xref: /linux/fs/binfmt_elf_fdpic.c (revision f7511d5f66f01fc451747b24e79f3ada7a3af9af)
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 *, unsigned long limit);
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 (unlikely(retval != size))
140 		return retval < 0 ? retval : -ENOEXEC;
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 (unlikely(retval != phdr->p_filesz)) {
222 				if (retval >= 0)
223 					retval = -ENOEXEC;
224 				goto error;
225 			}
226 
227 			retval = -ENOENT;
228 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
229 				goto error;
230 
231 			kdebug("Using ELF interpreter %s", interpreter_name);
232 
233 			/* replace the program with the interpreter */
234 			interpreter = open_exec(interpreter_name);
235 			retval = PTR_ERR(interpreter);
236 			if (IS_ERR(interpreter)) {
237 				interpreter = NULL;
238 				goto error;
239 			}
240 
241 			/*
242 			 * If the binary is not readable then enforce
243 			 * mm->dumpable = 0 regardless of the interpreter's
244 			 * permissions.
245 			 */
246 			if (file_permission(interpreter, MAY_READ) < 0)
247 				bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
248 
249 			retval = kernel_read(interpreter, 0, bprm->buf,
250 					     BINPRM_BUF_SIZE);
251 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
252 				if (retval >= 0)
253 					retval = -ENOEXEC;
254 				goto error;
255 			}
256 
257 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
258 			break;
259 
260 		case PT_LOAD:
261 #ifdef CONFIG_MMU
262 			if (exec_params.load_addr == 0)
263 				exec_params.load_addr = phdr->p_vaddr;
264 #endif
265 			break;
266 		}
267 
268 	}
269 
270 	if (elf_check_const_displacement(&exec_params.hdr))
271 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
272 
273 	/* perform insanity checks on the interpreter */
274 	if (interpreter_name) {
275 		retval = -ELIBBAD;
276 		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
277 			goto error;
278 
279 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
280 
281 		/* read the interpreter's program header table */
282 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
283 		if (retval < 0)
284 			goto error;
285 	}
286 
287 	stack_size = exec_params.stack_size;
288 	if (stack_size < interp_params.stack_size)
289 		stack_size = interp_params.stack_size;
290 
291 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
292 		executable_stack = EXSTACK_ENABLE_X;
293 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
294 		executable_stack = EXSTACK_DISABLE_X;
295 	else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
296 		executable_stack = EXSTACK_ENABLE_X;
297 	else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
298 		executable_stack = EXSTACK_DISABLE_X;
299 	else
300 		executable_stack = EXSTACK_DEFAULT;
301 
302 	retval = -ENOEXEC;
303 	if (stack_size == 0)
304 		goto error;
305 
306 	if (elf_check_const_displacement(&interp_params.hdr))
307 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
308 
309 	/* flush all traces of the currently running executable */
310 	retval = flush_old_exec(bprm);
311 	if (retval)
312 		goto error;
313 
314 	/* there's now no turning back... the old userspace image is dead,
315 	 * defunct, deceased, etc. after this point we have to exit via
316 	 * error_kill */
317 	set_personality(PER_LINUX_FDPIC);
318 	set_binfmt(&elf_fdpic_format);
319 
320 	current->mm->start_code = 0;
321 	current->mm->end_code = 0;
322 	current->mm->start_stack = 0;
323 	current->mm->start_data = 0;
324 	current->mm->end_data = 0;
325 	current->mm->context.exec_fdpic_loadmap = 0;
326 	current->mm->context.interp_fdpic_loadmap = 0;
327 
328 	current->flags &= ~PF_FORKNOEXEC;
329 
330 #ifdef CONFIG_MMU
331 	elf_fdpic_arch_lay_out_mm(&exec_params,
332 				  &interp_params,
333 				  &current->mm->start_stack,
334 				  &current->mm->start_brk);
335 
336 	retval = setup_arg_pages(bprm, current->mm->start_stack,
337 				 executable_stack);
338 	if (retval < 0) {
339 		send_sig(SIGKILL, current, 0);
340 		goto error_kill;
341 	}
342 #endif
343 
344 	/* load the executable and interpreter into memory */
345 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
346 				    "executable");
347 	if (retval < 0)
348 		goto error_kill;
349 
350 	if (interpreter_name) {
351 		retval = elf_fdpic_map_file(&interp_params, interpreter,
352 					    current->mm, "interpreter");
353 		if (retval < 0) {
354 			printk(KERN_ERR "Unable to load interpreter\n");
355 			goto error_kill;
356 		}
357 
358 		allow_write_access(interpreter);
359 		fput(interpreter);
