Fix noop receive of raw send stream

Currently, the noop receive code fails to work with raw send streams
and resuming send streams. This happens because zfs_receive_impl()
reads the DRR_BEGIN payload without reading the payload itself.
Normally, the kernel expects to read this itself, but in this case
the recv_skip() code runs instead and it is not prepared to handle
the stream being left at any place other than the beginning of a
record.

This patch resolves this issue by manually reading the DRR_BEGIN
payload in the dry-run case. This patch also includes a number of
small fixups in this code path.

Fixes: #9173

Signed-off-by: Tom Caputi <[email protected]>

lib/libzfs/libzfs_sendrecv.c

@@ -4042,10 +4042,11 @@ recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
 {
 	dmu_replay_record_t *drr;
 	void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
+	uint64_t payload_size;
 	char errbuf[1024];
 
 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
-	    "cannot receive:"));
+	    "cannot receive"));
 
 	/* XXX would be great to use lseek if possible... */
 	drr = buf;
@@ -4072,9 +4073,14 @@ recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
 				drr->drr_u.drr_object.drr_bonuslen =
 				    BSWAP_32(drr->drr_u.drr_object.
 				    drr_bonuslen);
+				drr->drr_u.drr_object.drr_raw_bonuslen =
+				    BSWAP_32(drr->drr_u.drr_object.
+				    drr_raw_bonuslen);
 			}
-			(void) recv_read(hdl, fd, buf,
-			    P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
+
+			payload_size =
+			    DRR_OBJECT_PAYLOAD_SIZE(&drr->drr_u.drr_object);
+			(void) recv_read(hdl, fd, buf, payload_size,
 			    B_FALSE, NULL);
 			break;
 
@@ -4087,7 +4093,7 @@ recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
 				    BSWAP_64(
 				    drr->drr_u.drr_write.drr_compressed_size);
 			}
-			uint64_t payload_size =
+			payload_size =
 			    DRR_WRITE_PAYLOAD_SIZE(&drr->drr_u.drr_write);
 			(void) recv_read(hdl, fd, buf,
 			    payload_size, B_FALSE, NULL);
@@ -4096,9 +4102,15 @@ recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
 			if (byteswap) {
 				drr->drr_u.drr_spill.drr_length =
 				    BSWAP_64(drr->drr_u.drr_spill.drr_length);
+				drr->drr_u.drr_spill.drr_compressed_size =
+				    BSWAP_64(drr->drr_u.drr_spill.
+				    drr_compressed_size);
 			}
-			(void) recv_read(hdl, fd, buf,
-			    drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
+
+			payload_size =
+			    DRR_SPILL_PAYLOAD_SIZE(&drr->drr_u.drr_spill);
+			(void) recv_read(hdl, fd, buf, payload_size,
+			    B_FALSE, NULL);
 			break;
 		case DRR_WRITE_EMBEDDED:
 			if (byteswap) {
@@ -4841,6 +4853,21 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
 	}
 
 	if (flags->dryrun) {
+		void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
+
+		/*
+		 * We have read the DRR_BEGIN record, but we have
+		 * not yet read the payload. For non-dryrun sends
+		 * this will be done by the kernel, so we must
+		 * emulate that here, before attempting to read
+		 * more records.
+		 */
+		err = recv_read(hdl, infd, buf, drr->drr_payloadlen,
+		    flags->byteswap, NULL);
+		free(buf);
+		if (err != 0)
+			goto out;
+
 		err = recv_skip(hdl, infd, flags->byteswap);
 		goto out;
 	}

tests/zfs-tests/tests/functional/cli_root/zfs_receive/zfs_receive_raw.ksh

@@ -36,6 +36,7 @@
 # 9. Verify the key is unavailable
 # 10. Attempt to load the key and mount the dataset
 # 11. Verify the checksum of the file is the same as the original
+# 12. Verify 'zfs receive -n' works with the raw stream
 #
 
 verify_runnable "both"
@@ -88,4 +89,6 @@ typeset cksum2=$(md5digest /$TESTPOOL/$TESTFS1/c1/$TESTFILE0)
 [[ "$cksum2" == "$checksum" ]] || \
 	log_fail "Checksums differ ($cksum2 != $checksum)"
 
+log_must eval "zfs send -w $snap | zfs receive -n $TESTPOOL/$TESTFS3"
+
 log_pass "ZFS can receive streams from raw sends"