fdtools.c - OpenGrok cross reference for /dports/dns/yadifa/yadifa-2.5.3-10333/lib/dnscore/src/fdtools.c

/*------------------------------------------------------------------------------
 *
 * Copyright (c) 2011-2021, EURid vzw. All rights reserved.
 * The YADIFA TM software product is provided under the BSD 3-clause license:
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 *        * Redistributions of source code must retain the above copyright
 *          notice, this list of conditions and the following disclaimer.
 *        * Redistributions in binary form must reproduce the above copyright
 *          notice, this list of conditions and the following disclaimer in the
 *          documentation and/or other materials provided with the distribution.
 *        * Neither the name of EURid nor the names of its contributors may be
 *          used to endorse or promote products derived from this software
 *          without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 *
 *------------------------------------------------------------------------------
 *
 */

/** @defgroup dnscoretools Generic Tools
 *  @ingroup dnscore
 *  @brief
 *
 * @{
 */

#include "dnscore/dnscore-config.h"
#include <stdlib.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <dirent.h>
#include <sys/socket.h>

#define FDTOOLS_C_ 1

#include "dnscore/fdtools.h"
#include "dnscore/zalloc.h"
#include "dnscore/ptr_set.h"
#include "dnscore/timems.h"
#include "dnscore/logger.h"
#include "dnscore/mutex.h"

 /* GLOBAL VARIABLES */

extern logger_handle *g_system_logger;
#define MODULE_MSG_HANDLE g_system_logger

#if DEBUG
#define DEBUG_FD_OPEN_CLOSE_MONITOR 1
#else
#define DEBUG_FD_OPEN_CLOSE_MONITOR 0
#endif

#define FDTRACK_TAG 0x4b434152544446
#define MTIMESET_TAG 0x544553454d49544d

#if DEBUG
// avoids logging these operations in the logger.
// this prevents a self-deadlock if the logger limit is reached (2^20 lines)
bool logger_is_self();
#endif

#if DEBUG
static void
admin_warn(const char *pathname)
{
    uid_t uid = getuid();
    uid_t euid = getuid();
    gid_t gid = getgid();
    gid_t egid = getegid();

    bool is_admin = (uid==0)||(euid==0)||(gid==0)||(egid==0);

    if(is_admin)
    {
        printf("It is unlikely file or directory creation should be made as an admin '%s' (DEBUG)\n", pathname);
        fflush(NULL);
    }
}
#endif

#if DEBUG_FD_OPEN_CLOSE_MONITOR

// file descriptor track enabled

#include "dnscore/u32_set.h"
#include "dnscore/zalloc.h"

static group_mutex_t fd_mtx = GROUP_MUTEX_INITIALIZER;

static u32_set fd_to_name = U32_SET_EMPTY;

struct fd_track
{
    stacktrace opener;
    stacktrace closer;
    char name[256];
};

typedef struct fd_track fd_track;

static void fd_set_name(int fd, const char *name)
{
    fd_track *track;

    group_mutex_lock(&fd_mtx, GROUP_MUTEX_WRITE);

    log_debug6("file descriptor: %i is '%s'", fd, name);

    u32_node *node = u32_set_insert(&fd_to_name, fd);

    if(node->value == NULL)
    {
        ZALLOC_OBJECT_OR_DIE(track, fd_track, FDTRACK_TAG);
        node->value = track;
        track->opener = 0;
        track->closer = 0;
        track->name[0] = '\0';
    }
    else
    {
        track = (fd_track*)node->value;
        if(track->name[0] != '\0')
        {
            log_err("file descriptor: %i was associated with '%s'", fd, track->name);
        }
    }

    strcpy_ex(track->name, name, sizeof(track->name));
    track->opener = debug_stacktrace_get();
    track->name[sizeof(track->name) - 1] = '\0';

    group_mutex_unlock(&fd_mtx, GROUP_MUTEX_WRITE);
}

static void fd_clear_name(int fd)
{
    group_mutex_lock(&fd_mtx, GROUP_MUTEX_WRITE);

    u32_node *node = u32_set_find(&fd_to_name, fd);
    if(node != NULL)
    {
        yassert(node->value != NULL);

        fd_track *track = (fd_track*)node->value;
        stacktrace st = debug_stacktrace_get();

        if(track->name[0] != '\0')
        {
            log_debug6("file descriptor: %i is '%s' no more", fd, track->name);
        }
        else
        {
            log_debug6("file descriptor: %i is being closed by", fd);
            debug_stacktrace_log(MODULE_MSG_HANDLE, MSG_DEBUG6, st);
            log_debug6("file descriptor: %i was closed already by", fd);
            debug_stacktrace_log(MODULE_MSG_HANDLE, MSG_DEBUG6, track->closer);
            log_debug6("file descriptor: %i was last opened by", fd);
            debug_stacktrace_log(MODULE_MSG_HANDLE, MSG_DEBUG6, track->opener);
        }

        track->closer = st;

        track->name[0] = '\0';
    }
    else
    {
        stacktrace st = debug_stacktrace_get();

