflow_cleared,
+ postfix, ns->ln_agent_entries_repaired,
ns->ln_success_count,
time_phase1,
time_phase2);
struct lfsck_namespace *ns = com->lc_file_ram;
struct lfsck_assistant_data *lad = com->lc_data;
struct dt_object *root;
- struct dt_object *dto;
int rc;
ENTRY;
memset(ns, 0, sizeof(*ns));
} else {
__u32 count = ns->ln_success_count;
- __u64 last_time = ns->ln_time_last_complete;
+ time64_t last_time = ns->ln_time_last_complete;
memset(ns, 0, sizeof(*ns));
ns->ln_success_count = count;
}
ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
ns->ln_status = LS_INIT;
+ ns->ln_time_latest_reset = ktime_get_real_seconds();
- lfsck_object_put(env, com->lc_obj);
- com->lc_obj = NULL;
- dto = lfsck_namespace_load_one_trace_file(env, com, root,
- LFSCK_NAMESPACE, true);
- if (IS_ERR(dto))
- GOTO(out, rc = PTR_ERR(dto));
+ rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
+ &dt_lfsck_namespace_features,
+ LFSCK_NAMESPACE, true);
+ if (rc)
+ GOTO(out, rc);
- com->lc_obj = dto;
- rc = lfsck_namespace_load_sub_trace_files(env, com, true);
+ rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
+ LFSCK_NAMESPACE, true);
if (rc != 0)
GOTO(out, rc);
- lad->lad_incomplete = 0;
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
CFS_RESET_BITMAP(lad->lad_bitmap);
rc = lfsck_namespace_store(env, com);
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_lmv *llmv = lfsck->li_lmv;
struct lfsck_namespace_req *lnr;
- __u32 size =
- sizeof(*lnr) + LFSCK_TMPBUF_LEN;
- bool wakeup = false;
+ struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
+ __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
+ int rc;
+ bool wakeup = false;
ENTRY;
if (llmv == NULL)
RETURN_EXIT;
+ rc = dt_attr_get(env, lfsck->li_obj_dir, la);
+ if (rc)
+ RETURN_EXIT;
+
OBD_ALLOC(lnr, size);
if (lnr == NULL) {
ns->ln_striped_dirs_skipped++;
}
lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
- NULL, lfsck->li_pos_current.lp_oit_cookie, true);
+ la, lfsck->li_pos_current.lp_oit_cookie, true);
if (IS_ERR(lso)) {
OBD_FREE(lnr, size);
ns->ln_striped_dirs_skipped++;
lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
lnr->lnr_size = size;
+ lnr->lnr_type = lso->lso_attr.la_mode;
spin_lock(&lad->lad_lock);
- if (lad->lad_assistant_status < 0) {
+ if (lad->lad_assistant_status < 0 ||
+ unlikely(!thread_is_running(&lfsck->li_thread) ||
+ !thread_is_running(&lad->lad_thread))) {
spin_unlock(&lad->lad_lock);
lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
ns->ln_striped_dirs_skipped++;
ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
} else {
ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
- ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
- HALF_SEC - lfsck->li_time_last_checkpoint);
- ns->ln_time_last_checkpoint = cfs_time_current_sec();
+ ns->ln_run_time_phase1 += ktime_get_seconds() -
+ lfsck->li_time_last_checkpoint;
+ ns->ln_time_last_checkpoint = ktime_get_real_seconds();
ns->ln_items_checked += com->lc_new_checked;
com->lc_new_checked = 0;
}
}
down_write(&com->lc_sem);
- ns->ln_time_latest_start = cfs_time_current_sec();
+ ns->ln_time_latest_start = ktime_get_real_seconds();
spin_lock(&lfsck->li_lock);
if (ns->ln_flags & LF_SCANNED_ONCE) {
struct linkea_data ldata = { NULL };
__u32 idx = lfsck_dev_idx(lfsck);
int rc;
+ bool remote = false;
ENTRY;
rc = lfsck_links_read(env, obj, &ldata);
GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
}
- if (rc == -ENODATA) {
+ if (rc && rc != -ENODATA)
+ GOTO(out, rc);
+
+ if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
rc = lfsck_namespace_check_for_double_scan(env, com, obj);
GOTO(out, rc);
}
- if (rc != 0)
- GOTO(out, rc);
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL) {
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+ cname, pfid);
