ACPI: ibm-acpi: break fan_read into separate functions
[linux-2.6/cjktty.git] / arch / mips / kernel / irixelf.c
blobab12c8f0151852e74a66f781fe875ffc362b2368
1 /*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
6 * irixelf.c: Code to load IRIX ELF executables conforming to the MIPS ABI.
7 * Based off of work by Eric Youngdale.
9 * Copyright (C) 1993 - 1994 Eric Youngdale <ericy@cais.com>
10 * Copyright (C) 1996 - 2004 David S. Miller <dm@engr.sgi.com>
11 * Copyright (C) 2004 - 2005 Steven J. Hill <sjhill@realitydiluted.com>
13 #include <linux/module.h>
14 #include <linux/fs.h>
15 #include <linux/stat.h>
16 #include <linux/sched.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/a.out.h>
20 #include <linux/errno.h>
21 #include <linux/init.h>
22 #include <linux/signal.h>
23 #include <linux/binfmts.h>
24 #include <linux/string.h>
25 #include <linux/file.h>
26 #include <linux/fcntl.h>
27 #include <linux/ptrace.h>
28 #include <linux/slab.h>
29 #include <linux/shm.h>
30 #include <linux/personality.h>
31 #include <linux/elfcore.h>
32 #include <linux/smp_lock.h>
34 #include <asm/mipsregs.h>
35 #include <asm/namei.h>
36 #include <asm/prctl.h>
37 #include <asm/uaccess.h>
39 #define DLINFO_ITEMS 12
41 #include <linux/elf.h>
43 #undef DEBUG
45 static int load_irix_binary(struct linux_binprm * bprm, struct pt_regs * regs);
46 static int load_irix_library(struct file *);
47 static int irix_core_dump(long signr, struct pt_regs * regs,
48 struct file *file);
50 static struct linux_binfmt irix_format = {
51 NULL, THIS_MODULE, load_irix_binary, load_irix_library,
52 irix_core_dump, PAGE_SIZE
55 #ifndef elf_addr_t
56 #define elf_addr_t unsigned long
57 #endif
59 #ifdef DEBUG
60 /* Debugging routines. */
61 static char *get_elf_p_type(Elf32_Word p_type)
63 int i = (int) p_type;
65 switch(i) {
66 case PT_NULL: return("PT_NULL"); break;
67 case PT_LOAD: return("PT_LOAD"); break;
68 case PT_DYNAMIC: return("PT_DYNAMIC"); break;
69 case PT_INTERP: return("PT_INTERP"); break;
70 case PT_NOTE: return("PT_NOTE"); break;
71 case PT_SHLIB: return("PT_SHLIB"); break;
72 case PT_PHDR: return("PT_PHDR"); break;
73 case PT_LOPROC: return("PT_LOPROC/REGINFO"); break;
74 case PT_HIPROC: return("PT_HIPROC"); break;
75 default: return("PT_BOGUS"); break;
79 static void print_elfhdr(struct elfhdr *ehp)
81 int i;
83 printk("ELFHDR: e_ident<");
84 for(i = 0; i < (EI_NIDENT - 1); i++) printk("%x ", ehp->e_ident[i]);
85 printk("%x>\n", ehp->e_ident[i]);
86 printk(" e_type[%04x] e_machine[%04x] e_version[%08lx]\n",
87 (unsigned short) ehp->e_type, (unsigned short) ehp->e_machine,
88 (unsigned long) ehp->e_version);
89 printk(" e_entry[%08lx] e_phoff[%08lx] e_shoff[%08lx] "
90 "e_flags[%08lx]\n",
91 (unsigned long) ehp->e_entry, (unsigned long) ehp->e_phoff,
92 (unsigned long) ehp->e_shoff, (unsigned long) ehp->e_flags);
93 printk(" e_ehsize[%04x] e_phentsize[%04x] e_phnum[%04x]\n",
94 (unsigned short) ehp->e_ehsize, (unsigned short) ehp->e_phentsize,
95 (unsigned short) ehp->e_phnum);
96 printk(" e_shentsize[%04x] e_shnum[%04x] e_shstrndx[%04x]\n",
97 (unsigned short) ehp->e_shentsize, (unsigned short) ehp->e_shnum,
98 (unsigned short) ehp->e_shstrndx);
101 static void print_phdr(int i, struct elf_phdr *ep)
103 printk("PHDR[%d]: p_type[%s] p_offset[%08lx] p_vaddr[%08lx] "
104 "p_paddr[%08lx]\n", i, get_elf_p_type(ep->p_type),
105 (unsigned long) ep->p_offset, (unsigned long) ep->p_vaddr,
106 (unsigned long) ep->p_paddr);
107 printk(" p_filesz[%08lx] p_memsz[%08lx] p_flags[%08lx] "
108 "p_align[%08lx]\n", (unsigned long) ep->p_filesz,
109 (unsigned long) ep->p_memsz, (unsigned long) ep->p_flags,
110 (unsigned long) ep->p_align);
113 static void dump_phdrs(struct elf_phdr *ep, int pnum)
115 int i;
117 for(i = 0; i < pnum; i++, ep++) {
118 if((ep->p_type == PT_LOAD) ||
119 (ep->p_type == PT_INTERP) ||
120 (ep->p_type == PT_PHDR))
121 print_phdr(i, ep);
124 #endif /* DEBUG */
126 static void set_brk(unsigned long start, unsigned long end)
128 start = PAGE_ALIGN(start);
129 end = PAGE_ALIGN(end);
130 if (end <= start)
131 return;
132 down_write(&current->mm->mmap_sem);
133 do_brk(start, end - start);
134 up_write(&current->mm->mmap_sem);
138 /* We need to explicitly zero any fractional pages
139 * after the data section (i.e. bss). This would
140 * contain the junk from the file that should not
141 * be in memory.