        log_debug6("file descriptor: %i is untracked and is being closed by", fd);
        debug_stacktrace_log(MODULE_MSG_HANDLE, MSG_DEBUG6, st);
    }

    group_mutex_unlock(&fd_mtx, GROUP_MUTEX_WRITE);
}
#endif

/**
 * Writes fully the buffer to the fd
 * It will only return a short count for system errors.
 * ie: fs full, non-block would block, fd invalid/closed, ...
 */

ssize_t
writefully(int fd, const void *buf, size_t count)
{
    const u8* start = (const u8*)buf;
    const u8* current = start;
    ssize_t n;

    while(count > 0)
    {
        if((n = write(fd, current, count)) <= 0)
        {
            if(n == 0)
            {
                break;
            }

            int err = errno;

            if(err == EINTR)
            {
                continue;
            }

            if(err == EAGAIN) /** @note It is nonsense to call writefully with a non-blocking fd */
            {
                if(current - start > 0)
                {
                    break;
                }

                return MAKE_ERRNO_ERROR(ETIMEDOUT);
            }

            if(err == ENOSPC)
            {
                // the disk is full : wait a bit, hope the admin catches it, try again later
                sleep((rand()&7) + 1);
                continue;
            }

            if(current - start > 0)
            {
                break;
            }

            return MAKE_ERRNO_ERROR(err);
        }

        current += n;
        count -= n;
    }

    return current - start;
}

/**
 * Reads fully the buffer from the fd
 * It will only return a short count for system errors.
 * ie: fs full, non-block would block, fd invalid/closed, ...
 */

ssize_t
readfully(int fd, void *buf, size_t length)
{
    u8* start = (u8*)buf;
    u8* current = start;
    ssize_t n;

    while(length > 0)
    {
        if((n = read(fd, current, length)) <= 0)
        {
            if(n == 0) // end of file
            {
                break;
            }

            int err = errno;

            if(err == EINTR)
            {
                continue;
            }

            if(err == EAGAIN) /** @note It is nonsense to call readfully with a non-blocking fd */
            {
                break;
            }

            if(current - start > 0)
            {
                break;
            }

            return -1;
        }

        current += n;
        length -= n;
    }

    return current - start;
}

/**
 * Writes fully the buffer to the fd
 * It will only return a short count for system errors.
 * ie: fs full, non-block would block, fd invalid/closed, ...
 */

ssize_t
writefully_limited(int fd, const void *buf, size_t count, double minimum_rate_us)
{
    const u8* start = (const u8*)buf;
    const u8* current = start;
    ssize_t n;

    u64 tstart = timeus();

    // ASSUMED : timeout set on fd for read & write

    while(count > 0)
    {
        if((n = write(fd, current, count)) <= 0)
        {
            if(n == 0)
            {
                break;
            }

            int err = errno;

            if(err == EINTR)
            {
                continue;
            }

            if(err == EAGAIN)
            {
                /*
                 * Measure the current elapsed time
                 * Measure the current bytes
                 * compare with the minimum rate
                 * act on it
                 */

                /* t is in us */

                u64 now = timeus();

                u64 time_elapsed_u64 = now - tstart;

                if(time_elapsed_u64 >= ONE_SECOND_US)
                {
                    double time_elapsed_us = time_elapsed_u64;

                    double bytes_written = (current - (u8*)buf)  * ONE_SECOND_US_F;

                    double expected_bytes_written = minimum_rate_us * time_elapsed_us;

                    if(bytes_written < expected_bytes_written)  /* bytes/time < minimum_rate */
                    {
#if DEBUG
                        log_warn("writefully_limited: rate of %fBps < %fBps (%fµs)", bytes_written, expected_bytes_written, time_elapsed_us);
#else
                        log_debug("writefully_limited: rate of %fBps < %fBps (%fµs)", bytes_written, expected_bytes_written, time_elapsed_us);
#endif
                        return TCP_RATE_TOO_SLOW;
                    }
                }

                continue;
            }

            if(err == ENOSPC)
            {
                // the disk is full : wait a bit, hope the admin catches it, try again later
                sleep((rand()&7) + 1);
                continue;
            }

            if(current - start > 0)
            {
                break;
            }

            return -1;
        }

        current += n;
        count -= n;
    }

    return current - start;
}

/**
 * Reads fully the buffer from the fd
 * It will only return a short count for system errors.
 * ie: fs full, non-block would block, fd invalid/closed, ...
 */

ssize_t
readfully_limited(int fd, void *buf, size_t count, double minimum_rate_us)
{
    u8* start = (u8*)buf;
    u8* current = start;
    ssize_t n;

    u64 tstart = timeus();

    // ASSUME : timeout set on fd for read & write

    while(count > 0)
    {
        if((n = read(fd, current, count)) <= 0)
        {
            if(n == 0)
            {
                break;
            }

            int err = errno;

            if(err == EINTR)
            {
                continue;
            }

            if(err == EAGAIN) /** @note It is nonsense to call readfully with a non-blocking fd */
            {
                /*
                 * Measure the current elapsed time
                 * Measure the current bytes
                 * compare with the minimum rate
                 * act on it
                 */

                u64 now = timeus();

