vdso

include/common/utils.h

@@ -14,19 +14,19 @@
 #if defined(__has_builtin)
 #if __has_builtin(__builtin_types_compatible_p)
 // Get pointer to parent object given pointer to a struct or union member.
-#define CONTAINER_OF(ptr, type, member) \
+#define containerof(ptr, type, member) \
     ({ \
-        _Static_assert( \
+        static_assert( \
-            __builtin_types_compatible_p(__typeof__(*(ptr)), __typeof__(((type*)0)->member)), \
+            __builtin_types_compatible_p(typeof(*(ptr)), typeof(((type*)0)->member)), \
-            "Incompatible types for CONTAINER_OF" \
+            "Incompatible types for containerof" \
         ); \
         (type*)((size_t)ptr - offsetof(type, member)); \
     })
 #endif
 #endif
-#ifndef CONTAINER_OF
+#ifndef containerof
 // Get pointer to parent object given pointer to a struct or union member.
-#define CONTAINER_OF(ptr, type, member) (type*)((size_t)ptr - offsetof(type, member))
+#define containerof(ptr, type, member) (type*)((size_t)ptr - offsetof(type, member))
 #endif
 
 #define CONCAT(A, B)  _CONCAT(A, B)

include/zinnia/archctl.h

@@ -2,7 +2,6 @@
 #define ZINNIA_ARCHCTL_H
 
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -11,16 +10,14 @@ extern "C" {
 typedef enum {
     // Does nothing.
     ZN_ARCHCTL_NONE = 0,
-    // On x86_64, sets the FSBASE MSR to the value.
+    // On x86_64, sets the FSBASE MSR.
     ZN_ARCHCTL_SET_FSBASE = 1,
-    // On x86_64, sets the GSBASE MSR to the value.
+    // On x86_64, sets the GSBASE MSR.
     ZN_ARCHCTL_SET_GSBASE = 2,
 } zn_archctl_t;
 
 // Performs an architecture-dependent operation identified by `op`.
-static inline zn_status_t zn_archctl(zn_archctl_t op, void* value) {
+zn_status_t zn_archctl(zn_archctl_t op, void* value);
-    return zn_syscall2(op, value, ZN_SYSCALL_ARCHCTL);
-}
 
 #ifdef __cplusplus
 }

include/zinnia/channel.h

@@ -3,7 +3,6 @@
 
 #include <zinnia/handle.h>
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
 #include <stddef.h>
 
 #ifdef __cplusplus
@@ -15,28 +14,20 @@ enum zn_channel_flags {
 };
 
 // Creates a new channel.
-static inline zn_status_t zn_channel_create(
+zn_status_t zn_channel_create(enum zn_channel_flags flags, zn_handle_t* endpoint0, zn_handle_t* endpoint1);
-    enum zn_channel_flags flags,
-    zn_handle_t* endpoint0,
-    zn_handle_t* endpoint1
-) {
-    return zn_syscall3(flags, endpoint0, endpoint1, ZN_SYSCALL_CHANNEL_CREATE);
-}
 
 // Notifies the peer that there is new data available.
 // The kernel only copies the first `num_handles` handles and `num_bytes` bytes.
-static inline zn_status_t zn_channel_write(
+zn_status_t zn_channel_write(
     zn_handle_t channel,
     const zn_handle_t* handles,
     const void* bytes,
     size_t num_handles,
     size_t num_bytes
-) {
+);
-    return zn_syscall5(channel, handles, bytes, num_handles, num_bytes, ZN_SYSCALL_CHANNEL_WRITE);
-}
 
 // Waits for a new message to appear in the channel.
-static inline zn_status_t zn_channel_read(
+zn_status_t zn_channel_read(
     zn_handle_t channel,
     zn_handle_t* handles,
     void* bytes,
@@ -44,18 +35,7 @@ static inline zn_status_t zn_channel_read(
     size_t num_bytes,
     size_t* read_handles,
     size_t* read_bytes
-) {
+);
-    return zn_syscall7(
-        channel,
-        handles,
-        bytes,
-        num_handles,
-        num_bytes,
-        read_handles,
-        read_bytes,
-        ZN_SYSCALL_CHANNEL_READ
-    );
-}
 
 #ifdef __cplusplus
 }

include/zinnia/futex.h

@@ -2,7 +2,6 @@
 #define ZINNIA_FUTEX_H
 
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
 #include <stdint.h>
 
 #ifdef __cplusplus
@@ -11,14 +10,10 @@ extern "C" {
 
 // Blocks the calling thread if *addr == expected.
 // Returns immediately if the value at addr has changed.
-static inline zn_status_t zn_futex_wait(int32_t* addr, int32_t expected) {
+zn_status_t zn_futex_wait(int32_t* addr, int32_t expected);
-    return zn_syscall2(addr, expected, ZN_SYSCALL_FUTEX_WAIT);
-}
 
 // Wakes up to `count` threads waiting on the given futex address.
-static inline zn_status_t zn_futex_wake(int32_t* addr, int32_t count) {
+zn_status_t zn_futex_wake(int32_t* addr, int32_t count);
-    return zn_syscall2(addr, count, ZN_SYSCALL_FUTEX_WAKE);
-}
 
 #ifdef __cplusplus
 }

include/zinnia/handle.h

@@ -3,7 +3,6 @@
 
 #include <zinnia/rights.h>
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
 #include <stdint.h>
 
 #ifdef __cplusplus
@@ -20,22 +19,16 @@ typedef int32_t zn_handle_t;
 #define ZN_HANDLE_ZERO_VMO    ((zn_handle_t)(-4))
 
 // Checks an object handle for validity.
-static inline zn_status_t zn_handle_validate(zn_handle_t handle) {
+zn_status_t zn_handle_validate(zn_handle_t handle);
-    return zn_syscall1(handle, ZN_SYSCALL_HANDLE_VALIDATE);
-}
 
 // Drops an object handle. All further references using this handle are invalid.
 // The numerical value may become valid again, but it is an error to keep using it.
 // The underlying object gets freed only once no other handle references it.
-static inline zn_status_t zn_handle_drop(zn_handle_t handle) {
+zn_status_t zn_handle_drop(zn_handle_t handle);
-    return zn_syscall1(handle, ZN_SYSCALL_HANDLE_DROP);
-}
 
