How can I create directory tree in C++/Linux?

Question

Easy with Boost.Filesystem: create_directories

#include <boost/filesystem.hpp>
//...
boost::filesystem::create_directories("/tmp/a/b/c");

Returns: true if a new directory was created, otherwise false.

With C++17 or later, there's the standard header <filesystem> with function std::filesystem::create_directories which should be used in modern C++ programs. The C++ standard functions do not have the POSIX-specific explicit permissions (mode) argument, though.

However, here's a C function that can be compiled with C++ compilers.

/*
@(#)File:           mkpath.c
@(#)Purpose:        Create all directories in path
@(#)Author:         J Leffler
@(#)Copyright:      (C) JLSS 1990-2020
@(#)Derivation:     mkpath.c 1.16 2020/06/19 15:08:10
*/

/*TABSTOP=4*/

#include "posixver.h"
#include "mkpath.h"
#include "emalloc.h"

#include <errno.h>
#include <string.h>
/* "sysstat.h" == <sys/stat.h> with fixup for (old) Windows - inc mode_t */
#include "sysstat.h"

typedef struct stat Stat;

static int do_mkdir(const char *path, mode_t mode)
{
    Stat            st;
    int             status = 0;

    if (stat(path, &st) != 0)
    {
        /* Directory does not exist. EEXIST for race condition */
        if (mkdir(path, mode) != 0 && errno != EEXIST)
            status = -1;
    }
    else if (!S_ISDIR(st.st_mode))
    {
        errno = ENOTDIR;
        status = -1;
    }

    return(status);
}

/**
** mkpath - ensure all directories in path exist
** Algorithm takes the pessimistic view and works top-down to ensure
** each directory in path exists, rather than optimistically creating
** the last element and working backwards.
*/
int mkpath(const char *path, mode_t mode)
{
    char           *pp;
    char           *sp;
    int             status;
    char           *copypath = STRDUP(path);

    status = 0;
    pp = copypath;
    while (status == 0 && (sp = strchr(pp, '/')) != 0)
    {
        if (sp != pp)
        {
            /* Neither root nor double slash in path */
            *sp = '\0';
            status = do_mkdir(copypath, mode);
            *sp = '/';
        }
        pp = sp + 1;
    }
    if (status == 0)
        status = do_mkdir(path, mode);
    FREE(copypath);
    return (status);
}

#ifdef TEST

#include <stdio.h>
#include <unistd.h>

/*
** Stress test with parallel running of mkpath() function.
** Before the EEXIST test, code would fail.
** With the EEXIST test, code does not fail.
**
** Test shell script
** PREFIX=mkpath.$$
** NAME=./$PREFIX/sa/32/ad/13/23/13/12/13/sd/ds/ww/qq/ss/dd/zz/xx/dd/rr/ff/ff/ss/ss/ss/ss/ss/ss/ss/ss
** : ${MKPATH:=mkpath}
** ./$MKPATH $NAME &
** [...repeat a dozen times or so...]
** ./$MKPATH $NAME &
** wait
** rm -fr ./$PREFIX/
*/

int main(int argc, char **argv)
{
    int             i;

    for (i = 1; i < argc; i++)
    {
        for (int j = 0; j < 20; j++)
        {
            if (fork() == 0)
            {
                int rc = mkpath(argv[i], 0777);
                if (rc != 0)
                    fprintf(stderr, "%d: failed to create (%d: %s): %s\n",
                            (int)getpid(), errno, strerror(errno), argv[i]);
                exit(rc == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
            }
        }
        int status;
        int fail = 0;
        while (wait(&status) != -1)
        {
            if (WEXITSTATUS(status) != 0)
                fail = 1;
        }
        if (fail == 0)
            printf("created: %s\n", argv[i]);
    }
    return(0);
}

#endif /* TEST */

The macros STRDUP() and FREE() are error-checking versions of strdup() and free(), declared in emalloc.h (and implemented in emalloc.c and estrdup.c). The "sysstat.h" header deals with broken versions of <sys/stat.h> and can be replaced by <sys/stat.h> on modern Unix systems (but there were many issues back in 1990). And "mkpath.h" declares mkpath().

The change between v1.12 (original version of the answer) and v1.13 (amended version of the answer) was the test for EEXIST in do_mkdir(). This was pointed out as necessary by Switch — thank you, Switch. The test code has been upgraded and reproduced the problem on a MacBook Pro (2.3GHz Intel Core i7, running Mac OS X 10.7.4), and suggests that the problem is fixed in the revision (but testing can only show the presence of bugs, never their absence). The code shown is now v1.16; there have been cosmetic or administrative changes made since v1.13 (such as use mkpath.h instead of jlss.h and include <unistd.h> unconditionally in the test code only). It's reasonable to argue that "sysstat.h" should be replaced by <sys/stat.h> unless you have an unusually recalcitrant system.