360 		interpreter = NULL;
361 	}
362 
363 #ifdef CONFIG_MMU
364 	if (!current->mm->start_brk)
365 		current->mm->start_brk = current->mm->end_data;
366 
367 	current->mm->brk = current->mm->start_brk =
368 		PAGE_ALIGN(current->mm->start_brk);
369 
370 #else
371 	/* create a stack and brk area big enough for everyone
372 	 * - the brk heap starts at the bottom and works up
373 	 * - the stack starts at the top and works down
374 	 */
375 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
376 	if (stack_size < PAGE_SIZE * 2)
377 		stack_size = PAGE_SIZE * 2;
378 
379 	down_write(&current->mm->mmap_sem);
380 	current->mm->start_brk = do_mmap(NULL, 0, stack_size,
381 					 PROT_READ | PROT_WRITE | PROT_EXEC,
382 					 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
383 					 0);
384 
385 	if (IS_ERR_VALUE(current->mm->start_brk)) {
386 		up_write(&current->mm->mmap_sem);
387 		retval = current->mm->start_brk;
388 		current->mm->start_brk = 0;
389 		goto error_kill;
390 	}
391 
392 	/* expand the stack mapping to use up the entire allocation granule */
393 	fullsize = ksize((char *) current->mm->start_brk);
394 	if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
395 				    fullsize, 0, 0)))
396 		stack_size = fullsize;
397 	up_write(&current->mm->mmap_sem);
398 
399 	current->mm->brk = current->mm->start_brk;
400 	current->mm->context.end_brk = current->mm->start_brk;
401 	current->mm->context.end_brk +=
402 		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
403 	current->mm->start_stack = current->mm->start_brk + stack_size;
404 #endif
405 
406 	compute_creds(bprm);
407 	current->flags &= ~PF_FORKNOEXEC;
408 	if (create_elf_fdpic_tables(bprm, current->mm,
409 				    &exec_params, &interp_params) < 0)
410 		goto error_kill;
411 
412 	kdebug("- start_code  %lx", current->mm->start_code);
413 	kdebug("- end_code    %lx", current->mm->end_code);
414 	kdebug("- start_data  %lx", current->mm->start_data);
415 	kdebug("- end_data    %lx", current->mm->end_data);
416 	kdebug("- start_brk   %lx", current->mm->start_brk);
417 	kdebug("- brk         %lx", current->mm->brk);
418 	kdebug("- start_stack %lx", current->mm->start_stack);
419 
420 #ifdef ELF_FDPIC_PLAT_INIT
421 	/*
422 	 * The ABI may specify that certain registers be set up in special
423 	 * ways (on i386 %edx is the address of a DT_FINI function, for
424 	 * example.  This macro performs whatever initialization to
425 	 * the regs structure is required.
426 	 */
427 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
428 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
429 			    dynaddr);
430 #endif
431 
432 	/* everything is now ready... get the userspace context ready to roll */
433 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
434 	start_thread(regs, entryaddr, current->mm->start_stack);
435 
436 	if (unlikely(current->ptrace & PT_PTRACED)) {
437 		if (current->ptrace & PT_TRACE_EXEC)
438 			ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
439 		else
440 			send_sig(SIGTRAP, current, 0);
441 	}
442 
443 	retval = 0;
444 
445 error:
446 	if (interpreter) {
447 		allow_write_access(interpreter);
448 		fput(interpreter);
449 	}
450 	kfree(interpreter_name);
451 	kfree(exec_params.phdrs);
452 	kfree(exec_params.loadmap);
453 	kfree(interp_params.phdrs);
454 	kfree(interp_params.loadmap);
455 	return retval;
456 
457 	/* unrecoverable error - kill the process */
458 error_kill:
459 	send_sig(SIGSEGV, current, 0);
460 	goto error;
461 
462 }
463 
464 /*****************************************************************************/
465 /*
466  * present useful information to the program
467  */
468 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
469 				   struct mm_struct *mm,
470 				   struct elf_fdpic_params *exec_params,
471 				   struct elf_fdpic_params *interp_params)
472 {
473 	unsigned long sp, csp, nitems;
474 	elf_caddr_t __user *argv, *envp;
475 	size_t platform_len = 0, len;
476 	char *k_platform;
477 	char __user *u_platform, *p;
478 	long hwcap;
479 	int loop;
480 
481 	/* we're going to shovel a whole load of stuff onto the stack */
482 #ifdef CONFIG_MMU
483 	sp = bprm->p;
484 #else
485 	sp = mm->start_stack;
486 
487 	/* stack the program arguments and environment */
488 	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
489 		return -EFAULT;
490 #endif
491 
492 	/* get hold of platform and hardware capabilities masks for the machine
493 	 * we are running on.  In some cases (Sparc), this info is impossible
494 	 * to get, in others (i386) it is merely difficult.