143 static void padzero(unsigned long elf_bss)
145 unsigned long nbyte;
147 nbyte = elf_bss & (PAGE_SIZE-1);
148 if (nbyte) {
149 nbyte = PAGE_SIZE - nbyte;
150 clear_user((void __user *) elf_bss, nbyte);
154 static unsigned long * create_irix_tables(char * p, int argc, int envc,
155 struct elfhdr * exec, unsigned int load_addr,
156 unsigned int interp_load_addr, struct pt_regs *regs,
157 struct elf_phdr *ephdr)
159 elf_addr_t *argv;
160 elf_addr_t *envp;
161 elf_addr_t *sp, *csp;
163 #ifdef DEBUG
164 printk("create_irix_tables: p[%p] argc[%d] envc[%d] "
165 "load_addr[%08x] interp_load_addr[%08x]\n",
166 p, argc, envc, load_addr, interp_load_addr);
167 #endif
168 sp = (elf_addr_t *) (~15UL & (unsigned long) p);
169 csp = sp;
170 csp -= exec ? DLINFO_ITEMS*2 : 2;
171 csp -= envc+1;
172 csp -= argc+1;
173 csp -= 1; /* argc itself */
174 if ((unsigned long)csp & 15UL) {
175 sp -= (16UL - ((unsigned long)csp & 15UL)) / sizeof(*sp);
179 * Put the ELF interpreter info on the stack
181 #define NEW_AUX_ENT(nr, id, val) \
182 __put_user ((id), sp+(nr*2)); \
183 __put_user ((val), sp+(nr*2+1)); \
185 sp -= 2;
186 NEW_AUX_ENT(0, AT_NULL, 0);
188 if(exec) {
189 sp -= 11*2;
191 NEW_AUX_ENT (0, AT_PHDR, load_addr + exec->e_phoff);
192 NEW_AUX_ENT (1, AT_PHENT, sizeof (struct elf_phdr));
193 NEW_AUX_ENT (2, AT_PHNUM, exec->e_phnum);
194 NEW_AUX_ENT (3, AT_PAGESZ, ELF_EXEC_PAGESIZE);
195 NEW_AUX_ENT (4, AT_BASE, interp_load_addr);
196 NEW_AUX_ENT (5, AT_FLAGS, 0);
197 NEW_AUX_ENT (6, AT_ENTRY, (elf_addr_t) exec->e_entry);
198 NEW_AUX_ENT (7, AT_UID, (elf_addr_t) current->uid);
199 NEW_AUX_ENT (8, AT_EUID, (elf_addr_t) current->euid);
200 NEW_AUX_ENT (9, AT_GID, (elf_addr_t) current->gid);
201 NEW_AUX_ENT (10, AT_EGID, (elf_addr_t) current->egid);
203 #undef NEW_AUX_ENT
205 sp -= envc+1;
206 envp = sp;
207 sp -= argc+1;
208 argv = sp;
210 __put_user((elf_addr_t)argc,--sp);
211 current->mm->arg_start = (unsigned long) p;
212 while (argc-->0) {
213 __put_user((unsigned long)p,argv++);
214 p += strlen_user(p);
216 __put_user((unsigned long) NULL, argv);
217 current->mm->arg_end = current->mm->env_start = (unsigned long) p;
218 while (envc-->0) {
219 __put_user((unsigned long)p,envp++);
220 p += strlen_user(p);
222 __put_user((unsigned long) NULL, envp);
223 current->mm->env_end = (unsigned long) p;
224 return sp;
228 /* This is much more generalized than the library routine read function,
229 * so we keep this separate. Technically the library read function
230 * is only provided so that we can read a.out libraries that have
231 * an ELF header.
233 static unsigned int load_irix_interp(struct elfhdr * interp_elf_ex,
234 struct file * interpreter,
235 unsigned int *interp_load_addr)
237 struct elf_phdr *elf_phdata = NULL;
238 struct elf_phdr *eppnt;
239 unsigned int len;
240 unsigned int load_addr;
241 int elf_bss;
242 int retval;
243 unsigned int last_bss;
244 int error;
245 int i;
246 unsigned int k;
248 elf_bss = 0;
249 last_bss = 0;
250 error = load_addr = 0;
252 #ifdef DEBUG
253 print_elfhdr(interp_elf_ex);
254 #endif
256 /* First of all, some simple consistency checks */
257 if ((interp_elf_ex->e_type != ET_EXEC &&
258 interp_elf_ex->e_type != ET_DYN) ||
259 !interpreter->f_op->mmap) {
260 printk("IRIX interp has bad e_type %d\n", interp_elf_ex->e_type);
261 return 0xffffffff;
264 /* Now read in all of the header information */
265 if(sizeof(struct elf_phdr) * interp_elf_ex->e_phnum > PAGE_SIZE) {
266 printk("IRIX interp header bigger than a page (%d)\n",
267 (sizeof(struct elf_phdr) * interp_elf_ex->e_phnum));
268 return 0xffffffff;
271 elf_phdata = kmalloc(sizeof(struct elf_phdr) * interp_elf_ex->e_phnum,
272 GFP_KERNEL);
274 if(!elf_phdata) {
275 printk("Cannot kmalloc phdata for IRIX interp.\n");
276 return 0xffffffff;
279 /* If the size of this structure has changed, then punt, since
280 * we will be doing the wrong thing.
