LU-14119 osd: delete stale OI mapping entry

Once LMA check shows OI mapping entry is stale, delete it from
OI table, as can avoid removing whole OI files.

Don't add OI mapping into cache until osd_fid_lookup(), because
the mapping in OI is not trustable until FID in LMA is checked,
otherwise it may mislead LFSCK.

Signed-off-by: Lai Siyao <lai.siyao@whamcloud.com>
Change-Id: I4b50dcc02149d485e4bf4a361ca2994daa280feb
Reviewed-on: https://review.whamcloud.com/41741
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Hongchao Zhang <hongchao@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c
index 972f2b9..00e3ebd 100644 (file)
--- a/lustre/osd-ldiskfs/osd_handler.c
+++ b/lustre/osd-ldiskfs/osd_handler.c
@@ -933,7 +933,6 @@ struct osd_check_lmv_buf {
        struct dir_context ctx;
        struct osd_thread_info *oclb_info;
        struct osd_device *oclb_dev;
        struct dir_context ctx;
        struct osd_thread_info *oclb_info;
        struct osd_device *oclb_dev;

-       struct osd_idmap_cache *oclb_oic;

        int oclb_items;
        bool oclb_found;
 };
        int oclb_items;
        bool oclb_found;
 };


@@ -959,7 +958,6 @@ static int osd_stripe_dir_filldir(void *buf,
        struct lu_fid *fid = &oti->oti_fid3;
        struct osd_inode_id *id = &oti->oti_id3;
        struct osd_device *dev = oclb->oclb_dev;
        struct lu_fid *fid = &oti->oti_fid3;
        struct osd_inode_id *id = &oti->oti_id3;
        struct osd_device *dev = oclb->oclb_dev;

-       struct osd_idmap_cache *oic = oclb->oclb_oic;

        struct inode *inode;
 
        oclb->oclb_items++;
        struct inode *inode;
 
        oclb->oclb_items++;


@@ -982,10 +980,7 @@ static int osd_stripe_dir_filldir(void *buf,
 
        iput(inode);
        osd_add_oi_cache(oti, dev, id, fid);
 
        iput(inode);
        osd_add_oi_cache(oti, dev, id, fid);

-       oic->oic_fid = *fid;
-       oic->oic_lid = *id;
-       oic->oic_dev = dev;
-       osd_oii_insert(dev, oic, true);
+       osd_oii_insert(dev, fid, id, true);

        oclb->oclb_found = true;
 
        return 1;
        oclb->oclb_found = true;
 
        return 1;


@@ -1028,7 +1023,7 @@ static int osd_stripe_dir_filldir(void *buf,
  *    the correct OI mapping for the slave MDT-object.
  */
 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
  *    the correct OI mapping for the slave MDT-object.
  */
 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,

-                        struct inode *inode, struct osd_idmap_cache *oic)
+                        struct inode *inode)

 {
        struct lu_buf *buf = &oti->oti_big_buf;
        struct dentry *dentry = &oti->oti_obj_dentry;
 {
        struct lu_buf *buf = &oti->oti_big_buf;
        struct dentry *dentry = &oti->oti_obj_dentry;


@@ -1038,7 +1033,6 @@ static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
                .ctx.actor = osd_stripe_dir_filldir,
                .oclb_info = oti,
                .oclb_dev = dev,
                .ctx.actor = osd_stripe_dir_filldir,
                .oclb_info = oti,
                .oclb_dev = dev,

-               .oclb_oic = oic,

                .oclb_found = false,
        };
        int rc = 0;
                .oclb_found = false,
        };
        int rc = 0;


@@ -1092,9 +1086,9 @@ again:
 out:
        if (rc < 0)
                CDEBUG(D_LFSCK,
 out:
        if (rc < 0)
                CDEBUG(D_LFSCK,

