LibC: Implement getitimer

For some reason this does not work on 32 bit version, so it is disabled
on that platform. I'll have to look into it later to find the bug :)

kernel/arch/i686/Thread.S

@@ -41,31 +41,15 @@ start_userspace_thread:
 	call get_thread_start_sp
 	movl %eax, %esp
 
-	# STACK LAYOUT
+	# ds, es = user data
-	#   entry
-	#   argc
-	#   argv
-	#   envp
-	#   userspace stack
-
-	call get_userspace_thread_stack_top
-
 	movw $(0x20 | 3), %bx
 	movw %bx, %ds
 	movw %bx, %es
-	movw %bx, %fs
+	# gs = thread local
+	movw $(0x30 | 3), %bx
 	movw %bx, %gs
+	# fs = 0
 	xorw %bx, %bx
+	movw %bx, %fs
 
-	popl %edi
-	popl %esi
-	popl %edx
-	popl %ecx
-	popl %ebx
-
-	pushl $(0x20 | 3)
-	pushl %eax
-	pushl $0x202
-	pushl $(0x18 | 3)
-	pushl %ebx
 	iret

kernel/arch/i686/interrupts.S

@@ -40,8 +40,7 @@ isr_stub:
 	movl 56(%esp), %edx // isr number
 
 	movl %esp, %ebp
-	subl $15, %esp
+	andl $-16, %esp
-	andl $0xFFFFFFF0, %esp
 
 	pushl %eax
 	pushl %ebx
@@ -61,14 +60,13 @@ irq_stub:
 	load_kernel_segments
 	cld
 
-	movl 40(%esp), %eax # interrupt number
+	movl 40(%esp), %edi # interrupt number
 
 	movl %esp, %ebp
-	subl $15, %esp
+	andl $-16, %esp
-	andl $0xFFFFFFF0, %esp
 
 	subl $12, %esp
-	pushl %eax
+	pushl %edi
 	call cpp_irq_handler
 
 	movl %ebp, %esp
@@ -83,16 +81,15 @@ asm_yield_handler:
 	pushal
 	cld
 
-	movl %esp,     %eax # interrupt registers ptr
+	leal 32(%esp), %edi # interrupt stack ptr
-	leal 32(%esp), %ebx # interrupt stack ptr
+	movl %esp,     %esi # interrupt registers ptr
 
 	movl %esp, %ebp
-	subl $15, %esp
+	andl $-16, %esp
-	andl $0xFFFFFFF0, %esp
 
 	subl $8, %esp
-	pushl %eax
+	pushl %esi
-	pushl %ebx
+	pushl %edi
 	call cpp_yield_handler
 
 	movl %ebp, %esp
@@ -107,8 +104,7 @@ asm_ipi_handler:
 	cld
 
 	movl %esp, %ebp
-	subl $15, %esp
+	andl $-16, %esp
-	andl $0xFFFFFFF0, %esp
 
 	call cpp_ipi_handler
 
@@ -117,7 +113,6 @@ asm_ipi_handler:
 	pop_userspace
 	iret
 
-
 .global asm_timer_handler
 asm_timer_handler:
 	push_userspace
@@ -125,8 +120,7 @@ asm_timer_handler:
 	cld
 
 	movl %esp, %ebp
-	subl $15, %esp
+	andl $-16, %esp
-	andl $0xFFFFFFF0, %esp
 
 	call cpp_timer_handler
 

kernel/arch/x86_64/Thread.S

@@ -32,24 +32,4 @@ start_userspace_thread:
 	call get_thread_start_sp
 	movq %rax, %rsp
 
-	# STACK LAYOUT
-	#   entry
-	#   argc
-	#   argv
-	#   envp
-	#   userspace stack
-
-	call get_userspace_thread_stack_top
-
-	popq %rdi
-	popq %rsi
-	popq %rdx
-	popq %rcx
-	popq %rbx
-
-	pushq $(0x20 | 3)
-	pushq %rax
-	pushq $0x202
-	pushq $(0x18 | 3)
-	pushq %rbx
 	iretq

kernel/include/kernel/ELF.h

@@ -8,8 +8,15 @@ namespace Kernel::ELF
 
 	struct LoadResult
 	{
-		bool has_interpreter;
+		struct TLS
+		{
+			vaddr_t addr;
+			size_t size;
+		};
+
+		bool open_execfd;
 		vaddr_t entry_point;
+		BAN::Optional<TLS> master_tls;
 		BAN::Vector<BAN::UniqPtr<MemoryRegion>> regions;
 	};
 

kernel/include/kernel/FS/Inode.h

@@ -169,6 +169,7 @@ namespace Kernel
 	private:
 		BAN::WeakPtr<SharedFileData> m_shared_region;
 		friend class FileBackedRegion;
+		friend class OpenFileDescriptorSet;
 		friend class SharedFileData;
 		friend class TTY;
 	};

kernel/include/kernel/GDT.h

@@ -128,6 +128,10 @@ namespace Kernel
 #endif
 		}
 
+#if ARCH(i686)
+		void set_tls(uintptr_t addr);
+#endif
+
 	private:
 		GDT() = default;
 
@@ -149,8 +153,8 @@ namespace Kernel
 		BAN::Array<SegmentDescriptor, 7> m_gdt; // null, kernel code, kernel data, user code, user data, tss low, tss high
 		static constexpr uint16_t m_tss_offset = 0x28;
 #elif ARCH(i686)
-		BAN::Array<SegmentDescriptor, 7> m_gdt; // null, kernel code, kernel data, user code, user data, processor data, tss
+		BAN::Array<SegmentDescriptor, 8> m_gdt; // null, kernel code, kernel data, user code, user data, processor data, tls, tss
-		static constexpr uint16_t m_tss_offset = 0x30;
+		static constexpr uint16_t m_tss_offset = 0x38;
 #endif
 		TaskStateSegment m_tss;
 		const GDTR m_gdtr {

kernel/include/kernel/Memory/Types.h

@@ -4,8 +4,10 @@
 
 #if ARCH(x86_64)
 	#define KERNEL_OFFSET 0xFFFFFFFF80000000
+	#define USERSPACE_END 0xFFFF800000000000
 #elif ARCH(i686)
 	#define KERNEL_OFFSET 0xC0000000
+	#define USERSPACE_END 0xC0000000
 #else
 	#error
 #endif

kernel/include/kernel/Memory/VirtualRange.h

@@ -26,6 +26,16 @@ namespace Kernel
 		size_t size() const { return m_size; }
 		PageTable::flags_t flags() const { return m_flags; }
 
+		paddr_t paddr_of(vaddr_t vaddr) const
+		{
+			ASSERT(vaddr % PAGE_SIZE == 0);
+			const size_t index = (vaddr - m_vaddr) / PAGE_SIZE;
+			ASSERT(index < m_paddrs.size());
+			const paddr_t paddr = m_paddrs[index];
+			ASSERT(paddr);
+			return paddr;
+		}
+
 		bool contains(vaddr_t address) const { return vaddr() <= address && address < vaddr() + size(); }
 
 		BAN::ErrorOr<void> allocate_page_for_demand_paging(vaddr_t address);

kernel/include/kernel/Networking/TCPSocket.h

@@ -91,7 +91,7 @@ namespace Kernel
 			bool						has_ghost_byte { false };
 
 			uint32_t					data_size	{ 0 }; // number of bytes in this buffer
-			uint8_t						scale		{ 1 }; // window scale
+			uint8_t						scale_shift	{ 0 }; // window scale
 			BAN::UniqPtr<VirtualRange>	buffer;
 		};
 
@@ -99,8 +99,8 @@ namespace Kernel
 		{
 			uint32_t					mss				{ 0 }; // maximum segment size
 			uint16_t					non_scaled_size	{ 0 }; // window size without scaling
-			uint8_t						scale			{ 0 }; // window scale
+			uint8_t						scale_shift		{ 0 }; // window scale
-			uint32_t 					scaled_size() const { return (uint32_t)non_scaled_size << scale; }
+			uint32_t 					scaled_size() const { return (uint32_t)non_scaled_size << scale_shift; }
 
 			uint32_t					start_seq		{ 0 }; // sequence number of first byte in buffer
 			uint32_t					current_seq		{ 0 }; // sequence number of next send
@@ -118,6 +118,7 @@ namespace Kernel
 		{
 			sockaddr_storage	address;
 			socklen_t			address_len;
+			bool				has_window_scale;
 		};
 
 		struct PendingConnection

kernel/include/kernel/OpenFileDescriptorSet.h

@@ -47,6 +47,9 @@ namespace Kernel
 
 		BAN::ErrorOr<size_t> read_dir_entries(int fd, struct dirent* list, size_t list_len);
 
+		BAN::ErrorOr<size_t> recvfrom(int fd, BAN::ByteSpan buffer, sockaddr* address, socklen_t* address_len);
+		BAN::ErrorOr<size_t> sendto(int fd, BAN::ConstByteSpan buffer, const sockaddr* address, socklen_t address_len);
+
 		BAN::ErrorOr<VirtualFileSystem::File> file_of(int) const;
 		BAN::ErrorOr<BAN::StringView> path_of(int) const;
 		BAN::ErrorOr<BAN::RefPtr<Inode>> inode_of(int);

kernel/include/kernel/Process.h

@@ -6,6 +6,7 @@
 #include <BAN/StringView.h>
 #include <BAN/Vector.h>
 #include <kernel/Credentials.h>
+#include <kernel/ELF.h>
 #include <kernel/FS/Inode.h>
 #include <kernel/Lock/Mutex.h>
 #include <kernel/Memory/Heap.h>
@@ -34,15 +35,6 @@ namespace Kernel
 	public:
 		using entry_t = Thread::entry_t;
 
-		struct userspace_info_t
-		{
-			uintptr_t entry { 0 };
-			int argc { 0 };
-			char** argv { nullptr };
-			char** envp { nullptr };
-			int file_fd { -1 };
-		};
-
 	public:
 		static Process* create_kernel();
 		static Process* create_kernel(entry_t, void*);
@@ -183,7 +175,9 @@ namespace Kernel
 		BAN::ErrorOr<long> sys_sigprocmask(int how, const sigset_t* set, sigset_t* oset);
 
 		BAN::ErrorOr<long> sys_yield();
-		BAN::ErrorOr<long> sys_pthread_create(const pthread_attr_t* __restrict attr, void (*entry)(void*), void* arg);
+		BAN::ErrorOr<long> sys_set_tls(void*);
+		BAN::ErrorOr<long> sys_get_tls();
+		BAN::ErrorOr<long> sys_pthread_create(const pthread_attr_t* attr, void (*entry)(void*), void* arg);
 		BAN::ErrorOr<long> sys_pthread_exit(void* value);
 		BAN::ErrorOr<long> sys_pthread_join(pthread_t thread, void** value);
 		BAN::ErrorOr<long> sys_pthread_self();
@@ -208,7 +202,6 @@ namespace Kernel
 		size_t proc_environ(off_t offset, BAN::ByteSpan) const;
 
 		bool is_userspace() const { return m_is_userspace; }
-		const userspace_info_t& userspace_info() const { return m_userspace_info; }
 
 		// Returns error if page could not be allocated
 		// Returns true if the page was allocated successfully
@@ -227,6 +220,13 @@ namespace Kernel
 		Process(const Credentials&, pid_t pid, pid_t parent, pid_t sid, pid_t pgrp);
 		static Process* create_process(const Credentials&, pid_t parent, pid_t sid = 0, pid_t pgrp = 0);
 
+		struct TLSResult
+		{
+			BAN::UniqPtr<MemoryRegion> region;
+			vaddr_t addr;
+		};
+		static BAN::ErrorOr<TLSResult> initialize_thread_local_storage(PageTable&, ELF::LoadResult::TLS master_tls);
+
 		struct FileParent
 		{
 			VirtualFileSystem::File parent;
@@ -315,7 +315,6 @@ namespace Kernel
 		BAN::Vector<BAN::String> m_environ;
 
 		bool m_is_userspace { false };
-		userspace_info_t m_userspace_info;
 
 		SpinLock m_child_exit_lock;
 		BAN::Vector<ChildExitStatus> m_child_exit_statuses;

kernel/include/kernel/Processor.h

@@ -110,6 +110,8 @@ namespace Kernel
 		static void send_smp_message(ProcessorID, const SMPMessage&, bool send_ipi = true);
 		static void broadcast_smp_message(const SMPMessage&);
 
+		static void load_tls();
+
 	private:
 		Processor() = default;
 		~Processor() { ASSERT_NOT_REACHED(); }

kernel/include/kernel/Thread.h

@@ -7,6 +7,8 @@
 #include <kernel/Memory/VirtualRange.h>
 #include <kernel/ThreadBlocker.h>
 
+#include <LibELF/AuxiliaryVector.h>
+
 #include <signal.h>
 #include <sys/types.h>
 
@@ -41,9 +43,10 @@ namespace Kernel
 		BAN::ErrorOr<Thread*> pthread_create(entry_t, void*);
 
 		BAN::ErrorOr<Thread*> clone(Process*, uintptr_t sp, uintptr_t ip);
-		void setup_exec();
 		void setup_process_cleanup();
 
+		BAN::ErrorOr<void> initialize_userspace(vaddr_t entry, BAN::Span<BAN::String> argv, BAN::Span<BAN::String> envp, BAN::Span<LibELF::AuxiliaryVector> auxv);
+
 		// Returns true, if thread is going to trigger signal
 		bool is_interrupted_by_signal() const;
 
@@ -85,6 +88,9 @@ namespace Kernel
 
 		bool is_userspace() const { return m_is_userspace; }
 
+		void set_tls(vaddr_t tls) { m_tls = tls; }
+		vaddr_t get_tls() const { return m_tls; }
+
 		size_t virtual_page_count() const { return (m_kernel_stack ? (m_kernel_stack->size() / PAGE_SIZE) : 0) + (m_userspace_stack ? (m_userspace_stack->size() / PAGE_SIZE) : 0); }
 		size_t physical_page_count() const { return virtual_page_count(); }
 
@@ -100,7 +106,7 @@ namespace Kernel
 	private:
 		Thread(pid_t tid, Process*);
 
-		void setup_exec_impl(uintptr_t entry, uintptr_t arg0, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3);
+		void setup_exec(vaddr_t ip, vaddr_t sp);
 
 		static void on_exit_trampoline(Thread*);
 		void on_exit();
@@ -118,6 +124,8 @@ namespace Kernel
 		bool                       m_is_userspace         { false };
 		bool                       m_delete_process       { false };
 
+		vaddr_t                    m_tls                  { 0 };
+
 		SchedulerQueue::Node*      m_scheduler_node       { nullptr };
 
 		InterruptStack             m_interrupt_stack      { };

kernel/kernel/ELF.cpp

@@ -8,6 +8,7 @@
 
 #include <ctype.h>
 #include <fcntl.h>
+#include <pthread.h>
 
 namespace Kernel::ELF
 {
@@ -108,13 +109,13 @@ namespace Kernel::ELF
 		auto file_header = TRY(read_and_validate_file_header(inode));
 		auto program_headers = TRY(read_program_headers(inode, file_header));
 
