kernel+libc: Remove (de)allocate_memory and replace it with POSIX mmap

kernel/CMakeLists.txt

@@ -24,7 +24,7 @@ set(SOURCES
     src/sys/exit.cpp
     src/sys/console_write.cpp
     src/sys/clock_gettime.cpp
-    src/sys/allocate_memory.cpp
+    src/sys/mmap.cpp
     src/sys/usleep.cpp
     src/sys/open.cpp
     src/sys/file.cpp

kernel/src/memory/UserVM.cpp

@@ -78,30 +78,33 @@ Result<u64> UserVM::alloc_several_pages(usize count)
     return VM_BASE + index * ARCH_PAGE_SIZE;
 }
 
-Result<void> UserVM::free_one_page(u64 address)
+Result<bool> UserVM::free_one_page(u64 address)
 {
     if (address < VM_BASE) return err(EINVAL);
     const u64 index = (address - VM_BASE) / ARCH_PAGE_SIZE;
     if (index > (MAX_VM_SIZE * 8)) return err(EINVAL);
 
-    if (!m_bitmap.get(index)) return err(EFAULT);
+    // NOTE: POSIX says munmap() should silently do nothing if the address is not mapped, instead of throwing an error
+    // like EFAULT.
+    if (!m_bitmap.get(index)) return false;
 
     m_bitmap.set(index, false);
 
-    return {};
+    return true;
 }
 
-Result<void> UserVM::free_several_pages(u64 address, usize count)
+Result<bool> UserVM::free_several_pages(u64 address, usize count)
 {
     if (address < VM_BASE) return err(EINVAL);
     const u64 index = (address - VM_BASE) / ARCH_PAGE_SIZE;
     if ((index + count) > (MAX_VM_SIZE * 8)) return err(EINVAL);
 
-    if (!TRY(m_bitmap.try_match_region(index, count, true))) return err(EFAULT);
+    // NOTE: Same as above.
+    if (!TRY(m_bitmap.try_match_region(index, count, true))) return false;
 
     m_bitmap.clear_region(index, count, false);
 
-    return {};
+    return true;
 }
 
 UserVM::~UserVM()

kernel/src/memory/UserVM.h

@@ -12,8 +12,8 @@ class UserVM
     Result<u64> alloc_one_page();
     Result<u64> alloc_several_pages(usize count);
 
-    Result<void> free_one_page(u64 address);
-    Result<void> free_several_pages(u64 address, usize count);
+    Result<bool> free_one_page(u64 address);
+    Result<bool> free_several_pages(u64 address, usize count);
 
     static Result<OwnedPtr<UserVM>> try_create();
 

kernel/src/sys/allocate_memory.cpp

@@ -1,53 +0,0 @@
-#include "Log.h"
-#include "arch/MMU.h"
-#include "memory/MemoryManager.h"
-#include "sys/Syscall.h"
-#include "thread/Scheduler.h"
-#include <bits/mmap-flags.h>
-#include <luna/Alignment.h>
-
-constexpr uintptr_t USERSPACE_HEAP_BASE = 0x3000000;
-
-Result<u64> sys_allocate_memory(Registers*, SyscallArgs args)
-{
-    usize size = (usize)args[0];
-    int flags = (int)args[1];
-
-    if (flags < 0) return err(EINVAL);
-    if (size == 0) return 0;
-
-    size = align_up<ARCH_PAGE_SIZE>(size);
-
-    Thread* current = Scheduler::current();
-
-    u64 address = TRY(current->vm_allocator->alloc_several_pages(size / ARCH_PAGE_SIZE));
-
-    int mmu_flags = MMU::User | MMU::NoExecute;
-    if (flags & PROT_WRITE) mmu_flags |= MMU::ReadWrite;
-    if (flags & PROT_EXEC) mmu_flags &= ~MMU::NoExecute;
-    if (flags == PROT_NONE) mmu_flags = MMU::NoExecute;
-
-    kdbgln("allocate_memory: allocating memory at %#lx, size=%zu", address, size);
-
-    return MemoryManager::alloc_at(address, size / ARCH_PAGE_SIZE, mmu_flags);
-}
-
-Result<u64> sys_deallocate_memory(Registers*, SyscallArgs args)
-{
-    u64 address = (u64)args[0];
-    usize size = (usize)args[1];
-
-    if (size == 0) return 0;
-
-    size = align_up<ARCH_PAGE_SIZE>(size);
-
-    Thread* current = Scheduler::current();
-
-    TRY(current->vm_allocator->free_several_pages(address, size / ARCH_PAGE_SIZE));
-
-    kdbgln("deallocate_memory: deallocating memory at %#lx, size=%zu", address, size);
-
-    TRY(MemoryManager::unmap_owned(address, size / ARCH_PAGE_SIZE));
-
-    return { 0 };
-}

