xref: /linux/fs/binfmt_elf_fdpic.c (revision 2b8232ce512105e28453f301d1510de8363bccd1)
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2  *
3  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  * Derived from binfmt_elf.c
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version
10  * 2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/module.h>
14 
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/elf.h>
34 #include <linux/elf-fdpic.h>
35 #include <linux/elfcore.h>
36 
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/pgalloc.h>
40 
41 typedef char *elf_caddr_t;
42 
43 #if 0
44 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
45 #else
46 #define kdebug(fmt, ...) do {} while(0)
47 #endif
48 
49 #if 0
50 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
51 #else
52 #define kdcore(fmt, ...) do {} while(0)
53 #endif
54 
55 MODULE_LICENSE("GPL");
56 
57 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
58 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
59 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
60 			      struct mm_struct *, const char *);
61 
62 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
63 				   struct elf_fdpic_params *,
64 				   struct elf_fdpic_params *);
65 
66 #ifndef CONFIG_MMU
67 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
68 					    unsigned long *);
69 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
70 						   struct file *,
71 						   struct mm_struct *);
72 #endif
73 
74 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
75 					     struct file *, struct mm_struct *);
76 
77 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
78 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *);
79 #endif
80 
81 static struct linux_binfmt elf_fdpic_format = {
82 	.module		= THIS_MODULE,
83 	.load_binary	= load_elf_fdpic_binary,
84 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
85 	.core_dump	= elf_fdpic_core_dump,
86 #endif
87 	.min_coredump	= ELF_EXEC_PAGESIZE,
88 };
89 
90 static int __init init_elf_fdpic_binfmt(void)
91 {
92 	return register_binfmt(&elf_fdpic_format);
93 }
94 
95 static void __exit exit_elf_fdpic_binfmt(void)
96 {
97 	unregister_binfmt(&elf_fdpic_format);
98 }
99 
100 core_initcall(init_elf_fdpic_binfmt);
101 module_exit(exit_elf_fdpic_binfmt);
102 
103 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
104 {
105 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
106 		return 0;
107 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
108 		return 0;
109 	if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
110 		return 0;
111 	if (!file->f_op || !file->f_op->mmap)
112 		return 0;
113 	return 1;
114 }
115 
116 /*****************************************************************************/
117 /*
118  * read the program headers table into memory
119  */
120 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
121 				 struct file *file)
122 {
123 	struct elf32_phdr *phdr;
124 	unsigned long size;
125 	int retval, loop;
126 
127 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
128 		return -ENOMEM;
129 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
130 		return -ENOMEM;
131 
132 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
133 	params->phdrs = kmalloc(size, GFP_KERNEL);
134 	if (!params->phdrs)
135 		return -ENOMEM;
136 
137 	retval = kernel_read(file, params->hdr.e_phoff,
138 			     (char *) params->phdrs, size);
139 	if (retval < 0)
140 		return retval;
141 
142 	/* determine stack size for this binary */
143 	phdr = params->phdrs;
144 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
145 		if (phdr->p_type != PT_GNU_STACK)
146 			continue;
147 
148 		if (phdr->p_flags & PF_X)
149 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
150 		else
151 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
152 
153 		params->stack_size = phdr->p_memsz;
154 		break;
155 	}
156 
157 	return 0;
158 }
159 
160 /*****************************************************************************/
161 /*
162  * load an fdpic binary into various bits of memory
163  */
164 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
165 				 struct pt_regs *regs)
166 {
167 	struct elf_fdpic_params exec_params, interp_params;
168 	struct elf_phdr *phdr;
169 	unsigned long stack_size, entryaddr;
170 #ifndef CONFIG_MMU
171 	unsigned long fullsize;
172 #endif
173 #ifdef ELF_FDPIC_PLAT_INIT
174 	unsigned long dynaddr;
175 #endif
176 	struct file *interpreter = NULL; /* to shut gcc up */
177 	char *interpreter_name = NULL;
178 	int executable_stack;
179 	int retval, i;
180 
181 	kdebug("____ LOAD %d ____", current->pid);
182 
183 	memset(&exec_params, 0, sizeof(exec_params));
184 	memset(&interp_params, 0, sizeof(interp_params));
185 
186 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
187 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
188 
189 	/* check that this is a binary we know how to deal with */
190 	retval = -ENOEXEC;
191 	if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
192 		goto error;
193 
194 	/* read the program header table */
195 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
196 	if (retval < 0)
197 		goto error;
198 
199 	/* scan for a program header that specifies an interpreter */
200 	phdr = exec_params.phdrs;
201 
202 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
203 		switch (phdr->p_type) {
204 		case PT_INTERP:
205 			retval = -ENOMEM;
206 			if (phdr->p_filesz > PATH_MAX)
207 				goto error;
208 			retval = -ENOENT;
209 			if (phdr->p_filesz < 2)
210 				goto error;
211 
212 			/* read the name of the interpreter into memory */
213 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
214 			if (!interpreter_name)
215 				goto error;
216 
217 			retval = kernel_read(bprm->file,
218 					     phdr->p_offset,
219 					     interpreter_name,
220 					     phdr->p_filesz);
221 			if (retval < 0)
222 				goto error;
223 
224 			retval = -ENOENT;
225 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
226 				goto error;
227 
228 			kdebug("Using ELF interpreter %s", interpreter_name);
229 
230 			/* replace the program with the interpreter */
231 			interpreter = open_exec(interpreter_name);
232 			retval = PTR_ERR(interpreter);
233 			if (IS_ERR(interpreter)) {
234 				interpreter = NULL;
235 				goto error;
236 			}
237 
238 			/*
239 			 * If the binary is not readable then enforce
240 			 * mm->dumpable = 0 regardless of the interpreter's
241 			 * permissions.