-		vaddr_t executable_end = 0;
+		size_t exec_max_offset { 0 };
 		BAN::String interpreter;
 
 		for (const auto& program_header : program_headers)
 		{
 			if (program_header.p_type == PT_LOAD)
-				executable_end = BAN::Math::max<vaddr_t>(executable_end, program_header.p_vaddr + program_header.p_memsz);
+				exec_max_offset = BAN::Math::max<vaddr_t>(exec_max_offset, program_header.p_vaddr + program_header.p_memsz);
 			else if (program_header.p_type == PT_INTERP)
 			{
 				BAN::Vector<uint8_t> interp_buffer;
@@ -140,9 +141,6 @@ namespace Kernel::ELF
 			}
 		}
 
-		if (file_header.e_type == ET_DYN)
-			executable_end = 0x400000;
-
 		if (!interpreter.empty())
 		{
 			auto interpreter_inode = TRY(VirtualFileSystem::get().file_from_absolute_path(credentials, interpreter, O_EXEC)).inode;
@@ -164,15 +162,16 @@ namespace Kernel::ELF
 		}
 
 		const vaddr_t load_base_vaddr =
-			[&file_header, executable_end]() -> vaddr_t
+			[&file_header, exec_max_offset]() -> vaddr_t
 			{
 				if (file_header.e_type == ET_EXEC)
 					return 0;
 				if (file_header.e_type == ET_DYN)
-					return (executable_end + PAGE_SIZE - 1) & PAGE_ADDR_MASK;
+					return (exec_max_offset + PAGE_SIZE - 1) & PAGE_ADDR_MASK;
 				ASSERT_NOT_REACHED();
 			}();
 
+		vaddr_t last_loaded_address = 0;
 		BAN::Vector<BAN::UniqPtr<MemoryRegion>> memory_regions;
 		for (const auto& program_header : program_headers)
 		{
@@ -241,10 +240,57 @@ namespace Kernel::ELF
 
 				TRY(memory_regions.emplace_back(BAN::move(region)));
 			}
+
+			last_loaded_address = BAN::Math::max(last_loaded_address, pheader_base + program_header.p_memsz);
 		}
 
 		LoadResult result;
-		result.has_interpreter = !interpreter.empty();
+
+		for (const auto& program_header : program_headers)
+		{
+			if (program_header.p_type != PT_TLS)
+				continue;
+
+			if (!BAN::Math::is_power_of_two(program_header.p_align))
+				return BAN::Error::from_errno(EINVAL);
+
+			size_t region_size = program_header.p_memsz;
+			if (auto rem = region_size % program_header.p_align)
+				region_size += program_header.p_align - rem;
+
+			size_t offset = 0;
+			if (auto rem = region_size % alignof(uthread))
+				offset = alignof(uthread) - rem;
+
+			auto region = TRY(MemoryBackedRegion::create(
+				page_table,
+				offset + region_size,
+				{ .start = last_loaded_address, .end = USERSPACE_END },
+				MemoryRegion::Type::PRIVATE,
+				PageTable::Flags::UserSupervisor | PageTable::Flags::Present
+			));
+
+			for (vaddr_t vaddr = region->vaddr(); vaddr < region->vaddr() + offset + region->size(); vaddr += PAGE_SIZE)
+				TRY(region->allocate_page_containing(vaddr, false));
+
+			if (program_header.p_filesz > 0)
+			{
+				BAN::Vector<uint8_t> file_data_buffer;
+				TRY(file_data_buffer.resize(program_header.p_filesz));
+				if (TRY(inode->read(program_header.p_offset, file_data_buffer.span())) != file_data_buffer.size())
+					return BAN::Error::from_errno(EFAULT);
+				TRY(region->copy_data_to_region(offset, file_data_buffer.data(), file_data_buffer.size()));
+			}
+
+			result.master_tls = LoadResult::TLS {
+				.addr = region->vaddr(),
+				.size = region->size(),
+			};
+
+			TRY(memory_regions.emplace_back(BAN::move(region)));
+		}
+
+		result.open_execfd = !interpreter.empty();
 		result.entry_point = load_base_vaddr + file_header.e_entry;
 		result.regions = BAN::move(memory_regions);
 		return BAN::move(result);

kernel/kernel/GDT.cpp

@@ -26,12 +26,20 @@ namespace Kernel
 		gdt->write_entry(0x20, 0x00000000, 0xFFFFF, 0xF2, data_flags);	// user data
 #if ARCH(i686)
 		gdt->write_entry(0x28, reinterpret_cast<uint32_t>(processor), sizeof(Processor), 0x92, 0x4); // processor data
+		gdt->write_entry(0x30, 0x00000000, 0x00000, 0x00, 0x0);			// tls
 #endif
 		gdt->write_tss();
 
 		return gdt;
 	}
 
+#if ARCH(i686)
+	void GDT::set_tls(uintptr_t addr)
+	{
+		write_entry(0x30, addr, 0xFFFF, 0xF2, 0xC);
+	}
+#endif
+
 	void GDT::write_entry(uint8_t offset, uint32_t base, uint32_t limit, uint8_t access, uint8_t flags)
 	{
 		ASSERT(offset % sizeof(SegmentDescriptor) == 0);

kernel/kernel/IDT.cpp

@@ -248,11 +248,7 @@ namespace Kernel
 		}
 
 		if (Thread::current().has_process() && Process::current().is_userspace())
-		{
+			process_name = Process::current().name();
-			const char* const* argv = Process::current().userspace_info().argv;
-			if (argv && *argv)
-				process_name = *argv;
-		}
 
 #if ARCH(x86_64)
 		dwarnln(

kernel/kernel/Networking/TCPSocket.cpp

@@ -32,7 +32,7 @@ namespace Kernel
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
 			true
 		));
-		socket->m_recv_window.scale = 12; // use PAGE_SIZE windows
+		socket->m_recv_window.scale_shift = PAGE_SIZE_SHIFT; // use PAGE_SIZE windows
 		socket->m_send_window.buffer = TRY(VirtualRange::create_to_vaddr_range(
 			PageTable::kernel(),
 			KERNEL_OFFSET,
@@ -80,7 +80,6 @@ namespace Kernel
 		auto connection = m_pending_connections.front();
 		m_pending_connections.pop();
 
-
 		auto listen_key = ListenKey(
 			reinterpret_cast<const sockaddr*>(&connection.target.address),
 			connection.target.address_len
@@ -104,6 +103,9 @@ namespace Kernel
 		return_inode->m_next_state = State::SynReceived;
 		return_inode->m_mutex.unlock();
 
+		if (!return_inode->m_connection_info->has_window_scale)
+			return_inode->m_recv_window.scale_shift = 0;
+
 		TRY(m_listen_children.emplace(listen_key, return_inode));
 
 		const uint64_t wake_time_ms = SystemTimer::get().ms_since_boot() + 5000;
@@ -155,7 +157,7 @@ namespace Kernel
 		if (!is_bound())
 			TRY(m_network_layer.bind_socket_to_unused(this, address, address_len));
 
-		m_connection_info.emplace(sockaddr_storage {}, address_len);
+		m_connection_info.emplace(sockaddr_storage {}, address_len, true);
 		memcpy(&m_connection_info->address, address, address_len);
 
 		m_next_flags = SYN;
@@ -224,51 +226,33 @@ namespace Kernel
 
 	BAN::ErrorOr<size_t> TCPSocket::sendto_impl(BAN::ConstByteSpan message, const sockaddr* address, socklen_t address_len)
 	{
+		(void)address;
+		(void)address_len;
+
 		if (address)
 			return BAN::Error::from_errno(EISCONN);
 		if (!m_has_connected)
 			return BAN::Error::from_errno(ENOTCONN);
 
-		if (message.size() > m_send_window.buffer->size())
+		while (m_send_window.data_size == m_send_window.buffer->size())
-		{
-			size_t nsent = 0;
-			while (nsent < message.size())
-			{
-				const size_t to_send = BAN::Math::min<size_t>(message.size() - nsent, m_send_window.buffer->size());
-				TRY(sendto_impl(message.slice(nsent, to_send), address, address_len));
-				nsent += to_send;
-			}
-			return nsent;
-		}
-
-		while (true)
 		{
 			if (m_state != State::Established)
 				return return_with_maybe_zero();
-			if (m_send_window.data_size + message.size() <= m_send_window.buffer->size())
-				break;
 			LockFreeGuard free(m_mutex);
 			TRY(Thread::current().block_or_eintr_or_timeout_ms(m_thread_blocker, 100, false));
 		}
 
+		const size_t to_send = BAN::Math::min<size_t>(message.size(), m_send_window.buffer->size() - m_send_window.data_size);
+
 		{
 			auto* buffer = reinterpret_cast<uint8_t*>(m_send_window.buffer->vaddr());
-			memcpy(buffer + m_send_window.data_size, message.data(), message.size());
+			memcpy(buffer + m_send_window.data_size, message.data(), to_send);
-			m_send_window.data_size += message.size();
+			m_send_window.data_size += to_send;
 		}
 
-		const uint32_t target_ack = m_send_window.start_seq + m_send_window.data_size;
 		m_thread_blocker.unblock();
 
-		while (m_send_window.current_ack < target_ack)
+		return to_send;
-		{
-			if (m_state != State::Established)
-				return return_with_maybe_zero();
-			LockFreeGuard free(m_mutex);
-			TRY(Thread::current().block_or_eintr_or_timeout_ms(m_thread_blocker, 100, false));
-		}
-
-		return message.size();
 	}
 
 	bool TCPSocket::can_read_impl() const
@@ -377,16 +361,18 @@ namespace Kernel
 		header.seq_number = m_send_window.current_seq + m_send_window.has_ghost_byte;
 		header.ack_number = m_recv_window.start_seq + m_recv_window.data_size + m_recv_window.has_ghost_byte;
 		header.data_offset = (sizeof(TCPHeader) + m_tcp_options_bytes) / sizeof(uint32_t);
-		header.window_size = BAN::Math::min<size_t>(0xFFFF, m_recv_window.buffer->size() >> m_recv_window.scale);
+		header.window_size = BAN::Math::min<size_t>(0xFFFF, m_recv_window.buffer->size() >> m_recv_window.scale_shift);
 		header.flags = m_next_flags;
 		if (header.flags & FIN)
 			m_send_window.has_ghost_byte = true;
 		m_next_flags = 0;
 
-		if (m_state == State::Closed)
+		if (m_state == State::Closed || m_state == State::SynReceived)
 		{
 			add_tcp_header_option<0, TCPOption::MaximumSeqmentSize>(header, m_interface->payload_mtu() - m_network_layer.header_size());
-			add_tcp_header_option<4, TCPOption::WindowScale>(header, m_recv_window.scale);
+
+			if (m_connection_info->has_window_scale)
+				add_tcp_header_option<4, TCPOption::WindowScale>(header, m_recv_window.scale_shift);
 			header.window_size = BAN::Math::min<size_t>(0xFFFF, m_recv_window.buffer->size());
 
 			m_send_window.mss = 1440;
@@ -463,9 +449,12 @@ namespace Kernel
 				if (options.maximum_seqment_size.has_value())
 					m_send_window.mss = *options.maximum_seqment_size;
 				if (options.window_scale.has_value())
-					m_send_window.scale = *options.window_scale;
+					m_send_window.scale_shift = *options.window_scale;
 				else
-					m_recv_window.scale = 1;
+				{
+					m_recv_window.scale_shift = 0;
+					m_connection_info->has_window_scale = false;
+				}
 
 				m_send_window.start_seq = m_send_window.current_seq;
 				m_send_window.current_ack = m_send_window.current_seq;
@@ -492,6 +481,7 @@ namespace Kernel
 						ConnectionInfo connection_info;
 						memcpy(&connection_info.address, sender, sender_len);
 						connection_info.address_len = sender_len;
+						connection_info.has_window_scale = parse_tcp_options(header).window_scale.has_value();
 						MUST(m_pending_connections.emplace(
 							connection_info,
 							header.seq_number + 1

kernel/kernel/OpenFileDescriptorSet.cpp

@@ -1,5 +1,6 @@
 #include <kernel/FS/Pipe.h>
 #include <kernel/FS/VirtualFileSystem.h>
+#include <kernel/Lock/LockGuard.h>
 #include <kernel/Networking/NetworkManager.h>
 #include <kernel/OpenFileDescriptorSet.h>
 
@@ -298,11 +299,13 @@ namespace Kernel
 	{
 		TRY(validate_fd(fd));
 		auto& open_file = m_open_files[fd];
+		if (open_file.inode()->mode().ifsock())
+			return recvfrom(fd, buffer, nullptr, nullptr);
 		if (!(open_file.status_flags() & O_RDONLY))
 			return BAN::Error::from_errno(EBADF);
 		if ((open_file.status_flags() & O_NONBLOCK) && !open_file.inode()->can_read())
 			return 0;
-		size_t nread = TRY(open_file.inode()->read(open_file.offset(), buffer));
+		const size_t nread = TRY(open_file.inode()->read(open_file.offset(), buffer));
 		open_file.offset() += nread;
 		return nread;
 	}
@@ -311,13 +314,15 @@ namespace Kernel
 	{
 		TRY(validate_fd(fd));
 		auto& open_file = m_open_files[fd];
+		if (open_file.inode()->mode().ifsock())
+			return sendto(fd, buffer, nullptr, 0);
 		if (!(open_file.status_flags() & O_WRONLY))
 			return BAN::Error::from_errno(EBADF);
 		if ((open_file.status_flags() & O_NONBLOCK) && !open_file.inode()->can_write())
-			return 0;
+			return BAN::Error::from_errno(EWOULDBLOCK);
 		if (open_file.status_flags() & O_APPEND)
 			open_file.offset() = open_file.inode()->size();
-		size_t nwrite = TRY(open_file.inode()->write(open_file.offset(), buffer));
+		const size_t nwrite = TRY(open_file.inode()->write(open_file.offset(), buffer));
 		open_file.offset() += nwrite;
 		return nwrite;
 	}
@@ -340,6 +345,43 @@ namespace Kernel
 		}
 	}
 