 // Creates a new handle pointing to the same object as pointed to
 // by the original handle. The new handle may obtain new rights.
-static inline zn_status_t zn_handle_clone(zn_handle_t handle, zn_rights_t cloned_rights, zn_handle_t* cloned) {
+zn_status_t zn_handle_clone(zn_handle_t handle, zn_rights_t cloned_rights, zn_handle_t* cloned);
-    return zn_syscall3(handle, cloned_rights, cloned, ZN_SYSCALL_HANDLE_CLONE);
-}
 
 #ifdef __cplusplus
 }

include/zinnia/ipc.h

@@ -1,26 +0,0 @@
-#ifndef ZINNIA_IPC_H
-#define ZINNIA_IPC_H
-
-#include <zinnia/handle.h>
-#include <zinnia/rights.h>
-#include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-// Defines an IPC window.
-struct zn_ipcw {
-    zn_handle_t handles[16];
-};
-
-static inline zn_status_t zn_ipcw_init(struct zn_ipcw** ipcw) {
-    return ZN_ERR_UNSUPPORTED;
-}
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif

include/zinnia/mem.h

@@ -3,7 +3,6 @@
 
 #include <zinnia/handle.h>
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
 #include <stddef.h>
 #include <stdint.h>
 
@@ -19,7 +18,7 @@ enum zn_vm_flags {
     ZN_VM_MAP_WRITE = 1 << 1,
     // Mapping is executable.
     ZN_VM_MAP_EXEC = 1 << 2,
-    // Shared memory (not cloned on calls to fork).
+    // Shared memory.
     ZN_VM_MAP_SHARED = 1 << 3,
 
     // Map at a fixed address, replacing whatever is in the way.
@@ -31,31 +30,23 @@ enum zn_vm_flags {
 };
 
 // Creates a new virtual memory object.
-static inline zn_status_t zn_vmo_create(size_t length, zn_handle_t* out) {
+zn_status_t zn_vmo_create(size_t length, zn_handle_t* out);
-    return zn_syscall2(length, out, ZN_SYSCALL_VMO_CREATE);
-}
 
 // Creates a new virtual memory object which points to a contiguous phyiscal memory region.
-static inline zn_status_t zn_vmo_create_phys(uintptr_t phys_addr, size_t length, zn_handle_t* out) {
+zn_status_t zn_vmo_create_phys(uintptr_t phys_addr, size_t length, zn_handle_t* out);
-    return zn_syscall3(phys_addr, length, out, ZN_SYSCALL_VMO_CREATE_PHYS);
-}
 
 // Maps `bytes` of `vmo` at `vmo_offset` to `addr` in `vas`.
-static inline zn_status_t zn_vmo_map(
+zn_status_t zn_vmo_map(
     zn_handle_t vmo,
     zn_handle_t vas,
     uintptr_t vmo_offset,
     uintptr_t* addr,
     size_t bytes,
     enum zn_vm_flags flags
-) {
+);
-    return zn_syscall6(vmo, vas, vmo_offset, addr, bytes, flags, ZN_SYSCALL_VMO_MAP);
-}
 
 // Creates a new virtual address space.
-static inline zn_status_t zn_vas_create(zn_handle_t* out) {
+zn_status_t zn_vas_create(zn_handle_t* out);
-    return zn_syscall1(out, ZN_SYSCALL_VAS_CREATE);
-}
 
 #ifdef __cplusplus
 }

include/zinnia/namespace.h

@@ -9,15 +9,11 @@ extern "C" {
 #endif
 
 // Creates a new namespace.
-static inline zn_status_t zn_ns_create(zn_handle_t* out) {
+zn_status_t zn_ns_create(zn_handle_t* out);
-    return ZN_ERR_UNSUPPORTED;
-}
 
 // Moves a handle to another namespace.
 // The handle must have the `ZN_RIGHT_MOVE` right.
-static inline zn_status_t zn_ns_move(zn_handle_t handle, zn_handle_t target_ns) {
+zn_status_t zn_ns_move(zn_handle_t handle, zn_handle_t target_ns);
-    return ZN_ERR_UNSUPPORTED;
-}
 
 #ifdef __cplusplus
 }

include/zinnia/random.h

@@ -2,7 +2,6 @@
 #define ZINNIA_RANDOM_H
 
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
 #include <stddef.h>
 
 #ifdef __cplusplus
@@ -14,14 +13,10 @@ enum {
 };
 
 // Stores random bytes in a buffer.
-static inline zn_status_t zn_random_get(void* addr, size_t len) {
+zn_status_t zn_random_get(void* addr, size_t len);
-    return zn_syscall2(addr, len, ZN_SYSCALL_RANDOM_GET);
-}
 
 // Adds entropy to the RNG.
-static inline zn_status_t zn_random_entropy(const void* addr, size_t len) {
+zn_status_t zn_random_entropy(const void* addr, size_t len);
-    return zn_syscall2(addr, len, ZN_SYSCALL_RANDOM_ENTROPY);
-}
 
 #ifdef __cplusplus
 }

include/zinnia/status.h

@@ -34,36 +34,7 @@ typedef enum {
     ZN_ERR_ALREADY_EXISTS = -11,
 } zn_status_t;
 
-static inline const char* zn_status_to_string(zn_status_t err) {
+const char* zn_status_to_string(zn_status_t err);
-    switch (err) {
-    case ZN_OK:
-        return "No error (OK)";
-    case ZN_ERR_INTERNAL:
-        return "An internal error occured (INTERNAL)";
-    case ZN_ERR_BAD_SYSCALL:
-        return "Syscall number is not a recognized syscall (BAD_SYSCALL)";
-    case ZN_ERR_UNSUPPORTED:
-        return "This operation is not supported or implemented yet (UNSUPPORTED)";
-    case ZN_ERR_NO_MEMORY:
-        return "System does not have enough free memory for this operation (NO_MEMORY)";
-    case ZN_ERR_NO_HANDLES:
-        return "Process can not own any more handles (NO_HANDLES)";
-    case ZN_ERR_INVALID:
-        return "One or more of the provided arguments is not valid (INVALID)";
-    case ZN_ERR_BAD_RANGE:
-        return "Argument is outside of the range for valid values (BAD_RANGE)";
-    case ZN_ERR_BAD_HANDLE:
-        return "Object handle does not name a valid object (BAD_HANDLE)";
-    case ZN_ERR_BAD_PERMS:
-        return "Object has insufficient permissions for this operation (BAD_PERMS)";
-    case ZN_ERR_BAD_BUFFER:
-        return "Buffer is not large enough or doesn't point to a valid memory region (BAD_BUFFER)";
-    case ZN_ERR_ALREADY_EXISTS:
-        return "The resource already exists (ALREADY_EXISTS)";
-    }
-
-    return "Unknown error code";
-}
 
