/* $NetBSD: elf2aout.c,v 1.9 2000/01/11 20:33:21 is Exp $ */ /* * Copyright (c) 1995 * Ted Lemon (hereinafter referred to as the author) * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* elf2aout.c This program converts an elf executable to a NetBSD a.out executable. The minimal symbol table is copied, but the debugging symbols and other informational sections are not. */ #include #include #include #include #include #include #include #include #include #include #include #include struct sect { unsigned long vaddr; unsigned long len; }; void combine __P((struct sect *, struct sect *, int)); int phcmp __P((const void *, const void *)); char *saveRead __P((int file, off_t offset, off_t len, char *name)); void copy __P((int, int, off_t, off_t)); void translate_syms __P((int, int, off_t, off_t, off_t, off_t)); int *symTypeTable; int main(int argc, char **argv) { Elf32_Ehdr ex; Elf32_Phdr *ph; Elf32_Shdr *sh; char *shstrtab; int strtabix, symtabix; int i; struct sect text, data, bss; struct exec aex; int infile, outfile; unsigned long cur_vma = ULONG_MAX; int symflag = 0; strtabix = symtabix = 0; text.len = data.len = bss.len = 0; text.vaddr = data.vaddr = bss.vaddr = 0; /* Check args... */ if (argc < 3 || argc > 4) { usage: fprintf(stderr, "usage: elf2aout [-s]\n"); exit(1); } if (argc == 4) { if (strcmp(argv[3], "-s")) goto usage; symflag = 1; } /* Try the input file... */ if ((infile = open(argv[1], O_RDONLY)) < 0) { fprintf(stderr, "Can't open %s for read: %s\n", argv[1], strerror(errno)); exit(1); } /* Read the header, which is at the beginning of the file... */ i = read(infile, &ex, sizeof ex); if (i != sizeof ex) { fprintf(stderr, "ex: %s: %s.\n", argv[1], i ? strerror(errno) : "End of file reached"); exit(1); } /* Read the program headers... */ ph = (Elf32_Phdr *) saveRead(infile, ex.e_phoff, ex.e_phnum * sizeof(Elf32_Phdr), "ph"); /* Read the section headers... */ sh = (Elf32_Shdr *) saveRead(infile, ex.e_shoff, ex.e_shnum * sizeof(Elf32_Shdr), "sh"); /* Read in the section string table. */ shstrtab = saveRead(infile, sh[ex.e_shstrndx].sh_offset, sh[ex.e_shstrndx].sh_size, "shstrtab"); /* Find space for a table matching ELF section indices to a.out symbol * types. */ symTypeTable = (int *) malloc(ex.e_shnum * sizeof(int)); if (!symTypeTable) { fprintf(stderr, "symTypeTable: can't allocate.\n"); exit(1); } memset(symTypeTable, 0, ex.e_shnum * sizeof(int)); /* Look for the symbol table and string table... Also map section * indices to symbol types for a.out */ for (i = 0; i < ex.e_shnum; i++) { char *name = shstrtab + sh[i].sh_name; if (!strcmp(name, ".symtab")) symtabix = i; else if (!strcmp(name, ".strtab")) strtabix = i; else if (!strcmp(name, ".text") || !strcmp(name, ".rodata")) symTypeTable[i] = N_TEXT; else if (!strcmp(name, ".data") || !strcmp(name, ".sdata") || !strcmp(name, ".lit4") || !strcmp(name, ".lit8")) symTypeTable[i] = N_DATA; else if (!strcmp(name, ".bss") || !strcmp(name, ".sbss")) symTypeTable[i] = N_BSS; } /* Figure out if we can cram the program header into an a.out * header... Basically, we can't handle anything but loadable * segments, but we can ignore some kinds of segments. We can't * handle holes in the address space, and we handle start addresses * other than 0x1000 by hoping that the loader will know where to load * - a.out doesn't have an explicit load address. Segments may be * out of order, so we sort them first. */ qsort(ph, ex.e_phnum, sizeof(Elf32_Phdr), phcmp); for (i = 0; i < ex.e_phnum; i++) { /* Section types we can ignore... */ if (ph[i].p_type == PT_NULL || ph[i].p_type == PT_NOTE || ph[i].p_type == PT_PHDR || ph[i].p_type == PT_MIPS_REGINFO) continue; /* Section types we can't handle... */ else if (ph[i].p_type != PT_LOAD) errx(1, "Program header %d type %d can't be converted.", i, ph[i].p_type); /* Writable (data) segment? */ if (ph[i].p_flags & PF_W) { struct sect ndata, nbss; ndata.vaddr = ph[i].p_vaddr; ndata.len = ph[i].p_filesz; nbss.vaddr = ph[i].p_vaddr + ph[i].p_filesz; nbss.len = ph[i].p_memsz - ph[i].p_filesz; combine(&data, &ndata, 0); combine(&bss, &nbss, 1); } else { struct sect ntxt; ntxt.vaddr = ph[i].p_vaddr; ntxt.len = ph[i].p_filesz; combine(&text, &ntxt, 0); } /* Remember the lowest segment start address. */ if (ph[i].p_vaddr < cur_vma) cur_vma = ph[i].p_vaddr; } /* Sections must be in order to be converted... */ if (text.vaddr > data.vaddr || data.vaddr > bss.vaddr || text.vaddr + text.len > data.vaddr || data.vaddr + data.len > bss.vaddr) { fprintf(stderr, "Sections ordering prevents a.out conversion.