X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Futils%2Fliblustreapi.c;h=ca5debe6f3bfeb0d840eca9d6ad0d47bce68dd3d;hp=0f85d294b7891abcb953eb6b833b6b1d16b602ae;hb=9b0b7264a8d7a6a2abe681d15fefc88c27d20c1e;hpb=0f38a0b9db4f7c6c101ee55fe7b94d49a32eabe9

diff --git a/lustre/utils/liblustreapi.c b/lustre/utils/liblustreapi.c
index 0f85d29..ca5debe 100644
--- a/lustre/utils/liblustreapi.c
+++ b/lustre/utils/liblustreapi.c
@@ -400,9 +400,54 @@ int llapi_ioctl_unpack(struct obd_ioctl_data *data, char *pbuf, int max_len)
 
 /* XXX: llapi_xxx() functions return negative values upon failure */
 
-int llapi_stripe_limit_check(unsigned long long stripe_size, int stripe_offset,
-				int stripe_count, int stripe_pattern)
+int llapi_layout_search_ost(__u32 ost, char *pname, char *fsname)
 {
+	char ostname[MAX_OBD_NAME + 64];
+	char *pool_name = pname;
+	int rc = 0;
+
+	/**
+	 * The current policy is that the pool does not have to exist at the
+	 * setstripe time, see sanity-pfl/-flr tests.
+	 * If this logic will change, re-enable it.
+	 *
+	 * if (pname && strlen(pname) == 0)
+	 */
+		pool_name = NULL;
+
+	snprintf(ostname, sizeof(ostname), "%s-OST%04x_UUID",
+		 fsname, ost);
+	rc = llapi_search_ost(fsname, pool_name, ostname);
+	if (rc <= 0) {
+		if (rc == 0)
+			rc = -ENODEV;
+
+		llapi_error(LLAPI_MSG_ERROR, rc,
+			    "%s: cannot find OST %s in %s", __func__, ostname,
+			    pool_name != NULL ? "pool" : "system");
+		return rc;
+	}
+
+	return 0;
+}
+
+/**
+ * Verify the setstripe parameters before using.
+ * This is a pair method for comp_args_to_layout()/llapi_layout_sanity_cb()
+ * when just 1 component or a non-PFL layout is given.
+ *
+ * \param[in] param		stripe parameters
+ * \param[in] pool_name		pool name
+ * \param[in] fsname		lustre FS name
+ *
+ * \retval			0, success
+ *				< 0, error code on failre
+ */
+static int llapi_stripe_param_verify(const struct llapi_stripe_param *param,
+				     char **pool_name,
+				     char *fsname)
+{
+	int count;
 	static int page_size;
 	int rc = 0;
 
@@ -420,33 +465,94 @@ int llapi_stripe_limit_check(unsigned long long stripe_size, int stripe_offset,
 					  page_size, LOV_MIN_STRIPE_SIZE);
 		}
 	}
-	if (!llapi_stripe_size_is_aligned(stripe_size)) {
+	if (!llapi_stripe_size_is_aligned(param->lsp_stripe_size)) {
 		rc = -EINVAL;
 		llapi_error(LLAPI_MSG_ERROR, rc,
 			    "error: bad stripe_size %llu, must be an even multiple of %d bytes",
-			    (unsigned long long)stripe_size, page_size);
+			    param->lsp_stripe_size, page_size);
 		goto out;
 	}
-	if (!llapi_stripe_index_is_valid(stripe_offset)) {
+	if (!llapi_stripe_index_is_valid(param->lsp_stripe_offset)) {
 		rc = -EINVAL;
 		llapi_error(LLAPI_MSG_ERROR, rc, "error: bad stripe offset %d",
-				stripe_offset);
+			    param->lsp_stripe_offset);
 		goto out;
 	}
-	if (!llapi_stripe_count_is_valid(stripe_count)) {
+	if (llapi_stripe_size_is_too_big(param->lsp_stripe_size)) {
 		rc = -EINVAL;
-		llapi_error(LLAPI_MSG_ERROR, rc, "error: bad stripe count %d",
-				stripe_count);
+		llapi_error(LLAPI_MSG_ERROR, rc,
+			    "error: stripe size '%llu' over 4GB limit",
+			    param->lsp_stripe_size);
 		goto out;
 	}
-	if (llapi_stripe_size_is_too_big(stripe_size)) {
+
+	count = param->lsp_stripe_count;
+	if (param->lsp_stripe_pattern == LOV_PATTERN_MDT) {
 		rc = -EINVAL;
 		llapi_error(LLAPI_MSG_ERROR, rc,
-			    "error: stripe size '%llu' over 4GB limit",
-			    (unsigned long long)stripe_size);
+			    "Invalid pattern: %d, must be specified with -E\n",
+			    param->lsp_stripe_pattern);
 		goto out;
+	} else {
+		if (!llapi_stripe_count_is_valid(count)) {
+			rc = -EINVAL;
+			llapi_error(LLAPI_MSG_ERROR, rc,
+				    "Invalid stripe count %d\n", count);
+			goto out;
+		}
 	}
 