242 			 */
243 			if (file_permission(interpreter, MAY_READ) < 0)
244 				bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
245 
246 			retval = kernel_read(interpreter, 0, bprm->buf,
247 					     BINPRM_BUF_SIZE);
248 			if (retval < 0)
249 				goto error;
250 
251 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
252 			break;
253 
254 		case PT_LOAD:
255 #ifdef CONFIG_MMU
256 			if (exec_params.load_addr == 0)
257 				exec_params.load_addr = phdr->p_vaddr;
258 #endif
259 			break;
260 		}
261 
262 	}
263 
264 	if (elf_check_const_displacement(&exec_params.hdr))
265 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
266 
267 	/* perform insanity checks on the interpreter */
268 	if (interpreter_name) {
269 		retval = -ELIBBAD;
270 		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
271 			goto error;
272 
273 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
274 
275 		/* read the interpreter's program header table */
276 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
277 		if (retval < 0)
278 			goto error;
279 	}
280 
281 	stack_size = exec_params.stack_size;
282 	if (stack_size < interp_params.stack_size)
283 		stack_size = interp_params.stack_size;
284 
285 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
286 		executable_stack = EXSTACK_ENABLE_X;
287 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
288 		executable_stack = EXSTACK_DISABLE_X;
289 	else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
290 		executable_stack = EXSTACK_ENABLE_X;
291 	else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
292 		executable_stack = EXSTACK_DISABLE_X;
293 	else
294 		executable_stack = EXSTACK_DEFAULT;
295 
296 	retval = -ENOEXEC;
297 	if (stack_size == 0)
298 		goto error;
299 
300 	if (elf_check_const_displacement(&interp_params.hdr))
301 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
302 
303 	/* flush all traces of the currently running executable */
304 	retval = flush_old_exec(bprm);
305 	if (retval)
306 		goto error;
307 
308 	/* there's now no turning back... the old userspace image is dead,
309 	 * defunct, deceased, etc. after this point we have to exit via
310 	 * error_kill */
311 	set_personality(PER_LINUX_FDPIC);
312 	set_binfmt(&elf_fdpic_format);
313 
314 	current->mm->start_code = 0;
315 	current->mm->end_code = 0;
316 	current->mm->start_stack = 0;
317 	current->mm->start_data = 0;
318 	current->mm->end_data = 0;
319 	current->mm->context.exec_fdpic_loadmap = 0;
320 	current->mm->context.interp_fdpic_loadmap = 0;
321 
322 	current->flags &= ~PF_FORKNOEXEC;
323 
324 #ifdef CONFIG_MMU
325 	elf_fdpic_arch_lay_out_mm(&exec_params,
326 				  &interp_params,
327 				  &current->mm->start_stack,
328 				  &current->mm->start_brk);
329 
330 	retval = setup_arg_pages(bprm, current->mm->start_stack,
331 				 executable_stack);
332 	if (retval < 0) {
333 		send_sig(SIGKILL, current, 0);
334 		goto error_kill;
335 	}
336 #endif
337 
338 	/* load the executable and interpreter into memory */
339 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
340 				    "executable");
341 	if (retval < 0)
342 		goto error_kill;
343 
344 	if (interpreter_name) {
345 		retval = elf_fdpic_map_file(&interp_params, interpreter,
346 					    current->mm, "interpreter");
347 		if (retval < 0) {
348 			printk(KERN_ERR "Unable to load interpreter\n");
349 			goto error_kill;
350 		}
351 
352 		allow_write_access(interpreter);
353 		fput(interpreter);
354 		interpreter = NULL;
355 	}
356 
357 #ifdef CONFIG_MMU
358 	if (!current->mm->start_brk)
359 		current->mm->start_brk = current->mm->end_data;
360 
361 	current->mm->brk = current->mm->start_brk =
362 		PAGE_ALIGN(current->mm->start_brk);
363 
364 #else
365 	/* create a stack and brk area big enough for everyone
366 	 * - the brk heap starts at the bottom and works up
367 	 * - the stack starts at the top and works down
368 	 */
369 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
370 	if (stack_size < PAGE_SIZE * 2)
371 		stack_size = PAGE_SIZE * 2;
372 
373 	down_write(&current->mm->mmap_sem);
374 	current->mm->start_brk = do_mmap(NULL, 0, stack_size,
375 					 PROT_READ | PROT_WRITE | PROT_EXEC,
376 					 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
377 					 0);
378 
379 	if (IS_ERR_VALUE(current->mm->start_brk)) {
380 		up_write(&current->mm->mmap_sem);
381 		retval = current->mm->start_brk;
382 		current->mm->start_brk = 0;
383 		goto error_kill;
384 	}
385 
386 	/* expand the stack mapping to use up the entire allocation granule */
387 	fullsize = ksize((char *) current->mm->start_brk);
388 	if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
389 				    fullsize, 0, 0)))
390 		stack_size = fullsize;
391 	up_write(&current->mm->mmap_sem);
392 
393 	current->mm->brk = current->mm->start_brk;
394 	current->mm->context.end_brk = current->mm->start_brk;
395 	current->mm->context.end_brk +=
396 		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
397 	current->mm->start_stack = current->mm->start_brk + stack_size;
398 #endif
399 
400 	compute_creds(bprm);
401 	current->flags &= ~PF_FORKNOEXEC;
402 	if (create_elf_fdpic_tables(bprm, current->mm,
403 				    &exec_params, &interp_params) < 0)
404 		goto error_kill;
405 
406 	kdebug("- start_code  %lx", current->mm->start_code);
407 	kdebug("- end_code    %lx", current->mm->end_code);
408 	kdebug("- start_data  %lx", current->mm->start_data);
409 	kdebug("- end_data    %lx", current->mm->end_data);
410 	kdebug("- start_brk   %lx", current->mm->start_brk);
411 	kdebug("- brk         %lx", current->mm->brk);
412 	kdebug("- start_stack %lx", current->mm->start_stack);
413 
414 #ifdef ELF_FDPIC_PLAT_INIT
415 	/*
416 	 * The ABI may specify that certain registers be set up in special
417 	 * ways (on i386 %edx is the address of a DT_FINI function, for
418 	 * example.  This macro performs whatever initialization to
419 	 * the regs structure is required.
420 	 */
421 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
422 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
423 			    dynaddr);
424 #endif
425 
426 	/* everything is now ready... get the userspace context ready to roll */
427 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
428 	start_thread(regs, entryaddr, current->mm->start_stack);
429 
430 	if (unlikely(current->ptrace & PT_PTRACED)) {
431 		if (current->ptrace & PT_TRACE_EXEC)
432 			ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
433 		else
434 			send_sig(SIGTRAP, current, 0);
435 	}
436 
437 	retval = 0;
438 
439 error:
440 	if (interpreter) {
441 		allow_write_access(interpreter);
442 		fput(interpreter);
443 	}
444 	kfree(interpreter_name);
445 	kfree(exec_params.phdrs);
446 	kfree(exec_params.loadmap);
447 	kfree(interp_params.phdrs);
448 	kfree(interp_params.loadmap);
449 	return retval;
450 
451 	/* unrecoverable error - kill the process */
452 error_kill:
453 	send_sig(SIGSEGV, current, 0);
454 	goto error;
455 
456 }
457 
458 /*****************************************************************************/
459 /*
460  * present useful information to the program
461  */
462 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
463 				   struct mm_struct *mm,
464 				   struct elf_fdpic_params *exec_params,
465 				   struct elf_fdpic_params *interp_params)
466 {
467 	unsigned long sp, csp, nitems;
468 	elf_caddr_t __user *argv, *envp;
469 	size_t platform_len = 0, len;
470 	char *k_platform;
471 	char __user *u_platform, *p;
472 	long hwcap;
473 	int loop;
474 
475 	/* we're going to shovel a whole load of stuff onto the stack */
476 #ifdef CONFIG_MMU
477 	sp = bprm->p;
478 #else
479 	sp = mm->start_stack;
480 
481 	/* stack the program arguments and environment */
482 	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
483 		return -EFAULT;
484 #endif
485 
486 	/* get hold of platform and hardware capabilities masks for the machine
487 	 * we are running on.  In some cases (Sparc), this info is impossible
488 	 * to get, in others (i386) it is merely difficult.
489 	 */
490 	hwcap = ELF_HWCAP;
491 	k_platform = ELF_PLATFORM;
492 	u_platform = NULL;
493 
494 	if (k_platform) {
495 		platform_len = strlen(k_platform) + 1;
496 		sp -= platform_len;
497 		u_platform = (char __user *) sp;
498 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
499 			return -EFAULT;
500 	}
501 
502 #if defined(__i386__) && defined(CONFIG_SMP)
503 	/* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
504 	 * by the processes running on the same package. One thing we can do is
505 	 * to shuffle the initial stack for them.
