Reimplement efile_drv as a dirty NIF

This improves the latency of file operations as dirty schedulers
are a bit more eager to run jobs than async threads, and use a
single global queue rather than per-thread queues, eliminating the
risk of a job stalling behind a long-running job on the same thread
while other async threads sit idle.

There's no such thing as a free lunch though; the lowered latency
comes at the cost of increased busy-waiting which may have an
adverse effect on some applications. This behavior can be tweaked
with the +sbwt flag, but unfortunately it affects all types of
schedulers and not just dirty ones. We plan to add type-specific
flags at a later stage.

sendfile has been moved to inet_drv to lessen the effect of a nasty
race; the cooperation between inet_drv and efile has never been
airtight and the socket dying at the wrong time (Regardless of
reason) could result in fd aliasing. Moving it to the inet driver
makes it impossible to trigger this by closing the socket in the
middle of a sendfile operation, while still allowing it to be
aborted -- something that can't be done if it stays in the file
driver.

The race still occurs if the controlling process dies in the short
window between dispatching the sendfile operation and the dup(2)
call in the driver, but it's much less likely to happen now.

A proper fix is in the works.

--

Notable functional differences:

* The use_threads option for file:sendfile/5 no longer has any
  effect.

* The file-specific DTrace probes have been removed. The same
  effect can be achieved with normal tracing together with the
  nif__entry/nif__return probes to track scheduling.