495 	 */
496 	hwcap = ELF_HWCAP;
497 	k_platform = ELF_PLATFORM;
498 	u_platform = NULL;
499 
500 	if (k_platform) {
501 		platform_len = strlen(k_platform) + 1;
502 		sp -= platform_len;
503 		u_platform = (char __user *) sp;
504 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
505 			return -EFAULT;
506 	}
507 
508 #if defined(__i386__) && defined(CONFIG_SMP)
509 	/* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
510 	 * by the processes running on the same package. One thing we can do is
511 	 * to shuffle the initial stack for them.
512 	 *
513 	 * the conditionals here are unneeded, but kept in to make the code
514 	 * behaviour the same as pre change unless we have hyperthreaded
515 	 * processors. This keeps Mr Marcelo Person happier but should be
516 	 * removed for 2.5
517 	 */
518 	if (smp_num_siblings > 1)
519 		sp = sp - ((current->pid % 64) << 7);
520 #endif
521 
522 	sp &= ~7UL;
523 
524 	/* stack the load map(s) */
525 	len = sizeof(struct elf32_fdpic_loadmap);
526 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
527 	sp = (sp - len) & ~7UL;
528 	exec_params->map_addr = sp;
529 
530 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
531 		return -EFAULT;
532 
533 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
534 
535 	if (interp_params->loadmap) {
536 		len = sizeof(struct elf32_fdpic_loadmap);
537 		len += sizeof(struct elf32_fdpic_loadseg) *
538 			interp_params->loadmap->nsegs;
539 		sp = (sp - len) & ~7UL;
540 		interp_params->map_addr = sp;
541 
542 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
543 				 len) != 0)
544 			return -EFAULT;
545 
546 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
547 	}
548 
549 	/* force 16 byte _final_ alignment here for generality */
550 #define DLINFO_ITEMS 13
551 
552 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0);
553 #ifdef DLINFO_ARCH_ITEMS
554 	nitems += DLINFO_ARCH_ITEMS;
555 #endif
556 
557 	csp = sp;
558 	sp -= nitems * 2 * sizeof(unsigned long);
559 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
560 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
561 	sp -= 1 * sizeof(unsigned long);		/* argc */
562 
563 	csp -= sp & 15UL;
564 	sp -= sp & 15UL;
565 
566 	/* put the ELF interpreter info on the stack */
567 #define NEW_AUX_ENT(nr, id, val)					\
568 	do {								\
569 		struct { unsigned long _id, _val; } __user *ent;	\
570 									\
571 		ent = (void __user *) csp;				\
572 		__put_user((id), &ent[nr]._id);				\
573 		__put_user((val), &ent[nr]._val);			\
574 	} while (0)
575 
576 	csp -= 2 * sizeof(unsigned long);
577 	NEW_AUX_ENT(0, AT_NULL, 0);
578 	if (k_platform) {
579 		csp -= 2 * sizeof(unsigned long);
580 		NEW_AUX_ENT(0, AT_PLATFORM,
581 			    (elf_addr_t) (unsigned long) u_platform);
582 	}
583 
584 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
585 	NEW_AUX_ENT( 0, AT_HWCAP,	hwcap);
586 	NEW_AUX_ENT( 1, AT_PAGESZ,	PAGE_SIZE);
587 	NEW_AUX_ENT( 2, AT_CLKTCK,	CLOCKS_PER_SEC);
588 	NEW_AUX_ENT( 3, AT_PHDR,	exec_params->ph_addr);
589 	NEW_AUX_ENT( 4, AT_PHENT,	sizeof(struct elf_phdr));
590 	NEW_AUX_ENT( 5, AT_PHNUM,	exec_params->hdr.e_phnum);
591 	NEW_AUX_ENT( 6,	AT_BASE,	interp_params->elfhdr_addr);
592 	NEW_AUX_ENT( 7, AT_FLAGS,	0);
593 	NEW_AUX_ENT( 8, AT_ENTRY,	exec_params->entry_addr);
594 	NEW_AUX_ENT( 9, AT_UID,		(elf_addr_t) current->uid);
595 	NEW_AUX_ENT(10, AT_EUID,	(elf_addr_t) current->euid);
596 	NEW_AUX_ENT(11, AT_GID,		(elf_addr_t) current->gid);
597 	NEW_AUX_ENT(12, AT_EGID,	(elf_addr_t) current->egid);
598 
599 #ifdef ARCH_DLINFO
600 	/* ARCH_DLINFO must come last so platform specific code can enforce
601 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
602 	 */
603 	ARCH_DLINFO;
604 #endif
605 #undef NEW_AUX_ENT
606 
607 	/* allocate room for argv[] and envv[] */
608 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
609 	envp = (elf_caddr_t __user *) csp;
610 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
611 	argv = (elf_caddr_t __user *) csp;
612 
613 	/* stack argc */
614 	csp -= sizeof(unsigned long);
615 	__put_user(bprm->argc, (unsigned long __user *) csp);
616 
617 	BUG_ON(csp != sp);
618 