282 if(interp_elf_ex->e_phentsize != 32) {
283 printk("IRIX interp e_phentsize == %d != 32 ",
284 interp_elf_ex->e_phentsize);
285 kfree(elf_phdata);
286 return 0xffffffff;
289 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
290 (char *) elf_phdata,
291 sizeof(struct elf_phdr) * interp_elf_ex->e_phnum);
293 #ifdef DEBUG
294 dump_phdrs(elf_phdata, interp_elf_ex->e_phnum);
295 #endif
297 eppnt = elf_phdata;
298 for(i=0; i<interp_elf_ex->e_phnum; i++, eppnt++) {
299 if(eppnt->p_type == PT_LOAD) {
300 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
301 int elf_prot = 0;
302 unsigned long vaddr = 0;
303 if (eppnt->p_flags & PF_R) elf_prot = PROT_READ;
304 if (eppnt->p_flags & PF_W) elf_prot |= PROT_WRITE;
305 if (eppnt->p_flags & PF_X) elf_prot |= PROT_EXEC;
306 elf_type |= MAP_FIXED;
307 vaddr = eppnt->p_vaddr;
309 pr_debug("INTERP do_mmap(%p, %08lx, %08lx, %08lx, %08lx, %08lx) ",
310 interpreter, vaddr,
311 (unsigned long) (eppnt->p_filesz + (eppnt->p_vaddr & 0xfff)),
312 (unsigned long) elf_prot, (unsigned long) elf_type,
313 (unsigned long) (eppnt->p_offset & 0xfffff000));
314 down_write(&current->mm->mmap_sem);
315 error = do_mmap(interpreter, vaddr,
316 eppnt->p_filesz + (eppnt->p_vaddr & 0xfff),
317 elf_prot, elf_type,
318 eppnt->p_offset & 0xfffff000);
319 up_write(&current->mm->mmap_sem);
321 if(error < 0 && error > -1024) {
322 printk("Aieee IRIX interp mmap error=%d\n", error);
323 break; /* Real error */
325 pr_debug("error=%08lx ", (unsigned long) error);
326 if(!load_addr && interp_elf_ex->e_type == ET_DYN) {
327 load_addr = error;
328 pr_debug("load_addr = error ");
331 /* Find the end of the file mapping for this phdr, and keep
332 * track of the largest address we see for this.
334 k = eppnt->p_vaddr + eppnt->p_filesz;
335 if(k > elf_bss) elf_bss = k;
337 /* Do the same thing for the memory mapping - between
338 * elf_bss and last_bss is the bss section.
340 k = eppnt->p_memsz + eppnt->p_vaddr;
341 if(k > last_bss) last_bss = k;
342 pr_debug("\n");
346 /* Now use mmap to map the library into memory. */
347 if(error < 0 && error > -1024) {
348 pr_debug("got error %d\n", error);
349 kfree(elf_phdata);
350 return 0xffffffff;
353 /* Now fill out the bss section. First pad the last page up
354 * to the page boundary, and then perform a mmap to make sure
355 * that there are zero-mapped pages up to and including the
356 * last bss page.
358 pr_debug("padzero(%08lx) ", (unsigned long) (elf_bss));
359 padzero(elf_bss);
360 len = (elf_bss + 0xfff) & 0xfffff000; /* What we have mapped so far */
362 pr_debug("last_bss[%08lx] len[%08lx]\n", (unsigned long) last_bss,
363 (unsigned long) len);
365 /* Map the last of the bss segment */
366 if (last_bss > len) {
367 down_write(&current->mm->mmap_sem);
368 do_brk(len, (last_bss - len));
369 up_write(&current->mm->mmap_sem);
371 kfree(elf_phdata);
373 *interp_load_addr = load_addr;
374 return ((unsigned int) interp_elf_ex->e_entry);
377 /* Check sanity of IRIX elf executable header. */
378 static int verify_binary(struct elfhdr *ehp, struct linux_binprm *bprm)
380 if (memcmp(ehp->e_ident, ELFMAG, SELFMAG) != 0)
381 return -ENOEXEC;
383 /* First of all, some simple consistency checks */
384 if((ehp->e_type != ET_EXEC && ehp->e_type != ET_DYN) ||
385 !bprm->file->f_op->mmap) {
386 return -ENOEXEC;
389 /* XXX Don't support N32 or 64bit binaries yet because they can
390 * XXX and do execute 64 bit instructions and expect all registers
391 * XXX to be 64 bit as well. We need to make the kernel save
392 * XXX all registers as 64bits on cpu's capable of this at
393 * XXX exception time plus frob the XTLB exception vector.
395 if((ehp->e_flags & EF_MIPS_ABI2))
396 return -ENOEXEC;
398 return 0;
402 * This is where the detailed check is performed. Irix binaries
403 * use interpreters with 'libc.so' in the name, so this function
404 * can differentiate between Linux and Irix binaries.
406 static inline int look_for_irix_interpreter(char **name,
407 struct file **interpreter,
408 struct elfhdr *interp_elf_ex,
409 struct elf_phdr *epp,
410 struct linux_binprm *bprm, int pnum)
412 int i;
413 int retval = -EINVAL;
414 struct file *file = NULL;
416 *name = NULL;
417 for(i = 0; i < pnum; i++, epp++) {
418 if (epp->p_type != PT_INTERP)
419 continue;
421 /* It is illegal to have two interpreters for one executable. */
422 if (*name != NULL)
423 goto out;
425 *name = kmalloc(epp->p_filesz + strlen(IRIX_EMUL), GFP_KERNEL);
426 if (!*name)
427 return -ENOMEM;
429 strcpy(*name, IRIX_EMUL);
430 retval = kernel_read(bprm->file, epp->p_offset, (*name + 16),
431 epp->p_filesz);
432 if (retval < 0)
433 goto out;
435 file = open_exec(*name);
436 if (IS_ERR(file)) {
437 retval = PTR_ERR(file);
438 goto out;
440 retval = kernel_read(file, 0, bprm->buf, 128);
441 if (retval < 0)
442 goto dput_and_out;
444 *interp_elf_ex = *(struct elfhdr *) bprm->buf;
446 *interpreter = file;
447 return 0;
449 dput_and_out:
450 fput(file);
451 out:
452 kfree(*name);
453 return retval;
456 static inline int verify_irix_interpreter(struct elfhdr *ihp)
458 if (memcmp(ihp->e_ident, ELFMAG, SELFMAG) != 0)
459 return -ELIBBAD;
460 return 0;
463 #define EXEC_MAP_FLAGS (MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE)