(You are hereby given permission to use this code for any purpose with attribution.)

This code is available in my SOQ (Stack Overflow Questions) repository on GitHub as files mkpath.c and mkpath.h (etc.) in the src/so-0067-5039 sub-directory.

system("mkdir -p /tmp/a/b/c")

is the shortest way I can think of (in terms of the length of code, not necessarily execution time).

It's not cross-platform but will work under Linux.

Here is my example of code (it works for both Windows and Linux):

#include <iostream>
#include <string>
#include <sys/stat.h> // stat
#include <errno.h>    // errno, ENOENT, EEXIST
#if defined(_WIN32)
#include <direct.h>   // _mkdir
#endif

bool isDirExist(const std::string& path)
{
#if defined(_WIN32)
    struct _stat info;
    if (_stat(path.c_str(), &info) != 0)
    {
        return false;
    }
    return (info.st_mode & _S_IFDIR) != 0;
#else 
    struct stat info;
    if (stat(path.c_str(), &info) != 0)
    {
        return false;
    }
    return (info.st_mode & S_IFDIR) != 0;
#endif
}

bool makePath(const std::string& path)
{
#if defined(_WIN32)
    int ret = _mkdir(path.c_str());
#else
    mode_t mode = 0755;
    int ret = mkdir(path.c_str(), mode);
#endif
    if (ret == 0)
        return true;

    switch (errno)
    {
    case ENOENT:
        // parent didn't exist, try to create it
        {
            int pos = path.find_last_of('/');
            if (pos == std::string::npos)
#if defined(_WIN32)
                pos = path.find_last_of('\\');
            if (pos == std::string::npos)
#endif
                return false;
            if (!makePath( path.substr(0, pos) ))
                return false;
        }
        // now, try to create again
#if defined(_WIN32)
        return 0 == _mkdir(path.c_str());
#else 
        return 0 == mkdir(path.c_str(), mode);
#endif

    case EEXIST:
        // done!
        return isDirExist(path);

    default:
        return false;
    }
}

int main(int argc, char* ARGV[])
{
    for (int i=1; i<argc; i++)
    {
        std::cout << "creating " << ARGV[i] << " ... " << (makePath(ARGV[i]) ? "OK" : "failed") << std::endl;
    }
    return 0;
}

Usage:

$ makePath 1/2 folderA/folderB/folderC
creating 1/2 ... OK
creating folderA/folderB/folderC ... OK

#include <sys/types.h>
#include <sys/stat.h>

int status;
...
status = mkdir("/tmp/a/b/c", S_IRWXU | S_IRWXG | S_IROTH | S_IXOTH);

From here. You may have to do separate mkdirs for /tmp, /tmp/a, /tmp/a/b/ and then /tmp/a/b/c because there isn't an equivalent of the -p flag in the C api. Be sure and ignore the EEXISTS errno while you're doing the upper level ones.

It should be noted that starting from C++17 filesystem interface is part of the standard library. This means that one can have following to create directories:

#include <filesystem>

std::filesystem::create_directories("/a/b/c/d")

More info here: https://en.cppreference.com/w/cpp/filesystem/create_directory

Additionally, with gcc, one needs to "-std=c++17" to CFLAGS. And "-lstdc++fs" to LDLIBS. The latter potentially is not going to be required in the future.

This is similar to the previous but works forward through the string instead of recursively backwards. Leaves errno with the right value for last failure. If there's a leading slash, there's an extra time through the loop which could have been avoided via one find_first_of() outside the loop or by detecting the leading / and setting pre to 1. The efficiency is the same whether we get set up by a first loop or a pre loop call, and the complexity would be (slightly) higher when using the pre-loop call.

#include <iostream>
#include <string>
#include <sys/stat.h>

int
mkpath(std::string s,mode_t mode)
{
    size_t pos=0;
    std::string dir;
    int mdret;

    if(s[s.size()-1]!='/'){
        // force trailing / so we can handle everything in loop
        s+='/';
    }

    while((pos=s.find_first_of('/',pos))!=std::string::npos){
        dir=s.substr(0,pos++);
        if(dir.size()==0) continue; // if leading / first time is 0 length
        if((mdret=mkdir(dir.c_str(),mode)) && errno!=EEXIST){
            return mdret;
        }
    }
    return mdret;
}

int main()
{
    int mkdirretval;
    mkdirretval=mkpath("./foo/bar",0755);
    std::cout << mkdirretval << '\n';

}