619 	/* fill in the argv[] array */
620 #ifdef CONFIG_MMU
621 	current->mm->arg_start = bprm->p;
622 #else
623 	current->mm->arg_start = current->mm->start_stack -
624 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
625 #endif
626 
627 	p = (char __user *) current->mm->arg_start;
628 	for (loop = bprm->argc; loop > 0; loop--) {
629 		__put_user((elf_caddr_t) p, argv++);
630 		len = strnlen_user(p, MAX_ARG_STRLEN);
631 		if (!len || len > MAX_ARG_STRLEN)
632 			return -EINVAL;
633 		p += len;
634 	}
635 	__put_user(NULL, argv);
636 	current->mm->arg_end = (unsigned long) p;
637 
638 	/* fill in the envv[] array */
639 	current->mm->env_start = (unsigned long) p;
640 	for (loop = bprm->envc; loop > 0; loop--) {
641 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
642 		len = strnlen_user(p, MAX_ARG_STRLEN);
643 		if (!len || len > MAX_ARG_STRLEN)
644 			return -EINVAL;
645 		p += len;
646 	}
647 	__put_user(NULL, envp);
648 	current->mm->env_end = (unsigned long) p;
649 
650 	mm->start_stack = (unsigned long) sp;
651 	return 0;
652 }
653 
654 /*****************************************************************************/
655 /*
656  * transfer the program arguments and environment from the holding pages onto
657  * the stack
658  */
659 #ifndef CONFIG_MMU
660 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
661 					    unsigned long *_sp)
662 {
663 	unsigned long index, stop, sp;
664 	char *src;
665 	int ret = 0;
666 
667 	stop = bprm->p >> PAGE_SHIFT;
668 	sp = *_sp;
669 
670 	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
671 		src = kmap(bprm->page[index]);
672 		sp -= PAGE_SIZE;
673 		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
674 			ret = -EFAULT;
675 		kunmap(bprm->page[index]);
676 		if (ret < 0)
677 			goto out;
678 	}
679 
680 	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
681 
682 out:
683 	return ret;
684 }
685 #endif
686 
687 /*****************************************************************************/
688 /*
689  * load the appropriate binary image (executable or interpreter) into memory
690  * - we assume no MMU is available
691  * - if no other PIC bits are set in params->hdr->e_flags
692  *   - we assume that the LOADable segments in the binary are independently relocatable
693  *   - we assume R/O executable segments are shareable
694  * - else
695  *   - we assume the loadable parts of the image to require fixed displacement
696  *   - the image is not shareable
697  */
698 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
699 			      struct file *file,
700 			      struct mm_struct *mm,
701 			      const char *what)
702 {
703 	struct elf32_fdpic_loadmap *loadmap;
704 #ifdef CONFIG_MMU
705 	struct elf32_fdpic_loadseg *mseg;
706 #endif
707 	struct elf32_fdpic_loadseg *seg;
708 	struct elf32_phdr *phdr;
709 	unsigned long load_addr, stop;
710 	unsigned nloads, tmp;
711 	size_t size;
712 	int loop, ret;
713 
714 	/* allocate a load map table */
715 	nloads = 0;
716 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
717 		if (params->phdrs[loop].p_type == PT_LOAD)
718 			nloads++;
719 
720 	if (nloads == 0)
721 		return -ELIBBAD;
722 
723 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
724 	loadmap = kzalloc(size, GFP_KERNEL);
725 	if (!loadmap)
726 		return -ENOMEM;
727 
728 	params->loadmap = loadmap;
729 
730 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
731 	loadmap->nsegs = nloads;
732 
733 	load_addr = params->load_addr;
734 	seg = loadmap->segs;
735 
736 	/* map the requested LOADs into the memory space */
737 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
738 	case ELF_FDPIC_FLAG_CONSTDISP:
739 	case ELF_FDPIC_FLAG_CONTIGUOUS:
740 #ifndef CONFIG_MMU
741 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
742 		if (ret < 0)
743 			return ret;
744 		break;
745 #endif
746 	default:
747 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
748 		if (ret < 0)
749 			return ret;
750 		break;
751 	}
752 
753 	/* map the entry point */
754 	if (params->hdr.e_entry) {
755 		seg = loadmap->segs;
756 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
757 			if (params->hdr.e_entry >= seg->p_vaddr &&
