Start adding block device stuff and file system stuff. Added test for block driver as a process.

risc_v/src/block.rs

@@ -6,7 +6,8 @@
 use crate::{kmem::{kfree, kmalloc},
             page::{zalloc, PAGE_SIZE},
             virtio,
-            virtio::{Descriptor, MmioOffsets, Queue, StatusField, VIRTIO_RING_SIZE}};
+			virtio::{Descriptor, MmioOffsets, Queue, StatusField, VIRTIO_RING_SIZE}};
+use crate::process::{set_running, set_waiting};
 use core::mem::size_of;
 
 #[repr(C)]
@@ -79,6 +80,13 @@ pub struct Request {
 	data:   Data,
 	status: Status,
 	head:   u16,
+
+	// Do not change anything above this line.
+	// This is the PID of watcher. We store the PID
+	// because it is possible that the process DIES
+	// before we get here. If we used a pointer, we
+	// may dereference invalid memory.
+	watcher: u16,
 }
 
 // Internal block device structure
@@ -118,6 +126,14 @@ pub const VIRTIO_BLK_F_CONFIG_WCE: u32 = 11;
 pub const VIRTIO_BLK_F_DISCARD: u32 = 13;
 pub const VIRTIO_BLK_F_WRITE_ZEROES: u32 = 14;
 
+// We might get several types of errors, but they can be enumerated here.
+pub enum BlockErrors {
+	Success = 0,
+	BlockDeviceNotFound,
+	InvalidArgument,
+	ReadOnly
+}
+
 // Much like with processes, Rust requires some initialization
 // when we declare a static. In this case, we use the Option
 // value type to signal that the variable exists, but not the
@@ -247,13 +263,16 @@ pub fn fill_next_descriptor(bd: &mut BlockDevice, desc: Descriptor) -> u16 {
 /// a multiple of 512, but we don't really check that.
 /// We DO however, check that we aren't writing to an R/O device. This would
 /// cause a I/O error if we tried to write to a R/O device.
-pub fn block_op(dev: usize, buffer: *mut u8, size: u32, offset: u64, write: bool) {
+pub fn block_op(dev: usize, buffer: *mut u8, size: u32, offset: u64, write: bool, watcher: u16) -> Result<u32, BlockErrors> {
 	unsafe {
 		if let Some(bdev) = BLOCK_DEVICES[dev - 1].as_mut() {
 			// Check to see if we are trying to write to a read only device.
-			if true == bdev.read_only && true == write {
+			if bdev.read_only && write {
 				println!("Trying to write to read/only!");
-				return;
+				return Err(BlockErrors::ReadOnly);
+			}
+			if size % 512 != 0 {
+				return Err(BlockErrors::InvalidArgument);
 			}
 			let sector = offset / 512;
 			// TODO: Before we get here, we are NOT allowed to schedule a read or
@@ -268,7 +287,7 @@ pub fn block_op(dev: usize, buffer: *mut u8, size: u32, offset: u64, write: bool
 			let head_idx = fill_next_descriptor(bdev, desc);
 			(*blk_request).header.sector = sector;
 			// A write is an "out" direction, whereas a read is an "in" direction.
-			(*blk_request).header.blktype = if true == write {
+			(*blk_request).header.blktype = if write {
 				VIRTIO_BLK_T_OUT
 			}
 			else {
@@ -280,10 +299,11 @@ pub fn block_op(dev: usize, buffer: *mut u8, size: u32, offset: u64, write: bool
 			(*blk_request).data.data = buffer;
 			(*blk_request).header.reserved = 0;
 			(*blk_request).status.status = 111;
+			(*blk_request).watcher = watcher;
 			let desc = Descriptor { addr:  buffer as u64,
 			                        len:   size,
 			                        flags: virtio::VIRTIO_DESC_F_NEXT
-			                               | if false == write {
+			                               | if !write {
 				                               virtio::VIRTIO_DESC_F_WRITE
 			                               }
 			                               else {
@@ -301,16 +321,20 @@ pub fn block_op(dev: usize, buffer: *mut u8, size: u32, offset: u64, write: bool
 			// The only queue a block device has is 0, which is the request
 			// queue.
 			bdev.dev.add(MmioOffsets::QueueNotify.scale32()).write_volatile(0);
+			Ok(size)
+		}
+		else {
+			Err(BlockErrors::BlockDeviceNotFound)
 		}
 	}
 }
 
