kernel: add linux compatibility layer

Makefile

@@ -161,6 +161,11 @@ ifeq ($(CONFIG_USE_FAKE_CLOCK), 1)
 	libk_c_files  += $(kernel_src_dir)/time/fake_clock.c
 endif
 
+ifeq ($(CONFIG_LINUX_COMPAT), 1)
+	KERNEL_CONFIG += -DCONFIG_LINUX_COMPAT
+	LIBC_CONFIG   += -DCONFIG_LINUX_COMPAT
+endif
+
 # This file exists in a Docker container because we copy it in `Dockerfile`.
 in_docker = $(wildcard /tmp/install-linux-deps)
 ifneq ($(in_docker),)
@@ -435,6 +440,7 @@ what: ## display some information about the current configuration
 	echo "OS_NAME : $(OS_NAME)"
 	echo "ARCH    : $(ARCH)"
 	echo ""
+	echo "CONFIG_LINUX_COMPAT   = $(CONFIG_LINUX_COMPAT)"
 	echo "CONFIG_SEMIHOSTING    = $(CONFIG_SEMIHOSTING)"
 	echo "CONFIG_USE_DLMALLOC   = $(CONFIG_USE_DLMALLOC)"
 	echo "CONFIG_USE_FAKE_CLOCK = $(CONFIG_USE_FAKE_CLOCK)"

Makefile-cfg.include

@@ -6,6 +6,8 @@ LLVM_SUFFIX ?=
 
 # When set to 1, enable the Undefined Behavior SANitizer.
 UBSAN                 ?=
+# When set to 1, enable Linux (binary) compatibility.
+CONFIG_LINUX_COMPAT   ?=
 # When set to 1, enable semi-hosting mode (QEMU, mainly).
 CONFIG_SEMIHOSTING    ?=
 # When set to 1, use dlmalloc for malloc/free/realloc (instead of liballoc).

data/initrd/etc/hosts

@@ -0,0 +1 @@
+140.82.121.4 github

data/initrd/etc/passwd

@@ -0,0 +1 @@
+root:x:0:0::/:none

include/kernel/proc/descriptor.h

@@ -2,9 +2,11 @@
 #ifndef PROC_DESCRIPTOR_H
 #define PROC_DESCRIPTOR_H
 
+#include <arpa/inet.h>
 #include <fs/vfs.h>
 #include <stdbool.h>
 #include <stdint.h>
+#include <sys/socket.h>
 #include <sys/types.h>
 
 #define STDIN  0
@@ -23,6 +25,8 @@ typedef struct descriptor
   uint32_t type;
   uint32_t protocol;
   uint16_t port;
+  struct sockaddr_in addr;
+  socklen_t addr_len;
 } descriptor_t;
 
 /**
@@ -82,4 +86,8 @@ int descriptor_udp_lookup(uint16_t port);
  */
 bool is_protocol_supported(uint32_t type, uint32_t protocol);
 
+void duplicate_descriptor(int oldfd, int newfd);
+
+int descriptor_raw_lookup(uint32_t protocol, in_addr_t src_addr);
+
 #endif

