Whamcloud - gitweb
LU-4612 lvfs: correct call to pop_ctxt
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_compat.c
index 2403aa1..6ec1129 100644 (file)
@@ -62,19 +62,12 @@ static void osd_push_ctxt(const struct osd_device *dev,
                           struct lvfs_run_ctxt *newctxt,
                           struct lvfs_run_ctxt *save)
 {
-        OBD_SET_CTXT_MAGIC(newctxt);
-        newctxt->pwdmnt = dev->od_mnt;
-        newctxt->pwd = dev->od_mnt->mnt_root;
-        newctxt->fs = get_ds();
+       OBD_SET_CTXT_MAGIC(newctxt);
+       newctxt->pwdmnt = dev->od_mnt;
+       newctxt->pwd = dev->od_mnt->mnt_root;
+       newctxt->fs = get_ds();
 
-        push_ctxt(save, newctxt, NULL);
-}
-
-static void osd_pop_ctxt(const struct osd_device *dev,
-                        struct lvfs_run_ctxt *new,
-                        struct lvfs_run_ctxt *save)
-{
-       pop_ctxt(save, new, NULL);
+       push_ctxt(save, newctxt);
 }
 
 /* utility to make a directory */
@@ -112,7 +105,7 @@ static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
                GOTO(out_up, dchild);
        }
 
-       err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
+       err = vfs_mkdir(dir->d_inode, dchild, mode);
        if (err)
                GOTO(out_err, err);
 
@@ -131,7 +124,7 @@ int osd_last_rcvd_subdir_count(struct osd_device *osd)
         struct dentry        *dlast;
         loff_t                off;
         int                   rc = 0;
-       int                   count = FILTER_SUBDIR_COUNT;
+       int                   count = OBJ_SUBDIR_COUNT;
 
         ENTRY;
 
@@ -182,8 +175,6 @@ static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 
        omm = dev->od_mdt_map;
 
-       LASSERT(dev->od_fsops);
-
        parent = osd_sb(dev)->s_root;
        osd_push_ctxt(dev, &new, &save);
 
@@ -202,7 +193,7 @@ static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
        GOTO(cleanup, rc);
 
 cleanup:
-       pop_ctxt(&save, &new, NULL);
+       pop_ctxt(&save, &new);
        if (rc) {
                if (omm->omm_remote_parent != NULL)
                        dput(omm->omm_remote_parent);
@@ -259,10 +250,7 @@ int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
                                   NULL);
        CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
               name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
-       LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
-                osd_name(osd), parent->d_inode->i_ino,
-                parent->d_inode->i_nlink);
-       parent->d_inode->i_nlink++;
+       ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
        mark_inode_dirty(parent->d_inode);
        mutex_unlock(&parent->d_inode->i_mutex);
        RETURN(rc);
@@ -296,7 +284,8 @@ int osd_delete_from_remote_parent(const struct lu_env *env,
        dentry = osd_child_dentry_by_inode(env, parent->d_inode,
                                           name, strlen(name));
        mutex_lock(&parent->d_inode->i_mutex);
-       bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
+       bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
+                                   NULL, NULL);
        if (bh == NULL) {
                mutex_unlock(&parent->d_inode->i_mutex);
                RETURN(-ENOENT);
@@ -304,10 +293,7 @@ int osd_delete_from_remote_parent(const struct lu_env *env,
        CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
               name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
        rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
-       LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
-                osd_name(osd), parent->d_inode->i_ino,
-                parent->d_inode->i_nlink);
-       parent->d_inode->i_nlink--;
+       ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
        mark_inode_dirty(parent->d_inode);
        mutex_unlock(&parent->d_inode->i_mutex);
        brelse(bh);
@@ -339,7 +325,8 @@ int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
        dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
                                           name, strlen(name));
        mutex_lock(&parent->d_inode->i_mutex);
-       bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
+       bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
+                                   NULL, NULL);
        if (bh == NULL) {
                rc = -ENOENT;
        } else {
@@ -393,12 +380,23 @@ static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
        rwlock_init(&dev->od_ost_map->om_seq_list_lock);
        sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
 