506 	 *
507 	 * the conditionals here are unneeded, but kept in to make the code
508 	 * behaviour the same as pre change unless we have hyperthreaded
509 	 * processors. This keeps Mr Marcelo Person happier but should be
510 	 * removed for 2.5
511 	 */
512 	if (smp_num_siblings > 1)
513 		sp = sp - ((current->pid % 64) << 7);
514 #endif
515 
516 	sp &= ~7UL;
517 
518 	/* stack the load map(s) */
519 	len = sizeof(struct elf32_fdpic_loadmap);
520 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
521 	sp = (sp - len) & ~7UL;
522 	exec_params->map_addr = sp;
523 
524 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
525 		return -EFAULT;
526 
527 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
528 
529 	if (interp_params->loadmap) {
530 		len = sizeof(struct elf32_fdpic_loadmap);
531 		len += sizeof(struct elf32_fdpic_loadseg) *
532 			interp_params->loadmap->nsegs;
533 		sp = (sp - len) & ~7UL;
534 		interp_params->map_addr = sp;
535 
536 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
537 				 len) != 0)
538 			return -EFAULT;
539 
540 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
541 	}
542 
543 	/* force 16 byte _final_ alignment here for generality */
544 #define DLINFO_ITEMS 13
545 
546 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0);
547 #ifdef DLINFO_ARCH_ITEMS
548 	nitems += DLINFO_ARCH_ITEMS;
549 #endif
550 
551 	csp = sp;
552 	sp -= nitems * 2 * sizeof(unsigned long);
553 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
554 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
555 	sp -= 1 * sizeof(unsigned long);		/* argc */
556 
557 	csp -= sp & 15UL;
558 	sp -= sp & 15UL;
559 
560 	/* put the ELF interpreter info on the stack */
561 #define NEW_AUX_ENT(nr, id, val)					\
562 	do {								\
563 		struct { unsigned long _id, _val; } __user *ent;	\
564 									\
565 		ent = (void __user *) csp;				\
566 		__put_user((id), &ent[nr]._id);				\
567 		__put_user((val), &ent[nr]._val);			\
568 	} while (0)
569 
570 	csp -= 2 * sizeof(unsigned long);
571 	NEW_AUX_ENT(0, AT_NULL, 0);
572 	if (k_platform) {
573 		csp -= 2 * sizeof(unsigned long);
574 		NEW_AUX_ENT(0, AT_PLATFORM,
575 			    (elf_addr_t) (unsigned long) u_platform);
576 	}
577 
578 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
579 	NEW_AUX_ENT( 0, AT_HWCAP,	hwcap);
580 	NEW_AUX_ENT( 1, AT_PAGESZ,	PAGE_SIZE);
581 	NEW_AUX_ENT( 2, AT_CLKTCK,	CLOCKS_PER_SEC);
582 	NEW_AUX_ENT( 3, AT_PHDR,	exec_params->ph_addr);
583 	NEW_AUX_ENT( 4, AT_PHENT,	sizeof(struct elf_phdr));
584 	NEW_AUX_ENT( 5, AT_PHNUM,	exec_params->hdr.e_phnum);
585 	NEW_AUX_ENT( 6,	AT_BASE,	interp_params->elfhdr_addr);
586 	NEW_AUX_ENT( 7, AT_FLAGS,	0);
587 	NEW_AUX_ENT( 8, AT_ENTRY,	exec_params->entry_addr);
588 	NEW_AUX_ENT( 9, AT_UID,		(elf_addr_t) current->uid);
589 	NEW_AUX_ENT(10, AT_EUID,	(elf_addr_t) current->euid);
590 	NEW_AUX_ENT(11, AT_GID,		(elf_addr_t) current->gid);
591 	NEW_AUX_ENT(12, AT_EGID,	(elf_addr_t) current->egid);
592 
593 #ifdef ARCH_DLINFO
594 	/* ARCH_DLINFO must come last so platform specific code can enforce
595 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
596 	 */
597 	ARCH_DLINFO;
598 #endif
599 #undef NEW_AUX_ENT
600 
601 	/* allocate room for argv[] and envv[] */
602 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
603 	envp = (elf_caddr_t __user *) csp;
604 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
605 	argv = (elf_caddr_t __user *) csp;
606 
607 	/* stack argc */
608 	csp -= sizeof(unsigned long);
609 	__put_user(bprm->argc, (unsigned long __user *) csp);
610 
611 	BUG_ON(csp != sp);
612 
613 	/* fill in the argv[] array */
614 #ifdef CONFIG_MMU
615 	current->mm->arg_start = bprm->p;
616 #else
617 	current->mm->arg_start = current->mm->start_stack -
618 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
619 #endif
620 
621 	p = (char __user *) current->mm->arg_start;
622 	for (loop = bprm->argc; loop > 0; loop--) {
623 		__put_user((elf_caddr_t) p, argv++);
624 		len = strnlen_user(p, MAX_ARG_STRLEN);
625 		if (!len || len > MAX_ARG_STRLEN)
626 			return -EINVAL;
627 		p += len;
628 	}
629 	__put_user(NULL, argv);
630 	current->mm->arg_end = (unsigned long) p;
631 
632 	/* fill in the envv[] array */
633 	current->mm->env_start = (unsigned long) p;
634 	for (loop = bprm->envc; loop > 0; loop--) {
635 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
636 		len = strnlen_user(p, MAX_ARG_STRLEN);
637 		if (!len || len > MAX_ARG_STRLEN)
638 			return -EINVAL;
639 		p += len;
640 	}
641 	__put_user(NULL, envp);
642 	current->mm->env_end = (unsigned long) p;
643 
644 	mm->start_stack = (unsigned long) sp;
645 	return 0;
646 }
647 
648 /*****************************************************************************/
649 /*
650  * transfer the program arguments and environment from the holding pages onto
651  * the stack
652  */
653 #ifndef CONFIG_MMU
654 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
655 					    unsigned long *_sp)
656 {
657 	unsigned long index, stop, sp;
658 	char *src;
659 	int ret = 0;
660 
661 	stop = bprm->p >> PAGE_SHIFT;
662 	sp = *_sp;
663 
664 	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
665 		src = kmap(bprm->page[index]);
666 		sp -= PAGE_SIZE;
667 		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
668 			ret = -EFAULT;
669 		kunmap(bprm->page[index]);
670 		if (ret < 0)
671 			goto out;
672 	}
673 
674 	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
675 
676 out:
677 	return ret;
678 }
679 #endif
680 
681 /*****************************************************************************/
682 /*
683  * load the appropriate binary image (executable or interpreter) into memory
684  * - we assume no MMU is available
685  * - if no other PIC bits are set in params->hdr->e_flags
686  *   - we assume that the LOADable segments in the binary are independently relocatable
687  *   - we assume R/O executable segments are shareable
688  * - else
689  *   - we assume the loadable parts of the image to require fixed displacement
690  *   - the image is not shareable
691  */
692 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
693 			      struct file *file,
694 			      struct mm_struct *mm,
695 			      const char *what)
696 {
697 	struct elf32_fdpic_loadmap *loadmap;
698 #ifdef CONFIG_MMU
699 	struct elf32_fdpic_loadseg *mseg;
700 #endif
701 	struct elf32_fdpic_loadseg *seg;
702 	struct elf32_phdr *phdr;
703 	unsigned long load_addr, stop;
704 	unsigned nloads, tmp;
705 	size_t size;
706 	int loop, ret;
707 
708 	/* allocate a load map table */
709 	nloads = 0;
710 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
711 		if (params->phdrs[loop].p_type == PT_LOAD)
712 			nloads++;
713 
714 	if (nloads == 0)
