Support for longnames for files/directories (Linux part)

This patch adds the ability for zfs to support file/dir name up to 1023
bytes. This number is chosen so we can support up to 255 4-byte
characters. This new feature is represented by the new feature flag
feature@longname.

A new dataset property "longname" is also introduced to toggle longname
support for each dataset individually. This property can be disabled,
even if it contains longname files. In such case, new file cannot be
created with longname but existing longname files can still be looked
up.

Note that, to my knowledge native Linux filesystems don't support name
longer than 255 bytes. So there might be programs not able to work with
longname.

Note that NFS server may needs to use exportfs_get_name to reconnect
dentries, and the buffer being passed is limit to NAME_MAX+1 (256). So
NFS may not work when longname is enabled.

Note, FreeBSD vfs layer imposes a limit of 255 name lengh, so even
though we add code to support it here, it won't actually work.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Chunwei Chen <david.chen@nutanix.com>
Closes #15921

module/zfs/dmu_recv.c

@@ -602,6 +602,13 @@ recv_begin_check_feature_flags_impl(uint64_t featureflags, spa_t *spa)
 	    !spa_feature_is_enabled(spa, SPA_FEATURE_REDACTED_DATASETS))
 		return (SET_ERROR(ENOTSUP));
 
+	/*
+	 * If the LONGNAME is not enabled on the target, fail that request.
+	 */
+	if ((featureflags & DMU_BACKUP_FEATURE_LONGNAME) &&
+	    !spa_feature_is_enabled(spa, SPA_FEATURE_LONGNAME))
+		return (SET_ERROR(ENOTSUP));
+
 	return (0);
 }
 
@@ -990,6 +997,16 @@ dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
 	dmu_buf_will_dirty(newds->ds_dbuf, tx);
 	dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT;
 
+	/*
+	 * Activate longname feature if received
+	 */
+	if (featureflags & DMU_BACKUP_FEATURE_LONGNAME &&
+	    !dsl_dataset_feature_is_active(newds, SPA_FEATURE_LONGNAME)) {
+		dsl_dataset_activate_feature(newds->ds_object,
+		    SPA_FEATURE_LONGNAME, (void *)B_TRUE, tx);
+		newds->ds_feature[SPA_FEATURE_LONGNAME] = (void *)B_TRUE;
+	}
+
 	/*
 	 * If we actually created a non-clone, we need to create the objset
 	 * in our new dataset. If this is a raw send we postpone this until

module/zfs/dmu_send.c

@@ -2011,6 +2011,10 @@ setup_featureflags(struct dmu_send_params *dspp, objset_t *os,
 	if (dsl_dataset_feature_is_active(to_ds, SPA_FEATURE_LARGE_DNODE)) {
 		*featureflags |= DMU_BACKUP_FEATURE_LARGE_DNODE;
 	}
+
+	if (dsl_dataset_feature_is_active(to_ds, SPA_FEATURE_LONGNAME)) {
+		*featureflags |= DMU_BACKUP_FEATURE_LONGNAME;
+	}
 	return (0);
 }
 

module/zfs/dsl_dataset.c

@@ -494,7 +494,8 @@ dsl_dataset_get_snapname(dsl_dataset_t *ds)
 		return (err);
 	headphys = headdbuf->db_data;
 	err = zap_value_search(dp->dp_meta_objset,
-	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
+	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname,
+	    sizeof (ds->ds_snapname));
 	if (err != 0 && zfs_recover == B_TRUE) {
 		err = 0;
 		(void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname),

module/zfs/dsl_dir.c

@@ -239,7 +239,8 @@ dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
 				err = zap_value_search(dp->dp_meta_objset,
 				    dsl_dir_phys(dd->dd_parent)->
 				    dd_child_dir_zapobj,
-				    ddobj, 0, dd->dd_myname);
+				    ddobj, 0, dd->dd_myname,
+				    sizeof (dd->dd_myname));
 			}
 			if (err != 0)
 				goto errout;

module/zfs/zap.c

@@ -832,7 +832,12 @@ zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l,
 static int
 fzap_checkname(zap_name_t *zn)
 {
-	if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
+	uint32_t maxnamelen = zn->zn_normbuf_len;
+	uint64_t len = (uint64_t)zn->zn_key_orig_numints * zn->zn_key_intlen;
+	/* Only allow directory zap to have longname */
+	if (len > maxnamelen ||
+	    (len > ZAP_MAXNAMELEN &&
+	    zn->zn_zap->zap_dnode->dn_type != DMU_OT_DIRECTORY_CONTENTS))
 		return (SET_ERROR(ENAMETOOLONG));
 	return (0);
 }
@@ -1102,7 +1107,7 @@ zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
 
 int
 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
-    char *name)
+    char *name, uint64_t namelen)
 {
 	zap_cursor_t zc;
 	int err;
@@ -1110,12 +1115,13 @@ zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
 	if (mask == 0)
 		mask = -1ULL;
 
