kvm tools: Fix large disk images on 32-bit

Use read()/write() instead of mmap() for virtio block device emulation to
support large disk images on 32-bit.

Reported-and-tested-by: Asias He <asias.hejun@gmail.com>
Tested-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>

Makefile

@@ -15,6 +15,7 @@ TAGS = ctags
 OBJS	+= 8250-serial.o
 OBJS	+= blk-virtio.o
 OBJS	+= cpuid.o
+OBJS	+= read-write.o
 OBJS	+= disk-image.o
 OBJS	+= interrupt.o
 OBJS	+= ioport.o
@@ -40,6 +41,9 @@ ifeq ($(uname_M),x86_64)
 endif
 endif
 
+DEFINES	+= -D_FILE_OFFSET_BITS=64
+DEFINES	+= -D_GNU_SOURCE
+
 CFLAGS	+= $(CPPFLAGS) $(DEFINES) -Iinclude -I../../include -Os -g
 
 WARNINGS += -Werror

disk-image.c

@@ -1,5 +1,6 @@
 #include "kvm/disk-image.h"
 
+#include "kvm/read-write.h"
 #include "kvm/util.h"
 
 #include <sys/types.h>
@@ -12,19 +13,6 @@
 #include <unistd.h>
 #include <fcntl.h>
 
-static const char QCOW_MAGIC[]	= { 'Q', 'F', 'I', 0xfb };
-
-struct qcow_header {
-	uint8_t			magic[5];
-};
-
-static bool disk_image__is_qcow(struct disk_image *self)
-{
-	struct qcow_header *header = self->mmap;
-
-	return !memcmp(header->magic, QCOW_MAGIC, ARRAY_SIZE(QCOW_MAGIC));
-}
-
 struct disk_image *disk_image__open(const char *filename)
 {
 	struct disk_image *self;
@@ -34,7 +22,7 @@ struct disk_image *disk_image__open(const char *filename)
 	if (!self)
 		return NULL;
 
-	self->fd	= open(filename, O_RDONLY);
+	self->fd	= open(filename, O_RDWR);
 	if (self->fd < 0)
 		goto failed_free;
 
@@ -43,13 +31,6 @@ struct disk_image *disk_image__open(const char *filename)
 
 	self->size	= st.st_size;
 
-	self->mmap	= mmap(NULL, self->size, PROT_READ|PROT_WRITE, MAP_PRIVATE, self->fd, 0);
-	if (self->mmap == MAP_FAILED)
-		goto failed_close_fd;
-
-	if (disk_image__is_qcow(self))
-		die("QCOW disk image format is not supported.");
-
 	return self;
 
 failed_close_fd:
@@ -62,9 +43,6 @@ failed_free:
 
 void disk_image__close(struct disk_image *self)
 {
-	if (munmap(self->mmap, self->size) < 0)
-		warning("munmap() failed");
-
 	if (close(self->fd) < 0)
 		warning("close() failed");
 
@@ -78,7 +56,11 @@ int disk_image__read_sector(struct disk_image *self, uint64_t sector, void *dst,
 	if (offset + dst_len > self->size)
 		return -1;
 
-	memcpy(dst, self->mmap + offset, dst_len);
+	if (lseek(self->fd, offset, SEEK_SET) < 0)
+		return -1;
+
+	if (read_in_full(self->fd, dst, dst_len) < 0)
+		return -1;
 
 	return 0;
 }
@@ -90,7 +72,11 @@ int disk_image__write_sector(struct disk_image *self, uint64_t sector, void *src
 	if (offset + src_len > self->size)
 		return -1;
 
-	memcpy(self->mmap + offset, src, src_len);
+	if (lseek(self->fd, offset, SEEK_SET) < 0)
+		return -1;
+
+	if (write_in_full(self->fd, src, src_len) < 0)
+		return -1;
 
 	return 0;
 }

include/kvm/disk-image.h

@@ -7,7 +7,6 @@
 #define SECTOR_SIZE		(1UL << SECTOR_SHIFT)
 
 struct disk_image {
-	void		*mmap;
 	int		fd;
 	uint64_t	size;
 };

include/kvm/read-write.h

@@ -0,0 +1,13 @@
+#ifndef KVM_READ_WRITE_H
+#define KVM_READ_WRITE_H
+
+#include <sys/types.h>
+#include <unistd.h>
+
+ssize_t xread(int fd, void *buf, size_t count);
+ssize_t xwrite(int fd, const void *buf, size_t count);
+
+ssize_t read_in_full(int fd, void *buf, size_t count);
+ssize_t write_in_full(int fd, const void *buf, size_t count);
+
+#endif /* KVM_READ_WRITE_H */

read-write.c

@@ -0,0 +1,79 @@
+#include "kvm/read-write.h"
+
+#include <sys/types.h>
+#include <unistd.h>
+#include <string.h>
+#include <errno.h>
+
+/* Same as read(2) except that this function never returns EAGAIN or EINTR. */
+ssize_t xread(int fd, void *buf, size_t count)
+{
+	ssize_t nr;
+
+restart:
+	nr = read(fd, buf, count);
+	if ((nr < 0) && ((errno == EAGAIN) || (errno == EINTR)))
+		goto restart;
+
+	return nr;
+}
+
+/* Same as write(2) except that this function never returns EAGAIN or EINTR. */
+ssize_t xwrite(int fd, const void *buf, size_t count)
+{
+	ssize_t nr;
+
+restart:
+	nr = write(fd, buf, count);
+	if ((nr < 0) && ((errno == EAGAIN) || (errno == EINTR)))
+		goto restart;
+
+	return nr;
+}
+
+ssize_t read_in_full(int fd, void *buf, size_t count)
+{
+	ssize_t total = 0;
+	char *p = buf;
+
+	while (count > 0) {
+		ssize_t nr;
+
+		nr = xread(fd, p, count);
+		if (nr <= 0) {
+			if (total > 0)
+				return total;
+
+			return -1;
+		}
+
+		count -= nr;
+		total += nr;
+		p += nr;
+	}
+
+	return total;
+}
+
+ssize_t write_in_full(int fd, const void *buf, size_t count)
+{
+	const char *p = buf;
+	ssize_t total = 0;
+
+	while (count > 0) {
+		ssize_t nr;
+
+		nr = xwrite(fd, p, count);
+		if (nr < 0)
+			return -1;
+		if (nr == 0) {
+			errno = ENOSPC;
+			return -1;
+		}
+		count -= nr;
+		total += nr;
+		p += nr;
+	}
+
+	return total;
+}