 #ifdef __cplusplus
 }

include/zinnia/system.h

@@ -2,33 +2,24 @@
 #define ZINNIA_SYSTEM_H
 
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
+#include <stdint.h>
 #include <string.h>
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-static inline void zn_log(const char* message, size_t len) {
+// Logs a message through the kernel.
-    zn_syscall2(message, len, ZN_SYSCALL_LOG);
+// Messages longer than an unspecified size will get truncated by the kernel.
-}
+void zn_log(const char* message, size_t len);
 
 // Returns the page size of the system.
-static inline size_t zn_page_size() {
+size_t zn_page_size();
-    size_t size = 0;
-    zn_syscall1(&size, ZN_SYSCALL_PAGE_SIZE);
-    return size;
-}
 
-static inline zn_status_t zn_powerctl() {
+zn_status_t zn_powerctl();
-    return ZN_ERR_UNSUPPORTED;
-}
 
-static inline uint64_t zn_time() {
+// Returns the current timestamp in nanoseconds.
-    uint64_t time = 0;
+uint64_t zn_time();
-    zn_syscall1(&time, ZN_SYSCALL_TIME);
-    return time;
-}
 
 #ifdef __cplusplus
 }

include/zinnia/thread.h

@@ -3,7 +3,6 @@
 
 #include <zinnia/handle.h>
 #include <zinnia/status.h>
-#include <zinnia/syscall_stubs.h>
 #include <stddef.h>
 #include <stdint.h>
 
@@ -16,25 +15,19 @@ enum zn_thread_flags {
 };
 
 // Creates a new thread in the given namespace.
-static inline zn_status_t zn_thread_create(
+zn_status_t zn_thread_create(
     zn_handle_t ns,
     const char* name,
     size_t name_len,
     enum zn_thread_flags flags,
     zn_handle_t* out
-) {
+);
-    return zn_syscall5(ns, name, name_len, flags, out, ZN_SYSCALL_THREAD_CREATE);
-}
 
 // Starts executing a thread.
-static inline zn_status_t zn_thread_start(zn_handle_t thread, uintptr_t ip, uintptr_t sp, uintptr_t arg) {
+zn_status_t zn_thread_start(zn_handle_t thread, uintptr_t ip, uintptr_t sp, uintptr_t arg);
-    return zn_syscall4(thread, ip, sp, arg, ZN_SYSCALL_THREAD_START);
-}
 
 // Stops execution of the calling thread.
-static inline void zn_thread_exit(void) {
+void zn_thread_exit(void);
-    zn_syscall0(ZN_SYSCALL_THREAD_EXIT);
-}
 
 #ifdef __cplusplus
 }

kernel/arch/x86_64/hpet.c

@@ -26,12 +26,12 @@ struct hpet {
 #define HPET_COUNTER (0xF0 / sizeof(uint64_t))
 
 static void hpet_reset(struct clock* c) {
-    struct hpet* hpet = CONTAINER_OF(c, struct hpet, clock);
+    struct hpet* hpet = containerof(c, struct hpet, clock);
     hpet->regs[HPET_COUNTER] = 0;
 }
 
 static uint64_t hpet_get_elapsed_ns(struct clock* c) {
-    struct hpet* hpet = CONTAINER_OF(c, struct hpet, clock);
+    struct hpet* hpet = containerof(c, struct hpet, clock);
     uint64_t counter = hpet->regs[HPET_COUNTER];
     return (counter * (uint64_t)hpet->period / 1'000'000);
 }

kernel/boot/main.c

@@ -26,7 +26,7 @@ static zn_status_t load_server(
     struct task** out
 ) {
     struct vas* space;
-    zn_status_t status = vma_new(&space);
+    zn_status_t status = vas_new(&space);
     if (status != ZN_OK)
         return status;
 
@@ -68,7 +68,7 @@ static void kernel_main_task(uintptr_t arg0) {
     assert(channel_create(&init_ep, &other_ep) == ZN_OK, "Failed to create init channel\n");
 
     struct vas* init_space;
-    assert(vma_new(&init_space) == ZN_OK, "");
+    assert(vas_new(&init_space) == ZN_OK, "");
 
     struct boot_file* file_data = &info->files[0];
     struct paged_vmo* init_file;

kernel/core/elf.c

@@ -67,7 +67,7 @@ zn_status_t elf_load(struct exec_info* info, struct task** out) {
                 return status;
 
             // We map more than we copied so the rest is filled with zeroed pages.
-            status = vma_map_vmo(info->space, &phdr_obj->object, phdr.p_vaddr, phdr.p_memsz, prot, phdr.p_offset);
+            status = vas_map_vmo(info->space, &phdr_obj->object, phdr.p_vaddr, phdr.p_memsz, prot, phdr.p_offset);
             if (status)
                 return status;
         }
@@ -182,7 +182,7 @@ zn_status_t elf_load(struct exec_info* info, struct task** out) {
     uintptr_t argc = num_argv;
     vmo_write(&stack->object, stack_off, &argc, sizeof(argc), nullptr);
 