--

OTP-14256

1 files changed, 957 insertions, 0 deletions

diff --git a/erts/emulator/nifs/unix/unix_prim_file.c b/erts/emulator/nifs/unix/unix_prim_file.c
new file mode 100644
index 0000000000..57c8ef62e1
--- /dev/null
+++ b/erts/emulator/nifs/unix/unix_prim_file.c
@@ -0,0 +1,957 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Ericsson 2017. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+#include "erl_nif.h"
+#include "config.h"
+#include "sys.h"
+
+#ifdef VALGRIND
+#  include <valgrind/memcheck.h>
+#endif
+
+#include "prim_file_nif.h"
+
+#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
+#define __DARWIN__ 1
+#endif
+
+#if defined(__DARWIN__) || defined(HAVE_LINUX_FALLOC_H) || defined(HAVE_POSIX_FALLOCATE)
+#include <fcntl.h>
+#endif
+
+#ifdef HAVE_LINUX_FALLOC_H
+#include <linux/falloc.h>
+#endif
+
+#include <utime.h>
+
+/* Macros for testing file types. */
+#ifdef NO_UMASK
+#define FILE_MODE 0644
+#define DIR_MODE  0755
+#else
+#define FILE_MODE 0666
+#define DIR_MODE  0777
+#endif
+
+/* Old platforms might not have IOV_MAX defined. */
+#if !defined(IOV_MAX) && defined(UIO_MAXIOV)
+#define IOV_MAX UIO_MAXIOV
+#elif !defined(IOV_MAX)
+#define IOV_MAX 16
+#endif
+
+typedef struct {
+    efile_data_t common;
+    int fd;
+} efile_unix_t;
+
+static int has_invalid_null_termination(const ErlNifBinary *path) {
+    const char *null_pos, *end_pos;
+
+    null_pos = memchr(path->data, '\0', path->size);
+    end_pos = (const char*)&path->data[path->size] - 1;
+
+    if(null_pos == NULL) {
+        return 1;
+    }
+
+    /* prim_file:internal_name2native sometimes feeds us data that is "doubly"
+     * NUL-terminated, so we'll accept any number of trailing NULs so long as
+     * they aren't interrupted by anything else. */
+    while(null_pos < end_pos && (*null_pos) == '\0') {
+        null_pos++;
+    }
+
+    return null_pos != end_pos;
+}
+
+posix_errno_t efile_marshal_path(ErlNifEnv *env, ERL_NIF_TERM path, efile_path_t *result) {
+    if(!enif_inspect_binary(env, path, result)) {
+        return EINVAL;
+    }
+
+    if(has_invalid_null_termination(result)) {
+        return EINVAL;
+    }
+
+    return 0;
+}
+
+ERL_NIF_TERM efile_get_handle(ErlNifEnv *env, efile_data_t *d) {
+    efile_unix_t *u = (efile_unix_t*)d;
+
+    ERL_NIF_TERM result;
+    unsigned char *bits;
+
+    bits = enif_make_new_binary(env, sizeof(u->fd), &result);
+    memcpy(bits, &u->fd, sizeof(u->fd));
+
+    return result;
+}
+
+static int open_file_type_check(const efile_path_t *path, int fd) {
+    struct stat file_info;
+    int error;
+
+#ifndef HAVE_FSTAT
+    error = stat((const char*)path->data, &file_info);
+    (void)fd;
+#else
+    error = fstat(fd, &file_info);
+    (void)path;
+#endif
+
+    if(error < 0) {
+        /* If we failed to stat assume success and let the next call handle the
+         * error. The old driver checked whether the file was to be used
+         * immediately in a read within the call, but the new implementation
+         * never does that. */
+         return 1;
+    } else {
+        /* The old driver tolerated opening /dev/null despite the "no devices"
+         * limitation. It provided no explanation for this but we still need
+         * to match the behavior. We're checking through stat(2) instead of
+         * comparing the name to account for links. */
+        struct stat null_device_info;
+        int is_dev_null;
+
+        is_dev_null = (stat("/dev/null", &null_device_info) == 0);
+        is_dev_null &= (file_info.st_ino == null_device_info.st_ino);
+        is_dev_null &= (file_info.st_dev == null_device_info.st_dev);
+
+        if(is_dev_null) {
+            return 1;
+        }
+    }
+
+    if(!S_ISREG(file_info.st_mode)) {
+        return 0;
+    }
+
+    return 1;
+}
+
+posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes,
+        ErlNifResourceType *nif_type, efile_data_t **d) {
+
+    int flags, fd;
+
+    flags = 0;
+
+    if(modes & EFILE_MODE_READ && !(modes & EFILE_MODE_WRITE)) {
+        flags |= O_RDONLY;
+    } else if(modes & EFILE_MODE_WRITE && !(modes & EFILE_MODE_READ)) {
+        if(!(modes & EFILE_MODE_NO_TRUNCATE)) {
