From 32aa37d5dcfedf90a14696a5a6eec76bc029cffa Mon Sep 17 00:00:00 2001
From: Andreas Kling <kling@serenityos.org>
Date: Wed, 17 Nov 2021 19:33:00 +0100
Subject: [PATCH] Kernel+LibC: Add msync() system call

This allows userspace to trigger a full (FIXME) flush of a shared file
mapping to disk. We iterate over all the mapped pages in the VMObject
and write them out to the underlying inode, one by one. This is rather
naive, and there's lots of room for improvement.

Note that shared file mappings are currently not possible since mmap()
returns ENOTSUP for PROT_WRITE+MAP_SHARED. That restriction will be
removed in a subsequent commit. :^)
---
 Kernel/API/POSIX/sys/mman.h           |  4 ++++
 Kernel/API/Syscall.h                  |  1 +
 Kernel/Memory/SharedInodeVMObject.cpp | 19 +++++++++++++++++++
 Kernel/Memory/SharedInodeVMObject.h   |  2 ++
 Kernel/Process.h                      |  1 +
 Kernel/Syscalls/mmap.cpp              | 17 +++++++++++++++++
 Userland/Libraries/LibC/sys/mman.cpp  |  6 ++++++
 Userland/Libraries/LibC/sys/mman.h    |  1 +
 8 files changed, 51 insertions(+)

diff --git a/Kernel/API/POSIX/sys/mman.h b/Kernel/API/POSIX/sys/mman.h
index 3c996e9699..65a6618385 100644
--- a/Kernel/API/POSIX/sys/mman.h
+++ b/Kernel/API/POSIX/sys/mman.h
@@ -34,6 +34,10 @@ extern "C" {
 #define MADV_SET_VOLATILE 0x100
 #define MADV_SET_NONVOLATILE 0x200
 
+#define MS_SYNC 1
+#define MS_ASYNC 2
+#define MS_INVALIDATE 4
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/Kernel/API/Syscall.h b/Kernel/API/Syscall.h
index c321b4b7f9..40e55642b1 100644
--- a/Kernel/API/Syscall.h
+++ b/Kernel/API/Syscall.h
@@ -124,6 +124,7 @@ enum class NeedsBigProcessLock {
     S(mount, NeedsBigProcessLock::Yes)                      \
     S(mprotect, NeedsBigProcessLock::Yes)                   \
     S(mremap, NeedsBigProcessLock::Yes)                     \
+    S(msync, NeedsBigProcessLock::Yes)                      \
     S(msyscall, NeedsBigProcessLock::Yes)                   \
     S(munmap, NeedsBigProcessLock::Yes)                     \
     S(open, NeedsBigProcessLock::Yes)                       \
diff --git a/Kernel/Memory/SharedInodeVMObject.cpp b/Kernel/Memory/SharedInodeVMObject.cpp
index 605af54e18..15ad1c39e2 100644
--- a/Kernel/Memory/SharedInodeVMObject.cpp
+++ b/Kernel/Memory/SharedInodeVMObject.cpp
@@ -5,6 +5,7 @@
  */
 
 #include <Kernel/FileSystem/Inode.h>
+#include <Kernel/Locking/Spinlock.h>
 #include <Kernel/Memory/SharedInodeVMObject.h>
 
 namespace Kernel::Memory {
@@ -34,4 +35,22 @@ SharedInodeVMObject::SharedInodeVMObject(SharedInodeVMObject const& other)
 {
 }
 