715 		return -ELIBBAD;
716 
717 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
718 	loadmap = kzalloc(size, GFP_KERNEL);
719 	if (!loadmap)
720 		return -ENOMEM;
721 
722 	params->loadmap = loadmap;
723 
724 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
725 	loadmap->nsegs = nloads;
726 
727 	load_addr = params->load_addr;
728 	seg = loadmap->segs;
729 
730 	/* map the requested LOADs into the memory space */
731 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
732 	case ELF_FDPIC_FLAG_CONSTDISP:
733 	case ELF_FDPIC_FLAG_CONTIGUOUS:
734 #ifndef CONFIG_MMU
735 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
736 		if (ret < 0)
737 			return ret;
738 		break;
739 #endif
740 	default:
741 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
742 		if (ret < 0)
743 			return ret;
744 		break;
745 	}
746 
747 	/* map the entry point */
748 	if (params->hdr.e_entry) {
749 		seg = loadmap->segs;
750 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
751 			if (params->hdr.e_entry >= seg->p_vaddr &&
752 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
753 				params->entry_addr =
754 					(params->hdr.e_entry - seg->p_vaddr) +
755 					seg->addr;
756 				break;
757 			}
758 		}
759 	}
760 
761 	/* determine where the program header table has wound up if mapped */
762 	stop = params->hdr.e_phoff;
763 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
764 	phdr = params->phdrs;
765 
766 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
767 		if (phdr->p_type != PT_LOAD)
768 			continue;
769 
770 		if (phdr->p_offset > params->hdr.e_phoff ||
771 		    phdr->p_offset + phdr->p_filesz < stop)
772 			continue;
773 
774 		seg = loadmap->segs;
775 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
776 			if (phdr->p_vaddr >= seg->p_vaddr &&
777 			    phdr->p_vaddr + phdr->p_filesz <=
778 			    seg->p_vaddr + seg->p_memsz) {
779 				params->ph_addr =
780 					(phdr->p_vaddr - seg->p_vaddr) +
781 					seg->addr +
782 					params->hdr.e_phoff - phdr->p_offset;
783 				break;
784 			}
785 		}
786 		break;
787 	}
788 
789 	/* determine where the dynamic section has wound up if there is one */
790 	phdr = params->phdrs;
791 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
792 		if (phdr->p_type != PT_DYNAMIC)
793 			continue;
794 
795 		seg = loadmap->segs;
796 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
797 			if (phdr->p_vaddr >= seg->p_vaddr &&
798 			    phdr->p_vaddr + phdr->p_memsz <=
799 			    seg->p_vaddr + seg->p_memsz) {
800 				params->dynamic_addr =
801 					(phdr->p_vaddr - seg->p_vaddr) +
802 					seg->addr;
803 
804 				/* check the dynamic section contains at least
805 				 * one item, and that the last item is a NULL
806 				 * entry */
807 				if (phdr->p_memsz == 0 ||
808 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
809 					goto dynamic_error;
810 
811 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
812 				if (((Elf32_Dyn *)
813 				     params->dynamic_addr)[tmp - 1].d_tag != 0)
814 					goto dynamic_error;
815 				break;
816 			}
817 		}
818 		break;
819 	}
820 
821 	/* now elide adjacent segments in the load map on MMU linux
822 	 * - on uClinux the holes between may actually be filled with system
823 	 *   stuff or stuff from other processes
824 	 */
825 #ifdef CONFIG_MMU
826 	nloads = loadmap->nsegs;
827 	mseg = loadmap->segs;
828 	seg = mseg + 1;
829 	for (loop = 1; loop < nloads; loop++) {
830 		/* see if we have a candidate for merging */
831 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
832 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
833 			if (load_addr == (seg->addr & PAGE_MASK)) {
834 				mseg->p_memsz +=
835 					load_addr -
836 					(mseg->addr + mseg->p_memsz);
837 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
838 				mseg->p_memsz += seg->p_memsz;
839 				loadmap->nsegs--;
840 				continue;
841 			}
842 		}
843 
844 		mseg++;
845 		if (mseg != seg)
846 			*mseg = *seg;
847 	}
848 #endif
849 
850 	kdebug("Mapped Object [%s]:", what);
851 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
852 	kdebug("- entry    : %lx", params->entry_addr);
853 	kdebug("- PHDR[]   : %lx", params->ph_addr);
854 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
855 	seg = loadmap->segs;
856 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
857 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
858 		       loop,
859 		       seg->addr, seg->addr + seg->p_memsz - 1,
860 		       seg->p_vaddr, seg->p_memsz);
861 
862 	return 0;
863 
864 dynamic_error:
865 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
866 	       what, file->f_path.dentry->d_inode->i_ino);
867 	return -ELIBBAD;
868 }
869 
870 /*****************************************************************************/
871 /*
872  * map a file with constant displacement under uClinux
873  */
874 #ifndef CONFIG_MMU
875 static int elf_fdpic_map_file_constdisp_on_uclinux(
876 	struct elf_fdpic_params *params,
877 	struct file *file,
878 	struct mm_struct *mm)
879 {
880 	struct elf32_fdpic_loadseg *seg;
881 	struct elf32_phdr *phdr;
882 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
883 	loff_t fpos;
884 	int loop, ret;
885 
886 	load_addr = params->load_addr;
887 	seg = params->loadmap->segs;
888 
889 	/* determine the bounds of the contiguous overall allocation we must
890 	 * make */
891 	phdr = params->phdrs;
892 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
893 		if (params->phdrs[loop].p_type != PT_LOAD)
894 			continue;
895 
896 		if (base > phdr->p_vaddr)
897 			base = phdr->p_vaddr;
898 		if (top < phdr->p_vaddr + phdr->p_memsz)
899 			top = phdr->p_vaddr + phdr->p_memsz;
900 	}
901 
902 	/* allocate one big anon block for everything */
903 	mflags = MAP_PRIVATE;
904 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
905 		mflags |= MAP_EXECUTABLE;
906 
907 	down_write(&mm->mmap_sem);
908 	maddr = do_mmap(NULL, load_addr, top - base,
909 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
910 	up_write(&mm->mmap_sem);
911 	if (IS_ERR_VALUE(maddr))
912 		return (int) maddr;
913 
914 	if (load_addr != 0)
915 		load_addr += PAGE_ALIGN(top - base);
916 
917 	/* and then load the file segments into it */
918 	phdr = params->phdrs;
919 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
920 		if (params->phdrs[loop].p_type != PT_LOAD)
921 			continue;
922 
923 		fpos = phdr->p_offset;
924 
925 		seg->addr = maddr + (phdr->p_vaddr - base);
926 		seg->p_vaddr = phdr->p_vaddr;
927 		seg->p_memsz = phdr->p_memsz;
928 
929 		ret = file->f_op->read(file, (void *) seg->addr,
930 				       phdr->p_filesz, &fpos);
931 		if (ret < 0)
932 			return ret;
933 
934 		/* map the ELF header address if in this segment */
935 		if (phdr->p_offset == 0)
