fix several issues relating to elf and registers and add mmap

src/arch/x86_64/isr.asm

@@ -5,6 +5,12 @@ section .text
 
 
 	push rbp
+	mov  rbp, [rsp+24] ; save CS for usermode checking
+	; check and patch ss if it is invalid because of the syscall instruction
+	cmp qword [rsp+48], 0x33
+	jne .keepgoing
+	mov qword [rsp+48], 0x43
+	.keepgoing:
 	push rsi
 	push rdi
 	push r15
@@ -22,12 +28,13 @@ section .text
 	
 
 
-	mov rax, ds
+	
-	cmp rax,0x40
+	cmp rbp,0x3b
 	jne .noswapgs
 	swapgs
 	.noswapgs:
 
+	mov rax, ds
 	push rax
 	mov rax, es
 	push rax
@@ -38,10 +45,10 @@ section .text
 	
 	; now use kernel segments for data
 
-	mov rax, 0x30
+	;mov rax, 0x38
-	mov ds, rax
+	;mov ds, rax
-	mov es, rax
+	;mov es, rax
-	mov ss, rax
+	;mov ss, rax
 	mov rdi,cr2
 	push rdi
 
@@ -52,18 +59,15 @@ section .text
 	
 	add rsp, 24 ; cr2 gs fs
 
+	cmp rbp,0x3b
+	jne .noswapgspop
+	swapgs
+	.noswapgspop:
+	
 	pop rax
-	cmp rax,0x43
-	je .dontpopsegs
 	mov es,rax
 	pop rax
 	mov ds,rax
-	swapgs
-	jmp .poprest
-	.dontpopsegs:
-	add rsp,8 ; ds
-	
-	.poprest:
 	pop rax
 	pop rbx
 	pop rcx

src/arch/x86_64/syscall.asm

@@ -2,12 +2,13 @@ section .rodata
 
 
 extern syscall_libc_log
-func_count equ 1
+extern syscall_mmap
+func_count equ 2
 
 
 func_table:
 	dq syscall_libc_log
-
+	dq syscall_mmap
 section .text
 global asm_syscall_entry
 
@@ -46,10 +47,13 @@ asm_syscall_entry:
 	push r9
 	push r10
 
+
+
 	; system V C abi expects the third param to be in rcx
 	; but the param is in rdx because of the syscall instruction
 
-	mov rcx, rdx
+	mov rcx, r10
+
 
 	call [func_table + rax * 8]
 

src/include/kernel/elf.h

@@ -60,6 +60,8 @@ typedef struct{
 #define PH_TYPE_DYNAMIC 2
 #define PH_TYPE_INTERPRETER 3
 #define PH_TYPE_NOTE 4
+#define PH_TYPE_SHLIB 5
+#define PH_TYPE_PHDR 6
 
 typedef struct{
 	uint32_t type;

src/include/kernel/syscalls.h

@@ -7,5 +7,6 @@ typedef struct {
 } syscallret;
 
 syscallret syscall_libc_log(char* str);
+syscallret syscall_mmap(void* hint, size_t len, int prot, int flags, int fd, off_t offset);
 
 #endif

src/include/kernel/vmm.h

@@ -7,8 +7,10 @@
 
 #define KERNEL_SPACE_START 0xFFFF800000000000
 #define KERNEL_SPACE_END   0xFFFFFFFFFFFFFFFF
-#define USER_SPACE_START   0x1000
+#define USER_SPACE_START   0x0
-#define USER_SPACE_END     0x00007FFFFFFFFFFF
+#define USER_SPACE_END     (void*)0x00007FFFFFFFFFFF
+
+#define USER_ALLOC_START   (void*)0x1000
 
 #define VMM_TYPE_FREE 0
 #define VMM_TYPE_ANON 1
@@ -60,5 +62,6 @@ bool		vmm_map(void* paddr, void* vaddr, size_t pagec, size_t mmuflags);
 void*		vmm_alloc(size_t pagec, size_t mmuflags);
 bool		vmm_setfree(void* addr, size_t pagec);
 bool		vmm_allocnowat(void* addr, size_t mmuflags, size_t size);
+void*		vmm_allocfrom(void* addr, size_t mmuflags, size_t size);
 vmm_context*	vmm_newcontext();
 #endif

src/sys/mm/vmm.c

@@ -106,6 +106,17 @@ static char* debugnames[] = {
 	"FILE"
 };
 