-        LASSERT(dev->od_fsops);
         osd_push_ctxt(dev, &new, &save);
 
-        d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
+       d = ll_lookup_one_len("O", rootd, strlen("O"));
        if (IS_ERR(d))
                GOTO(cleanup, rc = PTR_ERR(d));
+       if (d->d_inode == NULL) {
+               dput(d);
+               /* The lookup() may be called again inside simple_mkdir().
+                * Since the repeated lookup() only be called for "/O" at
+                * mount time, it will not affect the whole performance. */
+               d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
+               if (IS_ERR(d))
+                       GOTO(cleanup, rc = PTR_ERR(d));
+
+               /* It is quite probably that the device is new formatted. */
+               dev->od_maybe_new = 1;
+       }
 
        inode = d->d_inode;
        ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
@@ -413,7 +411,7 @@ static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
        GOTO(cleanup, rc);
 
 cleanup:
-       osd_pop_ctxt(dev, &new, &save);
+       pop_ctxt(&save, &new);
         if (IS_ERR(d)) {
                 OBD_FREE_PTR(dev->od_ost_map);
                 RETURN(PTR_ERR(d));
@@ -522,10 +520,9 @@ static int osd_obj_update_entry(struct osd_thread_info *info,
                                struct dentry *dir, const char *name,
                                const struct lu_fid *fid,
                                const struct osd_inode_id *id,
-                               struct thandle *th)
+                               handle_t *th)
 {
        struct inode               *parent = dir->d_inode;
-       struct osd_thandle         *oh;
        struct dentry              *child;
        struct ldiskfs_dir_entry_2 *de;
        struct buffer_head         *bh;
@@ -537,9 +534,8 @@ static int osd_obj_update_entry(struct osd_thread_info *info,
        int                         rc;
        ENTRY;
 
-       oh = container_of(th, struct osd_thandle, ot_super);
-       LASSERT(oh->ot_handle != NULL);
-       LASSERT(oh->ot_handle->h_transaction != NULL);
+       LASSERT(th != NULL);
+       LASSERT(th->h_transaction != NULL);
 
        child = &info->oti_child_dentry;
        child->d_parent = dir;
@@ -549,7 +545,7 @@ static int osd_obj_update_entry(struct osd_thread_info *info,
 
        ll_vfs_dq_init(parent);
        mutex_lock(&parent->i_mutex);
-       bh = osd_ldiskfs_find_entry(parent, child, &de, NULL);
+       bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
        if (bh == NULL)
                GOTO(out, rc = -ENOENT);
 
@@ -600,12 +596,12 @@ update:
         * is still valid. Since it was referenced by an invalid entry,
         * making it as invisible temporary may be not worse. OI scrub
         * will process it later. */
-       rc = ldiskfs_journal_get_write_access(oh->ot_handle, bh);
+       rc = ldiskfs_journal_get_write_access(th, bh);
        if (rc != 0)
                GOTO(out, rc);
 
        de->inode = cpu_to_le32(id->oii_ino);
-       rc = ldiskfs_journal_dirty_metadata(oh->ot_handle, bh);
+       rc = ldiskfs_journal_dirty_metadata(th, bh);
 
        GOTO(out, rc);
 
@@ -618,19 +614,17 @@ out:
 static int osd_obj_del_entry(struct osd_thread_info *info,
                             struct osd_device *osd,
                             struct dentry *dird, char *name,
-                            struct thandle *th)
+                            handle_t *th)
 {
        struct ldiskfs_dir_entry_2 *de;
        struct buffer_head         *bh;
-       struct osd_thandle         *oh;
        struct dentry              *child;
        struct inode               *dir = dird->d_inode;
        int                         rc;
        ENTRY;
 