465 static inline void map_executable(struct file *fp, struct elf_phdr *epp, int pnum,
466 unsigned int *estack, unsigned int *laddr,
467 unsigned int *scode, unsigned int *ebss,
468 unsigned int *ecode, unsigned int *edata,
469 unsigned int *ebrk)
471 unsigned int tmp;
472 int i, prot;
474 for(i = 0; i < pnum; i++, epp++) {
475 if(epp->p_type != PT_LOAD)
476 continue;
478 /* Map it. */
479 prot = (epp->p_flags & PF_R) ? PROT_READ : 0;
480 prot |= (epp->p_flags & PF_W) ? PROT_WRITE : 0;
481 prot |= (epp->p_flags & PF_X) ? PROT_EXEC : 0;
482 down_write(&current->mm->mmap_sem);
483 (void) do_mmap(fp, (epp->p_vaddr & 0xfffff000),
484 (epp->p_filesz + (epp->p_vaddr & 0xfff)),
485 prot, EXEC_MAP_FLAGS,
486 (epp->p_offset & 0xfffff000));
487 up_write(&current->mm->mmap_sem);
489 /* Fixup location tracking vars. */
490 if((epp->p_vaddr & 0xfffff000) < *estack)
491 *estack = (epp->p_vaddr & 0xfffff000);
492 if(!*laddr)
493 *laddr = epp->p_vaddr - epp->p_offset;
494 if(epp->p_vaddr < *scode)
495 *scode = epp->p_vaddr;
497 tmp = epp->p_vaddr + epp->p_filesz;
498 if(tmp > *ebss)
499 *ebss = tmp;
500 if((epp->p_flags & PF_X) && *ecode < tmp)
501 *ecode = tmp;
502 if(*edata < tmp)
503 *edata = tmp;
505 tmp = epp->p_vaddr + epp->p_memsz;
506 if(tmp > *ebrk)
507 *ebrk = tmp;
512 static inline int map_interpreter(struct elf_phdr *epp, struct elfhdr *ihp,
513 struct file *interp, unsigned int *iladdr,
514 int pnum, mm_segment_t old_fs,
515 unsigned int *eentry)
517 int i;
519 *eentry = 0xffffffff;
520 for(i = 0; i < pnum; i++, epp++) {
521 if(epp->p_type != PT_INTERP)
522 continue;
524 /* We should have fielded this error elsewhere... */
525 if(*eentry != 0xffffffff)
526 return -1;
528 set_fs(old_fs);
529 *eentry = load_irix_interp(ihp, interp, iladdr);
530 old_fs = get_fs();
531 set_fs(get_ds());
533 fput(interp);
535 if (*eentry == 0xffffffff)
536 return -1;
538 return 0;
542 * IRIX maps a page at 0x200000 that holds information about the
543 * process and the system, here we map the page and fill the
544 * structure
546 static void irix_map_prda_page(void)
548 unsigned long v;
549 struct prda *pp;
551 down_write(&current->mm->mmap_sem);
552 v = do_brk (PRDA_ADDRESS, PAGE_SIZE);
553 up_write(&current->mm->mmap_sem);
555 if (v < 0)
556 return;
558 pp = (struct prda *) v;
559 pp->prda_sys.t_pid = current->pid;
560 pp->prda_sys.t_prid = read_c0_prid();
561 pp->prda_sys.t_rpid = current->pid;
563 /* We leave the rest set to zero */
568 /* These are the functions used to load ELF style executables and shared
569 * libraries. There is no binary dependent code anywhere else.
571 static int load_irix_binary(struct linux_binprm * bprm, struct pt_regs * regs)
573 struct elfhdr elf_ex, interp_elf_ex;
574 struct file *interpreter;
575 struct elf_phdr *elf_phdata, *elf_ihdr, *elf_ephdr;
576 unsigned int load_addr, elf_bss, elf_brk;
577 unsigned int elf_entry, interp_load_addr = 0;
578 unsigned int start_code, end_code, end_data, elf_stack;
579 int retval, has_interp, has_ephdr, size, i;
580 char *elf_interpreter;
581 mm_segment_t old_fs;
583 load_addr = 0;
584 has_interp = has_ephdr = 0;
585 elf_ihdr = elf_ephdr = NULL;
586 elf_ex = *((struct elfhdr *) bprm->buf);
587 retval = -ENOEXEC;
589 if (verify_binary(&elf_ex, bprm))
590 goto out;
593 * Telling -o32 static binaries from Linux and Irix apart from each
594 * other is difficult. There are 2 differences to be noted for static
595 * binaries from the 2 operating systems:
597 * 1) Irix binaries have their .text section before their .init
598 * section. Linux binaries are just the opposite.
600 * 2) Irix binaries usually have <= 12 sections and Linux
601 * binaries have > 20.
603 * We will use Method #2 since Method #1 would require us to read in
604 * the section headers which is way too much overhead. This appears
605 * to work for everything we have ran into so far. If anyone has a
606 * better method to tell the binaries apart, I'm listening.
608 if (elf_ex.e_shnum > 20)
609 goto out;
611 #ifdef DEBUG
612 print_elfhdr(&elf_ex);
613 #endif
615 /* Now read in all of the header information */
616 size = elf_ex.e_phentsize * elf_ex.e_phnum;
617 if (size > 65536)
618 goto out;
619 elf_phdata = kmalloc(size, GFP_KERNEL);
620 if (elf_phdata == NULL) {
621 retval = -ENOMEM;
622 goto out;
625 retval = kernel_read(bprm->file, elf_ex.e_phoff, (char *)elf_phdata, size);
626 if (retval < 0)
627 goto out_free_ph;
629 #ifdef DEBUG
630 dump_phdrs(elf_phdata, elf_ex.e_phnum);
631 #endif
633 /* Set some things for later. */
634 for(i = 0; i < elf_ex.e_phnum; i++) {
635 switch(elf_phdata[i].p_type) {
636 case PT_INTERP:
637 has_interp = 1;
638 elf_ihdr = &elf_phdata[i];
639 break;
640 case PT_PHDR:
641 has_ephdr = 1;
642 elf_ephdr = &elf_phdata[i];
643 break;
647 pr_debug("\n");
649 elf_bss = 0;
650 elf_brk = 0;
652 elf_stack = 0xffffffff;
653 elf_interpreter = NULL;
654 start_code = 0xffffffff;
655 end_code = 0;
656 end_data = 0;
659 * If we get a return value, we change the value to be ENOEXEC
660 * so that we can exit gracefully and the main binary format
661 * search loop in 'fs/exec.c' will move onto the next handler
662 * which should be the normal ELF binary handler.