-                      "%s: cannot check LMV, ino = %lu/%u "DFID": rc = %d\n",
+                      "%s: cannot check LMV, ino = %lu/%u: rc = %d\n",

                       osd_ino2name(inode), inode->i_ino, inode->i_generation,
                       osd_ino2name(inode), inode->i_ino, inode->i_generation,

-                      PFID(&oic->oic_fid), rc);
+                      rc);

        else
                rc = 0;
 
        else
                rc = 0;
 


@@ -1252,16 +1246,15 @@ trigger:
                if (scrub->os_partial_scan && !scrub->os_in_join)
                        goto join;
 
                if (scrub->os_partial_scan && !scrub->os_in_join)
                        goto join;
 

-               osd_add_oi_cache(info, dev, id, fid);

                if (IS_ERR_OR_NULL(inode) || result) {
                if (IS_ERR_OR_NULL(inode) || result) {

-                       osd_oii_insert(dev, oic, result == -ENOENT);
+                       osd_oii_insert(dev, fid, id, result == -ENOENT);

                        GOTO(out, result = -EINPROGRESS);
                }
 
                LASSERT(remote);
                LASSERT(obj->oo_inode == inode);
 
                        GOTO(out, result = -EINPROGRESS);
                }
 
                LASSERT(remote);
                LASSERT(obj->oo_inode == inode);
 

-               osd_oii_insert(dev, oic, true);
+               osd_oii_insert(dev, fid, id, true);

                goto found;
        }
 
                goto found;
        }
 


@@ -1283,16 +1276,15 @@ join:
        if (rc1 && rc1 != -EALREADY)
                GOTO(out, result = -EREMCHG);
 
        if (rc1 && rc1 != -EALREADY)
                GOTO(out, result = -EREMCHG);
 

-       osd_add_oi_cache(info, dev, id, fid);

        if (IS_ERR_OR_NULL(inode) || result) {
        if (IS_ERR_OR_NULL(inode) || result) {

-               osd_oii_insert(dev, oic, result == -ENOENT);
+               osd_oii_insert(dev, fid, id, result == -ENOENT);

                GOTO(out, result = -EINPROGRESS);
        }
 
        LASSERT(remote);
        LASSERT(obj->oo_inode == inode);
 
                GOTO(out, result = -EINPROGRESS);
        }
 
        LASSERT(remote);
        LASSERT(obj->oo_inode == inode);
 

-       osd_oii_insert(dev, oic, true);
+       osd_oii_insert(dev, fid, id, true);

        goto found;
 
 check_lma:
        goto found;
 
 check_lma:


@@ -1389,6 +1381,8 @@ check_lma:
 
        if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
                result = -EREMCHG;
 
        if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
                result = -EREMCHG;

+               osd_scrub_refresh_mapping(info, dev, fid, id, DTO_INDEX_DELETE,
+                                         true, 0, NULL);

                goto trigger;
        }
 
                goto trigger;
        }
 


@@ -1423,7 +1417,7 @@ found:
 
        if (S_ISDIR(inode->i_mode) &&
            (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
 
        if (S_ISDIR(inode->i_mode) &&
            (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))

-               osd_check_lmv(info, dev, inode, oic);
+               osd_check_lmv(info, dev, inode);

 
        result = osd_attach_jinode(inode);
        if (result)
 
        result = osd_attach_jinode(inode);
        if (result)


@@ -5770,11 +5764,9 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
 
 static int
 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
 
 static int
 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,

-                     struct osd_idmap_cache *oic)
+                     const struct lu_fid *fid, struct osd_inode_id *id)

 {
        struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 {
        struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;

-       struct lu_fid *fid = &oic->oic_fid;
-       struct osd_inode_id *id = &oic->oic_lid;

        struct inode *inode = NULL;
        int once = 0;
        bool insert;
        struct inode *inode = NULL;
        int once = 0;
        bool insert;


@@ -5837,7 +5829,7 @@ trigger:
                        }
                }
 
                        }
                }
 

-               rc = osd_oii_insert(dev, oic, insert);
+               rc = osd_oii_insert(dev, fid, id, insert);

                /*
                 * There is race condition between osd_oi_lookup and OI scrub.
                 * The OI scrub finished just after osd_oi_lookup() failure.
                /*
                 * There is race condition between osd_oi_lookup and OI scrub.
                 * The OI scrub finished just after osd_oi_lookup() failure.