+ErrorOr<void> SharedInodeVMObject::sync()
+{
+    SpinlockLocker locker(m_lock);
+
+    for (size_t page_index = 0; page_index < page_count(); ++page_index) {
+        auto& physical_page = m_physical_pages[page_index];
+        if (!physical_page)
+            continue;
+
+        u8 page_buffer[PAGE_SIZE];
+        MM.copy_physical_page(*physical_page, page_buffer);
+
+        TRY(m_inode->write_bytes(page_index * PAGE_SIZE, PAGE_SIZE, UserOrKernelBuffer::for_kernel_buffer(page_buffer), nullptr));
+    }
+
+    return {};
+}
+
 }
diff --git a/Kernel/Memory/SharedInodeVMObject.h b/Kernel/Memory/SharedInodeVMObject.h
index c596addbf9..81d6a93ca0 100644
--- a/Kernel/Memory/SharedInodeVMObject.h
+++ b/Kernel/Memory/SharedInodeVMObject.h
@@ -18,6 +18,8 @@ public:
     static ErrorOr<NonnullRefPtr<SharedInodeVMObject>> try_create_with_inode(Inode&);
     virtual ErrorOr<NonnullRefPtr<VMObject>> try_clone() override;
 
+    ErrorOr<void> sync();
+
 private:
     virtual bool is_shared_inode() const override { return true; }
 
diff --git a/Kernel/Process.h b/Kernel/Process.h
index 3fe8fe2f55..bc9ff41679 100644
--- a/Kernel/Process.h
+++ b/Kernel/Process.h
@@ -313,6 +313,7 @@ public:
     ErrorOr<FlatPtr> sys$mprotect(Userspace<void*>, size_t, int prot);
     ErrorOr<FlatPtr> sys$madvise(Userspace<void*>, size_t, int advice);
     ErrorOr<FlatPtr> sys$msyscall(Userspace<void*>);
+    ErrorOr<FlatPtr> sys$msync(Userspace<void*>, size_t, int flags);
     ErrorOr<FlatPtr> sys$purge(int mode);
     ErrorOr<FlatPtr> sys$select(Userspace<const Syscall::SC_select_params*>);
     ErrorOr<FlatPtr> sys$poll(Userspace<const Syscall::SC_poll_params*>);
diff --git a/Kernel/Syscalls/mmap.cpp b/Kernel/Syscalls/mmap.cpp
index 68a001890d..1b3e99b755 100644
--- a/Kernel/Syscalls/mmap.cpp
+++ b/Kernel/Syscalls/mmap.cpp
@@ -589,4 +589,21 @@ ErrorOr<FlatPtr> Process::sys$msyscall(Userspace<void*> address)
     region->set_syscall_region(true);
     return 0;
 }
+
+ErrorOr<FlatPtr> Process::sys$msync(Userspace<void*> address, size_t size, [[maybe_unused]] int flags)
+{
+    // FIXME: We probably want to sync all mappings in the address+size range.
+    auto* region = address_space().find_region_from_range(Memory::VirtualRange { address.vaddr(), size });
+    if (!region)
+        return EINVAL;
+
+    auto& vmobject = region->vmobject();
+    if (!vmobject.is_shared_inode())
+        return 0;
+
+    auto& inode_vmobject = static_cast<Memory::SharedInodeVMObject&>(vmobject);
+    TRY(inode_vmobject.sync());
+    return 0;
+}
+
 }
diff --git a/Userland/Libraries/LibC/sys/mman.cpp b/Userland/Libraries/LibC/sys/mman.cpp
index 457ddd4393..fa2479f4f6 100644
--- a/Userland/Libraries/LibC/sys/mman.cpp
+++ b/Userland/Libraries/LibC/sys/mman.cpp
@@ -89,4 +89,10 @@ int mlock(const void*, size_t)
     dbgln("FIXME: Implement mlock()");
     return 0;
 }
+
+int msync(void* address, size_t size, int flags)
+{
+    int rc = syscall(SC_msync, address, size, flags);
+    __RETURN_WITH_ERRNO(rc, rc, -1);
+}
 }
diff --git a/Userland/Libraries/LibC/sys/mman.h b/Userland/Libraries/LibC/sys/mman.h
index 2f694ff541..c1eacbe127 100644
--- a/Userland/Libraries/LibC/sys/mman.h
+++ b/Userland/Libraries/LibC/sys/mman.h
@@ -20,5 +20,6 @@ int set_mmap_name(void*, size_t, const char*);
 int madvise(void*, size_t, int advice);
 void* allocate_tls(const char* initial_data, size_t);
 int mlock(const void*, size_t);
+int msync(void*, size_t, int flags);
 
 __END_DECLS