X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flfsck%2Flfsck_lib.c;h=6ea74b8764e83fc2649aa4bb66f99534fa67fb3c;hp=de7b849644b40398f39093ccbae7c4e62311a68b;hb=76f0977b7ea5d46836cb459deb7b9ad9e781d585;hpb=32a11dbd310b44d2a6e084c48310f21ef6649145

diff --git a/lustre/lfsck/lfsck_lib.c b/lustre/lfsck/lfsck_lib.c
index de7b849..6ea74b8 100644
--- a/lustre/lfsck/lfsck_lib.c
+++ b/lustre/lfsck/lfsck_lib.c
@@ -322,6 +322,468 @@ void lfsck_component_cleanup(const struct lu_env *env,
 	lfsck_component_put(env, com);
 }
 
+int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
+		    struct lu_fid *fid, bool locked)
+{
+	struct lfsck_bookmark	*bk = &lfsck->li_bookmark_ram;
+	int			 rc = 0;
+	ENTRY;
+
+	if (!locked)
+		mutex_lock(&lfsck->li_mutex);
+
+	rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
+	if (rc >= 0) {
+		bk->lb_last_fid = *fid;
+		/* We do not care about whether the subsequent sub-operations
+		 * failed or not. The worst case is that one FID is lost that
+		 * is not a big issue for the LFSCK since it is relative rare
+		 * for LFSCK create. */
+		rc = lfsck_bookmark_store(env, lfsck);
+	}
+
+	if (!locked)
+		mutex_unlock(&lfsck->li_mutex);
+
+	RETURN(rc);
+}
+
+static const char dot[] = ".";
+static const char dotdot[] = "..";
+
+static int lfsck_create_lpf_local(const struct lu_env *env,
+				  struct lfsck_instance *lfsck,
+				  struct dt_object *parent,
+				  struct dt_object *child,
+				  struct lu_attr *la,
+				  struct dt_object_format *dof,
+				  const char *name)
+{
+	struct dt_device	*dev	= lfsck->li_bottom;
+	struct lfsck_bookmark	*bk	= &lfsck->li_bookmark_ram;
+	struct dt_object	*bk_obj = lfsck->li_bookmark_obj;
+	const struct lu_fid	*cfid	= lu_object_fid(&child->do_lu);
+	struct thandle		*th	= NULL;
+	loff_t			 pos	= 0;
+	int			 len	= sizeof(struct lfsck_bookmark);
+	int			 rc	= 0;
+	ENTRY;
+
+	th = dt_trans_create(env, dev);
+	if (IS_ERR(th))
+		RETURN(PTR_ERR(th));
+
+	/* 1a. create child */
+	rc = dt_declare_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 2a. increase child nlink */
+	rc = dt_declare_ref_add(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 3a. insert name into parent dir */
+	rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid,
+			       (const struct dt_key *)name, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 4a. increase parent nlink */
+	rc = dt_declare_ref_add(env, parent, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 5a. update bookmark */
+	rc = dt_declare_record_write(env, bk_obj, len, 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, child, 0);
+	/* 1b.1 create child */
+	rc = dt_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	if (unlikely(!dt_try_as_dir(env, child)))
+		GOTO(unlock, rc = -ENOTDIR);
+
+	/* 1b.2 insert dot into child dir */
+	rc = dt_insert(env, child, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 1b.3 insert dotdot into child dir */
+	rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID,
+		       (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 2b. increase child nlink */
+	rc = dt_ref_add(env, child, th);
+	dt_write_unlock(env, child);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 3b. insert name into parent dir */
+	rc = dt_insert(env, parent, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	dt_write_lock(env, parent, 0);