+	/* Make sure we have a good pool */
+	if (*pool_name != NULL) {
+		if (!llapi_pool_name_is_valid(pool_name, fsname)) {
+			rc = -EINVAL;
+			llapi_error(LLAPI_MSG_ERROR, rc,
+				    "Pool '%s' is not on filesystem '%s'",
+				    *pool_name, fsname);
+			goto out;
+		}
+
+		/* Make sure the pool exists and is non-empty */
+		rc = llapi_search_ost(fsname, *pool_name, NULL);
+		if (rc < 1) {
+			char *err = rc == 0 ? "has no OSTs" : "does not exist";
+
+			rc = -EINVAL;
+			llapi_error(LLAPI_MSG_ERROR, rc, "pool '%s.%s' %s",
+				    fsname, *pool_name, err);
+			goto out;
+		}
+		rc = 0;
+	}
+
+	/* sanity check of target list */
+	if (param->lsp_is_specific) {
+		bool found = false;
+		int i;
+
+		for (i = 0; i < count; i++) {
+			rc = llapi_layout_search_ost(param->lsp_osts[i],
+						     *pool_name, fsname);
+			if (rc)
+				goto out;
+
+			/* Make sure stripe offset is in OST list. */
+			if (param->lsp_osts[i] == param->lsp_stripe_offset)
+				found = true;
+		}
+		if (!found) {
+			rc = -EINVAL;
+			llapi_error(LLAPI_MSG_ERROR, rc,
+				    "%s: stripe offset '%d' is not in the target list",
+				    __func__, param->lsp_stripe_offset);
+			goto out;
+		}
+	} else if (param->lsp_stripe_offset != -1) {
+		rc = llapi_layout_search_ost(param->lsp_stripe_offset,
+					     *pool_name, fsname);
+		if (rc)
+			goto out;
+	}
 out:
 	errno = -rc;
 	return rc;
@@ -570,9 +676,9 @@ int llapi_file_open_param(const char *name, int flags, mode_t mode,
 			  const struct llapi_stripe_param *param)
 {
 	char fsname[MAX_OBD_NAME + 1] = { 0 };
-	char *pool_name = param->lsp_pool;
 	struct lov_user_md *lum = NULL;
-	size_t lum_size = sizeof(*lum);
+	char *pool_name = param->lsp_pool;
+	size_t lum_size;
 	int fd, rc;
 
 	/* Make sure we are on a Lustre file system */
@@ -585,82 +691,17 @@ int llapi_file_open_param(const char *name, int flags, mode_t mode,
 	}
 
 	/* Check if the stripe pattern is sane. */
-	rc = llapi_stripe_limit_check(param->lsp_stripe_size,
-				      param->lsp_stripe_offset,
-				      param->lsp_stripe_count,
-				      param->lsp_stripe_pattern);
+	rc = llapi_stripe_param_verify(param, &pool_name, fsname);
 	if (rc != 0)
 		return rc;
 
-	/* Make sure we have a good pool */
-	if (pool_name != NULL) {
-		/*
-		 * in case user gives the full pool name <fsname>.<poolname>,
-		 * strip the fsname
-		 */
-		char *ptr = strchr(pool_name, '.');
-
-		if (ptr != NULL) {
-			*ptr = '\0';
-			if (strcmp(pool_name, fsname) != 0) {
-				*ptr = '.';
-				llapi_err_noerrno(LLAPI_MSG_ERROR,
-					"Pool '%s' is not on filesystem '%s'",
-					pool_name, fsname);
-				return -EINVAL;
-			}
-			pool_name = ptr + 1;
-		}
-
-		/* Make sure the pool exists and is non-empty */
-		rc = llapi_search_ost(fsname, pool_name, NULL);
-		if (rc < 1) {
-			char *err = rc == 0 ? "has no OSTs" : "does not exist";
-
-			llapi_err_noerrno(LLAPI_MSG_ERROR, "pool '%s.%s' %s",
-					  fsname, pool_name, err);
-			return -EINVAL;
-		}
-
-		lum_size = sizeof(struct lov_user_md_v3);
-	}
-
-	/* sanity check of target list */
-	if (param->lsp_is_specific) {
-		char ostname[MAX_OBD_NAME + 64];
-		bool found = false;
-		int i;
-
-		for (i = 0; i < param->lsp_stripe_count; i++) {
-			snprintf(ostname, sizeof(ostname), "%s-OST%04x_UUID",
-				 fsname, param->lsp_osts[i]);
-			rc = llapi_search_ost(fsname, pool_name, ostname);
-			if (rc <= 0) {
-				if (rc == 0)
-					rc = -ENODEV;
-
-				llapi_error(LLAPI_MSG_ERROR, rc,
-					    "%s: cannot find OST %s in %s",
-					    __func__, ostname,
-					    pool_name != NULL ?
-					    "pool" : "system");
-				return rc;
-			}
-
-			/* Make sure stripe offset is in OST list. */
-			if (param->lsp_osts[i] == param->lsp_stripe_offset)
-				found = true;
-		}
-		if (!found) {
-			llapi_error(LLAPI_MSG_ERROR, -EINVAL,
-				    "%s: stripe offset '%d' is not in the target list",
-				    __func__, param->lsp_stripe_offset);
-			return -EINVAL;
-		}
-
+	if (param->lsp_is_specific)
 		lum_size = lov_user_md_size(param->lsp_stripe_count,
 					    LOV_USER_MAGIC_SPECIFIC);
-	}
+	else if (pool_name)
+		lum_size = sizeof(struct lov_user_md_v3);
+	else
+		lum_size = sizeof(*lum);
 
 	lum = calloc(1, lum_size);
 	if (lum == NULL)
@@ -735,6 +776,18 @@ retry_open:
 	return fd;
 }
 
