Stub for 64-bit shlib

modified:   stub/Makefile
	modified:   stub/src/amd64-expand.S
	modified:   stub/src/amd64-linux.elf-entry.S
	modified:   stub/src/amd64-linux.elf-so_entry.S
	modified:   stub/src/amd64-linux.elf-so_fold.S
	modified:   stub/src/amd64-linux.elf-so_main.c
	modified:   stub/src/arm64-expand.S
	modified:   stub/src/upxfd_linux.c

src/stub/Makefile

@@ -421,11 +421,13 @@ tc.amd64-linux.elf.gcc += -fno-exceptions -fno-asynchronous-unwind-tables
 tc.amd64-linux.elf.gcc += -Wall -W -Wcast-align -Wcast-qual -Wstrict-prototypes -Wwrite-strings -Werror
 
 amd64-linux.elf-entry.h: $(srcdir)/src/$$T.S
+	@echo; echo TARGET=$@  PATH=$(PATH); echo
 	$(call tc,gcc) -c -x assembler-with-cpp $< -o tmp/$T.bin
 	$(call tc,f-embed_objinfo,tmp/$T.bin)
 	$(call tc,bin2h) tmp/$T.bin $@
 
 amd64-linux.elf-so_entry.h: $(srcdir)/src/$$T.S
+	@echo; echo TARGET=$@  PATH=$(PATH); echo
 	$(call tc,gcc) -c -x assembler-with-cpp $< -o tmp/$T.bin
 	$(call tc,f-embed_objinfo,tmp/$T.bin)
 	$(call tc,bin2h) tmp/$T.bin $@
@@ -433,17 +435,32 @@ amd64-linux.elf-so_entry.h: $(srcdir)/src/$$T.S
 amd64-linux.elf-fold.h : $(srcdir)/src/$$T.lds \
 	    tmp/$$T.o \
 	    tmp/amd64-expand.o \
+	    tmp/amd64-linux.elf-upxfd_linux.o \
 	    tmp/amd64-linux.elf-main2.o
+	@echo; echo TARGET=$@  PATH=$(PATH); echo
 # FIXME: multiarch-ld-2.18 creates a huge file here, so use 2.17
 	multiarch-ld-2.17 -r -T $(srcdir)/src/$T.lds -Map tmp/$T.map $(filter %.o,$^) -o tmp/$T.bin
 	$(call tc,f-embed_objinfo_without_xstrip_keep_dot_text,tmp/$T.bin)
 	$(call tc,bin2h) tmp/$T.bin $@
 
-amd64-linux.elf-so_fold.h : tmp/$$T.o tmp/amd64-linux.elf-so_main.o $(srcdir)/src/$$T.lds tmp/amd64-expand.o
+amd64-linux.elf-so_fold.h : $(srcdir)/src/$$T.lds \
-	multiarch-ld-2.17 -r          -T $(srcdir)/src/$T.lds -Map tmp/$T.map $(filter %.o,$^) -o tmp/$T.bin
+	    tmp/$$T.o \
+	    tmp/amd64-expand.o \
+	    tmp/amd64-linux.elf-upxfd_linux.o \
+	    tmp/amd64-linux.elf-so_main.o
+	@echo; echo TARGET=$@  PATH=$(PATH); echo
+# FIXME: multiarch-ld-2.18 creates a huge file here, so use 2.17
+	$(call tc,ld)     -r -T $(srcdir)/src/$T.lds -Map tmp/$T.map $(filter %.o,$^) -o tmp/$T.bin
+#	multiarch-ld-2.17 -r -T $(srcdir)/src/$T.lds -Map tmp/$T.map $(filter %.o,$^) -o tmp/$T.bin
+	ls -l tmp/$T.bin
 	$(call tc,f-embed_objinfo_without_xstrip,tmp/$T.bin)
 	$(call tc,bin2h) tmp/$T.bin $@
 
+tmp/amd64-linux.elf-so_fold.o : $(srcdir)/src/$$T.S
+	$(call tc,gcc) -c $< -o tmp/$T.o
+	$(call tc,objcopy) -R .text $@
+	$(call tc,f-objstrip,$@)
+
 tmp/amd64-expand.o: $(srcdir)/src/$$T.S
 	$(call tc,gcc) -c $< -o $@
 
@@ -451,11 +468,6 @@ tmp/amd64-linux.elf-fold.o : $(srcdir)/src/$$T.S
 	$(call tc,gcc) -c $< -o $@
 	$(call tc,f-objstrip,$@)
 