+	/* 4b. increase parent nlink */
+	rc = dt_ref_add(env, parent, th);
+	dt_write_unlock(env, parent);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	bk->lb_lpf_fid = *cfid;
+	lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
+
+	/* 5b. update bookmark */
+	rc = dt_record_write(env, bk_obj,
+			     lfsck_buf_get(env, bk, len), &pos, th);
+
+	GOTO(stop, rc);
+
+unlock:
+	dt_write_unlock(env, child);
+
+stop:
+	dt_trans_stop(env, dev, th);
+
+	return rc;
+}
+
+static int lfsck_create_lpf_remote(const struct lu_env *env,
+				   struct lfsck_instance *lfsck,
+				   struct dt_object *parent,
+				   struct dt_object *child,
+				   struct lu_attr *la,
+				   struct dt_object_format *dof,
+				   const char *name)
+{
+	struct lfsck_bookmark	*bk	= &lfsck->li_bookmark_ram;
+	struct dt_object	*bk_obj = lfsck->li_bookmark_obj;
+	const struct lu_fid	*cfid	= lu_object_fid(&child->do_lu);
+	struct thandle		*th	= NULL;
+	struct dt_device	*dev;
+	loff_t			 pos	= 0;
+	int			 len	= sizeof(struct lfsck_bookmark);
+	int			 rc	= 0;
+	ENTRY;
+
+	/* Create .lustre/lost+found/MDTxxxx. */
+
+	/* XXX: Currently, cross-MDT create operation needs to create the child
+	 *	object firstly, then insert name into the parent directory. For
+	 *	this case, the child object resides on current MDT (local), but
+	 *	the parent ".lustre/lost+found" may be on remote MDT. It is not
+	 *	easy to contain all the sub-modifications orderly within single
+	 *	transaction.
+	 *
+	 *	To avoid more inconsistency, we split the create operation into
+	 *	two transactions:
+	 *
+	 *	1) create the child locally.
+	 *	2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
+	 *	   remotely and update the lfsck_bookmark::lb_lpf_fid locally.
+	 *
+	 *	If 1) done but 2) failed, then the worst case is that we lose
+	 *	one object locally, which is not a big issue. (can be repaird
+	 *	by LFSCK phase III) */
+
+	/* Transaction I: */
+
+	dev = lfsck->li_bottom;
+	th = dt_trans_create(env, dev);
+	if (IS_ERR(th))
+		RETURN(PTR_ERR(th));
+
+	/* 1a. create child locally. */
+	rc = dt_declare_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 2a. increase child nlink locally. */
+	rc = dt_declare_ref_add(env, child, 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, child, 0);
+	/* 1b. create child locally. */
+	rc = dt_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	if (unlikely(!dt_try_as_dir(env, child)))
+		GOTO(unlock, rc = -ENOTDIR);
+
+	/* 2b.1 insert dot into child dir locally. */
+	rc = dt_insert(env, child, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 2b.2 insert dotdot into child dir locally. */
+	rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID,
+		       (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 2b.3 increase child nlink locally. */
+	rc = dt_ref_add(env, child, th);
+	dt_write_unlock(env, child);
+	dt_trans_stop(env, dev, th);
+	if (rc != 0)
+		RETURN(rc);
+
+	/* Transaction II: */
+
+	dev = lfsck->li_next;
+	th = dt_trans_create(env, dev);
+	if (IS_ERR(th))
+		RETURN(PTR_ERR(th));
+
+	/* 3a. insert name into parent dir remotely. */
+	rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid,
+			       (const struct dt_key *)name, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 4a. increase parent nlink remotely. */