+	BAN::ErrorOr<size_t> OpenFileDescriptorSet::recvfrom(int fd, BAN::ByteSpan buffer, sockaddr* address, socklen_t* address_len)
+	{
+		TRY(validate_fd(fd));
+		auto& open_file = m_open_files[fd];
+		if (!open_file.inode()->mode().ifsock())
+			return BAN::Error::from_errno(ENOTSOCK);
+		LockGuard _(open_file.inode()->m_mutex);
+		if ((open_file.status_flags() & O_NONBLOCK) && !open_file.inode()->can_read())
+			return BAN::Error::from_errno(EWOULDBLOCK);
+		return open_file.inode()->recvfrom(buffer, address, address_len);
+	}
+
+	BAN::ErrorOr<size_t> OpenFileDescriptorSet::sendto(int fd, BAN::ConstByteSpan buffer, const sockaddr* address, socklen_t address_len)
+	{
+		TRY(validate_fd(fd));
+		auto& open_file = m_open_files[fd];
+		if (!open_file.inode()->mode().ifsock())
+			return BAN::Error::from_errno(ENOTSOCK);
+		if ((open_file.status_flags() & O_NONBLOCK) && !open_file.inode()->can_write())
+			return BAN::Error::from_errno(EWOULDBLOCK);
+
+		LockGuard _(open_file.inode()->m_mutex);
+
+		size_t total_sent = 0;
+		while (total_sent < buffer.size())
+		{
+			if ((open_file.status_flags() & O_NONBLOCK) && !open_file.inode()->can_write())
+				return total_sent;
+			const size_t nsend = TRY(open_file.inode()->sendto(buffer.slice(total_sent), address, address_len));
+			if (nsend == 0)
+				return 0;
+			total_sent += nsend;
+		}
+
+		return total_sent;
+	}
+
 	BAN::ErrorOr<VirtualFileSystem::File> OpenFileDescriptorSet::file_of(int fd) const
 	{
 		TRY(validate_fd(fd));

kernel/kernel/Process.cpp

@@ -17,9 +17,12 @@
 #include <kernel/Terminal/PseudoTerminal.h>
 #include <kernel/Timer/Timer.h>
 
+#include <LibELF/AuxiliaryVector.h>
+
 #include <LibInput/KeyboardLayout.h>
 
 #include <fcntl.h>
+#include <pthread.h>
 #include <stdio.h>
 #include <sys/banan-os.h>
 #include <sys/sysmacros.h>
@@ -116,76 +119,61 @@ namespace Kernel
 		process->m_working_directory = VirtualFileSystem::get().root_file();
 		process->m_page_table = BAN::UniqPtr<PageTable>::adopt(MUST(PageTable::create_userspace()));
 
-		TRY(process->m_cmdline.push_back({}));
+		TRY(process->m_cmdline.emplace_back());
 		TRY(process->m_cmdline.back().append(path));
+		for (auto argument : arguments)
+		{
+			TRY(process->m_cmdline.emplace_back());
+			TRY(process->m_cmdline.back().append(argument));
+		}
 
 		LockGuard _(process->m_process_lock);
 
-		auto executable_inode = TRY(process->find_file(AT_FDCWD, path.data(), O_EXEC)).inode;
+		auto executable_file = TRY(process->find_file(AT_FDCWD, path.data(), O_EXEC));
+		auto executable_inode = executable_file.inode;
+
 		auto executable = TRY(ELF::load_from_inode(executable_inode, process->m_credentials, process->page_table()));
 		process->m_mapped_regions = BAN::move(executable.regions);
 
-		char** argv_addr = nullptr;
-		{
-			size_t needed_bytes = sizeof(char*) + path.size() + 1;
-			for (auto argument : arguments)
-				needed_bytes += sizeof(char*) + argument.size() + 1;
-			needed_bytes += sizeof(char*);
-
-			if (auto rem = needed_bytes % PAGE_SIZE)
-				needed_bytes += PAGE_SIZE - rem;
-
-			auto argv_region = MUST(MemoryBackedRegion::create(
-				process->page_table(),
-				needed_bytes,
-				{ .start = 0x400000, .end = KERNEL_OFFSET },
-				MemoryRegion::Type::PRIVATE,
-				PageTable::Flags::UserSupervisor | PageTable::Flags::Present
-			));
-			argv_addr = reinterpret_cast<char**>(argv_region->vaddr());
-
-			uintptr_t offset = sizeof(char*) * (1 + arguments.size() + 1);
-			for (size_t i = 0; i <= arguments.size(); i++)
-			{
-				const uintptr_t addr = argv_region->vaddr() + offset;
-				TRY(argv_region->copy_data_to_region(i * sizeof(char*), reinterpret_cast<const uint8_t*>(&addr), sizeof(char*)));
-
-				const auto current = (i == 0) ? path : arguments[i - 1];
-				TRY(argv_region->copy_data_to_region(offset, reinterpret_cast<const uint8_t*>(current.data()), current.size()));
-
-				const uint8_t zero = 0;
-				TRY(argv_region->copy_data_to_region(offset + current.size(), &zero, 1));
-
-				offset += current.size() + 1;
-			}
-
-			const uintptr_t zero = 0;
-			TRY(argv_region->copy_data_to_region((1 + arguments.size()) * sizeof(char*), reinterpret_cast<const uint8_t*>(&zero), sizeof(char*)));
-
-			TRY(process->m_mapped_regions.push_back(BAN::move(argv_region)));
-		}
-
 		if (executable_inode->mode().mode & +Inode::Mode::ISUID)
 			process->m_credentials.set_euid(executable_inode->uid());
 		if (executable_inode->mode().mode & +Inode::Mode::ISGID)
 			process->m_credentials.set_egid(executable_inode->gid());
 
-		if (executable.has_interpreter)
+		BAN::Vector<LibELF::AuxiliaryVector> auxiliary_vector;
+		TRY(auxiliary_vector.reserve(1 + executable.open_execfd));
+
+		if (executable.open_execfd)
 		{
-			VirtualFileSystem::File file;
+			const int execfd = TRY(process->m_open_file_descriptors.open(BAN::move(executable_file), O_RDONLY));
-			TRY(file.canonical_path.append("<self>"));
+			TRY(auxiliary_vector.push_back({
-			file.inode = executable_inode;
+				.a_type = LibELF::AT_EXECFD,
-			process->m_userspace_info.file_fd = TRY(process->m_open_file_descriptors.open(BAN::move(file), O_RDONLY));
+				.a_un = { .a_val = static_cast<uint32_t>(execfd) },
+			}));
 		}
 
-		process->m_is_userspace = true;
+		TRY(auxiliary_vector.push_back({
-		process->m_userspace_info.entry = executable.entry_point;
+			.a_type = LibELF::AT_NULL,
-		process->m_userspace_info.argc = 1 + arguments.size();
+			.a_un = { .a_val = 0 },
-		process->m_userspace_info.argv = argv_addr;
+		}));
-		process->m_userspace_info.envp = nullptr;
+
+		BAN::Optional<vaddr_t> tls_addr;
+		if (executable.master_tls.has_value())
+		{
+			auto tls_result = TRY(process->initialize_thread_local_storage(process->page_table(), *executable.master_tls));
+			TRY(process->m_mapped_regions.emplace_back(BAN::move(tls_result.region)));
+			tls_addr = tls_result.addr;
+		}
 
 		auto* thread = MUST(Thread::create_userspace(process, process->page_table()));
-		thread->setup_exec();
+		MUST(thread->initialize_userspace(
+			executable.entry_point,
+			process->m_cmdline.span(),
+			process->m_environ.span(),
+			auxiliary_vector.span()
+		));
+		if (tls_addr.has_value())
+			thread->set_tls(*tls_addr);
 
 		process->add_thread(thread);
 		process->register_to_scheduler();
@@ -318,6 +306,63 @@ namespace Kernel
 		ASSERT_NOT_REACHED();
 	}
 
+	BAN::ErrorOr<Process::TLSResult> Process::initialize_thread_local_storage(PageTable& page_table, ELF::LoadResult::TLS master_tls)
+	{
+		const auto [master_addr, master_size] = master_tls;
+		ASSERT(master_size % alignof(uthread) == 0);
+
+		const size_t tls_size = master_size + PAGE_SIZE;
+
+		auto region = TRY(MemoryBackedRegion::create(
+			page_table,
+			tls_size,
+			{ .start = master_addr, .end = USERSPACE_END },
+			MemoryRegion::Type::PRIVATE,
+			PageTable::Flags::UserSupervisor | PageTable::Flags::ReadWrite | PageTable::Flags::Present
+		));
+
+		BAN::Vector<uint8_t> temp_buffer;
+		TRY(temp_buffer.resize(BAN::Math::min<size_t>(master_size, PAGE_SIZE)));
+
+		size_t bytes_copied = 0;
+		while (bytes_copied < master_size)
+		{
+			const size_t to_copy = BAN::Math::min(master_size - bytes_copied, temp_buffer.size());
+
+			const vaddr_t vaddr = master_addr + bytes_copied;
+			const paddr_t paddr = page_table.physical_address_of(vaddr & PAGE_ADDR_MASK);
+			PageTable::with_fast_page(paddr, [&] {
+				memcpy(temp_buffer.data(), PageTable::fast_page_as_ptr(vaddr % PAGE_SIZE), to_copy);
+			});
+
+			TRY(region->copy_data_to_region(bytes_copied, temp_buffer.data(), to_copy));
+			bytes_copied += to_copy;
+		}
+
+		const uthread uthread {
+			.self = reinterpret_cast<struct uthread*>(region->vaddr() + master_size),
+			.master_tls_addr = reinterpret_cast<void*>(master_addr),
+			.master_tls_size = master_size,
+		};
+		const uintptr_t dtv[2] { 1, region->vaddr() };
+
+		TRY(region->copy_data_to_region(
+			master_size,
+			reinterpret_cast<const uint8_t*>(&uthread),
+			sizeof(uthread)
+		));
+		TRY(region->copy_data_to_region(
+			master_size + sizeof(uthread),
+			reinterpret_cast<const uint8_t*>(&dtv),
+			sizeof(dtv)
+		));
+
+		TLSResult result;
+		result.addr = region->vaddr() + master_size;;
+		result.region = BAN::move(region);
+		return result;
+	}
+
 	size_t Process::proc_meminfo(off_t offset, BAN::ByteSpan buffer) const
 	{
 		ASSERT(offset >= 0);
@@ -534,7 +579,6 @@ namespace Kernel
 		forked->m_open_file_descriptors = BAN::move(*open_file_descriptors);
 		forked->m_mapped_regions = BAN::move(mapped_regions);
 		forked->m_is_userspace = m_is_userspace;
-		forked->m_userspace_info = m_userspace_info;
 		forked->m_has_called_exec = false;
 		memcpy(forked->m_signal_handlers, m_signal_handlers, sizeof(m_signal_handlers));
 
@@ -561,77 +605,69 @@ namespace Kernel
 
 			TRY(validate_string_access(path));
 
-			auto executable_file = TRY(find_file(AT_FDCWD, path, O_EXEC));
-			auto executable_inode = executable_file.inode;
-
 			BAN::Vector<BAN::String> str_argv;
 			for (int i = 0; argv && argv[i]; i++)
 			{
 				TRY(validate_pointer_access(argv + i, sizeof(char*), false));
 				TRY(validate_string_access(argv[i]));
-				TRY(str_argv.emplace_back(argv[i]));
+				TRY(str_argv.emplace_back());
+				TRY(str_argv.back().append(argv[i]));
 			}
 
 			BAN::Vector<BAN::String> str_envp;
 			for (int i = 0; envp && envp[i]; i++)
 			{
-				TRY(validate_pointer_access(envp + 1, sizeof(char*), false));
+				TRY(validate_pointer_access(envp + i, sizeof(char*), false));
 				TRY(validate_string_access(envp[i]));
-				TRY(str_envp.emplace_back(envp[i]));
+				TRY(str_envp.emplace_back());
+				TRY(str_envp.back().append(envp[i]));
 			}
 
+			auto executable_file = TRY(find_file(AT_FDCWD, path, O_EXEC));
+			auto executable_inode = executable_file.inode;
+
 			auto executable = TRY(ELF::load_from_inode(executable_inode, m_credentials, *new_page_table));
 			auto new_mapped_regions = BAN::move(executable.regions);
 
-			int file_fd = -1;
+			BAN::Vector<LibELF::AuxiliaryVector> auxiliary_vector;
-			if (executable.has_interpreter)
+			TRY(auxiliary_vector.reserve(1 + executable.open_execfd));
+
+			BAN::ScopeGuard execfd_guard([this, &auxiliary_vector] {
+				if (auxiliary_vector.empty())
+					return;
+				if (auxiliary_vector.front().a_type != LibELF::AT_EXECFD)
+					return;
+				MUST(m_open_file_descriptors.close(auxiliary_vector.front().a_un.a_val));
+			});
+
+			if (executable.open_execfd)
 			{
-				VirtualFileSystem::File file;
+				const int execfd = TRY(m_open_file_descriptors.open(BAN::move(executable_file), O_RDONLY));
-				file.canonical_path = BAN::move(executable_file.canonical_path);
+				TRY(auxiliary_vector.push_back({
-				file.inode = executable_inode;
+					.a_type = LibELF::AT_EXECFD,
-				file_fd = TRY(m_open_file_descriptors.open(BAN::move(file), O_RDONLY));
+					.a_un = { .a_val = static_cast<uint32_t>(execfd) },
+				}));
 			}
-			BAN::ScopeGuard file_closer([&] { if (file_fd != -1) MUST(m_open_file_descriptors.close(file_fd)); });
 
-			// allocate memory on the new process for arguments and environment
+			TRY(auxiliary_vector.push_back({
-			auto create_region =
+				.a_type = LibELF::AT_NULL,
-				[&](BAN::Span<BAN::String> container) -> BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>>
+				.a_un = { .a_val = 0 },
-				{
+			}));
-					size_t bytes = sizeof(char*);
-					for (auto& elem : container)
-						bytes += sizeof(char*) + elem.size() + 1;
-
-					if (auto rem = bytes % PAGE_SIZE)
-						bytes += PAGE_SIZE - rem;
-
-					auto region = TRY(MemoryBackedRegion::create(
-						*new_page_table,
-						bytes,
-						{ .start = executable.entry_point, .end = KERNEL_OFFSET },
-						MemoryRegion::Type::PRIVATE,
-						PageTable::Flags::UserSupervisor | PageTable::Flags::ReadWrite | PageTable::Flags::Present
-					));
-
-					size_t data_offset = sizeof(char*) * (container.size() + 1);
-					for (size_t i = 0; i < container.size(); i++)
-					{
-						uintptr_t ptr_addr = region->vaddr() + data_offset;
-						TRY(region->copy_data_to_region(sizeof(char*) * i, (const uint8_t*)&ptr_addr, sizeof(char*)));
-						TRY(region->copy_data_to_region(data_offset, (const uint8_t*)container[i].data(), container[i].size()));
-						data_offset += container[i].size() + 1;
-					}
-
-					uintptr_t null = 0;
-					TRY(region->copy_data_to_region(sizeof(char*) * container.size(), (const uint8_t*)&null, sizeof(char*)));
-
-					return BAN::UniqPtr<MemoryRegion>(BAN::move(region));
-				};
-
-			TRY(new_mapped_regions.reserve(new_mapped_regions.size() + 2));
-			MUST(new_mapped_regions.push_back(TRY(create_region(str_argv.span()))));
-			MUST(new_mapped_regions.push_back(TRY(create_region(str_envp.span()))));
 
 			auto* new_thread = TRY(Thread::create_userspace(this, *new_page_table));
+			TRY(new_thread->initialize_userspace(
+				executable.entry_point,
+				str_argv.span(),
+				str_envp.span(),
+				auxiliary_vector.span()
+			));
+
+			if (executable.master_tls.has_value())
+			{
+				auto tls_result = TRY(initialize_thread_local_storage(*new_page_table, *executable.master_tls));
+				TRY(new_mapped_regions.emplace_back(BAN::move(tls_result.region)));
+				new_thread->set_tls(tls_result.addr);
+			}
 