-    status = vma_map_vmo(info->space, &stack->object, stack_start, stack_size, ZN_VM_MAP_READ | ZN_VM_MAP_WRITE, 0);
+    status = vas_map_vmo(info->space, &stack->object, stack_start, stack_size, ZN_VM_MAP_READ | ZN_VM_MAP_WRITE, 0);
     if (status)
         return status;
 

kernel/core/meson.build

@@ -1,9 +1,9 @@
 kernel_sources += files(
     'channel.c',
     'clock.c',
-    'futex.c',
     'console.c',
     'elf.c',
+    'futex.c',
     'irq.c',
     'namespace.c',
     'panic.c',
@@ -12,4 +12,5 @@ kernel_sources += files(
     'sched.c',
     'spin.c',
     'uacpi_glue.c',
+    'vdso.c',
 )

kernel/core/namespace.c

@@ -2,6 +2,7 @@
 #include <zinnia/rights.h>
 #include <zinnia/status.h>
 #include <kernel/alloc.h>
+#include <kernel/mutex.h>
 #include <kernel/namespace.h>
 #include <stdatomic.h>
 
@@ -26,6 +27,7 @@ zn_status_t namespace_add_desc(struct namespace* namespace, struct namespace_des
         return ZN_ERR_INTERNAL;
 
     // Skip over handles that are already in use (e.g. hard-coded handles).
+    // mutex_lock(&namespace->mutex);
     zn_handle_t handle;
     do {
         handle = atomic_fetch_add_explicit(&namespace->next_handle, 1, memory_order_acq_rel);
@@ -35,6 +37,8 @@ zn_status_t namespace_add_desc(struct namespace* namespace, struct namespace_des
 
     zn_status_t status =
         hashmap_insert(&namespace->descriptors, handle, desc, hashmap_hash_default, hashmap_eq_default);
+    // mutex_unlock(&namespace->mutex);
+
     if (status != ZN_OK)
         return status;
 
@@ -45,6 +49,8 @@ zn_status_t namespace_add_desc(struct namespace* namespace, struct namespace_des
 zn_status_t namespace_add_desc_at(struct namespace* namespace, struct namespace_desc desc, zn_handle_t handle) {
     if (!namespace)
         return ZN_ERR_INTERNAL;
+
+    // Skip Invalid or reserved handles.
     if (handle <= ZN_HANDLE_INVALID)
         return ZN_ERR_BAD_HANDLE;
 

kernel/core/vdso.c

@@ -0,0 +1,10 @@
+#include <common/compiler.h>
+#include <kernel/vdso.h>
+#include <stdint.h>
+
+[[__used, __section(".vdso")]]
+const uint8_t vdso_bytes[] = {
+#embed VDSO_PATH
+};
+
+const size_t vdso_length = sizeof(vdso_bytes);

kernel/include/kernel/syscall_numbers.h

@@ -1,5 +1,4 @@
-#ifndef ZINNIA_SYSCALL_NUMBERS_H
+#pragma once
-#define ZINNIA_SYSCALL_NUMBERS_H
 
 typedef enum {
     ZN_SYSCALL_LOG,
@@ -32,5 +31,3 @@ typedef enum {
     ZN_SYSCALL_FUTEX_WAIT,
     ZN_SYSCALL_FUTEX_WAKE,
 } zn_syscall_t;
-
-#endif

kernel/include/kernel/syscall_stubs.h

@@ -1,14 +1,9 @@
-#ifndef ZINNIA_SYSCALL_STUBS_H
+#pragma once
-#define ZINNIA_SYSCALL_STUBS_H
 
 #include <zinnia/status.h>
-#include <zinnia/syscall_numbers.h>
+#include <kernel/syscall_numbers.h>
 #include <stdint.h>
 
-#ifdef __cplusplus
-extern "C" {
-#endif
-
 #ifdef __x86_64__
 typedef uint64_t zn_arg_t;
 #elif defined(__aarch64__)
@@ -78,6 +73,9 @@ typedef uint64_t zn_arg_t;
 #error "Unsupported architecture!"
 #endif
 
+// These symbols should not be used in the kernel, ever.
+#ifdef IN_VDSO
+
 static inline zn_status_t zn_syscall0(zn_syscall_t num) {
     register zn_syscall_t rnum asm(ASM_REG_NUM) = num;
     register zn_status_t value asm(ASM_REG_RET);
@@ -245,8 +243,4 @@ static inline zn_status_t zn_syscall8(
         num \
     )
 
-#ifdef __cplusplus
-}
-#endif
-
 #endif

kernel/include/kernel/vas.h

@@ -13,12 +13,12 @@ struct vas {
 };
 
 // Try to create a new, empty virtual address space.
-zn_status_t vma_new(struct vas** out);
+zn_status_t vas_new(struct vas** out);
 
 // Delete a virtual address space.
-void vma_delete(struct vas* vm);
+void vas_delete(struct vas* vm);
 
-zn_status_t vma_map_vmo(
+zn_status_t vas_map_vmo(
     struct vas* vm,
     struct vmo* vmo,
     uintptr_t addr,
@@ -28,12 +28,12 @@ zn_status_t vma_map_vmo(
 );
 
 // Remove any mappings in the given address range.
-zn_status_t vma_unmap(struct vas* vm, uintptr_t vaddr, size_t size);
+zn_status_t vas_unmap(struct vas* vm, uintptr_t vaddr, size_t size);
 
 // Change the protection of an existing mapping.
-zn_status_t vma_protect(struct vas* vm, uintptr_t vaddr, size_t size, enum zn_vm_flags flags);
+zn_status_t vas_protect(struct vas* vm, uintptr_t vaddr, size_t size, enum zn_vm_flags flags);
 
 // Try to handle a page fault.
-bool vma_page_fault(struct vas* vm, uintptr_t vaddr, enum zn_vm_flags access_type);
+bool vas_page_fault(struct vas* vm, uintptr_t vaddr, enum zn_vm_flags access_type);
 
 extern struct vas kernel_vas;

kernel/include/kernel/vdso.h

@@ -1,6 +1,30 @@
 #pragma once
 
+#include <stddef.h>
 #include <stdint.h>
 
-extern uint8_t __ld_vdso_start[];
+extern const uint8_t vdso_bytes[];
-extern uint8_t __ld_vdso_end[];
+extern const size_t vdso_length;
+
+// Data which is shared between the kernel and VAS.
+// The kernel updates its contents continuously, which is visible to every VAS.
+struct vdso_kernel_data {
+    struct {
+        uint64_t sec;
+        uint64_t nsec;
+    } time;
+};
+
+#ifdef IN_VDSO
+
+#include <kernel/syscall_numbers.h>
+#include <kernel/syscall_stubs.h>
+
+// Export a function under the current API version.
+#define VDSO_SYMBOL(fn)                     asm(".symver " #fn ",__vdso_" #fn "@@ZINNIA_0")
+// Export a function as a symbol for an older API version.
+#define VDSO_SYMBOL_LEGACY(fn, symbol, ver) asm(".symver " #fn ",__vdso_" #symbol "@ZINNIA_" ver)
+
+extern const volatile struct vdso_kernel_data vdso_data;
+
+#endif

kernel/kernel.ld

@@ -30,6 +30,10 @@ SECTIONS
         KEEP(*(.cmdline))
         __ld_cmdline_end = .;
 