+int llapi_file_is_encrypted(int fd)
+{
+	unsigned long flags;
+	int rc;
+
+	rc = ioctl(fd, FS_IOC_GETFLAGS, &flags);
+	if (rc == -1)
+		return -errno;
+
+	return !!(flags & LUSTRE_ENCRYPT_FL);
+}
+
 int llapi_file_open_pool(const char *name, int flags, int mode,
 			 unsigned long long stripe_size, int stripe_offset,
 			 int stripe_count, int stripe_pattern, char *pool_name)
@@ -962,6 +1015,8 @@ static inline void param2lmu(struct lmv_user_md *lmu,
 	lmu->lum_stripe_count = param->lsp_stripe_count;
 	lmu->lum_stripe_offset = param->lsp_stripe_offset;
 	lmu->lum_hash_type = param->lsp_stripe_pattern;
+	lmu->lum_max_inherit = param->lsp_max_inherit;
+	lmu->lum_max_inherit_rr = param->lsp_max_inherit_rr;
 	if (param->lsp_pool != NULL)
 		strncpy(lmu->lum_pool_name, param->lsp_pool, LOV_MAXPOOLNAME);
 	if (param->lsp_is_specific) {
@@ -1086,6 +1141,9 @@ int llapi_dir_create(const char *name, mode_t mode,
 	data.ioc_inlbuf2 = (char *)lmu;
 	data.ioc_inllen2 = lmu_size;
 	data.ioc_type = mode;
+	if (param->lsp_is_create)
+		/* borrow obdo1.o_flags to store this flag */
+		data.ioc_obdo1.o_flags = OBD_FL_OBDMDEXISTS;
 	rc = llapi_ioctl_pack(&data, &buf, sizeof(rawbuf));
 	if (rc) {
 		llapi_error(LLAPI_MSG_ERROR, rc,
@@ -1653,6 +1711,34 @@ free_path:
 /* wrapper for lfs.c and obd.c */
 int llapi_poollist(const char *name)
 {
+	int poolcount, rc, i;
+	char *buf, **pools;
+
+	rc = llapi_get_poolbuf(name, &buf, &pools, &poolcount);
+	if (rc)
+		return rc;
+
+	for (i = 0; i < poolcount; i++)
+		llapi_printf(LLAPI_MSG_NORMAL, "%s\n", pools[i]);
+	free(buf);
+
+	return 0;
+}
+
+/**
+ * Get buffer that holds uuids and the list of pools in a filesystem.
+ *
+ * \param name		filesystem name or path
+ * \param buf		bufffer that has to be freed if function returns 0
+ * \param pools		pointer to the list of pools in buffer
+ * \param poolcount	number of pools
+ *
+ * \return 0 when found at least 1 pool, i.e. poolcount  > 0
+ * \retval -error failure
+ */
+int llapi_get_poolbuf(const char *name, char **buf,
+		      char ***pools, int *poolcount)
+{
 	/*
 	 * list of pool names (assume that pool count is smaller
 	 * than OST count)
@@ -1660,7 +1746,7 @@ int llapi_poollist(const char *name)
 	char **list, *buffer = NULL, *fsname = (char *)name;
 	char *poolname = NULL, *tmp = NULL, data[16];
 	enum param_filter type = FILTER_BY_PATH;
-	int obdcount, bufsize, rc, nb, i;
+	int obdcount, bufsize, rc, nb;
 
 	if (name == NULL)
 		return -EINVAL;
@@ -1712,11 +1798,15 @@ retry_get_pools:
 		goto retry_get_pools;
 	}
 
-	for (i = 0; i < nb; i++)
-		llapi_printf(LLAPI_MSG_NORMAL, "%s\n", list[i]);
 	rc = (nb < 0 ? nb : 0);
+	if (!rc) {
+		*buf = buffer;
+		*pools = list;
+		*poolcount = nb;
+	}
 err:
-	if (buffer)
+	/* Don't free buffer, it will be used later */
+	if (rc && buffer)
 		free(buffer);
 	if (fsname != NULL && type == FILTER_BY_FS_NAME)
 		free(fsname);
@@ -1728,6 +1818,11 @@ typedef int (semantic_func_t)(char *path, int p, int *d,
 
 #define OBD_NOT_FOUND           (-1)
 
+static bool lmv_is_foreign(__u32 magic)
+{
+	return magic == LMV_MAGIC_FOREIGN;
+}
+
 static void find_param_fini(struct find_param *param)
 {
 	if (param->fp_migrate)
@@ -1825,6 +1920,8 @@ static int cb_get_dirstripe(char *path, int *d, struct find_param *param)
 	int ret;
 	bool did_nofollow = false;
 
+	if (!d || *d < 0)
+		return -ENOTDIR;
 again:
 	param->fp_lmv_md->lum_stripe_count = param->fp_lmv_stripe_count;
 	if (param->fp_get_default_lmv)
@@ -1867,7 +1964,7 @@ again:
 		int lmv_size;
 
 		/* if foreign LMV case, fake stripes number */
-		if (param->fp_lmv_md->lum_magic == LMV_MAGIC_FOREIGN) {
+		if (lmv_is_foreign(param->fp_lmv_md->lum_magic)) {
 			struct lmv_foreign_md *lfm;
 
 			lfm = (struct lmv_foreign_md *)param->fp_lmv_md;
@@ -1903,6 +2000,7 @@ again:
 		}
 		goto again;
 	}
+
 	return ret;
 }
 