 			ASSERT(Processor::get_interrupt_state() == InterruptState::Enabled);
 			Processor::set_interrupt_state(InterruptState::Disabled);
@@ -655,8 +691,8 @@ namespace Kernel
 			m_threads.front()->m_process = nullptr;
 			m_threads.front()->give_keep_alive_page_table(BAN::move(m_page_table));
 
+			MUST(Processor::scheduler().add_thread(new_thread));
 			m_threads.front() = new_thread;
-			MUST(Processor::scheduler().add_thread(m_threads.front()));
 
 			for (size_t i = 0; i < sizeof(m_signal_handlers) / sizeof(*m_signal_handlers); i++)
 			{
@@ -673,21 +709,13 @@ namespace Kernel
 			m_mapped_regions = BAN::move(new_mapped_regions);
 			m_page_table = BAN::move(new_page_table);
 
-			file_closer.disable();
+			execfd_guard.disable();
-
-			m_userspace_info.argc = str_argv.size();
-			m_userspace_info.argv = reinterpret_cast<char**>(m_mapped_regions[m_mapped_regions.size() - 2]->vaddr());
-			m_userspace_info.envp = reinterpret_cast<char**>(m_mapped_regions[m_mapped_regions.size() - 1]->vaddr());
-			m_userspace_info.entry = executable.entry_point;
-			m_userspace_info.file_fd = file_fd;
 
 			m_cmdline = BAN::move(str_argv);
 			m_environ = BAN::move(str_envp);
 		}
 
 		m_has_called_exec = true;
-
-		m_threads.front()->setup_exec();
 		Processor::yield();
 		ASSERT_NOT_REACHED();
 	}
@@ -1022,6 +1050,11 @@ namespace Kernel
 	BAN::ErrorOr<long> Process::sys_read(int fd, void* buffer, size_t count)
 	{
 		LockGuard _(m_process_lock);
+		if (count == 0)
+		{
+			TRY(m_open_file_descriptors.inode_of(fd));
+			return 0;
+		}
 		TRY(validate_pointer_access(buffer, count, true));
 		return TRY(m_open_file_descriptors.read(fd, BAN::ByteSpan((uint8_t*)buffer, count)));
 	}
@@ -1136,6 +1169,11 @@ namespace Kernel
 	BAN::ErrorOr<long> Process::sys_pread(int fd, void* buffer, size_t count, off_t offset)
 	{
 		LockGuard _(m_process_lock);
+		if (count == 0)
+		{
+			TRY(m_open_file_descriptors.inode_of(fd));
+			return 0;
+		}
 		TRY(validate_pointer_access(buffer, count, true));
 		auto inode = TRY(m_open_file_descriptors.inode_of(fd));
 		return TRY(inode->read(offset, { (uint8_t*)buffer, count }));
@@ -1321,16 +1359,8 @@ namespace Kernel
 		TRY(validate_pointer_access(arguments->message, arguments->length, false));
 		TRY(validate_pointer_access(arguments->dest_addr, arguments->dest_len, false));
 
-		auto inode = TRY(m_open_file_descriptors.inode_of(arguments->socket));
+		auto message = BAN::ConstByteSpan(static_cast<const uint8_t*>(arguments->message), arguments->length);
-		if (!inode->mode().ifsock())
+		return TRY(m_open_file_descriptors.sendto(arguments->socket, message, arguments->dest_addr, arguments->dest_len));
-			return BAN::Error::from_errno(ENOTSOCK);
-
-		const auto status_flags = TRY(m_open_file_descriptors.status_flags_of(arguments->socket));
-		if ((status_flags & O_NONBLOCK) && !inode->can_write())
-			return BAN::Error::from_errno(EAGAIN);
-
-		BAN::ConstByteSpan message { reinterpret_cast<const uint8_t*>(arguments->message), arguments->length };
-		return TRY(inode->sendto(message, arguments->dest_addr, arguments->dest_len));
 	}
 
 	BAN::ErrorOr<long> Process::sys_recvfrom(sys_recvfrom_t* arguments)
@@ -1349,16 +1379,8 @@ namespace Kernel
 			TRY(validate_pointer_access(arguments->address, *arguments->address_len, true));
 		}
 
-		auto inode = TRY(m_open_file_descriptors.inode_of(arguments->socket));
+		auto message = BAN::ByteSpan(static_cast<uint8_t*>(arguments->buffer), arguments->length);
-		if (!inode->mode().ifsock())
+		return TRY(m_open_file_descriptors.recvfrom(arguments->socket, message, arguments->address, arguments->address_len));
-			return BAN::Error::from_errno(ENOTSOCK);
-
-		const auto status_flags = TRY(m_open_file_descriptors.status_flags_of(arguments->socket));
-		if ((status_flags & O_NONBLOCK) && !inode->can_read())
-			return BAN::Error::from_errno(EAGAIN);
-
-		BAN::ByteSpan buffer { reinterpret_cast<uint8_t*>(arguments->buffer), arguments->length };
-		return TRY(inode->recvfrom(buffer, arguments->address, arguments->address_len));
 	}
 
 	BAN::ErrorOr<long> Process::sys_ioctl(int fildes, int request, void* arg)
@@ -1675,7 +1697,7 @@ namespace Kernel
 		else
 			page_flags |= PageTable::Flags::UserSupervisor;
 
-		AddressRange address_range { .start = 0x400000, .end = KERNEL_OFFSET };
+		AddressRange address_range { .start = 0x400000, .end = USERSPACE_END };
 		if (args->flags & MAP_FIXED)
 		{
 			vaddr_t base_addr = reinterpret_cast<vaddr_t>(args->addr);
@@ -1820,7 +1842,7 @@ namespace Kernel
 
 	BAN::ErrorOr<long> Process::sys_smo_map(SharedMemoryObjectManager::Key key)
 	{
-		auto region = TRY(SharedMemoryObjectManager::get().map_object(key, page_table(), { .start = 0x400000, .end = KERNEL_OFFSET }));
+		auto region = TRY(SharedMemoryObjectManager::get().map_object(key, page_table(), { .start = 0x400000, .end = USERSPACE_END }));
 
 		LockGuard _(m_process_lock);
 		TRY(m_mapped_regions.push_back(BAN::move(region)));
@@ -2090,12 +2112,28 @@ namespace Kernel
 		return 0;
 	}
 
-	BAN::ErrorOr<long> Process::sys_pthread_create(const pthread_attr_t* __restrict attr, void (*entry)(void*), void* arg)
+	BAN::ErrorOr<long> Process::sys_set_tls(void* addr)
 	{
-		if (attr != nullptr)
+		Thread::current().set_tls(reinterpret_cast<vaddr_t>(addr));
+		Processor::load_tls();
+		return 0;
+	}
+
+	BAN::ErrorOr<long> Process::sys_get_tls()
+	{
+		return Thread::current().get_tls();
+	}
+
+	BAN::ErrorOr<long> Process::sys_pthread_create(const pthread_attr_t* attr, void (*entry)(void*), void* arg)
+	{
+		if (attr)
 		{
-			dwarnln("pthread attr not supported");
+			TRY(validate_pointer_access(attr, sizeof(*attr), false));
-			return BAN::Error::from_errno(ENOTSUP);
+			if (*attr)
+			{
+				dwarnln("pthread attr not supported");
+				return BAN::Error::from_errno(ENOTSUP);
+			}
 		}
 
 		LockGuard _(m_process_lock);
@@ -2575,7 +2613,7 @@ namespace Kernel
 			goto unauthorized_access;
 
 		// trying to access kernel space memory
-		if (vaddr + size > KERNEL_OFFSET)
+		if (vaddr + size > USERSPACE_END)
 			goto unauthorized_access;
 
 		if (vaddr == 0)

kernel/kernel/Processor.cpp

@@ -10,7 +10,10 @@ extern Kernel::TerminalDriver* g_terminal_driver;
 namespace Kernel
 {
 
+#if ARCH(x86_64)
+	static constexpr uint32_t MSR_IA32_FS_BASE = 0xC0000100;
 	static constexpr uint32_t MSR_IA32_GS_BASE = 0xC0000101;
+#endif
 
 	ProcessorID          Processor::s_bsb_id                { PROCESSOR_NONE };
 	BAN::Atomic<uint8_t> Processor::s_processor_count       { 0 };
@@ -260,6 +263,18 @@ namespace Kernel
 		set_interrupt_state(state);
 	}
 
+	void Processor::load_tls()
+	{
+		const auto addr = scheduler().current_thread().get_tls();
+#if ARCH(x86_64)
+		uint32_t ptr_hi = addr >> 32;
+		uint32_t ptr_lo = addr & 0xFFFFFFFF;
+		asm volatile("wrmsr" :: "d"(ptr_hi), "a"(ptr_lo), "c"(MSR_IA32_FS_BASE));
+#elif ARCH(i686)
+		gdt().set_tls(addr);
+#endif
+	}
+
 	void Processor::send_smp_message(ProcessorID processor_id, const SMPMessage& message, bool send_ipi)
 	{
 		ASSERT(processor_id != current_id());

kernel/kernel/Scheduler.cpp

@@ -278,6 +278,7 @@ namespace Kernel
 			thread->m_state = Thread::State::Executing;
 
 		Processor::gdt().set_tss_stack(thread->kernel_stack_top());
+		Processor::load_tls();
 		*interrupt_stack     = thread->interrupt_stack();
 		*interrupt_registers = thread->interrupt_registers();
 

kernel/kernel/Syscall.cpp

@@ -44,11 +44,7 @@ namespace Kernel
 
 		BAN::ErrorOr<long> ret = BAN::Error::from_errno(ENOSYS);
 
-		const char* process_path = nullptr;
+		const char* process_path = Process::current().name();
-		if (Process::current().userspace_info().argc > 0 && Process::current().userspace_info().argv)
-			process_path = Process::current().userspace_info().argv[0];
-		if (process_path == nullptr)
-			process_path = "<null>";
 
 #if DUMP_ALL_SYSCALLS
 		dprintln("{} pid {}: {}", process_path, Process::current().pid(), s_syscall_names[syscall]);

kernel/kernel/Thread.cpp

@@ -29,11 +29,6 @@ namespace Kernel
 		return Thread::current().interrupt_stack().sp;
 	}
 
-	extern "C" uintptr_t get_userspace_thread_stack_top()
-	{
-		return Thread::current().userspace_stack_top() - 4 * sizeof(uintptr_t);
-	}
-
 	extern "C" void load_thread_sse()
 	{
 		Thread::current().load_sse();
@@ -108,7 +103,7 @@ namespace Kernel
 
 		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
 			page_table,
-			0x200000, KERNEL_OFFSET,
+			0x200000, USERSPACE_END,
 			kernel_stack_size,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
 			true
@@ -116,7 +111,7 @@ namespace Kernel
 
 		thread->m_userspace_stack = TRY(VirtualRange::create_to_vaddr_range(
 			page_table,
-			0x200000, KERNEL_OFFSET,
+			0x200000, USERSPACE_END,
 			userspace_stack_size,
 			PageTable::Flags::UserSupervisor | PageTable::Flags::ReadWrite | PageTable::Flags::Present,
 			true
@@ -176,11 +171,12 @@ namespace Kernel
 		save_sse();
 		memcpy(thread->m_sse_storage, m_sse_storage, sizeof(m_sse_storage));
 
-		thread->setup_exec_impl(
+		PageTable::with_fast_page(thread->userspace_stack().paddr_of(thread->userspace_stack_top() - PAGE_SIZE), [=] {
-			reinterpret_cast<uintptr_t>(entry),
+			PageTable::fast_page_as<void*>(PAGE_SIZE - sizeof(uintptr_t)) = arg;
-			reinterpret_cast<uintptr_t>(arg),
+		});
-			0, 0, 0
+
-		);
+		const vaddr_t entry_addr = reinterpret_cast<vaddr_t>(entry);
+		thread->setup_exec(entry_addr, thread->userspace_stack_top() - sizeof(uintptr_t));
 
 		return thread;
 	}
@@ -200,6 +196,8 @@ namespace Kernel
 		thread->m_kernel_stack = TRY(m_kernel_stack->clone(new_process->page_table()));
 		thread->m_userspace_stack = TRY(m_userspace_stack->clone(new_process->page_table()));
 
+		thread->m_tls = m_tls;
+
 		thread->m_state = State::NotStarted;
 
 		thread->m_interrupt_stack.ip = ip;
@@ -222,21 +220,112 @@ namespace Kernel
 		return thread;
 	}
 
-	void Thread::setup_exec()
+	BAN::ErrorOr<void> Thread::initialize_userspace(vaddr_t entry, BAN::Span<BAN::String> argv, BAN::Span<BAN::String> envp, BAN::Span<LibELF::AuxiliaryVector> auxv)
 	{
-		const auto& userspace_info = process().userspace_info();
+		// System V ABI: Initial process stack
-		ASSERT(userspace_info.entry);
 
