Cfxtrjtrk

I noticed that when I query the size of a device using open + lseek, everything is OK, but when I stat the device I get zero instead of the real device size. The device is clean without any file system and the first bytes of device start with some text like "1234567890ABC". What is wrong?

The code:

#include <sys/stat.h>

#include <dirent.h>



bool

GetFileSize(const char* pPath, uint64_t& Size)

{

    pPath = "/home/sw/.bashrc";

    pPath = "/dev/sda";



    struct stat buffer;

    if (stat(pPath, &buffer))

    {

        printf("Failed to stat file. Error: %s. FilePath: %sn", strerror(errno), pPath);

        return false;

    }



    printf("File size by stat: %" PRIu64 " WTF?n", buffer.st_size);



    //

    // Note: It's strange, but stat::st_size from the stat call is zero for devices

    //



    int File = open(pPath, O_RDONLY);

    if (File < 0)

    {

        printf("Failed to open file. Error: %s. FilePath: %sn", strerror(errno), pPath);

        return false;

    }



    long off = lseek(File, 0, SEEK_END);

    if (off == (off_t)-1)

    {

        printf("Failed to get file size. Error: %s. FilePath: %sn", strerror(errno), pPath);

        close(File);

        return false;

    }

    close(File);



    printf("File size by lseek: %" PRIu64 "n", off);

    fflush(stdout);



    Size = off;

    return true;

}

Output:

File size by stat: 0 WTF?

File size by lseek: 34359738368

If I use stat for a regular file then everything is OK (comment out the line with "/dev/sda"):

File size by stat: 4019 WTF?

File size by lseek: 4019

The devil is in the detail... For starters, there is the fundamental principle of Unix design: everything is a file, Nicely explained here.

The second is that the stat(2) call is giving you inode statistics stored on the filesystem about the device-special file which has a size of zero (think of it as lstat(2)). If you have a block-device that has a filesystem on it you get information about it using statfs(2) or getfsstat(2) or statvfs(2) in a filesystem/device independent way.

Dealing with special files (usually residing in /dev) has always been system specific and the manual pages reside in section 4. So if you want to manipulate a device directly you should read up on the specifics there. For instance, in Linux man 4 hd will show you how to programmatically interact with IDE block devices. Whereas man 4 sd will give you how to interact with scsi discs, etc.

Third thing, system calls are not supposed to be inconsistent in their functionality NOR their limitations.

Hope this has helped.

from this Unix Stack Exchange question:

Device files are not files per se. They're an I/O interface to use the devices in Unix-like operating systems. They use no space on disk, however, they still use an inode as reported by the stat command:

$ stat /dev/sda

      File: /dev/sda

      Size: 0               Blocks: 0          IO Block: 4096   block special file

Device: 6h/6d   Inode: 14628       Links: 1     Device type: 8,0

That solves the stat part.

the fact that you can seek in this "file" is not related. This isn't really a file, but you can open it and read from it. You can seek to it too. It allows to read the disk at the lowest level, so seeking is necessary (that's why it works, and why wouldn't it return the new position like any "real" file?).

According to this other UnixSE answer, you can get the device size by reading this /dev/sda/size file.

The length of a "device" such as /dev/sda is not specifed by the POSIX struct stat:

off_t st_size       For regular files, the file size in bytes. 



                    For symbolic links, the length in bytes of the 

                    pathname contained in the symbolic link. 



                    For a shared memory object, the length in bytes. 



                    For a typed memory object, the length in bytes. 



                    For other file types, the use of this field is 

                    unspecified. 

So POSIX has no requirement for the "size" of a disk device.

Linux likewise does not specify that stat() shall return the size of a disk device:

st_size

This field gives the size of the file (if it is a regular
file
or a symbolic link) in bytes. The size of a symbolic link is
the length of the pathname it contains, without a terminating
null byte.

On Linux, the documented way to get the size of a raw disk device that you can open is with the BLKGETSIZE ioctl. See the sd(4) manpage.

Note that this returns the size of the device in sectors. You might think that, for size in bytes, you have to multiply by the value returned by the BLKSSZGET ioctl, but if I'm reading the source code correctly, you actually have to multiply by 512 no matter what BLKSSZGET returns.

lseek is the backbone to C's fseek, so it has its own semantics, matching fseek - and quite detached from other areas of the Unix API. Provenance-wise, you'd expect lseek to act like fseek, and fseek is a C-library interface that came to be without being Unix-specific.

stat is Unix-specific, though, and does its own thing. It's a reasonable difference to expect if you think about provenance. Of course the problem is, then, that C APIs have very weak type models because C is one step short of making true type safety possible. As it is, you can be type-safe as long as it is possible to fit the value into intptr_t - because then its real type can be a pointer type to a struct "tag", and those are the only "small" type-safe values in C - and even then, only if you enable -Werror, i.e. turn all warnings into errors. Passing structs by value would have been nicer since there the type mismatches are hard errors, but initializing those is a pain:

typedef struct { const char *val; } file_path;

typedef struct { const char *val; } file_mode;

typedef FILE *file_handle;



file_path path = { "foo/bar" };

file_mode mode = { "w" };

file_handle fh = fopen(path, mode);

The path and mode arguments are type-safe, i.e. you couldn't pass a file_path to something that expects a date_string given as typedef struct { const char *val; } date_string. And although modern compilers do a wicked good job of optimizing the code above and essentially stripping the type-safe overhead, the ABIs lag behind this. And thus C is only mildly type-safe. Only with -Werror :)