snapraid/os/unix.h

// SPDX-License-Identifier: GPL-3.0-or-later
// Copyright (C) 2026 Andrea Mazzoleni

#ifndef __UNIX_H
#define __UNIX_H

#ifdef __linux__
#define HAVE_LINUX_DEVICE 1 /**< In Linux enables special device support. */
#define HAVE_DIRECT_IO 1 /**< Support O_DIRECT in open(). */
#include <sys/syscall.h>
#include <sched.h>
#endif

#define SYSLOG "syslog"

#define OS_LVL_CRITICAL LOG_CRIT
#define OS_LVL_ERROR LOG_ERR
#define OS_LVL_WARNING LOG_WARNING
#define OS_LVL_INFO LOG_INFO
#define OS_LVL_DEBUG LOG_DEBUG

#if !HAVE_STRUCT_DIRENT_D_TYPE
/* without d_type these values are internal markers populated from lstat(). */
#ifndef DT_UNKNOWN
#define DT_UNKNOWN 0
#endif
#ifndef DT_DIR
#define DT_DIR 4
#endif
#ifndef DT_REG
#define DT_REG 8
#endif
#ifndef DT_LNK
#define DT_LNK 10
#endif
#endif

/* implement close_range for glibc 2.33 or earlier */
#if defined(__linux__) && !defined(HAVE_CLOSE_RANGE)
#ifndef __NR_close_range
#define __NR_close_range 436
#endif
#define close_range close_range_impl
static inline int close_range_impl(unsigned int first, unsigned int last, unsigned int flags)
{
	return (int)syscall(__NR_close_range, first, last, flags);
}
#define HAVE_CLOSE_RANGE 1
#endif

#ifndef WEXITSTATUS
#define WEXITSTATUS(stat_val) ((unsigned)(stat_val) >> 8)
#endif
#ifndef WIFEXITED
#define WIFEXITED(stat_val) (((stat_val) & 255) == 0)
#endif

#define FOPEN_TEXT ""
#define FOPEN_BINARY ""

#define O_BINARY 0 /**< Not used in Unix. */
#define O_SEQUENTIAL 0 /**< In Unix posix_fadvise() shall be used. */

/**
 * If nanoseconds are not supported, we report the special STAT_NSEC_INVALID value,
 * to mark that it's undefined.
 */
#define STAT_NSEC_INVALID -1

/* check if we have nanoseconds support */
#if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
#define STAT_NSEC(st) ((int)(st)->st_mtim.tv_nsec) /* Linux */
#elif HAVE_STRUCT_STAT_ST_MTIMENSEC
#define STAT_NSEC(st) ((int)(st)->st_mtimensec) /* NetBSD */
#elif HAVE_STRUCT_STAT_ST_MTIMESPEC_TV_NSEC
#define STAT_NSEC(st) ((int)(st)->st_mtimespec.tv_nsec) /* FreeBSD, Mac OS X */
#else
#define STAT_NSEC(st) STAT_NSEC_INVALID
#endif

#if HAVE_POSIX_FADVISE
/**
 * Wrapper around posix_fadvise() that handles ENOENT by converting it to
 * ENOSYS.
 *
 * Why can posix_fadvise() return ENOENT?
 *
 * According to POSIX and the Linux kernel's standard fadvise64 implementation
 * (mm/fadvise.c), posix_fadvise() should only return EBADF (bad file
 * descriptor), ESPIPE (fd refers to a pipe or FIFO), or EINVAL (invalid
 * advice value or arguments). ENOENT is never generated by the kernel's
 * generic path.
 *
 * However, the VFS layer allows individual filesystems to provide their own
 * .fadvise() method, which is called instead of (or in addition to) the
 * generic implementation. Certain filesystem drivers can return ENOENT in
 * this method for reasons that are completely unrelated to the file not
 * existing (the file is clearly open, we have a valid fd):
 *
 *  - NFS: A stale NFS inode can be revalidated against the server while
 *    fadvise is being processed. If the server-side file has been removed or
 *    renamed, the revalidation discovers the inode is gone and returns
 *    -ENOENT, even though the client still holds an open file descriptor.
 *
 *  - FUSE-based filesystems (mergerfs, sshfs, rclone, etc.): The FUSE daemon
 *    receives the fadvise request and may forward it to an underlying backend.
 *    If that backend cannot find the file (e.g., the backing path has changed,
 *    the FUSE daemon is partially initialised, or the daemon translates the
 *    request and the target layer returns an error), the daemon may return
 *    ENOENT to the kernel.
 *
 *  - Overlayfs (used by Docker, Podman, LXC): The overlayfs .fadvise()
 *    implementation delegates to the lower or upper layer. If the lookup
 *    through the overlay stack fails to find the real inode (e.g., the upper
 *    directory entry is absent and the lower copy is in a weird state), the
 *    delegation can return ENOENT.
 *
 *  - Other network or virtual filesystems: Any out-of-tree or in-tree
 *    filesystem that implements .fadvise() and performs a path or inode
 *    lookup as part of processing the hint can propagate ENOENT if that
 *    internal lookup fails.
 *
 * Because posix_fadvise() is always an advisory hint, it never affects
 * correctness, only performance, it is safe to treat unexpected errors,
 * including ENOENT, as "not supported" (ENOSYS) and continue.
 */
int posix_fadvise_wrapper(int fd, off_t offset, off_t len, int advice);
#endif

#if !HAVE_EACCESS
/**
 * Check effective user's permissions for a file.
 * Conceptually identical to access(), but uses the effective UID/GID.
 */
int eaccess(const char* pathname, int mode);
#endif

/**
 * Open a file with the O_NOATIME flag to avoid to update the access time.
 */
int open_noatime(const char* file, int flags);

/**
 * Check if the specified file is hidden.
 */
int dirent_hidden(struct dirent* dd);

/**
 * Return a description of the file type.
 */
const char* stat_desc(struct stat* st);

/**
 * Return the alignment requirement for direct IO.
 */
size_t direct_size(void);

/****************************************************************************/
/* thread */

#if HAVE_PTHREAD_H
#include <pthread.h>
#endif

#if HAVE_PTHREAD
#define HAVE_THREAD 1
typedef pthread_t thread_id_t;
typedef pthread_mutex_t thread_mutex_t;
typedef pthread_cond_t thread_cond_t;
typedef pthread_key_t thread_key_t;
typedef pthread_rwlock_t thread_rwlock_t;
#endif

#endif