@@ -2873,7 +2971,8 @@ static void lov_dump_user_lmm_header(struct lov_user_md *lum, char *path,
 			     space, prefix,
 			     (uintmax_t)lmm_oi_id(&lum->lmm_oi));
 	}
-	if ((verbose & (VERBOSE_DETAIL | VERBOSE_DFID)) && !is_dir) {
+
+	if (verbose & (VERBOSE_DETAIL | VERBOSE_DFID)) {
 		__u64 seq;
 		__u32 oid;
 		__u32 ver;
@@ -2881,30 +2980,51 @@ static void lov_dump_user_lmm_header(struct lov_user_md *lum, char *path,
 		if (verbose & ~VERBOSE_DFID)
 			llapi_printf(LLAPI_MSG_NORMAL, "%slmm_fid:           ",
 				     space);
-		/*
-		 * This needs a bit of hand-holding since old 1.x lmm_oi
-		 * have { oi.oi_id = mds_inum, oi.oi_seq = 0 } and 2.x lmm_oi
-		 * have { oi.oi_id = mds_oid, oi.oi_seq = mds_seq } instead of
-		 * a real FID.  Ideally the 2.x code would have stored this
-		 * like a FID with { oi_id = mds_seq, oi_seq = mds_oid } so the
-		 * ostid union lu_fid { f_seq = mds_seq, f_oid = mds_oid }
-		 * worked properly (especially since IGIF FIDs use mds_inum as
-		 * the FID SEQ), but unfortunately that didn't happen.
-		 *
-		 * Print it to look like an IGIF FID, even though the fields
-		 * are reversed on disk, so that it makes sense to userspace.
-		 *
-		 * Don't use ostid_id() and ostid_seq(), since they assume the
-		 * oi_fid fields are in the right order.  This is why there are
-		 * separate lmm_oi_seq() and lmm_oi_id() routines for this.
-		 *
-		 * For newer layout types hopefully this will be a real FID.
-		 */
-		seq = lmm_oi_seq(&lum->lmm_oi) == 0 ?
-			lmm_oi_id(&lum->lmm_oi) : lmm_oi_seq(&lum->lmm_oi);
-		oid = lmm_oi_seq(&lum->lmm_oi) == 0 ?
-			0 : (__u32)lmm_oi_id(&lum->lmm_oi);
-		ver = (__u32)(lmm_oi_id(&lum->lmm_oi) >> 32);
+
+		if (is_dir) {
+			struct lu_fid dir_fid;
+
+			rc = llapi_path2fid(path, &dir_fid);
+			if (rc)
+				llapi_error(LLAPI_MSG_ERROR, rc,
+					    "Cannot determine directory fid.");
+
+			seq = dir_fid.f_seq;
+			oid = dir_fid.f_oid;
+			ver = dir_fid.f_ver;
+		} else {
+			/*
+			 * This needs a bit of hand-holding since old 1.x
+			 * lmm_oi have { oi.oi_id = mds_inum, oi.oi_seq = 0 }
+			 * and 2.x lmm_oi have { oi.oi_id = mds_oid,
+			 * oi.oi_seq = mds_seq } instead of a real FID.
+			 * Ideally the 2.x code would have stored this like a
+			 * FID with { oi_id = mds_seq, oi_seq = mds_oid } so
+			 * the ostid union lu_fid { f_seq = mds_seq,
+			 * f_oid = mds_oid } worked properly (especially since
+			 * IGIF FIDs use mds_inum as the FID SEQ), but
+			 * unfortunately that didn't happen.
+			 *
+			 * Print it to look like an IGIF FID, even though the
+			 * fields are reversed on disk, so that it makes sense
+			 * to userspace.
+			 *
+			 * Don't use ostid_id() and ostid_seq(), since they
+			 * assume the oi_fid fields are in the right order.
+			 * This is why there are separate lmm_oi_seq() and
+			 * lmm_oi_id() routines for this.
+			 *
+			 * For newer layout types hopefully this will be a
+			 * real FID.
+			 */
+			seq = lmm_oi_seq(&lum->lmm_oi) == 0 ?
+				lmm_oi_id(&lum->lmm_oi) :
+				lmm_oi_seq(&lum->lmm_oi);
+			oid = lmm_oi_seq(&lum->lmm_oi) == 0 ?
+			    0 : (__u32)lmm_oi_id(&lum->lmm_oi);
+			ver = (__u32)(lmm_oi_id(&lum->lmm_oi) >> 32);
+		}
+
 		if (yaml)
 			llapi_printf(LLAPI_MSG_NORMAL, DFID_NOBRACE"\n",
 				     (unsigned long long)seq, oid, ver);
@@ -3211,10 +3331,56 @@ void lmv_dump_user_lmm(struct lmv_user_md *lum, char *pool_name,
 			llapi_printf(LLAPI_MSG_NORMAL, ",bad_type");
 		if (flags & LMV_HASH_FLAG_LOST_LMV)
 			llapi_printf(LLAPI_MSG_NORMAL, ",lost_lmv");
+		if (flags & LMV_HASH_FLAG_FIXED)
+			llapi_printf(LLAPI_MSG_NORMAL, ",fixed");
+		if (flags & ~LMV_HASH_FLAG_KNOWN)
+			llapi_printf(LLAPI_MSG_NORMAL, ",unknown_%04x",
+				     flags & ~LMV_HASH_FLAG_KNOWN);
 
-		separator = "\n";
+		if (verbose & VERBOSE_HASH_TYPE && !yaml)
+			separator = " ";
+		else
+			separator = "\n";
+	}
+
+	if ((verbose & VERBOSE_INHERIT) && lum->lum_magic == LMV_USER_MAGIC) {
+		llapi_printf(LLAPI_MSG_NORMAL, "%s", separator);
+		if (verbose & ~VERBOSE_INHERIT)
+			llapi_printf(LLAPI_MSG_NORMAL, "lmv_max_inherit: ");
+		if (lum->lum_max_inherit == LMV_INHERIT_UNLIMITED)
+			llapi_printf(LLAPI_MSG_NORMAL, "-1");
+		else if (lum->lum_max_inherit == LMV_INHERIT_NONE)
+			llapi_printf(LLAPI_MSG_NORMAL, "0");
+		else
+			llapi_printf(LLAPI_MSG_NORMAL, "%hhu",
+				     lum->lum_max_inherit);
+		if (verbose & VERBOSE_INHERIT && !yaml)
+			separator = " ";
+		else
+			separator = "\n";
+	}
+
+	if ((verbose & VERBOSE_INHERIT_RR) &&
+	    lum->lum_magic == LMV_USER_MAGIC &&
+	    lum->lum_stripe_offset == LMV_OFFSET_DEFAULT) {
+		llapi_printf(LLAPI_MSG_NORMAL, "%s", separator);
+		if (verbose & ~VERBOSE_INHERIT_RR)
+			llapi_printf(LLAPI_MSG_NORMAL, "lmv_max_inherit_rr: ");
+		if (lum->lum_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED)
+			llapi_printf(LLAPI_MSG_NORMAL, "-1");
+		else if (lum->lum_max_inherit_rr == LMV_INHERIT_RR_NONE)
+			llapi_printf(LLAPI_MSG_NORMAL, "0");
+		else
+			llapi_printf(LLAPI_MSG_NORMAL, "%hhu",
+				     lum->lum_max_inherit_rr);
+		if (verbose & VERBOSE_INHERIT_RR && !yaml)
+			separator = " ";
+		else
+			separator = "\n";
 	}
 
