Kernel+LibC: Allow sys$mmap() callers to specify address alignment

This is exposed via the non-standard serenity_mmap() call in userspace.

Kernel/Process.cpp

@@ -151,12 +151,12 @@ bool Process::in_group(gid_t gid) const
     return m_gid == gid || m_extra_gids.contains(gid);
 }
 
-Range Process::allocate_range(VirtualAddress vaddr, size_t size)
+Range Process::allocate_range(VirtualAddress vaddr, size_t size, size_t alignment)
 {
     vaddr.mask(PAGE_MASK);
     size = PAGE_ROUND_UP(size);
     if (vaddr.is_null())
-        return page_directory().range_allocator().allocate_anywhere(size);
+        return page_directory().range_allocator().allocate_anywhere(size, alignment);
     return page_directory().range_allocator().allocate_specific(vaddr, size);
 }
 
@@ -185,11 +185,9 @@ Region& Process::allocate_split_region(const Region& source_region, const Range&
     return region;
 }
 
-Region* Process::allocate_region(VirtualAddress vaddr, size_t size, const String& name, int prot, bool commit)
+Region* Process::allocate_region(const Range& range, const String& name, int prot, bool commit)
 {
-    auto range = allocate_range(vaddr, size);
+    ASSERT(range.is_valid());
-    if (!range.is_valid())
-        return nullptr;
     auto& region = add_region(Region::create_user_accessible(range, name, prot_to_region_access_flags(prot)));
     region.map(page_directory());
     if (commit)
@@ -197,6 +195,14 @@ Region* Process::allocate_region(VirtualAddress vaddr, size_t size, const String
     return &region;
 }
 
+Region* Process::allocate_region(VirtualAddress vaddr, size_t size, const String& name, int prot, bool commit)
+{
+    auto range = allocate_range(vaddr, size);
+    if (!range.is_valid())
+        return nullptr;
+    return allocate_region(range, name, prot, commit);
+}
+
 Region* Process::allocate_file_backed_region(VirtualAddress vaddr, size_t size, NonnullRefPtr<Inode> inode, const String& name, int prot)
 {
     auto range = allocate_range(vaddr, size);
@@ -207,9 +213,10 @@ Region* Process::allocate_file_backed_region(VirtualAddress vaddr, size_t size,
     return &region;
 }
 
-Region* Process::allocate_region_with_vmobject(VirtualAddress vaddr, size_t size, NonnullRefPtr<VMObject> vmobject, size_t offset_in_vmobject, const String& name, int prot, bool user_accessible)
+Region* Process::allocate_region_with_vmobject(const Range& range, NonnullRefPtr<VMObject> vmobject, size_t offset_in_vmobject, const String& name, int prot, bool user_accessible)
 {
-    size_t end_in_vmobject = offset_in_vmobject + size;
+    ASSERT(range.is_valid());
+    size_t end_in_vmobject = offset_in_vmobject + range.size();
     if (end_in_vmobject < offset_in_vmobject) {
         dbgprintf("allocate_region_with_vmobject: Overflow (offset + size)\n");
         return nullptr;
@@ -222,9 +229,6 @@ Region* Process::allocate_region_with_vmobject(VirtualAddress vaddr, size_t size
         dbgprintf("allocate_region_with_vmobject: Attempt to allocate a region with an end past the end of its VMObject.\n");
         return nullptr;
     }
-    auto range = allocate_range(vaddr, size);
-    if (!range.is_valid())
-        return nullptr;
     offset_in_vmobject &= PAGE_MASK;
     Region* region;
     if (user_accessible)
@@ -235,6 +239,15 @@ Region* Process::allocate_region_with_vmobject(VirtualAddress vaddr, size_t size
     return region;
 }
 
+
+Region* Process::allocate_region_with_vmobject(VirtualAddress vaddr, size_t size, NonnullRefPtr<VMObject> vmobject, size_t offset_in_vmobject, const String& name, int prot, bool user_accessible)
+{
+    auto range = allocate_range(vaddr, size);
+    if (!range.is_valid())
+        return nullptr;
+    return allocate_region_with_vmobject(range, move(vmobject), offset_in_vmobject, name, prot, user_accessible);
+}
+
 bool Process::deallocate_region(Region& region)
 {
     InterruptDisabler disabler;
@@ -364,11 +377,15 @@ void* Process::sys$mmap(const Syscall::SC_mmap_params* user_params)
 
     void* addr = (void*)params.addr;
     size_t size = params.size;
+    size_t alignment = params.alignment;
     int prot = params.prot;
     int flags = params.flags;
     int fd = params.fd;
     int offset = params.offset;
 