936 			params->elfhdr_addr = seg->addr;
937 
938 		/* clear any space allocated but not loaded */
939 		if (phdr->p_filesz < phdr->p_memsz)
940 			clear_user((void *) (seg->addr + phdr->p_filesz),
941 				   phdr->p_memsz - phdr->p_filesz);
942 
943 		if (mm) {
944 			if (phdr->p_flags & PF_X) {
945 				if (!mm->start_code) {
946 					mm->start_code = seg->addr;
947 					mm->end_code = seg->addr +
948 						phdr->p_memsz;
949 				}
950 			} else if (!mm->start_data) {
951 				mm->start_data = seg->addr;
952 #ifndef CONFIG_MMU
953 				mm->end_data = seg->addr + phdr->p_memsz;
954 #endif
955 			}
956 
957 #ifdef CONFIG_MMU
958 			if (seg->addr + phdr->p_memsz > mm->end_data)
959 				mm->end_data = seg->addr + phdr->p_memsz;
960 #endif
961 		}
962 
963 		seg++;
964 	}
965 
966 	return 0;
967 }
968 #endif
969 
970 /*****************************************************************************/
971 /*
972  * map a binary by direct mmap() of the individual PT_LOAD segments
973  */
974 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
975 					     struct file *file,
976 					     struct mm_struct *mm)
977 {
978 	struct elf32_fdpic_loadseg *seg;
979 	struct elf32_phdr *phdr;
980 	unsigned long load_addr, delta_vaddr;
981 	int loop, dvset;
982 
983 	load_addr = params->load_addr;
984 	delta_vaddr = 0;
985 	dvset = 0;
986 
987 	seg = params->loadmap->segs;
988 
989 	/* deal with each load segment separately */
990 	phdr = params->phdrs;
991 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
992 		unsigned long maddr, disp, excess, excess1;
993 		int prot = 0, flags;
994 
995 		if (phdr->p_type != PT_LOAD)
996 			continue;
997 
998 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
999 		       (unsigned long) phdr->p_vaddr,
1000 		       (unsigned long) phdr->p_offset,
1001 		       (unsigned long) phdr->p_filesz,
1002 		       (unsigned long) phdr->p_memsz);
1003 
1004 		/* determine the mapping parameters */
1005 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1006 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1007 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1008 
1009 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1010 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1011 			flags |= MAP_EXECUTABLE;
1012 
1013 		maddr = 0;
1014 
1015 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1016 		case ELF_FDPIC_FLAG_INDEPENDENT:
1017 			/* PT_LOADs are independently locatable */
1018 			break;
1019 
1020 		case ELF_FDPIC_FLAG_HONOURVADDR:
1021 			/* the specified virtual address must be honoured */
1022 			maddr = phdr->p_vaddr;
1023 			flags |= MAP_FIXED;
1024 			break;
1025 
1026 		case ELF_FDPIC_FLAG_CONSTDISP:
1027 			/* constant displacement
1028 			 * - can be mapped anywhere, but must be mapped as a
1029 			 *   unit
1030 			 */
1031 			if (!dvset) {
1032 				maddr = load_addr;
1033 				delta_vaddr = phdr->p_vaddr;
1034 				dvset = 1;
1035 			} else {
1036 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1037 				flags |= MAP_FIXED;
1038 			}
1039 			break;
1040 
1041 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1042 			/* contiguity handled later */
1043 			break;
1044 
1045 		default:
1046 			BUG();
1047 		}
1048 
1049 		maddr &= PAGE_MASK;
1050 
1051 		/* create the mapping */
1052 		disp = phdr->p_vaddr & ~PAGE_MASK;
1053 		down_write(&mm->mmap_sem);
1054 		maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1055 				phdr->p_offset - disp);
1056 		up_write(&mm->mmap_sem);
1057 
1058 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1059 		       loop, phdr->p_memsz + disp, prot, flags,
1060 		       phdr->p_offset - disp, maddr);
1061 
1062 		if (IS_ERR_VALUE(maddr))
1063 			return (int) maddr;
1064 
1065 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1066 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1067 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1068 
1069 		seg->addr = maddr + disp;
1070 		seg->p_vaddr = phdr->p_vaddr;
1071 		seg->p_memsz = phdr->p_memsz;
1072 
1073 		/* map the ELF header address if in this segment */
1074 		if (phdr->p_offset == 0)
1075 			params->elfhdr_addr = seg->addr;
1076 
1077 		/* clear the bit between beginning of mapping and beginning of
1078 		 * PT_LOAD */
1079 		if (prot & PROT_WRITE && disp > 0) {
1080 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1081 			clear_user((void __user *) maddr, disp);
1082 			maddr += disp;
1083 		}
1084 
1085 		/* clear any space allocated but not loaded
1086 		 * - on uClinux we can just clear the lot
1087 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1088 		 *   extant in the file
1089 		 */
1090 		excess = phdr->p_memsz - phdr->p_filesz;
1091 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1092 
1093 #ifdef CONFIG_MMU
1094 		if (excess > excess1) {
1095 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1096 			unsigned long xmaddr;
1097 
1098 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1099 			down_write(&mm->mmap_sem);
1100 			xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1101 					 prot, flags, 0);
1102 			up_write(&mm->mmap_sem);
1103 
1104 			kdebug("mmap[%d] <anon>"
1105 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1106 			       loop, xaddr, excess - excess1, prot, flags,
1107 			       xmaddr);
1108 
1109 			if (xmaddr != xaddr)
1110 				return -ENOMEM;
1111 		}
1112 
1113 		if (prot & PROT_WRITE && excess1 > 0) {
1114 			kdebug("clear[%d] ad=%lx sz=%lx",
1115 			       loop, maddr + phdr->p_filesz, excess1);
1116 			clear_user((void __user *) maddr + phdr->p_filesz,
1117 				   excess1);
1118 		}
1119 
1120 #else
1121 		if (excess > 0) {
1122 			kdebug("clear[%d] ad=%lx sz=%lx",
1123 			       loop, maddr + phdr->p_filesz, excess);
1124 			clear_user((void *) maddr + phdr->p_filesz, excess);
1125 		}
1126 #endif
1127 
1128 		if (mm) {
1129 			if (phdr->p_flags & PF_X) {
1130 				if (!mm->start_code) {
1131 					mm->start_code = maddr;
1132 					mm->end_code = maddr + phdr->p_memsz;
1133 				}
1134 			} else if (!mm->start_data) {
1135 				mm->start_data = maddr;
1136 				mm->end_data = maddr + phdr->p_memsz;
1137 			}
1138 		}
1139 
1140 		seg++;
1141 	}
1142 
1143 	return 0;
1144 }
1145 
1146 /*****************************************************************************/
1147 /*
1148  * ELF-FDPIC core dumper
1149  *
1150  * Modelled on fs/exec.c:aout_core_dump()
1151  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1152  *
1153  * Modelled on fs/binfmt_elf.c core dumper
1154  */
1155 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1156 
1157 /*
1158  * These are the only things you should do on a core-file: use only these
1159  * functions to write out all the necessary info.