+        __ld_vdso_start = .;
+        KEEP(*(.vdso))
+        __ld_vdso_end = .;
+
         *(.rodata .rodata.*)
         *(.got .got.*)
 

kernel/mem/meson.build

@@ -5,7 +5,7 @@ kernel_sources += files(
     'pmm.c',
     'slab.c',
     'usercopy.c',
+    'vas.c',
     'virt.c',
-    'vma.c',
     'vmo.c',
 )

kernel/mem/vas.c

@@ -5,7 +5,7 @@
 #include <assert.h>
 
 // Try to create a new, empty virtual address space.
-zn_status_t vma_new(struct vas** out) {
+zn_status_t vas_new(struct vas** out) {
     debug_assert(out != nullptr, "Missing out");
 
     zn_status_t res;
@@ -30,23 +30,23 @@ err0:
 }
 
 // Delete a virtual address space.
-void vma_delete(struct vas* vm) {}
+void vas_delete(struct vas* vm) {}
 
 // Remove any mappings in the given address range.
 // May fail if there is no memory and a mapping needs to be split.
-zn_status_t vma_unmap(struct vas* vm, uintptr_t vaddr, size_t size);
+zn_status_t vas_unmap(struct vas* vm, uintptr_t vaddr, size_t size);
 
 // Change the protection of an existing mapping.
 // May fail if there is no memory and a mapping needs to be split.
-zn_status_t vma_protect(struct vas* vm, uintptr_t vaddr, size_t size, enum zn_vm_flags flags);
+zn_status_t vas_protect(struct vas* vm, uintptr_t vaddr, size_t size, enum zn_vm_flags flags);
 
 // Try to handle a page fault.
-bool vma_page_fault(struct vas* vm, uintptr_t vaddr, enum zn_vm_flags access_type) {
+bool vas_page_fault(struct vas* vm, uintptr_t vaddr, enum zn_vm_flags access_type) {
     // TODO
     return false;
 }
 
-zn_status_t vma_map_vmo(
+zn_status_t vas_map_vmo(
     struct vas* vas,
     struct vmo* vmo,
     uintptr_t addr,

kernel/mem/vmo.c

@@ -101,7 +101,7 @@ zn_status_t vmo_copy(
 }
 
 static zn_status_t paged_get_page(struct vmo* vmo, uintptr_t offset_idx, struct page** out) {
-    struct paged_vmo* paged = CONTAINER_OF(vmo, struct paged_vmo, object);
+    struct paged_vmo* paged = containerof(vmo, struct paged_vmo, object);
 
     // Try to find the requested page in the cache.
     struct page_list* iter;
@@ -177,7 +177,7 @@ struct phys_range_vmo {
 };
 
 static zn_status_t phys_range_get_page(struct vmo* vmo, uintptr_t offset_idx, struct page** out) {
-    struct phys_range_vmo* p = CONTAINER_OF(vmo, struct phys_range_vmo, object);
+    struct phys_range_vmo* p = containerof(vmo, struct phys_range_vmo, object);
     *out = &vm_pfndb[(p->start / arch_mem_page_size()) + offset_idx];
     return ZN_OK;
 }

kernel/meson.build

@@ -34,6 +34,7 @@ subdir('core')
 subdir('syscalls')
 
 subdir('klibc')
+subdir('vdso')
 
 executable(
     'zinnia',
@@ -44,11 +45,12 @@ executable(
     ],
     c_args: kernel_c_args,
     link_args: kernel_link_args,
-    link_depends: kernel_lds,
+    link_depends: [
+        kernel_lds,
+        vdso,
+    ],
     sources: [
         kernel_sources,
         uacpi_sources,
     ],
-    install: true,
-    install_dir: '/boot',
 )

kernel/syscalls/dispatch.c

@@ -1,9 +1,9 @@
 #include <zinnia/status.h>
-#include <zinnia/syscall_numbers.h>
 #include <common/compiler.h>
 #include <common/utils.h>
 #include <kernel/print.h>
 #include <kernel/sched.h>
+#include <kernel/syscall_numbers.h>
 #include <kernel/syscalls.h>
 #include <kernel/types.h>
 

kernel/syscalls/vmo.c

@@ -118,7 +118,7 @@ zn_status_t syscall_vmo_map(struct arch_context* ctx) {
         if (!usercopy_write(addr, &target_addr, sizeof(target_addr)))
             return ZN_ERR_BAD_BUFFER;
     }
-    vma_map_vmo(vas, vmo, target_addr, bytes, flags, vmo_offset);
+    vas_map_vmo(vas, vmo, target_addr, bytes, flags, vmo_offset);
 
     return ZN_OK;
 }

kernel/vdso/archctl.c

@@ -0,0 +1,8 @@
+#include <zinnia/archctl.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_archctl(zn_archctl_t op, void* value) {
+    return zn_syscall2(op, value, ZN_SYSCALL_ARCHCTL);
+}
+
+VDSO_SYMBOL(zn_archctl);

