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

2 files changed, 1477 insertions, 0 deletions

diff --git a/erts/emulator/nifs/common/prim_file_nif.c b/erts/emulator/nifs/common/prim_file_nif.c
new file mode 100644
index 0000000000..6874f41d75
--- /dev/null
+++ b/erts/emulator/nifs/common/prim_file_nif.c
@@ -0,0 +1,1237 @@
+/*
+ * %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%
+ */
+
+#define STATIC_ERLANG_NIF 1
+
+#include "erl_nif.h"
+#include "config.h"
+#include "sys.h"
+
+#ifdef VALGRIND
+#  include <valgrind/memcheck.h>
+#endif
+
+#include "erl_driver.h"
+#include "prim_file_nif.h"
+
+/* NIF interface declarations */
+static int load(ErlNifEnv *env, void** priv_data, ERL_NIF_TERM load_info);
+static int upgrade(ErlNifEnv *env, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info);
+static void unload(ErlNifEnv *env, void* priv_data);
+
+static ErlNifResourceType *efile_resource_type;
+
+static ERL_NIF_TERM am_ok;
+static ERL_NIF_TERM am_error;
+static ERL_NIF_TERM am_continue;
+
+static ERL_NIF_TERM am_file_info;
+
+/* File modes */
+static ERL_NIF_TERM am_read;
+static ERL_NIF_TERM am_write;
+static ERL_NIF_TERM am_exclusive;
+static ERL_NIF_TERM am_append;
+static ERL_NIF_TERM am_sync;
+static ERL_NIF_TERM am_skip_type_check;
+
+/* enum efile_access_t; read and write are defined above.*/
+static ERL_NIF_TERM am_read_write;
+static ERL_NIF_TERM am_none;
+
+/* enum efile_advise_t */
+static ERL_NIF_TERM am_normal;
+static ERL_NIF_TERM am_random;
+static ERL_NIF_TERM am_sequential;
+static ERL_NIF_TERM am_will_need;
+static ERL_NIF_TERM am_dont_need;
+static ERL_NIF_TERM am_no_reuse;
+
+/* enum efile_filetype_t */
+static ERL_NIF_TERM am_device;
+static ERL_NIF_TERM am_directory;
+static ERL_NIF_TERM am_regular;
+static ERL_NIF_TERM am_symlink;
+static ERL_NIF_TERM am_other;
+
+/* enum efile_seek_t, 'eof' marker. */
+static ERL_NIF_TERM am_bof;
+static ERL_NIF_TERM am_cur;
+static ERL_NIF_TERM am_eof;
+
+static ERL_NIF_TERM read_info_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_permissions_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_owner_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_time_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM read_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM list_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM make_hard_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM make_soft_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM rename_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM make_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM del_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM del_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM get_device_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM get_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM read_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM get_handle_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM altname_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+/* All file handle operations are passed through a wrapper that handles state
+ * transitions, marking it as busy during the course of the operation, and
+ * closing on completion if the owner died in the middle of an operation.
+ *
+ * This is pretty ugly but required as there's no way to tell when it's safe to
+ * asynchronously close a file; the event could have fired just before landing
+ * in a system call which will fail with EBADF at best or alias a newly opened
+ * fd at worst.
+ *
+ * The old driver got away with enqueueing the close operation on the same
+ * async queue as all of its other operations, but since dirty schedulers use a
+ * single global queue there's no natural way to schedule an asynchronous close
+ * "behind" other operations.
+ *
+ * The states may transition as follows:
+ *
+ * IDLE ->
+ *      BUSY (file_handle_wrapper) |
+ *      CLOSED (owner_death_callback)
+ *
+ * BUSY ->
+ *      IDLE (file_handle_wrapper)
+ *      CLOSED (close_nif_impl)
+ *      CLOSE_PENDING (owner_death_callback)
+ *
+ * CLOSE_PENDING ->
+ *      CLOSED (file_handle_wrapper)
+ */
+
+typedef ERL_NIF_TERM (*file_op_impl_t)(efile_data_t *d, ErlNifEnv *env,
+    int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM file_handle_wrapper(file_op_impl_t operation, ErlNifEnv *env,
+    int argc, const ERL_NIF_TERM argv[]);
+
+#define WRAP_FILE_HANDLE_EXPORT(name) \
+    static ERL_NIF_TERM name ## _impl (efile_data_t *d, ErlNifEnv *env, \
+        int argc, const ERL_NIF_TERM argv[]);\
+    static ERL_NIF_TERM name(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) { \
+        return file_handle_wrapper( name ## _impl , env, argc, argv); \
+    }
+
+WRAP_FILE_HANDLE_EXPORT(close_nif)
+WRAP_FILE_HANDLE_EXPORT(read_nif)
+WRAP_FILE_HANDLE_EXPORT(write_nif)
+WRAP_FILE_HANDLE_EXPORT(pread_nif)
+WRAP_FILE_HANDLE_EXPORT(pwrite_nif)
+WRAP_FILE_HANDLE_EXPORT(seek_nif)
+WRAP_FILE_HANDLE_EXPORT(sync_nif)
+WRAP_FILE_HANDLE_EXPORT(truncate_nif)
+WRAP_FILE_HANDLE_EXPORT(allocate_nif)
+WRAP_FILE_HANDLE_EXPORT(advise_nif)
+WRAP_FILE_HANDLE_EXPORT(get_handle_nif)
+WRAP_FILE_HANDLE_EXPORT(ipread_s32bu_p32bu_nif)
+
+static ErlNifFunc nif_funcs[] = {
+    /* File handle ops */