+	separator = "\n";
+
 	if (verbose & VERBOSE_OBJID && lum->lum_magic != LMV_USER_MAGIC) {
 		llapi_printf(LLAPI_MSG_NORMAL, "%s", separator);
 		if (lum->lum_stripe_count > 0)
@@ -3504,7 +3670,7 @@ static inline void
 lov_v1v3_pool_name(struct lov_user_md *v1, char *pool_name)
 {
 	if (v1->lmm_magic == LOV_USER_MAGIC_V3)
-		snprintf(pool_name, LOV_MAXPOOLNAME, "%s",
+		snprintf(pool_name, LOV_MAXPOOLNAME + 1, "%s",
 			 ((struct lov_user_md_v3 *)v1)->lmm_pool_name);
 	else
 		pool_name[0] = '\0';
@@ -4005,7 +4171,7 @@ static void llapi_lov_dump_user_lmm(struct find_param *param, char *path,
 	}
 }
 
-int llapi_file_get_stripe(const char *path, struct lov_user_md *lum)
+static int llapi_file_get_stripe1(const char *path, struct lov_user_md *lum)
 {
 	const char *fname;
 	char *dname;
@@ -4047,6 +4213,31 @@ out_free:
 	return rc;
 }
 
+int llapi_file_get_stripe(const char *path, struct lov_user_md *lum)
+{
+	char *canon_path = NULL;
+	int rc, rc2;
+
+	rc = llapi_file_get_stripe1(path, lum);
+	if (!(rc == -ENOTTY || rc == -ENODATA))
+		goto out;
+
+	/* Handle failure due to symlinks by dereferencing path manually. */
+	canon_path = canonicalize_file_name(path);
+	if (canon_path == NULL)
+		goto out; /* Keep original rc. */
+
+	rc2 = llapi_file_get_stripe1(canon_path, lum);
+	if (rc2 < 0)
+		goto out; /* Keep original rc. */
+
+	rc = 0;
+out:
+	free(canon_path);
+
+	return rc;
+}
+
 int llapi_file_lookup(int dirfd, const char *name)
 {
 	struct obd_ioctl_data data = { 0 };
@@ -4491,7 +4682,7 @@ static int find_check_foreign(struct find_param *param)
 		struct lmv_foreign_md *lfm;
 
 		lfm = (void *)param->fp_lmv_md;
-		if (lfm->lfm_magic != LMV_MAGIC_FOREIGN) {
+		if (lmv_is_foreign(lfm->lfm_magic)) {
 			if (param->fp_foreign_type == LU_FOREIGN_TYPE_UNKNOWN)
 				return param->fp_exclude_foreign ? 1 : -1;
 			return -1;
@@ -4697,6 +4888,35 @@ static int fget_projid(int fd, int *projid)
 	return 0;
 }
 
+/*
+ * Check that the file's permissions in *st matches the one in find_param
+ */
+static int check_file_permissions(const struct find_param *param,
+			mode_t mode)
+{
+	int decision = 0;
+
+	mode &= 07777;
+
+	switch (param->fp_perm_sign) {
+	case LFS_FIND_PERM_EXACT:
+		decision = (mode == param->fp_perm);
+		break;
+	case LFS_FIND_PERM_ALL:
+		decision = ((mode & param->fp_perm) == param->fp_perm);
+		break;
+	case LFS_FIND_PERM_ANY:
+		decision = ((mode & param->fp_perm) != 0);
+		break;
+	}
+
+	if ((param->fp_exclude_perm && decision)
+		|| (!param->fp_exclude_perm && !decision))
+		return -1;
+	else
+		return 1;
+}
+
 static int cb_find_init(char *path, int p, int *dp,
 			void *data, struct dirent64 *de)
 {
@@ -4726,16 +4946,21 @@ static int cb_find_init(char *path, int p, int *dp,
 	}
 
 	/* See if we can check the file type from the dirent. */
-	if (param->fp_type != 0 && de != NULL && de->d_type != DT_UNKNOWN) {
-		checked_type = 1;
+	if (de != NULL && de->d_type != DT_UNKNOWN) {
+		if (param->fp_type != 0) {
+			checked_type = 1;
 
-		if (DTTOIF(de->d_type) == param->fp_type) {
-			if (param->fp_exclude_type)
-				goto decided;
-		} else {
-			if (!param->fp_exclude_type)
-				goto decided;
+			if (DTTOIF(de->d_type) == param->fp_type) {
+				if (param->fp_exclude_type)
+					goto decided;
+			} else {
+				if (!param->fp_exclude_type)
+					goto decided;
+			}
 		}
+		if ((param->fp_check_mdt_count || param->fp_hash_type ||
+		     param->fp_check_hash_flag) && de->d_type != DT_DIR)
+			goto decided;
 	}
 
 	ret = 0;
@@ -4751,30 +4976,73 @@ static int cb_find_init(char *path, int p, int *dp,
 	    param->fp_check_size || param->fp_check_blocks ||
 	    find_check_lmm_info(param) ||
 	    param->fp_check_mdt_count || param->fp_hash_type ||
-	    param->fp_check_hash_flag)
+	    param->fp_check_hash_flag || param->fp_perm_sign)
 		decision = 0;
 
 	if (param->fp_type != 0 && checked_type == 0)
 		decision = 0;
 