+static void debug_dumpusermappings(){
+	vmm_mapping* map = arch_getcls()->context->userstart;
+	
+	printf("User mappings:\n");
+
+	while(map != NULL){
+		printf("START: %016p END: %016p MMUFLAGS: %016p DATA: %016p OFFSET: %lu TYPE: %s\n", map->start, map->end, map->mmuflags, map->data, map->offset, debugnames[map->type]);
+		map = map->next;
+	}
+
+}
 static void debug_dumpkernelmappings(){
 	vmm_mapping* map = kmapstart;
 	
@@ -375,6 +386,31 @@ void* vmm_alloc(size_t pagec, size_t mmuflags){
 	return addr;
 }
 
+void* vmm_allocfrom(void* addr, size_t mmuflags, size_t pagec){
+	
+	int* lock;
+	vmm_mapping** start;
+	
+	getcontextinfo(addr, &lock, &start);
+
+	spinlock_acquire(lock);
+	
+	addr = findfirstfreearea(findmappingfromaddr(*start, addr), pagec);
+
+	if(!addr)
+		goto _done;
+	
+	if(!setmap(start, addr, pagec, mmuflags, VMM_TYPE_ANON, 0, 0))
+		addr = NULL;
+
+	_done:
+	
+	spinlock_release(lock);
+
+	return addr;
+	
+}
+
 bool vmm_unmap(void* addr, size_t pagec){
 	
 	int* lock;
@@ -448,6 +484,14 @@ bool vmm_allocnowat(void* addr, size_t mmuflags, size_t size){
 
 	spinlock_acquire(lock);
 	
+	vmm_mapping* m = findmappingfromaddr(*start, addr);
+
+	if(m->type != VMM_TYPE_FREE || (addr + size*PAGE_SIZE-1) > m->end){
+		spinlock_release(lock);
+		return false;
+	}
+
+
 	// XXX allocate 1 page at a time to reduce fragmentation
 
 	void* paddr = pmm_alloc(size);
@@ -481,7 +525,6 @@ bool vmm_dealwithrequest(void* addr){
 
 	spinlock_acquire(lock);
 
-
 	bool status;
 	vmm_mapping* map = findmappingfromaddr(*start, addr);
 	void* paddr;
@@ -498,7 +541,7 @@ bool vmm_dealwithrequest(void* addr){
 
 	addr = (size_t)addr & ~(0xFFF);
 
-	printf("Demand paging %p\n", addr);
+	//printf("Demand paging %p\n", addr);
 	
 	
 

src/sys/sched/elf.c

@@ -57,7 +57,7 @@ static void* stacksetup(char* interp, char** argv, char** env, elf_header64 head
 	size_t initialsize = 0;
 	size_t argdatasize = 0;
 	size_t envdatasize = 0;
-	size_t auxvsize    = 5*8; // FIXME better way of keeping track of this
+	size_t auxvsize    = 5*16; // FIXME better way of keeping track of this
 
 	while(argv[argc]){
 		argdatasize += strlen(argv[argc]) + 1;
@@ -90,8 +90,7 @@ static void* stacksetup(char* interp, char** argv, char** env, elf_header64 head
 
 	// get the addresses to copy the data to
 
-	elf_ph64* phdrdatastart = STACK_TOP - header.ph_count * header.ph_size;
+	char* argdatastart = STACK_TOP  - argdatasize;
-	char* argdatastart = (void*) phdrdatastart - argdatasize;
 	char* envdatastart = argdatastart - envdatasize;
 	
 	auxv64_t* auxvstart = (void*)(((uintptr_t)envdatastart & (~0xF)) - auxvsize);
@@ -108,11 +107,9 @@ static void* stacksetup(char* interp, char** argv, char** env, elf_header64 head
 	// AT_PHDR
 	
 	auxvstart->a_type = AT_PHDR;
-	auxvstart->a_val = (uint64_t)phdrdatastart;
+	auxvstart->a_val = (uint64_t)phs;
 	auxvstart++;
 