kernel/vdso/channel.c

@@ -0,0 +1,43 @@
+#include <zinnia/channel.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_channel_create(enum zn_channel_flags flags, zn_handle_t* endpoint0, zn_handle_t* endpoint1) {
+    return zn_syscall3(flags, endpoint0, endpoint1, ZN_SYSCALL_CHANNEL_CREATE);
+}
+
+VDSO_SYMBOL(zn_channel_create);
+
+zn_status_t zn_channel_write(
+    zn_handle_t channel,
+    const zn_handle_t* handles,
+    const void* bytes,
+    size_t num_handles,
+    size_t num_bytes
+) {
+    return zn_syscall5(channel, handles, bytes, num_handles, num_bytes, ZN_SYSCALL_CHANNEL_WRITE);
+}
+
+VDSO_SYMBOL(zn_channel_write);
+
+zn_status_t zn_channel_read(
+    zn_handle_t channel,
+    zn_handle_t* handles,
+    void* bytes,
+    size_t num_handles,
+    size_t num_bytes,
+    size_t* read_handles,
+    size_t* read_bytes
+) {
+    return zn_syscall7(
+        channel,
+        handles,
+        bytes,
+        num_handles,
+        num_bytes,
+        read_handles,
+        read_bytes,
+        ZN_SYSCALL_CHANNEL_READ
+    );
+}
+
+VDSO_SYMBOL(zn_channel_read);

kernel/vdso/data.c

@@ -0,0 +1,7 @@
+#include <common/compiler.h>
+#include <kernel/vdso.h>
+
+[[__used, __section(".vdso_data")]]
+const volatile struct vdso_kernel_data vdso_data = {0};
+
+static_assert(sizeof(vdso_data) <= 0x1000);

kernel/vdso/futex.c

@@ -0,0 +1,14 @@
+#include <zinnia/futex.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_futex_wait(int32_t* addr, int32_t expected) {
+    return zn_syscall2(addr, expected, ZN_SYSCALL_FUTEX_WAIT);
+}
+
+VDSO_SYMBOL(zn_futex_wait);
+
+zn_status_t zn_futex_wake(int32_t* addr, int32_t count) {
+    return zn_syscall2(addr, count, ZN_SYSCALL_FUTEX_WAKE);
+}
+
+VDSO_SYMBOL(zn_futex_wake);

kernel/vdso/handle.c

@@ -0,0 +1,20 @@
+#include <zinnia/handle.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_handle_validate(zn_handle_t handle) {
+    return zn_syscall1(handle, ZN_SYSCALL_HANDLE_VALIDATE);
+}
+
+VDSO_SYMBOL(zn_handle_validate);
+
+zn_status_t zn_handle_drop(zn_handle_t handle) {
+    return zn_syscall1(handle, ZN_SYSCALL_HANDLE_DROP);
+}
+
+VDSO_SYMBOL(zn_handle_drop);
+
+zn_status_t zn_handle_clone(zn_handle_t handle, zn_rights_t cloned_rights, zn_handle_t* cloned) {
+    return zn_syscall3(handle, cloned_rights, cloned, ZN_SYSCALL_HANDLE_CLONE);
+}
+
+VDSO_SYMBOL(zn_handle_clone);

kernel/vdso/mem.c

@@ -0,0 +1,33 @@
+#include <zinnia/mem.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_vmo_create(size_t length, zn_handle_t* out) {
+    return zn_syscall2(length, out, ZN_SYSCALL_VMO_CREATE);
+}
+
+VDSO_SYMBOL(zn_vmo_create);
+
+zn_status_t zn_vmo_create_phys(uintptr_t phys_addr, size_t length, zn_handle_t* out) {
+    return zn_syscall3(phys_addr, length, out, ZN_SYSCALL_VMO_CREATE_PHYS);
+}
+
+VDSO_SYMBOL(zn_vmo_create_phys);
+
+zn_status_t zn_vmo_map(
+    zn_handle_t vmo,
+    zn_handle_t vas,
+    uintptr_t vmo_offset,
+    uintptr_t* addr,
+    size_t bytes,
+    enum zn_vm_flags flags
+) {
+    return zn_syscall6(vmo, vas, vmo_offset, addr, bytes, flags, ZN_SYSCALL_VMO_MAP);
+}
+
+VDSO_SYMBOL(zn_vmo_map);
+
+zn_status_t zn_vas_create(zn_handle_t* out) {
+    return zn_syscall1(out, ZN_SYSCALL_VAS_CREATE);
+}
+
+VDSO_SYMBOL(zn_vas_create);

kernel/vdso/meson.build

@@ -0,0 +1,48 @@
+vdso_c_args = [
+    '-ffreestanding',
+    '-fno-stack-protector',
+    '-fno-omit-frame-pointer',
+    '-fPIE',
+    '-fPIC',
+    '-nostdinc',
+    '-DIN_VDSO',
+]
+
+vdso_lds = meson.current_source_dir() / 'vdso.ld'
+
+vdso_link_args = [
+    '-ffreestanding',
+    '-nostdlib',
+    '-static',
+    '-T' + vdso_lds,
+]
+
+# Build a shared library, which we are going to embed next.
+# TODO: We need to ensure there are no relocations inside the vDSO.
+vdso = shared_library(
+    'zinnia-vdso',
+    name_prefix: '',
+    c_args: vdso_c_args,
+    include_directories: [
+        common_includes,
+        kernel_includes,
+    ],
+    link_args: vdso_link_args,
+    link_depends: vdso_lds,
+    sources: [
+        'archctl.c',
+        'channel.c',
+        'data.c',
+        'futex.c',
+        'handle.c',
+        'mem.c',
+        'namespace.c',
+        'random.c',
+        'status.c',
+        'system.c',
+        'thread.c',
+    ],
+)
+
+# Tell the kernel where the built binary lives so we can #embed it.
+kernel_c_args += ['-DVDSO_PATH="' + vdso.full_path() + '"']

kernel/vdso/namespace.c

@@ -0,0 +1,14 @@
+#include <zinnia/namespace.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_ns_create(zn_handle_t* out) {
+    return ZN_ERR_UNSUPPORTED;
+}
+
+VDSO_SYMBOL(zn_ns_create);
+
+zn_status_t zn_ns_move(zn_handle_t handle, zn_handle_t target_ns) {
+    return ZN_ERR_UNSUPPORTED;
+}
+
+VDSO_SYMBOL(zn_ns_move);

kernel/vdso/random.c

@@ -0,0 +1,14 @@
+#include <zinnia/random.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_random_get(void* addr, size_t len) {
+    return zn_syscall2(addr, len, ZN_SYSCALL_RANDOM_GET);
+}
+
+VDSO_SYMBOL(zn_random_get);
+
+zn_status_t zn_random_entropy(const void* addr, size_t len) {
+    return zn_syscall2(addr, len, ZN_SYSCALL_RANDOM_ENTROPY);
+}
+
+VDSO_SYMBOL(zn_random_entropy);