-	zap_attribute_t *za = zap_attribute_alloc();
+	zap_attribute_t *za = zap_attribute_long_alloc();
 	for (zap_cursor_init(&zc, os, zapobj);
 	    (err = zap_cursor_retrieve(&zc, za)) == 0;
 	    zap_cursor_advance(&zc)) {
 		if ((za->za_first_integer & mask) == (value & mask)) {
-			(void) strlcpy(name, za->za_name, MAXNAMELEN);
+			if (strlcpy(name, za->za_name, namelen) >= namelen)
+				err = SET_ERROR(ENAMETOOLONG);
 			break;
 		}
 	}
@@ -1130,7 +1136,7 @@ zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
 	zap_cursor_t zc;
 	int err = 0;
 
-	zap_attribute_t *za = zap_attribute_alloc();
+	zap_attribute_t *za = zap_attribute_long_alloc();
 	for (zap_cursor_init(&zc, os, fromobj);
 	    zap_cursor_retrieve(&zc, za) == 0;
 	    (void) zap_cursor_advance(&zc)) {
@@ -1155,7 +1161,7 @@ zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
 	zap_cursor_t zc;
 	int err = 0;
 
-	zap_attribute_t *za = zap_attribute_alloc();
+	zap_attribute_t *za = zap_attribute_long_alloc();
 	for (zap_cursor_init(&zc, os, fromobj);
 	    zap_cursor_retrieve(&zc, za) == 0;
 	    (void) zap_cursor_advance(&zc)) {
@@ -1180,7 +1186,7 @@ zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
 	zap_cursor_t zc;
 	int err = 0;
 
-	zap_attribute_t *za = zap_attribute_alloc();
+	zap_attribute_t *za = zap_attribute_long_alloc();
 	for (zap_cursor_init(&zc, os, fromobj);
 	    zap_cursor_retrieve(&zc, za) == 0;
 	    (void) zap_cursor_advance(&zc)) {

module/zfs/zap_micro.c

@@ -131,12 +131,12 @@ zap_hash(zap_name_t *zn)
 }
 
 static int
-zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
+zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags,
+    size_t outlen)
 {
 	ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
 
 	size_t inlen = strlen(name) + 1;
-	size_t outlen = ZAP_MAXNAMELEN;
 
 	int err = 0;
 	(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
@@ -149,23 +149,39 @@ zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
 boolean_t
 zap_match(zap_name_t *zn, const char *matchname)
 {
+	boolean_t res = B_FALSE;
 	ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
 
 	if (zn->zn_matchtype & MT_NORMALIZE) {
-		char norm[ZAP_MAXNAMELEN];
+		size_t namelen = zn->zn_normbuf_len;
+		char normbuf[ZAP_MAXNAMELEN];
+		char *norm = normbuf;
+
+		/*
+		 * Cannot allocate this on-stack as it exceed the stack-limit of
+		 * 1024.
+		 */
+		if (namelen > ZAP_MAXNAMELEN)
+			norm = kmem_alloc(namelen, KM_SLEEP);
 
 		if (zap_normalize(zn->zn_zap, matchname, norm,
-		    zn->zn_normflags) != 0)
-			return (B_FALSE);
-
-		return (strcmp(zn->zn_key_norm, norm) == 0);
+		    zn->zn_normflags, namelen) != 0) {
+			res = B_FALSE;
+		} else {
+			res = (strcmp(zn->zn_key_norm, norm) == 0);
+		}
+		if (norm != normbuf)
+			kmem_free(norm, namelen);
 	} else {
-		return (strcmp(zn->zn_key_orig, matchname) == 0);
+		res = (strcmp(zn->zn_key_orig, matchname) == 0);
 	}
+	return (res);
 }
 
 static kmem_cache_t *zap_name_cache;
 static kmem_cache_t *zap_attr_cache;
+static kmem_cache_t *zap_name_long_cache;
+static kmem_cache_t *zap_attr_long_cache;
 
 void
 zap_init(void)
@@ -177,6 +193,14 @@ zap_init(void)
 	zap_attr_cache = kmem_cache_create("zap_attr_cache",
 	    sizeof (zap_attribute_t) + ZAP_MAXNAMELEN,  0, NULL,
 	    NULL, NULL, NULL, NULL, 0);
+
+	zap_name_long_cache = kmem_cache_create("zap_name_long",
+	    sizeof (zap_name_t) + ZAP_MAXNAMELEN_NEW, 0, NULL, NULL,
+	    NULL, NULL, NULL, 0);
+
+	zap_attr_long_cache = kmem_cache_create("zap_attr_long_cache",
+	    sizeof (zap_attribute_t) + ZAP_MAXNAMELEN_NEW,  0, NULL,
+	    NULL, NULL, NULL, NULL, 0);
 }
 