 	if (decision == 0) {
-		if (d != -1 && (param->fp_check_mdt_count ||
-		    param->fp_hash_type || param->fp_check_foreign ||
-		    param->fp_check_hash_flag)) {
+		if (d != -1 &&
+		    (param->fp_check_mdt_count || param->fp_hash_type ||
+		     param->fp_check_hash_flag || param->fp_check_foreign)) {
 			param->fp_get_lmv = 1;
 			ret = cb_get_dirstripe(path, &d, param);
 			if (ret != 0) {
-				/*
-				 * XXX this works to decide for foreign
-				 * criterion only
-				 */
-				if (errno == ENODATA &&
-				    param->fp_check_foreign) {
-					if (param->fp_exclude_foreign)
-						goto foreign;
-					goto decided;
+				if (errno == ENODATA) {
+					ret = 0;
+					if (param->fp_check_mdt_count ||
+					    param->fp_hash_type ||
+					    param->fp_check_hash_flag) {
+						param->fp_lmv_md->lum_stripe_count = 0;
+						param->fp_lmv_md->lum_hash_type = 0;
+					}
+					if (param->fp_check_foreign) {
+						if (param->fp_exclude_foreign)
+							goto print;
+						goto decided;
+					}
+				} else {
+					return ret;
 				}
-				return ret;
+			}
+
+			if (param->fp_check_mdt_count) {
+				if (lmv_is_foreign(param->fp_lmv_md->lum_magic))
+					goto decided;
+
+				decision = find_value_cmp(param->fp_lmv_md->lum_stripe_count,
+							  param->fp_mdt_count,
+							  param->fp_mdt_count_sign,
+							  param->fp_exclude_mdt_count, 1, 0);
+				if (decision == -1)
+					goto decided;
+			}
+
+			if (param->fp_hash_type) {
+				__u32 found;
+				__u32 type = param->fp_lmv_md->lum_hash_type &
+					LMV_HASH_TYPE_MASK;
+
+				if (lmv_is_foreign(param->fp_lmv_md->lum_magic))
+					goto decided;
+
+				found = (1 << type) & param->fp_hash_type;
+				if ((found && param->fp_exclude_hash_type) ||
+				    (!found && !param->fp_exclude_hash_type))
+					goto decided;
+			}
+
+			if (param->fp_check_hash_flag) {
+				__u32 flags = param->fp_lmv_md->lum_hash_type &
+					~LMV_HASH_TYPE_MASK;
+
+				if (lmv_is_foreign(param->fp_lmv_md->lum_magic))
+					goto decided;
+
+				if (!(flags & param->fp_hash_inflags) ||
+				    (flags & param->fp_hash_exflags))
+					goto decided;
 			}
 		}
 
@@ -4839,7 +5107,18 @@ static int cb_find_init(char *path, int p, int *dp,
 		}
 	}
 
+	/* Check the file permissions from the stat info */
+	if (param->fp_perm_sign) {
+		decision = check_file_permissions(param, lmd->lmd_stx.stx_mode);
+		if (decision == -1)
+			goto decided;
+	}
+
 	if (param->fp_type && !checked_type) {
+		if ((param->fp_check_mdt_count || param->fp_check_hash_flag ||
+		     param->fp_hash_type) && !S_ISDIR(lmd->lmd_stx.stx_mode))
+			goto decided;
+
 		if ((lmd->lmd_stx.stx_mode & S_IFMT) == param->fp_type) {
 			if (param->fp_exclude_type)
 				goto decided;
@@ -4903,61 +5182,12 @@ static int cb_find_init(char *path, int p, int *dp,
 			goto decided;
 	}
 
-	if (param->fp_check_mdt_count) {
-		if (param->fp_lmv_md->lum_magic == LMV_MAGIC_FOREIGN) {
-			decision = -1;
-			goto decided;
-		}
-
-		decision = find_value_cmp(
-				param->fp_lmv_md->lum_stripe_count,
-				param->fp_mdt_count,
-				param->fp_mdt_count_sign,
-				param->fp_exclude_mdt_count, 1, 0);
-		if (decision == -1)
-			goto decided;
-	}
-
 	if (param->fp_check_layout) {
 		decision = find_check_layout(param);
 		if (decision == -1)
 			goto decided;
 	}
 
-	if (param->fp_hash_type) {
-		__u32 found;
-		__u32 type = param->fp_lmv_md->lum_hash_type &
-			     LMV_HASH_TYPE_MASK;
-
-		if (param->fp_lmv_md->lum_magic == LMV_MAGIC_FOREIGN) {
-			decision = -1;
-			goto decided;
-		}
-
-		found = (1 << type) & param->fp_hash_type;
-		if ((found && param->fp_exclude_hash_type) ||
-		    (!found && !param->fp_exclude_hash_type)) {
-			decision = -1;
-			goto decided;
-		}
-	}
-
-	if (param->fp_check_hash_flag) {
-		__u32 flags = param->fp_lmv_md->lum_hash_type &
-			      ~LMV_HASH_TYPE_MASK;
-
-		if (param->fp_lmv_md->lum_magic == LMV_MAGIC_FOREIGN) {
-			decision = -1;
-			goto decided;
-		}
-
-		if (!(flags & param->fp_hash_inflags) ||
-		     (flags & param->fp_hash_exflags)) {
-			decision = -1;
-			goto decided;
-		}
-	}
-
 	/* If an OBD UUID is specified but none matches, skip this file. */
 	if ((param->fp_obd_uuid && param->fp_obd_index == OBD_NOT_FOUND) ||
 	    (param->fp_mdt_uuid && param->fp_mdt_index == OBD_NOT_FOUND))
@@ -5184,12 +5414,9 @@ obd_matches:
 			goto decided;
 	}
 
-foreign:
-	llapi_printf(LLAPI_MSG_NORMAL, "%s", path);
-	if (param->fp_zero_end)
-		llapi_printf(LLAPI_MSG_NORMAL, "%c", '\0');
-	else
-		llapi_printf(LLAPI_MSG_NORMAL, "\n");
+print:
+	llapi_printf(LLAPI_MSG_NORMAL, "%s%c", path,
+		     param->fp_zero_end ? '\0' : '\n');
 
 decided:
 	ret = 0;
@@ -5247,6 +5474,9 @@ static int cb_migrate_mdt_init(char *path, int p, int *dp,
 	data.ioc_inlbuf2 = (char *)lmu;
 	data.ioc_inllen2 = lmv_user_md_size(lmu->lum_stripe_count,
 					    lmu->lum_magic);
+	/* reach bottom? */
+	if (param->fp_depth == param->fp_max_depth)
+		data.ioc_type = MDS_MIGRATE_NSONLY;
 	ret = llapi_ioctl_pack(&data, &rawbuf, sizeof(raw));
 	if (ret != 0) {
 		llapi_error(LLAPI_MSG_ERROR, ret,
@@ -5288,6 +5518,11 @@ migrate:
 	}
 
 out:
+	/* Do not get down anymore? */
+	if (param->fp_depth == param->fp_max_depth)
+		ret = 1;
+	param->fp_depth++;
+
 	if (dp != NULL) {
 		/*
 		 * If the directory is being migration, we need
@@ -5821,39 +6056,6 @@ int llapi_get_connect_flags(const char *mnt, __u64 *flags)
 }
 