-	memcpy(phdrdatastart, phs, header.ph_count * header.ph_size);
-
 	// AT_PHENT
 
 	auxvstart->a_type = AT_PHENT;
@@ -123,6 +120,7 @@ static void* stacksetup(char* interp, char** argv, char** env, elf_header64 head
 
 	auxvstart->a_type = AT_PHNUM;
 	auxvstart->a_val = header.ph_count;
+	auxvstart++;
 
 	// AT_ENTRY
 	
@@ -133,6 +131,7 @@ static void* stacksetup(char* interp, char** argv, char** env, elf_header64 head
 	// AT_NULL
 
 	auxvstart->a_type = AT_NULL;
+	auxvstart->a_val  = 0;
 
 	// copy the data
 
@@ -174,6 +173,7 @@ int elf_load(thread_t* thread, vnode_t* node, char** argv, char** env){
 	
 	elf_header64 header;	
 	elf_header64 progheader;
+	vnode_t* prognode = node;
 	elf_ph64 ph; 
 
 	int err = 0;
@@ -193,7 +193,7 @@ int elf_load(thread_t* thread, vnode_t* node, char** argv, char** env){
 	// and also save the executable's program headers
 	// for auxv
 
-	elf_ph64 progphs[header.ph_count];
+	elf_ph64* progphs;
 
 	for(size_t currentph = 0; currentph < header.ph_count; ++currentph){	
 		readc = vfs_read(&err, node, &ph, header.ph_size, header.ph_pos + currentph*header.ph_size);
@@ -202,7 +202,8 @@ int elf_load(thread_t* thread, vnode_t* node, char** argv, char** env){
 		if(readc != header.ph_size)
 			return EINVAL;
 		
-		memcpy(progphs+currentph, &ph, sizeof(elf_ph64));
+		if(ph.type == PH_TYPE_PHDR)
+			progphs = (elf_ph64*)ph.memaddr;
 		
 		if(ph.type == PH_TYPE_INTERPRETER){
 
@@ -210,7 +211,7 @@ int elf_load(thread_t* thread, vnode_t* node, char** argv, char** env){
 
 			progheader = header;
 
-			char* interp = alloc(ph.fsize + 1);
+			interp = alloc(ph.fsize + 1);
 			if(!interp)
 				return ENOMEM;
 
@@ -300,9 +301,63 @@ int elf_load(thread_t* thread, vnode_t* node, char** argv, char** env){
 
 	}
 	
+	// also load the executable memory image if we're using an interpreter
+	
+	if(interp){
+		for(size_t currentph = 0; currentph < header.ph_count; ++currentph){
+
+			// read ph
+
+			readc = vfs_read(&err, prognode, &ph, progheader.ph_size, progheader.ph_pos + currentph*progheader.ph_size);
+
+			if(err) goto _fail;
+
+			if(readc != progheader.ph_size){
+				err = EINVAL;
+				goto _fail;
+			}
+
+			if(ph.type != PH_TYPE_LOAD)
+				continue;
+
+			if(!vmm_allocnowat((void*)ph.memaddr, phflagtommuflag(ph.flags), ph.msize)){
+				err = ENOMEM;
+				goto _fail;
+			}
+
+			// set up temporary mmu flags to copy data
+
+			arch_mmu_changeflags(arch_getcls()->context->context, ph.memaddr, ARCH_MMU_MAP_READ | ARCH_MMU_MAP_WRITE, ph.msize / PAGE_SIZE + (ph.msize % PAGE_SIZE ? 1 : 0));
+
+			// read program data
+
+			readc = vfs_read(&err, prognode, (void*)ph.memaddr, ph.fsize, ph.offset);
+
+			if(err)
+				goto _fail;
+
+			if(readc != ph.fsize){
+				err = EINVAL;
+				goto _fail;
+			}
+
+			// zero if needed
+
+			size_t zerocount = ph.msize - ph.fsize;
+
+			if(zerocount){
+				void* zeroaddr = ph.memaddr + ph.fsize;
+				memset(zeroaddr, 0, zerocount);
+			}
+
+			// reset proper mmu flags
+
+			arch_mmu_changeflags(arch_getcls()->context->context, ph.memaddr, phflagtommuflag(ph.flags), ph.msize / PAGE_SIZE + (ph.msize % PAGE_SIZE ? 1 : 0));
 