758 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
759 				params->entry_addr =
760 					(params->hdr.e_entry - seg->p_vaddr) +
761 					seg->addr;
762 				break;
763 			}
764 		}
765 	}
766 
767 	/* determine where the program header table has wound up if mapped */
768 	stop = params->hdr.e_phoff;
769 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
770 	phdr = params->phdrs;
771 
772 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
773 		if (phdr->p_type != PT_LOAD)
774 			continue;
775 
776 		if (phdr->p_offset > params->hdr.e_phoff ||
777 		    phdr->p_offset + phdr->p_filesz < stop)
778 			continue;
779 
780 		seg = loadmap->segs;
781 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
782 			if (phdr->p_vaddr >= seg->p_vaddr &&
783 			    phdr->p_vaddr + phdr->p_filesz <=
784 			    seg->p_vaddr + seg->p_memsz) {
785 				params->ph_addr =
786 					(phdr->p_vaddr - seg->p_vaddr) +
787 					seg->addr +
788 					params->hdr.e_phoff - phdr->p_offset;
789 				break;
790 			}
791 		}
792 		break;
793 	}
794 
795 	/* determine where the dynamic section has wound up if there is one */
796 	phdr = params->phdrs;
797 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
798 		if (phdr->p_type != PT_DYNAMIC)
799 			continue;
800 
801 		seg = loadmap->segs;
802 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
803 			if (phdr->p_vaddr >= seg->p_vaddr &&
804 			    phdr->p_vaddr + phdr->p_memsz <=
805 			    seg->p_vaddr + seg->p_memsz) {
806 				params->dynamic_addr =
807 					(phdr->p_vaddr - seg->p_vaddr) +
808 					seg->addr;
809 
810 				/* check the dynamic section contains at least
811 				 * one item, and that the last item is a NULL
812 				 * entry */
813 				if (phdr->p_memsz == 0 ||
814 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
815 					goto dynamic_error;
816 
817 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
818 				if (((Elf32_Dyn *)
819 				     params->dynamic_addr)[tmp - 1].d_tag != 0)
820 					goto dynamic_error;
821 				break;
822 			}
823 		}
824 		break;
825 	}
826 
827 	/* now elide adjacent segments in the load map on MMU linux
828 	 * - on uClinux the holes between may actually be filled with system
829 	 *   stuff or stuff from other processes
830 	 */
831 #ifdef CONFIG_MMU
832 	nloads = loadmap->nsegs;
833 	mseg = loadmap->segs;
834 	seg = mseg + 1;
835 	for (loop = 1; loop < nloads; loop++) {
836 		/* see if we have a candidate for merging */
837 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
838 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
839 			if (load_addr == (seg->addr & PAGE_MASK)) {
840 				mseg->p_memsz +=
841 					load_addr -
842 					(mseg->addr + mseg->p_memsz);
843 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
844 				mseg->p_memsz += seg->p_memsz;
845 				loadmap->nsegs--;
846 				continue;
847 			}
848 		}
849 
850 		mseg++;
851 		if (mseg != seg)
852 			*mseg = *seg;
853 	}
854 #endif
855 
856 	kdebug("Mapped Object [%s]:", what);
857 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
858 	kdebug("- entry    : %lx", params->entry_addr);
859 	kdebug("- PHDR[]   : %lx", params->ph_addr);
860 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
861 	seg = loadmap->segs;
862 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
863 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
864 		       loop,
865 		       seg->addr, seg->addr + seg->p_memsz - 1,
866 		       seg->p_vaddr, seg->p_memsz);
867 
868 	return 0;
869 
870 dynamic_error:
871 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
872 	       what, file->f_path.dentry->d_inode->i_ino);
873 	return -ELIBBAD;
874 }
875 
876 /*****************************************************************************/
877 /*
878  * map a file with constant displacement under uClinux
879  */
880 #ifndef CONFIG_MMU
881 static int elf_fdpic_map_file_constdisp_on_uclinux(
882 	struct elf_fdpic_params *params,
883 	struct file *file,
884 	struct mm_struct *mm)
885 {
886 	struct elf32_fdpic_loadseg *seg;
887 	struct elf32_phdr *phdr;
888 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
889 	loff_t fpos;