664 retval = look_for_irix_interpreter(&elf_interpreter, &interpreter,
665 &interp_elf_ex, elf_phdata, bprm,
666 elf_ex.e_phnum);
667 if (retval) {
668 retval = -ENOEXEC;
669 goto out_free_file;
672 if (elf_interpreter) {
673 retval = verify_irix_interpreter(&interp_elf_ex);
674 if(retval)
675 goto out_free_interp;
678 /* OK, we are done with that, now set up the arg stuff,
679 * and then start this sucker up.
681 retval = -E2BIG;
682 if (!bprm->sh_bang && !bprm->p)
683 goto out_free_interp;
685 /* Flush all traces of the currently running executable */
686 retval = flush_old_exec(bprm);
687 if (retval)
688 goto out_free_dentry;
690 /* OK, This is the point of no return */
691 current->mm->end_data = 0;
692 current->mm->end_code = 0;
693 current->mm->mmap = NULL;
694 current->flags &= ~PF_FORKNOEXEC;
695 elf_entry = (unsigned int) elf_ex.e_entry;
697 /* Do this so that we can load the interpreter, if need be. We will
698 * change some of these later.
700 setup_arg_pages(bprm, STACK_TOP, EXSTACK_DEFAULT);
701 current->mm->start_stack = bprm->p;
703 /* At this point, we assume that the image should be loaded at
704 * fixed address, not at a variable address.
706 old_fs = get_fs();
707 set_fs(get_ds());
709 map_executable(bprm->file, elf_phdata, elf_ex.e_phnum, &elf_stack,
710 &load_addr, &start_code, &elf_bss, &end_code,
711 &end_data, &elf_brk);
713 if(elf_interpreter) {
714 retval = map_interpreter(elf_phdata, &interp_elf_ex,
715 interpreter, &interp_load_addr,
716 elf_ex.e_phnum, old_fs, &elf_entry);
717 kfree(elf_interpreter);
718 if(retval) {
719 set_fs(old_fs);
720 printk("Unable to load IRIX ELF interpreter\n");
721 send_sig(SIGSEGV, current, 0);
722 retval = 0;
723 goto out_free_file;
727 set_fs(old_fs);
729 kfree(elf_phdata);
730 set_personality(PER_IRIX32);
731 set_binfmt(&irix_format);
732 compute_creds(bprm);
733 current->flags &= ~PF_FORKNOEXEC;
734 bprm->p = (unsigned long)
735 create_irix_tables((char *)bprm->p, bprm->argc, bprm->envc,
736 (elf_interpreter ? &elf_ex : NULL),
737 load_addr, interp_load_addr, regs, elf_ephdr);
738 current->mm->start_brk = current->mm->brk = elf_brk;
739 current->mm->end_code = end_code;
740 current->mm->start_code = start_code;
741 current->mm->end_data = end_data;
742 current->mm->start_stack = bprm->p;
744 /* Calling set_brk effectively mmaps the pages that we need for the
745 * bss and break sections.
747 set_brk(elf_bss, elf_brk);
750 * IRIX maps a page at 0x200000 which holds some system
751 * information. Programs depend on this.
753 irix_map_prda_page();
755 padzero(elf_bss);
757 pr_debug("(start_brk) %lx\n" , (long) current->mm->start_brk);
758 pr_debug("(end_code) %lx\n" , (long) current->mm->end_code);
759 pr_debug("(start_code) %lx\n" , (long) current->mm->start_code);
760 pr_debug("(end_data) %lx\n" , (long) current->mm->end_data);
761 pr_debug("(start_stack) %lx\n" , (long) current->mm->start_stack);
762 pr_debug("(brk) %lx\n" , (long) current->mm->brk);
764 #if 0 /* XXX No fucking way dude... */
765 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
766 * and some applications "depend" upon this behavior.
767 * Since we do not have the power to recompile these, we
768 * emulate the SVr4 behavior. Sigh.
770 down_write(&current->mm->mmap_sem);
771 (void) do_mmap(NULL, 0, 4096, PROT_READ | PROT_EXEC,
772 MAP_FIXED | MAP_PRIVATE, 0);
773 up_write(&current->mm->mmap_sem);
774 #endif
776 start_thread(regs, elf_entry, bprm->p);
777 if (current->ptrace & PT_PTRACED)
778 send_sig(SIGTRAP, current, 0);
779 return 0;
780 out:
781 return retval;
783 out_free_dentry:
784 allow_write_access(interpreter);
785 fput(interpreter);
786 out_free_interp:
787 kfree(elf_interpreter);
788 out_free_file:
789 out_free_ph:
790 kfree (elf_phdata);
791 goto out;
794 /* This is really simpleminded and specialized - we are loading an
795 * a.out library that is given an ELF header.