-tmp/amd64-linux.elf-so_fold.o : $(srcdir)/src/$$T.S
-	$(call tc,gcc) -c $< -o $@
-	$(call tc,objcopy) -R .text $@
-	$(call tc,f-objstrip,$@)
-
 tmp/amd64-linux.elf-main.o : $(srcdir)/src/$$T.c
 	$(call tc,gcc) -c -Os $< -o $@
 	$(call tc,f-objstrip,$@)
@@ -834,6 +846,11 @@ tmp/arm64-expand.o: $(srcdir)/src/$$T.S
 	$(call tc,gcc) -c $< -o $@
 	arm64-linux-objdump-2.25 -Dr $(tc_objdump_disasm_options) tmp/$T.o | $(RTRIM) > tmp/$T.o.disasm
 
+tmp/amd64-linux.elf-upxfd_linux.o : $(srcdir)/src/upxfd_linux.c
+	$(call tc,gcc) -c -O $< -o $@
+	$(call tc,objcopy) --rename-section .text=UMF_LINUX -R .comment -R .data -R .bss -R .note.GNU-stack $@
+	$(call tc,objdump) -Dr $(tc_objdump_disasm_options) $@ | $(RTRIM) > $@.disasm
+
 tmp/arm64-linux.elf-main2.o : $(srcdir)/src/$$T.c $(srcdir)/src/amd64-linux.elf-main2.c
 	$(call tc,gcc) -c -Os $(srcdir)/src/arm64-linux.elf-main2.c -o $@
 	$(call tc,f-objstrip,$@)

src/stub/src/amd64-expand.S

@@ -229,3 +229,7 @@ eof_n2e:
         movl %edi,(%rcx)  // actual length used at dst  XXX: 4GB
         sub %rsi,%rax  // src -= eof;  // return 0: good; else: bad
         ret
+
+        .balign 4
+upx_mmap_and_fd: .globl upx_mmap_and_fd
+    // UMF_LINUX goes here

src/stub/src/amd64-linux.elf-entry.S

@@ -129,9 +129,9 @@ D_PMASK= 0*NBPW
 #define arg2l esi
 #define arg3l edx
 // Create anonymous temporary file on mfd; like upxfd_create
-        push $'u'|('p'<<8)|('x'<<16)|(0<<24)  // MATCH_22
+        push $'u'|('p'<<8)|('X'<<16)|(0<<24)  // MATCH_22
-        push %rsp; pop %arg1  // "upx"
+        push %rsp; pop %arg1  // "upX"
-        mov $MFD_EXEC,%arg2l
+        push $MFD_EXEC; pop %arg2
 0: // try memfd_create
         movl $__NR_memfd_create,%eax; syscall
         test %eax,%eax; jns ok_memfd  // success

src/stub/src/amd64-linux.elf-so_entry.S

@@ -58,6 +58,7 @@ __NR_mmap=      9  // 64-bit mode only!  /usr/include/asm/unistd_64.h
 __NR_mprotect= 10
 __NR_munmap=   11
 __NR_memfd_create= 0x13f  // 319
+  MFD_EXEC= 0x10
 
 __NR_write=  1
 __NR_close=  3
@@ -204,9 +205,20 @@ eof_n2b:
 
         pop %rbp  // MATCH_45
 
-        push $0; pop %arg2
+#define arg2l esi
-        call 0f; .asciz "upx"; 0: pop %arg1
+// Create anonymous temporary file on mfd; like upxfd_create
-        push $__NR_memfd_create; call do_sys
+        push $'u'|('p'<<8)|('X'<<16)|(0<<24)  // MATCH_22
+        push %rsp; pop %arg1  // "upX"
+        push $MFD_EXEC; pop %arg2
+0: // try memfd_create
+        movl $__NR_memfd_create,%eax; syscall
+        test %eax,%eax; jns ok_memfd  // success
+        test %arg2l,%arg2l; jz no_memfd  // memfd_create failed twice
+        xor %arg2l,%arg2l; jmp 0b  // try again without MFD_EXEC
+no_memfd:  // so try /dev/shm
+        hlt  // FIXME /dev/shm
+ok_memfd:
+        pop %rcx  // MATCH_22  discard "upx"
 
         push %rax; pop %arg1  // mfd
         push %rsp; pop %arg2  // buffer

src/stub/src/amd64-linux.elf-so_fold.S

@@ -98,6 +98,7 @@ __NR_read=  0
 __NR_write= 1
 //__NR_open=  2
 __NR_openat= 257
+  FD_CWD= -100
 __NR_close= 3
 