include/libc/sys/linux_compat.h

@@ -0,0 +1,112 @@
+// https://chromium.googlesource.com/chromiumos/docs/+/HEAD/constants/syscalls.md
+#ifndef SYS_LINUX_COMPAT_H
+#define SYS_LINUX_COMPAT_H
+
+#ifdef __x86_64__
+
+#define SYSCALL_READ           0
+#define SYSCALL_WRITE          1
+#define SYSCALL_OPEN           2
+#define SYSCALL_CLOSE          3
+#define SYSCALL_FSTAT          5
+#define SYSCALL_POLL           7
+#define SYSCALL_LSEEK          8
+#define SYSCALL_MMAP           9
+#define SYSCALL_BRK            12
+#define SYSCALL_RT_SIGACTION   13
+#define SYSCALL_RT_SIGPROCMASK 14
+#define SYSCALL_IOCTL          16
+#define SYSCALL_WRITEV         20
+#define SYSCALL_DUP2           33
+#define SYSCALL_NANOSLEEP      35
+#define SYSCALL_GETPID         39
+#define SYSCALL_SOCKET         41
+#define SYSCALL_CONNECT        42
+#define SYSCALL_SENDTO         44
+#define SYSCALL_RECVFROM       45
+#define SYSCALL_BIND           49
+#define SYSCALL_GETSOCKNAME    51
+#define SYSCALL_SETSOCKOPT     54
+#define SYSCALL_EXECV          59
+#define SYSCALL_EXIT           60
+#define SYSCALL_UNAME          63
+#define SYSCALL_FCNTL          72
+#define SYSCALL_GETCWD         79
+#define SYSCALL_GETTIMEOFDAY   96
+#define SYSCALL_SETUID         105
+#define SYSCALL_SETGID         106
+#define SYSCALL_GETEUID        107
+#define SYSCALL_ARCH_PRCTL     158
+#define SYSCALL_REBOOT         169
+#define SYSCALL_SET_TID_ADDR   218
+#define SYSCALL_EXIT_GROUP     231
+#define SYSCALL_OPENAT         257
+
+#elif __arm__
+
+#define SYSCALL_EXIT         1
+#define SYSCALL_READ         3
+#define SYSCALL_WRITE        4
+#define SYSCALL_OPEN         5
+#define SYSCALL_CLOSE        6
+#define SYSCALL_EXECV        11
+#define SYSCALL_LSEEK        19
+#define SYSCALL_GETPID       20
+#define SYSCALL_BRK          45
+#define SYSCALL_GETEUID      49
+#define SYSCALL_IOCTL        54
+#define SYSCALL_DUP2         63
+#define SYSCALL_GETTIMEOFDAY 78
+#define SYSCALL_REBOOT       88
+#define SYSCALL_FSTAT        108
+#define SYSCALL_WRITEV       146
+#define SYSCALL_EXIT_GROUP   248
+#define SYSCALL_SET_TID_ADDR 256
+#define SYSCALL_SOCKET       281
+#define SYSCALL_SENDTO       290
+#define SYSCALL_RECVFROM     292
+#define SYSCALL_OPENAT       322
+
+// Not available on AArch32:
+//
+// - SYSCALL_ARCH_PRCTL
+
+#elif __aarch64__
+
+#define SYSCALL_FCNTL        25
+#define SYSCALL_IOCTL        29
+#define SYSCALL_OPENAT       56
+#define SYSCALL_CLOSE        57
+#define SYSCALL_LSEEK        62
+#define SYSCALL_READ         63
+#define SYSCALL_WRITE        64
+#define SYSCALL_WRITEV       66
+#define SYSCALL_FSTAT        80
+#define SYSCALL_EXIT         93
+#define SYSCALL_EXIT_GROUP   94
+#define SYSCALL_SET_TID_ADDR 96
+#define SYSCALL_REBOOT       142
+#define SYSCALL_GETTIMEOFDAY 169
+#define SYSCALL_GETPID       172
+#define SYSCALL_GETEUID      175
+#define SYSCALL_SOCKET       198
+#define SYSCALL_SENDTO       206
+#define SYSCALL_RECVFROM     207
+#define SYSCALL_BRK          214
+#define SYSCALL_MUNMAP       215
+#define SYSCALL_EXECV        221
+#define SYSCALL_MMAP         222
+
+// Not available on AArch64:
+//
+// - SYSCALL_OPEN
+// - SYSCALL_DUP2
+// - SYSCALL_ARCH_PRCTL
+
+#endif
+
+// Not available outside ArvernOS:
+#define SYSCALL_TEST           348
+#define SYSCALL_GETHOSTBYNAME2 349
+
+#endif

include/libc/sys/syscall.h

@@ -11,6 +11,12 @@
 #include <sys/time.h>
 #include <sys/types.h>
 
+#ifdef CONFIG_LINUX_COMPAT
+
+#include <sys/linux_compat.h>
+
+#else // CONFIG_LINUX_COMPAT
+
 #define SYSCALL_TEST           1
 #define SYSCALL_WRITE          2
 #define SYSCALL_READ           3
@@ -29,6 +35,8 @@
 #define SYSCALL_EXIT           16
 #define SYSCALL_OPENAT         17
 
+#endif // CONFIG_LINUX_COMPAT
+
 #define SYSCALL_SET_ERRNO()                                                    \
   if (retval < 0) {                                                            \
     errno = -retval;                                                           \

src/kernel/arch/aarch32/board/raspi2/asm/proc.asm

@@ -22,3 +22,10 @@ arch_task_switch:
 ret_from_fork:
   bl task_schedule_tail
   blx r4
+
+#ifdef CONFIG_LINUX_COMPAT
+.global linux_compat_start
+linux_compat_start:
+  mov sp, r1
+  blx r0
+#endif

src/kernel/arch/aarch64/board/raspi3/asm/proc.asm

@@ -32,3 +32,10 @@ arch_task_switch:
   ldr x30, [x8]
   mov sp, x9
   ret
+
+#ifdef CONFIG_LINUX_COMPAT
+.global linux_compat_start
+linux_compat_start:
+  mov sp, x1
+  blr x0
+#endif

src/kernel/arch/x86_64/Makefile.include

@@ -27,7 +27,7 @@ QEMU = qemu-system-x86_64
 # Options for the different tools
 ###############################################################################
 