-		setup_exec_impl(
+		ASSERT(m_is_userspace);
-			userspace_info.entry,
+		ASSERT(m_userspace_stack);
-			userspace_info.argc,
+
-			reinterpret_cast<uintptr_t>(userspace_info.argv),
+		size_t needed_size = 0;
-			reinterpret_cast<uintptr_t>(userspace_info.envp),
+
-			userspace_info.file_fd
+		// argc
-		);
+		needed_size += sizeof(uintptr_t);
+
+		// argv
+		needed_size += (argv.size() + 1) * sizeof(uintptr_t);
+		for (auto arg : argv)
+			needed_size += arg.size() + 1;
+
+		// envp
+		needed_size += (envp.size() + 1) * sizeof(uintptr_t);
+		for (auto env : envp)
+			needed_size += env.size() + 1;
+
+		// auxv
+		needed_size += auxv.size() * sizeof(LibELF::AuxiliaryVector);
+
+		if (needed_size > m_userspace_stack->size())
+			return BAN::Error::from_errno(ENOBUFS);
+
+		vaddr_t vaddr = userspace_stack_top() - needed_size;
+
+		const auto stack_copy_buf =
+			[this](BAN::ConstByteSpan buffer, vaddr_t vaddr) -> void
+			{
+				ASSERT(vaddr + buffer.size() <= userspace_stack_top());
+
+				size_t bytes_copied = 0;
+				while (bytes_copied < buffer.size())
+				{
+					const size_t to_copy = BAN::Math::min<size_t>(buffer.size() - bytes_copied, PAGE_SIZE - (vaddr % PAGE_SIZE));
+
+					PageTable::with_fast_page(userspace_stack().paddr_of(vaddr & PAGE_ADDR_MASK), [=]() {
+						memcpy(PageTable::fast_page_as_ptr(vaddr % PAGE_SIZE), buffer.data() + bytes_copied, to_copy);
+					});
+
+					vaddr += to_copy;
+					bytes_copied += to_copy;
+				}
+			};
+
+		const auto stack_push_buf =
+			[&stack_copy_buf, &vaddr](BAN::ConstByteSpan buffer) -> void
+			{
+				stack_copy_buf(buffer, vaddr);
+				vaddr += buffer.size();
+			};
+
+		const auto stack_push_uint =
+			[&stack_push_buf](uintptr_t value) -> void
+			{
+				stack_push_buf(BAN::ConstByteSpan::from(value));
+			};
+
+		const auto stack_push_str =
+			[&stack_push_buf](BAN::StringView string) -> void
+			{
+				const uint8_t* string_u8 = reinterpret_cast<const uint8_t*>(string.data());
+				stack_push_buf(BAN::ConstByteSpan(string_u8, string.size() + 1));
+			};
+
+		// argc
+		stack_push_uint(argv.size());
+
+		// argv
+		const vaddr_t argv_vaddr = vaddr;
+		vaddr += argv.size() * sizeof(uintptr_t);
+		stack_push_uint(0);
+
+		// envp
+		const vaddr_t envp_vaddr = vaddr;
+		vaddr += envp.size() * sizeof(uintptr_t);
+		stack_push_uint(0);
+
+		// auxv
+		for (auto aux : auxv)
+			stack_push_buf(BAN::ConstByteSpan::from(aux));
+
+		// information
+		for (size_t i = 0; i < argv.size(); i++)
+		{
+			stack_copy_buf(BAN::ConstByteSpan::from(vaddr), argv_vaddr + i * sizeof(uintptr_t));
+			stack_push_str(argv[i]);
+		}
+		for (size_t i = 0; i < envp.size(); i++)
+		{
+			stack_copy_buf(BAN::ConstByteSpan::from(vaddr), envp_vaddr + i * sizeof(uintptr_t));
+			stack_push_str(envp[i]);
+		}
+
+		ASSERT(vaddr == userspace_stack_top());
+
+		setup_exec(entry, userspace_stack_top() - needed_size);
+
+		return {};
 	}
 
-	void Thread::setup_exec_impl(uintptr_t entry, uintptr_t arg0, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3)
+	void Thread::setup_exec(vaddr_t ip, vaddr_t sp)
 	{
 		ASSERT(is_userspace());
 		m_state = State::NotStarted;
@@ -244,13 +333,13 @@ namespace Kernel
 		// Signal mask is inherited
 
 		// Initialize stack for returning
-		PageTable::with_fast_page(process().page_table().physical_address_of(kernel_stack_top() - PAGE_SIZE), [&] {
+		PageTable::with_fast_page(kernel_stack().paddr_of(kernel_stack_top() - PAGE_SIZE), [=] {
-			uintptr_t sp = PageTable::fast_page() + PAGE_SIZE;
+			uintptr_t cur_sp = PageTable::fast_page() + PAGE_SIZE;
-			write_to_stack(sp, entry);
+			write_to_stack(cur_sp, 0x20 | 3);
-			write_to_stack(sp, arg3);
+			write_to_stack(cur_sp, sp);
-			write_to_stack(sp, arg2);
+			write_to_stack(cur_sp, 0x202);
-			write_to_stack(sp, arg1);
+			write_to_stack(cur_sp, 0x18 | 3);
-			write_to_stack(sp, arg0);
+			write_to_stack(cur_sp, ip);
 		});
 
 		m_interrupt_stack.ip = reinterpret_cast<vaddr_t>(start_userspace_thread);
@@ -286,7 +375,7 @@ namespace Kernel
 		m_signal_pending_mask = 0;
 		m_signal_block_mask = ~0ull;
 
-		PageTable::with_fast_page(process().page_table().physical_address_of(kernel_stack_top() - PAGE_SIZE), [&] {
+		PageTable::with_fast_page(kernel_stack().paddr_of(kernel_stack_top() - PAGE_SIZE), [&] {
 			uintptr_t sp = PageTable::fast_page() + PAGE_SIZE;
 			write_to_stack(sp, this);
 			write_to_stack(sp, &Thread::on_exit_trampoline);

userspace/libraries/LibC/CMakeLists.txt

@@ -58,6 +58,10 @@ target_compile_definitions(objlibc PRIVATE __arch=${BANAN_ARCH})
 target_compile_options(objlibc PRIVATE -O2 -g -Wstack-usage=512 -fno-exceptions -fpic -nolibc)
 target_compile_options(objlibc PUBLIC -Wall -Wextra -Werror -Wno-error=stack-usage=)
 
+if("${BANAN_ARCH}" STREQUAL "i686")
+	target_compile_definitions(objlibc PRIVATE __disable_thread_local_storage)
+endif()
+
 function(add_crtx crtx)
 	add_custom_target(${crtx}
 		COMMAND ${CMAKE_CXX_COMPILER} -c -o ${CMAKE_INSTALL_LIBDIR}/${crtx}.o ${CMAKE_CURRENT_SOURCE_DIR}/arch/${BANAN_ARCH}/${crtx}.S
@@ -71,6 +75,7 @@ add_crtx(crtn)
 
 banan_include_headers(objlibc ban)
 banan_include_headers(objlibc kernel)
+banan_include_headers(objlibc libelf)
 
 banan_install_headers(objlibc)
 

userspace/libraries/LibC/arch/i686/crt0.S

@@ -2,40 +2,38 @@
 
 .global _start
 _start:
-	pushl $0
+	# get argc, argv, envp
-	pushl %edi
+	movl  (%esp), %edi
-	pushl %esi
+	leal 4(%esp), %esi
-	pushl %edx
+	leal 4(%esi, %edi, 4), %edx
-
-	# STACK LAYOUT
-	#   null
-	#   argc
-	#   argv
-	#   envp
 
+	# align stack
+	andl $-16, %esp
 	xorl %ebp, %ebp
 
-	# init libc (envp already as argument)
+	# save argc, argv, envp
+	subl $4, %esp
+	pushl %edx
+	pushl %esi
+	pushl %edi
+
+	subl $4, %esp
+
+	pushl $__fini_array_end
+	pushl $__fini_array_start
+	pushl $_fini
+
+	pushl $__init_array_end
+	pushl $__init_array_start
+	pushl $_init
+
+	pushl %edx
+
 	call _init_libc
 
-	# call global constructors
+	addl $(4 * 8), %esp
-	movl $_init, %eax
-	testl %eax, %eax
-	jz 1f
-	call *%eax
- 1:
 
-	movl $__init_array_start, %ebx
+	# argc, argv, envp already on stack
-	jmp 2f
- 1: call *(%ebx)
-	addl $4, %ebx
- 2: cmpl $__init_array_end, %ebx
-	jne 1b
-
-	# call main
-	movl 0(%esp), %eax
-	xchgl %eax, 8(%esp)
-	movl %eax, (%esp)
 	call main
 
 	subl $12, %esp

userspace/libraries/LibC/arch/x86_64/crt0.S

@@ -2,41 +2,40 @@
 
 .global _start
 _start:
-	pushq $0
+	# get argc, argv, envp
+	movq  (%rsp), %rdi
+	leaq 8(%rsp), %rsi
+	leaq 8(%rsi, %rdi, 8), %rdx
+
+	# align stack
+	andq $-16, %rsp
+	xorq %rbp, %rbp
+
+	# save argc, argv, envp
+	subq $8, %rsp
 	pushq %rdi
 	pushq %rsi
 	pushq %rdx
 
-	# STACK LAYOUT
+	movq %rdx, %rdi # environ
-	#   null
-	#   argc
-	#   argv
-	#   envp
 
-	xorq %rbp, %rbp
+	pushq $__fini_array_end
+	pushq $__fini_array_start
+	pushq $_fini
+
+	pushq $__init_array_end
+	pushq $__init_array_start
+	pushq $_init
 
-	# init libc
-	movq 0(%rsp), %rdi
 	call _init_libc
 
-	# call global constructors
+	addq $(6 * 8), %rsp
-	movq $_init, %rax
-	testq %rax, %rax
-	jz 1f
-	call *%rax
- 1:
-
-	movq $__init_array_start, %rbx
-	jmp 2f
- 1: call *(%rbx)
-	addq $8, %rbx
- 2: cmpq $__init_array_end, %rbx
-	jne 1b
 
 	# call main
-	movq 16(%rsp), %rdi
+	popq %rdx
-	movq  8(%rsp), %rsi
+	popq %rsi
-	movq  0(%rsp), %rdx
+	popq %rdi
+	addq $8, %rsp
 	call main
 
 	# call exit

userspace/libraries/LibC/errno.cpp

@@ -1,6 +1,10 @@
 #include <errno.h>
 
+#if __disable_thread_local_storage
 static int s_errno = 0;
+#else
+static thread_local int s_errno = 0;
+#endif
 
 int* __errno_location()
 {

userspace/libraries/LibC/icxxabi.cpp

@@ -1,41 +1,39 @@
-#include <BAN/Assert.h>
-#include <stdint.h>
 #include <stddef.h>
 
 #define ATEXIT_MAX_FUNCS 128
 
 struct atexit_func_entry_t
 {
-	void(*func)(void*);
+	void (*func)(void*);
 	void* arg;
 	void* dso_handle;
 };
 
-static atexit_func_entry_t __atexit_funcs[ATEXIT_MAX_FUNCS];
+static atexit_func_entry_t s_atexit_funcs[ATEXIT_MAX_FUNCS];
-static size_t __atexit_func_count = 0;
+static size_t s_atexit_func_count = 0;
 
 extern "C" int __cxa_atexit(void(*func)(void*), void* arg, void* dso_handle)
 {
-	if (__atexit_func_count >= ATEXIT_MAX_FUNCS)
+	if (s_atexit_func_count >= ATEXIT_MAX_FUNCS)
 		return -1;
-	auto& atexit_func = __atexit_funcs[__atexit_func_count++];
+	s_atexit_funcs[s_atexit_func_count++] = {
-	atexit_func.func = func;
+		.func = func,
-	atexit_func.arg = arg;
+		.arg = arg,
-	atexit_func.dso_handle = dso_handle;
+		.dso_handle = dso_handle,
+	};
 	return 0;
 };
 
-extern "C" void __cxa_finalize(void* f)
+extern "C" void __cxa_finalize(void* dso_handle)
 {
-	for (size_t i = __atexit_func_count; i > 0; i--)
+	for (size_t i = s_atexit_func_count; i > 0; i--)
 	{
-		auto& atexit_func = __atexit_funcs[i - 1];
+		auto& atexit_func = s_atexit_funcs[i - 1];
 		if (atexit_func.func == nullptr)
 			continue;
-		if (f == nullptr || f == atexit_func.func)
+		if (dso_handle && dso_handle != atexit_func.dso_handle)
-		{
+			continue;
-			atexit_func.func(atexit_func.arg);
+		atexit_func.func(atexit_func.arg);
-			atexit_func.func = nullptr;
+		atexit_func.func = nullptr;
-		}
 	}
 };

userspace/libraries/LibC/include/netdb.h

@@ -54,6 +54,7 @@ struct hostent
 	int		h_addrtype;		/* Address type. */
 	int		h_length;		/* The length, in bytes, of the address. */
 	char**	h_addr_list;	/* A pointer to an array of pointers to network addresses (in network byte order) for the host, terminated by a null pointer. */
+#define h_addr h_addr_list[0] /* Backwards compatibility */
 };
 
 struct netent

userspace/libraries/LibC/include/pthread.h

@@ -8,6 +8,7 @@
 __BEGIN_DECLS
 
 #include <sched.h>
+#include <stdint.h>
 #include <time.h>
 
 #define __need_size_t
@@ -27,6 +28,14 @@ __BEGIN_DECLS
 #define __need_pthread_t
 #include <sys/types.h>
 
+struct uthread
+{
+	struct uthread* self;
+	void* master_tls_addr;
+	size_t master_tls_size;
+	uintptr_t dtv[];
+};
+
 #define PTHREAD_BARRIER_SERIAL_THREAD	1
 #define PTHREAD_CANCEL_ASYNCHRONOUS		2
 #define PTHREAD_CANCEL_ENABLE			3

userspace/libraries/LibC/include/sys/syscall.h

@@ -91,6 +91,8 @@ __BEGIN_DECLS
 	O(SYS_SYMLINKAT,		symlinkat)		\
 	O(SYS_HARDLINKAT,		hardlinkat)		\
 	O(SYS_YIELD,			yield)			\
+	O(SYS_SET_TLS,			set_tls)		\
+	O(SYS_GET_TLS,			get_tls)		\
 	O(SYS_PTHREAD_CREATE,	pthread_create)	\
 	O(SYS_PTHREAD_EXIT,		pthread_exit)	\
 	O(SYS_PTHREAD_JOIN,		pthread_join)	\

userspace/libraries/LibC/malloc.cpp

@@ -2,6 +2,7 @@
 
 #include <assert.h>
 #include <errno.h>
+#include <pthread.h>
 #include <stdlib.h>
 #include <string.h>
 #include <sys/mman.h>
@@ -50,19 +51,29 @@ struct malloc_pool_t
 	bool contains(malloc_node_t* node) { return start <= (uint8_t*)node && (uint8_t*)node < end(); }
 };
 
-static malloc_pool_t s_malloc_pools[s_malloc_pool_count];
+struct malloc_info_t
-
-void _init_malloc()
 {
-	size_t pool_size = s_malloc_pool_size_initial;
+	consteval malloc_info_t()
-	for (size_t i = 0; i < s_malloc_pool_count; i++)
 	{
-		s_malloc_pools[i].start = nullptr;
+		size_t pool_size = s_malloc_pool_size_initial;
-		s_malloc_pools[i].size = pool_size;
+		for (auto& pool : pools)
-		s_malloc_pools[i].free_list = nullptr;;
+		{
-		pool_size *= s_malloc_pool_size_multiplier;
+			pool = {
+				.start = nullptr,
+				.size = pool_size,
+				.free_list = nullptr,
+			};
+			pool_size *= s_malloc_pool_size_multiplier;
+		}
 	}
-}
+
+	malloc_pool_t pools[s_malloc_pool_count];
+};
+
+static malloc_info_t s_malloc_info;
+static auto& s_malloc_pools = s_malloc_info.pools;
+
+static pthread_spinlock_t s_malloc_lock;
 
 static bool allocate_pool(size_t pool_index)
 {
@@ -192,19 +203,31 @@ void* malloc(size_t size)
 
 	// try to find any already existing pools that we can allocate in
 	for (size_t i = first_usable_pool; i < s_malloc_pool_count; i++)
-		if (s_malloc_pools[i].start != nullptr)
+	{
-			if (void* ret = allocate_from_pool(i, size))
+		if (s_malloc_pools[i].start == nullptr)
-				return ret;
+			continue;
+		pthread_spin_lock(&s_malloc_lock);
+		void* ret = allocate_from_pool(i, size);
+		pthread_spin_unlock(&s_malloc_lock);
+		if (ret != nullptr)
+			return ret;
+	}
 