+            flags |= O_TRUNC;
+        }
+
+        flags |= O_WRONLY | O_CREAT;
+    } else if(modes & EFILE_MODE_READ_WRITE) {
+        flags |= O_RDWR | O_CREAT;
+    } else {
+        return EINVAL;
+    }
+
+    if(modes & EFILE_MODE_APPEND) {
+        flags &= ~O_TRUNC;
+        flags |= O_APPEND;
+    }
+
+    if(modes & EFILE_MODE_EXCLUSIVE) {
+        flags |= O_EXCL;
+    }
+
+    if(modes & EFILE_MODE_SYNC) {
+#ifndef O_SYNC
+        return ENOTSUP;
+#else
+        flags |= O_SYNC;
+#endif
+    }
+
+    do {
+        fd = open((const char*)path->data, flags, FILE_MODE);
+    } while(fd == -1 && errno == EINTR);
+
+    if(fd != -1) {
+        efile_unix_t *u;
+
+        if(!(modes & EFILE_MODE_SKIP_TYPE_CHECK) && !open_file_type_check(path, fd)) {
+            close(fd);
+
+            /* This is blatantly incorrect, but we're documented as returning
+             * this for everything that isn't a file. */
+            return EISDIR;
+        }
+
+        u = (efile_unix_t*)enif_alloc_resource(nif_type, sizeof(efile_unix_t));
+        u->fd = fd;
+
+        EFILE_INIT_RESOURCE(&u->common, modes);
+        (*d) = &u->common;
+
+        return 0;
+    }
+
+    (*d) = NULL;
+    return errno;
+}
+
+int efile_close(efile_data_t *d) {
+    efile_unix_t *u = (efile_unix_t*)d;
+    int fd;
+
+    ASSERT(erts_atomic32_read_nob(&d->state) == EFILE_STATE_CLOSED);
+    ASSERT(u->fd != -1);
+
+    fd = u->fd;
+    u->fd = -1;
+
+    /* close(2) either always closes (*BSD, Linux) or leaves the fd in an
+     * undefined state (POSIX 2008, Solaris), so we must not retry on EINTR. */
+
+    if(close(fd) < 0) {
+        u->common.posix_errno = errno;
+        return 0;
+    }
+
+    return 1;
+}
+
+static void shift_iov(SysIOVec **iov, int *iovlen, ssize_t shift) {
+    SysIOVec *head_vec = (*iov);
+
+    ASSERT(shift >= 0);
+
+    while(shift > 0) {
+        ASSERT(head_vec < &(*iov)[*iovlen]);
+
+        if(shift < head_vec->iov_len) {
+            head_vec->iov_base = (char*)head_vec->iov_base + shift;
+            head_vec->iov_len -= shift;
+            break;
+        } else {
+            shift -= head_vec->iov_len;
+            head_vec++;
+        }
+    }
+
+    (*iovlen) -= head_vec - (*iov);
+    (*iov) = head_vec;
+}
+
+Sint64 efile_readv(efile_data_t *d, SysIOVec *iov, int iovlen) {
+    efile_unix_t *u = (efile_unix_t*)d;
+
+    Sint64 bytes_read;
+    ssize_t result;
+
+    bytes_read = 0;
+
+    do {
+        int use_fallback = 0;
+
+        if(iovlen < 1) {
+            result = 0;
+            break;
+        }
+
+        /* writev(2) implies readv(2) */
+#ifdef HAVE_WRITEV
+        result = readv(u->fd, iov, MIN(IOV_MAX, iovlen));
+
+        /* Fall back to using read(2) if readv(2) reports that the combined
+         * size of iov is greater than SSIZE_T_MAX. */
+        use_fallback = (result < 0 && errno == EINVAL);
+#else
+        use_fallback = 1;
+#endif
+
+        if(use_fallback) {
+            result = read(u->fd, iov->iov_base, iov->iov_len);
+        }
+
+        if(result > 0) {
+            shift_iov(&iov, &iovlen, result);
+            bytes_read += result;
+        }
+    } while(result > 0 || (result < 0 && errno == EINTR));
+
+    u->common.posix_errno = errno;
+
+    if(result == 0 && bytes_read > 0) {
+        return bytes_read;
+    }
+
+    return result;
+}
+
+Sint64 efile_writev(efile_data_t *d, SysIOVec *iov, int iovlen) {
+    efile_unix_t *u = (efile_unix_t*)d;
+
+    Sint64 bytes_written;
+    ssize_t result;
+
+    bytes_written = 0;
+
+    do {
+        int use_fallback = 0;
+
+        if(iovlen < 1) {
+            result = 0;
+            break;
+        }
+
+#ifdef HAVE_WRITEV
+        result = writev(u->fd, iov, MIN(IOV_MAX, iovlen));
+
+        /* Fall back to using write(2) if writev(2) reports that the combined
+         * size of iov is greater than SSIZE_T_MAX. */
+        use_fallback = (result < 0 && errno == EINVAL);
+#else
+        use_fallback = 1;
+#endif
+
+        if(use_fallback) {
+            result = write(u->fd, iov->iov_base, iov->iov_len);
+        }
+
+        if(result > 0) {