@@ -5850,7 +5842,7 @@ trigger:
                if (!S_ISDIR(inode->i_mode))
                        rc = 0;
                else
                if (!S_ISDIR(inode->i_mode))
                        rc = 0;
                else

-                       rc = osd_check_lmv(oti, dev, inode, oic);
+                       rc = osd_check_lmv(oti, dev, inode);

 
                GOTO(out, rc);
        }
 
                GOTO(out, rc);
        }


@@ -5875,10 +5867,10 @@ out:
 
 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
                               struct osd_device *dev,
 
 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
                               struct osd_device *dev,

-                              struct osd_idmap_cache *oic,

                               struct lu_fid *fid, __u32 ino)
 {
        struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
                               struct lu_fid *fid, __u32 ino)
 {
        struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;

+       struct osd_idmap_cache *oic = &oti->oti_cache;

        struct inode *inode;
        int rc;
 
        struct inode *inode;
        int rc;
 


@@ -6069,13 +6061,12 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
        if (!IS_ERR(bh)) {
                struct osd_thread_info *oti = osd_oti_get(env);
                struct osd_inode_id *id = &oti->oti_id;
        if (!IS_ERR(bh)) {
                struct osd_thread_info *oti = osd_oti_get(env);
                struct osd_inode_id *id = &oti->oti_id;

-               struct osd_idmap_cache *oic = &oti->oti_cache;

                struct osd_device *dev = osd_obj2dev(obj);
 
                ino = le32_to_cpu(de->inode);
                if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
                        brelse(bh);
                struct osd_device *dev = osd_obj2dev(obj);
 
                ino = le32_to_cpu(de->inode);
                if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
                        brelse(bh);

-                       rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
+                       rc = osd_fail_fid_lookup(oti, dev, fid, ino);

                        GOTO(out, rc);
                }
 
                        GOTO(out, rc);
                }
 


@@ -6103,19 +6094,24 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
                        osd_id_gen(id, ino, OSD_OII_NOGEN);
                }
 
                        osd_id_gen(id, ino, OSD_OII_NOGEN);
                }
 

-               if (rc != 0 || osd_remote_fid(env, dev, fid)) {
-                       fid_zero(&oic->oic_fid);
-
+               if (rc != 0 || osd_remote_fid(env, dev, fid))

                        GOTO(out, rc);
                        GOTO(out, rc);

-               }

 
 

-               osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
-               rc = osd_consistency_check(oti, dev, oic);
-               if (rc == -ENOENT)
-                       fid_zero(&oic->oic_fid);
-               else
+               rc = osd_consistency_check(oti, dev, fid, id);
+               if (rc != -ENOENT) {

                        /* Other error should not affect lookup result. */
                        rc = 0;
                        /* Other error should not affect lookup result. */
                        rc = 0;

+
+                       /* Normal file mapping should be added into OI cache
+                        * after FID in LMA check, but for local files like
+                        * hsm_actions, their FIDs are not stored in OI files,
+                        * see osd_initial_OI_scrub(), and here is the only
+                        * place to load mapping into OI cache.
+                        */
+                       if (!fid_is_namespace_visible(fid))
+                               osd_add_oi_cache(osd_oti_get(env),
+                                                osd_obj2dev(obj), id, fid);
+               }

        } else {
                rc = PTR_ERR(bh);
        }
        } else {
                rc = PTR_ERR(bh);
        }

diff --git a/lustre/osd-ldiskfs/osd_internal.h b/lustre/osd-ldiskfs/osd_internal.h
index b6e5a1e..f9db66f 100644 (file)
--- a/lustre/osd-ldiskfs/osd_internal.h
+++ b/lustre/osd-ldiskfs/osd_internal.h
@@ -800,8 +800,8 @@ int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
 void osd_scrub_stop(struct osd_device *dev);
 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
 void osd_scrub_stop(struct osd_device *dev);
 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);