 	// allocate new pool
 	for (size_t i = first_usable_pool; i < s_malloc_pool_count; i++)
 	{
 		if (s_malloc_pools[i].start != nullptr)
 			continue;
-		if (!allocate_pool(i))
+
+		pthread_spin_lock(&s_malloc_lock);
+		void* ret = nullptr;
+		if (allocate_pool(i))
+			ret = allocate_from_pool(i, size);
+		pthread_spin_unlock(&s_malloc_lock);
+
+		if (ret == nullptr)
 			break;
-		// NOTE: always works since we just created the pool
+		return ret;
-		return allocate_from_pool(i, size);
 	}
 
 	errno = ENOMEM;
@@ -250,6 +273,8 @@ void free(void* ptr)
 	if (ptr == nullptr)
 		return;
 
+	pthread_spin_lock(&s_malloc_lock);
+
 	auto* node = node_from_data_pointer(ptr);
 
 	node->allocated = false;
@@ -271,6 +296,8 @@ void free(void* ptr)
 	node->prev_free = nullptr;
 	node->next_free = pool.free_list;
 	pool.free_list = node;
+
+	pthread_spin_unlock(&s_malloc_lock);
 }
 
 void* calloc(size_t nmemb, size_t size)

userspace/libraries/LibC/pthread.cpp

@@ -2,55 +2,195 @@
 #include <BAN/Atomic.h>
 #include <BAN/PlacementNew.h>
 
+#include <kernel/Arch.h>
+
 #include <errno.h>
 #include <pthread.h>
 #include <stdlib.h>
 #include <string.h>
+#include <sys/mman.h>
 #include <sys/syscall.h>
 #include <unistd.h>
 
 struct pthread_trampoline_info_t
 {
+	struct uthread* uthread;
 	void* (*start_routine)(void*);
 	void* arg;
 };
 
 // stack is 16 byte aligned on entry, this `call` is used to align it
-extern "C" void pthread_trampoline(void*);
+extern "C" void _pthread_trampoline(void*);
-asm("pthread_trampoline: call pthread_trampoline_cpp");
+asm(
+#if ARCH(x86_64)
+"_pthread_trampoline:"
+	"popq %rdi;"
+	"andq $-16, %rsp;"
+	"xorq %rbp, %rbp;"
+	"call _pthread_trampoline_cpp"
+#elif ARCH(i686)
+"_pthread_trampoline:"
+	"ud2;"
+	"popl %edi;"
+	"andl $-16, %esp;"
+	"xorl %ebp, %ebp;"
+	"subl $12, %esp;"
+	"pushl %edi;"
+	"call _pthread_trampoline_cpp"
+#endif
+);
 
-extern "C" void pthread_trampoline_cpp(void* arg)
+extern "C" void _pthread_trampoline_cpp(void* arg)
 {
-	pthread_trampoline_info_t info;
+	auto info = *reinterpret_cast<pthread_trampoline_info_t*>(arg);
-	memcpy(&info, arg, sizeof(pthread_trampoline_info_t));
+	syscall(SYS_SET_TLS, info.uthread);
 	free(arg);
-
 	pthread_exit(info.start_routine(info.arg));
 	ASSERT_NOT_REACHED();
 }
 
-int pthread_create(pthread_t* __restrict thread, const pthread_attr_t* __restrict attr, void* (*start_routine)(void*), void* __restrict arg)
+static uthread* get_uthread()
+{
+	uthread* result;
+#if ARCH(x86_64)
+	asm volatile("movq %%fs:0, %0" : "=r"(result));
+#elif ARCH(i686)
+	asm volatile("movl %%gs:0, %0" : "=r"(result));
+#endif
+	return result;
+}
+
+static void free_uthread(uthread* uthread)
+{
+	if (uthread->dtv[0] == 0)
+		return free(uthread);
+
+	uint8_t* tls_addr = reinterpret_cast<uint8_t*>(uthread) - uthread->master_tls_size;
+	const size_t tls_size = uthread->master_tls_size
+		+ sizeof(struct uthread)
+		+ (uthread->dtv[0] + 1) * sizeof(uintptr_t);
+	munmap(tls_addr, tls_size);
+}
+
+#if not __disable_thread_local_storage
+struct pthread_cleanup_t
+{
+	void (*routine)(void*);
+	void* arg;
+	pthread_cleanup_t* next;
+};
+
+static thread_local pthread_cleanup_t* s_cleanup_stack = nullptr;
+
+void pthread_cleanup_pop(int execute)
+{
+	ASSERT(s_cleanup_stack);
+
+	auto* cleanup = s_cleanup_stack;
+	s_cleanup_stack = cleanup->next;
+
+	if (execute)
+		cleanup->routine(cleanup->arg);
+
+	free(cleanup);
+}
+
+void pthread_cleanup_push(void (*routine)(void*), void* arg)
+{
+	auto* cleanup = static_cast<pthread_cleanup_t*>(malloc(sizeof(pthread_cleanup_t)));
+	ASSERT(cleanup);
+
+	cleanup->routine = routine;
+	cleanup->arg = arg;
+	cleanup->next = s_cleanup_stack;
+
+	s_cleanup_stack = cleanup;
+}
+#endif
+
+int pthread_attr_init(pthread_attr_t* attr)
+{
+	*attr = 0;
+	return 0;
+}
+
+int pthread_create(pthread_t* __restrict thread_id, const pthread_attr_t* __restrict attr, void* (*start_routine)(void*), void* __restrict arg)
 {
 	auto* info = static_cast<pthread_trampoline_info_t*>(malloc(sizeof(pthread_trampoline_info_t)));
 	if (info == nullptr)
-		return -1;
+		return errno;
-	info->start_routine = start_routine;
-	info->arg = arg;
 
-	const auto ret = syscall(SYS_PTHREAD_CREATE, attr, pthread_trampoline, info);
+	*info = {
-	if (ret == -1)
+		.uthread = nullptr,
+		.start_routine = start_routine,
+		.arg = arg,
+	};
+
+	long syscall_ret = 0;
+
+	if (uthread* self = get_uthread(); self->master_tls_addr == nullptr)
 	{
-		free(info);
+		uthread* uthread = static_cast<struct uthread*>(malloc(sizeof(struct uthread) + sizeof(uintptr_t)));
-		return -1;
+		if (uthread == nullptr)
+			goto pthread_create_error;
+		uthread->self = uthread;
+		uthread->master_tls_addr = nullptr;
+		uthread->master_tls_size = 0;
+		uthread->dtv[0] = 0;
+
+		info->uthread = uthread;
+	}
+	else
+	{
+		const size_t module_count = self->dtv[0];
+
+		const size_t tls_size = self->master_tls_size
+			+ sizeof(uthread)
+			+ (module_count + 1) * sizeof(uintptr_t);
+
+		uint8_t* tls_addr = static_cast<uint8_t*>(mmap(nullptr, tls_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0));
+		if (tls_addr == MAP_FAILED)
+			goto pthread_create_error;
+		memcpy(tls_addr, self->master_tls_addr, self->master_tls_size);
+
+		uthread* uthread = reinterpret_cast<struct uthread*>(tls_addr + self->master_tls_size);
+		uthread->self = uthread;
+		uthread->master_tls_addr = self->master_tls_addr;
+		uthread->master_tls_size = self->master_tls_size;
+
+		const uintptr_t self_addr = reinterpret_cast<uintptr_t>(self);
+		const uintptr_t uthread_addr = reinterpret_cast<uintptr_t>(uthread);
+
+		uthread->dtv[0] = module_count;
+		for (size_t i = 1; i <= module_count; i++)
+			uthread->dtv[i] = self->dtv[i] - self_addr + uthread_addr;
+
+		info->uthread = uthread;
 	}
 
-	if (thread)
+	syscall_ret = syscall(SYS_PTHREAD_CREATE, attr, _pthread_trampoline, info);
-		*thread = ret;
+	if (syscall_ret == -1)
+		goto pthread_create_error;
+
+	if (thread_id)
+		*thread_id = syscall_ret;
 	return 0;
+
+pthread_create_error:
+	const int return_code = errno;
+	if (info->uthread)
+		free_uthread(info->uthread);
+	free(info);
+	return return_code;
 }
 
 void pthread_exit(void* value_ptr)
 {
+#if not __disable_thread_local_storage
+	while (s_cleanup_stack)
+		pthread_cleanup_pop(1);
+#endif
+	free_uthread(get_uthread());
 	syscall(SYS_PTHREAD_EXIT, value_ptr);
 	ASSERT_NOT_REACHED();
 }
@@ -62,7 +202,14 @@ int pthread_join(pthread_t thread, void** value_ptr)
 
 pthread_t pthread_self(void)
 {
+#if __disable_thread_local_storage
 	return syscall(SYS_PTHREAD_SELF);
+#else
+	static thread_local pthread_t s_pthread_self { -1 };
+	if (s_pthread_self == -1) [[unlikely]]
+		s_pthread_self = syscall(SYS_PTHREAD_SELF);
+	return s_pthread_self;
+#endif
 }
 
 static inline BAN::Atomic<pthread_t>& pthread_spin_get_atomic(pthread_spinlock_t* lock)
@@ -123,3 +270,21 @@ int pthread_spin_unlock(pthread_spinlock_t* lock)
 	atomic.store(0, BAN::MemoryOrder::memory_order_release);
 	return 0;
 }
+
+struct tls_index
+{
+	unsigned long int ti_module;
+	unsigned long int ti_offset;
+};
+
+extern "C" void* __tls_get_addr(tls_index* ti)
+{
+	return reinterpret_cast<void*>(get_uthread()->dtv[ti->ti_module] + ti->ti_offset);
+}
+
+#if ARCH(i686)
+extern "C" void* __attribute__((__regparm__(1))) ___tls_get_addr(tls_index* ti)
+{
+	return reinterpret_cast<void*>(get_uthread()->dtv[ti->ti_module] + ti->ti_offset);
+}
+#endif

userspace/libraries/LibC/stdio.cpp

@@ -94,7 +94,8 @@ static int drop_read_buffer(FILE* file)
 	return 0;
 }
 
-void _init_stdio()
+__attribute__((constructor))
+static void _init_stdio()
 {
 	for (size_t i = 0; i < FOPEN_MAX; i++)
 	{

userspace/libraries/LibC/stdlib.cpp

@@ -14,16 +14,14 @@
 #include <strings.h>
 #include <sys/stat.h>
 #include <sys/syscall.h>
+#include <sys/weak_alias.h>
 #include <unistd.h>
 
 #include <icxxabi.h>
 
-extern "C" char** environ;
+char** __environ;
-
+weak_alias(__environ, environ);
-extern "C" __attribute__((weak)) void _fini();
+static bool s_environ_malloced = false;
-
-static void (*at_exit_funcs[64])();
-static uint32_t at_exit_funcs_count = 0;
 
 void abort(void)
 {
@@ -41,11 +39,8 @@ void abort(void)
 
 void exit(int status)
 {
-	for (uint32_t i = at_exit_funcs_count; i > 0; i--)
-		at_exit_funcs[i - 1]();
 	fflush(nullptr);
 	__cxa_finalize(nullptr);
-	if (_fini) _fini();
 	_exit(status);
 	ASSERT_NOT_REACHED();
 }
@@ -62,13 +57,8 @@ int abs(int val)
 
 int atexit(void (*func)(void))
 {
-	if (at_exit_funcs_count > sizeof(at_exit_funcs) / sizeof(*at_exit_funcs))
+	void* func_addr = reinterpret_cast<void*>(func);
-	{
+	return __cxa_atexit([](void* func_ptr) { reinterpret_cast<void (*)(void)>(func_ptr)(); }, func_addr, nullptr);
-		errno = ENOBUFS;
-		return -1;
-	}
-	at_exit_funcs[at_exit_funcs_count++] = func;
-	return 0;
 }
 
 static constexpr int get_base_digit(char c, int base)
@@ -516,27 +506,35 @@ int putenv(char* string)
 		return -1;
 	}
 
-	if (!environ)
+	if (!s_environ_malloced)
 	{
-		environ = (char**)malloc(sizeof(char*) * 2);
+		size_t env_count = 0;
-		if (!environ)
+		while (environ[env_count])
+			env_count++;
+
+		char** new_environ = static_cast<char**>(malloc((env_count + 1) * sizeof(char*)));
+		if (new_environ == nullptr)
 			return -1;
-		environ[0] = string;
+		for (size_t i = 0; i < env_count; i++)
-		environ[1] = nullptr;
+		{
-		return 0;
+			const size_t bytes = strlen(environ[i]) + 1;
+			new_environ[i] = (char*)malloc(bytes);
+			memcpy(new_environ[i], environ[i], bytes);
+		}
+		new_environ[env_count] = nullptr;
+
+		environ = new_environ;
+		s_environ_malloced = true;
 	}
 
-	int cnt = 0;
+	const char* eq_addr = strchr(string, '=');
-	for (int i = 0; string[i]; i++)
+	if (eq_addr == nullptr)
-		if (string[i] == '=')
-			cnt++;
-	if (cnt != 1)
 	{
 		errno = EINVAL;
 		return -1;
 	}
 
-	int namelen = strchr(string, '=') - string;
+	size_t namelen = eq_addr - string;
 	for (int i = 0; environ[i]; i++)
 	{
 		if (strncmp(environ[i], string, namelen + 1) == 0)
@@ -547,15 +545,15 @@ int putenv(char* string)
 		}
 	}
 
-	int env_count = 0;
+	size_t env_count = 0;
 	while (environ[env_count])
 		env_count++;
 
-	char** new_envp = (char**)malloc(sizeof(char*) * (env_count + 2));
+	char** new_envp = static_cast<char**>(malloc(sizeof(char*) * (env_count + 2)));
 	if (new_envp == nullptr)
 		return -1;
 
-	for (int i = 0; i < env_count; i++)
+	for (size_t i = 0; i < env_count; i++)
 		new_envp[i] = environ[i];
 	new_envp[env_count] = string;
 	new_envp[env_count + 1] = nullptr;

userspace/libraries/LibC/sys/time.cpp

@@ -17,6 +17,11 @@ int gettimeofday(struct timeval* __restrict tp, void* __restrict tzp)
 	return 0;
 }
 
+int getitimer(int which, struct itimerval* value)
+{
+	return setitimer(which, nullptr, value);
+}
+
 int setitimer(int which, const struct itimerval* __restrict value, struct itimerval* __restrict ovalue)
 {
 	return syscall(SYS_SETITIMER, which, value, ovalue);

userspace/libraries/LibC/unistd.cpp

@@ -1,10 +1,12 @@
 #include <BAN/Assert.h>
 #include <BAN/Debug.h>
+
 #include <kernel/Memory/Types.h>
 #include <kernel/Syscall.h>
 
 #include <errno.h>
 #include <fcntl.h>
+#include <pthread.h>
 #include <pwd.h>
 #include <stdarg.h>
 #include <stdio.h>
@@ -16,36 +18,48 @@
 #include <termios.h>
 #include <unistd.h>
 