-pub fn read(dev: usize, buffer: *mut u8, size: u32, offset: u64) {
+pub fn read(dev: usize, buffer: *mut u8, size: u32, offset: u64) -> Result<u32, BlockErrors> {
-	block_op(dev, buffer, size, offset, false);
+	block_op(dev, buffer, size, offset, false, 0)
 }
 
-pub fn write(dev: usize, buffer: *mut u8, size: u32, offset: u64) {
+pub fn write(dev: usize, buffer: *mut u8, size: u32, offset: u64) -> Result<u32, BlockErrors> {
-	block_op(dev, buffer, size, offset, true);
+	block_op(dev, buffer, size, offset, true, 0)
 }
 
 /// Here we handle block specific interrupts. Here, we need to check
@@ -324,10 +348,17 @@ pub fn pending(bd: &mut BlockDevice) {
 		while bd.ack_used_idx != queue.used.idx {
 			let ref elem = queue.used.ring[bd.ack_used_idx as usize % VIRTIO_RING_SIZE];
 			bd.ack_used_idx = bd.ack_used_idx.wrapping_add(1);
+			// Requests stay resident on the heap until this function, so we can recapture the address here
 			let rq = queue.desc[elem.id as usize].addr as *const Request;
+
+			// A process might be waiting for this interrupt. Awaken the process attached here.
+			let pid_of_watcher = (*rq).watcher;
+			// A PID of 0 means that we don't have a watcher.
+			if pid_of_watcher > 0 {
+				set_running(pid_of_watcher);
+				// TODO: Set GpA0 to the value of the return status.
+			}
 			kfree(rq as *mut u8);
-			// TODO: Awaken the process that will need this I/O. This is
-			// the purpose of the waiting state.
 		}
 	}
 }
@@ -344,3 +375,14 @@ pub fn handle_interrupt(idx: usize) {
 		}
 	}
 }
+
+pub fn process_read(pid: u16, dev: usize, buffer: *mut u8, size: u32, offset: u64) -> Result<u32, BlockErrors> {
+	println!("Process read {}, {}, 0x{:x}, {}, {}", pid, dev, buffer as usize, size, offset);
+	set_waiting(pid);
+	block_op(dev, buffer, size, offset, false, pid)
+}
+
+pub fn process_write(pid: u16, dev: usize, buffer: *mut u8, size: u32, offset: u64) -> Result<u32, BlockErrors> {
+	set_waiting(pid);
+	block_op(dev, buffer, size, offset, true, pid)
+}

risc_v/src/fs.rs

@@ -27,8 +27,10 @@ pub struct Stat {
 
 /// A file descriptor
 pub struct Descriptor {
-    blockdev: usize,
+    pub blockdev: usize,
-    node: u32,
+    pub node: u32,
+    pub loc: u32,
+    pub size: u32,
 }
 
 pub enum FsError {

risc_v/src/kmem.rs

@@ -89,6 +89,18 @@ pub fn init() {
 	}
 }
 
+/// talloc and tfree will allocate a reference to a given type
+/// This helps when creating structures. 
+pub fn talloc<T>() -> Option<&'static mut T> {
+	unsafe {
+		(kzmalloc(size_of::<T>()) as *mut T).as_mut()
+	}
+}
+
+pub fn tfree<T>(p: *mut T) {
+	kfree(p as *mut u8);
+}
+
 /// Allocate sub-page level allocation based on bytes and zero the memory
 pub fn kzmalloc(sz: usize) -> *mut u8 {
 	let size = align_val(sz, 3);