 void
@@ -184,33 +208,47 @@ zap_fini(void)
 {
 	kmem_cache_destroy(zap_name_cache);
 	kmem_cache_destroy(zap_attr_cache);
+	kmem_cache_destroy(zap_name_long_cache);
+	kmem_cache_destroy(zap_attr_long_cache);
 }
 
 static zap_name_t *
-zap_name_alloc(zap_t *zap)
+zap_name_alloc(zap_t *zap, boolean_t longname)
 {
-	zap_name_t *zn = kmem_cache_alloc(zap_name_cache, KM_SLEEP);
+	kmem_cache_t *cache = longname ? zap_name_long_cache : zap_name_cache;
+	zap_name_t *zn = kmem_cache_alloc(cache, KM_SLEEP);
+
 	zn->zn_zap = zap;
+	zn->zn_normbuf_len = longname ? ZAP_MAXNAMELEN_NEW : ZAP_MAXNAMELEN;
 	return (zn);
 }
 
 void
 zap_name_free(zap_name_t *zn)
 {
-	kmem_cache_free(zap_name_cache, zn);
+	if (zn->zn_normbuf_len == ZAP_MAXNAMELEN) {
+		kmem_cache_free(zap_name_cache, zn);
+	} else {
+		ASSERT3U(zn->zn_normbuf_len, ==, ZAP_MAXNAMELEN_NEW);
+		kmem_cache_free(zap_name_long_cache, zn);
+	}
 }
 
 static int
 zap_name_init_str(zap_name_t *zn, const char *key, matchtype_t mt)
 {
 	zap_t *zap = zn->zn_zap;
+	size_t key_len = strlen(key) + 1;
+
+	/* Make sure zn is allocated for longname if key is long */
+	IMPLY(key_len > ZAP_MAXNAMELEN,
+	    zn->zn_normbuf_len == ZAP_MAXNAMELEN_NEW);
 
 	zn->zn_key_intlen = sizeof (*key);
 	zn->zn_key_orig = key;
-	zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
+	zn->zn_key_orig_numints = key_len;
 	zn->zn_matchtype = mt;
 	zn->zn_normflags = zap->zap_normflags;
-	zn->zn_normbuf_len = ZAP_MAXNAMELEN;
 
 	/*
 	 * If we're dealing with a case sensitive lookup on a mixed or
@@ -226,7 +264,7 @@ zap_name_init_str(zap_name_t *zn, const char *key, matchtype_t mt)
 		 * what the hash is computed from.
 		 */
 		if (zap_normalize(zap, key, zn->zn_normbuf,
-		    zap->zap_normflags) != 0)
+		    zap->zap_normflags, zn->zn_normbuf_len) != 0)
 			return (SET_ERROR(ENOTSUP));
 		zn->zn_key_norm = zn->zn_normbuf;
 		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
@@ -245,7 +283,7 @@ zap_name_init_str(zap_name_t *zn, const char *key, matchtype_t mt)
 		 * what the matching is based on.  (Not the hash!)
 		 */
 		if (zap_normalize(zap, key, zn->zn_normbuf,
-		    zn->zn_normflags) != 0)
+		    zn->zn_normflags, zn->zn_normbuf_len) != 0)
 			return (SET_ERROR(ENOTSUP));
 		zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
 	}
@@ -256,7 +294,8 @@ zap_name_init_str(zap_name_t *zn, const char *key, matchtype_t mt)
 zap_name_t *
 zap_name_alloc_str(zap_t *zap, const char *key, matchtype_t mt)
 {
-	zap_name_t *zn = zap_name_alloc(zap);
+	size_t key_len = strlen(key) + 1;
+	zap_name_t *zn = zap_name_alloc(zap, (key_len > ZAP_MAXNAMELEN));
 	if (zap_name_init_str(zn, key, mt) != 0) {
 		zap_name_free(zn);
 		return (NULL);
@@ -491,7 +530,7 @@ mzap_open(dmu_buf_t *db)
 		zfs_btree_create_custom(&zap->zap_m.zap_tree, mze_compare,
 		    mze_find_in_buf, sizeof (mzap_ent_t), 512);
 
-		zap_name_t *zn = zap_name_alloc(zap);
+		zap_name_t *zn = zap_name_alloc(zap, B_FALSE);
 		for (uint16_t i = 0; i < zap->zap_m.zap_num_chunks; i++) {
 			mzap_ent_phys_t *mze =
 			    &zap_m_phys(zap)->mz_chunk[i];
@@ -698,7 +737,7 @@ mzap_upgrade(zap_t **zapp, const void *tag, dmu_tx_t *tx, zap_flags_t flags)
 
 	fzap_upgrade(zap, tx, flags);
 