+            shift_iov(&iov, &iovlen, result);
+            bytes_written += result;
+        }
+    } while(result > 0 || (result < 0 && errno == EINTR));
+
+    u->common.posix_errno = errno;
+
+    if(result == 0 && bytes_written > 0) {
+        return bytes_written;
+    }
+
+    return result;
+}
+
+Sint64 efile_preadv(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen) {
+    efile_unix_t *u = (efile_unix_t*)d;
+
+    Uint64 bytes_read;
+    Sint64 result;
+
+#if !defined(HAVE_PREADV) && !defined(HAVE_PREAD)
+    /* This function is documented as leaving the file position undefined, but
+     * the old driver always reset it so there's probably code in the wild that
+     * relies on this behavior. */
+    off_t original_position = lseek(u->fd, 0, SEEK_CUR);
+
+    if(original_position < 0 || lseek(u->fd, offset, SEEK_SET) < 0) {
+        u->common.posix_errno = errno;
+        return -1;
+    }
+#endif
+
+    bytes_read = 0;
+
+    do {
+        if(iovlen < 1) {
+            result = 0;
+            break;
+        }
+
+#if defined(HAVE_PREADV)
+        result = preadv(u->fd, iov, MIN(IOV_MAX, iovlen), offset);
+#elif defined(HAVE_PREAD)
+        result = pread(u->fd, iov->iov_base, iov->iov_len, offset);
+#else
+        result = read(u->fd, iov->iov_base, iov->iov_len);
+#endif
+
+        if(result > 0) {
+            shift_iov(&iov, &iovlen, result);
+            bytes_read += result;
+            offset += result;
+        }
+    } while(result > 0 || (result < 0 && errno == EINTR));
+
+    u->common.posix_errno = errno;
+
+#if !defined(HAVE_PREADV) && !defined(HAVE_PREAD)
+    if(result >= 0) {
+        if(lseek(u->fd, original_position, SEEK_SET) < 0) {
+            u->common.posix_errno = errno;
+            return -1;
+        }
+    }
+#endif
+
+    if(result == 0 && bytes_read > 0) {
+        return bytes_read;
+    }
+
+    return result;
+}
+
+Sint64 efile_pwritev(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen) {
+    efile_unix_t *u = (efile_unix_t*)d;
+
+    Sint64 bytes_written;
+    ssize_t result;
+
+#if !defined(HAVE_PWRITEV) && !defined(HAVE_PWRITE)
+    off_t original_position = lseek(u->fd, 0, SEEK_CUR);
+
+    if(original_position < 0 || lseek(u->fd, offset, SEEK_SET) < 0) {
+        u->common.posix_errno = errno;
+        return -1;
+    }
+#endif
+
+    bytes_written = 0;
+
+    do {
+        if(iovlen < 1) {
+            result = 0;
+            break;
+        }
+
+#if defined(HAVE_PWRITEV)
+        result = pwritev(u->fd, iov, MIN(IOV_MAX, iovlen), offset);
+#elif defined(HAVE_PWRITE)
+        result = pwrite(u->fd, iov->iov_base, iov->iov_len, offset);
+#else
+        result = write(u->fd, iov->iov_base, iov->iov_len);
+#endif
+
+        if(result > 0) {
+            shift_iov(&iov, &iovlen, result);
+            bytes_written += result;
+            offset += result;
+        }
+    } while(result > 0 || (result < 0 && errno == EINTR));
+
+    u->common.posix_errno = errno;
+
+#if !defined(HAVE_PWRITEV) && !defined(HAVE_PWRITE)
+    if(result >= 0) {
+        if(lseek(u->fd, original_position, SEEK_SET) < 0) {
+            u->common.posix_errno = errno;
+            return -1;
+        }
+    }
+#endif
+
+    if(result == 0 && bytes_written > 0) {
+        return bytes_written;
+    }
+
+    return result;
+}
+
+int efile_seek(efile_data_t *d, enum efile_seek_t seek, Sint64 offset, Sint64 *new_position) {
+    efile_unix_t *u = (efile_unix_t*)d;
+    off_t result;
+    int whence;
+
+    switch(seek) {
+        case EFILE_SEEK_BOF: whence = SEEK_SET; break;
+        case EFILE_SEEK_CUR: whence = SEEK_CUR; break;
+        case EFILE_SEEK_EOF: whence = SEEK_END; break;
+        default: ERTS_INTERNAL_ERROR("Invalid seek parameter");
+    }
+
+    result = lseek(u->fd, offset, whence);
+
+    /*
+     * The man page for lseek (on SunOs 5) says:
+     *
+     * "if fildes is a remote file descriptor and offset is negative, lseek()
+     * returns the file pointer even if it is negative."
+     */
+    if(result < 0 && errno == 0) {
+        errno = EINVAL;
+    }
+
+    if(result < 0) {
+        u->common.posix_errno = errno;