 __NR_mmap=      9
@@ -149,6 +150,13 @@ sysgo:  # NOTE: kernel demands 4th arg in %sys4, NOT %arg4
         syscall; cmp $-4096,%rax; jb 0f; int3; 0:
         ret
 
+open: .globl open
+        mov %arg3,%arg4
+        mov %arg2,%arg3
+        mov %arg1,%arg2
+        mov $FD_CWD,%arg1
+        jmp openat
+
 Punmap: .globl Punmap  // page-align the lo end
         mov %arg1,%rax; and $0xfff,%eax
         sub %rax,%arg1

src/stub/src/amd64-linux.elf-so_main.c

@@ -40,8 +40,6 @@ extern size_t Pwrite(unsigned, void const *, size_t);
 
 extern void f_int3(int arg);
 
-#define DEBUG 1
-
 #ifndef DEBUG  //{
 #define DEBUG 0
 #endif  //}
@@ -384,10 +382,10 @@ make_hatch(
         ) >> ((pf & (PF_R|PF_W|PF_X))<<2) ))
 
 #undef PAGE_MASK
-static unsigned long
+static ElfW(Addr)
 get_page_mask(void)  // the mask which KEEPS the page, discards the offset
 {
-    unsigned long rv = ~0xffful;  // default to (PAGE_SIZE == 4KiB)
+    ElfW(Addr) rv = ~0xffful;  // default to (PAGE_SIZE == 4KiB)
     int fd = openat(0, addr_string("/proc/self/auxv"), O_RDONLY, 0);
     if (0 <= fd) {
         ElfW(auxv_t) data[40];
@@ -454,16 +452,25 @@ fini_SELinux(
 }
 
 unsigned
-prep_SELinux(unsigned size, char *ptr, unsigned len) // returns mfd
+prep_SELinux(unsigned size, char *ptr, ElfW(Addr) page_mask) // returns mfd
 {
     // Cannot set PROT_EXEC except via mmap() into a region (Linux "vma")
     // that has never had PROT_WRITE.  So use a Linux-only "memory file"
     // to hold the contents.
-    char *val = upx_mmap_and_fd(ptr, size, nullptr);
+    unsigned saved[65536/sizeof(unsigned)];
-    unsigned mfd = 0xfff & (unsigned)(long)val;
+    char *page = (char *)(page_mask & (ElfW(Addr))ptr);
+    unsigned frag = (unsigned)(ptr - page);
+    if (frag) {
+        memcpy(saved, page, frag);
+    }
+    char *val = upx_mmap_and_fd(page, frag + size, nullptr);
+    unsigned mfd = 0xfff & (unsigned)(ElfW(Addr))val;
     val -= mfd; --mfd;
-    if (len)
+    if (val != page) {
-        Pwrite(mfd, ptr, len);  // Save lo fragment of contents on first page.
+        my_bkpt((void const *)0x1262, val, page, ptr, frag);
+    }
+    if (frag)
+        write(mfd, saved, frag);  // Save lo fragment of contents on page.
     return mfd;
 }
 
@@ -491,7 +498,7 @@ upx_so_main(  // returns &escape_hatch
     ElfW(Ehdr) *elf_tmp  // scratch for ElfW(Ehdr) and ElfW(Phdrs)
 )
 {
-    unsigned long const page_mask = get_page_mask();
+    ElfW(Addr) const page_mask = get_page_mask();
     char *const va_load = (char *)&so_info->off_reloc - so_info->off_reloc;
     So_info so_infc;  // So_info Copy
     memcpy(&so_infc, so_info, sizeof(so_infc));  // before de-compression overwrites
@@ -548,59 +555,41 @@ upx_so_main(  // returns &escape_hatch
 