 /**
- * Get a 64-bit value representing the version of file data pointed by fd.
- *
- * Each write or truncate, flushed on OST, will change this value. You can use
- * this value to verify if file data was modified. This only checks the file
- * data, not metadata.
- *
- * \param  flags  0: no flush pages, usually used it the process has already
- *		    taken locks;
- *                LL_DV_RD_FLUSH: OSTs will take LCK_PR to flush dirty pages
- *                  from clients;
- *                LL_DV_WR_FLUSH: OSTs will take LCK_PW to flush all caching
- *                  pages from clients.
- *
- * \retval 0 on success.
- * \retval -errno on error.
- */
-int llapi_get_data_version(int fd, __u64 *data_version, __u64 flags)
-{
-	int rc;
-	struct ioc_data_version idv;
-
-	idv.idv_flags = (__u32)flags;
-
-	rc = ioctl(fd, LL_IOC_DATA_VERSION, &idv);
-	if (rc)
-		rc = -errno;
-	else
-		*data_version = idv.idv_version;
-
-	return rc;
-}
-
-/**
  * Flush cached pages from all clients.
  *
  * \param fd	File descriptor
@@ -5867,301 +6069,3 @@ int llapi_file_flush(int fd)
 	return llapi_get_data_version(fd, &dv, LL_DV_WR_FLUSH);
 }
 