+        return 0;
+    }
+
+    (*new_position) = result;
+
+    return 1;
+}
+
+int efile_sync(efile_data_t *d, int data_only) {
+    efile_unix_t *u = (efile_unix_t*)d;
+
+#if defined(HAVE_FDATASYNC) && !defined(__DARWIN__)
+    if(data_only) {
+        if(fdatasync(u->fd) < 0) {
+            u->common.posix_errno = errno;
+            return 0;
+        }
+
+        return 1;
+    }
+#endif
+
+#if defined(__DARWIN__) && defined(F_FULLFSYNC)
+    if(fcntl(u->fd, F_FULLFSYNC) < 0) {
+#else
+    if(fsync(u->fd) < 0) {
+#endif
+        u->common.posix_errno = errno;
+        return 0;
+    }
+
+    return 1;
+}
+
+int efile_advise(efile_data_t *d, Sint64 offset, Sint64 length, enum efile_advise_t advise) {
+    efile_unix_t *u = (efile_unix_t*)d;
+#ifdef HAVE_POSIX_FADVISE
+    int p_advise;
+
+    switch(advise) {
+        case EFILE_ADVISE_NORMAL: p_advise = POSIX_FADV_NORMAL; break;
+        case EFILE_ADVISE_RANDOM: p_advise = POSIX_FADV_RANDOM; break;
+        case EFILE_ADVISE_SEQUENTIAL: p_advise = POSIX_FADV_SEQUENTIAL; break;
+        case EFILE_ADVISE_WILL_NEED: p_advise = POSIX_FADV_WILLNEED; break;
+        case EFILE_ADVISE_DONT_NEED: p_advise = POSIX_FADV_DONTNEED; break;
+        case EFILE_ADVISE_NO_REUSE: p_advise = POSIX_FADV_NOREUSE; break;
+        default:
+            u->common.posix_errno = EINVAL;
+            return 0;
+    }
+
+    if(posix_fadvise(u->fd, offset, length, p_advise) < 0) {
+        u->common.posix_errno = errno;
+        return 0;
+    }
+
+    return 1;
+#else
+    /* We'll pretend to support this syscall as it's only a recommendation even
+     * on systems that do support it. */
+    return 1;
+#endif
+}
+
+int efile_allocate(efile_data_t *d, Sint64 offset, Sint64 length) {
+    efile_unix_t *u = (efile_unix_t*)d;
+    int ret = -1;
+
+    /* We prefer OS-specific methods, but fall back to posix_fallocate on
+     * failure. It's unclear whether this has any practical benefit on
+     * modern systems, but the old driver did it. */
+
+#if defined(HAVE_FALLOCATE)
+    /* Linux-specific */
+    do {
+        ret = fallocate(u->fd, FALLOC_FL_KEEP_SIZE, offset, length);
+    } while(ret < 0 && errno == EINTR);
+#elif defined(F_PREALLOCATE)
+    /* Mac-specific */
+    fstore_t fs = {};
+
+    fs.fst_flags = F_ALLOCATECONTIG;
+    fs.fst_posmode = F_VOLPOSMODE;
+    fs.fst_offset = offset;
+    fs.fst_length = length;
+
+    ret = fcntl(u->fd, F_PREALLOCATE, &fs);
+    if(ret < 0) {
+        fs.fst_flags = F_ALLOCATEALL;
+        ret = fcntl(u->fd, F_PREALLOCATE, &fs);
+    }
+#elif !defined(HAVE_POSIX_FALLOCATE)
+    u->common.posix_errno = ENOTSUP;
+    return 0;
+#endif
+
+#ifdef HAVE_POSIX_FALLOCATE
+    if(ret < 0) {
+        do {
+            ret = posix_fallocate(u->fd, offset, length);
+
+            /* On Linux and Solaris for example, posix_fallocate() returns a
+             * positive error number on error and it does not set errno. On
+             * FreeBSD however (9.0 at least), it returns -1 on error and it
+             * sets errno. */
+            if (ret > 0) {
+                errno = ret;
+                ret = -1;
+            }
+        } while(ret < 0 && errno == EINTR);
+    }
+#endif
+
+    if(ret < 0) {
+        u->common.posix_errno = errno;
+        return 0;
+    }
+
+    return 1;
+}
+
+int efile_truncate(efile_data_t *d) {
+    efile_unix_t *u = (efile_unix_t*)d;
+    off_t offset;
+
+    offset = lseek(u->fd, 0, SEEK_CUR);
+
+    if(offset < 0) {
+        u->common.posix_errno = errno;
+        return 0;
+    }
+
+    if(ftruncate(u->fd, offset) < 0) {
+        u->common.posix_errno = errno;
+        return 0;
+    }
+
+    return 1;
+}
+
+posix_errno_t efile_read_info(const efile_path_t *path, int follow_links, efile_fileinfo_t *result) {
+    struct stat data;
+
+    if(follow_links) {
+        if(stat((const char*)path->data, &data) < 0) {
+            return errno;
+        }
+    } else {
+        if(lstat((const char*)path->data, &data) < 0) {
+            return errno;
+        }
+    }
+
+    if(S_ISCHR(data.st_mode) || S_ISBLK(data.st_mode)) {
+        result->type = EFILE_FILETYPE_DEVICE;