1160  */
1161 static int dump_write(struct file *file, const void *addr, int nr)
1162 {
1163 	return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1164 }
1165 
1166 static int dump_seek(struct file *file, loff_t off)
1167 {
1168 	if (file->f_op->llseek) {
1169 		if (file->f_op->llseek(file, off, SEEK_SET) != off)
1170 			return 0;
1171 	} else {
1172 		file->f_pos = off;
1173 	}
1174 	return 1;
1175 }
1176 
1177 /*
1178  * Decide whether a segment is worth dumping; default is yes to be
1179  * sure (missing info is worse than too much; etc).
1180  * Personally I'd include everything, and use the coredump limit...
1181  *
1182  * I think we should skip something. But I am not sure how. H.J.
1183  */
1184 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1185 {
1186 	int dump_ok;
1187 
1188 	/* Do not dump I/O mapped devices or special mappings */
1189 	if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1190 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1191 		return 0;
1192 	}
1193 
1194 	/* If we may not read the contents, don't allow us to dump
1195 	 * them either. "dump_write()" can't handle it anyway.
1196 	 */
1197 	if (!(vma->vm_flags & VM_READ)) {
1198 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1199 		return 0;
1200 	}
1201 
1202 	/* By default, dump shared memory if mapped from an anonymous file. */
1203 	if (vma->vm_flags & VM_SHARED) {
1204 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1205 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1206 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1207 			       vma->vm_flags, dump_ok ? "yes" : "no");
1208 			return dump_ok;
1209 		}
1210 
1211 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1212 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1213 		       vma->vm_flags, dump_ok ? "yes" : "no");
1214 		return dump_ok;
1215 	}
1216 
1217 #ifdef CONFIG_MMU
1218 	/* By default, if it hasn't been written to, don't write it out */
1219 	if (!vma->anon_vma) {
1220 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1221 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1222 		       vma->vm_flags, dump_ok ? "yes" : "no");
1223 		return dump_ok;
1224 	}
1225 #endif
1226 
1227 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1228 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1229 	       dump_ok ? "yes" : "no");
1230 	return dump_ok;
1231 }
1232 
1233 /* An ELF note in memory */
1234 struct memelfnote
1235 {
1236 	const char *name;
1237 	int type;
1238 	unsigned int datasz;
1239 	void *data;
1240 };
1241 
1242 static int notesize(struct memelfnote *en)
1243 {
1244 	int sz;
1245 
1246 	sz = sizeof(struct elf_note);
1247 	sz += roundup(strlen(en->name) + 1, 4);
1248 	sz += roundup(en->datasz, 4);
1249 
1250 	return sz;
1251 }
1252 
1253 /* #define DEBUG */
1254 
1255 #define DUMP_WRITE(addr, nr)	\
1256 	do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1257 #define DUMP_SEEK(off)	\
1258 	do { if (!dump_seek(file, (off))) return 0; } while(0)
1259 
1260 static int writenote(struct memelfnote *men, struct file *file)
1261 {
1262 	struct elf_note en;
1263 
1264 	en.n_namesz = strlen(men->name) + 1;
1265 	en.n_descsz = men->datasz;
1266 	en.n_type = men->type;
1267 
1268 	DUMP_WRITE(&en, sizeof(en));
1269 	DUMP_WRITE(men->name, en.n_namesz);
1270 	/* XXX - cast from long long to long to avoid need for libgcc.a */
1271 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1272 	DUMP_WRITE(men->data, men->datasz);
1273 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1274 
1275 	return 1;
1276 }
1277 #undef DUMP_WRITE
1278 #undef DUMP_SEEK
1279 
1280 #define DUMP_WRITE(addr, nr)	\
1281 	if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1282 		goto end_coredump;
1283 #define DUMP_SEEK(off)	\
1284 	if (!dump_seek(file, (off))) \
1285 		goto end_coredump;
1286 
1287 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1288 {
1289 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1290 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1291 	elf->e_ident[EI_DATA] = ELF_DATA;
1292 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1293 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1294 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1295 
1296 	elf->e_type = ET_CORE;
1297 	elf->e_machine = ELF_ARCH;
1298 	elf->e_version = EV_CURRENT;
1299 	elf->e_entry = 0;
1300 	elf->e_phoff = sizeof(struct elfhdr);
1301 	elf->e_shoff = 0;
1302 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1303 	elf->e_ehsize = sizeof(struct elfhdr);
1304 	elf->e_phentsize = sizeof(struct elf_phdr);
1305 	elf->e_phnum = segs;
1306 	elf->e_shentsize = 0;
1307 	elf->e_shnum = 0;
1308 	elf->e_shstrndx = 0;
1309 	return;
1310 }
1311 
1312 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1313 {
1314 	phdr->p_type = PT_NOTE;
1315 	phdr->p_offset = offset;
1316 	phdr->p_vaddr = 0;
1317 	phdr->p_paddr = 0;
1318 	phdr->p_filesz = sz;
1319 	phdr->p_memsz = 0;
1320 	phdr->p_flags = 0;
1321 	phdr->p_align = 0;
1322 	return;
1323 }
1324 
1325 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1326 		unsigned int sz, void *data)
1327 {
1328 	note->name = name;
1329 	note->type = type;
1330 	note->datasz = sz;
1331 	note->data = data;
1332 	return;
1333 }
1334 
1335 /*
1336  * fill up all the fields in prstatus from the given task struct, except
1337  * registers which need to be filled up seperately.
1338  */
1339 static void fill_prstatus(struct elf_prstatus *prstatus,
1340 			  struct task_struct *p, long signr)
1341 {
1342 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1343 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1344 	prstatus->pr_sighold = p->blocked.sig[0];
1345 	prstatus->pr_pid = p->pid;
1346 	prstatus->pr_ppid = p->parent->pid;
1347 	prstatus->pr_pgrp = process_group(p);
1348 	prstatus->pr_sid = process_session(p);
1349 	if (thread_group_leader(p)) {
1350 		/*
1351 		 * This is the record for the group leader.  Add in the
1352 		 * cumulative times of previous dead threads.  This total
1353 		 * won't include the time of each live thread whose state
1354 		 * is included in the core dump.  The final total reported
1355 		 * to our parent process when it calls wait4 will include
1356 		 * those sums as well as the little bit more time it takes
1357 		 * this and each other thread to finish dying after the
1358 		 * core dump synchronization phase.