kernel/vdso/status.c

@@ -0,0 +1,35 @@
+#include <zinnia/status.h>
+#include <kernel/vdso.h>
+
+const char* zn_status_to_string(zn_status_t err) {
+    switch (err) {
+    case ZN_OK:
+        return "No error (OK)";
+    case ZN_ERR_INTERNAL:
+        return "An internal error occured (INTERNAL)";
+    case ZN_ERR_BAD_SYSCALL:
+        return "Syscall number is not a recognized syscall (BAD_SYSCALL)";
+    case ZN_ERR_UNSUPPORTED:
+        return "This operation is not supported or implemented yet (UNSUPPORTED)";
+    case ZN_ERR_NO_MEMORY:
+        return "System does not have enough free memory for this operation (NO_MEMORY)";
+    case ZN_ERR_NO_HANDLES:
+        return "Process can not own any more handles (NO_HANDLES)";
+    case ZN_ERR_INVALID:
+        return "One or more of the provided arguments is not valid (INVALID)";
+    case ZN_ERR_BAD_RANGE:
+        return "Argument is outside of the range for valid values (BAD_RANGE)";
+    case ZN_ERR_BAD_HANDLE:
+        return "Object handle does not name a valid object (BAD_HANDLE)";
+    case ZN_ERR_BAD_PERMS:
+        return "Object has insufficient permissions for this operation (BAD_PERMS)";
+    case ZN_ERR_BAD_BUFFER:
+        return "Buffer is not large enough or doesn't point to a valid memory region (BAD_BUFFER)";
+    case ZN_ERR_ALREADY_EXISTS:
+        return "The resource already exists (ALREADY_EXISTS)";
+    }
+
+    return "Unknown error code";
+}
+
+VDSO_SYMBOL(zn_status_to_string);

kernel/vdso/system.c

@@ -0,0 +1,28 @@
+#include <zinnia/system.h>
+#include <kernel/vdso.h>
+
+void zn_log(const char* message, size_t len) {
+    zn_syscall2(message, len, ZN_SYSCALL_LOG);
+}
+
+VDSO_SYMBOL(zn_log);
+
+size_t zn_page_size() {
+    size_t size = 0;
+    zn_syscall1(&size, ZN_SYSCALL_PAGE_SIZE);
+    return size;
+}
+
+VDSO_SYMBOL(zn_page_size);
+
+zn_status_t zn_powerctl() {
+    return ZN_ERR_UNSUPPORTED;
+}
+
+VDSO_SYMBOL(zn_powerctl);
+
+uint64_t zn_time() {
+    return vdso_data.time.nsec;
+}
+
+VDSO_SYMBOL(zn_time);

kernel/vdso/thread.c

@@ -0,0 +1,26 @@
+#include <zinnia/thread.h>
+#include <kernel/vdso.h>
+
+zn_status_t zn_thread_create(
+    zn_handle_t ns,
+    const char* name,
+    size_t name_len,
+    enum zn_thread_flags flags,
+    zn_handle_t* out
+) {
+    return zn_syscall5(ns, name, name_len, flags, out, ZN_SYSCALL_THREAD_CREATE);
+}
+
+VDSO_SYMBOL(zn_thread_create);
+
+zn_status_t zn_thread_start(zn_handle_t thread, uintptr_t ip, uintptr_t sp, uintptr_t arg) {
+    return zn_syscall4(thread, ip, sp, arg, ZN_SYSCALL_THREAD_START);
+}
+
+VDSO_SYMBOL(zn_thread_start);
+
+void zn_thread_exit(void) {
+    zn_syscall0(ZN_SYSCALL_THREAD_EXIT);
+}
+
+VDSO_SYMBOL(zn_thread_exit);

kernel/vdso/vdso.ld

@@ -0,0 +1,58 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+
+VERSION {
+    ZINNIA_0 {
+        global:
+            __vdso*;
+        local: *;
+    };
+}
+
+PHDRS {
+    text    PT_LOAD    FILEHDR PHDRS FLAGS(5);
+    dynamic PT_DYNAMIC FLAGS(4);
+    note    PT_NOTE    FLAGS(4);
+}
+
+SECTIONS
+{
+    . = SIZEOF_HEADERS;
+
+    .text : {
+        *(.text .text.*)
+        *(.plt .plt.*)
+    } :text
+
+    .rodata : {
+        *(.rodata .rodata.*)
+        *(.got .got.*)
+    } :text
+
+    .data ALIGN(CONSTANT(MAXPAGESIZE)) : {
+        *(.vdso_data)
+    } :text
+
+    .dynsym : {
+        *(.dynsym)
+    } :text
+
+    .dynstr : {
+        *(.dynstr)
+    } :text
+
+    .dynamic : {
+        *(.dynamic)
+    } :text :dynamic
+
+    /DISCARD/ :
+    {
+        *(.data*)
+        *(.bss*)
+        *(.comment*)
+        *(.eh_frame*)
+        *(.rel.eh_frame*)
+        *(.gcc_except_table*)
+        *(.note.gnu.property)
+        *(.interp)
+    }
+}

lib/meson.build

@@ -0,0 +1,7 @@
+libzinnia = static_library(
+    'zinnia',
+    sources: [
+        'vdso.c',
+    ],
+    include_directories: common_includes,
+)