+    if (alignment & ~PAGE_MASK)
+        return (void*)-EINVAL;
+
     if (!is_user_range(VirtualAddress(addr), size))
         return (void*)-EFAULT;
 
@@ -407,15 +424,19 @@ void* Process::sys$mmap(const Syscall::SC_mmap_params* user_params)
 
     Region* region = nullptr;
 
+    auto range = allocate_range(VirtualAddress(addr), size, alignment);
+    if (!range.is_valid())
+        return (void*)-ENOMEM;
+
     if (map_purgeable) {
         auto vmobject = PurgeableVMObject::create_with_size(size);
-        region = allocate_region_with_vmobject(VirtualAddress(addr), size, vmobject, 0, !name.is_null() ? name : "mmap (purgeable)", prot);
+        region = allocate_region_with_vmobject(range, vmobject, 0, !name.is_null() ? name : "mmap (purgeable)", prot);
         if (!region && (!map_fixed && addr != 0))
             region = allocate_region_with_vmobject({}, size, vmobject, 0, !name.is_null() ? name : "mmap (purgeable)", prot);
     } else if (map_anonymous) {
-        region = allocate_region(VirtualAddress(addr), size, !name.is_null() ? name : "mmap", prot, false);
+        region = allocate_region(range, !name.is_null() ? name : "mmap", prot, false);
         if (!region && (!map_fixed && addr != 0))
-            region = allocate_region({}, size, !name.is_null() ? name : "mmap", prot, false);
+            region = allocate_region(allocate_range({}, size), !name.is_null() ? name : "mmap", prot, false);
     } else {
         if (offset < 0)
             return (void*)-EINVAL;

Kernel/Process.h

@@ -364,6 +364,8 @@ public:
     Region* allocate_region_with_vmobject(VirtualAddress, size_t, NonnullRefPtr<VMObject>, size_t offset_in_vmobject, const String& name, int prot, bool user_accessible = true);
     Region* allocate_file_backed_region(VirtualAddress, size_t, NonnullRefPtr<Inode>, const String& name, int prot);
     Region* allocate_region(VirtualAddress, size_t, const String& name, int prot = PROT_READ | PROT_WRITE, bool commit = true);
+    Region* allocate_region_with_vmobject(const Range&, NonnullRefPtr<VMObject>, size_t offset_in_vmobject, const String& name, int prot, bool user_accessible = true);
+    Region* allocate_region(const Range&, const String& name, int prot = PROT_READ | PROT_WRITE, bool commit = true);
     bool deallocate_region(Region& region);
 
     Region& allocate_split_region(const Region& source_region, const Range&, size_t offset_in_vmobject);
@@ -405,7 +407,7 @@ private:
     Process(Thread*& first_thread, const String& name, uid_t, gid_t, pid_t ppid, RingLevel, RefPtr<Custody> cwd = nullptr, RefPtr<Custody> executable = nullptr, TTY* = nullptr, Process* fork_parent = nullptr);
     static pid_t allocate_pid();
 
-    Range allocate_range(VirtualAddress, size_t);
+    Range allocate_range(VirtualAddress, size_t, size_t alignment = PAGE_SIZE);
 
     Region& add_region(NonnullOwnPtr<Region>);
 

Kernel/Syscall.h

@@ -240,6 +240,7 @@ struct StringListArgument {
 struct SC_mmap_params {
     uint32_t addr;
     uint32_t size;
+    uint32_t alignment;
     int32_t prot;
     int32_t flags;
     int32_t fd;

Kernel/VM/RangeAllocator.cpp

@@ -94,7 +94,7 @@ void RangeAllocator::carve_at_index(int index, const Range& range)
         m_available_ranges.insert(index + 1, move(remaining_parts[1]));
 }
 
