ZFS send fails to dump objects larger than 128PiB

When dumping objects larger than 128PiB it's possible for do_dump() to
miscalculate the FREE_RECORD offset due to an integer overflow
condition: this prevents the receiving end from correctly restoring
the dumped object.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6760

module/zfs/bpobj.c

@@ -261,7 +261,7 @@ bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
 	}
 	if (free) {
 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
-		    (i + 1) * sizeof (blkptr_t), -1ULL, tx));
+		    (i + 1) * sizeof (blkptr_t), DMU_OBJECT_END, tx));
 	}
 	if (err || !bpo->bpo_havesubobj || bpo->bpo_phys->bpo_subobjs == 0)
 		goto out;
@@ -339,7 +339,7 @@ bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
 	if (free) {
 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
 		    bpo->bpo_phys->bpo_subobjs,
-		    (i + 1) * sizeof (uint64_t), -1ULL, tx));
+		    (i + 1) * sizeof (uint64_t), DMU_OBJECT_END, tx));
 	}
 
 out:

module/zfs/dmu.c

@@ -967,7 +967,7 @@ dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
 	if (err)
 		return (err);
 	ASSERT(offset < UINT64_MAX);
-	ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
+	ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset);
 	dnode_free_range(dn, offset, size, tx);
 	dnode_rele(dn, FTAG);
 	return (0);

module/zfs/dmu_send.c

@@ -223,9 +223,6 @@ dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 	    (object == dsp->dsa_last_data_object &&
 	    offset > dsp->dsa_last_data_offset));
 
-	if (length != -1ULL && offset + length < offset)
-		length = -1ULL;
-
 	/*
 	 * If there is a pending op, but it's not PENDING_FREE, push it out,
 	 * since free block aggregation can only be done for blocks of the
@@ -242,19 +239,22 @@ dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 
 	if (dsp->dsa_pending_op == PENDING_FREE) {
 		/*
-		 * There should never be a PENDING_FREE if length is -1
-		 * (because dump_dnode is the only place where this
-		 * function is called with a -1, and only after flushing
-		 * any pending record).
+		 * There should never be a PENDING_FREE if length is
+		 * DMU_OBJECT_END (because dump_dnode is the only place where
+		 * this function is called with a DMU_OBJECT_END, and only after
+		 * flushing any pending record).
 		 */
-		ASSERT(length != -1ULL);
+		ASSERT(length != DMU_OBJECT_END);
 		/*
 		 * Check to see whether this free block can be aggregated
 		 * with pending one.
 		 */
 		if (drrf->drr_object == object && drrf->drr_offset +
 		    drrf->drr_length == offset) {
-			drrf->drr_length += length;
+			if (offset + length < offset)
+				drrf->drr_length = DMU_OBJECT_END;
+			else
+				drrf->drr_length += length;
 			return (0);
 		} else {
 			/* not a continuation.  Push out pending record */
@@ -268,9 +268,12 @@ dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 	dsp->dsa_drr->drr_type = DRR_FREE;
 	drrf->drr_object = object;
 	drrf->drr_offset = offset;
-	drrf->drr_length = length;
+	if (offset + length < offset)
+		drrf->drr_length = DMU_OBJECT_END;
+	else
+		drrf->drr_length = length;
 	drrf->drr_toguid = dsp->dsa_toguid;
-	if (length == -1ULL) {
+	if (length == DMU_OBJECT_END) {
 		if (dump_record(dsp, NULL, 0) != 0)
 			return (SET_ERROR(EINTR));
 	} else {
@@ -587,7 +590,7 @@ dump_dnode(dmu_sendarg_t *dsp, const blkptr_t *bp, uint64_t object,
 
 	/* Free anything past the end of the file. */
 	if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
-	    (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
+	    (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), DMU_OBJECT_END) != 0)
 		return (SET_ERROR(EINTR));
 	if (dsp->dsa_err != 0)
 		return (SET_ERROR(EINTR));
@@ -771,7 +774,9 @@ do_dump(dmu_sendarg_t *dsa, struct send_block_record *data)
 	} else if (BP_IS_HOLE(bp)) {
 		uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level);
 		uint64_t offset = zb->zb_blkid * span;
-		err = dump_free(dsa, zb->zb_object, offset, span);
+		/* Don't dump free records for offsets > DMU_OBJECT_END */
+		if (zb->zb_blkid == 0 || span <= DMU_OBJECT_END / zb->zb_blkid)
+			err = dump_free(dsa, zb->zb_object, offset, span);
 	} else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 		return (0);
 	} else if (type == DMU_OT_DNODE) {
@@ -2860,7 +2865,7 @@ receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf)
 {
 	int err;
 