-int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
-                  int insert);
+int osd_oii_insert(struct osd_device *dev, const struct lu_fid *fid,
+                  struct osd_inode_id *id, int insert);

 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
                   struct osd_inode_id *id);
 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
                   struct osd_inode_id *id);
 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);


@@ -828,6 +828,13 @@ int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 
 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
                       __u64 seq);
 
 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
                       __u64 seq);

+int osd_scrub_refresh_mapping(struct osd_thread_info *info,
+                             struct osd_device *dev,
+                             const struct lu_fid *fid,
+                             const struct osd_inode_id *id,
+                             int ops, bool force,
+                             enum oi_check_flags flags, bool *exist);
+

 /* osd_quota_fmt.c */
 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
                       int type, uint blk, int depth, uint index,
 /* osd_quota_fmt.c */
 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
                       int type, uint blk, int depth, uint index,

diff --git a/lustre/osd-ldiskfs/osd_scrub.c b/lustre/osd-ldiskfs/osd_scrub.c
index 0352876..ecc7631 100644 (file)
--- a/lustre/osd-ldiskfs/osd_scrub.c
+++ b/lustre/osd-ldiskfs/osd_scrub.c
@@ -64,12 +64,12 @@ static inline int osd_scrub_has_window(struct lustre_scrub *scrub,
  * \retval   0, changed successfully
  * \retval -ve, on error
  */
  * \retval   0, changed successfully
  * \retval -ve, on error
  */

-static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
-                                    struct osd_device *dev,
-                                    const struct lu_fid *fid,
-                                    const struct osd_inode_id *id,
-                                    int ops, bool force,
-                                    enum oi_check_flags flags, bool *exist)
+int osd_scrub_refresh_mapping(struct osd_thread_info *info,
+                             struct osd_device *dev,
+                             const struct lu_fid *fid,
+                             const struct osd_inode_id *id,
+                             int ops, bool force,
+                             enum oi_check_flags flags, bool *exist)

 {
        handle_t *th;
        int       rc;
 {
        handle_t *th;
        int       rc;


@@ -3019,8 +3019,8 @@ const struct dt_index_operations osd_otable_ops = {
 
 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
 
 
 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
 

-int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
-                  int insert)
+int osd_oii_insert(struct osd_device *dev, const struct lu_fid *fid,
+                  struct osd_inode_id *id, int insert)

 {
        struct osd_inconsistent_item *oii;
        struct osd_scrub *oscrub = &dev->od_scrub;
 {
        struct osd_inconsistent_item *oii;
        struct osd_scrub *oscrub = &dev->od_scrub;


@@ -3033,7 +3033,9 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
                RETURN(-ENOMEM);
 
        INIT_LIST_HEAD(&oii->oii_list);
                RETURN(-ENOMEM);
 
        INIT_LIST_HEAD(&oii->oii_list);

-       oii->oii_cache = *oic;
+       oii->oii_cache.oic_fid = *fid;
+       oii->oii_cache.oic_lid = *id;
+       oii->oii_cache.oic_dev = dev;

        oii->oii_insert = insert;
 
        spin_lock(&lscrub->os_lock);
        oii->oii_insert = insert;
 
        spin_lock(&lscrub->os_lock);

diff --git a/lustre/osd-zfs/osd_internal.h b/lustre/osd-zfs/osd_internal.h
index cd50e1d..f5754c1 100644 (file)
--- a/lustre/osd-zfs/osd_internal.h
+++ b/lustre/osd-zfs/osd_internal.h
@@ -700,6 +700,23 @@ int osd_oii_insert(const struct lu_env *env, struct osd_device *dev,
 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
                   uint64_t *oid);
 