+    {"open_nif", 2, open_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"close_nif", 1, close_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"read_nif", 2, read_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"write_nif", 2, write_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"pread_nif", 3, pread_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"pwrite_nif", 3, pwrite_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"seek_nif", 3, seek_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"sync_nif", 2, sync_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"truncate_nif", 1, truncate_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"allocate_nif", 3, allocate_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"advise_nif", 4, advise_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+
+    /* Filesystem ops */
+    {"make_hard_link_nif", 2, make_hard_link_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"make_soft_link_nif", 2, make_soft_link_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"rename_nif", 2, rename_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"read_info_nif", 2, read_info_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"set_permissions_nif", 2, set_permissions_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"set_owner_nif", 3, set_owner_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"set_time_nif", 4, set_time_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"read_link_nif", 1, read_link_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"list_dir_nif", 1, list_dir_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"make_dir_nif", 1, make_dir_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"del_file_nif", 1, del_file_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"del_dir_nif", 1, del_dir_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"get_device_cwd_nif", 1, get_device_cwd_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"set_cwd_nif", 1, set_cwd_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"get_cwd_nif", 0, get_cwd_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+
+    /* These operations are equivalent to chained calls of other operations,
+     * but have been moved down to avoid excessive rescheduling. */
+    {"ipread_s32bu_p32bu_nif", 3, ipread_s32bu_p32bu_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"read_file_nif", 1, read_file_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+
+    /* Internal ops. */
+    {"get_handle_nif", 1, get_handle_nif},
+    {"altname_nif", 1, altname_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+};
+
+ERL_NIF_INIT(prim_file, nif_funcs, load, NULL, upgrade, unload)
+
+static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlNifMonitor* mon);
+static void gc_callback(ErlNifEnv *env, void* data);
+
+static int load(ErlNifEnv *env, void** priv_data, ERL_NIF_TERM load_info)
+{
+    ErlNifResourceTypeInit callbacks;
+
+    am_ok = enif_make_atom(env, "ok");
+    am_error = enif_make_atom(env, "error");
+    am_continue = enif_make_atom(env, "continue");
+
+    am_read = enif_make_atom(env, "read");
+    am_write = enif_make_atom(env, "write");
+    am_exclusive = enif_make_atom(env, "exclusive");
+    am_append = enif_make_atom(env, "append");
+    am_sync = enif_make_atom(env, "sync");
+    am_skip_type_check = enif_make_atom(env, "skip_type_check");
+
+    am_read_write = enif_make_atom(env, "read_write");
+    am_none = enif_make_atom(env, "none");
+
+    am_normal = enif_make_atom(env, "normal");
+    am_random = enif_make_atom(env, "random");
+    am_sequential = enif_make_atom(env, "sequential");
+    am_will_need = enif_make_atom(env, "will_need");
+    am_dont_need = enif_make_atom(env, "dont_need");
+    am_no_reuse = enif_make_atom(env, "no_reuse");
+
+    am_device = enif_make_atom(env, "device");
+    am_directory = enif_make_atom(env, "directory");
+    am_regular = enif_make_atom(env, "regular");
+    am_symlink = enif_make_atom(env, "symlink");
+    am_other = enif_make_atom(env, "other");
+
+    am_file_info = enif_make_atom(env, "file_info");
+
+    am_bof = enif_make_atom(env, "bof");
+    am_cur = enif_make_atom(env, "cur");
+    am_eof = enif_make_atom(env, "eof");
+
+    callbacks.down = owner_death_callback;
+    callbacks.dtor = gc_callback;
+    callbacks.stop = NULL;
+
+    efile_resource_type = enif_open_resource_type_x(env, "efile", &callbacks,
+        ERL_NIF_RT_CREATE, NULL);
+
+    *priv_data = NULL;
+
+    return 0;
+}
+
+static void unload(ErlNifEnv *env, void* priv_data)
+{
+
+}
+
+static int upgrade(ErlNifEnv *env, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info)
+{
+    if(*old_priv_data != NULL) {
+        return -1; /* Don't know how to do that */
+    }
+    if(*priv_data != NULL) {
+        return -1; /* Don't know how to do that */
+    }
+    if(load(env, priv_data, load_info)) {
+        return -1;
+    }
+    return 0;
+}
+
+static ERL_NIF_TERM posix_error_to_tuple(ErlNifEnv *env, posix_errno_t posix_errno) {
+    ERL_NIF_TERM error = enif_make_atom(env, erl_errno_id(posix_errno));
+    return enif_make_tuple2(env, am_error, error);
+}
+
+static int get_file_data(ErlNifEnv *env, ERL_NIF_TERM opaque, efile_data_t **d) {
+    return enif_get_resource(env, opaque, efile_resource_type, (void **)d);
+}
+
+static ERL_NIF_TERM file_handle_wrapper(file_op_impl_t operation, ErlNifEnv *env,
+        int argc, const ERL_NIF_TERM argv[]) {
+
+    efile_data_t *d;
+
+    enum efile_state_t previous_state;
+    ERL_NIF_TERM result;
+
+    if(argc < 1 || !get_file_data(env, argv[0], &d)) {
+        return enif_make_badarg(env);
+    }
+
+    previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+        EFILE_STATE_BUSY, EFILE_STATE_IDLE);
+
+    if(previous_state == EFILE_STATE_IDLE) {
+        result = operation(d, env, argc - 1, &argv[1]);
+
+        previous_state = erts_atomic32_cmpxchg_relb(&d->state,
+            EFILE_STATE_IDLE, EFILE_STATE_BUSY);