-Range RangeAllocator::allocate_anywhere(size_t size)
+Range RangeAllocator::allocate_anywhere(size_t size, size_t alignment)
 {
 #ifdef VM_GUARD_PAGES
     // NOTE: We pad VM allocations with a guard page on each side.
@@ -104,26 +104,32 @@ Range RangeAllocator::allocate_anywhere(size_t size)
     size_t effective_size = size;
     size_t offset_from_effective_base = 0;
 #endif
+
     for (int i = 0; i < m_available_ranges.size(); ++i) {
         auto& available_range = m_available_ranges[i];
-        if (available_range.size() < effective_size)
+        // FIXME: This check is probably excluding some valid candidates when using a large alignment.
+        if (available_range.size() < (effective_size + alignment))
             continue;
-        Range allocated_range(available_range.base().offset(offset_from_effective_base), size);
+
-        if (available_range.size() == effective_size) {
+        uintptr_t initial_base = available_range.base().offset(offset_from_effective_base).get();
+        uintptr_t aligned_base = round_up_to_power_of_two(initial_base, alignment);
+
+        Range allocated_range(VirtualAddress(aligned_base), size);
+        if (available_range == allocated_range) {
 #ifdef VRA_DEBUG
-            dbgprintf("VRA: Allocated perfect-fit anywhere(%u): %x\n", size, allocated_range.base().get());
+            dbgprintf("VRA: Allocated perfect-fit anywhere(%zu, %zu): %x\n", size, alignment, allocated_range.base().get());
 #endif
             m_available_ranges.remove(i);
             return allocated_range;
         }
         carve_at_index(i, allocated_range);
 #ifdef VRA_DEBUG
-        dbgprintf("VRA: Allocated anywhere(%u): %x\n", size, allocated_range.base().get());
+        dbgprintf("VRA: Allocated anywhere(%zu, %zu): %x\n", size, alignment, allocated_range.base().get());
         dump();
 #endif
         return allocated_range;
     }
-    kprintf("VRA: Failed to allocate anywhere: %u\n", size);
+    kprintf("VRA: Failed to allocate anywhere: %zu, %zu\n", size, alignment);
     return {};
 }
 

Kernel/VM/RangeAllocator.h

@@ -84,7 +84,7 @@ public:
     void initialize_with_range(VirtualAddress, size_t);
     void initialize_from_parent(const RangeAllocator&);
 
-    Range allocate_anywhere(size_t);
+    Range allocate_anywhere(size_t, size_t alignment = PAGE_SIZE);
     Range allocate_specific(VirtualAddress, size_t);
     void deallocate(Range);
 

Libraries/LibC/mman.cpp

@@ -32,14 +32,9 @@
 
 extern "C" {
 
-void* mmap(void* addr, size_t size, int prot, int flags, int fd, off_t offset)
+void* serenity_mmap(void* addr, size_t size, int prot, int flags, int fd, off_t offset, size_t alignment, const char* name)
 {
-    return mmap_with_name(addr, size, prot, flags, fd, offset, nullptr);
+    Syscall::SC_mmap_params params { (u32)addr, size, alignment, prot, flags, fd, offset, { name, name ? strlen(name) : 0 } };
-}
-
-void* mmap_with_name(void* addr, size_t size, int prot, int flags, int fd, off_t offset, const char* name)
-{
-    Syscall::SC_mmap_params params { (u32)addr, size, prot, flags, fd, offset, { name, name ? strlen(name) : 0 } };
     int rc = syscall(SC_mmap, &params);
     if (rc < 0 && -rc < EMAXERRNO) {
         errno = -rc;
@@ -48,6 +43,16 @@ void* mmap_with_name(void* addr, size_t size, int prot, int flags, int fd, off_t
     return (void*)rc;
 }
 
+void* mmap(void* addr, size_t size, int prot, int flags, int fd, off_t offset)
+{
+    return serenity_mmap(addr, size, prot, flags, fd, offset, PAGE_SIZE, nullptr);
+}
+
+void* mmap_with_name(void* addr, size_t size, int prot, int flags, int fd, off_t offset, const char* name)
+{
+    return serenity_mmap(addr, size, prot, flags, fd, offset, PAGE_SIZE, name);
+}
+
 int munmap(void* addr, size_t size)
 {
     int rc = syscall(SC_munmap, addr, size);

Libraries/LibC/mman.h

@@ -53,6 +53,7 @@ __BEGIN_DECLS
 
 void* mmap(void* addr, size_t, int prot, int flags, int fd, off_t);
 void* mmap_with_name(void* addr, size_t, int prot, int flags, int fd, off_t, const char* name);
+void* serenity_mmap(void* addr, size_t, int prot, int flags, int fd, off_t, size_t alignment, const char* name);
 int munmap(void*, size_t);
 int mprotect(void*, size_t, int prot);
 int set_mmap_name(void*, size_t, const char*);