1359 		 */
1360 		cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1361 				   &prstatus->pr_utime);
1362 		cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1363 				   &prstatus->pr_stime);
1364 	} else {
1365 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1366 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1367 	}
1368 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1369 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1370 
1371 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1372 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1373 }
1374 
1375 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1376 		       struct mm_struct *mm)
1377 {
1378 	unsigned int i, len;
1379 
1380 	/* first copy the parameters from user space */
1381 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1382 
1383 	len = mm->arg_end - mm->arg_start;
1384 	if (len >= ELF_PRARGSZ)
1385 		len = ELF_PRARGSZ - 1;
1386 	if (copy_from_user(&psinfo->pr_psargs,
1387 		           (const char __user *) mm->arg_start, len))
1388 		return -EFAULT;
1389 	for (i = 0; i < len; i++)
1390 		if (psinfo->pr_psargs[i] == 0)
1391 			psinfo->pr_psargs[i] = ' ';
1392 	psinfo->pr_psargs[len] = 0;
1393 
1394 	psinfo->pr_pid = p->pid;
1395 	psinfo->pr_ppid = p->parent->pid;
1396 	psinfo->pr_pgrp = process_group(p);
1397 	psinfo->pr_sid = process_session(p);
1398 
1399 	i = p->state ? ffz(~p->state) + 1 : 0;
1400 	psinfo->pr_state = i;
1401 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1402 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1403 	psinfo->pr_nice = task_nice(p);
1404 	psinfo->pr_flag = p->flags;
1405 	SET_UID(psinfo->pr_uid, p->uid);
1406 	SET_GID(psinfo->pr_gid, p->gid);
1407 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1408 
1409 	return 0;
1410 }
1411 
1412 /* Here is the structure in which status of each thread is captured. */
1413 struct elf_thread_status
1414 {
1415 	struct list_head list;
1416 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1417 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1418 	struct task_struct *thread;
1419 #ifdef ELF_CORE_COPY_XFPREGS
1420 	elf_fpxregset_t xfpu;		/* NT_PRXFPREG */
1421 #endif
1422 	struct memelfnote notes[3];
1423 	int num_notes;
1424 };
1425 
1426 /*
1427  * In order to add the specific thread information for the elf file format,
1428  * we need to keep a linked list of every thread's pr_status and then create
1429  * a single section for them in the final core file.
1430  */
1431 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1432 {
1433 	struct task_struct *p = t->thread;
1434 	int sz = 0;
1435 
1436 	t->num_notes = 0;
1437 
1438 	fill_prstatus(&t->prstatus, p, signr);
1439 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1440 
1441 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1442 		  &t->prstatus);
1443 	t->num_notes++;
1444 	sz += notesize(&t->notes[0]);
1445 
1446 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1447 	if (t->prstatus.pr_fpvalid) {
1448 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1449 			  &t->fpu);
1450 		t->num_notes++;
1451 		sz += notesize(&t->notes[1]);
1452 	}
1453 
1454 #ifdef ELF_CORE_COPY_XFPREGS
1455 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1456 		fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu),
1457 			  &t->xfpu);
1458 		t->num_notes++;
1459 		sz += notesize(&t->notes[2]);
1460 	}
1461 #endif
1462 	return sz;
1463 }
1464 
1465 /*
1466  * dump the segments for an MMU process
1467  */
1468 #ifdef CONFIG_MMU
1469 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1470 			   unsigned long *limit, unsigned long mm_flags)
1471 {
1472 	struct vm_area_struct *vma;
1473 
1474 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1475 		unsigned long addr;
1476 
1477 		if (!maydump(vma, mm_flags))
1478 			continue;
1479 
1480 		for (addr = vma->vm_start;
1481 		     addr < vma->vm_end;
1482 		     addr += PAGE_SIZE
1483 		     ) {
1484 			struct vm_area_struct *vma;
1485 			struct page *page;
1486 
1487 			if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1488 					   &page, &vma) <= 0) {
1489 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1490 			}
1491 			else if (page == ZERO_PAGE(addr)) {
1492 				page_cache_release(page);
1493 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1494 			}
1495 			else {
1496 				void *kaddr;
1497 
1498 				flush_cache_page(vma, addr, page_to_pfn(page));
1499 				kaddr = kmap(page);
1500 				if ((*size += PAGE_SIZE) > *limit ||
1501 				    !dump_write(file, kaddr, PAGE_SIZE)
1502 				    ) {
1503 					kunmap(page);
1504 					page_cache_release(page);
1505 					return -EIO;
1506 				}
1507 				kunmap(page);
1508 				page_cache_release(page);
1509 			}
1510 		}
1511 	}
1512 
1513 	return 0;
1514 
1515 end_coredump:
1516 	return -EFBIG;
1517 }
1518 #endif
1519 
1520 /*
1521  * dump the segments for a NOMMU process
1522  */
1523 #ifndef CONFIG_MMU
1524 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1525 			   unsigned long *limit, unsigned long mm_flags)
1526 {
1527 	struct vm_list_struct *vml;
1528 
1529 	for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1530 	struct vm_area_struct *vma = vml->vma;
1531 
1532 		if (!maydump(vma, mm_flags))
1533 			continue;
1534 
1535 		if ((*size += PAGE_SIZE) > *limit)
1536 			return -EFBIG;
1537 
1538 		if (!dump_write(file, (void *) vma->vm_start,
1539 				vma->vm_end - vma->vm_start))
1540 			return -EIO;
1541 	}
1542 
1543 	return 0;
1544 }
1545 #endif
1546 
1547 /*
1548  * Actual dumper
1549  *
1550  * This is a two-pass process; first we find the offsets of the bits,
1551  * and then they are actually written out.  If we run out of core limit
1552  * we just truncate.
1553  */
1554 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1555 			       struct file *file)
1556 {
1557 #define	NUM_NOTES	6
1558 	int has_dumped = 0;
1559 	mm_segment_t fs;
1560 	int segs;
1561 	size_t size = 0;
1562 	int i;
1563 	struct vm_area_struct *vma;
1564 	struct elfhdr *elf = NULL;
1565 	loff_t offset = 0, dataoff;
1566 	unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1567 	int numnote;
1568 	struct memelfnote *notes = NULL;
1569 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1570 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1571  	struct task_struct *g, *p;
1572  	LIST_HEAD(thread_list);
1573  	struct list_head *t;
1574 	elf_fpregset_t *fpu = NULL;
1575 #ifdef ELF_CORE_COPY_XFPREGS
1576 	elf_fpxregset_t *xfpu = NULL;
1577 #endif
1578 	int thread_status_size = 0;
1579 #ifndef CONFIG_MMU
1580 	struct vm_list_struct *vml;
1581 #endif
1582 	elf_addr_t *auxv;
1583 	unsigned long mm_flags;
1584 
1585 	/*
1586 	 * We no longer stop all VM operations.
1587 	 *
1588 	 * This is because those proceses that could possibly change map_count
1589 	 * or the mmap / vma pages are now blocked in do_exit on current
1590 	 * finishing this core dump.
1591 	 *
1592 	 * Only ptrace can touch these memory addresses, but it doesn't change
1593 	 * the map_count or the pages allocated. So no possibility of crashing
1594 	 * exists while dumping the mm->vm_next areas to the core file.