+
+        ASSERT(previous_state != EFILE_STATE_IDLE);
+
+        if(previous_state == EFILE_STATE_CLOSE_PENDING) {
+            /* This is the only point where a change from CLOSE_PENDING is
+             * possible, and we're running synchronously, so we can't race with
+             * anything else here. */
+            erts_atomic32_set_acqb(&d->state, EFILE_STATE_CLOSED);
+            efile_close(d);
+        }
+    } else {
+        /* CLOSE_PENDING should be impossible at this point since it requires
+         * a transition from BUSY; the only valid state here is CLOSED. */
+        ASSERT(previous_state == EFILE_STATE_CLOSED);
+
+        result = posix_error_to_tuple(env, EINVAL);
+    }
+
+    return result;
+}
+
+static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlNifMonitor* mon) {
+    efile_data_t *d = (efile_data_t*)obj;
+
+    (void)env;
+    (void)pid;
+    (void)mon;
+
+    for(;;) {
+        enum efile_state_t previous_state;
+
+        previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+            EFILE_STATE_CLOSED, EFILE_STATE_IDLE);
+
+        switch(previous_state) {
+        case EFILE_STATE_IDLE:
+            efile_close(d);
+            return;
+        case EFILE_STATE_CLOSE_PENDING:
+        case EFILE_STATE_CLOSED:
+            /* We're either already closed or managed to mark ourselves for
+             * closure in the previous iteration. */
+            return;
+        case EFILE_STATE_BUSY:
+            /* Schedule ourselves to be closed once the current operation
+             * finishes, retrying the [IDLE -> CLOSED] transition in case we
+             * narrowly passed the [BUSY -> IDLE] one. */
+            erts_atomic32_cmpxchg_nob(&d->state,
+                EFILE_STATE_CLOSE_PENDING, EFILE_STATE_BUSY);
+            break;
+        }
+    }
+}
+
+static void gc_callback(ErlNifEnv *env, void* data) {
+    efile_data_t *d = (efile_data_t*)data;
+
+    enum efile_state_t previous_state;
+
+    (void)env;
+
+    previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+        EFILE_STATE_CLOSED, EFILE_STATE_IDLE);
+
+    ASSERT(previous_state != EFILE_STATE_CLOSE_PENDING &&
+        previous_state != EFILE_STATE_BUSY);
+
+    if(previous_state == EFILE_STATE_IDLE) {
+        efile_close(d);
+    }
+}
+
+static ERL_NIF_TERM efile_filetype_to_atom(enum efile_filetype_t type) {
+    switch(type) {
+        case EFILE_FILETYPE_DEVICE: return am_device;
+        case EFILE_FILETYPE_DIRECTORY: return am_directory;
+        case EFILE_FILETYPE_REGULAR: return am_regular;
+        case EFILE_FILETYPE_SYMLINK: return am_symlink;
+        case EFILE_FILETYPE_OTHER: return am_other;
+    }
+
+    return am_other;
+}
+
+static ERL_NIF_TERM efile_access_to_atom(enum efile_access_t type) {
+    if(type & EFILE_ACCESS_READ && !(type & EFILE_ACCESS_WRITE)) {
+        return am_read;
+    } else if(type & EFILE_ACCESS_WRITE && !(type & EFILE_ACCESS_READ)) {
+        return am_write;
+    } else if(type & EFILE_ACCESS_READ_WRITE) {
+        return am_read_write;
+    }
+
+    return am_none;
+}
+
+static enum efile_modes_t efile_translate_modelist(ErlNifEnv *env, ERL_NIF_TERM list) {
+    enum efile_modes_t modes;
+    ERL_NIF_TERM head, tail;
+
+    modes = 0;
+
+    while(enif_get_list_cell(env, list, &head, &tail)) {
+        if(enif_is_identical(head, am_read)) {
+            modes |= EFILE_MODE_READ;
+        } else if(enif_is_identical(head, am_write)) {
+            modes |= EFILE_MODE_WRITE;
+        } else if(enif_is_identical(head, am_exclusive)) {
+            modes |= EFILE_MODE_EXCLUSIVE;
+        } else if(enif_is_identical(head, am_append)) {
+            modes |= EFILE_MODE_APPEND;
+        } else if(enif_is_identical(head, am_sync)) {
+            modes |= EFILE_MODE_SYNC;
+        } else if(enif_is_identical(head, am_skip_type_check)) {
+            modes |= EFILE_MODE_SKIP_TYPE_CHECK;
+        } else {
+            /* Modes like 'raw', 'ram', 'delayed_writes' etc are handled
+             * further up the chain. */
+        }
+
+        list = tail;
+    }
+
+    if(modes & (EFILE_MODE_APPEND | EFILE_MODE_EXCLUSIVE)) {
+        /* 'append' and 'exclusive' are documented as "open for writing." */
+        modes |= EFILE_MODE_WRITE;
+    } else if(!(modes & EFILE_MODE_READ_WRITE)) {
+        /* Defaulting to read if !(W|R) is undocumented, but specifically
+         * tested against in file_SUITE. */
+        modes |= EFILE_MODE_READ;
+    }
+
+    return modes;
+}
+
+static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+    efile_data_t *d;
+
+    ErlNifPid controlling_process;
+    enum efile_modes_t modes;
+    ERL_NIF_TERM result;
+    efile_path_t path;
+
+    if(argc != 2 || !enif_is_list(env, argv[1])) {
+        return enif_make_badarg(env);
+    }
+
+    modes = efile_translate_modelist(env, argv[1]);
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_open(&path, modes, efile_resource_type, &d))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    result = enif_make_resource(env, d);
+    enif_release_resource(d);
+
+    enif_self(env, &controlling_process);
+
+    if(enif_monitor_process(env, d, &controlling_process, &d->monitor)) {
+        return posix_error_to_tuple(env, EINVAL);
+    }
+
+    return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM close_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    enum efile_state_t previous_state;
+
+    if(argc != 0) {
+        return enif_make_badarg(env);
+    }
+
+    previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+        EFILE_STATE_CLOSED, EFILE_STATE_BUSY);