risc_v/src/main.rs

@@ -122,6 +122,7 @@ pub fn id_map_range(root: &mut page::Table,
 extern "C" {
 	fn switch_to_user(frame: usize) -> !;
 }
+
 fn rust_switch_to_user(frame: usize) -> ! {
 	unsafe {
 		switch_to_user(frame);
@@ -149,18 +150,8 @@ extern "C" fn kinit() {
 	}
 	// Set up virtio. This requires a working heap and page-grained allocator.
 	virtio::probe();
-	// Let's test the block driver!
+	// Test the block driver!
-	println!("Testing block driver.");
+	process::add_kernel_process(test::test_block);
-	let buffer = kmem::kmalloc(512);
-	block::read(8, buffer, 512, 0);
-	for i in 0..48 {
-		print!(" {:02x}", unsafe { buffer.add(i).read() });
-		if 0 == ((i+1) % 24) {
-			println!();
-		}
-	}
-	kmem::kfree(buffer);
-	println!("Block driver done");
 	// We schedule the next context switch using a multiplier of 1
 	// Block testing code removed.
 	trap::schedule_next_context_switch(1);
@@ -192,7 +183,6 @@ pub mod syscall;
 pub mod trap;
 pub mod uart;
 pub mod virtio;
-#[cfg(test)]
 pub mod test;
 
 

risc_v/src/minixfs.rs

@@ -3,77 +3,232 @@
 // Stephen Marz
 // 16 March 2020
 
-use crate::fs::{Descriptor, FileSystem, Stat, FsError};
+use crate::{block,
-
+            fs::{Descriptor, FileSystem, FsError, Stat},
+            kmem::{kfree, kmalloc, talloc, tfree}};
 
 use alloc::string::String;
-
+use core::{mem::size_of, ptr::null_mut};
 
 pub const MAGIC: u16 = 0x4d5a;
+pub const BLOCK_SIZE: u32 = 1024;
+pub const NUM_IPTRS: u32 = BLOCK_SIZE / 4;
 
+/// The superblock describes the file system on the disk. It gives
+/// us all the information we need to read the file system and navigate
+/// the file system, including where to find the inodes and zones (blocks).
 #[repr(C)]
 pub struct SuperBlock {
-    ninodes: u32,
+	pub ninodes:         u32,
-    pad0: u16,
+	pub pad0:            u16,
-    imap_blocks: u16,
+	pub imap_blocks:     u16,
-    zmap_blocks: u16,
+	pub zmap_blocks:     u16,
-    first_data_zone: u16,
+	pub first_data_zone: u16,
-    log_zone_size: u16,
+	pub log_zone_size:   u16,
-    pad1: u16,
+	pub pad1:            u16,
-    max_size: u32,
+	pub max_size:        u32,
-    zones: u32,
+	pub zones:           u32,
-    magic: u16,
+	pub magic:           u16,
-    pad2: u16,
+	pub pad2:            u16,
-    block_size: u16,
+	pub block_size:      u16,
-    disk_version: u8,
+	pub disk_version:    u8,
 }
 
+/// An inode stores the "meta-data" to a file. The mode stores the permissions
+/// AND type of file. This is how we differentiate a directory from a file. A file
+/// size is in here too, which tells us how many blocks we need to read. Finally, the
+/// zones array points to where we can find the blocks, which is where the data
+/// is contained for the file.
 #[repr(C)]
+#[derive(Copy, Clone)]
 pub struct Inode {
-    mode: u16,
+	pub mode:   u16,
-    nlinks: u16,
+	pub nlinks: u16,
-    uid: u16,
+	pub uid:    u16,
-    gid: u16,
+	pub gid:    u16,
-    size: u32,
+	pub size:   u32,
-    atime: u32,
+	pub atime:  u32,
-    mtime: u32,
+	pub mtime:  u32,
-    ctime: u32,
+	pub ctime:  u32,
-    zones: [u32; 10],
+	pub zones:  [u32; 10],
 }
 
+/// Notice that an inode does not contain the name of a file. This is because
+/// more than one file name may refer to the same inode. These are called "hard links"
+/// Instead, a DirEntry essentially associates a file name with an inode as shown in
+/// the structure below.
 #[repr(C)]
 pub struct DirEntry {
-    inode: u32,
+	pub inode: u32,
-    name: [u8; 60],
+	pub name:  [u8; 60],
 }
 