-/*
- * Fetch layout version from OST objects. Layout version on OST objects are
- * only set when the file is a mirrored file AND after the file has been
- * written at least once.
- *
- * It actually fetches the least layout version from the objects.
- */
-int llapi_get_ost_layout_version(int fd, __u32 *layout_version)
-{
-	int rc;
-	struct ioc_data_version idv = { 0 };
-
-	rc = ioctl(fd, LL_IOC_DATA_VERSION, &idv);
-	if (rc)
-		rc = -errno;
-	else
-		*layout_version = idv.idv_layout_version;
-
-	return rc;
-}
-
-/*
- * Create a file without any name and open it for read/write
- *
- * - file is created as if it were a standard file in the given \a directory
- * - file does not appear in \a directory and mtime does not change because
- *   the filename is handled specially by the Lustre MDS.
- * - file is destroyed at final close
- *
- * \param[in]	directory	directory from which to inherit layout/MDT idx
- * \param[in]	mdt_idx		MDT index on which the file is created,
- *				\a idx == -1 means no specific MDT is requested
- * \param[in]	mode		standard open(2) mode
- * \param[in]	stripe_param	stripe parameters. May be NULL.
- *
- * \retval	a file descriptor on success.
- * \retval	-errno on error.
- */
-int llapi_create_volatile_param(const char *directory, int mdt_idx,
-				int open_flags, mode_t mode,
-				const struct llapi_stripe_param *stripe_param)
-{
-	char file_path[PATH_MAX];
-	int saved_errno = errno;
-	int fd;
-	unsigned int rnumber;
-	int rc;
-
-	do {
-		rnumber = random();
-		if (mdt_idx == -1)
-			rc = snprintf(file_path, sizeof(file_path),
-				      "%s/" LUSTRE_VOLATILE_HDR "::%.4X",
-				      directory, rnumber);
-		else
-			rc = snprintf(file_path, sizeof(file_path),
-				      "%s/" LUSTRE_VOLATILE_HDR ":%.4X:%.4X",
-				      directory, mdt_idx, rnumber);
-
-		if (rc < 0 || rc >= sizeof(file_path))
-			return -ENAMETOOLONG;
-
-		/*
-		 * Either open O_WRONLY or O_RDWR, creating RDONLY
-		 * is non-sensical here
-		 */
-		if ((open_flags & O_ACCMODE) == O_RDONLY)
-			open_flags = O_RDWR | (open_flags & ~O_ACCMODE);
-
-		open_flags |= O_CREAT | O_EXCL | O_NOFOLLOW;
-
-		if (stripe_param != NULL) {
-			fd = llapi_file_open_param(file_path, open_flags,
-						   mode, stripe_param);
-			if (fd < 0)
-				rc = fd;
-		} else {
-			fd = open(file_path, open_flags, mode);
-			if (fd < 0)
-				rc = -errno;
-		}
-	} while (fd < 0 && rc == -EEXIST);
-
-	if (fd < 0) {
-		llapi_error(LLAPI_MSG_ERROR, rc,
-			    "Cannot create volatile file '%s' in '%s'",
-			    file_path + strlen(directory) + 1 +
-			    LUSTRE_VOLATILE_HDR_LEN,
-			    directory);
-		return rc;
-	}
-
-	/*
-	 * Unlink file in case this wasn't a Lustre filesystem and the magic
-	 * volatile filename wasn't handled as intended. The effect is the
-	 * same. If volatile open was supported then we expect unlink() to
-	 * return -ENOENT.
-	 */
-	(void)unlink(file_path);
-
-	/*
-	 * Since we are returning successfully we restore errno (and
-	 * mask out possible EEXIST from open() and ENOENT from unlink().
-	 */
-	errno = saved_errno;
-
-	return fd;
-}
-
-/*
- * Create a file without any name open it for read/write
- *
- * - file is created as if it were a standard file in the given \a directory
- * - file does not appear in \a directory and mtime does not change because
- *   the filename is handled specially by the Lustre MDS.
- * - file is removed at final close
- * - file modes are rw------- since it doesn't make sense to have a read-only
- *   or write-only file that cannot be opened again.
- * - if user wants another mode it must use fchmod() on the open file, no
- *   security problems arise because it cannot be opened by another process.
- *
- * \param[in]	directory	directory from which to inherit layout/MDT idx
- * \param[in]	idx		MDT index on which the file is created,
- *				\a idx == -1 means no specific MDT is requested
- * \param[in]	open_flags	standard open(2) flags
- *
- * \retval	a file descriptor on success.
- * \retval	-errno on error.
- */
-int llapi_create_volatile_idx(const char *directory, int mdt_idx,
-			      int open_flags)
-{
-	return llapi_create_volatile_param(directory, mdt_idx, open_flags,
-					   S_IRUSR | S_IWUSR, NULL);
-}
-
-/**
- * Swap the layouts between 2 file descriptors
- * the 2 files must be open for writing
- * first fd received the ioctl, second fd is passed as arg
- * this is assymetric but avoid use of root path for ioctl
- */
-int llapi_fswap_layouts_grouplock(int fd1, int fd2, __u64 dv1, __u64 dv2,
-				  int gid, __u64 flags)
-{
-	struct lustre_swap_layouts	lsl;
-	struct stat			st1;
-	struct stat			st2;
-	int				rc;
-
-	if (flags & (SWAP_LAYOUTS_KEEP_ATIME | SWAP_LAYOUTS_KEEP_MTIME)) {
-		rc = fstat(fd1, &st1);
-		if (rc < 0)
-			return -errno;
-
-		rc = fstat(fd2, &st2);
-		if (rc < 0)
-			return -errno;
-	}
-	lsl.sl_fd = fd2;
-	lsl.sl_flags = flags;
-	lsl.sl_gid = gid;
-	lsl.sl_dv1 = dv1;
-	lsl.sl_dv2 = dv2;
-	rc = ioctl(fd1, LL_IOC_LOV_SWAP_LAYOUTS, &lsl);
-	if (rc < 0)
-		return -errno;
-
-	if (flags & (SWAP_LAYOUTS_KEEP_ATIME | SWAP_LAYOUTS_KEEP_MTIME)) {
-		struct timeval	tv1[2];
-		struct timeval	tv2[2];
-
-		memset(tv1, 0, sizeof(tv1));
-		memset(tv2, 0, sizeof(tv2));
-
-		if (flags & SWAP_LAYOUTS_KEEP_ATIME) {
-			tv1[0].tv_sec = st1.st_atime;
-			tv2[0].tv_sec = st2.st_atime;
-		} else {
-			tv1[0].tv_sec = st2.st_atime;
-			tv2[0].tv_sec = st1.st_atime;
-		}
-
-		if (flags & SWAP_LAYOUTS_KEEP_MTIME) {
-			tv1[1].tv_sec = st1.st_mtime;
-			tv2[1].tv_sec = st2.st_mtime;
-		} else {
-			tv1[1].tv_sec = st2.st_mtime;
-			tv2[1].tv_sec = st1.st_mtime;
-		}
-
-		rc = futimes(fd1, tv1);
-		if (rc < 0)
-			return -errno;
-
-		rc = futimes(fd2, tv2);
-		if (rc < 0)
-			return -errno;
-	}
-
-	return 0;
-}
-
-int llapi_fswap_layouts(int fd1, int fd2, __u64 dv1, __u64 dv2, __u64 flags)
-{
-	int	rc;
-	int	grp_id;
-
-	do
-		grp_id = random();
-	while (grp_id == 0);
-
-	rc = llapi_fswap_layouts_grouplock(fd1, fd2, dv1, dv2, grp_id, flags);
-	if (rc < 0)
-		return rc;
-
-	return 0;
-}
-
-/**
- * Swap the layouts between 2 files
- * the 2 files are open in write
- */
-int llapi_swap_layouts(const char *path1, const char *path2,
-		       __u64 dv1, __u64 dv2, __u64 flags)
-{
-	int	fd1, fd2, rc;
-
-	fd1 = open(path1, O_WRONLY | O_LOV_DELAY_CREATE);
-	if (fd1 < 0) {
-		rc = -errno;
-		llapi_error(LLAPI_MSG_ERROR, rc,
-			    "error: cannot open '%s' for write", path1);
-		goto out;
-	}
-
-	fd2 = open(path2, O_WRONLY | O_LOV_DELAY_CREATE);
-	if (fd2 < 0) {
-		rc = -errno;
-		llapi_error(LLAPI_MSG_ERROR, rc,
-			    "error: cannot open '%s' for write", path2);
-		goto out_close;
-	}
-
-	rc = llapi_fswap_layouts(fd1, fd2, dv1, dv2, flags);
-	if (rc < 0)
-		llapi_error(LLAPI_MSG_ERROR, rc,
-			    "error: cannot swap layout between '%s' and '%s'",
-			    path1, path2);
-
-	close(fd2);
-out_close:
-	close(fd1);
-out:
-	return rc;
-}
-
-/**
- * Take group lock.
- *
- * \param fd   File to lock.
- * \param gid  Group Identifier.
- *
- * \retval 0 on success.
- * \retval -errno on failure.
- */
-int llapi_group_lock(int fd, int gid)
-{
-	int rc;
-
-	rc = ioctl(fd, LL_IOC_GROUP_LOCK, gid);
-	if (rc < 0) {
-		rc = -errno;
-		llapi_error(LLAPI_MSG_ERROR, rc, "cannot get group lock");
-	}
-	return rc;
-}
-
-/**
- * Put group lock.
- *
- * \param fd   File to unlock.
- * \param gid  Group Identifier.
- *
- * \retval 0 on success.
- * \retval -errno on failure.
- */
-int llapi_group_unlock(int fd, int gid)
-{
-	int rc;
-
-	rc = ioctl(fd, LL_IOC_GROUP_UNLOCK, gid);
-	if (rc < 0) {
-		rc = -errno;
-		llapi_error(LLAPI_MSG_ERROR, rc, "cannot put group lock");
-	}
-	return rc;
-}