+
+    ASSERT(previous_state == EFILE_STATE_CLOSE_PENDING ||
+        previous_state == EFILE_STATE_BUSY);
+
+    if(previous_state == EFILE_STATE_BUSY) {
+        enif_demonitor_process(env, d, &d->monitor);
+
+        if(!efile_close(d)) {
+            return posix_error_to_tuple(env, d->posix_errno);
+        }
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM read_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    Sint64 bytes_read, block_size;
+    SysIOVec read_vec[1];
+    ErlNifBinary result;
+
+    if(argc != 1 || !enif_is_number(env, argv[0])) {
+        return enif_make_badarg(env);
+    }
+
+    if(!enif_get_int64(env, argv[0], &block_size) || block_size < 0) {
+        return posix_error_to_tuple(env, EINVAL);
+    }
+
+    if(!enif_alloc_binary(block_size, &result)) {
+        return posix_error_to_tuple(env, ENOMEM);
+    }
+
+    read_vec[0].iov_base = result.data;
+    read_vec[0].iov_len = result.size;
+
+    bytes_read = efile_readv(d, read_vec, 1);
+    ASSERT(bytes_read <= block_size);
+
+    if(bytes_read < 0) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    } else if(bytes_read == 0) {
+        enif_release_binary(&result);
+        return am_eof;
+    }
+
+    if(bytes_read < block_size && !enif_realloc_binary(&result, bytes_read)) {
+        ERTS_INTERNAL_ERROR("Failed to shrink read result.");
+    }
+
+    return enif_make_tuple2(env, am_ok, enif_make_binary(env, &result));
+}
+
+static ERL_NIF_TERM write_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    ErlNifIOVec vec, *input = &vec;
+    Sint64 bytes_written;
+    ERL_NIF_TERM tail;
+
+    if(argc != 1 || !enif_inspect_iovec(env, 64, argv[0], &tail, &input)) {
+        return enif_make_badarg(env);
+    }
+
+    bytes_written = efile_writev(d, input->iov, input->iovcnt);
+
+    if(bytes_written < 0) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    }
+
+    if(!enif_is_empty_list(env, tail)) {
+        ASSERT(bytes_written > 0);
+        return enif_make_tuple2(env, am_continue, tail);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM pread_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    Sint64 bytes_read, block_size, offset;
+    SysIOVec read_vec[1];
+    ErlNifBinary result;
+
+    if(argc != 2 || !enif_is_number(env, argv[0])
+                 || !enif_is_number(env, argv[1])) {
+        return enif_make_badarg(env);
+    }
+
+    if(!enif_get_int64(env, argv[0], &offset) ||
+       !enif_get_int64(env, argv[1], &block_size) ||
+       (offset < 0 || block_size < 0)) {
+        return posix_error_to_tuple(env, EINVAL);
+    }
+
+    if(!enif_alloc_binary(block_size, &result)) {
+        return posix_error_to_tuple(env, ENOMEM);
+    }
+
+    read_vec[0].iov_base = result.data;
+    read_vec[0].iov_len = result.size;
+
+    bytes_read = efile_preadv(d, offset, read_vec, 1);
+
+    if(bytes_read < 0) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    } else if(bytes_read == 0) {
+        enif_release_binary(&result);
+        return am_eof;
+    }
+
+    if(bytes_read < block_size && !enif_realloc_binary(&result, bytes_read)) {
+        ERTS_INTERNAL_ERROR("Failed to shrink pread result.");
+    }
+
+    return enif_make_tuple2(env, am_ok, enif_make_binary(env, &result));
+}
+
+static ERL_NIF_TERM pwrite_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    ErlNifIOVec vec, *input = &vec;
+    Sint64 bytes_written, offset;
+    ERL_NIF_TERM tail;
+
+    if(argc != 2 || !enif_is_number(env, argv[0])
+                 || !enif_inspect_iovec(env, 64, argv[1], &tail, &input)) {
+        return enif_make_badarg(env);
+    }
+
+    if(!enif_get_int64(env, argv[0], &offset) || offset < 0) {
+        return posix_error_to_tuple(env, EINVAL);
+    }
+
+    bytes_written = efile_pwritev(d, offset, input->iov, input->iovcnt);
+
+    if(bytes_written < 0) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    }
+
+    if(!enif_is_empty_list(env, tail)) {
+        ASSERT(bytes_written > 0);
+        return enif_make_tuple3(env, am_continue,
+            enif_make_int64(env, bytes_written), tail);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM seek_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    Sint64 new_position, offset;
+    enum efile_seek_t seek;
+
+    if(argc != 2 || !enif_get_int64(env, argv[1], &offset)) {
+        return enif_make_badarg(env);
+    }
+
+    if(enif_is_identical(argv[0], am_bof)) {
+        seek = EFILE_SEEK_BOF;
+    } else if(enif_is_identical(argv[0], am_cur)) {
+        seek = EFILE_SEEK_CUR;
+    } else if(enif_is_identical(argv[0], am_eof)) {
+        seek = EFILE_SEEK_EOF;
+    } else {
+        return enif_make_badarg(env);
+    }
+
+    if(!efile_seek(d, seek, offset, &new_position)) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    }
+
+    return enif_make_tuple2(env, am_ok, enif_make_uint64(env, new_position));
+}
+
+static ERL_NIF_TERM sync_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    int data_only;
+
+    if(argc != 1 || !enif_get_int(env, argv[0], &data_only)) {
+        return enif_make_badarg(env);
+    }
+
+    if(!efile_sync(d, data_only)) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM truncate_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    if(argc != 0) {
+        return enif_make_badarg(env);
+    }
+
+    if(!efile_truncate(d)) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM allocate_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    Sint64 offset, length;
+
+    if(argc != 2 || !enif_is_number(env, argv[0])
+                 || !enif_is_number(env, argv[1])) {
+        return enif_make_badarg(env);
+    }
+
+    if(!enif_get_int64(env, argv[0], &offset) ||
+       !enif_get_int64(env, argv[1], &length) ||
+       (offset < 0 || length < 0)) {
+        return posix_error_to_tuple(env, EINVAL);
+    }
+
+    if(!efile_allocate(d, offset, length)) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM advise_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    enum efile_advise_t advise;
+    Sint64 offset, length;
+
+    if(argc != 3 || !enif_is_number(env, argv[0])
+                 || !enif_is_number(env, argv[1])) {
+        return enif_make_badarg(env);
+    }
+
+    if(!enif_get_int64(env, argv[0], &offset) ||
+       !enif_get_int64(env, argv[1], &length) ||
+       (offset < 0 || length < 0)) {
+        return posix_error_to_tuple(env, EINVAL);
+    }
+
+    if(enif_is_identical(argv[2], am_normal)) {
+        advise = EFILE_ADVISE_NORMAL;
+    } else if(enif_is_identical(argv[2], am_random)) {
+        advise = EFILE_ADVISE_RANDOM;
+    } else if(enif_is_identical(argv[2], am_sequential)) {
+        advise = EFILE_ADVISE_SEQUENTIAL;
+    } else if(enif_is_identical(argv[2], am_will_need)) {
+        advise = EFILE_ADVISE_WILL_NEED;
+    } else if(enif_is_identical(argv[2], am_dont_need)) {
+        advise = EFILE_ADVISE_DONT_NEED;
+    } else if(enif_is_identical(argv[2], am_no_reuse)) {
+        advise = EFILE_ADVISE_NO_REUSE;
+    } else {
+        /* The tests check for EINVAL instead of badarg. Sigh. */
+        return posix_error_to_tuple(env, EINVAL);
+    }
+
+    if(!efile_advise(d, offset, length, advise)) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    }
+
+    return am_ok;
+}
+
+/* This undocumented function reads a pointer and then reads the data block
+ * described by said pointer. It was reverse-engineered from the old
+ * implementation so while all tests pass it may not be entirely correct. Our
+ * current understanding is as follows:
+ *
+ * Pointer layout:
+ *
+ *     <<Size:1/integer-unit:32, Offset:1/integer-unit:32>>
+ *
+ * Where Offset is the -absolute- address to the data block.
+ *
+ * *) If we fail to read the pointer block in its entirety, we return eof.
+ * *) If the provided max_payload_size is larger than Size, we return eof.
+ * *) If we fail to read any data whatsoever at Offset, we return
+ *    {ok, {Size, Offset, eof}}
+ * *) Otherwise, we return {ok, {Size, Offset, Data}}. Note that the size
+ *    of Data may be smaller than Size if we encounter EOF before we could
+ *    read the entire block.
+ *
+ * On errors we'll return {error, posix()} regardless of whether they
+ * happened before or after reading the pointer block. */
+static ERL_NIF_TERM ipread_s32bu_p32bu_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    Sint64 payload_offset, payload_size;
+
+    SysIOVec read_vec[1];
+    Sint64 bytes_read;
+
+    ErlNifBinary payload;
+
+    if(argc != 2 || !enif_is_number(env, argv[0])
+                 || !enif_is_number(env, argv[1])) {
+        return enif_make_badarg(env);
+    }
+
+    {
+        Sint64 max_payload_size, pointer_offset;
+        unsigned char pointer_block[8];
+
+        if(!enif_get_int64(env, argv[0], &pointer_offset) ||
+           !enif_get_int64(env, argv[1], &max_payload_size) ||
+           (pointer_offset < 0 || max_payload_size >= 1u << 31)) {
+            return posix_error_to_tuple(env, EINVAL);
+        }
+
+        read_vec[0].iov_base = pointer_block;
+        read_vec[0].iov_len = sizeof(pointer_block);
+
+        bytes_read = efile_preadv(d, pointer_offset, read_vec, 1);
+
+        if(bytes_read < 0) {
+            return posix_error_to_tuple(env, d->posix_errno);
+        } else if(bytes_read < sizeof(pointer_block)) {
+            return am_eof;
+        }
+
+        payload_size = (Uint32)get_int32(&pointer_block[0]);
+        payload_offset = (Uint32)get_int32(&pointer_block[4]);
+
+        if(payload_size > max_payload_size) {
+            return am_eof;
+        }
+    }
+
+    if(!enif_alloc_binary(payload_size, &payload)) {
+        return posix_error_to_tuple(env, ENOMEM);
+    }
+
+    read_vec[0].iov_base = payload.data;
+    read_vec[0].iov_len = payload.size;
+
+    bytes_read = efile_preadv(d, payload_offset, read_vec, 1);
+
+    if(bytes_read < 0) {
+        return posix_error_to_tuple(env, d->posix_errno);
+    } else if(bytes_read == 0) {
+        enif_release_binary(&payload);
+
+        return enif_make_tuple2(env, am_ok,
+                enif_make_tuple3(env,
+                    enif_make_uint(env, payload_size),
+                    enif_make_uint(env, payload_offset),
+                    am_eof));
+    }
+
+    if(bytes_read < payload.size && !enif_realloc_binary(&payload, bytes_read)) {
+        ERTS_INTERNAL_ERROR("Failed to shrink ipread payload.");
+    }
+
+    return enif_make_tuple2(env, am_ok,
+        enif_make_tuple3(env,
+            enif_make_uint(env, payload_size),
+            enif_make_uint(env, payload_offset),
+            enif_make_binary(env, &payload)));
+}
+
+static ERL_NIF_TERM get_handle_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    if(argc != 0) {
+        return enif_make_badarg(env);
+    }
+
+    return efile_get_handle(env, d);
+}
+
+static ERL_NIF_TERM read_info_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_fileinfo_t info = {0};
+    efile_path_t path;
+    int follow_links;
+
+    if(argc != 2 || !enif_get_int(env, argv[1], &follow_links)) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_read_info(&path, follow_links, &info))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    /* #file_info as declared in file.hrl */
+    return enif_make_tuple(env, 14,
+        am_file_info,
+        enif_make_uint64(env, info.size),
+        efile_filetype_to_atom(info.type),
+        efile_access_to_atom(info.access),
+        enif_make_int64(env, MAX(EFILE_MIN_FILETIME, info.a_time)),
+        enif_make_int64(env, MAX(EFILE_MIN_FILETIME, info.m_time)),
+        enif_make_int64(env, MAX(EFILE_MIN_FILETIME, info.c_time)),
+        enif_make_uint(env, info.mode),
+        enif_make_uint(env, info.links),
+        enif_make_uint(env, info.major_device),
+        enif_make_uint(env, info.minor_device),
+        enif_make_uint(env, info.inode),
+        enif_make_uint(env, info.uid),
+        enif_make_uint(env, info.gid)
+    );
+}
+
+static ERL_NIF_TERM set_permissions_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+    Uint32 permissions;
+
+    if(argc != 2 || !enif_get_uint(env, argv[1], &permissions)) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_set_permissions(&path, permissions))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM set_owner_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+    Uint32 uid, gid;
+
+    if(argc != 3 || !enif_get_uint(env, argv[1], &uid)
+                 || !enif_get_uint(env, argv[2], &gid)) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_set_owner(&path, uid, gid))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM set_time_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    Sint64 accessed, modified, created;
+    efile_path_t path;
+
+    if(argc != 4 || !enif_get_int64(env, argv[1], &accessed)
+                 || !enif_get_int64(env, argv[2], &modified)
+                 || !enif_get_int64(env, argv[3], &created)) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_set_time(&path, accessed, modified, created))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM read_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+    ERL_NIF_TERM result;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_read_link(env, &path, &result))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM list_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+    ERL_NIF_TERM result;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_list_dir(env, &path, &result))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM rename_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t existing_path, new_path;
+
+    if(argc != 2) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &existing_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_marshal_path(env, argv[1], &new_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_rename(&existing_path, &new_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM make_hard_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t existing_path, new_path;
+
+    if(argc != 2) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &existing_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_marshal_path(env, argv[1], &new_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_make_hard_link(&existing_path, &new_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM make_soft_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t existing_path, new_path;
+
+    if(argc != 2) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &existing_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_marshal_path(env, argv[1], &new_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_make_soft_link(&existing_path, &new_path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM make_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_make_dir(&path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM del_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_del_file(&path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM del_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_del_dir(&path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+static ERL_NIF_TERM get_device_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    ERL_NIF_TERM result;
+    int device_index;
+
+    if(argc != 1 || !enif_get_int(env, argv[0], &device_index)) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_get_device_cwd(env, device_index, &result))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM get_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+    ERL_NIF_TERM result;
+
+    if(argc != 0) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_get_cwd(env, &result))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM set_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_set_cwd(&path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return am_ok;
+}
+
+/** @brief Reads an entire file into \c result, stopping after \c size bytes or
+ * EOF. It will read until EOF if size is 0. */
+static posix_errno_t read_file(efile_data_t *d, size_t size, ErlNifBinary *result) {
+    size_t initial_buffer_size;
+    ssize_t bytes_read;
+
+    if(size == 0) {
+        initial_buffer_size = 16 << 10;
+    } else {
+        initial_buffer_size = size;
+    }
+
+    if(!enif_alloc_binary(initial_buffer_size, result)) {
+        return ENOMEM;
+    }
+
+    bytes_read = 0;
+
+    for(;;) {
+        ssize_t block_bytes_read;
+        SysIOVec read_vec[1];
+
+        read_vec[0].iov_base = result->data + bytes_read;
+        read_vec[0].iov_len = result->size - bytes_read;
+
+        block_bytes_read = efile_readv(d, read_vec, 1);
+
+        if(block_bytes_read < 0) {
+            enif_release_binary(result);
+            return d->posix_errno;
+        }
+
+        bytes_read += block_bytes_read;
+
+        if(block_bytes_read < (result->size - bytes_read)) {
+            /* EOF */
+            break;
+        } else if(bytes_read == size) {
+            break;
+        }
+
+        if(!enif_realloc_binary(result, bytes_read * 2)) {
+            enif_release_binary(result);
+            return ENOMEM;
+        }
+    }
+
+    /* The file may have shrunk since we queried its size, so we have to do
+     * this even when the size is known. */
+    if(bytes_read < result->size && !enif_realloc_binary(result, bytes_read)) {
+        ERTS_INTERNAL_ERROR("Failed to shrink read_file result.");
+    }
+
+    return 0;
+}
+
+static ERL_NIF_TERM read_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_fileinfo_t info = {0};
+    efile_path_t path;
+    efile_data_t *d;
+
+    ErlNifBinary result;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_read_info(&path, 1, &info))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_open(&path, EFILE_MODE_READ, efile_resource_type, &d))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    posix_errno = read_file(d, info.size, &result);
+    enif_release_resource(d);
+
+    if(posix_errno) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return enif_make_tuple2(env, am_ok, enif_make_binary(env, &result));
+}
+
+static ERL_NIF_TERM altname_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+    posix_errno_t posix_errno;
+
+    efile_path_t path;
+    ERL_NIF_TERM result;
+
+    if(argc != 1) {
+        return enif_make_badarg(env);
+    }
+
+    if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+        return posix_error_to_tuple(env, posix_errno);
+    } else if((posix_errno = efile_altname(env, &path, &result))) {
+        return posix_error_to_tuple(env, posix_errno);
+    }
+
+    return enif_make_tuple2(env, am_ok, result);
+}
diff --git a/erts/emulator/nifs/common/prim_file_nif.h b/erts/emulator/nifs/common/prim_file_nif.h
new file mode 100644
index 0000000000..cc9bc8f5c3
--- /dev/null
+++ b/erts/emulator/nifs/common/prim_file_nif.h
@@ -0,0 +1,240 @@
+/*
+ * %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%
+ */
+
+typedef int posix_errno_t;
+
+enum efile_modes_t {
+    EFILE_MODE_READ = (1 << 0),
+    EFILE_MODE_WRITE = (1 << 1), /* Implies truncating file when used alone. */
+    EFILE_MODE_APPEND = (1 << 2),
+    EFILE_MODE_EXCLUSIVE = (1 << 3),
+    EFILE_MODE_SYNC = (1 << 4),
+
+    EFILE_MODE_SKIP_TYPE_CHECK = (1 << 5), /* Special for device files on Unix. */
+    EFILE_MODE_NO_TRUNCATE = (1 << 6), /* Special for reopening on VxWorks. */
+
+    EFILE_MODE_READ_WRITE = EFILE_MODE_READ | EFILE_MODE_WRITE
+};
+
+enum efile_access_t {
+    EFILE_ACCESS_NONE = 0,
+    EFILE_ACCESS_READ = 1,
+    EFILE_ACCESS_WRITE = 2,
+    EFILE_ACCESS_READ_WRITE = EFILE_ACCESS_READ | EFILE_ACCESS_WRITE
+};
+
+enum efile_seek_t {
+    EFILE_SEEK_BOF,
+    EFILE_SEEK_CUR,
+    EFILE_SEEK_EOF
+};
+
+enum efile_filetype_t {
+    EFILE_FILETYPE_DEVICE,
+    EFILE_FILETYPE_DIRECTORY,
+    EFILE_FILETYPE_REGULAR,
+    EFILE_FILETYPE_SYMLINK,
+    EFILE_FILETYPE_OTHER
+};
+
+enum efile_advise_t {
+    EFILE_ADVISE_NORMAL,
+    EFILE_ADVISE_RANDOM,
+    EFILE_ADVISE_SEQUENTIAL,
+    EFILE_ADVISE_WILL_NEED,
+    EFILE_ADVISE_DONT_NEED,
+    EFILE_ADVISE_NO_REUSE
+};
+
+enum efile_state_t {
+    EFILE_STATE_IDLE = 0,
+    EFILE_STATE_BUSY = 1,
+    EFILE_STATE_CLOSE_PENDING = 2,
+    EFILE_STATE_CLOSED = 3
+};
+
+typedef struct {
+    Sint64 size;            /* Size of file */
+    Uint32 type;            /* Type of file -- one of EFILE_FILETYPE_*. */
+    Uint32 access;          /* Access to file -- one of EFILE_ACCESS_*. */
+    Uint32 mode;            /* Access permissions -- bit field. */
+    Uint32 links;           /* Number of links to file. */
+    Uint32 major_device;    /* Major device or file system. */
+    Uint32 minor_device;    /* Minor device (for devices). */
+    Uint32 inode;           /* Inode number. */
+    Uint32 uid;             /* User id of owner. */
+    Uint32 gid;             /* Group id of owner. */
+    Sint64 a_time;          /* Last time the file was accessed. */
+    Sint64 m_time;          /* Last time the file was modified. */
+    Sint64 c_time;          /* Windows: creation time, Unix: last inode
+                             * change. */
+} efile_fileinfo_t;
+
+/* The smallest value that can be converted freely between universal, local,
+ * and POSIX time, as required by read_file_info/2. Corresponds to
+ * {{1902,1,1},{0,0,0}} */
+#define EFILE_MIN_FILETIME -2145916800
+
+/* Initializes an efile_data_t; must be used in efile_open on success. */
+#define EFILE_INIT_RESOURCE(__d, __modes) do { \
+        erts_atomic32_init_acqb(&(__d)->state, EFILE_STATE_IDLE); \
+        (__d)->posix_errno = 0; \
+        (__d)->modes = __modes; \
+    } while(0)
+
+typedef struct {
+    erts_atomic32_t state;
+
+    posix_errno_t posix_errno;
+    enum efile_modes_t modes;
+
+    ErlNifMonitor monitor;
+} efile_data_t;
+
+typedef ErlNifBinary efile_path_t;
+
+/* @brief Translates the given "raw name" into the format expected by the APIs
+ * used by the underlying implementation. The result is transient and does not
+ * need to be released.
+ *
+ * This may change the structure of the path and its results should never be
+ * passed on to the user. Refer to the OS-specific implementation for details.
+ *
+ * @param path The term to translate; it must have been encoded with
+ * prim_file:internal_native2name for compatibility reasons. */
+posix_errno_t efile_marshal_path(ErlNifEnv *env, ERL_NIF_TERM path, efile_path_t *result);
+
+/* @brief Returns the underlying handle as an implementation-defined term.
+ *
+ * This is an internal function intended to support tests and tricky
+ * operations like sendfile(2). */
+ERL_NIF_TERM efile_get_handle(ErlNifEnv *env, efile_data_t *d);
+
+/* @brief Read until EOF or the given iovec has been filled.
+ *
+ * @return -1 on failure, or the number of bytes read on success. The return
+ * value will be 0 if no bytes could be read before EOF or the end of the
+ * iovec. */
+Sint64 efile_readv(efile_data_t *d, SysIOVec *iov, int iovlen);
+
+/* @brief Write the entirety of the given iovec.
+ *
+ * @return -1 on failure, or the number of bytes written on success. "Partial"
+ * failures will be reported with -1 and not the number of bytes we managed to
+ * write to disk before the failure. */
+Sint64 efile_writev(efile_data_t *d, SysIOVec *iov, int iovlen);
+
+/* @brief As \c efile_readv, but starting from a file offset. */
+Sint64 efile_preadv(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen);
+
+/* @brief As \c efile_writev, but starting from a file offset. */
+Sint64 efile_pwritev(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen);
+
+int efile_seek(efile_data_t *d, enum efile_seek_t seek, Sint64 offset, Sint64 *new_position);
+
+int efile_sync(efile_data_t *d, int data_only);
+
+int efile_advise(efile_data_t *d, Sint64 offset, Sint64 length, enum efile_advise_t advise);
+int efile_allocate(efile_data_t *d, Sint64 offset, Sint64 length);
+int efile_truncate(efile_data_t *d);
+
+posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes,
+        ErlNifResourceType *nif_type, efile_data_t **d);
+
+/** @brief Closes a file. The file must have entered the CLOSED state prior to
+ * calling this to prevent double close. */
+int efile_close(efile_data_t *d);
+
+/* **** **** **** **** **** **** **** **** **** **** **** **** **** **** **** */
+
+posix_errno_t efile_read_info(const efile_path_t *path, int follow_link, efile_fileinfo_t *result);
+
+/** @brief Sets the file times to the given values. Refer to efile_fileinfo_t
+ * for a description of each. */
+posix_errno_t efile_set_time(const efile_path_t *path, Sint64 a_time, Sint64 m_time, Sint64 c_time);
+
+/** @brief On Unix, this sets the file permissions according to the docs for
+ * file:write_file_info/2. On Windows it uses the "owner write permission" flag
+ * to toggle whether the file is read-only or not. */
+posix_errno_t efile_set_permissions(const efile_path_t *path, Uint32 permissions);
+
+/** @brief On Unix, this will set the owner/group to the given values. It will
+ * do nothing on other platforms. */
+posix_errno_t efile_set_owner(const efile_path_t *path, Uint32 owner, Uint32 group);
+
+/** @brief Resolves the final path of the given link. */
+posix_errno_t efile_read_link(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result);
+
+/** @brief Lists the contents of the given directory.
+ * @param result [out] A list of all the directory/file names contained in the
+ * given directory. */
+posix_errno_t efile_list_dir(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result);
+
+/** @brief Changes the name of an existing file or directory, from old_path
+ * to new_path.
+ *
+ * If old_path and new_path refer to the same file or directory, it does
+ * nothing and returns success. Otherwise if new_path already exists, it will
+ * be deleted and replaced by src subject to the following conditions:
+ *
+ *     If old_path is a directory, new_path may be an empty directory.
+ *     If old_path is a file, new_path may be a file.
+ *
+ * Neither of these are guaranteed to be atomic. In any other situation where
+ * new_path already exists, the rename will fail.
+ *
+ * Some possible error codes:
+ *
+ * - EACCES:     Either paths or one of their parent directories can't be read
+ *               and/or written.
+ * - EEXIST:     new_path is a non-empty directory.
+ * - EINVAL:     old_path is a root directory or new_path is a subdirectory
+ *               of new_path.
+ * - EISDIR:     new_path is a directory, but old_path is not.
+ * - ENOTDIR:    old_path is a directory, but new_path is not.
+ * - ENOENT:     old_path doesn't exist, or either path is "".
+ * - EXDEV:      The paths are on different filesystems.
+ *
+ * The implementation of rename may allow cross-filesystem renames,
+ * but the caller should be prepared to emulate it with copy and
+ * delete if errno is EXDEV. */
+posix_errno_t efile_rename(const efile_path_t *old_path, const efile_path_t *new_path);
+
+posix_errno_t efile_make_hard_link(const efile_path_t *existing_path, const efile_path_t *new_path);
+posix_errno_t efile_make_soft_link(const efile_path_t *existing_path, const efile_path_t *new_path);
+posix_errno_t efile_make_dir(const efile_path_t *path);
+
+posix_errno_t efile_del_file(const efile_path_t *path);
+posix_errno_t efile_del_dir(const efile_path_t *path);
+
+posix_errno_t efile_get_cwd(ErlNifEnv *env, ERL_NIF_TERM *result);
+posix_errno_t efile_set_cwd(const efile_path_t *path);
+
+/** @brief A Windows-specific function for returning the working directory of a
+ * given device.
+ *
+ * @param device_index The drive index; 1 for A, 2 for B, etc.
+ * @param result [out] The working directory of the given device
+ */
+posix_errno_t efile_get_device_cwd(ErlNifEnv *env, int device_index, ERL_NIF_TERM *result);
+
+/** @brief A Windows-specific function for returning the 8.3-name of a given
+ * file or directory. */
+posix_errno_t efile_altname(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result);