\n"); exit(1); } /* If there's a data section but no text section, then the loader * combined everything into one section. That needs to be the text * section, so just make the data section zero length following text. */ if (data.len && !text.len) { text = data; data.vaddr = text.vaddr + text.len; data.len = 0; } /* If there is a gap between text and data, we'll fill it when we copy * the data, so update the length of the text segment as represented * in a.out to reflect that, since a.out doesn't allow gaps in the * program address space. */ if (text.vaddr + text.len < data.vaddr) text.len = data.vaddr - text.vaddr; /* We now have enough information to cons up an a.out header... */ aex.a_midmag = htonl((symflag << 26) | (MID_PMAX << 16) | OMAGIC); if (ex.e_machine == EM_PPC) aex.a_midmag = htonl((symflag << 26) | (MID_POWERPC << 16) | OMAGIC); aex.a_text = text.len; aex.a_data = data.len; aex.a_bss = bss.len; aex.a_entry = ex.e_entry; aex.a_syms = (sizeof(struct nlist) * (symtabix != -1 ? sh[symtabix].sh_size / sizeof(Elf32_Sym) : 0)); aex.a_trsize = 0; aex.a_drsize = 0; /* Make the output file... */ if ((outfile = open(argv[2], O_WRONLY | O_CREAT, 0777)) < 0) { fprintf(stderr, "Unable to create %s: %s\n", argv[2], strerror(errno)); exit(1); } /* Truncate file... */ if (ftruncate(outfile, 0)) { warn("ftruncate %s", argv[2]); } /* Write the header... */ i = write(outfile, &aex, sizeof aex); if (i != sizeof aex) { perror("aex: write"); exit(1); } /* Copy the loadable sections. Zero-fill any gaps less than 64k; * complain about any zero-filling, and die if we're asked to * zero-fill more than 64k. */ for (i = 0; i < ex.e_phnum; i++) { /* Unprocessable sections were handled above, so just verify * that the section can be loaded before copying. */ if (ph[i].p_type == PT_LOAD && ph[i].p_filesz) { if (cur_vma != ph[i].p_vaddr) { unsigned long gap = ph[i].p_vaddr - cur_vma; char obuf[1024]; if (gap > 65536) errx(1, "Intersegment gap (%ld bytes) too large.", (long) gap); #ifdef DEBUG warnx("Warning: %ld byte intersegment gap.", (long)gap); #endif memset(obuf, 0, sizeof obuf); while (gap) { int count = write(outfile, obuf, (gap > sizeof obuf ? sizeof obuf : gap)); if (count < 0) { fprintf(stderr, "Error writing gap: %s\n", strerror(errno)); exit(1); } gap -= count; } } copy(outfile, infile, ph[i].p_offset, ph[i].p_filesz); cur_vma = ph[i].p_vaddr + ph[i].p_filesz; } } /* Copy and translate the symbol table... */ translate_syms(outfile, infile, sh[symtabix].sh_offset, sh[symtabix].sh_size, sh[strtabix].sh_offset, sh[strtabix].sh_size); /* Looks like we won... */ exit(0); } /* translate_syms (out, in, offset, size) Read the ELF symbol table from in at offset; translate it into a.out nlist format and write it to out. */ void translate_syms(out, in, symoff, symsize, stroff, strsize) int out, in; off_t symoff, symsize; off_t stroff, strsize; { #define SYMS_PER_PASS 64 Elf32_Sym inbuf[64]; struct nlist outbuf[64]; int i, remaining, cur; char *oldstrings; char *newstrings, *nsp; int newstringsize; /* Zero the unused fields in the output buffer.. */ memset(outbuf, 0, sizeof outbuf); /* Find number of symbols to process... */ remaining = symsize / sizeof(Elf32_Sym); /* Suck in the old string table... */ oldstrings = saveRead(in, stroff, strsize, "string table"); /* Allocate space for the new one. XXX We make the wild assumption * that no two symbol table entries will point at the same place in * the string table - if that assumption is bad, this could easily * blow up. */ newstringsize = strsize + remaining; newstrings = (char *) malloc(newstringsize); if (!newstrings) { fprintf(stderr, "No memory for new string table!