kernel/src/sys/mmap.cpp

@@ -0,0 +1,77 @@
+#include "Log.h"
+#include "arch/MMU.h"
+#include "memory/MemoryManager.h"
+#include "sys/Syscall.h"
+#include "thread/Scheduler.h"
+#include <bits/mmap-flags.h>
+#include <luna/Alignment.h>
+
+constexpr uintptr_t USERSPACE_HEAP_BASE = 0x3000000;
+
+Result<u64> sys_mmap(Registers*, SyscallArgs args)
+{
+    void* addr = (void*)args[0];
+    usize len = (usize)args[1];
+    int prot = (int)args[2];
+    int flags = (int)args[3];
+
+    if (len == 0) return err(EINVAL);
+
+    if (flags < 0) return err(EINVAL);
+
+    // We support only anonymous mappings for now.
+    if ((flags & MAP_ANONYMOUS) != MAP_ANONYMOUS)
+    {
+        kwarnln("mmap: FIXME: attempt to mmap file instead of anonymous memory");
+        return err(ENOTSUP);
+    }
+
+    if (flags & MAP_SHARED)
+    {
+        kwarnln("mmap: FIXME: attempt to mmap shared memory");
+        return err(ENOTSUP);
+    }
+
+    len = align_up<ARCH_PAGE_SIZE>(len);
+
+    Thread* current = Scheduler::current();
+
+    u64 address;
+    if (!addr) address = TRY(current->vm_allocator->alloc_several_pages(len / ARCH_PAGE_SIZE));
+    else
+    {
+        kwarnln("mmap: FIXME: tried to mmap at a given address, instead of letting us choose");
+        return err(ENOTSUP);
+    }
+
+    int mmu_flags = MMU::User | MMU::NoExecute;
+    if (prot & PROT_WRITE) mmu_flags |= MMU::ReadWrite;
+    if (prot & PROT_EXEC) mmu_flags &= ~MMU::NoExecute;
+    if (prot == PROT_NONE) mmu_flags = MMU::NoExecute;
+
+    kdbgln("mmap: mapping memory at %#lx, size=%zu", address, len);
+
+    return MemoryManager::alloc_at(address, len / ARCH_PAGE_SIZE, mmu_flags);
+}
+
+Result<u64> sys_munmap(Registers*, SyscallArgs args)
+{
+    u64 address = (u64)args[0];
+    usize size = (usize)args[1];
+
+    if (size == 0) return err(EINVAL);
+    if (!is_aligned<ARCH_PAGE_SIZE>(size)) return err(EINVAL);
+
+    Thread* current = Scheduler::current();
+
+    bool ok = TRY(current->vm_allocator->free_several_pages(address, size / ARCH_PAGE_SIZE));
+
+    // POSIX says munmap should silently do nothing if the memory was not already mapped.
+    if (!ok) return 0;
+
+    kdbgln("munmap: unmapping memory at %#lx, size=%zu", address, size);
+
+    TRY(MemoryManager::unmap_owned(address, size / ARCH_PAGE_SIZE));
+
+    return { 0 };
+}

libc/include/bits/mmap-flags.h

@@ -8,4 +8,8 @@
 #define PROT_WRITE 2
 #define PROT_EXEC 4
 
+#define MAP_PRIVATE 0
+#define MAP_SHARED 1
+#define MAP_ANONYMOUS 2
+
 #endif

libc/include/sys/mman.h

@@ -15,8 +15,11 @@ extern "C"
 {
 #endif
 
-    void* allocate_memory(size_t size, int flags);
-    int deallocate_memory(void* address, size_t size);
+    /* Map a file into memory. */
+    void* mmap(void* addr, size_t len, int prot, int flags, int fd, off_t offset);
+
+    /* Remove a memory mapping. */
+    int munmap(void* addr, size_t len);
 
 #ifdef __cplusplus
 }

libc/src/sys/mman.cpp

@@ -7,29 +7,29 @@
 
 extern "C"
 {
-    void* allocate_memory(size_t size, int flags)
+    void* mmap(void* addr, size_t len, int prot, int flags, int, off_t)
     {
-        long rc = syscall(SYS_allocate_memory, size, flags);
+        long rc = syscall(SYS_mmap, addr, len, prot, flags);
         __errno_return(rc, void*);
     }
 
-    int deallocate_memory(void* address, size_t size)
+    int munmap(void* addr, size_t len)
     {
-        long rc = syscall(SYS_deallocate_memory, address, size);
+        long rc = syscall(SYS_munmap, addr, len);
         __errno_return(rc, int);
     }
 }
 
 Result<void*> allocate_pages_impl(usize count)
 {
-    long rc = syscall(SYS_allocate_memory, count * PAGE_SIZE, PROT_READ | PROT_WRITE);
+    long rc = syscall(SYS_mmap, nullptr, count * PAGE_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS);
     if (rc < 0) { return err((int)-rc); }
     return (void*)rc;
 }
 
 Result<void> release_pages_impl(void* address, usize count)
 {
-    long rc = syscall(SYS_deallocate_memory, address, count * PAGE_SIZE);
+    long rc = syscall(SYS_munmap, address, count * PAGE_SIZE);
     if (rc < 0) { return err((int)-rc); }
     return {};
 }

libluna/include/luna/Syscall.h

@@ -1,8 +1,8 @@
 #pragma once
 
 #define enumerate_syscalls(_e)                                                                                         \
-    _e(exit) _e(console_write) _e(clock_gettime) _e(allocate_memory) _e(deallocate_memory) _e(usleep) _e(open)         \
-        _e(close) _e(read) _e(getpid) _e(write) _e(lseek) _e(mkdir)
+    _e(exit) _e(console_write) _e(clock_gettime) _e(mmap) _e(munmap) _e(usleep) _e(open) _e(close) _e(read) _e(getpid) \
+        _e(write) _e(lseek) _e(mkdir)
 
 enum Syscalls
 {