+	rc = dt_declare_ref_add(env, parent, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 5a. decrease child nlink for dotdot locally if former remote
+	 *     update failed. */
+	rc = dt_declare_ref_del(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 6a. decrease child nlink for dot locally if former remote
+	 *     update failed. */
+	rc = dt_declare_ref_del(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 7a. destroy child locally if former remote update failed. */
+	rc = dt_declare_destroy(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 8a. update bookmark locally. */
+	rc = dt_declare_record_write(env, bk_obj, len, 0, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	rc = dt_trans_start(env, dev, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 3b. insert name into parent dir remotely. */
+	rc = dt_insert(env, parent, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+	if (rc == 0) {
+		dt_write_lock(env, parent, 0);
+		/* 4b. increase parent nlink remotely. */
+		rc = dt_ref_add(env, parent, th);
+		dt_write_unlock(env, parent);
+	}
+	if (rc != 0) {
+		/* 5b. decrease child nlink for dotdot locally. */
+		dt_ref_del(env, child, th);
+		/* 6b. decrease child nlink for dot locally. */
+		dt_ref_del(env, child, th);
+		/* 7b. destroy child locally. */
+		dt_destroy(env, child, th);
+		GOTO(stop, rc);
+	}
+
+	bk->lb_lpf_fid = *cfid;
+	lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
+
+	/* 8b. update bookmark locally. */
+	rc = dt_record_write(env, bk_obj,
+			     lfsck_buf_get(env, bk, len), &pos, th);
+
+	GOTO(stop, rc);
+
+unlock:
+	dt_write_unlock(env, child);
+stop:
+	dt_trans_stop(env, dev, th);
+
+	return rc;
+}
+
+/* Do NOT create .lustre/lost+found/MDTxxxx when register the lfsck instance,
+ * because the MDT0 maybe not reaady for sequence allocation yet. We do that
+ * only when it is required, such as orphan OST-objects repairing. */
+int lfsck_create_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
+{
+	struct lfsck_bookmark	 *bk	= &lfsck->li_bookmark_ram;
+	struct lfsck_thread_info *info	= lfsck_env_info(env);
+	struct lu_fid		 *cfid	= &info->lti_fid2;
+	struct lu_attr		 *la	= &info->lti_la;
+	struct dt_object_format  *dof	= &info->lti_dof;
+	struct dt_object	 *parent = NULL;
+	struct dt_object	 *child	= NULL;
+	char			  name[8];
+	int			  node	= lfsck_dev_idx(lfsck->li_bottom);
+	int			  rc	= 0;
+	ENTRY;
+
+	LASSERT(lfsck->li_master);
+
+	sprintf(name, "MDT%04x", node);
+	if (node == 0) {
+		parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
+						  &LU_LPF_FID);
+	} else {
+		struct lfsck_tgt_desc *ltd;
+
+		ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
+		if (unlikely(ltd == NULL))
+			RETURN(-ENODEV);
+
+		parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
+						  &LU_LPF_FID);
+		lfsck_tgt_put(ltd);
+	}
+	if (IS_ERR(parent))
+		RETURN(PTR_ERR(parent));
+
+	if (unlikely(!dt_try_as_dir(env, parent)))
+		GOTO(out, rc = -ENOTDIR);
+
+	mutex_lock(&lfsck->li_mutex);
+	if (lfsck->li_lpf_obj != NULL)
+		GOTO(unlock, rc = 0);
+
+	if (fid_is_zero(&bk->lb_lpf_fid)) {
+		/* There is corner case that: in former LFSCK scanning we have
+		 * created the .lustre/lost+found/MDTxxxx but failed to update
+		 * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
+		 * it from MDT0 firstly. */