+ if (!fid_is_sane(pfid)) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_PARENT, true);
+ } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
+ GOTO(out, rc);
+ } else {
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(pfid), range);
+ if ((rc == -ENOENT) ||
+ (!rc && range->lsr_index != idx)) {
+ remote = true;
+ break;
+ }
+ }
+ if (rc)
+ GOTO(out, rc);
+
+ linkea_next_entry(&ldata);
+ }
+
+ if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+ (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_AGENT_ENTRY, true);
+ if (rc)
+ GOTO(out, rc);
+ }
/* Record multiple-linked object. */
if (ldata.ld_leh->leh_reccount > 1) {
GOTO(out, rc);
}
- linkea_first_entry(&ldata);
- linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
- if (!fid_is_sane(pfid)) {
+ if (remote)
rc = lfsck_namespace_trace_update(env, com, fid,
- LNTF_CHECK_PARENT, true);
- } else {
- fld_range_set_mdt(range);
- rc = fld_local_lookup(env, ss->ss_server_fld,
- fid_seq(pfid), range);
- if ((rc == -ENOENT) ||
- (rc == 0 && range->lsr_index != idx))
- rc = lfsck_namespace_trace_update(env, com, fid,
- LNTF_CHECK_LINKEA, true);
- else
- rc = lfsck_namespace_check_for_double_scan(env, com,
- obj);
- }
+ LNTF_CHECK_LINKEA, true);
+ else
+ rc = lfsck_namespace_check_for_double_scan(env, com, obj);
GOTO(out, rc);
l_wait_event(mthread->t_ctl_waitq,
lad->lad_prefetched < bk->lb_async_windows ||
!thread_is_running(mthread) ||
- thread_is_stopped(athread),
+ !thread_is_running(athread),
&lwi);
- if (unlikely(!thread_is_running(mthread)) ||
- thread_is_stopped(athread))
+ if (unlikely(!thread_is_running(mthread) ||
+ !thread_is_running(athread)))
return 0;
if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
}
spin_lock(&lad->lad_lock);
- if (lad->lad_assistant_status < 0) {
+ if (lad->lad_assistant_status < 0 ||
+ unlikely(!thread_is_running(mthread) ||
+ !thread_is_running(athread))) {
spin_unlock(&lad->lad_lock);
lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
return lad->lad_assistant_status;
spin_unlock(&lfsck->li_lock);
if (!init) {
- ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
- HALF_SEC - lfsck->li_time_last_checkpoint);
- ns->ln_time_last_checkpoint = cfs_time_current_sec();
+ ns->ln_run_time_phase1 += ktime_get_seconds() -
+ lfsck->li_time_last_checkpoint;
+ ns->ln_time_last_checkpoint = ktime_get_real_seconds();
ns->ln_items_checked += com->lc_new_checked;
com->lc_new_checked = 0;
}
if (ns->ln_status == LS_SCANNING_PHASE1) {
struct lfsck_position pos;
- const struct dt_it_ops *iops;
- cfs_duration_t duration = cfs_time_current() -
- lfsck->li_time_last_checkpoint;
- __u64 checked = ns->ln_items_checked + com->lc_new_checked;
- __u64 speed = checked;
- __u64 new_checked = com->lc_new_checked *
- msecs_to_jiffies(MSEC_PER_SEC);
- __u32 rtime = ns->ln_run_time_phase1 +
- cfs_duration_sec(duration + HALF_SEC);
+ time64_t duration = ktime_get_seconds() -
+ lfsck->li_time_last_checkpoint;
+ u64 checked = ns->ln_items_checked + com->lc_new_checked;
+ u64 speed = checked;
+ u64 new_checked = com->lc_new_checked;
+ time64_t rtime = ns->ln_run_time_phase1 + duration;
if (duration != 0)
- do_div(new_checked, duration);
+ new_checked = div64_s64(new_checked, duration);
if (rtime != 0)
- do_div(speed, rtime);
+ speed = div64_s64(speed, rtime);
- lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
+ lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
+ bk->lb_param & LPF_DRYRUN);
seq_printf(m, "average_speed_phase1: %llu items/sec\n"
"average_speed_phase2: N/A\n"
"average_speed_total: %llu items/sec\n"
speed,
new_checked);
- LASSERT(lfsck->li_di_oit != NULL);
-
- iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
-
- /* The low layer otable-based iteration position may NOT