-       oh = container_of(th, struct osd_thandle, ot_super);
-       LASSERT(oh->ot_handle != NULL);
-       LASSERT(oh->ot_handle->h_transaction != NULL);
+       LASSERT(th != NULL);
+       LASSERT(th->h_transaction != NULL);
 
 
        child = &info->oti_child_dentry;
@@ -643,9 +637,9 @@ static int osd_obj_del_entry(struct osd_thread_info *info,
        ll_vfs_dq_init(dir);
        mutex_lock(&dir->i_mutex);
        rc = -ENOENT;
-       bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
+       bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
        if (bh) {
-               rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
+               rc = ldiskfs_delete_entry(th, dir, de, bh);
                brelse(bh);
        }
        mutex_unlock(&dir->i_mutex);
@@ -657,37 +651,38 @@ int osd_obj_add_entry(struct osd_thread_info *info,
                      struct osd_device *osd,
                      struct dentry *dir, char *name,
                      const struct osd_inode_id *id,
-                     struct thandle *th)
+                     handle_t *th)
 {
-        struct osd_thandle *oh;
-        struct dentry *child;
-        struct inode *inode;
-        int rc;
+       struct dentry *child;
+       struct inode *inode;
+       int rc;
 
-        ENTRY;
+       ENTRY;
+
+       if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
+               RETURN(0);
 
-        oh = container_of(th, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle != NULL);
-        LASSERT(oh->ot_handle->h_transaction != NULL);
+       LASSERT(th != NULL);
+       LASSERT(th->h_transaction != NULL);
 
-        inode = &info->oti_inode;
-        inode->i_sb = osd_sb(osd);
+       inode = &info->oti_inode;
+       inode->i_sb = osd_sb(osd);
        osd_id_to_inode(inode, id);
        inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 
-        child = &info->oti_child_dentry;
-        child->d_name.hash = 0;
-        child->d_name.name = name;
-        child->d_name.len = strlen(name);
-        child->d_parent = dir;
-        child->d_inode = inode;
+       child = &info->oti_child_dentry;
+       child->d_name.hash = 0;
+       child->d_name.name = name;
+       child->d_name.len = strlen(name);
+       child->d_parent = dir;
+       child->d_inode = inode;
 
        if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
                inode->i_ino++;
 
        ll_vfs_dq_init(dir->d_inode);
        mutex_lock(&dir->d_inode->i_mutex);
-       rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
+       rc = osd_ldiskfs_add_entry(th, child, inode, NULL);
        mutex_unlock(&dir->d_inode->i_mutex);
 
        RETURN(rc);
@@ -903,7 +898,7 @@ int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 
        dir = d_seq->d_inode;
        mutex_lock(&dir->i_mutex);
-       bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
+       bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
        mutex_unlock(&dir->i_mutex);
 
        if (bh == NULL)
@@ -924,7 +919,7 @@ int osd_obj_map_insert(struct osd_thread_info *info,
                       struct osd_device *osd,
                       const struct lu_fid *fid,
                       const struct osd_inode_id *id,
-                      struct thandle *th)
+                      handle_t *th)
 {
        struct osd_obj_map      *map;
        struct osd_obj_seq      *osd_seq;
@@ -967,7 +962,7 @@ again:
 }
 
 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
-                      const struct lu_fid *fid, struct thandle *th)
+                      const struct lu_fid *fid, handle_t *th)
 {
        struct osd_obj_map      *map;
        struct osd_obj_seq      *osd_seq;
@@ -1001,7 +996,7 @@ int osd_obj_map_update(struct osd_thread_info *info,
                       struct osd_device *osd,
                       const struct lu_fid *fid,
                       const struct osd_inode_id *id,
-                      struct thandle *th)
+                      handle_t *th)
 {
        struct osd_obj_seq      *osd_seq;
        struct dentry           *d;
@@ -1025,6 +1020,122 @@ int osd_obj_map_update(struct osd_thread_info *info,
        RETURN(rc);
 }
 