-char** __environ;
+struct init_funcs_t
-extern char** environ __attribute__((weak, alias("__environ")));
-
-extern void _init_malloc();
-extern void _init_stdio();
-extern "C" void _init_libc(char** _environ)
 {
-	static bool is_initialized = false;
+	void (*func)();
-	if (is_initialized)
+	void (**array_start)();
-		return;
+	void (**array_end)();
-	is_initialized = true;
+};
 
-	_init_malloc();
+extern "C" char** environ;
-	_init_stdio();
 
-	if (!_environ)
+extern "C" void _init_libc(char** environ, init_funcs_t init_funcs, init_funcs_t fini_funcs)
-		return;
+{
+	if (::environ == nullptr)
+		::environ = environ;
 
-	size_t env_count = 0;
+	if (syscall(SYS_GET_TLS) == 0)
-	while (_environ[env_count])
-		env_count++;
-
-	environ = (char**)malloc(sizeof(char*) * env_count + 1);
-	for (size_t i = 0; i < env_count; i++)
 	{
-		size_t bytes = strlen(_environ[i]) + 1;
+		alignas(uthread) static uint8_t storage[sizeof(uthread) + sizeof(uintptr_t)];
-		environ[i] = (char*)malloc(bytes);
+
-		memcpy(environ[i], _environ[i], bytes);
+		uthread& uthread = *reinterpret_cast<struct uthread*>(storage);
+		uthread = {
+			.self = &uthread,
+			.master_tls_addr = nullptr,
+			.master_tls_size = 0,
+		};
+		uthread.dtv[0] = 0;
+
+		syscall(SYS_SET_TLS, &uthread);
 	}
-	environ[env_count] = nullptr;
+
+	// call global constructors
+	if (init_funcs.func)
+		init_funcs.func();
+	const size_t init_array_count = init_funcs.array_end - init_funcs.array_start;
+	for (size_t i = 0; i < init_array_count; i++)
+		init_funcs.array_start[i]();
+
+	// register global destructors
+	const size_t fini_array_count = fini_funcs.array_end - fini_funcs.array_start;
+	for (size_t i = 0; i < fini_array_count; i++)
+		atexit(fini_funcs.array_start[i]);
+	if (fini_funcs.func)
+		atexit(fini_funcs.func);
 }
 
 void _exit(int status)

userspace/libraries/LibELF/include/LibELF/AuxiliaryVector.h

@@ -0,0 +1,28 @@
+#pragma once
+
+#include <stdint.h>
+
+namespace LibELF
+{
+
+	struct AuxiliaryVector
+	{
+		uint32_t a_type;
+		union
+		{
+			uint32_t a_val;
+			void* a_ptr;
+		} a_un;
+	};
+
+	enum AuxiliaryVectorValues
+	{
+		AT_NULL   = 0,
+		AT_IGNORE = 1,
+		AT_EXECFD = 2,
+		AT_PHDR   = 3,
+		AT_PHENT  = 4,
+		AT_PHNUM  = 5,
+	};
+
+}

userspace/libraries/LibELF/include/LibELF/Values.h

@@ -109,6 +109,7 @@ namespace LibELF
 		STT_FUNC	= 2,
 		STT_SECTION	= 3,
 		STT_FILE	= 4,
+		STT_TLS		= 6,
 		STT_LOOS	= 10,
 		STT_HIOS	= 12,
 		STT_LOPROC	= 13,
@@ -124,6 +125,7 @@ namespace LibELF
 		PT_NOTE		= 4,
 		PT_SHLIB	= 5,
 		PT_PHDR		= 6,
+		PT_TLS		= 7,
 		PT_LOOS		= 0x60000000,
 		PT_GNU_EH_FRAME = 0x6474E550,
 		PT_GNU_STACK    = 0x6474E551,
@@ -194,6 +196,26 @@ namespace LibELF
 		R_386_RELATIVE = 8,
 		R_386_GOTOFF   = 9,
 		R_386_GOTPC    = 10,
+		R_386_TLS_TPOFF		= 14,
+		R_386_TLS_IE		= 15,
+		R_386_TLS_GOTIE		= 16,
+		R_386_TLS_LE		= 17,
+		R_386_TLS_GD		= 18,
+		R_386_TLS_LDM		= 19,
+		R_386_TLS_GD_32		= 24,
+		R_386_TLS_GD_PUSH	= 25,
+		R_386_TLS_GD_CALL	= 26,
+		R_386_TLS_GD_POP	= 27,
+		R_386_TLS_LDM_32	= 28,
+		R_386_TLS_LDM_PUSH	= 29,
+		R_386_TLS_LDM_CALL	= 30,
+		R_386_TLS_LDM_POP	= 31,
+		R_386_TLS_LDO_32	= 32,
+		R_386_TLS_IE_32		= 33,
+		R_386_TLS_LE_32		= 34,
+		R_386_TLS_DTPMOD32	= 35,
+		R_386_TLS_DTPOFF32	= 36,
+		R_386_TLS_TPOFF32	= 37,
 	};
 
 #define ELF64_R_SYM(i) ((i) >> 32)

userspace/programs/DynamicLoader/main.cpp

@@ -1,56 +1,66 @@
 #include "utils.h"
 
+#include <LibELF/AuxiliaryVector.h>
 #include <LibELF/Types.h>
 #include <LibELF/Values.h>
 
 #include <fcntl.h>
 #include <limits.h>
+#include <pthread.h>
 #include <sys/mman.h>
 #include <sys/syscall.h>
 #include <unistd.h>
 
+#if defined(__x86_64__)
+	#define ELF_R_SYM ELF64_R_SYM
+#elif defined(__i686__)
+	#define ELF_R_SYM ELF32_R_SYM
+#endif
+
 extern "C"
 __attribute__((naked))
 void _start()
 {
 #if defined(__x86_64__)
 	asm volatile(
-		"xorq %rbp, %rbp;"
+		"movq  (%rsp), %rdi;"
+		"leaq 8(%rsp), %rsi;"
+		"leaq 8(%rsi, %rdi, 8), %rdx;"
+
+		"movq %rsp, %rbp;"
+		"andq $-16, %rsp;"
+
 		"call _entry;"
+
+		"movq %rbp, %rsp;"
+		"xorq %rbp, %rbp;"
+
+		"jmp *%rax;"
+
 		"ud2;"
 	);
 #elif defined(__i686__)
 	asm volatile(
-		"xorl %ebp, %ebp;"
+		"movl  (%esp), %edi;"
-		"pushl %ecx;"
+		"leal 4(%esp), %esi;"
+		"leal 4(%esi, %edi, 4), %edx;"
+
+		"movl %esp, %ebp;"
+		"andl $-16, %esp;"
+
+		"subl $4, %esp;"
 		"pushl %edx;"
 		"pushl %esi;"
 		"pushl %edi;"
-		"call _entry;"
-		"ud2;"
-	);
-#else
-	#error "unsupported architecture"
-#endif
-}
 
-__attribute__((naked, noreturn))
+		"call _entry;"
-static void call_entry_point(int, char**, char**, uintptr_t)
+
-{
+		"movl %ebp, %esp;"
-#if defined(__x86_64__)
+		"xorl %ebp, %ebp;"
-	asm volatile(
+
-		"andq $-16, %rsp;"
+		"jmp *%eax;"
-		"jmp *%rcx;"
+
-	);
+		"ud2;"
-#elif defined(__i686__)
-	asm volatile(
-		"addl $4, %esp;"
-		"popl %edi;"
-		"popl %esi;"
-		"popl %edx;"
-		"popl %ecx;"
-		"andl $-16, %esp;"
-		"jmp *%ecx;"
 	);
 #else
 	#error "unsupported architecture"
@@ -136,8 +146,15 @@ static void resolve_symbol_trampoline()
 struct LoadedElf
 {
 	ElfNativeFileHeader file_header;
+	ElfNativeProgramHeader tls_header;
 	ElfNativeDynamic* dynamics;
 
+	uint8_t* tls_addr;
+	size_t tls_module;
+	size_t tls_offset;
+
+	int fd;
+
 	uintptr_t base;
 
 	uintptr_t hash;
@@ -163,8 +180,8 @@ struct LoadedElf
 	uintptr_t init_array;
 	size_t init_arraysz;
 
-	bool has_called_init;
+	bool is_calling_init;
-	bool is_relocated;
+	bool is_relocating;
 
 	char path[PATH_MAX];
 };
@@ -207,20 +224,71 @@ static ElfNativeSymbol* find_symbol(const LoadedElf& elf, const char* name)
 }
 
 template<typename RelocT> requires BAN::is_same_v<RelocT, ElfNativeRelocation> || BAN::is_same_v<RelocT, ElfNativeRelocationA>
-static void handle_copy_relocation(const LoadedElf& elf, const RelocT& reloc)
+static bool is_tls_relocation(const RelocT& reloc)
 {
 #if defined(__x86_64__)
-	if (ELF64_R_TYPE(reloc.r_info) != R_X86_64_COPY)
+	switch (ELF64_R_TYPE(reloc.r_info))
-		return;
+	{
-	const uint32_t symbol_index = ELF64_R_SYM(reloc.r_info);
+		case R_X86_64_DTPMOD64:
+		case R_X86_64_DTPOFF64:
+		case R_X86_64_TPOFF64:
+		case R_X86_64_TLSGD:
+		case R_X86_64_TLSLD:
+		case R_X86_64_DTPOFF32:
+		case R_X86_64_GOTTPOFF:
+		case R_X86_64_TPOFF32:
+			return true;
+	}
 #elif defined(__i686__)
-	if (ELF32_R_TYPE(reloc.r_info) != R_386_COPY)
+	switch (ELF32_R_TYPE(reloc.r_info))
-		return;
+	{
-	const uint32_t symbol_index = ELF32_R_SYM(reloc.r_info);
+		case R_386_TLS_TPOFF:
+		case R_386_TLS_IE:
+		case R_386_TLS_GOTIE:
+		case R_386_TLS_LE:
+		case R_386_TLS_GD:
+		case R_386_TLS_LDM:
+		case R_386_TLS_GD_32:
+		case R_386_TLS_GD_PUSH:
+		case R_386_TLS_GD_CALL:
+		case R_386_TLS_GD_POP:
+		case R_386_TLS_LDM_32:
+		case R_386_TLS_LDM_PUSH:
+		case R_386_TLS_LDM_CALL:
+		case R_386_TLS_LDM_POP:
+		case R_386_TLS_LDO_32:
+		case R_386_TLS_IE_32:
+		case R_386_TLS_LE_32:
+		case R_386_TLS_DTPMOD32:
+		case R_386_TLS_DTPOFF32:
+		case R_386_TLS_TPOFF32:
+			return true;
+	}
 #else
 	#error "unsupported architecture"
 #endif
+	return false;
+}
 
+template<typename RelocT> requires BAN::is_same_v<RelocT, ElfNativeRelocation> || BAN::is_same_v<RelocT, ElfNativeRelocationA>
+static bool is_copy_relocation(const RelocT& reloc)
+{
+#if defined(__x86_64__)
+	return ELF64_R_TYPE(reloc.r_info) == R_X86_64_COPY;
+#elif defined(__i686__)
+	return ELF32_R_TYPE(reloc.r_info) == R_386_COPY;
+#else
+	#error "unsupported architecture"
+#endif
+}
+
+template<typename RelocT> requires BAN::is_same_v<RelocT, ElfNativeRelocation> || BAN::is_same_v<RelocT, ElfNativeRelocationA>
+static void handle_copy_relocation(const LoadedElf& elf, const RelocT& reloc)
+{
+	if (!is_copy_relocation(reloc))
+		return;
+
+	const uint32_t symbol_index = ELF_R_SYM(reloc.r_info);
 	if (symbol_index == 0)
 		print_error_and_exit("copy relocation without a symbol", 0);
 
@@ -259,25 +327,60 @@ static void handle_copy_relocation(const LoadedElf& elf, const RelocT& reloc)
 }
 
 template<typename RelocT> requires BAN::is_same_v<RelocT, ElfNativeRelocation> || BAN::is_same_v<RelocT, ElfNativeRelocationA>
-static uintptr_t handle_relocation(const LoadedElf& elf, const RelocT& reloc, bool resolve_symbols)
+static void handle_tls_relocation(const LoadedElf& elf, const RelocT& reloc)
 {
-	uintptr_t symbol_address = 0;
+	if (!is_tls_relocation(reloc))
+		return;
+
+	if (ELF_R_SYM(reloc.r_info))
+		print_error_and_exit("tls relocation with symbol index", 0);
+
+	if (elf.tls_addr == nullptr)
+		print_error_and_exit("tls relocation without tls", 0);
+
 
 #if defined(__x86_64__)
-	if (ELF64_R_TYPE(reloc.r_info) == R_X86_64_COPY)
+	switch (ELF64_R_TYPE(reloc.r_info))
-		return 0;
+	{
-	const uint32_t symbol_index = ELF64_R_SYM(reloc.r_info);
+		case R_X86_64_DTPMOD64:
+			*reinterpret_cast<uint64_t*>(elf.base + reloc.r_offset) = elf.tls_module;
+			break;
+		default:
+			print(STDERR_FILENO, "unsupported tls reloc type ");
+			print_uint(STDERR_FILENO, ELF64_R_TYPE(reloc.r_info));
+			print(STDERR_FILENO, " in ");
+			print(STDERR_FILENO, elf.path);
+			print_error_and_exit("", 0);
+	}
 #elif defined(__i686__)
-	if (ELF32_R_TYPE(reloc.r_info) == R_386_COPY)
+	switch (ELF32_R_TYPE(reloc.r_info))
-		return 0;
+	{
-	const uint32_t symbol_index = ELF32_R_SYM(reloc.r_info);
+		case R_386_TLS_DTPMOD32:
+			*reinterpret_cast<uint32_t*>(elf.base + reloc.r_offset) = elf.tls_module;
+			break;
+		default:
+			print(STDERR_FILENO, "unsupported tls reloc type ");
+			print_uint(STDERR_FILENO, ELF64_R_TYPE(reloc.r_info));
+			print(STDERR_FILENO, " in ");
+			print(STDERR_FILENO, elf.path);
+			print_error_and_exit("", 0);
+	}
 #else
 	#error "unsupported architecture"
 #endif
+}
 
+template<typename RelocT> requires BAN::is_same_v<RelocT, ElfNativeRelocation> || BAN::is_same_v<RelocT, ElfNativeRelocationA>
+static uintptr_t handle_relocation(const LoadedElf& elf, const RelocT& reloc, bool resolve_symbols)
+{
+	if (is_copy_relocation(reloc) || is_tls_relocation(reloc))
+		return 0;
+
+	const uint32_t symbol_index = ELF_R_SYM(reloc.r_info);
 	if (resolve_symbols == !symbol_index)
 		return 0;
 