+    } else if(S_ISDIR(data.st_mode)) {
+        result->type = EFILE_FILETYPE_DIRECTORY;
+    } else if(S_ISREG(data.st_mode)) {
+        result->type = EFILE_FILETYPE_REGULAR;
+    } else if(S_ISLNK(data.st_mode)) {
+        result->type = EFILE_FILETYPE_SYMLINK;
+    } else {
+        result->type = EFILE_FILETYPE_OTHER;
+    }
+
+    result->a_time = (Sint64)data.st_atime;
+    result->m_time = (Sint64)data.st_mtime;
+    result->c_time = (Sint64)data.st_ctime;
+    result->size = data.st_size;
+
+    result->major_device = data.st_dev;
+    result->minor_device = data.st_rdev;
+    result->links = data.st_nlink;
+    result->inode = data.st_ino;
+    result->mode = data.st_mode;
+    result->uid = data.st_uid;
+    result->gid = data.st_gid;
+
+#ifndef NO_ACCESS
+    result->access = EFILE_ACCESS_NONE;
+
+    if(access((const char*)path->data, R_OK) == 0) {
+        result->access |= EFILE_ACCESS_READ;
+    }
+    if(access((const char*)path->data, W_OK) == 0) {
+        result->access |= EFILE_ACCESS_WRITE;
+    }
+#else
+    /* Just look at read/write access for owner. */
+    result->access = ((data.st_mode >> 6) & 07) >> 1;
+#endif
+
+    return 0;
+}
+
+posix_errno_t efile_set_permissions(const efile_path_t *path, Uint32 permissions) {
+    const mode_t MUTABLE_MODES = (S_ISUID | S_ISGID | S_IRWXU | S_IRWXG | S_IRWXO);
+    mode_t new_modes = permissions & MUTABLE_MODES;
+
+    if(chmod((const char*)path->data, new_modes) < 0) {
+        new_modes &= ~(S_ISUID | S_ISGID);
+
+        if (chmod((const char*)path->data, new_modes) < 0) {
+            return errno;
+        }
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_set_owner(const efile_path_t *path, Uint32 owner, Uint32 group) {
+    if(chown((const char*)path->data, owner, group) < 0) {
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_set_time(const efile_path_t *path, Sint64 a_time, Sint64 m_time, Sint64 c_time) {
+    struct utimbuf tval;
+
+    tval.actime = (time_t)a_time;
+    tval.modtime = (time_t)m_time;
+
+    (void)c_time;
+
+    if(utime((const char*)path->data, &tval) < 0) {
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_read_link(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+    ErlNifBinary result_bin;
+
+    if(!enif_alloc_binary(256, &result_bin)) {
+        return ENOMEM;
+    }
+
+    for(;;) {
+        ssize_t bytes_copied;
+
+        bytes_copied = readlink((const char*)path->data, (char*)result_bin.data,
+            result_bin.size);
+
+        if(bytes_copied <= 0) {
+            posix_errno_t saved_errno = errno;
+            enif_release_binary(&result_bin);
+            return saved_errno;
+        } else if(bytes_copied < result_bin.size) {
+            if(!enif_realloc_binary(&result_bin, bytes_copied)) {
+                enif_release_binary(&result_bin);
+                return ENOMEM;
+            }
+
+            (*result) = enif_make_binary(env, &result_bin);
+
+            return 0;
+        }
+
+        /* The result didn't fit into the buffer, so we'll try again with a
+         * larger one. */
+
+        if(!enif_realloc_binary(&result_bin, result_bin.size * 2)) {
+            enif_release_binary(&result_bin);
+            return ENOMEM;
+        }
+    }
+}
+
+static int is_ignored_name(int name_length, const char *name) {
+    if(name_length == 1 && name[0] == '.') {
+        return 1;
+    } else if(name_length == 2 && memcmp(name, "..", 2) == 0) {
+        return 1;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_list_dir(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+    ERL_NIF_TERM list_head;
+
+    struct dirent *dir_entry;
+    DIR *dir_stream;
+
+    dir_stream = opendir((const char*)path->data);
+    if(dir_stream == NULL) {
+        posix_errno_t saved_errno = errno;
+        *result = enif_make_list(env, 0);
+        return saved_errno;
+    }
+
+    list_head = enif_make_list(env, 0);
+    dir_entry = readdir(dir_stream);
+
+    while(dir_entry != NULL) {
+        int name_length = strlen(dir_entry->d_name);
+
+        if(!is_ignored_name(name_length, dir_entry->d_name)) {
+            unsigned char *name_bytes;