lib/vdso.c

@@ -0,0 +1,138 @@
+#include <zinnia/archctl.h>
+#include <zinnia/channel.h>
+#include <zinnia/futex.h>
+#include <zinnia/handle.h>
+#include <zinnia/mem.h>
+#include <zinnia/namespace.h>
+#include <zinnia/random.h>
+#include <zinnia/status.h>
+#include <zinnia/system.h>
+#include <zinnia/thread.h>
+#include <common/elf.h>
+#include <sys/auxv.h>
+
+static uint8_t vdso_symbol_bind(uint8_t info) {
+    return ELF64_ST_BIND(info);
+}
+
+static uint8_t vdso_symbol_type(uint8_t info) {
+    return ELF64_ST_TYPE(info);
+}
+
+struct vdso_info {
+    uintptr_t base;
+    const char* strings;
+    const struct elf64_sym* symbols;
+    size_t symbol_count;
+};
+
+static bool vdso_info_init(struct vdso_info* info) {
+    const uintptr_t base = getauxval(AT_SYSINFO_EHDR);
+    if (!base)
+        return false;
+
+    const struct elf64_ehdr* ehdr = (const void*)base;
+    if (ehdr->e_ident[EI_MAG0] != ELF_MAG[0] || ehdr->e_ident[EI_MAG1] != ELF_MAG[1] ||
+        ehdr->e_ident[EI_MAG2] != ELF_MAG[2] || ehdr->e_ident[EI_MAG3] != ELF_MAG[3] ||
+        ehdr->e_ident[EI_CLASS] != ELFCLASS64 || ehdr->e_ident[EI_DATA] != ELFDATA2LSB ||
+        ehdr->e_machine != ELF_ARCH_MACHINE)
+        return false;
+
+    const struct elf64_dyn* dynamic = nullptr;
+    for (size_t i = 0; i < ehdr->e_phnum; i++) {
+        const struct elf64_phdr* phdr = (const void*)(base + ehdr->e_phoff + (i * ehdr->e_phentsize));
+        if (phdr->p_type == PT_DYNAMIC) {
+            dynamic = (const void*)(base + phdr->p_vaddr);
+            break;
+        }
+    }
+    if (!dynamic)
+        return false;
+
+    const char* strings = nullptr;
+    const struct elf64_sym* symbols = nullptr;
+    const struct elf_hash* hash = nullptr;
+
+    for (const struct elf64_dyn* entry = dynamic; entry->d_tag != DT_NULL; entry++) {
+        switch (entry->d_tag) {
+        case DT_STRTAB:
+            strings = (const void*)(base + entry->d_un.d_ptr);
+            break;
+        case DT_SYMTAB:
+            symbols = (const void*)(base + entry->d_un.d_ptr);
+            break;
+        case DT_HASH:
+            hash = (const void*)(base + entry->d_un.d_ptr);
+            break;
+        default:
+            break;
+        }
+    }
+
+    if (!strings || !symbols || !hash)
+        return false;
+
+    *info = (struct vdso_info){
+        .base = base,
+        .strings = strings,
+        .symbols = symbols,
+        .symbol_count = hash->n_chain,
+    };
+    return true;
+}
+
+static void* vdso_lookup(const char* name) {
+    struct vdso_info info;
+    if (!vdso_info_init(&info))
+        return nullptr;
+
+    for (size_t i = 0; i < info.symbol_count; i++) {
+        const struct elf64_sym* symbol = &info.symbols[i];
+        if (vdso_symbol_bind(symbol->st_info) != STB_GLOBAL || vdso_symbol_type(symbol->st_info) != STT_FUNC)
+            continue;
+
+        const char* symbol_name = info.strings + symbol->st_name;
+        const char* lhs = symbol_name;
+        const char* rhs = name;
+        while (*lhs && *lhs == *rhs) {
+            lhs++;
+            rhs++;
+        }
+        if (*lhs == 0 && *rhs == 0)
+            return (void*)(info.base + symbol->st_value);
+    }
+
+    return nullptr;
+}
+
+#define VDSO_IFUNC(name) \
+    static typeof(name)* resolve_##name(void) { \
+        return (typeof(name)*)vdso_lookup("__vdso_" #name); \
+    } \
+    typeof(name) name __attribute__((ifunc("resolve_" #name)));
+
+VDSO_IFUNC(zn_archctl);
+VDSO_IFUNC(zn_channel_create);
+VDSO_IFUNC(zn_channel_write);
+VDSO_IFUNC(zn_channel_read);
+VDSO_IFUNC(zn_futex_wait);
+VDSO_IFUNC(zn_futex_wake);
+VDSO_IFUNC(zn_handle_validate);
+VDSO_IFUNC(zn_handle_drop);
+VDSO_IFUNC(zn_handle_clone);
+VDSO_IFUNC(zn_vmo_create);
+VDSO_IFUNC(zn_vmo_create_phys);
+VDSO_IFUNC(zn_vmo_map);
+VDSO_IFUNC(zn_vas_create);
+VDSO_IFUNC(zn_ns_create);
+VDSO_IFUNC(zn_ns_move);
+VDSO_IFUNC(zn_random_get);
+VDSO_IFUNC(zn_random_entropy);
+VDSO_IFUNC(zn_status_to_string);
+VDSO_IFUNC(zn_log);
+VDSO_IFUNC(zn_page_size);
+VDSO_IFUNC(zn_powerctl);
+VDSO_IFUNC(zn_time);
+VDSO_IFUNC(zn_thread_create);
+VDSO_IFUNC(zn_thread_start);
+VDSO_IFUNC(zn_thread_exit);

servers/devd/meson.build

@@ -11,5 +11,6 @@ executable(
         common_includes,
         uacpi_includes,
     ],
+    link_with: libzinnia,
     install: true,
 )

servers/initd/meson.build

@@ -3,6 +3,7 @@ executable(
     'main.c',
     # The init process is linked statically because the dynamic loader can't be read before posixd is up.
     link_args: '-static',
+    link_with: libzinnia,
     include_directories: common_includes,
     install: true,
 )

servers/unix/main.c

@@ -1,3 +1,4 @@
+#include <zinnia/archctl.h>
 #include <zinnia/channel.h>
 #include <zinnia/handle.h>
 #include <zinnia/mem.h>
@@ -54,6 +55,7 @@ int other_fn(void* arg) {
 }
 
 int main(int argc, char* argv[], char* envp[]) {
+    zn_archctl(ZN_ARCHCTL_SET_FSBASE, nullptr);
     zn_printf("unix: starting...\n");
     const char* thread_name = "unix_second_thread";
 

servers/unix/meson.build

@@ -4,7 +4,7 @@ executable(
     'unix',
     'main.c',
     'uname.c',
-    link_args: '-static',
+    link_with: libzinnia,
     include_directories: [common_includes, 'include'],
     install: true,
 )