890 	int loop, ret;
891 
892 	load_addr = params->load_addr;
893 	seg = params->loadmap->segs;
894 
895 	/* determine the bounds of the contiguous overall allocation we must
896 	 * make */
897 	phdr = params->phdrs;
898 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
899 		if (params->phdrs[loop].p_type != PT_LOAD)
900 			continue;
901 
902 		if (base > phdr->p_vaddr)
903 			base = phdr->p_vaddr;
904 		if (top < phdr->p_vaddr + phdr->p_memsz)
905 			top = phdr->p_vaddr + phdr->p_memsz;
906 	}
907 
908 	/* allocate one big anon block for everything */
909 	mflags = MAP_PRIVATE;
910 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
911 		mflags |= MAP_EXECUTABLE;
912 
913 	down_write(&mm->mmap_sem);
914 	maddr = do_mmap(NULL, load_addr, top - base,
915 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
916 	up_write(&mm->mmap_sem);
917 	if (IS_ERR_VALUE(maddr))
918 		return (int) maddr;
919 
920 	if (load_addr != 0)
921 		load_addr += PAGE_ALIGN(top - base);
922 
923 	/* and then load the file segments into it */
924 	phdr = params->phdrs;
925 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
926 		if (params->phdrs[loop].p_type != PT_LOAD)
927 			continue;
928 
929 		fpos = phdr->p_offset;
930 
931 		seg->addr = maddr + (phdr->p_vaddr - base);
932 		seg->p_vaddr = phdr->p_vaddr;
933 		seg->p_memsz = phdr->p_memsz;
934 
935 		ret = file->f_op->read(file, (void *) seg->addr,
936 				       phdr->p_filesz, &fpos);
937 		if (ret < 0)
938 			return ret;
939 
940 		/* map the ELF header address if in this segment */
941 		if (phdr->p_offset == 0)
942 			params->elfhdr_addr = seg->addr;
943 
944 		/* clear any space allocated but not loaded */
945 		if (phdr->p_filesz < phdr->p_memsz)
946 			clear_user((void *) (seg->addr + phdr->p_filesz),
947 				   phdr->p_memsz - phdr->p_filesz);
948 
949 		if (mm) {
950 			if (phdr->p_flags & PF_X) {
951 				if (!mm->start_code) {
952 					mm->start_code = seg->addr;
953 					mm->end_code = seg->addr +
954 						phdr->p_memsz;
955 				}
956 			} else if (!mm->start_data) {
957 				mm->start_data = seg->addr;
958 #ifndef CONFIG_MMU
959 				mm->end_data = seg->addr + phdr->p_memsz;
960 #endif
961 			}
962 
963 #ifdef CONFIG_MMU
964 			if (seg->addr + phdr->p_memsz > mm->end_data)
965 				mm->end_data = seg->addr + phdr->p_memsz;
966 #endif
967 		}
968 
969 		seg++;
970 	}
971 
972 	return 0;
973 }
974 #endif
975 
976 /*****************************************************************************/
977 /*
978  * map a binary by direct mmap() of the individual PT_LOAD segments
979  */
980 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
981 					     struct file *file,
982 					     struct mm_struct *mm)
983 {
984 	struct elf32_fdpic_loadseg *seg;
985 	struct elf32_phdr *phdr;
986 	unsigned long load_addr, delta_vaddr;
987 	int loop, dvset;
988 
989 	load_addr = params->load_addr;
990 	delta_vaddr = 0;
991 	dvset = 0;
992 
993 	seg = params->loadmap->segs;
994 
995 	/* deal with each load segment separately */
996 	phdr = params->phdrs;
997 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
998 		unsigned long maddr, disp, excess, excess1;
999 		int prot = 0, flags;
1000 
1001 		if (phdr->p_type != PT_LOAD)
1002 			continue;
1003 
1004 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1005 		       (unsigned long) phdr->p_vaddr,
1006 		       (unsigned long) phdr->p_offset,
1007 		       (unsigned long) phdr->p_filesz,
1008 		       (unsigned long) phdr->p_memsz);
1009 
1010 		/* determine the mapping parameters */
1011 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1012 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1013 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1014 
1015 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1016 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1017 			flags |= MAP_EXECUTABLE;
1018 
1019 		maddr = 0;
1020 
1021 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1022 		case ELF_FDPIC_FLAG_INDEPENDENT:
1023 			/* PT_LOADs are independently locatable */
1024 			break;
1025 
1026 		case ELF_FDPIC_FLAG_HONOURVADDR:
1027 			/* the specified virtual address must be honoured */