+		rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
+			       (const struct dt_key *)name, BYPASS_CAPA);
+		if (rc != 0 && rc != -ENOENT)
+			GOTO(unlock, rc);
+
+		if (rc == 0) {
+			bk->lb_lpf_fid = *cfid;
+			rc = lfsck_bookmark_store(env, lfsck);
+		} else {
+			rc = lfsck_fid_alloc(env, lfsck, cfid, true);
+		}
+		if (rc != 0)
+			GOTO(unlock, rc);
+	} else {
+		*cfid = bk->lb_lpf_fid;
+	}
+
+	child = lfsck_object_find_by_dev(env, lfsck->li_bottom, cfid);
+	if (IS_ERR(child))
+		GOTO(unlock, rc = PTR_ERR(child));
+
+	if (dt_object_exists(child) != 0) {
+		if (unlikely(!dt_try_as_dir(env, child)))
+			GOTO(unlock, rc = -ENOTDIR);
+
+		lfsck->li_lpf_obj = child;
+		GOTO(unlock, rc = 0);
+	}
+
+	memset(la, 0, sizeof(*la));
+	la->la_atime = la->la_mtime = la->la_ctime = cfs_time_current_sec();
+	la->la_mode = S_IFDIR | S_IRWXU;
+	la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
+		       LA_UID | LA_GID;
+	memset(dof, 0, sizeof(*dof));
+	dof->dof_type = dt_mode_to_dft(S_IFDIR);
+
+	if (node == 0)
+		rc = lfsck_create_lpf_local(env, lfsck, parent, child, la,
+					    dof, name);
+	else
+		rc = lfsck_create_lpf_remote(env, lfsck, parent, child, la,
+					     dof, name);
+	if (rc == 0)
+		lfsck->li_lpf_obj = child;
+
+	GOTO(unlock, rc);
+
+unlock:
+	mutex_unlock(&lfsck->li_mutex);
+	if (rc != 0 && child != NULL && !IS_ERR(child))
+		lu_object_put(env, &child->do_lu);
+out:
+	if (parent != NULL && !IS_ERR(parent))
+		lu_object_put(env, &parent->do_lu);
+
+	return rc;
+}
+
+static int lfsck_fid_init(struct lfsck_instance *lfsck)
+{
+	struct lfsck_bookmark	*bk	= &lfsck->li_bookmark_ram;
+	struct seq_server_site	*ss;
+	char			*prefix;
+	int			 rc	= 0;
+	ENTRY;
+
+	ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
+	if (unlikely(ss == NULL))
+		RETURN(-ENODEV);
+
+	OBD_ALLOC_PTR(lfsck->li_seq);
+	if (lfsck->li_seq == NULL)
+		RETURN(-ENOMEM);
+
+	OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
+	if (prefix == NULL)
+		GOTO(out, rc = -ENOMEM);
+
+	snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
+	rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
+			     ss->ss_server_seq);
+	OBD_FREE(prefix, MAX_OBD_NAME + 7);
+	if (rc != 0)
+		GOTO(out, rc);
+
+	if (fid_is_sane(&bk->lb_last_fid))
+		lfsck->li_seq->lcs_fid = bk->lb_last_fid;
+
+	RETURN(0);
+
+out:
+	OBD_FREE_PTR(lfsck->li_seq);
+	lfsck->li_seq = NULL;
+
+	return rc;
+}
+
+static void lfsck_fid_fini(struct lfsck_instance *lfsck)
+{
+	if (lfsck->li_seq != NULL) {
+		seq_client_fini(lfsck->li_seq);
+		OBD_FREE_PTR(lfsck->li_seq);
+		lfsck->li_seq = NULL;
+	}
+}
+
 void lfsck_instance_cleanup(const struct lu_env *env,
 			    struct lfsck_instance *lfsck)
 {
@@ -370,11 +832,18 @@ void lfsck_instance_cleanup(const struct lu_env *env,
 		lfsck->li_bookmark_obj = NULL;
 	}
 
+	if (lfsck->li_lpf_obj != NULL) {
+		lu_object_put(env, &lfsck->li_lpf_obj->do_lu);
+		lfsck->li_lpf_obj = NULL;
+	}
+
 	if (lfsck->li_los != NULL) {
 		local_oid_storage_fini(env, lfsck->li_los);
 		lfsck->li_los = NULL;
 	}
 
+	lfsck_fid_fini(lfsck);
+
 	OBD_FREE_PTR(lfsck);
 }
 
@@ -2031,6 +2500,10 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key,
 		GOTO(out, rc);
 
 	if (master) {
+		rc = lfsck_fid_init(lfsck);
+		if (rc < 0)
+			GOTO(out, rc);
+
 		rc = lfsck_namespace_setup(env, lfsck);
 		if (rc < 0)
 			GOTO(out, rc);