d_type
is a speed optimization to save on lstat(2)
calls, when it's supported.
As the readdir
(3) man page points out, not all filesystems return real info in the d_type
field (typically because it would take an extra disk seek to read the inode, as is the case for XFS if you didn't use mkfs.xfs -n ftype=1
(implied by -m crc=1
which is not yet the default). Filesystems that always set DT_UNKNOWN
are common in real life, and not something that you can ignore. XFS is not the only example.
You always need code that will fall back to using lstat
(2) if d_type==DT_UNKNOWN
, if the filename alone isn't enough to decide it's uninteresting. (This is the case for some callers, like find -name
or expanding globs like *.c
, which is why readdir
doesn't incur the overhead of filling it in if it would take an extra disk read.)
The Linux getdents(2)
man page has an example program that does what you're trying to do, including a chained-ternary-operator block to decode the d_type
field into text strings. (As the other answers point out, your mistake is printing it out as an character, rather than comparing it against DT_REG
, DT_DIR
, etc.)
Anyway, the other answers mostly covered things, but missed the critical detail that you NEED a fallback for the case when d_type == DT_UNKNOWN
(0 on Linux. d_type
is stored in what used to be a padding byte, until Linux 2.6.4).
To be portable, your code needs to check that struct dirent
even HAS a d_type
field, if you use it, or your code won't even compile outside of GNU and BSD systems. (see readdir(3)
)
I wrote an example for finding directories with readdir, using d_type
with a fallback to stat
when d_type isn't available at compile time, when it's DT_UNKNOWN, and for symlinks.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…