This is the program I used to check the behavior of SEEK_DATA and SEEK_HOLE.
#define _FILE_OFFSET_BITS 64
#define _LARGEFILE64_SOURCE 1
#include <sys/types.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <stdint.h>
#include <fcntl.h>
#include <errno.h>
#include <stdio.h>
/*
* This function is suitable for macOS and Solaris' lseek behavior,
* since start is always set to the beginning of a data region or a hole.
* Please see the logics in main()
*/
static int find_allocation(int fd, off_t start, off_t *data, off_t*hole)
{
off_t offs;
offs = lseek(fd, start, SEEK_DATA);
if (offs < 0) {
return -errno;
}
if (offs < start) {
return -EIO;
}
if (offs > start) {
/* D2: start is at the beginning of hole */
*hole = start;
*data = offs;
return 0;
}
/* D1: start is at beginning of data region */
offs = lseek(fd, start, SEEK_HOLE);
if (offs < 0) {
return -errno;
}
if (offs < start) {
return -EIO;
}
if (offs > start) {
*data = start;
*hole = offs;
return 0;
}
return -EIO;
}
int main(int argc, const char* argv[])
{
off_t first_data = -1, trailing_hole = -1;
off_t cursor = 0;
if (argc != 2) {
fprintf(stderr, "Usage: %s path\n", argv[0]);
return 1;
}
int fd = open(argv[1], O_RDONLY);
if (fd < 0) {
perror("Cannot open the file");
return 1;
}
off_t filesize = lseek(fd, 0, SEEK_END);
if (filesize < 0) {
perror("Cannot get the file size");
return 1;
}
if (filesize < 2) {
fprintf(stderr, "Filesize is too small.\n");
return 1;
}
printf ("Filesize: %lld\n", (long long)filesize);
while(cursor < filesize) {
off_t data, hole;
int res;
/* cursor is always at the beginning of a data region or a hole */
res = find_allocation(fd, cursor, &data, &hole);
if (res < 0 && res != -ENXIO) {
fprintf(stderr, "The filesystem or platform being checked does not
support SEEK_DATA or SEEK_HOLE.\n");
perror("");
return 1;
}
if (res == -ENXIO) {
/* we are at the trailing hole */
trailing_hole = cursor;
break;
}
if (data == cursor && (hole - data) > 1) {
/* the length of the data region must be greater than 1. */
if (first_data == -1) {
first_data = cursor;
}
}
if (data == cursor) {
cursor = hole;
} else {
cursor = data;
}
}
if (first_data >= 0) {
printf("Checking for SEEK_DATA by using the data region at %lld... ",
(long long)first_data);
/* first_data plus 1 makes the offset in the middle of a data region */
errno = 0;
off_t offs = lseek(fd, first_data + 1, SEEK_DATA);
if (offs < 0 && errno != ENXIO) {
printf("Error\n");
perror(" Msg");
} else if (errno == ENXIO || offs > first_data + 1) {
/* offs is set to the next data region. This is macOS's behavior */
printf("macOS\n");
} else if (offs == first_data + 1) {
printf("Linux\n");
} else {
printf("Unknown behavior\n");
}
} else {
fprintf(stderr, "There is no data region which is suitable to be
checked.\n");
}
if (trailing_hole >= 0) {
printf("Checking for SEEK_HOLE by using the trailing hole at %lld... ",
(long long)trailing_hole);
off_t offs = lseek(fd, trailing_hole, SEEK_HOLE);
if (offs < 0) {
printf("Error\n");
perror(" Msg");
} else if (offs == filesize) {
/* offs is set to EOF. This is Solaris' behavior */
printf("Solaris\n");
} else if (offs == trailing_hole) {
printf("Linux\n");
} else {
printf("Unknown behavior\n");
}
} else {
fprintf(stderr, "There is no trailing hole which is suitable to be
checked.\n");
}
return 0;
}
Peter Maydell - 2018/9/8 11:34 PM:
> On 8 September 2018 at 15:15, Yan-Jie Wang <jaywang0...@gmail.com> wrote:
>> In macOS, lseek with SEEK_DATA behaves differently.
>> It seeks to the next data region even though offset is in the middle of
>> a data region. In addition, there may be many data regions without any
>> hole among them, like this: |---Data---|---Data---|
>>
>> Because of this, qemu-img convert with raw images as input may create
>> corrupted images in macOS especially for large files, and qemu-img
>> map may also report wrong things. This patch fixes this undesired
>> behaviors.
>
> Hi. I have two general questions here:
> (1) is this behaviour of SEEK_DATA specific to macOS, or do the
> other BSDs (FreeBSD, OpenBSD, NetBSD) also have it ?
I have installed FreeBSD in Virtualbox and checked the behavior of lseek in
FreeBSD.
The behavior of SEEK_DATA is the same as the one in Linux.
> (2) is there a way to determine which flavour of SEEK_DATA we
> have as a configure-time test rather than having to hardcode
> an OS-specific #ifdef ?
macOS can be installed on HFS+ or APFS filesystem. Only APFS supports SEEK_DATA
and SEEK_HOLE.
If we try to build qemu on HFS+ filesystem, it is not possible to detect the
behavior
of SEEK_DATA and SEEK_HOLE on configure-time. lseek with SEEK_DATA or
SEEK_HOLE returns
errors when the file being checked is on HFS+ filesystem. (I have checked it by
formatting
my USB thumb drive to HFS+ filesystem and running the program provided at the
top of this email on the file located at the HFS+ filesystem on my thumb drive.)
>
> thanks
> -- PMM
>