-QEMU_OPTIONS += -m 512M
+QEMU_OPTIONS += -m 512M -cpu IvyBridge
 QEMU_OPTIONS += -serial file:$(log_file)
 QEMU_OPTIONS += -netdev user,id=u1,ipv6=off,dhcpstart=10.0.2.20
 QEMU_OPTIONS += -device rtl8139,netdev=u1

src/kernel/arch/x86_64/asm/boot.asm

@@ -28,11 +28,37 @@ start:
   ; load the 64-bit GDT
   lgdt [gdt64.pointer]
 
+%ifdef CONFIG_LINUX_COMPAT
+  call enable_sse
+
+  ; enable the following instructions: rdfsbase, rdgsbase, wrfsbase, wrgsbase
+  mov eax, cr4
+  or eax, 1 << 16
+  mov cr4, eax
+%endif
+
   jmp gdt64.kernel_code:long_mode_start
 
-  ; Should not be reached.
+  ; should not be reached
   hlt
 
+enable_sse:
+  mov eax, 0x1        ; check for SSE
+  cpuid
+  test edx, 1 << 25
+  jz .no_sse          ; after this, SSE can be enabled
+  mov eax, cr0
+  and ax, 0xFFFB      ; clear coprocessor emulation CR0.EM
+  or ax, 0x2          ; set coprocessor monitoring  CR0.MP
+  mov cr0, eax
+  mov eax, cr4
+  or ax, 3 << 9       ; set CR4.OSFXSR and CR4.OSXMMEXCPT at the same time
+  mov cr4, eax
+  ret
+
+.no_sse:
+  ret
+
 ; -----------------------------------------------------------------------------
 ; make sure the kernel was really loaded by a Multiboot compliant bootloader
 %define MULTIBOOT2_MAGIC_VALUE 0x36d76289

src/kernel/arch/x86_64/asm/k_syscall.asm

@@ -1,42 +1,78 @@
 global syscall_handler
 
+%define RSP0  0
+%define CS    8
+%define SS   16
+
 syscall_handler:
   sti
 
-  push rcx     ; save the return address
-  push r11     ; save the flags
+%ifdef CONFIG_LINUX_COMPAT
+  ; Save the current rsp,cs and ss to be used in the iret frame at the end.
+  mov gs:[RSP0], rsp
+  mov gs:[CS], cs
+  mov gs:[SS], ss
+%endif
+
+  ; Save the return address, flags, base pointer for sysret.
+  push rcx
+  push r11
   push rbp
+
+  ; Save callee-saved registers.
   push rbx
-  push rdx
-  push rdi
-  push rsi
-  push r8
-  push r9
-  push r10
   push r12
   push r13
   push r14
   push r15
 
+  ; Save the syscall params.
+  push rdi
+  push rsi
+  push rdx
+  push r10
+  push r8
+  push r9
+
   mov rcx, r10 ; fix 3rd syscall arg
 
   extern syscall_handlers
   call [rax * 8 + syscall_handlers]
 
-  pop r15
-  pop r14
-  pop r13
-  pop r12
-  pop r10
   pop r9
   pop r8
+  pop r10
+  pop rdx
   pop rsi
   pop rdi
-  pop rdx
+
+  pop r15
+  pop r14
+  pop r13
+  pop r12
   pop rbx
+
   pop rbp
   pop r11
   pop rcx
 
+; This is needed because usermode has been broken recently (when the multi-
+; tasking code has been introduced). Therefore, we only have kernel mode (ring
+; 0). `sysret` always returns to ring 3 so we cannot use it when a syscall is
+; done from ring 0 (which happens with Linux unmodified binaries). That's why
+; we use `iretq`.
+%ifdef CONFIG_LINUX_COMPAT
+  push qword gs:[SS]
+  push qword gs:[RSP0]
+  push r11
+  push qword gs:[CS]
+  push rcx
+
+  cli
+
+  iretq
+%else
   cli
+
   o64 sysret
+%endif

src/kernel/arch/x86_64/asm/proc.asm

@@ -30,3 +30,11 @@ ret_from_fork:
   call task_schedule_tail
 
   call r12
+
+%ifdef CONFIG_LINUX_COMPAT
+global linux_compat_start
+
+linux_compat_start:
+  mov rsp, rsi
+  jmp rdi
+%endif