- * exactly match the namespace-based directory traversal
- * cookie. Generally, it is not a serious issue. But the
- * caller should NOT make assumption on that. */
- pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
- if (!lfsck->li_current_oit_processed)
- pos.lp_oit_cookie--;
-
- spin_lock(&lfsck->li_lock);
- if (lfsck->li_di_dir != NULL) {
- pos.lp_dir_cookie = lfsck->li_cookie_dir;
- if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
+ if (likely(lfsck->li_di_oit)) {
+ const struct dt_it_ops *iops =
+ &lfsck->li_obj_oit->do_index_ops->dio_it;
+
+ /* The low layer otable-based iteration position may NOT
+ * exactly match the namespace-based directory traversal
+ * cookie. Generally, it is not a serious issue. But the
+ * caller should NOT make assumption on that. */
+ pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
+ if (!lfsck->li_current_oit_processed)
+ pos.lp_oit_cookie--;
+
+ spin_lock(&lfsck->li_lock);
+ if (lfsck->li_di_dir) {
+ pos.lp_dir_cookie = lfsck->li_cookie_dir;
+ if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
+ fid_zero(&pos.lp_dir_parent);
+ pos.lp_dir_cookie = 0;
+ } else {
+ pos.lp_dir_parent =
+ *lfsck_dto2fid(lfsck->li_obj_dir);
+ }
+ } else {
fid_zero(&pos.lp_dir_parent);
pos.lp_dir_cookie = 0;
- } else {
- pos.lp_dir_parent =
- *lfsck_dto2fid(lfsck->li_obj_dir);
}
+ spin_unlock(&lfsck->li_lock);
} else {
- fid_zero(&pos.lp_dir_parent);
- pos.lp_dir_cookie = 0;
+ pos = ns->ln_pos_last_checkpoint;
}
- spin_unlock(&lfsck->li_lock);
+
lfsck_pos_dump(m, &pos, "current_position");
} else if (ns->ln_status == LS_SCANNING_PHASE2) {
- cfs_duration_t duration = cfs_time_current() -
- com->lc_time_last_checkpoint;
+ time64_t duration = ktime_get_seconds() -
+ com->lc_time_last_checkpoint;
__u64 checked = ns->ln_objs_checked_phase2 +
com->lc_new_checked;
__u64 speed1 = ns->ln_items_checked;
__u64 speed2 = checked;
__u64 speed0 = speed1 + speed2;
- __u64 new_checked = com->lc_new_checked *
- msecs_to_jiffies(MSEC_PER_SEC);
- __u32 rtime = ns->ln_run_time_phase2 +
- cfs_duration_sec(duration + HALF_SEC);
- __u32 time0 = ns->ln_run_time_phase1 + rtime;
+ __u64 new_checked = com->lc_new_checked;
+ time64_t rtime = ns->ln_run_time_phase2 + duration;
+ time64_t time0 = ns->ln_run_time_phase1 + rtime;
if (duration != 0)
- do_div(new_checked, duration);
+ new_checked = div64_s64(new_checked, duration);
if (ns->ln_run_time_phase1 != 0)
- do_div(speed1, ns->ln_run_time_phase1);
+ speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
else if (ns->ln_items_checked != 0)
time0++;
if (rtime != 0)
- do_div(speed2, rtime);
+ speed2 = div64_s64(speed2, rtime);
else if (checked != 0)
time0++;
if (time0 != 0)
- do_div(speed0, time0);
+ speed0 = div64_s64(speed0, time0);
lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
checked,
- ns->ln_run_time_phase1, rtime);
+ ns->ln_run_time_phase1, rtime,
+ bk->lb_param & LPF_DRYRUN);
seq_printf(m, "average_speed_phase1: %llu items/sec\n"
"average_speed_phase2: %llu objs/sec\n"
"average_speed_total: %llu items/sec\n"
__u64 speed1 = ns->ln_items_checked;
__u64 speed2 = ns->ln_objs_checked_phase2;
__u64 speed0 = speed1 + speed2;
- __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
+ time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
if (ns->ln_run_time_phase1 != 0)
- do_div(speed1, ns->ln_run_time_phase1);
+ speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
else if (ns->ln_items_checked != 0)
time0++;
if (ns->ln_run_time_phase2 != 0)
- do_div(speed2, ns->ln_run_time_phase2);
+ speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
else if (ns->ln_objs_checked_phase2 != 0)
time0++;
if (time0 != 0)
- do_div(speed0, time0);
+ speed0 = div64_s64(speed0, time0);
lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