                /* t is in us */

                u64 time_elapsed_u64 = now - tstart;

                double time_elapsed_us = time_elapsed_u64;

                if(time_elapsed_u64 >= ONE_SECOND_US)
                {
                    double bytes_read = (current - (u8*)buf) * ONE_SECOND_US_F;

                    double expected_bytes_read = minimum_rate_us * time_elapsed_us;

                    if(bytes_read < expected_bytes_read)  // bytes/time < minimum_rate
                    {
                        time_elapsed_us /= 1000000.0;
#if DEBUG
                        log_warn("readfully_limited: rate of %fBps < %fBps (%fµs) (DEBUG)", bytes_read, expected_bytes_read, time_elapsed_us);
#else
                        log_debug("readfully_limited: rate of %fBps < %fBps (%fµs)", bytes_read, expected_bytes_read, time_elapsed_us);
#endif
                        return TCP_RATE_TOO_SLOW;
                    }
                }

                continue;
            }

            if(current - start > 0)
            {
                break;
            }

            return -1;  /* EOF */
        }

        current += n;
        count -= n;
    }

    return current - start;
}

/**
 * Reads fully the buffer from the fd
 * It will only return a short count for system errors.
 * ie: fs full, non-block would block, fd invalid/closed, ...
 */

ssize_t
readfully_limited_ex(int fd, void *buf, size_t count, s64 timeout_us, double minimum_rate_us)
{
    u8* start = (u8*)buf;
    u8* current = start;
    ssize_t n;

    if(timeout_us <= 0)
    {
        timeout_us = 1;
    }

    s64 tstart = timeus();

    // ASSUME : timeout set on fd for read & write

    while(count > 0)
    {
        if((n = read(fd, current, count)) <= 0)
        {
            if(n == 0)
            {
                break;
            }

            int err = errno;

            if(err == EINTR)
            {
                continue;
            }

            if(err == EAGAIN) /** @note It is nonsense to call readfully with a non-blocking fd */
            {
                /*
                 * Measure the current elapsed time
                 * Measure the current bytes
                 * compare with the minimum rate
                 * act on it
                 */

                s64 now = timeus();

                /* t is in us */

                s64 time_elapsed_u64 = now - tstart;

                double time_elapsed_us = time_elapsed_u64;

                if(time_elapsed_u64 >= timeout_us)
                {
                    double bytes_read = (current - (u8*)buf);
                    bytes_read *= ONE_SECOND_US_F;
                    bytes_read /= time_elapsed_u64;

                    double expected_bytes_read = minimum_rate_us * time_elapsed_us;

                    if(bytes_read < expected_bytes_read)  // bytes/time < minimum_rate
                    {
                        time_elapsed_us /= 1000000.0;
#if DEBUG
                        log_warn("readfully_limited: rate of %fBps < %fBps (%fµs) (DEBUG)", bytes_read, expected_bytes_read, time_elapsed_us);
#else
                        log_debug("readfully_limited: rate of %fBps < %fBps (%fµs)", bytes_read, expected_bytes_read, time_elapsed_us);
#endif
                        return TCP_RATE_TOO_SLOW;
                    }
                }

                continue;
            }

            if(current - start > 0)
            {
                break;
            }

            return -1;  /* EOF */
        }

        current += n;
        count -= n;
    }

    return current - start;
}

/**
 * Reads an ASCII text line from fd, stops at EOF or '\n'
 */

ssize_t
readtextline(int fd, char *start, size_t count)
{
    char *current = start;
    const char * const limit = &start[count];

    while(current < limit)
    {
        ssize_t n;

        if((n = read(fd, current, 1)) > 0)
        {
            u8 c = *current;

            current++;
            count --;

            if(c == '\n')
            {
                break;
            }
        }
        else
        {
            if(n == 0)
            {
                break;
            }

            int err = errno;

            if(err == EINTR)
            {
                continue;
            }

            if(err == EAGAIN)
            {
                continue;
            }

            if(current - start > 0)
            {
                break;
            }

            return -1;
        }
    }

    return current - start;
}

/**
 * Deletes a file (see man 2 unlink).
 * Handles EINTR and other retry errors.
 * Safe to use in the logger thread as it only logs (debug) if the current
 * thread is not the logger's
 *
 * @param fd
 * @return
 */

int
unlink_ex(const char *folder, const char *filename)
{
    char fullpath[PATH_MAX];

    size_t l0 = strlen(folder);
    size_t l1 = strlen(filename);
    if(l0 + l1 < sizeof(fullpath))
    {
        memcpy(fullpath, folder, l0);
        fullpath[l0] = '/';
        memcpy(&fullpath[l0 + 1], filename, l1);
        fullpath[l0 + 1 + l1] = '\0';

        return unlink(fullpath);
    }
    else
    {
        errno = ENOMEM;
        return -1;
    }
}