1028 			maddr = phdr->p_vaddr;
1029 			flags |= MAP_FIXED;
1030 			break;
1031 
1032 		case ELF_FDPIC_FLAG_CONSTDISP:
1033 			/* constant displacement
1034 			 * - can be mapped anywhere, but must be mapped as a
1035 			 *   unit
1036 			 */
1037 			if (!dvset) {
1038 				maddr = load_addr;
1039 				delta_vaddr = phdr->p_vaddr;
1040 				dvset = 1;
1041 			} else {
1042 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1043 				flags |= MAP_FIXED;
1044 			}
1045 			break;
1046 
1047 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1048 			/* contiguity handled later */
1049 			break;
1050 
1051 		default:
1052 			BUG();
1053 		}
1054 
1055 		maddr &= PAGE_MASK;
1056 
1057 		/* create the mapping */
1058 		disp = phdr->p_vaddr & ~PAGE_MASK;
1059 		down_write(&mm->mmap_sem);
1060 		maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1061 				phdr->p_offset - disp);
1062 		up_write(&mm->mmap_sem);
1063 
1064 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1065 		       loop, phdr->p_memsz + disp, prot, flags,
1066 		       phdr->p_offset - disp, maddr);
1067 
1068 		if (IS_ERR_VALUE(maddr))
1069 			return (int) maddr;
1070 
1071 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1072 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1073 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1074 
1075 		seg->addr = maddr + disp;
1076 		seg->p_vaddr = phdr->p_vaddr;
1077 		seg->p_memsz = phdr->p_memsz;
1078 
1079 		/* map the ELF header address if in this segment */
1080 		if (phdr->p_offset == 0)
1081 			params->elfhdr_addr = seg->addr;
1082 
1083 		/* clear the bit between beginning of mapping and beginning of
1084 		 * PT_LOAD */
1085 		if (prot & PROT_WRITE && disp > 0) {
1086 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1087 			clear_user((void __user *) maddr, disp);
1088 			maddr += disp;
1089 		}
1090 
1091 		/* clear any space allocated but not loaded
1092 		 * - on uClinux we can just clear the lot
1093 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1094 		 *   extant in the file
1095 		 */
1096 		excess = phdr->p_memsz - phdr->p_filesz;
1097 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1098 
1099 #ifdef CONFIG_MMU
1100 		if (excess > excess1) {
1101 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1102 			unsigned long xmaddr;
1103 
1104 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1105 			down_write(&mm->mmap_sem);
1106 			xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1107 					 prot, flags, 0);
1108 			up_write(&mm->mmap_sem);
1109 
1110 			kdebug("mmap[%d] <anon>"
1111 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1112 			       loop, xaddr, excess - excess1, prot, flags,
1113 			       xmaddr);
1114 
1115 			if (xmaddr != xaddr)
1116 				return -ENOMEM;
1117 		}
1118 
1119 		if (prot & PROT_WRITE && excess1 > 0) {
1120 			kdebug("clear[%d] ad=%lx sz=%lx",
1121 			       loop, maddr + phdr->p_filesz, excess1);
1122 			clear_user((void __user *) maddr + phdr->p_filesz,
1123 				   excess1);
1124 		}
1125 
1126 #else
1127 		if (excess > 0) {
1128 			kdebug("clear[%d] ad=%lx sz=%lx",
1129 			       loop, maddr + phdr->p_filesz, excess);
1130 			clear_user((void *) maddr + phdr->p_filesz, excess);
1131 		}
1132 #endif
1133 
1134 		if (mm) {
1135 			if (phdr->p_flags & PF_X) {
1136 				if (!mm->start_code) {
1137 					mm->start_code = maddr;
1138 					mm->end_code = maddr + phdr->p_memsz;
1139 				}
1140 			} else if (!mm->start_data) {
1141 				mm->start_data = maddr;
1142 				mm->end_data = maddr + phdr->p_memsz;
1143 			}
1144 		}
1145 
1146 		seg++;
1147 	}
1148 
1149 	return 0;
1150 }
1151 
1152 /*****************************************************************************/
1153 /*
1154  * ELF-FDPIC core dumper
1155  *
1156  * Modelled on fs/exec.c:aout_core_dump()
1157  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1158  *
1159  * Modelled on fs/binfmt_elf.c core dumper
1160  */
1161 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1162 
1163 /*
1164  * These are the only things you should do on a core-file: use only these
1165  * functions to write out all the necessary info.
