* GPL HEADER END
*/
/*
- * Copyright (c) 2014, Intel Corporation.
+ * Copyright (c) 2014, 2017, Intel Corporation.
*/
/*
* lustre/lfsck/lfsck_striped_dir.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_lib.h>
#include <lustre_net.h>
#include <lustre_lmv.h>
-#include <lustre/lustre_user.h>
#include "lfsck_internal.h"
* \param[in] del_lmv true if need to drop the LMV EA
*
* \retval positive number if nothing to be done
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
static int lfsck_disable_master_lmv(const struct lu_env *env,
* striped directory to be handled and other information
*
* \retval positive number if nothing to be done
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
static int lfsck_remove_lmv(const struct lu_env *env,
* \param[in] index the old shard's index in the striped directory
* \param[in] flags the new shard's flags in the @lslr slot
*
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
static int lfsck_replace_lmv(const struct lu_env *env,
* we define the max depth can be called recursively
* (LFSCK_REC_LMV_MAX_DEPTH)
*
- * \retval zero for succeed
+ * \retval zero for success
* \retval "-ERANGE" for invalid @shard_idx
* \retval "-EEXIST" for the required lslr slot has been
* occupied by other shard
return rc > 0 ? 0 : rc;
}
-int lfsck_read_stripe_lmv(const struct lu_env *env, struct dt_object *obj,
+/**
+ * Read LMV from bottom object, so it doesn't contain stripe FIDs.
+ *
+ * TODO: test migrating/foreign directory lfsck
+ *
+ * \param[in] env thread env
+ * \param[in] lfsck lfsck instance
+ * \param[in] obj dt object
+ * \param[out] lmv LMV data pointer
+ *
+ * \retval 0 on success
+ * \retval -ENODATA on no LMV, corrupt LMV, dir is dead or foreign
+ * -ev on other failures
+ */
+int lfsck_read_stripe_lmv(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj,
struct lmv_mds_md_v1 *lmv)
{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_buf *buf = &info->lti_buf;
+ struct lmv_foreign_md *lfm;
int rc;
+ /* use bottom object to avoid reading in shard FIDs */
+ obj = lfsck_object_find_bottom(env, lfsck, lu_object_fid(&obj->do_lu));
+ if (IS_ERR(obj))
+ return PTR_ERR(obj);
+
dt_read_lock(env, obj, 0);
- rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lmv, sizeof(*lmv)),
- XATTR_NAME_LMV);
+ buf->lb_buf = lmv;
+ buf->lb_len = sizeof(*lmv);
+ rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LMV);
+ if (unlikely(rc == -ERANGE)) {
+ buf = &info->lti_big_buf;
+ /* this may be a foreign LMV */
+ rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LMV);
+ if (rc > sizeof(*lmv)) {
+ int rc1;
+
+ lu_buf_check_and_alloc(buf, rc);
+ rc1 = dt_xattr_get(env, obj, buf, XATTR_NAME_LMV);
+ if (rc != rc1)
+ rc = -ENODATA;
+ } else {
+ rc = -ENODATA;
+ }
+ }
dt_read_unlock(env, obj);
- if (rc != sizeof(*lmv))
- return rc > 0 ? -EINVAL : rc;
- lfsck_lmv_header_le_to_cpu(lmv, lmv);
- if ((lmv->lmv_magic == LMV_MAGIC &&
- !(lmv->lmv_hash_type & LMV_HASH_FLAG_MIGRATION)) ||
- (lmv->lmv_magic == LMV_MAGIC_STRIPE &&
- !(lmv->lmv_hash_type & LMV_HASH_FLAG_DEAD)))
- return 0;
+ lfsck_object_put(env, obj);