/**
 * Copies the absolute path of a file into a buffer.
 *
 * @param filename the file name
 * @param buffer the output buffer
 * @param buffer_size the size of the output buffer
 * @return the string length (without the terminator)
 */

ya_result
file_get_absolute_path(const char *filename, char *buffer, size_t buffer_size)
{
    ya_result ret;

    if(filename[0] == '/')
    {
        strcpy_ex(buffer, filename, buffer_size);
        ret = strlen(buffer);
        return ret;
    }
    else
    {
        if(getcwd(buffer, buffer_size) == NULL)
        {
            return ERRNO_ERROR;
        }

        size_t n = strlen(buffer);

        if(n < buffer_size)
        {
            ret = n + 1;

            buffer += n;
            buffer_size -= n;

            *buffer++ = '/';
            --buffer_size;

            n = strlen(filename);
            if(n < buffer_size)
            {
                memcpy(buffer, filename, n);
                buffer[n] = '\0';

                return ret + n;
            }
        }

        return ERROR; // not enough room
    }
}

#if DEBUG_BENCH_FD
static debug_bench_s debug_open;
static debug_bench_s debug_open_create;
static debug_bench_s debug_close;
static bool fdtools_debug_bench_register_done = FALSE;

static inline void fdtools_debug_bench_register()
{
    if(!fdtools_debug_bench_register_done)
    {
        fdtools_debug_bench_register_done = TRUE;
        debug_bench_register(&debug_open, "open");
        debug_bench_register(&debug_open_create, "open_create");
        debug_bench_register(&debug_close, "close");
    }
}
#endif

/**
 * Opens a file. (see man 2 open)
 * Handles EINTR and other retry errors.
 * Safe to use in the logger thread as it only logs (debug) if the current
 * thread is not the logger's
 *
 * @param fd
 * @return
 */

ya_result
open_ex(const char *pathname, int flags)
{
    int fd;

#if DEBUG
    if(!logger_is_self() && logger_is_running())
    {
        log_debug6("open_ex(%s,%o)", STRNULL(pathname), flags);
        errno = 0;
    }
#endif

    yassert(pathname != NULL);

#if DEBUG_BENCH_FD
    fdtools_debug_bench_register();
    u64 bench = debug_bench_start(&debug_open);
#endif

#if DNSCORE_FDTOOLS_CLOEXEC
    bool cloexec = (flags & O_CLOEXEC) != 0;
    flags &= ~O_CLOEXEC;
#endif

    while((fd = open(pathname, flags)) < 0)
    {
        int err = errno;

        if(err != EINTR)
        {
            break;
        }
    }

#if DNSCORE_FDTOOLS_CLOEXEC
    if((fd >= 0) && cloexec)
    {
        ya_result ret;

        if(FAIL(ret = fd_setcloseonexec(fd)))
        {
            log_warn("open_ex(%s,%o): failed to set CLOEXEC: %r", STRNULL(pathname), flags|O_CLOEXEC, ret);
        }
    }
#endif

#if DEBUG_BENCH_FD
    debug_bench_stop(&debug_open, bench);
#endif

#if DEBUG
    if(!logger_is_self() && logger_is_running())
    {
        log_debug6("open_ex(%s,%o): %r (%i)", STRNULL(pathname), flags, ERRNO_ERROR, fd);
#if DEBUG_FD_OPEN_CLOSE_MONITOR
        if(fd > 0)
        {
            fd_set_name(fd, pathname);
        }
#endif
    }
#endif

    return fd;
}

/**
 * Opens a file, create if it does not exist. (see man 2 open with O_CREAT)
 * Handles EINTR and other retry errors.
 * Safe to use in the logger thread as it only logs (debug) if the current
 * thread is not the logger's
 *
 * @param fd
 * @return
 */

ya_result
open_create_ex(const char *pathname, int flags, mode_t mode)
{
    int fd;

#if DEBUG
    if(!logger_is_self() && logger_is_running())
    {
        log_debug6("open_create_ex(%s,%o,%o)", STRNULL(pathname), flags, mode);
        errno = 0;
    }
#endif

#if DEBUG
    admin_warn(pathname);
#endif

    yassert(pathname != NULL);

#if DEBUG_BENCH_FD
    fdtools_debug_bench_register();
    u64 bench = debug_bench_start(&debug_open_create);
#endif

    while((fd = open(pathname, flags, mode)) < 0)
    {
        int err = errno;

        if(err != EINTR)
        {
            // do NOT set this to an error code other than -1
            //fd = MAKE_ERRNO_ERROR(err);
            break;
        }
    }