-	if (drrf->drr_length != -1ULL &&
+	if (drrf->drr_length != DMU_OBJECT_END &&
 	    drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
 		return (SET_ERROR(EINVAL));
 

tests/runfiles/linux.run

@@ -167,7 +167,8 @@ tests = ['zfs_rollback_001_pos', 'zfs_rollback_002_pos',
 [tests/functional/cli_root/zfs_send]
 tests = ['zfs_send_001_pos', 'zfs_send_002_pos', 'zfs_send_003_pos',
     'zfs_send_004_neg', 'zfs_send_005_pos', 'zfs_send_006_pos',
-    'zfs_send_007_pos', 'zfs_send_encrypted', 'zfs_send_raw']
+    'zfs_send_007_pos', 'zfs_send_encrypted', 'zfs_send_raw',
+    'zfs_send_sparse']
 
 [tests/functional/cli_root/zfs_set]
 tests = ['cache_001_pos', 'cache_002_neg', 'canmount_001_pos',

tests/zfs-tests/tests/functional/cli_root/zfs_send/Makefile.am

@@ -11,4 +11,5 @@ dist_pkgdata_SCRIPTS = \
 	zfs_send_006_pos.ksh \
 	zfs_send_007_pos.ksh \
 	zfs_send_encrypted.ksh \
-	zfs_send_raw.ksh
+	zfs_send_raw.ksh \
+	zfs_send_sparse.ksh

tests/zfs-tests/tests/functional/cli_root/zfs_send/zfs_send_sparse.ksh

@@ -0,0 +1,83 @@
+#!/bin/ksh -p
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright 2017, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+#
+# DESCRIPTION:
+# 'zfs send' should be able to send (big) sparse files correctly.
+#
+# STRATEGY:
+# 1. Create sparse files of various size
+# 2. Snapshot and send these sparse files
+# 3. Verify these files are received correctly and we don't trigger any issue
+#    like the one described in https://github.com/zfsonlinux/zfs/pull/6760
+#
+
+verify_runnable "both"
+
+function cleanup
+{
+        datasetexists $SENDFS && log_must zfs destroy -r $SENDFS
+        datasetexists $RECVFS && log_must zfs destroy -r $RECVFS
+}
+
+#
+# Write 1 random byte at $offset of "source" file in $sendfs dataset
+# Snapshot and send $sendfs dataset to $recvfs
+# Compare the received file with its source
+#
+function write_compare_files # <sendfs> <recvfs> <offset>
+{
+	typeset sendfs="$1"
+	typeset recvfs="$2"
+	typeset offset="$3"
+
+	# create source filesystem
+	log_must zfs create $sendfs
+	# write sparse file
+	sendfile="$(get_prop mountpoint $sendfs)/data.bin"
+	log_must dd if=/dev/urandom of=$sendfile bs=1 count=1 seek=$offset
+	# send/receive the file
+	log_must zfs snapshot $sendfs@snap
+	log_must eval "zfs send $sendfs@snap | zfs receive $recvfs"
+	# compare sparse files
+	recvfile="$(get_prop mountpoint $recvfs)/data.bin"
+	log_must cmp $sendfile $recvfile $offset $offset
+	sendsz=$(stat -c '%s' $sendfile)
+	recvsz=$(stat -c '%s' $recvfile)
+	if [[ $sendsz -ne $recvsz ]]; then
+		log_fail "$sendfile ($sendsz) and $recvfile ($recvsz) differ."
+	fi
+	# cleanup
+	log_must zfs destroy -r $sendfs
+	log_must zfs destroy -r $recvfs
+}
+
+log_assert "'zfs send' should be able to send (big) sparse files correctly."
+log_onexit cleanup
+
+SENDFS="$TESTPOOL/sendfs"
+RECVFS="$TESTPOOL/recvfs"
+OFF_T_MAX="$(echo '2 ^ 40 * 8 - 1' | bc)"
+
+for i in {1..60}; do
+	offset=$(echo "2 ^ $i" | bc)
+	[[ is_32bit ]] && [[ $offset -ge $OFF_T_MAX ]] && continue;
+	write_compare_files $SENDFS $RECVFS $offset
+done
+
+log_pass "'zfs send' sends (big) sparse files correctly."