+
+ if (rc > offsetof(typeof(*lfm), lfm_value) &&
+ *((__u32 *)buf->lb_buf) == LMV_MAGIC_FOREIGN) {
+ __u32 value_len;
+
+ lfm = buf->lb_buf;
+ value_len = le32_to_cpu(lfm->lfm_length);
+ CDEBUG(D_INFO,
+ "foreign LMV EA, magic %x, len %u, type %x, flags %x, for dir "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 (rc != value_len + offsetof(typeof(*lfm), lfm_value))
+ CDEBUG(D_LFSCK,
+ "foreign LMV EA internal size %u does not match EA full size %d for dir "DFID"\n",
+ value_len, rc, PFID(lfsck_dto2fid(obj)));
+
+ /* no further usage/decode of foreign LMV outside */
+ return -ENODATA;
+ }
- return -ENODATA;
+ if (rc == sizeof(*lmv)) {
+ rc = 0;
+ lfsck_lmv_header_le_to_cpu(lmv, lmv);
+ /* if LMV is corrupt, return -ENODATA */
+ if (lmv->lmv_magic != LMV_MAGIC_V1 &&
+ lmv->lmv_magic != LMV_MAGIC_STRIPE)
+ rc = -ENODATA;
+ } else if (rc >= 0) {
+ /* LMV is corrupt */
+ rc = -ENODATA;
+ }
+
+ return rc;
}
/**
* \retval negative error number on failure
*/
int lfsck_namespace_check_name(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
struct dt_object *parent,
struct dt_object *child,
const struct lu_name *cname)
int idx;
int rc;
- rc = lfsck_read_stripe_lmv(env, parent, lmv);
+ rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
if (rc != 0)
RETURN(rc == -ENODATA ? 0 : rc);
* \param[in] index the MDT index on which the LFSCK instance to be notified
*
* \retval positive number if nothing to be done
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
static int lfsck_namespace_notify_lmv_remote(const struct lu_env *env,
* \param[in] obj pointer to the striped directory to be rescanned
*
* \retval positive number if nothing to be done
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
int lfsck_namespace_notify_lmv_master_local(const struct lu_env *env,
if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
RETURN(0);
- rc = lfsck_read_stripe_lmv(env, obj, lmv4);
+ rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv4);
if (rc != 0)
RETURN(rc);
else
count = lmv4->lmv_stripe_count;
- OBD_ALLOC_LARGE(lslr, sizeof(struct lfsck_slave_lmv_rec) * count);
+ OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * count);
if (lslr == NULL) {
OBD_FREE_PTR(llu);
lfsck_lmv_put(env, llmv);
} else {
ns->ln_striped_dirs_repaired++;
+ llmv->ll_counted = 1;
spin_lock(&lfsck->li_lock);
list_add_tail(&llu->llu_link, &lfsck->li_list_lmv);
spin_unlock(&lfsck->li_lock);
* \param[in] flags to indicate which element(s) in the LMV EA will be set
*
* \retval positive number if nothing to be done
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
static int lfsck_namespace_set_lmv_master(const struct lu_env *env,
if (rc != 0)
GOTO(log, rc);
- rc = lfsck_read_stripe_lmv(env, obj, lmv3);
+ rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv3);
if (rc == -ENODATA) {
if (!(flags & LEF_SET_LMV_ALL))
GOTO(log, rc);
lmv3->lmv_magic = LMV_MAGIC;
lmv3->lmv_master_mdt_index = pidx;
+ lmv3->lmv_layout_version++;
if (flags & LEF_SET_LMV_ALL) {
rc = lfsck_allow_regenerate_master_lmv(env, com, obj,
* \param[in] name the name of the bad name hash