#if DEBUG_BENCH_FD
    debug_bench_stop(&debug_open_create, bench);
#endif

#if DEBUG
    if(!logger_is_self() && logger_is_running())
    {
        log_debug6("open_create_ex(%s,%o,%o): %r (%i)", STRNULL(pathname), flags, mode, ERRNO_ERROR, fd);
#if DEBUG_FD_OPEN_CLOSE_MONITOR
        if(fd > 0)
        {
            fd_set_name(fd, pathname);
        }
#endif
    }
#endif

    return fd;
}

int mkstemp_ex(char * tmp_name_template)
{
#ifndef WIN32
    int fd = mkstemp(tmp_name_template);
#if DEBUG
    if(!logger_is_self() && logger_is_running())
    {
#if DEBUG_FD_OPEN_CLOSE_MONITOR
        if(fd > 0)
        {
            fd_set_name(fd, tmp_name_template);
        }
#endif
    }
#endif
    return fd;
#else
    return -1;
#endif
}

/**
 * Opens a file, create if it does not exist. (see man 2 open with O_CREAT)
 * Handles EINTR and other retry errors.
 * This version of open_create_ex does NOT log anything, which is very important sometimes in the logger thread
 *
 * @param fd
 * @return
 */

ya_result
open_create_ex_nolog(const char *pathname, int flags, mode_t mode)
{
    int fd;
#if DEBUG
    admin_warn(pathname);
#endif
    while((fd = open(pathname, flags, mode)) < 0)
    {
        int err = errno;

        if(err != EINTR)
        {
            // do NOT set this to an error code other than -1
            //fd = MAKE_ERRNO_ERROR(err);
            break;
        }
    }

    return fd;
}

/**
 * Closes a file descriptor (see man 2 close)
 * Handles EINTR and other retry errors.
 * At return the file will be closed or not closable.
 *
 * @param fd
 * @return
 */

ya_result
#if !DNSCORE_HAS_CLOSE_EX_REF
close_ex(int fd)
#else
close_ex_ref(int* fdp)
#endif
{
    ya_result return_value = SUCCESS;

#if DNSCORE_HAS_CLOSE_EX_REF
    int fd = *fdp;

    if(fd == -2)
    {
        abort();
    }

    *fdp = -2;
#endif

#if DEBUG
    if(!logger_is_self() && logger_is_running())
    {
#if DEBUG_FD_OPEN_CLOSE_MONITOR
        if(fd > 0)
        {
            fd_clear_name(fd);
        }
#endif

        log_debug6("close_ex(%i)", fd);
        errno = 0;
    }
#endif

#if DEBUG_BENCH_FD
    fdtools_debug_bench_register();
    u64 bench = debug_bench_start(&debug_close);
#endif

    while(close(fd) < 0)
    {
        int err = errno;

        if(err != EINTR)
        {
            return_value = MAKE_ERRNO_ERROR(err);
            break;
        }
    }
#if DEBUG_BENCH_FD
    debug_bench_stop(&debug_close, bench);
#endif

#if DEBUG
    if(!logger_is_self() && logger_is_running())
    {
        log_debug6("close_ex(%i): %r", fd, return_value);
        if(FAIL(return_value))
        {
            logger_flush();
        }
    }
#endif

    return return_value;
}

/**
 * Closes a file descriptor (see man 2 close)
 * Handles EINTR and other retry errors.
 * At return the file will be closed or not closable.
 *
 * @param fd
 * @return
 */

ya_result
#if !DNSCORE_HAS_CLOSE_EX_REF
close_ex_nolog(int fd)
#else
close_ex_nolog_ref(int* fdp)
#endif
{
    ya_result return_value = SUCCESS;

#if DNSCORE_HAS_CLOSE_EX_REF
    int fd = *fdp;

    if(fd == -2)
    {
        abort();
    }

    *fdp = -2;
#endif

    while(close(fd) < 0)
    {
        int err = errno;

        if(err != EINTR)
        {
            return_value = MAKE_ERRNO_ERROR(err);
            break;
        }
    }

    return return_value;
}

int
fsync_ex(int fd)
{
#ifndef WIN32
    while(fsync(fd) < 0)
    {
        int err = errno;
        if(err != EINTR)
        {
            return ERRNO_ERROR;
        }
    }
#else
    FlushFileBuffers(fd);
#endif
    return SUCCESS;
}

int
fdatasync_ex(int fd)
{
#ifndef WIN32
#if defined(__linux__)
    while(fdatasync(fd) < 0)
#else
    while(fsync(fd) < 0)
#endif
    {
        int err = errno;
        if(err != EINTR)
        {
            return ERRNO_ERROR;
        }
    }
#else
    FlushFileBuffers(fd);
#endif
    return SUCCESS;
}

int dup_ex(int fd)
{
    int ret;
    while((ret = dup(fd)) < 0)
    {
        int err = errno;
        if(err != EINTR)
        {
            return ERRNO_ERROR;
        }
    }

    return ret;
}

int dup2_ex(int old_fd, int new_fd)
{
    int ret;
    while((ret = dup2(old_fd, new_fd)) < 0)
    {
        int err = errno;
        if(err != EINTR)
        {
            return ERRNO_ERROR;
        }
    }

    return ret;
}

int truncate_ex(const char *path, off_t len)
{
    int ret;
    while((ret = truncate(path, len)) < 0)
    {
        int err = errno;
        if(err != EINTR)
        {
            return ERRNO_ERROR;
        }
    }