-	zap_name_t *zn = zap_name_alloc(zap);
+	zap_name_t *zn = zap_name_alloc(zap, B_FALSE);
 	for (int i = 0; i < nchunks; i++) {
 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
 		if (mze->mze_name[0] == 0)
@@ -1625,21 +1664,38 @@ zap_remove_uint64_by_dnode(dnode_t *dn, const uint64_t *key, int key_numints,
 }
 
 
+static zap_attribute_t *
+zap_attribute_alloc_impl(boolean_t longname)
+{
+	zap_attribute_t *za;
+
+	za = kmem_cache_alloc((longname)? zap_attr_long_cache : zap_attr_cache,
+	    KM_SLEEP);
+	za->za_name_len = (longname)? ZAP_MAXNAMELEN_NEW : ZAP_MAXNAMELEN;
+	return (za);
+}
+
 zap_attribute_t *
 zap_attribute_alloc(void)
 {
-	uint32_t len = ZAP_MAXNAMELEN;
-	zap_attribute_t *za;
+	return (zap_attribute_alloc_impl(B_FALSE));
+}
 
-	za = kmem_cache_alloc(zap_attr_cache, KM_SLEEP);
-	za->za_name_len = len;
-	return (za);
+zap_attribute_t *
+zap_attribute_long_alloc(void)
+{
+	return (zap_attribute_alloc_impl(B_TRUE));
 }
 
 void
 zap_attribute_free(zap_attribute_t *za)
 {
-	kmem_cache_free(zap_attr_cache, za);
+	if (za->za_name_len == ZAP_MAXNAMELEN) {
+		kmem_cache_free(zap_attr_cache, za);
+	} else {
+		ASSERT3U(za->za_name_len, ==, ZAP_MAXNAMELEN_NEW);
+		kmem_cache_free(zap_attr_long_cache, za);
+	}
 }
 
 /*

module/zfs/zfs_ioctl.c

@@ -2594,6 +2594,41 @@ zfs_prop_set_special(const char *dsname, zprop_source_t source,
 		}
 		break;
 	}
+	case ZFS_PROP_LONGNAME:
+	{
+		zfsvfs_t *zfsvfs;
+
+		/*
+		 * Ignore the checks if the property is being applied as part of
+		 * 'zfs receive'. Because, we already check if the local pool
+		 * has SPA_FEATURE_LONGNAME enabled in dmu_recv_begin_check().
+		 */
+		if (source == ZPROP_SRC_RECEIVED) {
+			cmn_err(CE_NOTE, "Skipping ZFS_PROP_LONGNAME checks "
+			    "for dsname=%s\n", dsname);
+			err = -1;
+			break;
+		}
+
+		if ((err = zfsvfs_hold(dsname, FTAG, &zfsvfs, B_FALSE)) != 0) {
+			cmn_err(CE_WARN, "%s:%d Failed to hold for dsname=%s "
+			    "err=%d\n", __FILE__, __LINE__, dsname, err);
+			break;
+		}
+
+		if (!spa_feature_is_enabled(zfsvfs->z_os->os_spa,
+		    SPA_FEATURE_LONGNAME)) {
+			err = ENOTSUP;
+		} else {
+			/*
+			 * Set err to -1 to force the zfs_set_prop_nvlist code
+			 * down the default path to set the value in the nvlist.
+			 */
+			err = -1;
+		}
+		zfsvfs_rele(zfsvfs, FTAG);
+		break;
+	}
 	default:
 		err = -1;
 	}

module/zfs/zfs_znode.c

@@ -178,6 +178,7 @@ zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
 	dmu_buf_t *prevdb = NULL;
 	dmu_buf_t *sa_db = NULL;
 	char *path = buf + len - 1;
+	char *comp_buf;
 	int error;
 
 	*path = '\0';
@@ -193,9 +194,10 @@ zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
 		return (error);
 	}
 
+	comp_buf = kmem_alloc(ZAP_MAXNAMELEN_NEW + 2, KM_SLEEP);
 	for (;;) {
 		uint64_t pobj = 0;
-		char component[MAXNAMELEN + 2];
+		char *component = comp_buf;
 		size_t complen;
 		int is_xattrdir = 0;
 
@@ -219,7 +221,8 @@ zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
 			strcpy(component + 1, "<xattrdir>");
 		} else {
 			error = zap_value_search(osp, pobj, obj,
-			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
+			    ZFS_DIRENT_OBJ(-1ULL), component + 1,
+			    ZAP_MAXNAMELEN_NEW);
 			if (error != 0)
 				break;
 		}
@@ -250,6 +253,7 @@ zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
 	if (error == 0)
 		(void) memmove(buf, path, buf + len - path);
 
+	kmem_free(comp_buf, ZAP_MAXNAMELEN_NEW +2);
 	return (error);
 }