*
* \retval positive number if nothing to be done
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
int lfsck_namespace_repair_bad_name_hash(const struct lu_env *env,
if (IS_ERR(parent))
GOTO(log, rc = PTR_ERR(parent));
+ if (unlikely(!dt_object_exists(parent)))
+ /* The parent object was previously accessed when verifying
+ * the slave LMV EA. If this condition is true it is because
+ * the striped directory is being removed. */
+ GOTO(log, rc = 1);
+
*lmv2 = llmv->ll_lmv;
lmv2->lmv_hash_type = LMV_HASH_TYPE_UNKNOWN | LMV_HASH_FLAG_BAD_TYPE;
rc = lfsck_namespace_set_lmv_master(env, com, parent, lmv2,
__u16 type;
ENTRY;
- rc = lfsck_read_stripe_lmv(env, child, lmv);
+ rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
if (rc != 0)
RETURN(rc == -ENODATA ? 1 : rc);
ns->ln_flags |= LF_INCONSISTENT;
rc = lfsck_namespace_repair_bad_name_hash(env, com,
child, llmv, ent->lde_name);
- if (rc >= 0)
+ if (rc == 0)
ns->ln_name_hash_repaired++;
}
* \param[in] obj pointer to the object which LMV EA will be checked
* \param[in] llmv pointer to buffer holding the slave LMV EA
*
- * \retval zero for succeed
+ * \retval positive number if nothing to be done
+ * \retval zero for success
* \retval negative error number on failure
*/
int lfsck_namespace_verify_stripe_slave(const struct lu_env *env,
GOTO(out, rc);
}
+ CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ENGINE_DELAY, cfs_fail_val);
+
parent = lfsck_object_find_bottom(env, lfsck, pfid);
if (IS_ERR(parent)) {
rc = lfsck_namespace_trace_update(env, com, cfid,
GOTO(out, rc);
}
+ if (unlikely(!dt_object_exists(parent)))
+ GOTO(out, rc = 1);
+
if (unlikely(!dt_try_as_dir(env, parent)))
GOTO(out, rc = -ENOTDIR);
- rc = lfsck_read_stripe_lmv(env, parent, plmv);
+ rc = lfsck_read_stripe_lmv(env, lfsck, parent, plmv);
if (rc != 0) {
int rc1;
*
* \param[in] env pointer to the thread context
* \param[in] com pointer to the lfsck component
- * \paran[in] dir pointer to the striped directory or its shard to be
- * rescanned
* \param[in] lnr pointer to the namespace request that contains the
* striped directory or the shard
*
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
int lfsck_namespace_striped_dir_rescan(const struct lu_env *env,
struct lfsck_component *com,
- struct dt_object *dir,
struct lfsck_namespace_req *lnr)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lfsck_lmv *llmv = lnr->lnr_lmv;
struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
- const struct lu_fid *pfid = lfsck_dto2fid(dir);
+ struct lfsck_assistant_object *lso = lnr->lnr_lar.lar_parent;
+ const struct lu_fid *pfid = &lso->lso_fid;
+ struct dt_object *dir = NULL;
+ struct dt_object *obj = NULL;
struct lu_seq_range *range = &info->lti_range;
struct seq_server_site *ss = lfsck_dev_site(lfsck);
__u32 stripe_count;
if (llmv->ll_lmv_slave) {
if (llmv->ll_lmv_verified) {
ns->ln_striped_shards_scanned++;
- lfsck_namespace_trace_update(env, com,
- lfsck_dto2fid(dir),
+ lfsck_namespace_trace_update(env, com, pfid,
LNTF_UNCERTAIN_LMV |
LNTF_RECHECK_NAME_HASH, false);
}
}
if (llmv->ll_lmv_updated) {