    return ret;
}

int ftruncate_ex(int fd, off_t len)
{
    int ret;
    while((ret = ftruncate(fd, len)) < 0)
    {
        int err = errno;
        if(err != EINTR)
        {
            return ERRNO_ERROR;
        }
    }

    return ret;
}

/**
 * Returns the type of socket.
 *
 * @param fd the file descriptor of the socket
 * @return SOCK_STREAM, SOCK_DGRAM, SOCK_RAW or an errno error code like MAKE_ERRON_ERROR(EBADF) or MAKE_ERRON_ERROR(ENOTSOCK)
 */

ya_result
fd_getsockettype(int fd)
{
    int stype;
    socklen_t stype_len = sizeof(stype);
    int ret = getsockopt(fd, SOL_SOCKET, SO_TYPE, &stype, &stype_len);
    if(ret >= 0)
    {
        return stype; // SOCK_STREAM, SOCK_DGRAM, SOCK_RAW, ...
    }
    else
    {
        return ERRNO_ERROR; // expecting EBADF or ENOTSOCK
    }
}

s64
filesize(const char *name)
{
    struct stat s;
    if(stat(name, &s) >= 0) // MUST be stat
    {
        if(S_ISREG(s.st_mode))
        {
            return s.st_size;
        }
    }

    return (s64)ERRNO_ERROR;
}

/**
 * Checks for existence of a file/dir/link
 *
 * @param name the file name
 *
 * @return 1 if the file exists, 0 if the file does not exists, an error otherwise
 */

ya_result
file_exists(const char *name)
{
    struct stat s;
#ifndef WIN32
    if(lstat(name, &s) >= 0)    // MUST be lstat
    {
        return 1;
    }
#else
    if(stat(name, &s) >= 0)    // MUST be lstat
    {
        return 1;
    }
#endif

    int err = errno;

    if(err == ENOENT)
    {
        return 0;
    }

    return MAKE_ERRNO_ERROR(err);
}

/**
 *
 * Checks if a file exists and is a link
 *
 * @param name the file name
 *
 * @return  0 : not a link
 *          1 : a link
 *        < 0 : error
 */

ya_result
file_is_link(const char *name)
{
#ifndef WIN32
    struct stat s;
    if(lstat(name, &s) >= 0)    // MUST be lstat
    {
        return S_ISLNK(s.st_mode)?SUCCESS:ERROR;
    }

    return ERRNO_ERROR;
#else
    return 0;
#endif
}

/**
 *
 * Checks if a file exists and is a directory
 *
 * @param name the file name
 *
 * @return  0 : not a link
 *          1 : a link
 *        < 0 : error
 */

ya_result
file_is_directory(const char *name)
{
#ifndef WIN32
    struct stat s;
    if(lstat(name, &s) >= 0)    // MUST be lstat
    {
        return S_ISDIR(s.st_mode)?SUCCESS:ERROR;
    }

    return ERRNO_ERROR;
#else
    return 0;
#endif
}

/**
 *
 * Creates all directories on pathname.
 *
 * Could be optimised a bit :
 *
 *      try the biggest path first,
 *      going down until it works,
 *      then create back up.
 *
 * @param pathname
 * @param mode
 * @param flags
 *
 * @return
 */

int
mkdir_ex(const char *pathname, mode_t mode, u32 flags)
{
#if DEBUG
    log_debug("mkdir_ex(%s,%o)", pathname, mode);
#endif

#if DEBUG
    admin_warn(pathname);
#endif

    const char *s;
    char *t;

    char dir_path[PATH_MAX];

    s = pathname;
    t = dir_path;

    if(pathname[0] == '/')
    {
        t[0] = '/';
        t++;
        s++;
    }

    for(;;)
    {
        const char *p = (const char*)strchr(s, '/');

        bool last = (p == NULL);

        if(last)
        {
            if((flags & MKDIR_EX_PATH_TO_FILE) != 0)
            {
                return s - pathname;
            }

            p = s + strlen(s);
        }

        intptr n = (p - s);
        memcpy(t, s, n);
        t[n] = '\0';

        struct stat file_stat;
        if(stat(dir_path, &file_stat) < 0)
        {
            int err = errno;

            if(err != ENOENT)
            {
#if DEBUG
                log_debug("mkdir_ex(%s,%o): stat returned %r", pathname, mode, MAKE_ERRNO_ERROR(err));
#endif

                return MAKE_ERRNO_ERROR(err);
            }

            if(mkdir(dir_path, mode) < 0)
            {
#if DEBUG
                log_debug("mkdir_ex(%s,%o): mkdir(%s, %o) returned %r", pathname, mode, dir_path, mode, MAKE_ERRNO_ERROR(err));
#endif

                return ERRNO_ERROR;
            }
        }

        if(last)
        {
            s = &s[n];
            return s - pathname;
        }

        t[n++] = '/';