+	uintptr_t symbol_address = 0;
 	if (symbol_index)
 	{
 		const auto& symbol = *reinterpret_cast<ElfNativeSymbol*>(elf.symtab + symbol_index * elf.syment);
@@ -311,6 +414,9 @@ static uintptr_t handle_relocation(const LoadedElf& elf, const RelocT& reloc, bo
 				symbol_address = 0;
 			}
 		}
+
+		if (ELF_ST_TYPE(symbol.st_info) == STT_TLS)
+			print_error_and_exit("relocating TLS symbol", 0);
 	}
 
 	size_t size = 0;
@@ -417,10 +523,9 @@ static uintptr_t handle_relocation(const LoadedElf& elf, const RelocT& reloc, bo
 
 static void relocate_elf(LoadedElf& elf, bool lazy_load)
 {
-	// FIXME: handle circular dependencies
+	if (elf.is_relocating)
-
-	if (elf.is_relocated)
 		return;
+	elf.is_relocating = true;
 
 	// do copy relocations
 	if (elf.rel && elf.relent)
@@ -449,6 +554,14 @@ static void relocate_elf(LoadedElf& elf, bool lazy_load)
 		for (size_t i = 0; i < elf.relasz / elf.relaent; i++)
 			handle_relocation(elf, *reinterpret_cast<ElfNativeRelocationA*>(elf.rela + i * elf.relaent), true);
 
+	// do tls relocations
+	if (elf.rel && elf.relent)
+		for (size_t i = 0; i < elf.relsz / elf.relent; i++)
+			handle_tls_relocation(elf, *reinterpret_cast<ElfNativeRelocation*>(elf.rel + i * elf.relent));
+	if (elf.rela && elf.relaent)
+		for (size_t i = 0; i < elf.relasz / elf.relaent; i++)
+			handle_tls_relocation(elf, *reinterpret_cast<ElfNativeRelocationA*>(elf.rela + i * elf.relaent));
+
 	// do jumprel relocations
 	if (elf.jmprel && elf.pltrelsz)
 	{
@@ -498,8 +611,6 @@ static void relocate_elf(LoadedElf& elf, bool lazy_load)
 			}
 		}
 	}
-
-	elf.is_relocated = true;
 }
 
 extern "C"
@@ -527,17 +638,19 @@ static void handle_dynamic(LoadedElf& elf)
 
 		switch (dynamic.d_tag)
 		{
-			case DT_PLTGOT:     dynamic.d_un.d_ptr += elf.base; break;
+			case DT_PLTGOT:
-			case DT_HASH:       dynamic.d_un.d_ptr += elf.base; break;
+			case DT_HASH:
-			case DT_STRTAB:     dynamic.d_un.d_ptr += elf.base; break;
+			case DT_STRTAB:
-			case DT_SYMTAB:     dynamic.d_un.d_ptr += elf.base; break;
+			case DT_SYMTAB:
-			case DT_RELA:       dynamic.d_un.d_ptr += elf.base; break;
+			case DT_RELA:
-			case DT_INIT:       dynamic.d_un.d_ptr += elf.base; break;
+			case DT_INIT:
-			case DT_FINI:       dynamic.d_un.d_ptr += elf.base; break;
+			case DT_FINI:
-			case DT_REL:        dynamic.d_un.d_ptr += elf.base; break;
+			case DT_REL:
-			case DT_JMPREL:     dynamic.d_un.d_ptr += elf.base; break;
+			case DT_JMPREL:
-			case DT_INIT_ARRAY: dynamic.d_un.d_ptr += elf.base; break;
+			case DT_INIT_ARRAY:
-			case DT_FINI_ARRAY: dynamic.d_un.d_ptr += elf.base; break;
+			case DT_FINI_ARRAY:
+				dynamic.d_un.d_ptr += elf.base;
+				break;
 		}
 
 		switch (dynamic.d_tag)
@@ -593,8 +706,6 @@ static void handle_dynamic(LoadedElf& elf)
 
 		const auto& loaded_elf = load_elf(realpath, library_fd);
 		dynamic.d_un.d_ptr = reinterpret_cast<uintptr_t>(&loaded_elf);
-
-		syscall(SYS_CLOSE, library_fd);
 	}
 
 	// do relocations without symbols
@@ -798,6 +909,9 @@ static LoadedElf& load_elf(const char* path, int fd)
 		break;
 	}
 
+	ElfNativeProgramHeader tls_header {};
+	tls_header.p_type = PT_NULL;
+
 	for (size_t i = 0; i < file_header.e_phnum; i++)
 	{
 		ElfNativeProgramHeader program_header;
@@ -817,6 +931,9 @@ static LoadedElf& load_elf(const char* path, int fd)
 			case PT_GNU_RELRO:
 				print(STDDBG_FILENO, "TODO: PT_GNU_*\n");
 				break;
+			case PT_TLS:
+				tls_header = program_header;
+				break;
 			case PT_LOAD:
 				program_header.p_vaddr += base;
 				load_program_header(program_header, fd, needs_writable);
@@ -829,7 +946,9 @@ static LoadedElf& load_elf(const char* path, int fd)
 	}
 
 	auto& elf = s_loaded_files[s_loaded_file_count++];
+	elf.tls_header = tls_header;
 	elf.base = base;
+	elf.fd = fd;
 	elf.dynamics = nullptr;
 	memcpy(&elf.file_header, &file_header, sizeof(file_header));
 	strcpy(elf.path, path);
@@ -857,20 +976,165 @@ static LoadedElf& load_elf(const char* path, int fd)
 	return elf;
 }
 
-static void call_init_libc(LoadedElf& elf, char** envp)
+struct MasterTLS
 {
-	const auto* _init_libc = find_symbol(elf, "_init_libc");
+	uint8_t* addr;
-	if (_init_libc == nullptr)
+	size_t size;
-		return;
+	size_t module_count;
-	using _init_libc_t = void(*)(char**);
+};
-	reinterpret_cast<_init_libc_t>(elf.base + _init_libc->st_value)(envp);
+
+static MasterTLS initialize_master_tls()
+{
+	constexpr auto round =
+		[](size_t a, size_t b) -> size_t
+		{
+			return b * ((a + b - 1) / b);
+		};
+
+	size_t max_align = alignof(uthread);
+	size_t tls_m_offset = 0;
+	size_t tls_m_size = 0;
+	size_t module_count = 0;
+	for (size_t i = 0; i < s_loaded_file_count; i++)
+	{
+		const auto& tls_header = s_loaded_files[i].tls_header;
+		if (tls_header.p_type != PT_TLS)
+			continue;
+		if (tls_header.p_align == 0)
+			print_error_and_exit("TLS alignment is 0", 0);
+
+		max_align = max<size_t>(max_align, tls_header.p_align);
+		tls_m_offset = round(tls_m_offset + tls_header.p_memsz, tls_header.p_align);
+		tls_m_size = tls_header.p_memsz;
+
+		module_count++;
+	}
+
+	if (module_count == 0)
+		return { .addr = nullptr, .size = 0, .module_count = 0 };
+
+	size_t master_tls_size = tls_m_offset + tls_m_size;
+	if (auto rem = master_tls_size % max_align)
+		master_tls_size += max_align - rem;
+
+	uint8_t* master_tls_addr;
+
+	{
+		const sys_mmap_t mmap_args {
+			.addr = nullptr,
+			.len = master_tls_size,
+			.prot = PROT_READ | PROT_WRITE,
+			.flags = MAP_ANONYMOUS | MAP_PRIVATE,
+			.fildes = -1,
+			.off = 0,
+		};
+
+		const auto ret = syscall(SYS_MMAP, &mmap_args);
+		if (ret < 0)
+			print_error_and_exit("failed to allocate master TLS", ret);
+		master_tls_addr = reinterpret_cast<uint8_t*>(ret);
+	}
+
+	for (size_t i = 0, tls_offset = 0, tls_module = 1; i < s_loaded_file_count; i++)
+	{
+		const auto& tls_header = s_loaded_files[i].tls_header;
+		if (tls_header.p_type != PT_TLS)
+			continue;
+
+		tls_offset = round(tls_offset + tls_header.p_memsz, tls_header.p_align);
+
+		uint8_t* tls_buffer = master_tls_addr + master_tls_size - tls_offset;
+
+		if (tls_header.p_filesz > 0)
+		{
+			const int fd = s_loaded_files[i].fd;
+			if (auto ret = syscall(SYS_PREAD, fd, tls_buffer, tls_header.p_filesz, tls_header.p_offset); ret != static_cast<long>(tls_header.p_filesz))
+				print_error_and_exit("failed to read TLS data", ret);
+		}
+
+		memset(tls_buffer + tls_header.p_filesz, 0, tls_header.p_memsz - tls_header.p_filesz);
+
+		auto& elf = s_loaded_files[i];
+		elf.tls_addr = tls_buffer;
+		elf.tls_module = tls_module++;
+		elf.tls_offset = master_tls_size - tls_offset;
+	}
+
+	return { .addr = master_tls_addr, .size = master_tls_size, .module_count = module_count };
 }
 
-static void call_init_funcs(LoadedElf& elf, char** envp, bool skip)
+static void initialize_tls(MasterTLS master_tls)
 {
-	if (elf.has_called_init)
+	if (master_tls.addr == nullptr)
 		return;
 
+	const size_t tls_size = master_tls.size
+		+ sizeof(uthread)
+		+ (master_tls.module_count + 1) * sizeof(uintptr_t);
+
+	uint8_t* tls_addr;
+
+	{
+		const sys_mmap_t mmap_args {
+			.addr = nullptr,
+			.len = tls_size,
+			.prot = PROT_READ | PROT_WRITE,
+			.flags = MAP_ANONYMOUS | MAP_PRIVATE,
+			.fildes = -1,
+			.off = 0,
+		};
+
+		const auto ret = syscall(SYS_MMAP, &mmap_args);
+		if (ret < 0)
+			print_error_and_exit("failed to allocate master TLS", ret);
+		tls_addr = reinterpret_cast<uint8_t*>(ret);
+	}
+
+	memcpy(tls_addr, master_tls.addr, master_tls.size);
+
+	uthread* uthread = reinterpret_cast<struct uthread*>(tls_addr + master_tls.size);
+	uthread->self = uthread;
+	uthread->master_tls_addr = master_tls.addr;
+	uthread->master_tls_size = master_tls.size;
+
+	uthread->dtv[0] = master_tls.module_count;
+	for (size_t i = 0; i < s_loaded_file_count; i++)
+	{
+		const auto& elf = s_loaded_files[i];
+		if (elf.tls_addr == nullptr)
+			continue;
+		uthread->dtv[elf.tls_module] = reinterpret_cast<uintptr_t>(tls_addr) + elf.tls_offset;
+	}
+
+	syscall(SYS_SET_TLS, uthread);
+}
+
+static void initialize_environ(char** envp)
+{
+	uintptr_t environ = SYM_NOT_FOUND;
+	for (size_t i = 0; i < s_loaded_file_count; i++)
+	{
+		const auto* match = find_symbol(s_loaded_files[i], "environ");
+		if (match == nullptr)
+			continue;
+		if (environ == SYM_NOT_FOUND || ELF_ST_BIND(match->st_info) != STB_WEAK)
+			environ = s_loaded_files[i].base + match->st_value;
+		if (ELF_ST_BIND(match->st_info) != STB_WEAK)
+			break;
+	}
+
+	if (environ == SYM_NOT_FOUND)
+		return;
+
+	*reinterpret_cast<char***>(environ) = envp;
+}
+
+static void call_init_funcs(LoadedElf& elf, bool is_main_elf)
+{
+	if (elf.is_calling_init)
+		return;
+	elf.is_calling_init = true;
+
 	if (elf.dynamics)
 	{
 		for (size_t i = 0;; i++)
@@ -879,11 +1143,12 @@ static void call_init_funcs(LoadedElf& elf, char** envp, bool skip)
 			if (dynamic.d_tag == DT_NULL)
 				break;
 			if (dynamic.d_tag == DT_NEEDED)
-				call_init_funcs(*reinterpret_cast<LoadedElf*>(dynamic.d_un.d_ptr), envp, false);
+				call_init_funcs(*reinterpret_cast<LoadedElf*>(dynamic.d_un.d_ptr), false);
 		}
 	}
 
-	if (elf.has_called_init || skip)
+	// main executable calls its init functions in _start
+	if (is_main_elf)
 		return;
 
 	using init_t = void(*)();
@@ -891,19 +1156,33 @@ static void call_init_funcs(LoadedElf& elf, char** envp, bool skip)
 		reinterpret_cast<init_t>(elf.init)();
 	for (size_t i = 0; i < elf.init_arraysz / sizeof(init_t); i++)
 		reinterpret_cast<init_t*>(elf.init_array)[i]();
+}
 
-	if (strcmp(elf.path, "/usr/lib/libc.so") == 0)
+static LibELF::AuxiliaryVector* find_auxv(char** envp)
-		call_init_libc(elf, envp);
+{
+	if (envp == nullptr)
+		return nullptr;
 
-	elf.has_called_init = true;
+	char** null_env = envp;
+	while (*null_env)
+		null_env++;
+
+	return reinterpret_cast<LibELF::AuxiliaryVector*>(null_env + 1);
 }
 
 extern "C"
-__attribute__((used, noreturn))
+__attribute__((used))
-int _entry(int argc, char** argv, char** envp, int fd)
+uintptr_t _entry(int argc, char* argv[], char* envp[])
 {
-	const bool invoked_directly = (fd < 0);
+	int execfd = -1;
-	if (invoked_directly)
+	if (auto* auxv = find_auxv(envp))
+		for (auto* aux = auxv; aux->a_type != LibELF::AT_NULL; aux++)
+			if (aux->a_type == LibELF::AT_EXECFD) {
+				execfd = aux->a_un.a_val;
+				aux->a_type = LibELF::AT_IGNORE;
+			}
+
+	if (execfd == -1)
 	{
 		if (argc < 2)
 			print_error_and_exit("missing program name", 0);
@@ -911,17 +1190,23 @@ int _entry(int argc, char** argv, char** envp, int fd)
 		argc--;
 		argv++;
 
-		fd = syscall(SYS_OPENAT, AT_FDCWD, argv[0], O_RDONLY);
+		execfd = syscall(SYS_OPENAT, AT_FDCWD, argv[0], O_RDONLY);
-		if (fd < 0)
+		if (execfd < 0)
-			print_error_and_exit("could not open program", fd);
+			print_error_and_exit("could not open program", execfd);
 	}
 
 	init_random();
-	auto& elf = load_elf(argv[0], fd);
+	auto& elf = load_elf(argv[0], execfd);
-	syscall(SYS_CLOSE, fd);
 	fini_random();
 
+	const auto master_tls = initialize_master_tls();
 	relocate_elf(elf, true);
-	call_init_funcs(elf, envp, true);
+	initialize_tls(master_tls);
-	call_entry_point(argc, argv, envp, elf.base + elf.file_header.e_entry);
+	initialize_environ(envp);
+	call_init_funcs(elf, true);
+
+	for (size_t i = 0; i < s_loaded_file_count; i++)
+		syscall(SYS_CLOSE, s_loaded_files[i].fd);
+
+	return elf.base + elf.file_header.e_entry;
 }