-pub struct MinixFileSystem {
+// We need a BlockBuffer that can automatically be created and destroyed
-    sb: SuperBlock
+// in the lifetime of our read and write functions. In C, this would entail
+// goto statements that "unravel" all of the allocations that we made. Take
+// a look at the read() function to see why I thought this way would be better.
+pub struct BlockBuffer {
+	buffer: *mut u8,
+}
+
+impl BlockBuffer {
+	pub fn new(sz: u32) -> Self {
+		BlockBuffer { buffer: kmalloc(sz as usize), }
+	}
+
+	pub fn get_mut(&mut self) -> *mut u8 {
+		self.buffer
+	}
+
+	pub fn get(&self) -> *const u8 {
+		self.buffer
+	}
+}
+
+impl Default for BlockBuffer {
+	fn default() -> Self {
+		BlockBuffer { buffer: kmalloc(1024), }
+	}
+}
+
+impl Drop for BlockBuffer {
+	fn drop(&mut self) {
+		if !self.buffer.is_null() {
+			kfree(self.buffer);
+			self.buffer = null_mut();
+		}
+	}
+}
+
+/// The MinixFileSystem implements the FileSystem trait for the VFS.
+pub struct MinixFileSystem;
+
+impl MinixFileSystem {
+	/// Inodes are the meta-data of a file, including the mode (permissions and type) and
+	/// the file's size. They are stored above the data zones, but to figure out where we
+	/// need to go to get the inode, we first need the superblock, which is where we can
+	/// find all of the information about the filesystem itself.
+	pub fn get_inode(bdev: usize, inode_num: u32) -> Option<Inode> {
+		// When we read, everything needs to be a multiple of a sector (512 bytes)
+		// So, we need to have memory available that's at least 512 bytes, even if
+		// we only want 10 bytes or 32 bytes (size of an Inode).
+		let mut buffer = BlockBuffer::new(512);
+
+		// Here is a little memory trick. We have a reference and it will refer to the
+		// top portion of our buffer. Since we won't be using the super block and inode
+		// simultaneously, we can overlap the memory regions.
+		let super_block = unsafe { &*(buffer.get_mut() as *mut SuperBlock) };
+		let inode = unsafe { &*(buffer.get_mut() as *mut Inode) };
+
+		// Read from the block device. The size is 1 sector (512 bytes) and our offset is past
+		// the boot block (first 1024 bytes). This is where the superblock sits.
+		let result = block::read(bdev, buffer.get_mut(), 512, 1024);
+		for _ in 0..1000000 {
+				
+		}
+		if result.is_ok() && super_block.magic == MAGIC {
+			// If we get here, we successfully read what we think is the super block.
+			// The math here is 2 - one for the boot block, one for the super block. Then we
+			// have to skip the bitmaps blocks. We have a certain number of inode map blocks (imap)
+			// and zone map blocks (zmap).
+			// The inode comes to us as a NUMBER, not an index. So, we need to subtract 1.
+			let inode_offset = (2 + super_block.imap_blocks + super_block.zmap_blocks) as usize
+			                   * BLOCK_SIZE as usize + (inode_num as usize - 1) * size_of::<Inode>();
+
+			// Now, we read the inode itself.
+			let result = block::read(bdev, buffer.get_mut(), 512, inode_offset as u64);
+			for _ in 0..1000000 {
+				
+			}
+			if result.is_ok() {
+				println!("Inode sizex = {} {:o}", inode.size, inode.mode);
+				return Some(inode.clone());
+			}
+		}
+		// If we get here, some result wasn't OK. Either the super block
+		// or the inode itself.
+		None
+	}
 }
 
 impl FileSystem for MinixFileSystem {
-    fn init(_bdev: usize) -> bool {
+	fn init(_bdev: usize) -> bool {
-        false
+		false
-    }
+	}
-    fn open(_path: &String) -> Result<Descriptor, FsError> {
-        Err(FsError::FileNotFound)
-    }
-    fn read(_desc: &Descriptor, _buffer: *mut u8, _offset: u32, _size: u32) -> u32 {
-        0
-    }
-    fn write(_desc: &Descriptor, _buffer: *const u8, _offset: u32, _size: u32) -> u32 {
-        0
-    }
-    fn close(_desc: &mut Descriptor) {
 