        t = &t[n];
        s = &s[n];
    }
}

/**
 * Returns the modification time of the file in microseconds
 * This does not mean the precision of the time is that high.
 * This is only to simplify reading the time on a file.
 *
 * @param name the file name
 * @param timestamp a pointer to the timestamp
 * @return an error code
 */

ya_result
file_mtime(const char *name, s64 *timestamp)
{
    struct stat st;
    yassert(name != NULL);
    yassert(timestamp != NULL);
    if(stat(name, &st) >= 0)
    {
#ifdef WIN32
        s64 ts = ONE_SECOND_US * st.st_mtime;
#elif !__APPLE__
        s64 ts = (ONE_SECOND_US * st.st_mtim.tv_sec) + (st.st_mtim.tv_nsec / 1000LL);
#else
        s64 ts = (ONE_SECOND_US * st.st_mtimespec.tv_sec) + (st.st_mtimespec.tv_nsec / 1000LL);
#endif
        *timestamp = ts;
        return SUCCESS;
    }
    else
    {
        *timestamp = 0;
        return ERRNO_ERROR;
    }
}

/**
 * Returns the modification time of the file in microseconds
 * This does not mean the precision of the time is that high.
 * This is only to simplify reading the time on a file.
 *
 * @param name the file name
 * @param timestamp a pointer to the timestamp
 * @return an error code
 */

ya_result
fd_mtime(int fd, s64 *timestamp)
{
    struct stat st;
    yassert(timestamp != NULL);
    if(fstat(fd, &st) >= 0)
    {
#ifdef WIN32
        s64 ts = ONE_SECOND_US * st.st_mtime;
#elif !__APPLE__
        s64 ts = (ONE_SECOND_US * st.st_mtim.tv_sec) + (st.st_mtim.tv_nsec / 1000LL);
#else
        s64 ts = (ONE_SECOND_US * st.st_mtimespec.tv_sec) + (st.st_mtimespec.tv_nsec / 1000LL);
#endif
        *timestamp = ts;
        return SUCCESS;
    }
    else
    {
        *timestamp = 0;
        return ERRNO_ERROR;
    }
}

ya_result
fd_setcloseonexec(int fd)
{
#ifndef WIN32
    int ret = fcntl(fd, F_SETFD, FD_CLOEXEC);
    if(FAIL(ret))
    {
        ret = ERRNO_ERROR;
    }
    return ret;
#else
    return SUCCESS;
#endif
}

ya_result
fd_setnonblocking(int fd)
{
#ifndef WIN32
    int ret;
    if(ISOK(ret = fcntl(fd, F_GETFL, 0)))
    {
        fcntl(fd, F_SETFL, ret | O_NONBLOCK);
    }
    else
    {
        ret = ERRNO_ERROR;
    }

    return ret;
#else
    return SUCCESS;
#endif
}

/**
 * Fixes an issue with the dirent not always set as expected.
 *
 * The type can be set to DT_UNKNOWN instead of file or directory.
 * In that case the function will call stats to get the type.
 */

u8
dirent_get_type_from_fullpath(const char *fullpath)
{
    struct stat file_stat;
    u8 d_type;

    d_type = DT_UNKNOWN;

    while(stat(fullpath, &file_stat) < 0)
    {
        int e = errno;

        if(e != EINTR)
        {
            log_err("stat(%s): %r", fullpath, ERRNO_ERROR);
            break;
        }
    }

    if(S_ISREG(file_stat.st_mode))
    {
        d_type = DT_REG;
    }
    else if(S_ISDIR(file_stat.st_mode))
    {
        d_type = DT_DIR;
    }

    return d_type;
}

u8
dirent_get_file_type(const char *folder, const char *name)
{
    u8 d_type;
    char fullpath[PATH_MAX];

    d_type = DT_UNKNOWN;

    /*
     * If the FS OR the OS does not support d_type :
     */

    if(ISOK(snprintf(fullpath, sizeof(fullpath), "%s/%s", folder, name)))
    {
        d_type = dirent_get_type_from_fullpath(fullpath);
    }

    return d_type;
}

// typedef ya_result readdir_callback(const char *basedir, const char* file, u8 filetype, void *args);

static group_mutex_t readdir_mutex = GROUP_MUTEX_INITIALIZER;

ya_result
readdir_forall(const char *basedir, readdir_callback *func, void *args)
{
    DIR *dir;
    ya_result ret;
    size_t basedir_len = strlen(basedir);
    char *name;
    char path[PATH_MAX];
    memcpy(path, basedir, basedir_len);
    path[basedir_len] = '/';
    name = &path[basedir_len + 1];

    dir = opendir(basedir);

    if(dir == NULL)
    {
        return ERRNO_ERROR;
    }

    for(;;)
    {
        group_mutex_lock(&readdir_mutex, GROUP_MUTEX_WRITE);
        struct dirent *tmp = readdir(dir);

        if(tmp == NULL)
        {
            group_mutex_unlock(&readdir_mutex, GROUP_MUTEX_WRITE);

            ret = SUCCESS;

            break;
        }

        // ignore names "." and ".."