 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
                   uint64_t *oid);
 

+/**
+ * Basic transaction credit op
+ */
+enum dt_txn_op {
+       DTO_INDEX_INSERT,
+       DTO_INDEX_DELETE,
+       DTO_INDEX_UPDATE,
+       DTO_NR
+};
+
+int osd_scrub_refresh_mapping(const struct lu_env *env,
+                             struct osd_device *dev,
+                             const struct lu_fid *fid,
+                             uint64_t oid, enum dt_txn_op ops,
+                             bool force, const char *name);
+
+

 /* osd_xattr.c */
 int __osd_sa_xattr_schedule_update(const struct lu_env *env,
                                   struct osd_object *obj,
 /* osd_xattr.c */
 int __osd_sa_xattr_schedule_update(const struct lu_env *env,
                                   struct osd_object *obj,

diff --git a/lustre/osd-zfs/osd_object.c b/lustre/osd-zfs/osd_object.c
index 9817cc5..d1e4374 100644 (file)
--- a/lustre/osd-zfs/osd_object.c
+++ b/lustre/osd-zfs/osd_object.c
@@ -616,9 +616,12 @@ zget:
                GOTO(out, rc = 0);
 
        rc = osd_check_lma(env, obj);
                GOTO(out, rc = 0);
 
        rc = osd_check_lma(env, obj);

-       if ((!rc && !remote) || (rc != -EREMCHG))
+       if (rc != -EREMCHG)

                GOTO(out, rc);
 
                GOTO(out, rc);
 

+       osd_scrub_refresh_mapping(env, osd, fid, oid, DTO_INDEX_DELETE, true,
+                                 NULL);
+

 trigger:
        /* We still have chance to get the valid dnode: for the object that is
         * referenced by remote name entry, the object on the local MDT will be
 trigger:
        /* We still have chance to get the valid dnode: for the object that is
         * referenced by remote name entry, the object on the local MDT will be

diff --git a/lustre/osd-zfs/osd_scrub.c b/lustre/osd-zfs/osd_scrub.c
index 1f5e903..8579ee0 100644 (file)
--- a/lustre/osd-zfs/osd_scrub.c
+++ b/lustre/osd-zfs/osd_scrub.c
@@ -55,10 +55,6 @@
 #define OSD_OTABLE_MAX_HASH            ((1ULL << 48) - 1)
 #define OTABLE_PREFETCH                        256
 
 #define OSD_OTABLE_MAX_HASH            ((1ULL << 48) - 1)
 #define OTABLE_PREFETCH                        256
 

-#define DTO_INDEX_INSERT               1
-#define DTO_INDEX_DELETE               2
-#define DTO_INDEX_UPDATE               3
-

 static inline bool osd_scrub_has_window(struct osd_otable_it *it)
 {
        return it->ooi_prefetched < OTABLE_PREFETCH;
 static inline bool osd_scrub_has_window(struct osd_otable_it *it)
 {
        return it->ooi_prefetched < OTABLE_PREFETCH;


@@ -71,11 +67,11 @@ static inline bool osd_scrub_has_window(struct osd_otable_it *it)
  * \retval   0, changed successfully
  * \retval -ve, on error
  */
  * \retval   0, changed successfully
  * \retval -ve, on error
  */

-static int osd_scrub_refresh_mapping(const struct lu_env *env,
-                                    struct osd_device *dev,
-                                    const struct lu_fid *fid,
-                                    uint64_t oid, int ops,
-                                    bool force, const char *name)
+int osd_scrub_refresh_mapping(const struct lu_env *env,
+                             struct osd_device *dev,
+                             const struct lu_fid *fid,
+                             uint64_t oid, enum dt_txn_op ops,
+                             bool force, const char *name)

 {
        struct osd_thread_info *info = osd_oti_get(env);
        struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
 {
        struct osd_thread_info *info = osd_oti_get(env);
        struct zpl_direntry *zde = &info->oti_zde.lzd_reg;