797 static int load_irix_library(struct file *file)
799 struct elfhdr elf_ex;
800 struct elf_phdr *elf_phdata = NULL;
801 unsigned int len = 0;
802 int elf_bss = 0;
803 int retval;
804 unsigned int bss;
805 int error;
806 int i,j, k;
808 error = kernel_read(file, 0, (char *) &elf_ex, sizeof(elf_ex));
809 if (error != sizeof(elf_ex))
810 return -ENOEXEC;
812 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
813 return -ENOEXEC;
815 /* First of all, some simple consistency checks. */
816 if(elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
817 !file->f_op->mmap)
818 return -ENOEXEC;
820 /* Now read in all of the header information. */
821 if(sizeof(struct elf_phdr) * elf_ex.e_phnum > PAGE_SIZE)
822 return -ENOEXEC;
824 elf_phdata = kmalloc(sizeof(struct elf_phdr) * elf_ex.e_phnum, GFP_KERNEL);
825 if (elf_phdata == NULL)
826 return -ENOMEM;
828 retval = kernel_read(file, elf_ex.e_phoff, (char *) elf_phdata,
829 sizeof(struct elf_phdr) * elf_ex.e_phnum);
831 j = 0;
832 for(i=0; i<elf_ex.e_phnum; i++)
833 if((elf_phdata + i)->p_type == PT_LOAD) j++;
835 if(j != 1) {
836 kfree(elf_phdata);
837 return -ENOEXEC;
840 while(elf_phdata->p_type != PT_LOAD) elf_phdata++;
842 /* Now use mmap to map the library into memory. */
843 down_write(&current->mm->mmap_sem);
844 error = do_mmap(file,
845 elf_phdata->p_vaddr & 0xfffff000,
846 elf_phdata->p_filesz + (elf_phdata->p_vaddr & 0xfff),
847 PROT_READ | PROT_WRITE | PROT_EXEC,
848 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
849 elf_phdata->p_offset & 0xfffff000);
850 up_write(&current->mm->mmap_sem);
852 k = elf_phdata->p_vaddr + elf_phdata->p_filesz;
853 if (k > elf_bss) elf_bss = k;
855 if (error != (elf_phdata->p_vaddr & 0xfffff000)) {
856 kfree(elf_phdata);
857 return error;
860 padzero(elf_bss);
862 len = (elf_phdata->p_filesz + elf_phdata->p_vaddr+ 0xfff) & 0xfffff000;
863 bss = elf_phdata->p_memsz + elf_phdata->p_vaddr;
864 if (bss > len) {
865 down_write(&current->mm->mmap_sem);
866 do_brk(len, bss-len);
867 up_write(&current->mm->mmap_sem);
869 kfree(elf_phdata);
870 return 0;
873 /* Called through irix_syssgi() to map an elf image given an FD,
874 * a phdr ptr USER_PHDRP in userspace, and a count CNT telling how many
875 * phdrs there are in the USER_PHDRP array. We return the vaddr the
876 * first phdr was successfully mapped to.
878 unsigned long irix_mapelf(int fd, struct elf_phdr __user *user_phdrp, int cnt)
880 unsigned long type, vaddr, filesz, offset, flags;
881 struct elf_phdr __user *hp;
882 struct file *filp;
883 int i, retval;
885 pr_debug("irix_mapelf: fd[%d] user_phdrp[%p] cnt[%d]\n",
886 fd, user_phdrp, cnt);
888 /* First get the verification out of the way. */
889 hp = user_phdrp;
890 if (!access_ok(VERIFY_READ, hp, (sizeof(struct elf_phdr) * cnt))) {
891 pr_debug("irix_mapelf: bad pointer to ELF PHDR!\n");
893 return -EFAULT;
896 #ifdef DEBUG
897 dump_phdrs(user_phdrp, cnt);
898 #endif
900 for (i = 0; i < cnt; i++, hp++) {
901 if (__get_user(type, &hp->p_type))
902 return -EFAULT;
903 if (type != PT_LOAD) {
904 printk("irix_mapelf: One section is not PT_LOAD!\n");
905 return -ENOEXEC;
909 filp = fget(fd);
910 if (!filp)
911 return -EACCES;
912 if(!filp->f_op) {
913 printk("irix_mapelf: Bogon filp!\n");
914 fput(filp);
915 return -EACCES;
918 hp = user_phdrp;
919 for(i = 0; i < cnt; i++, hp++) {
920 int prot;
922 retval = __get_user(vaddr, &hp->p_vaddr);
923 retval |= __get_user(filesz, &hp->p_filesz);
924 retval |= __get_user(offset, &hp->p_offset);
925 retval |= __get_user(flags, &hp->p_flags);
926 if (retval)
927 return retval;
929 prot = (flags & PF_R) ? PROT_READ : 0;
930 prot |= (flags & PF_W) ? PROT_WRITE : 0;
931 prot |= (flags & PF_X) ? PROT_EXEC : 0;
933 down_write(&current->mm->mmap_sem);
934 retval = do_mmap(filp, (vaddr & 0xfffff000),
935 (filesz + (vaddr & 0xfff)),
936 prot, (MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE),
937 (offset & 0xfffff000));
938 up_write(&current->mm->mmap_sem);
940 if (retval != (vaddr & 0xfffff000)) {
941 printk("irix_mapelf: do_mmap fails with %d!\n", retval);
942 fput(filp);
943 return retval;
947 pr_debug("irix_mapelf: Success, returning %08lx\n",
948 (unsigned long) user_phdrp->p_vaddr);
950 fput(filp);
952 if (__get_user(vaddr, &user_phdrp->p_vaddr))
953 return -EFAULT;
955 return vaddr;
959 * ELF core dumper
961 * Modelled on fs/exec.c:aout_core_dump()
962 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
965 /* These are the only things you should do on a core-file: use only these
966 * functions to write out all the necessary info.
968 static int dump_write(struct file *file, const void __user *addr, int nr)
970 return file->f_op->write(file, (const char __user *) addr, nr, &file->f_pos) == nr;
973 static int dump_seek(struct file *file, off_t off)
975 if (file->f_op->llseek) {
976 if (file->f_op->llseek(file, off, 0) != off)
977 return 0;
978 } else
979 file->f_pos = off;
980 return 1;
983 /* Decide whether a segment is worth dumping; default is yes to be
984 * sure (missing info is worse than too much; etc).
985 * Personally I'd include everything, and use the coredump limit...
987 * I think we should skip something. But I am not sure how. H.J.
