implemenation of "df" command to find out free disk space information

Hi everyone,

df command can help to print out various disk space information for each file system and its underlying implementation relies on the statfs function.
I do not know why, in its implementation, the way of getting the available size of a file system is :

Code:

f_bavail x f_bsize / 1024 ;
// free blocks available to non-super user x block size of the file system / 1 Kbyte

And, why not use

Code:

f_bfree x f_bsize / 1024 
// total free blocks of the file system x a block size of the file system / 1 Kbyte

As the struct statfs defined, the f_bavail is the free block available to non-super user, and in the embedded (linux) world, most of the time, the system/program runs under the super user mode on the device. In this case, should we use f_bfree instead if we want to find out the total free space available of a file system?

Thanks~

Here's what I think is being referred to here:

On an ext2/3/4 filesystem -- which may or may not be the filesystem used in embedded devices, but it is on most normal linux boxes -- 5%* is "reserved for the super-user". This is supposedly to insure there will be space available to important privileged processes if for some reason the filesystem fills up.

AFAIK, it is not a good idea to run any OS in a filesytem that is more than 90% full -- it will result in significant, noticable performance degradation, so, most people don't do it. In reality, then, I would guess that 5% is never used -- which may be good, since that may reduce fragmentation (which man mke2fs claims).

Pretty sure it is these which are excluded from f_bavail. Based on what you present, it's still not totally clear that it is included in f_bfree, but in any case, IMO the idea is that those reserved super-user blocks are intended for an emergency and should never be intentionally used unless the other 95% is full (and such a situation would be an emergency). Ie, no don't include them in any kind of free or available space count -- either there is other space that should be used, or else these should still be left free for whatever critical operations that will need to occur either to shut the system down or free more space.

In fact, I'm sure of that, judging by the df manpage:

<total space>
The total size of the file system in 512-byte units. The exact
meaning of this figure is implementation-defined, but should
include <space used>, <space free>, plus any space reserved by
the system not normally available to a user.
<space free>
The total amount of space available within the file system for
the creation of new files by unprivileged users, in 512-byte
units. When this figure is less than or equal to zero, it shall
not be possible to create any new files on the file system with-
out first deleting others, unless the process has appropriate
privileges. The figure written may be less than zero.

By "not normally available to a user" I believe this includes normal activities of the super-user, Ie, even if there is only one user, the super-user, that user should not make use of the reserved space.

However, it is possible to create an ext2/3/4 filesystem with 0% reserved blocks (this is what I normally do with usb keys, mounted partitions, etc -- only the root filesystem has reserved space) which is probably the case in embedded environments. If there are no reserved blocks, the issue is moot.

*by default; you can set this amount when you create the filesystem

Hi MK27,

Thank you very much for your help. The information are very clear, informative, and to-the-point. Especially the illustration/analysis from the man page and your personal view. Thanks!

Whether to use f_bfree or f_bavail is a matter of your own policy. If you want to include the superuser blocks in the count, then include them. Otherwise, don't. It's a simple change to the source for df...

can u help me for "implementationof ls command in C"

Originally Posted by vanita

can u help me for "implementationof ls command in C"

Start here:
Accessing Directories - The GNU C Library