+int osd_obj_map_recover(struct osd_thread_info *info,
+                       struct osd_device *osd,
+                       struct inode *src_parent,
+                       struct dentry *src_child,
+                       const struct lu_fid *fid)
+{
+       struct osd_obj_seq         *osd_seq;
+       struct dentry              *tgt_parent;
+       struct dentry              *tgt_child = &info->oti_child_dentry;
+       struct inode               *dir;
+       struct inode               *inode     = src_child->d_inode;
+       struct ost_id              *ostid     = &info->oti_ostid;
+       handle_t                   *jh;
+       struct ldiskfs_dir_entry_2 *de;
+       struct buffer_head         *bh;
+       char                        name[32];
+       int                         dirn;
+       int                         rc        = 0;
+       ENTRY;
+
+       if (fid_is_last_id(fid)) {
+               osd_seq = osd_seq_load(info, osd, fid_seq(fid));
+               if (IS_ERR(osd_seq))
+                       RETURN(PTR_ERR(osd_seq));
+
+               tgt_parent = osd_seq->oos_root;
+               tgt_child->d_name.name = "LAST_ID";
+               tgt_child->d_name.len = strlen("LAST_ID");
+       } else {
+               fid_to_ostid(fid, ostid);
+               osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
+               if (IS_ERR(osd_seq))
+                       RETURN(PTR_ERR(osd_seq));
+
+               dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
+               tgt_parent = osd_seq->oos_dirs[dirn];
+               osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
+               tgt_child->d_name.name = name;
+               tgt_child->d_name.len = strlen(name);
+       }
+       LASSERT(tgt_parent != NULL);
+
+       dir = tgt_parent->d_inode;
+       tgt_child->d_name.hash = 0;
+       tgt_child->d_parent = tgt_parent;
+       tgt_child->d_inode = inode;
+
+       /* The non-initialized src_child may be destroyed. */
+       jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
+                               osd_dto_credits_noquota[DTO_INDEX_DELETE] +
+                               osd_dto_credits_noquota[DTO_INDEX_INSERT] +
+                               osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
+       if (IS_ERR(jh))
+               RETURN(PTR_ERR(jh));
+
+       ll_vfs_dq_init(src_parent);
+       ll_vfs_dq_init(dir);
+
+       mutex_lock(&src_parent->i_mutex);
+       mutex_lock(&dir->i_mutex);
+       bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
+       if (bh != NULL) {
+               /* XXX: If some other object occupied the same slot. And If such
+                *      inode is zero-sized and with SUID+SGID, then means it is
+                *      a new created one. Maybe we can remove it and insert the
+                *      original one back to the /O/<seq>/d<x>. But there are
+                *      something to be considered:
+                *
+                *      1) The OST-object under /lost+found has crashed LMA.
+                *         So it should not conflict with the current one.
+                *
+                *      2) There are race conditions that: someone may just want
+                *         to modify the current one. Even if the OI scrub takes
+                *         the object lock when remove the current one, it still
+                *         cause the modification to be lost becasue the target
+                *         has been removed when the RPC service thread waiting
+                *         for the lock.
+                *
+                *      So keep it there before we have suitable solution. */
+               brelse(bh);
+               mutex_unlock(&dir->i_mutex);
+               mutex_unlock(&src_parent->i_mutex);
+               ldiskfs_journal_stop(jh);
+
+               rc = -EEXIST;
+               /* If the src object has never been modified, then remove it. */
+               if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
+                   inode->i_mode & S_ISGID) {
+                       rc = vfs_unlink(src_parent, src_child);
+                       if (unlikely(rc == -ENOENT))
+                               rc = 0;
+               }
+               RETURN(rc);
+       }
+
+       bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
+                                   NULL, NULL);
+       if (unlikely(bh == NULL))
+               GOTO(unlock, rc = -ENOENT);
+
+       rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
+       brelse(bh);
+       if (rc != 0)
+               GOTO(unlock, rc);
+
+       rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
+
+       GOTO(unlock, rc);
+
+unlock:
+       mutex_unlock(&dir->i_mutex);
+       mutex_unlock(&src_parent->i_mutex);
+       ldiskfs_journal_stop(jh);
+       return rc;
+}
+
 static struct dentry *
 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
                     const struct lu_fid *fid, char **name)
@@ -1054,7 +1165,7 @@ osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
 
 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
                        const struct lu_fid *fid, const struct osd_inode_id *id,
-                       struct thandle *th)
+                       handle_t *th)
 {
        struct dentry   *root;
        char            *name;
@@ -1075,7 +1186,7 @@ int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
 
 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
                        const struct lu_fid *fid, const struct osd_inode_id *id,
-                       struct thandle *th)
+                       handle_t *th)
 {
        struct dentry   *root;
        char            *name;