989 static inline int maydump(struct vm_area_struct *vma)
991 if (!(vma->vm_flags & (VM_READ|VM_WRITE|VM_EXEC)))
992 return 0;
993 #if 1
994 if (vma->vm_flags & (VM_WRITE|VM_GROWSUP|VM_GROWSDOWN))
995 return 1;
996 if (vma->vm_flags & (VM_READ|VM_EXEC|VM_EXECUTABLE|VM_SHARED))
997 return 0;
998 #endif
999 return 1;
1002 /* An ELF note in memory. */
1003 struct memelfnote
1005 const char *name;
1006 int type;
1007 unsigned int datasz;
1008 void *data;
1011 static int notesize(struct memelfnote *en)
1013 int sz;
1015 sz = sizeof(struct elf_note);
1016 sz += roundup(strlen(en->name), 4);
1017 sz += roundup(en->datasz, 4);
1019 return sz;
1022 /* #define DEBUG */
1024 #define DUMP_WRITE(addr, nr) \
1025 if (!dump_write(file, (addr), (nr))) \
1026 goto end_coredump;
1027 #define DUMP_SEEK(off) \
1028 if (!dump_seek(file, (off))) \
1029 goto end_coredump;
1031 static int writenote(struct memelfnote *men, struct file *file)
1033 struct elf_note en;
1035 en.n_namesz = strlen(men->name);
1036 en.n_descsz = men->datasz;
1037 en.n_type = men->type;
1039 DUMP_WRITE(&en, sizeof(en));
1040 DUMP_WRITE(men->name, en.n_namesz);
1041 /* XXX - cast from long long to long to avoid need for libgcc.a */
1042 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1043 DUMP_WRITE(men->data, men->datasz);
1044 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1046 return 1;
1048 end_coredump:
1049 return 0;
1051 #undef DUMP_WRITE
1052 #undef DUMP_SEEK
1054 #define DUMP_WRITE(addr, nr) \
1055 if (!dump_write(file, (addr), (nr))) \
1056 goto end_coredump;
1057 #define DUMP_SEEK(off) \
1058 if (!dump_seek(file, (off))) \
1059 goto end_coredump;
1061 /* Actual dumper.
1063 * This is a two-pass process; first we find the offsets of the bits,
1064 * and then they are actually written out. If we run out of core limit
1065 * we just truncate.
1067 static int irix_core_dump(long signr, struct pt_regs * regs, struct file *file)
1069 int has_dumped = 0;
1070 mm_segment_t fs;
1071 int segs;
1072 int i;
1073 size_t size;
1074 struct vm_area_struct *vma;
1075 struct elfhdr elf;
1076 off_t offset = 0, dataoff;
1077 int limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1078 int numnote = 3;
1079 struct memelfnote notes[3];
1080 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1081 elf_fpregset_t fpu; /* NT_PRFPREG */
1082 struct elf_prpsinfo psinfo; /* NT_PRPSINFO */
1084 /* Count what's needed to dump, up to the limit of coredump size. */
1085 segs = 0;
1086 size = 0;
1087 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1088 if (maydump(vma))
1090 int sz = vma->vm_end-vma->vm_start;
1092 if (size+sz >= limit)
1093 break;
1094 else
1095 size += sz;
1098 segs++;
1100 #ifdef DEBUG
1101 printk("irix_core_dump: %d segs taking %d bytes\n", segs, size);
1102 #endif
1104 /* Set up header. */
1105 memcpy(elf.e_ident, ELFMAG, SELFMAG);
1106 elf.e_ident[EI_CLASS] = ELFCLASS32;
1107 elf.e_ident[EI_DATA] = ELFDATA2LSB;
1108 elf.e_ident[EI_VERSION] = EV_CURRENT;
1109 elf.e_ident[EI_OSABI] = ELF_OSABI;
1110 memset(elf.e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1112 elf.e_type = ET_CORE;
1113 elf.e_machine = ELF_ARCH;
1114 elf.e_version = EV_CURRENT;
1115 elf.e_entry = 0;
1116 elf.e_phoff = sizeof(elf);
1117 elf.e_shoff = 0;
1118 elf.e_flags = 0;
1119 elf.e_ehsize = sizeof(elf);
1120 elf.e_phentsize = sizeof(struct elf_phdr);
1121 elf.e_phnum = segs+1; /* Include notes. */
1122 elf.e_shentsize = 0;
1123 elf.e_shnum = 0;
1124 elf.e_shstrndx = 0;
1126 fs = get_fs();
1127 set_fs(KERNEL_DS);
1129 has_dumped = 1;
1130 current->flags |= PF_DUMPCORE;
1132 DUMP_WRITE(&elf, sizeof(elf));
1133 offset += sizeof(elf); /* Elf header. */
1134 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers. */
1136 /* Set up the notes in similar form to SVR4 core dumps made
1137 * with info from their /proc.
1139 memset(&psinfo, 0, sizeof(psinfo));
1140 memset(&prstatus, 0, sizeof(prstatus));
1142 notes[0].name = "CORE";
1143 notes[0].type = NT_PRSTATUS;
1144 notes[0].datasz = sizeof(prstatus);
1145 notes[0].data = &prstatus;
1146 prstatus.pr_info.si_signo = prstatus.pr_cursig = signr;
1147 prstatus.pr_sigpend = current->pending.signal.sig[0];
1148 prstatus.pr_sighold = current->blocked.sig[0];
1149 psinfo.pr_pid = prstatus.pr_pid = current->pid;
1150 psinfo.pr_ppid = prstatus.pr_ppid = current->parent->pid;
1151 psinfo.pr_pgrp = prstatus.pr_pgrp = process_group(current);
1152 psinfo.pr_sid = prstatus.pr_sid = current->signal->session;
1153 if (current->pid == current->tgid) {
1155 * This is the record for the group leader. Add in the
1156 * cumulative times of previous dead threads. This total
1157 * won't include the time of each live thread whose state
1158 * is included in the core dump. The final total reported
1159 * to our parent process when it calls wait4 will include
1160 * those sums as well as the little bit more time it takes
1161 * this and each other thread to finish dying after the
1162 * core dump synchronization phase.