#ifndef WIN32
        const char* tmp_name = tmp->d_name;
#else
        const char* tmp_name = tmp->name;
#endif

        if(tmp_name[0] == '.')
        {
            if(
                ((tmp_name[1] == '.') && (tmp_name[2] == '\0')) ||
                (tmp_name[1] == '\0')
                )
            {
                group_mutex_unlock(&readdir_mutex, GROUP_MUTEX_WRITE);
                continue;
            }
        }

        size_t name_len = strlen(tmp_name);

        if(name_len + basedir_len + 1 + 1 > sizeof(path))
        {
            group_mutex_unlock(&readdir_mutex, GROUP_MUTEX_WRITE);
            log_err("readdir_forall: '%s/%s' is bigger than expected (%i): skipping", basedir, tmp_name, sizeof(path));
            continue;
        }

        memcpy(name, tmp_name, name_len + 1);

        group_mutex_unlock(&readdir_mutex, GROUP_MUTEX_WRITE);

        u8 d_type = dirent_get_type_from_fullpath(path);

        if(FAIL(ret = func(basedir, name, d_type, args)))
        {
            return ret;
        }

        switch(ret)
        {
            case READDIR_CALLBACK_CONTINUE:
            {
                break;
            }
            case READDIR_CALLBACK_ENTER:
            {
                if(d_type == DT_DIR)
                {
                    if(FAIL(ret = readdir_forall(path, func, args)))
                    {
                        return ret;
                    }

                    if(ret == READDIR_CALLBACK_EXIT)
                    {
                        return ret;
                    }
                }
                break;
            }
            case READDIR_CALLBACK_EXIT:
            {
                return ret;
            }
            default:
            {
                // unhandled code
                break;
            }
        }
    }

    closedir(dir);

    return ret;
}

struct file_mtime_set_s
{
    ptr_set files_mtime;
    char *name;
    bool is_new;
};

typedef struct file_mtime_set_s file_mtime_set_t;

static ptr_set file_mtime_sets = { NULL, ptr_set_asciizp_node_compare};
static mutex_t file_mtime_sets_mtx;

file_mtime_set_t*
file_mtime_set_get_for_file(const char *filename)
{
    file_mtime_set_t *ret;
    mutex_lock(&file_mtime_sets_mtx);
    ptr_node *sets_node = ptr_set_insert(&file_mtime_sets, (char*)filename);
    if(sets_node->value != NULL)
    {
        ret = (file_mtime_set_t*)sets_node->value;
    }
    else
    {
        sets_node->key = strdup(filename);
        ZALLOC_OBJECT_OR_DIE(ret, file_mtime_set_t, MTIMESET_TAG);
        ret->files_mtime.root = NULL;
        ret->files_mtime.compare = ptr_set_asciizp_node_compare;
        ret->name = strdup(filename);
        ret->is_new = TRUE;
        sets_node->value = ret;
        file_mtime_set_add_file(ret, filename);
    }
    mutex_unlock(&file_mtime_sets_mtx);
    return ret;
}

void
file_mtime_set_add_file(file_mtime_set_t *ctx, const char *filename)
{
    s64 mtime;

    if(FAIL(file_mtime(filename, &mtime)))
    {
        mtime = MIN_S64;
    }

    ptr_node *node = ptr_set_insert(&ctx->files_mtime, (char*)filename);
    if(node->value == NULL)
    {
        node->key = strdup(filename);
        node->value_s64 = mtime;
    }
}

bool
file_mtime_set_modified(file_mtime_set_t *ctx)
{
    if(ctx->is_new)
    {
        ctx->is_new = FALSE;
        return TRUE;
    }

    ptr_set_iterator iter;
    ptr_set_iterator_init(&ctx->files_mtime, &iter);
    while(ptr_set_iterator_hasnext(&iter))
    {
        ptr_node *node = ptr_set_iterator_next_node(&iter);
        const char *filename = (const char*)node->key;
        s64 mtime;
        if(ISOK(file_mtime(filename, &mtime)))
        {
            if(node->value_s64 < mtime)
            {
                return TRUE;
            }
        }
        else
        {
            return TRUE;
        }
    }
    return FALSE;
}

void
file_mtime_set_clear(file_mtime_set_t *ctx)
{
    ptr_set_iterator iter;
    ptr_set_iterator_init(&ctx->files_mtime, &iter);
    while(ptr_set_iterator_hasnext(&iter))
    {
        ptr_node *node = ptr_set_iterator_next_node(&iter);
        free(node->key);
    }
    ptr_set_destroy(&ctx->files_mtime);
    file_mtime_set_add_file(ctx, ctx->name);
}

void
file_mtime_set_delete(file_mtime_set_t *ctx)
{
    mutex_lock(&file_mtime_sets_mtx);
    ptr_set_delete(&file_mtime_sets, ctx->name);
    mutex_unlock(&file_mtime_sets_mtx);
    free(ctx->name);
    file_mtime_set_clear(ctx);
    ZFREE_OBJECT(ctx);
}

/** @} */