+            ERL_NIF_TERM name_term;
+
+            name_bytes = enif_make_new_binary(env, name_length, &name_term);
+            sys_memcpy(name_bytes, dir_entry->d_name, name_length);
+
+            list_head = enif_make_list_cell(env, name_term, list_head);
+        }
+
+        dir_entry = readdir(dir_stream);
+    }
+
+    (*result) = list_head;
+    closedir(dir_stream);
+
+    return 0;
+}
+
+posix_errno_t efile_rename(const efile_path_t *old_path, const efile_path_t *new_path) {
+    if(rename((const char*)old_path->data, (const char*)new_path->data) < 0) {
+        if(errno == ENOTEMPTY) {
+            return EEXIST;
+        }
+
+        if(strcmp((const char*)old_path->data, "/") == 0) {
+            /* Alpha reports renaming / as EBUSY and Linux reports it as EACCES
+             * instead of EINVAL.*/
+             return EINVAL;
+        }
+
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_make_hard_link(const efile_path_t *existing_path, const efile_path_t *new_path) {
+    if(link((const char*)existing_path->data, (const char*)new_path->data) < 0) {
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_make_soft_link(const efile_path_t *existing_path, const efile_path_t *new_path) {
+    if(symlink((const char*)existing_path->data, (const char*)new_path->data) < 0) {
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_make_dir(const efile_path_t *path) {
+#ifdef NO_MKDIR_MODE
+    if(mkdir((const char*)path->data) < 0) {
+#else
+    if(mkdir((const char*)path->data, DIR_MODE) < 0) {
+#endif
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_del_file(const efile_path_t *path) {
+    if(unlink((const char*)path->data) < 0) {
+        /* Linux sets the wrong error code. */
+        if(errno == EISDIR) {
+            return EPERM;
+        }
+
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_del_dir(const efile_path_t *path) {
+    if(rmdir((const char*)path->data) < 0) {
+        posix_errno_t saved_errno = errno;
+
+        if(saved_errno == ENOTEMPTY) {
+            saved_errno = EEXIST;
+        }
+
+        /* The error code might be wrong if we're trying to delete the current
+         * directory. */
+        if(saved_errno == EEXIST) {
+            struct stat path_stat, cwd_stat;
+            int has_stat;
+
+            has_stat = (stat((const char*)path->data, &path_stat) == 0);
+            has_stat &= (stat(".", &cwd_stat) == 0);
+
+            if(has_stat && path_stat.st_ino == cwd_stat.st_ino) {
+                if(path_stat.st_dev == cwd_stat.st_dev) {
+                    return EINVAL;
+                }
+            }
+        }
+
+        return saved_errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_set_cwd(const efile_path_t *path) {
+    if(chdir((const char*)path->data) < 0) {
+        return errno;
+    }
+
+    return 0;
+}
+
+posix_errno_t efile_get_device_cwd(ErlNifEnv *env, int device_index, ERL_NIF_TERM *result) {
+    (void)device_index;
+    (void)result;
+    (void)env;
+
+    return ENOTSUP;
+}
+
+posix_errno_t efile_get_cwd(ErlNifEnv *env, ERL_NIF_TERM *result) {
+    ErlNifBinary result_bin;
+    size_t bytes_copied;
+
+    if(!enif_alloc_binary(256, &result_bin)) {
+        return ENOMEM;
+    }
+
+    while(getcwd((char*)result_bin.data, result_bin.size) == NULL) {
+        posix_errno_t saved_errno = errno;
+
+        if(saved_errno != ERANGE) {
+            enif_release_binary(&result_bin);
+            return saved_errno;
+        } else {
+            if(!enif_realloc_binary(&result_bin, result_bin.size * 2)) {
+                enif_release_binary(&result_bin);
+                return ENOMEM;
+            }
+        }
+    }
+
+    /* getcwd(2) guarantees null-termination. */
+    bytes_copied = strlen((const char*)result_bin.data);
+
+    if(!enif_realloc_binary(&result_bin, bytes_copied)) {
+        enif_release_binary(&result_bin);
+        return ENOMEM;
+    }
+
+    (*result) = enif_make_binary(env, &result_bin);
+
+    return 0;
+}
+
+posix_errno_t efile_altname(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+    (void)path;
+    (void)result;
+
+    return ENOTSUP;
+}