ns->ln_objs_checked_phase2,
ns->ln_run_time_phase1,
- ns->ln_run_time_phase2);
+ ns->ln_run_time_phase2,
+ bk->lb_param & LPF_DRYRUN);
seq_printf(m, "average_speed_phase1: %llu items/sec\n"
"average_speed_phase2: %llu objs/sec\n"
"average_speed_total: %llu items/sec\n"
static int lfsck_namespace_in_notify(const struct lu_env *env,
struct lfsck_component *com,
- struct lfsck_request *lr,
- struct thandle *th)
+ struct lfsck_request *lr)
{
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_namespace *ns = com->lc_file_ram;
- struct lfsck_assistant_data *lad = com->lc_data;
- struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
- struct lfsck_tgt_desc *ltd;
- int rc = 0;
- bool fail = false;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ int rc = 0;
+ bool fail = false;
ENTRY;
switch (lr->lr_event) {
- case LE_SKIP_NLINK_DECLARE: {
- struct dt_object *obj;
- struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
- int idx;
- __u8 flags = 0;
-
- LASSERT(th != NULL);
-
- idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
- obj = com->lc_sub_trace_objs[idx].lsto_obj;
- fid_cpu_to_be(key, &lr->lr_fid);
- mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
- rc = dt_declare_delete(env, obj,
- (const struct dt_key *)key, th);
- if (rc == 0)
- rc = dt_declare_insert(env, obj,
- (const struct dt_rec *)&flags,
- (const struct dt_key *)key, th);
- mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
-
- RETURN(rc);
- }
- case LE_SKIP_NLINK: {
- struct dt_object *obj;
- struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
- int idx;
- __u8 flags = 0;
- bool exist = false;
- ENTRY;
-
- LASSERT(th != NULL);
-
- idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
- obj = com->lc_sub_trace_objs[idx].lsto_obj;
- fid_cpu_to_be(key, &lr->lr_fid);
- mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
- rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
- (const struct dt_key *)key);
- if (rc == 0) {
- if (flags & LNTF_SKIP_NLINK) {
- mutex_unlock(
- &com->lc_sub_trace_objs[idx].lsto_mutex);
-
- RETURN(0);
- }
-
- exist = true;
- } else if (rc != -ENOENT) {
- GOTO(log, rc);
- }
-
- flags |= LNTF_SKIP_NLINK;
- if (exist) {
- rc = dt_delete(env, obj, (const struct dt_key *)key,
- th);
- if (rc != 0)
- GOTO(log, rc);
- }
-
- rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
- (const struct dt_key *)key, th, 1);
-
- GOTO(log, rc);
-
-log:
- mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
- CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
- " to be skipped for namespace double scan: rc = %d\n",
- lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
-
- if (rc != 0)
- /* If we cannot record this object in the LFSCK tracing,
- * we have to mark the LFSC as LF_INCOMPLETE, then the
- * LFSCK will skip nlink attribute verification for
- * all objects. */
- ns->ln_flags |= LF_INCOMPLETE;
-
- return 0;
- }
case LE_SET_LMV_MASTER: {
struct dt_object *obj;
struct dt_object *child,
struct lfsck_namespace_req *lnr)
{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lu_attr *la = &info->lti_la;
- struct dt_allocation_hint *hint = &info->lti_hint;
- struct dt_object_format *dof = &info->lti_dof;
- struct dt_insert_rec *rec = &info->lti_dt_rec;
- struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
- const struct lu_name *cname;
- const struct lu_fid *pfid = lfsck_dto2fid(parent);
- const struct lu_fid *cfid = lfsck_dto2fid(child);
- struct linkea_data ldata = { NULL };
- struct lfsck_lock_handle *llh = &info->lti_llh;
- struct lu_buf linkea_buf;
- struct lu_buf lmv_buf;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
- struct dt_device *dev = lfsck->li_next;
- struct thandle *th = NULL;
- int rc = 0;
- __u16 type = lnr->lnr_type;