+ if (dir == NULL) {
+ dir = lfsck_assistant_object_load(env, lfsck, lso);
+ if (IS_ERR(dir)) {
+ rc = PTR_ERR(dir);
+
+ RETURN(rc == -ENOENT ? 0 : rc);
+ }
+ }
+
lmv->lmv_layout_version++;
rc = lfsck_namespace_update_lmv(env, com, dir, lmv, false);
if (rc != 0)
RETURN(rc);
ns->ln_striped_dirs_scanned++;
- ns->ln_striped_dirs_repaired++;
+ if (!llmv->ll_counted)
+ ns->ln_striped_dirs_repaired++;
}
fld_range_set_mdt(range);
for (i = 0; i <= llmv->ll_max_filled_off; i++) {
- struct dt_object *obj = NULL;
struct lfsck_slave_lmv_rec *lslr = llmv->ll_lslr + i;
const struct lu_fid *cfid = &lslr->lslr_fid;
const struct lu_name *cname;
struct linkea_data ldata = { NULL };
- int len;
int rc1 = 0;
bool repair_linkea = false;
bool repair_lmvea = false;
if (fid_is_zero(cfid))
continue;
- len = snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
- DFID":%u", PFID(cfid), i);
- cname = lfsck_name_get_const(env, info->lti_tmpbuf, len);
- memcpy(lnr->lnr_name, info->lti_tmpbuf, len);
-
- obj = lfsck_object_find_bottom_nowait(env, lfsck, cfid);
+ lnr->lnr_fid = *cfid;
+ lnr->lnr_namelen = snprintf(lnr->lnr_name,
+ lnr->lnr_size - sizeof(*lnr),
+ DFID":%u", PFID(cfid), i);
+ cname = lfsck_name_get_const(env, lnr->lnr_name,
+ lnr->lnr_namelen);
+ obj = lfsck_object_find_bottom(env, lfsck, cfid);
if (IS_ERR(obj)) {
- if (lfsck_is_dead_obj(dir))
- RETURN(0);
+ if (dir == NULL) {
+ dir = lfsck_assistant_object_load(env, lfsck,
+ lso);
+ if (IS_ERR(dir)) {
+ if (PTR_ERR(dir) == -ENOENT)
+ RETURN(0);
+
+ dir = NULL;
+ }
+ } else if (lfsck_is_dead_obj(dir)) {
+ GOTO(out, rc = 0);
+ }
rc1 = PTR_ERR(obj);
goto next;
break;
}
- rc1 = lfsck_links_read(env, obj, &ldata);
+ rc1 = lfsck_links_read_with_rec(env, obj, &ldata);
if (rc1 == -ENOENT) {
create = true;
goto repair;
repair:
if (create) {
+ if (dir == NULL) {
+ dir = lfsck_assistant_object_load(env, lfsck,
+ lso);
+ if (IS_ERR(dir)) {
+ rc1 = PTR_ERR(dir);
+
+ if (rc1 == -ENOENT)
+ GOTO(out, rc = 0);
+
+ dir = NULL;
+ goto next;
+ }
+ }
+
rc1 = lfsck_namespace_repair_dangling(env, com, dir,
obj, lnr);
if (rc1 >= 0) {
}
if (rename) {
+ if (dir == NULL) {
+ dir = lfsck_assistant_object_load(env, lfsck,
+ lso);
+ if (IS_ERR(dir)) {
+ rc1 = PTR_ERR(dir);
+
+ if (rc1 == -ENOENT)
+ GOTO(out, rc = 0);
+
+ dir = NULL;
+ goto next;
+ }
+ }
+
rc1 = lfsck_namespace_repair_dirent(env, com, dir, obj,
info->lti_tmpbuf2, lnr->lnr_name,
lnr->lnr_type, true, false);
if (repair_linkea) {
struct lustre_handle lh = { 0 };
- rc1 = linkea_data_new(&ldata, &info->lti_big_buf);
+ rc1 = linkea_links_new(&ldata, &info->lti_big_buf,
+ cname, lfsck_dto2fid(dir));
if (rc1 != 0)
goto next;
- rc1 = linkea_add_buf(&ldata, cname, lfsck_dto2fid(dir));
- if (rc1 != 0)
- goto next;
+ if (dir == NULL) {
+ dir = lfsck_assistant_object_load(env, lfsck,
+ lso);
+ if (IS_ERR(dir)) {
+ rc1 = PTR_ERR(dir);
+
+ if (rc1 == -ENOENT)
+ GOTO(out, rc = 0);
+
+ dir = NULL;
+ goto next;
+ }
+ }
rc1 = lfsck_ibits_lock(env, lfsck, obj, &lh,
MDS_INODELOCK_UPDATE |
"%d "DFID" of the striped directory "DFID" with "
"dangling %s/%s, rename %s/%s, llinkea %s/%s, "