     for (; phdr < phdrN; ++phdr)
     if (phdr->p_type == PT_LOAD && !(phdr->p_flags & PF_W)) {
-        unsigned hi_offset = phdr->p_filesz + phdr->p_offset;
-        struct b_info al_bi;  // for aligned data from binfo
         // Need un-aligned read of b_info to determine compression sizes.
+        struct b_info al_bi;  // for aligned data from binfo
         x0.size = sizeof(struct b_info);
         xread(&x0, (char *)&al_bi, x0.size);  // aligned binfo
         x0.buf -= sizeof(al_bi);  // back up (the xread() was a peek)
-        x1.size = hi_offset - xct_off;
-        x1.buf = (void *)(hi_offset + base - al_bi.sz_unc);
         x0.size = al_bi.sz_cpr;
+        x1.size = al_bi.sz_unc;
+
+        DPRINTF("\\nphdr@%%p  p_offset=%%p  p_vaddr=%%p  p_filesz=%%p  p_memsz=%%p\\n",
+            phdr, phdr->p_offset, phdr->p_vaddr, phdr->p_filesz, phdr->p_memsz);
+        DPRINTF("x0=%%p  x1=%%p\\n", &x0, &x1);
 
         if  (!base) {
             base = (ElfW(Addr))va_load - phdr->p_vaddr;
             DPRINTF("base=%%p\\n", base);
         }
-        DPRINTF("phdr@%%p  p_offset=%%p  p_vaddr=%%p  p_filesz=%%p  p_memsz=%%p\\n",
+        x1.buf = (void *)(phdr->p_filesz + phdr->p_vaddr + base - x1.size);
-            phdr, phdr->p_offset, phdr->p_vaddr, phdr->p_filesz, phdr->p_memsz);
+        if ((phdr->p_filesz + phdr->p_offset) <= xct_off) { // hi_offset <= xct_off
-        DPRINTF("x0=%%p  x1=%%p\\n", &x0, &x1);
+            if (!n_load) {
-        //my_bkpt((void const *)0x1230, phdr, &x0, &x1);
+                ++n_load;
-
+                continue;  // 1st PT_LOAD is non-compressed loader tables ONLY!
-        unsigned frag = ~page_mask & (unsigned)(long)x1.buf;
+            }
-        unsigned mfd = 0;
+        }
-        if (!n_load) { // 1st PT_LOAD is special.
+        int mfd = 0;
-            // Already ELF headers are in place, perhaps already followed
-            // by non-compressed loader tables below xct_off.
-            if (xct_off < hi_offset) { // 1st PT_LOAD also has compressed code, too
         if (phdr->p_flags & PF_X) {
-                    mfd = prep_SELinux(x1.size, x1.buf, frag);
+            mfd = prep_SELinux(x1.size, x1.buf, page_mask);
         }
         else {
             underlay(x1.size, x1.buf, page_mask);  // also makes PROT_WRITE
         }
+        Extent xt = x1;
         unpackExtent(&x0, &x1);
         if (!hatch && phdr->p_flags & PF_X) {
             hatch = make_hatch(phdr, x1.buf, ~page_mask);
-                }
+            fini_SELinux(xt.size, xt.buf, phdr, mfd, base);  // FIXME: x1 was changed by unpackExtent!
-                my_bkpt((void const *)0x1235, &x1);
-                fini_SELinux(x1.size, x1.buf, phdr, mfd, base);  // FIXME: x1 changed!
-            }
-        }
-        else { // 2nd and later PT_LOADs
-            x1.buf = (void *)(phdr->p_vaddr + base);
-            x1.size = phdr->p_filesz;
-            if (phdr->p_flags & PF_X) {
-                mfd = prep_SELinux(x1.size, x1.buf, frag);
-            }
-            else {
-                underlay(x1.size, x1.buf, page_mask);  // also makes PROT_WRITE
-            }
-            unpackExtent(&x0, &x1);
-            if (!hatch && phdr->p_flags &PF_X) {
-                hatch = make_hatch(phdr, x1.buf, ~page_mask);
-            }
-            fini_SELinux(al_bi.sz_unc, (void *)(phdr->p_vaddr + base), phdr, mfd, base);
         }
         ++n_load;
     }

src/stub/src/arm64-expand.S

@@ -140,6 +140,9 @@ eof_lzma: .globl eof_lzma
         POP4(x2,x3, fp,lr)  // MATCH_94  x2= orig_dst; x3= plen_dst
         ret
 
+upx_mmap_and_fd: .globl upx_mmap_and_fd
+    // UMF_LINUX goes here
+
 #define M_NRV2B_LE32    2
 #define M_NRV2B_8    3
 #define M_NRV2D_LE32    5

src/stub/src/upxfd_linux.c

@@ -1,4 +1,4 @@
-/* upxfd_create.c -- workaround for 32-bit Android
+/* upxfd_create.c -- simplify upx_mmap_and_fd_linux for non-Android
 
    This file is part of the UPX executable compressor.