1595 	 */
1596 
1597 	/* alloc memory for large data structures: too large to be on stack */
1598 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1599 	if (!elf)
1600 		goto cleanup;
1601 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1602 	if (!prstatus)
1603 		goto cleanup;
1604 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1605 	if (!psinfo)
1606 		goto cleanup;
1607 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1608 	if (!notes)
1609 		goto cleanup;
1610 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1611 	if (!fpu)
1612 		goto cleanup;
1613 #ifdef ELF_CORE_COPY_XFPREGS
1614 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1615 	if (!xfpu)
1616 		goto cleanup;
1617 #endif
1618 
1619 	if (signr) {
1620 		struct elf_thread_status *tmp;
1621 		rcu_read_lock();
1622 		do_each_thread(g,p)
1623 			if (current->mm == p->mm && current != p) {
1624 				tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1625 				if (!tmp) {
1626 					rcu_read_unlock();
1627 					goto cleanup;
1628 				}
1629 				tmp->thread = p;
1630 				list_add(&tmp->list, &thread_list);
1631 			}
1632 		while_each_thread(g,p);
1633 		rcu_read_unlock();
1634 		list_for_each(t, &thread_list) {
1635 			struct elf_thread_status *tmp;
1636 			int sz;
1637 
1638 			tmp = list_entry(t, struct elf_thread_status, list);
1639 			sz = elf_dump_thread_status(signr, tmp);
1640 			thread_status_size += sz;
1641 		}
1642 	}
1643 
1644 	/* now collect the dump for the current */
1645 	fill_prstatus(prstatus, current, signr);
1646 	elf_core_copy_regs(&prstatus->pr_reg, regs);
1647 
1648 #ifdef CONFIG_MMU
1649 	segs = current->mm->map_count;
1650 #else
1651 	segs = 0;
1652 	for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1653 	    segs++;
1654 #endif
1655 #ifdef ELF_CORE_EXTRA_PHDRS
1656 	segs += ELF_CORE_EXTRA_PHDRS;
1657 #endif
1658 
1659 	/* Set up header */
1660 	fill_elf_fdpic_header(elf, segs + 1);	/* including notes section */
1661 
1662 	has_dumped = 1;
1663 	current->flags |= PF_DUMPCORE;
1664 
1665 	/*
1666 	 * Set up the notes in similar form to SVR4 core dumps made
1667 	 * with info from their /proc.
1668 	 */
1669 
1670 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1671 	fill_psinfo(psinfo, current->group_leader, current->mm);
1672 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1673 
1674 	numnote = 2;
1675 
1676 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1677 
1678 	i = 0;
1679 	do
1680 		i += 2;
1681 	while (auxv[i - 2] != AT_NULL);
1682 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1683 		  i * sizeof(elf_addr_t), auxv);
1684 
1685   	/* Try to dump the FPU. */
1686 	if ((prstatus->pr_fpvalid =
1687 	     elf_core_copy_task_fpregs(current, regs, fpu)))
1688 		fill_note(notes + numnote++,
1689 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1690 #ifdef ELF_CORE_COPY_XFPREGS
1691 	if (elf_core_copy_task_xfpregs(current, xfpu))
1692 		fill_note(notes + numnote++,
1693 			  "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu);
1694 #endif
1695 
1696 	fs = get_fs();
1697 	set_fs(KERNEL_DS);
1698 
1699 	DUMP_WRITE(elf, sizeof(*elf));
1700 	offset += sizeof(*elf);				/* Elf header */
1701 	offset += (segs+1) * sizeof(struct elf_phdr);	/* Program headers */
1702 
1703 	/* Write notes phdr entry */
1704 	{
1705 		struct elf_phdr phdr;
1706 		int sz = 0;
1707 
1708 		for (i = 0; i < numnote; i++)
1709 			sz += notesize(notes + i);
1710 
1711 		sz += thread_status_size;
1712 
1713 		fill_elf_note_phdr(&phdr, sz, offset);
1714 		offset += sz;
1715 		DUMP_WRITE(&phdr, sizeof(phdr));
1716 	}
1717 
1718 	/* Page-align dumped data */
1719 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1720 
1721 	/*
1722 	 * We must use the same mm->flags while dumping core to avoid
1723 	 * inconsistency between the program headers and bodies, otherwise an
1724 	 * unusable core file can be generated.
1725 	 */
1726 	mm_flags = current->mm->flags;
1727 
1728 	/* write program headers for segments dump */
1729 	for (
1730 #ifdef CONFIG_MMU
1731 		vma = current->mm->mmap; vma; vma = vma->vm_next
1732 #else
1733 			vml = current->mm->context.vmlist; vml; vml = vml->next
1734 #endif
1735 	     ) {
1736 		struct elf_phdr phdr;
1737 		size_t sz;
1738 
1739 #ifndef CONFIG_MMU
1740 		vma = vml->vma;
1741 #endif
1742 
1743 		sz = vma->vm_end - vma->vm_start;
1744 
1745 		phdr.p_type = PT_LOAD;
1746 		phdr.p_offset = offset;
1747 		phdr.p_vaddr = vma->vm_start;
1748 		phdr.p_paddr = 0;
1749 		phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1750 		phdr.p_memsz = sz;
1751 		offset += phdr.p_filesz;
1752 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1753 		if (vma->vm_flags & VM_WRITE)
1754 			phdr.p_flags |= PF_W;
1755 		if (vma->vm_flags & VM_EXEC)
1756 			phdr.p_flags |= PF_X;
1757 		phdr.p_align = ELF_EXEC_PAGESIZE;
1758 
1759 		DUMP_WRITE(&phdr, sizeof(phdr));
1760 	}
1761 
1762 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1763 	ELF_CORE_WRITE_EXTRA_PHDRS;
1764 #endif
1765 
1766  	/* write out the notes section */
1767 	for (i = 0; i < numnote; i++)
1768 		if (!writenote(notes + i, file))
1769 			goto end_coredump;
1770 
1771 	/* write out the thread status notes section */
1772 	list_for_each(t, &thread_list) {
1773 		struct elf_thread_status *tmp =
1774 				list_entry(t, struct elf_thread_status, list);
1775 
1776 		for (i = 0; i < tmp->num_notes; i++)
1777 			if (!writenote(&tmp->notes[i], file))
1778 				goto end_coredump;
1779 	}
1780 
1781 	DUMP_SEEK(dataoff);
1782 
1783 	if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1784 		goto end_coredump;
1785 
1786 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1787 	ELF_CORE_WRITE_EXTRA_DATA;
1788 #endif
1789 
1790 	if (file->f_pos != offset) {
1791 		/* Sanity check */
1792 		printk(KERN_WARNING
1793 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1794 		       file->f_pos, offset);
1795 	}
1796 
1797 end_coredump:
1798 	set_fs(fs);
1799 
1800 cleanup:
1801 	while (!list_empty(&thread_list)) {
1802 		struct list_head *tmp = thread_list.next;
1803 		list_del(tmp);
1804 		kfree(list_entry(tmp, struct elf_thread_status, list));
1805 	}
1806 
1807 	kfree(elf);
1808 	kfree(prstatus);
1809 	kfree(psinfo);
1810 	kfree(notes);
1811 	kfree(fpu);
1812 #ifdef ELF_CORE_COPY_XFPREGS
1813 	kfree(xfpu);
1814 #endif
1815 	return has_dumped;
1816 #undef NUM_NOTES
1817 }
1818 
1819 #endif		/* USE_ELF_CORE_DUMP */
1820