"repair_lmvea %s/%s: rc = %d\n", lfsck_lfsck2name(lfsck),
- i, PFID(cfid), PFID(&lnr->lnr_fid),
+ i, PFID(cfid), PFID(pfid),
create ? "yes" : "no", create_repaired ? "yes" : "no",
rename ? "yes" : "no", rename_repaired ? "yes" : "no",
repair_linkea ? "yes" : "no",
repair_lmvea ? "yes" : "no",
lmvea_repaired ? "yes" : "no", rc1);
- if (obj != NULL && !IS_ERR(obj))
+ if (obj != NULL && !IS_ERR(obj)) {
lfsck_object_put(env, obj);
+ obj = NULL;
+ }
if (rc1 < 0) {
rc = rc1;
}
}
- RETURN(rc);
+ GOTO(out, rc);
+
+out:
+ if (obj != NULL && !IS_ERR(obj))
+ lfsck_object_put(env, obj);
+
+ if (dir != NULL && !IS_ERR(dir))
+ lfsck_object_put(env, dir);
+
+ return rc;
}
/**
*
* \param[in] env pointer to the thread context
* \param[in] com pointer to the lfsck component
- * \param[in] dir pointer to the master MDT-object of the
- * striped directory
* \param[in] lnr pointer to the namespace request that contains the
* shard's name, parent object, parent's LMV, and ect.
*
- * \retval zero for succeed
+ * \retval zero for success
* \retval negative error number on failure
*/
int lfsck_namespace_handle_striped_master(const struct lu_env *env,
struct lfsck_component *com,
- struct dt_object *dir,
struct lfsck_namespace_req *lnr)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_namespace *ns = com->lc_file_ram;
struct lfsck_lmv *llmv = lnr->lnr_lmv;
- const struct lu_fid *pfid = lfsck_dto2fid(dir);
+ struct lfsck_assistant_object *lso = lnr->lnr_lar.lar_parent;
+ const struct lu_fid *pfid = &lso->lso_fid;
+ struct dt_object *dir;
struct dt_object *obj = NULL;
struct dt_device *dev = NULL;
int shard_idx = 0;
if (unlikely(llmv->ll_ignore))
RETURN(0);
+ dir = lfsck_assistant_object_load(env, lfsck, lso);
+ if (IS_ERR(dir)) {
+ rc = PTR_ERR(dir);
+
+ RETURN(rc == -ENOENT ? 0 : rc);
+ }
+
shard_idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
if (shard_idx < 0)
GOTO(fail_lmv, rc = shard_idx);
GOTO(out, rc);
}
- ltd = LTD_TGT(&lfsck->li_mdt_descs, shard_idx);
+ ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, shard_idx);
if (unlikely(ltd == NULL)) {
CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
"did not join the namespace LFSCK\n",
dev = ltd->ltd_tgt;
}
- obj = lfsck_object_find_by_dev_nowait(env, dev, &lnr->lnr_fid);
+ obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
if (IS_ERR(obj)) {
if (lfsck_is_dead_obj(dir))
RETURN(0);
GOTO(out, rc = 0);
}
- rc = lfsck_read_stripe_lmv(env, obj, lmv);
+ rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
if (unlikely(rc == -ENOENT))
/* It may happen when the remote object has been removed,
* but the local MDT does not aware of that. */
CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
"the shard: "DFID", parent "DFID", name %.*s: rc = %d\n",
lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
- PFID(lfsck_dto2fid(dir)),
- lnr->lnr_namelen, lnr->lnr_name, rc);
+ PFID(pfid), lnr->lnr_namelen, lnr->lnr_name, rc);
if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
if (obj != NULL && !IS_ERR(obj))
lfsck_object_put(env, obj);
+ lfsck_object_put(env, dir);
+
return rc;
}