-    }
+	fn open(_path: &String) -> Result<Descriptor, FsError> {
-    fn stat(_desc: &Descriptor) -> Stat {
+		Err(FsError::FileNotFound)
-        Stat {
+	}
-            mode: 0,
+
-            size: 0,
+	fn read(desc: &Descriptor, buffer: *mut u8, offset: u32, size: u32) -> u32 {
-            uid: 0,
+		let mut blocks_seen = 0u32;
-            gid: 0,
+		let offset_block = offset / BLOCK_SIZE;
-        }
+		let offset_byte = offset % BLOCK_SIZE;
-    }
+
+		let mut block_buffer = BlockBuffer::new(BLOCK_SIZE);
+		let stats = Self::stat(desc);
+		let inode_result = Self::get_inode(desc.blockdev, desc.node);
+		if inode_result.is_none() {
+			// The inode couldn't be read, for some reason.
+			return 0;
+		}
+		let inode = inode_result.unwrap();
+		// First, the _size parameter (now in bytes_left) is the size of the buffer, not
+		// necessarily the size of the file. If our buffer is bigger than the file, we're OK.
+		// If our buffer is smaller than the file, then we can only read up to the buffer size.
+		let mut bytes_left = if size > stats.size {
+			stats.size
+		}
+		else {
+			size
+		};
+		let mut bytes_left = 0;
+		let mut bytes_read = 0u32;
+		// In Rust, our for loop automatically "declares" i from 0 to < 7. The syntax
+		// 0..7 means 0 through to 7 but not including 7. If we want to include 7, we
+		// would use the syntax 0..=7.
+		for i in 0..7 {
+			// There are 7 direct zones in the Minix 3 file system. So, we can just read them
+			// one by one. Any zone that has the value 0 is skipped and we check the next
+			// zones. This might happen as we start writing and truncating.
+
+			// We really use this to keep track of when we need to actually start reading
+			// But an if statement probably takes more time than just incrementing it.
+			if offset_block <= blocks_seen {
+				// If we get here, then our offset is within our window that we want to see.
+				// We need to go to the direct pointer's index. That'll give us a block INDEX.
+				// That makes it easy since all we have to do is multiply the block size
+				// by whatever we get. If it's 0, we skip it and move on.
+				let zone_num = inode.zones[i];
+				if zone_num == 0 {
+					continue;
+				}
+				let zone_offset = zone_num * BLOCK_SIZE;
+				println!("Zone #{} -> #{} -> {}", i, zone_num, zone_offset);
+				if let Ok(_) = block::read(desc.blockdev, block_buffer.get_mut(), BLOCK_SIZE, zone_offset as u64) {
+					for _ in 0..100000 {}
+					println!("Offset = {:x}", unsafe {block_buffer.get_mut().add(32).read()});
+				}
+			}
+			blocks_seen += 1;
+		}
+
+		bytes_read
+	}
+
+	fn write(_desc: &Descriptor, _buffer: *const u8, _offset: u32, _size: u32) -> u32 {
+		0
+	}
+
+	fn close(_desc: &mut Descriptor) {}
+
+	fn stat(desc: &Descriptor) -> Stat {
+		let inode_result = Self::get_inode(desc.blockdev, desc.node);
+		// This could be a little dangerous, but the descriptor should be checked in open().
+		let inode = inode_result.unwrap();
+		Stat { mode: inode.mode,
+		       size: inode.size,
+		       uid:  inode.uid,
+		       gid:  inode.gid, }
+	}
 }
-

risc_v/src/test.rs

@@ -1,2 +1,27 @@
 // test.rs
 
+extern "C" {
+    fn make_syscall(sysno: usize, arg0: usize, arg1: usize, arg2: usize, arg3: usize);
+}
+
+pub fn test_block() {
+    // Let's test the block driver!
+    println!("Started test block process.");
+	let desc = crate::fs::Descriptor {
+		blockdev: 8,
+		node: 1,
+		loc: 0,
+		size: 500,
+	};
+    let buffer = crate::kmem::kmalloc(1024);
+    unsafe {
+        make_syscall(63, 8, buffer as usize, 1024, 1024);
+        for i in 0..32 {
+            print!("{:02x}  ", buffer.add(i).read());
+        }
+    }
+    println!();
+    crate::kmem::kfree(buffer);
+    println!("Test block finished");
+    loop {}
+}