1164 jiffies_to_timeval(current->utime + current->signal->utime,
1165 &prstatus.pr_utime);
1166 jiffies_to_timeval(current->stime + current->signal->stime,
1167 &prstatus.pr_stime);
1168 } else {
1169 jiffies_to_timeval(current->utime, &prstatus.pr_utime);
1170 jiffies_to_timeval(current->stime, &prstatus.pr_stime);
1172 jiffies_to_timeval(current->signal->cutime, &prstatus.pr_cutime);
1173 jiffies_to_timeval(current->signal->cstime, &prstatus.pr_cstime);
1175 if (sizeof(elf_gregset_t) != sizeof(struct pt_regs)) {
1176 printk("sizeof(elf_gregset_t) (%d) != sizeof(struct pt_regs) "
1177 "(%d)\n", sizeof(elf_gregset_t), sizeof(struct pt_regs));
1178 } else {
1179 *(struct pt_regs *)&prstatus.pr_reg = *regs;
1182 notes[1].name = "CORE";
1183 notes[1].type = NT_PRPSINFO;
1184 notes[1].datasz = sizeof(psinfo);
1185 notes[1].data = &psinfo;
1186 i = current->state ? ffz(~current->state) + 1 : 0;
1187 psinfo.pr_state = i;
1188 psinfo.pr_sname = (i < 0 || i > 5) ? '.' : "RSDZTD"[i];
1189 psinfo.pr_zomb = psinfo.pr_sname == 'Z';
1190 psinfo.pr_nice = task_nice(current);
1191 psinfo.pr_flag = current->flags;
1192 psinfo.pr_uid = current->uid;
1193 psinfo.pr_gid = current->gid;
1195 int i, len;
1197 set_fs(fs);
1199 len = current->mm->arg_end - current->mm->arg_start;
1200 len = len >= ELF_PRARGSZ ? ELF_PRARGSZ : len;
1201 (void *) copy_from_user(&psinfo.pr_psargs,
1202 (const char __user *)current->mm->arg_start, len);
1203 for (i = 0; i < len; i++)
1204 if (psinfo.pr_psargs[i] == 0)
1205 psinfo.pr_psargs[i] = ' ';
1206 psinfo.pr_psargs[len] = 0;
1208 set_fs(KERNEL_DS);
1210 strlcpy(psinfo.pr_fname, current->comm, sizeof(psinfo.pr_fname));
1212 /* Try to dump the FPU. */
1213 prstatus.pr_fpvalid = dump_fpu (regs, &fpu);
1214 if (!prstatus.pr_fpvalid) {
1215 numnote--;
1216 } else {
1217 notes[2].name = "CORE";
1218 notes[2].type = NT_PRFPREG;
1219 notes[2].datasz = sizeof(fpu);
1220 notes[2].data = &fpu;
1223 /* Write notes phdr entry. */
1225 struct elf_phdr phdr;
1226 int sz = 0;
1228 for(i = 0; i < numnote; i++)
1229 sz += notesize(&notes[i]);
1231 phdr.p_type = PT_NOTE;
1232 phdr.p_offset = offset;
1233 phdr.p_vaddr = 0;
1234 phdr.p_paddr = 0;
1235 phdr.p_filesz = sz;
1236 phdr.p_memsz = 0;
1237 phdr.p_flags = 0;
1238 phdr.p_align = 0;
1240 offset += phdr.p_filesz;
1241 DUMP_WRITE(&phdr, sizeof(phdr));
1244 /* Page-align dumped data. */
1245 dataoff = offset = roundup(offset, PAGE_SIZE);
1247 /* Write program headers for segments dump. */
1248 for(vma = current->mm->mmap, i = 0;
1249 i < segs && vma != NULL; vma = vma->vm_next) {
1250 struct elf_phdr phdr;
1251 size_t sz;
1253 i++;
1255 sz = vma->vm_end - vma->vm_start;
1257 phdr.p_type = PT_LOAD;
1258 phdr.p_offset = offset;
1259 phdr.p_vaddr = vma->vm_start;
1260 phdr.p_paddr = 0;
1261 phdr.p_filesz = maydump(vma) ? sz : 0;
1262 phdr.p_memsz = sz;
1263 offset += phdr.p_filesz;
1264 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1265 if (vma->vm_flags & VM_WRITE)
1266 phdr.p_flags |= PF_W;
1267 if (vma->vm_flags & VM_EXEC)
1268 phdr.p_flags |= PF_X;
1269 phdr.p_align = PAGE_SIZE;
1271 DUMP_WRITE(&phdr, sizeof(phdr));
1274 for(i = 0; i < numnote; i++)
1275 if (!writenote(&notes[i], file))
1276 goto end_coredump;
1278 set_fs(fs);
1280 DUMP_SEEK(dataoff);
1282 for(i = 0, vma = current->mm->mmap;
1283 i < segs && vma != NULL;
1284 vma = vma->vm_next) {
1285 unsigned long addr = vma->vm_start;
1286 unsigned long len = vma->vm_end - vma->vm_start;
1288 if (!maydump(vma))
1289 continue;
1290 i++;
1291 #ifdef DEBUG
1292 printk("elf_core_dump: writing %08lx %lx\n", addr, len);
1293 #endif
1294 DUMP_WRITE((void __user *)addr, len);
1297 if ((off_t) file->f_pos != offset) {
1298 /* Sanity check. */
1299 printk("elf_core_dump: file->f_pos (%ld) != offset (%ld)\n",
1300 (off_t) file->f_pos, offset);
1303 end_coredump:
1304 set_fs(fs);
1305 return has_dumped;
1308 static int __init init_irix_binfmt(void)
1310 extern int init_inventory(void);
1311 extern asmlinkage unsigned long sys_call_table;
1312 extern asmlinkage unsigned long sys_call_table_irix5;
1314 init_inventory();
1317 * Copy the IRIX5 syscall table (8000 bytes) into the main syscall
1318 * table. The IRIX5 calls are located by an offset of 8000 bytes
1319 * from the beginning of the main table.
1321 memcpy((void *) ((unsigned long) &sys_call_table + 8000),
1322 &sys_call_table_irix5, 8000);
1324 return register_binfmt(&irix_format);
1327 static void __exit exit_irix_binfmt(void)
1330 * Remove the Irix ELF loader.
1332 unregister_binfmt(&irix_format);
1335 module_init(init_irix_binfmt)
1336 module_exit(exit_irix_binfmt)