\n"); exit(1); } /* Initialize the table pointer... */ nsp = newstrings; /* Go the the start of the ELF symbol table... */ if (lseek(in, symoff, SEEK_SET) < 0) { perror("translate_syms: lseek"); exit(1); } /* Translate and copy symbols... */ while (remaining) { cur = remaining; if (cur > SYMS_PER_PASS) cur = SYMS_PER_PASS; remaining -= cur; if ((i = read(in, inbuf, cur * sizeof(Elf32_Sym))) != cur * sizeof(Elf32_Sym)) { if (i < 0) perror("translate_syms"); else fprintf(stderr, "translate_syms: premature end of file.\n"); exit(1); } /* Do the translation... */ for (i = 0; i < cur; i++) { int binding, type; /* Copy the symbol into the new table, but prepend an * underscore. */ *nsp = '_'; strcpy(nsp + 1, oldstrings + inbuf[i].st_name); outbuf[i].n_un.n_strx = nsp - newstrings + 4; nsp += strlen(nsp) + 1; type = ELF32_ST_TYPE(inbuf[i].st_info); binding = ELF32_ST_BIND(inbuf[i].st_info); /* Convert ELF symbol type/section/etc info into a.out * type info. */ if (type == STT_FILE) outbuf[i].n_type = N_FN; else if (inbuf[i].st_shndx == SHN_UNDEF) outbuf[i].n_type = N_UNDF; else if (inbuf[i].st_shndx == SHN_ABS) outbuf[i].n_type = N_ABS; else if (inbuf[i].st_shndx == SHN_COMMON || inbuf[i].st_shndx == SHN_MIPS_ACOMMON) outbuf[i].n_type = N_COMM; else outbuf[i].n_type = symTypeTable[inbuf[i].st_shndx]; if (binding == STB_GLOBAL) outbuf[i].n_type |= N_EXT; /* Symbol values in executables should be compatible. */ outbuf[i].n_value = inbuf[i].st_value; } /* Write out the symbols... */ if ((i = write(out, outbuf, cur * sizeof(struct nlist))) != cur * sizeof(struct nlist)) { fprintf(stderr, "translate_syms: write: %s\n", strerror(errno)); exit(1); } } /* Write out the string table length... */ if (write(out, &newstringsize, sizeof newstringsize) != sizeof newstringsize) { fprintf(stderr, "translate_syms: newstringsize: %s\n", strerror(errno)); exit(1); } /* Write out the string table... */ if (write(out, newstrings, newstringsize) != newstringsize) { fprintf(stderr, "translate_syms: newstrings: %s\n", strerror(errno)); exit(1); } } void copy(out, in, offset, size) int out, in; off_t offset, size; { char ibuf[4096]; int remaining, cur, count; /* Go the the start of the ELF symbol table... */ if (lseek(in, offset, SEEK_SET) < 0) { perror("copy: lseek"); exit(1); } remaining = size; while (remaining) { cur = remaining; if (cur > sizeof ibuf) cur = sizeof ibuf; remaining -= cur; if ((count = read(in, ibuf, cur)) != cur) { fprintf(stderr, "copy: read: %s\n", count ? strerror(errno) : "premature end of file"); exit(1); } if ((count = write(out, ibuf, cur)) != cur) { perror("copy: write"); exit(1); } } } /* Combine two segments, which must be contiguous. If pad is true, it's okay for there to be padding between. */ void combine(base, new, pad) struct sect *base, *new; int pad; { if (!base->len) *base = *new; else if (new->len) { if (base->vaddr + base->len != new->vaddr) { if (pad) base->len = new->vaddr - base->vaddr; else { fprintf(stderr, "Non-contiguous data can't be converted.\n"); exit(1); } } base->len += new->len; } } int phcmp(vh1, vh2) const void *vh1, *vh2; { Elf32_Phdr *h1, *h2; h1 = (Elf32_Phdr *) vh1; h2 = (Elf32_Phdr *) vh2; if (h1->p_vaddr > h2->p_vaddr) return 1; else if (h1->p_vaddr < h2->p_vaddr) return -1; else return 0; } char * saveRead(int file, off_t offset, off_t len, char *name) { char *tmp; int count; off_t off; if ((off = lseek(file, offset, SEEK_SET)) < 0) { fprintf(stderr, "%s: fseek: %s\n", name, strerror(errno)); exit(1); } if (!(tmp = (char *) malloc(len))) errx(1, "%s: Can't allocate %ld bytes.", name, (long)len); count = read(file, tmp, len); if (count != len) { fprintf(stderr, "%s: read: %s.\n", name, count ? strerror(errno) : "End of file reached"); exit(1); } return tmp; }