- bool create;
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_allocation_hint *hint = &info->lti_hint;
+ struct dt_object_format *dof = &info->lti_dof;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
+ const struct lu_name *cname;
+ const struct lu_fid *pfid = lfsck_dto2fid(parent);
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct linkea_data ldata = { NULL };
+ struct lfsck_lock_handle *llh = &info->lti_llh;
+ struct lustre_handle rlh = { 0 };
+ struct lustre_handle clh = { 0 };
+ struct lu_buf linkea_buf;
+ struct lu_buf lmv_buf;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ int rc = 0;
+ __u16 type = lnr->lnr_type;
+ bool create;
ENTRY;
cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
if (IS_ERR(child))
GOTO(log, rc = PTR_ERR(child));
- rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
- if (rc != 0)
- GOTO(log, rc);
-
- rc = linkea_add_buf(&ldata, cname, pfid);
+ rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
+ cname, pfid);
if (rc != 0)
GOTO(log, rc);
rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
- MDS_INODELOCK_UPDATE, LCK_PR);
+ MDS_INODELOCK_UPDATE, LCK_PW);
if (rc != 0)
GOTO(log, rc);
if (rc != 0)
GOTO(log, rc);
+ if (dt_object_remote(child)) {
+ rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
+ LCK_EX);
+ if (rc != 0)
+ GOTO(log, rc);
+ }
+
+ rc = lfsck_ibits_lock(env, lfsck, child, &clh,
+ MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc != 0)
+ GOTO(unlock_remote_lookup, rc);
+
/* Set the ctime as zero, then others can know it is created for
* repairing dangling name entry by LFSCK. And if the LFSCK made
* wrong decision and the real MDT-object has been found later,
* then the LFSCK has chance to fix the incosistency properly. */
memset(la, 0, sizeof(*la));
- la->la_mode = (type & S_IFMT) | 0600;
- la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
- LA_ATIME | LA_MTIME | LA_CTIME;
-
- child->do_ops->do_ah_init(env, hint, parent, child,
- la->la_mode & S_IFMT);
+ if (S_ISDIR(type))
+ la->la_mode = (type & S_IFMT) | 0700;
+ else
+ la->la_mode = (type & S_IFMT) | 0600;
+ la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
+
+ /*
+ * if it's directory, skip do_ah_init() to create a plain directory
+ * because it may have shards already, which will be inserted back
+ * later, besides, it may be remote, and creating stripe directory
+ * remotely is not supported.
+ */
+ if (S_ISREG(type))
+ child->do_ops->do_ah_init(env, hint, parent, child, type);
memset(dof, 0, sizeof(*dof));
dof->dof_type = dt_mode_to_dft(type);
th = dt_trans_create(env, dev);
if (IS_ERR(th))
- GOTO(log, rc = PTR_ERR(th));
+ GOTO(unlock_child, rc = PTR_ERR(th));
/* 1a. create child. */
rc = dt_declare_create(env, child, la, hint, dof, th);
lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
rc = dt_declare_xattr_set(env, child, &lmv_buf,
- XATTR_NAME_LMV, 0, th);
+ XATTR_NAME_LMV".set", 0, th);
if (rc != 0)
GOTO(stop, rc);
}
if (rc != 0)
GOTO(stop, rc);
+ /* 7a. if child is remote, delete and insert to generate local agent */
+ if (dt_object_remote(child)) {
+ rc = dt_declare_delete(env, parent,
+ (const struct dt_key *)lnr->lnr_name,
+ th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)lnr->lnr_name,
+ th);
+ if (rc)
+ GOTO(stop, rc);
+ }
+
rc = dt_trans_start_local(env, dev, th);
if (rc != 0)
GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
rec->rec_type = S_IFDIR;
rec->rec_fid = cfid;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dot, th, 1);
+ (const struct dt_key *)dot, th);
if (rc != 0)
GOTO(unlock, rc);
/* 4b. insert dotdot into child dir */
rec->rec_fid = pfid;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock, rc);
/* 5b. generate slave LMV EA. */
if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
- rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
- 0, th);
+ rc = dt_xattr_set(env, child, &lmv_buf,
+ XATTR_NAME_LMV".set", 0, th);
if (rc != 0)
GOTO(unlock, rc);
}
/* 6b. insert linkEA for child. */
rc = dt_xattr_set(env, child, &linkea_buf,
XATTR_NAME_LINK, 0, th);
+ if (rc)
+ GOTO(unlock, rc);
+
+ /* 7b. if child is remote, delete and insert to generate local agent */
+ if (dt_object_remote(child)) {
+ rc = dt_delete(env, parent,
+ (const struct dt_key *)lnr->lnr_name, th);
+ if (rc)
+ GOTO(unlock, rc);
+
+ rec->rec_type = type;
+ rec->rec_fid = cfid;
+ rc = dt_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)lnr->lnr_name, th);
+ if (rc)
+ GOTO(unlock, rc);
+ }
GOTO(unlock, rc);
stop:
dt_trans_stop(env, dev, th);
+unlock_child:
+ lfsck_ibits_unlock(&clh, LCK_EX);
+unlock_remote_lookup:
+ if (dt_object_remote(child))
+ lfsck_ibits_unlock(&rlh, LCK_EX);
log:
lfsck_unlock(llh);
CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
goto again;
}
+ LASSERT(handle != NULL);
+
if (dir == NULL) {
dir = lfsck_assistant_object_load(env, lfsck, lso);
if (IS_ERR(dir)) {
}
rc = linkea_add_buf(&ldata, cname, pfid);
- if (rc != 0)
- GOTO(stop, rc);
-
- rc = lfsck_links_write(env, obj, &ldata, handle);
- if (unlikely(rc == -ENOSPC) &&
- S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
- if (handle != NULL) {
- LASSERT(dt_write_locked(env, obj));
-
- dt_write_unlock(env, obj);
- dtlocked = false;
-
- dt_trans_stop(env, dev, handle);
- handle = NULL;
-
- lfsck_ibits_unlock(&lh, LCK_EX);
- }
-
- rc = lfsck_namespace_trace_update(env, com,
- &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
- if (rc != 0)
- /* If we cannot record this object in the
- * LFSCK tracing, we have to mark the LFSCK
- * as LF_INCOMPLETE, then the LFSCK will
- * skip nlink attribute verification for
- * all objects. */
- ns->ln_flags |= LF_INCOMPLETE;
-
- GOTO(out, rc = 0);
- }
-
+ if (rc == 0)
+ rc = lfsck_links_write(env, obj, &ldata, handle);
if (rc != 0)
GOTO(stop, rc);
if (log)
CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
"repaired the entry: "DFID", parent "DFID
- ", name %.*s\n", lfsck_lfsck2name(lfsck),
+ ", name %.*s, type %d\n",
+ lfsck_lfsck2name(lfsck),
PFID(&lnr->lnr_fid), PFID(pfid),
- lnr->lnr_namelen, lnr->lnr_name);
+ lnr->lnr_namelen, lnr->lnr_name, type);
switch (type) {
case LNIT_DANGLING:
/* b5. insert child's FID into the LFSCK trace file. */
rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
- (const struct dt_key *)key, th, 1);
+ (const struct dt_key *)key, th);
GOTO(stop, rc = (rc == 0 ? 1 : rc));
down_write(&com->lc_sem);
com->lc_new_checked++;
com->lc_new_scanned++;
- if (rc >= 0 && fid_is_sane(&fid))
+ if (rc >= 0)
ns->ln_fid_latest_scanned_phase2 = fid;
+
if (rc > 0)
ns->ln_objs_repaired_phase2++;
else if (rc < 0)
if (rc < 0 && bk->lb_param & LPF_FAILOUT)
GOTO(put, rc);
- if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
- cfs_time_current())) &&
+ if (unlikely(com->lc_time_next_checkpoint <=
+ ktime_get_seconds()) &&
com->lc_new_checked != 0) {
down_write(&com->lc_sem);
- ns->ln_run_time_phase2 +=
- cfs_duration_sec(cfs_time_current() +
- HALF_SEC - com->lc_time_last_checkpoint);
- ns->ln_time_last_checkpoint = cfs_time_current_sec();
+ ns->ln_run_time_phase2 += ktime_get_seconds() -
+ com->lc_time_last_checkpoint;
+ ns->ln_time_last_checkpoint = ktime_get_real_seconds();
ns->ln_objs_checked_phase2 += com->lc_new_checked;
com->lc_new_checked = 0;
- rc = lfsck_namespace_store(env, com);
+ lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
- if (rc != 0)
- GOTO(put, rc);
- com->lc_time_last_checkpoint = cfs_time_current();
+ com->lc_time_last_checkpoint = ktime_get_seconds();
com->lc_time_next_checkpoint =
com->lc_time_last_checkpoint +
- cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+ LFSCK_CHECKPOINT_INTERVAL;
}
lfsck_control_speed_by_self(com);
com->lc_new_checked = 0;
com->lc_new_scanned = 0;
- com->lc_time_last_checkpoint = cfs_time_current();
+ com->lc_time_last_checkpoint = ktime_get_seconds();
com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
- cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+ LFSCK_CHECKPOINT_INTERVAL;
i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
rc = lfsck_namespace_double_scan_one_trace_file(env, com,
struct lfsck_namespace *ns = com->lc_file_ram;
down_write(&com->lc_sem);
- ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
- HALF_SEC - com->lc_time_last_checkpoint);
- ns->ln_time_last_checkpoint = cfs_time_current_sec();
+ ns->ln_run_time_phase2 += ktime_get_seconds() -
+ com->lc_time_last_checkpoint;
+ ns->ln_time_last_checkpoint = ktime_get_real_seconds();
ns->ln_objs_checked_phase2 += com->lc_new_checked;
com->lc_new_checked = 0;
ns->ln_status = LS_PARTIAL;
else
ns->ln_status = LS_COMPLETED;
+ ns->ln_flags &= ~LF_SCANNED_ONCE;
if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
- ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
+ ns->ln_flags &= ~LF_INCONSISTENT;
ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
ns->ln_success_count++;
} else if (rc == 0) {
int rc = 0;
ENTRY;
- if (!lad->lad_incomplete)
+ if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
RETURN_EXIT;
set = ptlrpc_prep_set();
down_read(<ds->ltd_rw_sem);
cfs_foreach_bit(lad->lad_bitmap, idx) {
ltd = lfsck_ltd2tgt(ltds, idx);
- LASSERT(ltd != NULL);
+ if (unlikely(!ltd))
+ continue;
laia->laia_ltd = ltd;
rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
}
up_read(<ds->ltd_rw_sem);
- rc = ptlrpc_set_wait(set);
+ rc = ptlrpc_set_wait(env, set);
ptlrpc_set_destroy(set);
GOTO(out, rc);
LASSERT(S_ISDIR(lfsck_object_type(obj)));
- rc = lfsck_links_read(env, obj, &ldata);
+ rc = lfsck_links_read_with_rec(env, obj, &ldata);
if (rc == -ENODATA) {
dirty = true;
} else if (rc == 0) {
if (!dirty)
RETURN(rc);
- rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
- if (rc != 0)
- RETURN(rc);
-
- rc = linkea_add_buf(&ldata, cname, pfid);
+ rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
+ cname, pfid);
if (rc != 0)
RETURN(rc);
struct linkea_data ldata = { NULL };
int rc;
- rc = lfsck_links_read(env, obj, &ldata);
- if (rc != 0)
+ rc = lfsck_links_read_with_rec(env, obj, &ldata);
+ if (rc)
return rc;
linkea_first_entry(&ldata);
- if (ldata.ld_lee == NULL)
- return -ENODATA;
-
linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
+ if (!linkea_entry_is_valid(&ldata, cname, pfid))
+ return -EINVAL;
+
/* To guarantee the 'name' is terminated with '0'. */
memcpy(name, cname->ln_name, cname->ln_namelen);
name[cname->ln_namelen] = 0;
GOTO(stop, rc);
rc = dt_insert(env, dir, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)name, th);
if (rc == 0 && S_ISDIR(type) && !exists) {
dt_write_lock(env, dir, 0);
rc = dt_ref_add(env, dir, th);
obj = local_index_find_or_create(env, lfsck->li_los, root,
LFSCK_NAMESPACE,
S_IFREG | S_IRUGO | S_IWUSR,
- &dt_lfsck_features);
+ &dt_lfsck_namespace_features);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
com->lc_obj = obj;
rc = lfsck_namespace_load(env, com);
- if (rc == -ENODATA)
+ if (rc == -ENODATA) {
rc = lfsck_namespace_init(env, com);
- else if (rc < 0)
+ } else if (rc < 0) {
rc = lfsck_namespace_reset(env, com, true);
- else
- rc = lfsck_namespace_load_sub_trace_files(env, com, false);
+ } else {
+ rc = lfsck_load_sub_trace_files(env, com,
+ &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
+ if (rc)
+ rc = lfsck_namespace_reset(env, com, true);
+ }
if (rc != 0)
GOTO(out, rc);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff