Let's say I have a function defined in Rust, which looks like this:
#[no_mangle]
pub unsafe extern "C" fn do_something(
my_value: *mut MyStruct,
some_param: c_uint,
content: *mut *mut u8,
length: *mut c_uint,
capacity: *mut c_uint,
) -> *mut MyStruct {
// Do something and obtain an `ffi_result`...
let heap_data = ffi_result.as_mut_ptr();
// These values are passed as "out" parameters.
*length = ffi_result.len() as c_uint;
*capacity = ffi_result.capacity() as c_uint;
*content = heap_data;
// We intentionally "leak" this data to the heap.
// The caller is responsible for cleaning it up by calling another function.
std::mem::forget(ffi_result);
std::boxed::Box::into_raw(value_of_type_my_struct)
}
It takes in a pointer to a struct, a simple integer parameter, several out parameters that can later be used to create an Array and it returns a pointer to a struct.
Now I compile the rust library into a static library for the target aarch64-apple-ios. I set up a XCode project, add the static library as a dependency as explained here with an "Objective-C Bridging Header" where I import the following header file
#ifndef libmy_project_h
#define libmy_project_h
#include <stdint.h>
struct myStruct;
struct myStruct *do_something(struct myStruct *state, int someParam, char **content, int *length, int *capacity);
#endif
Up until this point everything seems to work fine and I have already successfully used this procedure for a whole bunch of other functions. However in this special case I can not figure out how to call this function from swift. I need to call the function from swift and pass content, length and capacity as out parameters so that I can later use the pointers to create an Array in Swift like so.
This is what I tried so far:
var content = UnsafeMutablePointer<UnsafeMutablePointer<CChar>?>(UnsafeMutablePointer(bitPattern: 0))
var length = UnsafeMutablePointer<Int32>(bitPattern: 0)
var capacity = UnsafeMutablePointer<Int32>(bitPattern: 0)
let my_struct = do_something(my_struct, Int32(some_param), content, length, capacity)
let buffer = UnsafeRawBufferPointer(start: content?.pointee, count: Int(length!.pointee))
var data = Array(repeating: UInt8(0), count: Int(length!.pointee))
data.withUnsafeMutableBytes { arrayPtr in
arrayPtr.copyBytes(from: buffer)
}
However now when I execute this swift snippet, I get an EXC_BAD_ACCESS error, which I think occurs because the pointers I manually created do not belong to the adress space of my application. How can I create pointers that I can use as out parameters?
P.S. For reference here is the same interop code in C#:
[DllImport("my_dll")]
private static extern IntPtr do_something(IntPtr state, uint someParam, out IntPtr content, out uint length, out uint capacity);
Which can be called like so:
IntPtr contentPointer;
uint length, capacity;
IntPtr my_struct = do_something(state, myParam, out contentPointer, out length, out capacity);
byte[] rawContent = new byte[length];
Marshal.Copy(contentPointer, rawContent, 0, (int)length);
// Free the data owned by rust with another FFI call:
free_do_something_result(contentPointer, length, capacity);
var length = UnsafeMutablePointer<Int32>(bitPattern: 0)
You need to pass storage for your out-parameters. This is defining a null pointer. When Rust tries to write the result to address 0, it crashes, since you don't have access to write there.
Instead of creating two layers of pointers, create a value of the type you want, and then pass the address (&) of that value; this will add the extra layer of pointer automatically.
// Create storage
var content: UnsafeMutablePointer<CChar>? // Could be NULL, so Optional
var length: Int32 = 0
var capacity: Int32 = 0
// Pass as references
do_something(&content, &length, &capacity)
// Copy the data
let data = Array(UnsafeRawBufferPointer(start: content, count: Int(length)))
content is still a pointer here because the thing being updated is a pointer. You're not providing storage for content, Rust is. But you do need to provide storage for the pointer (and that's what this does).
I can't compile your code because it's missing a lot (an MCVE would be much better here), so I can't test that this is doing exactly what you mean, but it should be close.
In your example, you're leaking the memory, but since your C# calls free_do_something_result (which I assume cleans it up), I assume you're actually doing the same in the Swift.
Related
I'm trying to optimize as much as possible an operation done on slices of u32 from arrays of u8. As such, I'm testing different options (for loops, iterators, using ByteOrder crate, etc.)
As part of these tests, I also wanted to check out if I could improve it using from_raw_parts standard function.
Here is my code:
use byteorder::{ByteOrder, BigEndian, LittleEndian};
use std::io::Read;
fn main(){
let random_bytes = (0..4).map(|_| { rand::random::<u8>() }).collect::<Vec<u8>>();
let random_bytes = random_bytes.as_slice();
let view = &random_bytes as *const _ as *const u32;
let slice: &[u32] = unsafe { std::slice::from_raw_parts(view, 1) };
println!("{:x?}", slice);
println!("{:x?}", LittleEndian::read_u32(&random_bytes[0..4]));
println!("{:x?}", BigEndian::read_u32(&random_bytes[0..4]));
println!("{:x?}", &random_bytes[0..4]);
}
I would've expected at least one one of the two Little or Big endian to be equal to the first print, but instead this does not seem to be the case, e.g. an example output
[d951db30]
143600ff
ff003614
[ff, 0, 36, 14]
What am I doing wrong?
The problem is here:
let view = &random_bytes as *const _ as *const u32;
A slice is a 2-machine-word struct that contains the pointer to data and the element count as members.
By doing &random_bytes, you are taking a reference to this slice structure (which itself contains a pointer and length), not acquiring the underlying pointer.
Slices have an as_ptr method that returns the pointer to data itself. When you use it, your code functions correctly:
use byteorder::{ByteOrder, BigEndian, LittleEndian};
use std::io::Read;
fn main(){
let random_bytes = (0..4).map(|_| { rand::random::<u8>() }).collect::<Vec<u8>>();
let random_bytes = random_bytes.as_slice();
let view = random_bytes.as_ptr() as *const u32;
let slice: &[u32] = unsafe { std::slice::from_raw_parts(view, 1) };
println!("{:x?}", slice);
println!("{:x?}", LittleEndian::read_u32(&random_bytes[0..4]));
println!("{:x?}", BigEndian::read_u32(&random_bytes[0..4]));
println!("{:x?}", &random_bytes[0..4]);
}
Output in playground:
[684a2f5b]
684a2f5b
5b2f4a68
[5b, 2f, 4a, 68]
i am trying to figure out how the code behind a basic kernel driver works.
I have the following struct:
static struct file_operations fops =
{
.open = dev_open,
.read = dev_read,
.write = dev_write,
.release = dev_release,
};
And my dev_open function is defined as:
static int dev_open(struct inode *, struct file *);
Now Im also familiar with the fact that the prototype for opening a device file is defined in the linux/fs.h:
http://lxr.linux.no/linux+v3.10/include/linux/fs.h#L1517
Here is the specific line from that link:
int (*open) (struct inode *, struct file *);
Now my question is what is the relationship between .open = dev_open, and int (*open) (struct inode *, struct file *);
which is defined in linux/fs.h? Is it passing the address of dev_open to the function pointer int (*open) defined in the linux/fs.h? There must be some relation or what is the point of defining the struct fops as type "file operation"?
A similar question was asked and answered here but i feel that my question was left out:
File operations in drivers
Thank you
I think this question is more about C than the Linux kernel.
Members of structure or union types cannot have function type, but they can have pointer to function type. For example, in the Linux kernel, the open member of struct file_operations needs to be declared with a pointer to function type: int (*open)(struct inode *, struct file *);. Declaring the member as int open(struct inode *, struct file *); is an error.
In this variable definition in Linux kernel code:
static struct file_operations fops =
{
.open = dev_open,
.read = dev_read,
.write = dev_write,
.release = dev_release,
};
Incidentally, the above should normally have the owner member initialized like so:
.owner = THIS_MODULE,
The expressions dev_open, dev_read, dev_write and dev_release are function designators being used as assignment expressions to initialize the members of fops. A function designator is an expression that has function type. Unless it is the operand of sizeof, _Alignof, or the unary & operator, a function designator is converted to a pointer to function type. Therefore, the above definition of variable foo is exactly equivalent to:
static struct file_operations fops =
{
.open = &dev_open,
.read = &dev_read,
.write = &dev_write,
.release = &dev_release,
};
(Don't forget to also initialize .owner = THIS_MODULE,.)
There, the function designators are operands of the unary & operator and so are not converted to pointer to function types implicitly, but the & operator is converting them to pointer to function types explicitly.
After the above initialization of fops, rc = fops.open(inode, file); indirectly calls dev_open(inode, file) and assigns the return value to rc. You may sometimes see this written in an older style: rc = (*fops.open)(inode, file);. They both do the same thing. The operand of the function call operator ( ) is in fact always a pointer to a function. In the case of rc = (*fops.open)(inode, file);, fops.open has a pointer to a function type. (*fops.open) dereferences fops.open to a function type but since (*fops.open) is a function designator it is implicitly converted back to a pointer to function type before the function call. Similarly, in the direct call rc = dev_open(inode, file);, dev_open is a function designator and so has a function type, but is implicitly converted to a pointer to function type before the function call.
My understanding is that Vala and Genie have reference counting rather than garbage collection.
Per Valadoc.org, this:
string path = Path.build_filename ("my", "full", "path/to.txt");
yields this:
a newly-allocated string that must be freed with g_free
Is this correct or is g_free not required due to reference counting?
If string is wrapped within an object will g_free on string be actioned on object destruction?
Valadoc contains the documentation from the original GLib, GObject, Gtk+, etc. libraries. It also contains additional Vala specific documentation on top of that.
You often see sentences which only makes sense in the context of the C programming language.
The Vala compiler does the memory management for you (most of the time). In this example a Vala string is actually translated by the Vala compiler to a gchar * in C where it has to be deallocated using g_free which the Vala compiler also does for you.
As a matter of fact strings are a bit different than objects as they are not reference counted, but copied instead.
Take this example in Vala:
int main () {
Object obj1 = new Object ();
Object obj2 = obj1;
string s1 = "";
string s2 = s1;
return 0;
}
The (manually cleaned up) code that valac produces in C looks like this:
int main (void) {
GObject* obj1 = g_object_new (G_TYPE_OBJECT, NULL);
GObject* obj2 = g_object_ref (obj1);
gchar* s1 = g_strdup ("");
gchar* s2 = g_strdup (s1);
g_free (s2);
g_free (s1);
g_object_unref (obj2);
g_object_unref (obj1);
return 0;
}
As you can see obj2 is a reference to the same object as in obj1 and the object is only destroyed when both references are unrefed.
The string s2 on the other side is a copy of the string stored in s1 (which is in turn a copy of the string literal "").
As you can also see the compiler does take care of such details for you. It makes the manual reference counting and manual string copying automatic.
I am using a 3rd party C library in my iOS application, which I am in the process of converting from Objective-C to Swift. I hit an obstacle when attempting to read one of the structs returned by the C library in Swift.
The struct looks similar to this:
typedef unsigned int LibUint;
typedef unsigned char LibUint8;
typedef struct RequestConfiguration_ {
LibUint8 names[30][128];
LibUint numberNames;
LibUint currentName;
} RequestConfiguration;
Which is imported into Swift as a Tuple containing 30 Tuples of 128 LibUint8 values. After a long time of trial and error using nested withUnsafePointer calls, I eventually began searching for solutions to iterating a Tuple in Swift.
What I ended up using is the following functions:
/**
* Perform iterator on every children of the type using reflection
*/
func iterateChildren<T>(reflectable: T, #noescape iterator: (String?, Any) -> Void) {
let mirror = Mirror(reflecting: reflectable)
for i in mirror.children {
iterator(i.label, i.value)
}
}
/**
* Returns a String containing the characters within the Tuple
*/
func libUint8TupleToString<T>(tuple: T) -> String {
var result = [CChar]()
let mirror = Mirror(reflecting: tuple)
for child in mirror.children {
let char = CChar(child.value as! LibUint8)
result.append(char)
// Null reached, skip the rest.
if char == 0 {
break;
}
}
// Always null terminate; faster than checking if last is null.
result.append(CChar(0))
return String.fromCString(result) ?? ""
}
/**
* Returns an array of Strings by decoding characters within the Tuple
*/
func libUint8StringsInTuple<T>(tuple: T, length: Int = 0) -> [String] {
var idx = 0
var strings = [String]()
iterateChildren(tuple) { (label, value) in
guard length > 0 && idx < length else { return }
let str = libUint8TupleToString(value)
strings.append(str)
idx++
}
return strings
}
Usage
func handleConfiguration(config: RequestConfiguration) {
// Declaration types are added for clarity
let names: [String] = libUint8StringsInTuple(config.names, config.numberNames)
let currentName: String = names[config.currentName]
}
My solution uses reflection to iterate the first Tuple, and reflection to iterate the second, because I was getting incorrect strings when using withUnsafePointer for the nested Tuples, which I assume is due to signage. Surely there must be a way to read the C strings in the array, using an UnsafePointer alike withUsafePointer(&struct.cstring) { String.fromCString(UnsafePointer($0)) }.
To be clear, I'm looking for the fastest way to read these C strings in Swift, even if that involves using Reflection.
Here is a possible solution:
func handleConfiguration(var config: RequestConfiguration) {
let numStrings = Int(config.numberNames)
let lenStrings = sizeofValue(config.names.0)
let names = (0 ..< numStrings).map { idx in
withUnsafePointer(&config.names) {
String.fromCString(UnsafePointer<CChar>($0) + idx * lenStrings) ?? ""
}
}
let currentName = names[Int(config.currentName)]
print(names, currentName)
}
It uses the fact that
LibUint8 names[30][128];
are 30*128 contiguous bytes in memory. withUnsafePointer(&config.names)
calls the closure with $0 as a pointer to the start of that
memory location, and
UnsafePointer<CChar>($0) + idx * lenStrings
is a pointer to the start of the idx-th subarray. The above code requires
that each subarray contains a NUL-terminated UTF-8 string.
The solution suggested by Martin R looks good to me and, as far as I can see from my limited testing, does work. However, as Martin pointed out, it requires that the strings be NUL-terminated UTF-8. Here are two more possible approaches. These follow the principle of handling the complexity of C data structures in C instead of dealing with it in Swift. Which of these approaches you choose depends on what specifically you are doing with RequestConfiguration in your app. If you are not comfortable programming in C, then a pure Swift approach, like the one suggested by Martin, might be a better choice.
For the purposes of this discussion, we will assume that the 3rd party C library has the following function for retrieving RequestConfiguration:
const RequestConfiguration * getConfig();
Approach 1: Make the RequestConfiguration object available to your Swift code, but extract names from it using the following C helper function:
const unsigned char * getNameFromConfig(const RequestConfiguration * rc, unsigned int nameIdx)
{
return rc->names[nameIdx];
}
Both this function's signature and the RequestConfiguration type must be available to the Swift code via the bridging header. You can then do something like this in Swift:
var cfg : UnsafePointer<RequestConfiguration> = getConfig()
if let s = String.fromCString(UnsafePointer<CChar>(getNameFromConfig(cfg, cfg.memory.currentName)))
{
print(s)
}
This approach is nice if you need the RequestConfiguration object available to Swift in order to check the number of names in multiple places, for example.
Approach 2: You just need to be able to get the name at a given position. In this case the RequestConfiguration type does not even need to be visible to Swift. You can write a helper C function like this:
const unsigned char * getNameFromConfig1(unsigned int idx)
{
const RequestConfiguration * p = getConfig();
return p->names[idx];
}
and use it in Swift as follows:
if let s = String.fromCString(UnsafePointer<CChar>(getNameFromConfig1(2)))
{
print(s)
}
This will print the name at position 2 (counting from 0). Of course, with this approach you might also want to have C helpers that return the count of names as well as the current name index.
Again, with these 2 approaches it is assumed the strings are NUL-terminated UTF-8. There are other approaches possible, these are just examples.
Also please note that the above assumes that you access RequestConfiguration as read-only. If you also want to modify it and make the changes visible to the 3rd party library C code, then it's a different ballgame.
var cpuInfo: processor_info_array_t = nil
var numCpuInfo: mach_msg_type_number_t = 0
var coresTotalUsage: Float = 0.0
var numCPUsU: natural_t = 0
let err = host_processor_info(mach_host_self(), PROCESSOR_CPU_LOAD_INFO, &numCPUsU, &cpuInfo, &numCpuInfo);
assert(err == KERN_SUCCESS, "Failed call to host_processor_info")
Hi, I am calling the above C API host_processor_info to get process load informations from swift, no problem there.
cpuInfo is a inout parameter (pointer) that, on return, will point to a structure containing the CPU information allocated by that API.
The caller is reponsible for deallocating the memory; I can do that easily from objective C but haven't had any luck in swift. I know I could wrap that call into an objective C extension but I'm trying to learn swift and would like, if possible, avoid the obj-c solution.
in obj-c I would deallocate with:
size_t cpuInfoSize = sizeof(integer_t) * numCpuInfo;
vm_deallocate(mach_task_self(), (vm_address_t) cpuInfo, cpuInfoSize)
cpuInfo in swift is an UnsafeMutablePointer not convertible into a vm_address_t.
Any help appreciated, thanks.
processor_info_array_t is a pointer type, and vm_address_t is an integer type
(ultimately an alias for UInt). (Judging from the comments in <i386/vm_types.h>
this might to be for historical reasons.)
The only way to convert a pointer to an integer (of the same size) in Swift is unsafeBitCast.
mach_init.h defines
extern mach_port_t mach_task_self_;
#define mach_task_self() mach_task_self_
Only the extern variable is visible in Swift, not the macro.
This gives:
let cpuInfoSize = vm_size_t(sizeof(integer_t)) * vm_size_t(numCpuInfo)
vm_deallocate(mach_task_self_, unsafeBitCast(cpuInfo, vm_address_t.self), cpuInfoSize)
In Swift 4, the equivalent code appears to be:
let cpuInfoSize = vm_size_t(MemoryLayout<integer_t>.stride * Int(numCpuInfo))
vm_deallocate(mach_task_self_, vm_address_t(bitPattern: cpuInfo), cpuInfoSize)
In particular, the initializer UInt(bitPattern:) is now apparently preferred to unsafeBitCast() to initialize an unsigned integer with the bit pattern of a pointer (I guess this usage is not longer considered "unsafe"). It correctly handles a nil pointer, returning 0 in this case.