Increase Linux pipe buffer on zfs recv size to the maximum system size

I noticed when reviewing documentation that it is possible for userspace
to use fctnl(fd, F_SETPIPE_SZ, (unsigned long) size) to change the
kernel pipe buffer size on Linux to increase the pipe size up to the
value specified in /proc/sys/fs/pipe-max-size. There are users using
mbuffer to improve zfs recv performance when piping over the network, so
it seems advantageous to integrate such functionality directly into the
zfs recv tool. This could have been configurable or we could have
changed the value back to the original (had we read it) after we were
done with the file descriptor,but I do not see a strong case for doing
either, so I went with a simple implementation.

Closes openzfs#1161

Signed-off-by: Richard Yao <[email protected]>

lib/libzfs/libzfs_sendrecv.c

@@ -43,6 +43,7 @@
 #include <sys/mnttab.h>
 #include <sys/avl.h>
 #include <sys/debug.h>
+#include <sys/stat.h>
 #include <stddef.h>
 #include <pthread.h>
 #include <umem.h>
@@ -3293,11 +3294,33 @@ int
 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, recvflags_t *flags,
     int infd, avl_tree_t *stream_avl)
 {
+	struct stat sb;
 	char *top_zfs = NULL;
 	int err;
 	int cleanup_fd;
 	uint64_t action_handle = 0;
 
+	if (fstat(infd, &sb) == -1) {
+		perror("fstat");
+		return (-2);
+	}
+
+	/*
+	 * It is not uncommon for gigabytes to be processed in zfs receive.
+	 * Speculatively increase the buffer size via Linux-specific fcntl()
+	 * call.
+	 */
+	if (S_ISFIFO(sb.st_mode)) {
+		FILE *procf = fopen("/proc/sys/fs/pipe-max-size", "r");
+
+		if (procf != NULL) {
+			unsigned long pipe_max_size;
+			fscanf(procf, "%lu", &pipe_max_size);
+			(void) fcntl(infd, F_SETPIPE_SZ, pipe_max_size);
+			fclose(procf);
+		}
+	}
+
 	cleanup_fd = open(ZFS_DEV, O_RDWR);
 	VERIFY(cleanup_fd >= 0);