 
+        	}
 	
+	}
 
 	void* stack = stacksetup(interp, argv, env, progheader, progphs);
 

src/sys/syscalls/libclog.c

@@ -17,7 +17,7 @@ syscallret syscall_libc_log(char* str){
 
 	strcpy(buf, str);
 
-	printf("%s", str);
+	printf("%s\n", str);
 
 	ret.errno = 0;
 	ret.ret = 0;

src/sys/syscalls/mmap.c

@@ -0,0 +1,97 @@
+#include <kernel/syscalls.h>
+#include <kernel/vmm.h>
+#include <sys/types.h>
+#include <errno.h>
+
+// linux abi
+
+#define PROT_NONE  0x00
+#define PROT_READ  0x01
+#define PROT_WRITE 0x02
+#define PROT_EXEC  0x04
+
+#define MAP_FILE    0x00
+#define MAP_SHARED    0x01
+#define MAP_PRIVATE   0x02
+#define MAP_FIXED     0x10
+#define MAP_ANON      0x20
+#define MAP_ANONYMOUS 0x20
+#define MAP_NORESERVE 0x4000
+
+#define MS_ASYNC 0x01
+#define MS_INVALIDATE 0x02
+#define MS_SYNC 0x04
+
+#define MCL_CURRENT 0x01
+#define MCL_FUTURE 0x02
+
+#define POSIX_MADV_NORMAL 0
+#define POSIX_MADV_RANDOM 1
+#define POSIX_MADV_SEQUENTIAL 2
+#define POSIX_MADV_WILLNEED 3
+#define POSIX_MADV_DONTNEED 4
+
+#define MADV_NORMAL 0
+#define MADV_RANDOM 1
+#define MADV_SEQUENTIAL 2
+#define MADV_WILLNEED 3
+#define MADV_DONTNEED 4
+#define MADV_FREE 8
+
+#define MREMAP_MAYMOVE 1
+#define MREMAP_FIXED 2
+
+#define MFD_CLOEXEC 1U
+#define MFD_ALLOW_SEALING 2U
+
+syscallret syscall_mmap(void* hint, size_t len, int prot, int flags, int fd, off_t offset){
+	syscallret retv;
+	
+	retv.ret = -1;
+	retv.errno = 0;
+
+	if(len == 0 || hint > USER_SPACE_END){
+		retv.errno = EINVAL;
+		return retv;
+	}
+
+	if(flags != (MAP_ANON | MAP_PRIVATE)){
+		retv.errno = ENOSYS;
+		return retv;
+		
+	}
+	
+	size_t mmuflags = 0;
+
+	if(prot & PROT_READ)
+		mmuflags |= ARCH_MMU_MAP_READ;
+	
+	if(prot & PROT_WRITE)
+		mmuflags |= ARCH_MMU_MAP_WRITE;
+	
+	if(!prot & PROT_EXEC)
+		mmuflags |= ARCH_MMU_MAP_NOEXEC;
+	
+	mmuflags |= ARCH_MMU_MAP_USER;
+
+	size_t plen = len / PAGE_SIZE + (len % PAGE_SIZE ? 1 : 0);
+
+	void* ret = NULL;
+
+	if(hint)
+		ret = vmm_allocnowat(hint, mmuflags, plen) ? hint : NULL; // TODO demand page
+	
+	if(!ret && !(flags & MAP_FIXED))
+		ret = vmm_allocfrom(USER_ALLOC_START, mmuflags, len);
+
+	if(!ret){
+		retv.errno = ENOMEM;
+		if(flags & MAP_ANON)
+			memset(ret, 0, plen);
+	}
+	else
+		retv.ret = ret;
+
+	return retv;
+
+}