Whamcloud - gitweb
LU-6824 ldiskfs: give warning with dir htree growing
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_oi.c
index 54a016f..ae3d5c1 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -40,7 +40,7 @@
  * Author: Nikita Danilov <nikita@clusterfs.com>
  */
 
-#define DEBUG_SUBSYSTEM S_MDS
+#define DEBUG_SUBSYSTEM S_OSD
 
 #include <linux/module.h>
 
@@ -139,6 +139,9 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
                return PTR_ERR(inode);
        }
 
+       ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
+       unlock_new_inode(inode);
+
        if (feat->dif_flags & DT_IND_VARKEY)
                rc = iam_lvar_create(inode, feat->dif_keysize_max,
                                     feat->dif_ptrsize, feat->dif_recsize_max,
@@ -148,7 +151,7 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
                                     feat->dif_ptrsize, feat->dif_recsize_max,
                                     jh);
        dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
-       rc = osd_ldiskfs_add_entry(jh, dentry, inode, NULL);
+       rc = osd_ldiskfs_add_entry(info, jh, dentry, inode, NULL);
        ldiskfs_journal_stop(jh);
        iput(inode);
        return rc;
@@ -228,7 +231,6 @@ static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
         if (IS_ERR(inode))
                 RETURN(PTR_ERR(inode));
 
-       ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
        /* 'What the @fid is' is not imporatant, because these objects
         * have no OI mappings, and only are visible inside the OSD.*/
        lu_igif_build(&info->oti_fid, inode->i_ino, inode->i_generation);
@@ -336,7 +338,54 @@ osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
        RETURN(count);
 }
 
-int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
+static int osd_remove_oi_one(struct dentry *parent, const char *name,
+                            int namelen)
+{
+       struct dentry *child;
+       int rc;
+
+       child = ll_lookup_one_len(name, parent, namelen);
+       if (IS_ERR(child)) {
+               rc = PTR_ERR(child);
+       } else {
+               rc = ll_vfs_unlink(parent->d_inode, child);
+               dput(child);
+       }
+
+       return rc == -ENOENT ? 0 : rc;
+}
+
+static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
+{
+       char name[16];
+       int namelen;
+       int rc;
+       int i;
+
+       for (i = 0; i < osd->od_scrub.os_file.sf_oi_count; i++) {
+               namelen = snprintf(name, sizeof(name), "%s.%d",
+                                  OSD_OI_NAME_BASE, i);
+               rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
+               if (rc != 0) {
+                       CERROR("%.16s: fail to remove the stale OI file %s: "
+                              "rc = %d\n",
+                              LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
+                              name, rc);
+                       return rc;
+               }
+       }
+
+       namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
+       rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
+       if (rc != 0)
+               CERROR("%.16s: fail to remove the stale OI file %s: rc = %d\n",
+                      LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name, name, rc);
+
+       return rc;
+}
+
+int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
+               bool restored)
 {
        struct osd_scrub  *scrub = &osd->od_scrub;
        struct scrub_file *sf = &scrub->os_file;
@@ -344,6 +393,12 @@ int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
        int                rc;
        ENTRY;
 
+       if (restored) {
+               rc = osd_remove_ois(info, osd);
+               if (rc != 0)
+                       return rc;
+       }
+
        OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
        if (oi == NULL)
                RETURN(-ENOMEM);
@@ -368,6 +423,7 @@ int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
        /* if previous failed then try found single OI from old filesystem */
        rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
        if (rc == 0) { /* found single OI from old filesystem */
+               ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
                if (sf->sf_success_count == 0)
                        /* XXX: There is one corner case that if the OI_scrub
                         *      file crashed or lost and we regard it upgrade,
@@ -511,11 +567,15 @@ int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
        if (!(flags & OI_CHECK_FLD))
                RETURN(0);
 
+       if (osd_seq_site(osd)->ss_server_fld == NULL)
+               RETURN(0);
+
        rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
        if (rc != 0) {
-               CERROR("%s: "DFID" lookup failed: rc = %d\n", osd_name(osd),
-                      PFID(fid), rc);
-               RETURN(rc);
+               if (rc != -ENOENT)
+                       CERROR("%s: lookup FLD "DFID": rc = %d\n",
+                              osd_name(osd), PFID(fid), rc);
+               RETURN(0);
        }
 
        if (fld_range_is_ost(range))
@@ -552,14 +612,23 @@ int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
        if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
                return osd_obj_map_lookup(info, osd, fid, id);
 
-       if (fid_is_fs_root(fid)) {
-               osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
-                          osd_sb(osd)->s_root->d_inode->i_generation);
-               return 0;
-       }
 
-       if (unlikely(fid_is_acct(fid)))
-               return osd_acct_obj_lookup(info, osd, fid, id);
+       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
+               int rc;
+               if (fid_is_fs_root(fid)) {
+                       osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
+                                  osd_sb(osd)->s_root->d_inode->i_generation);
+                       return 0;
+               }
+               if (unlikely(fid_is_acct(fid)))
+                       return osd_acct_obj_lookup(info, osd, fid, id);
+
+               /* For other special FIDs, try OI first, then do spec lookup */
+               rc = __osd_oi_lookup(info, osd, fid, id);
+               if (rc == -ENOENT)
+                       return osd_obj_spec_lookup(info, osd, fid, id);
+               return rc;
+       }
 
        if (!osd->od_igif_inoi && fid_is_igif(fid)) {
                osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
@@ -631,7 +700,7 @@ int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
                        return rc;
 
                if (unlikely(osd_id_eq(id, oi_id)))
-                       return 0;
+                       return 1;
 
                /* Check whether the mapping for oi_id is valid or not. */
                inode = osd_iget(info, osd, oi_id);
@@ -642,6 +711,13 @@ int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
                        return rc;
                }
 
+               /* The EA inode should NOT be in OI, old OI scrub may added
+                * such OI mapping by wrong, replace it. */
+               if (unlikely(osd_is_ea_inode(inode))) {
+                       iput(inode);
+                       goto update;
+               }
+
                rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
                iput(inode);
                if (rc == -ENODATA)