1166  */
1167 static int dump_write(struct file *file, const void *addr, int nr)
1168 {
1169 	return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1170 }
1171 
1172 static int dump_seek(struct file *file, loff_t off)
1173 {
1174 	if (file->f_op->llseek) {
1175 		if (file->f_op->llseek(file, off, SEEK_SET) != off)
1176 			return 0;
1177 	} else {
1178 		file->f_pos = off;
1179 	}
1180 	return 1;
1181 }
1182 
1183 /*
1184  * Decide whether a segment is worth dumping; default is yes to be
1185  * sure (missing info is worse than too much; etc).
1186  * Personally I'd include everything, and use the coredump limit...
1187  *
1188  * I think we should skip something. But I am not sure how. H.J.
1189  */
1190 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1191 {
1192 	int dump_ok;
1193 
1194 	/* Do not dump I/O mapped devices or special mappings */
1195 	if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1196 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1197 		return 0;
1198 	}
1199 
1200 	/* If we may not read the contents, don't allow us to dump
1201 	 * them either. "dump_write()" can't handle it anyway.
1202 	 */
1203 	if (!(vma->vm_flags & VM_READ)) {
1204 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1205 		return 0;
1206 	}
1207 
1208 	/* By default, dump shared memory if mapped from an anonymous file. */
1209 	if (vma->vm_flags & VM_SHARED) {
1210 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1211 			dump_ok = test_bit(MMF_DUMP_ANON_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 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1218 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1219 		       vma->vm_flags, dump_ok ? "yes" : "no");
1220 		return dump_ok;
1221 	}
1222 
1223 #ifdef CONFIG_MMU
1224 	/* By default, if it hasn't been written to, don't write it out */
1225 	if (!vma->anon_vma) {
1226 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1227 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1228 		       vma->vm_flags, dump_ok ? "yes" : "no");
1229 		return dump_ok;
1230 	}
1231 #endif
1232 
1233 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1234 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1235 	       dump_ok ? "yes" : "no");
1236 	return dump_ok;
1237 }
1238 
1239 /* An ELF note in memory */
1240 struct memelfnote
1241 {
1242 	const char *name;
1243 	int type;
1244 	unsigned int datasz;
1245 	void *data;
1246 };
1247 
1248 static int notesize(struct memelfnote *en)
1249 {
1250 	int sz;
1251 
1252 	sz = sizeof(struct elf_note);
1253 	sz += roundup(strlen(en->name) + 1, 4);
1254 	sz += roundup(en->datasz, 4);
1255 
1256 	return sz;
1257 }
1258 
1259 /* #define DEBUG */
1260 
1261 #define DUMP_WRITE(addr, nr)	\
1262 	do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1263 #define DUMP_SEEK(off)	\
1264 	do { if (!dump_seek(file, (off))) return 0; } while(0)
1265 
1266 static int writenote(struct memelfnote *men, struct file *file)
1267 {
1268 	struct elf_note en;
1269 
1270 	en.n_namesz = strlen(men->name) + 1;
1271 	en.n_descsz = men->datasz;
1272 	en.n_type = men->type;
1273 
1274 	DUMP_WRITE(&en, sizeof(en));
1275 	DUMP_WRITE(men->name, en.n_namesz);
1276 	/* XXX - cast from long long to long to avoid need for libgcc.a */
1277 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1278 	DUMP_WRITE(men->data, men->datasz);
1279 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1280 
1281 	return 1;
1282 }
1283 #undef DUMP_WRITE
1284 #undef DUMP_SEEK
1285 
1286 #define DUMP_WRITE(addr, nr)	\
1287 	if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1288 		goto end_coredump;
1289 #define DUMP_SEEK(off)	\
1290 	if (!dump_seek(file, (off))) \
1291 		goto end_coredump;
1292 
1293 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1294 {
1295 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1296 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1297 	elf->e_ident[EI_DATA] = ELF_DATA;
1298 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1299 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1300 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1301 
1302 	elf->e_type = ET_CORE;
1303 	elf->e_machine = ELF_ARCH;
1304 	elf->e_version = EV_CURRENT;
1305 	elf->e_entry = 0;
1306 	elf->e_phoff = sizeof(struct elfhdr);
1307 	elf->e_shoff = 0;
1308 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1309 	elf->e_ehsize = sizeof(struct elfhdr);
1310 	elf->e_phentsize = sizeof(struct elf_phdr);
1311 	elf->e_phnum = segs;
1312 	elf->e_shentsize = 0;
1313 	elf->e_shnum = 0;
1314 	elf->e_shstrndx = 0;
1315 	return;
1316 }
1317 
1318 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1319 {
1320 	phdr->p_type = PT_NOTE;
1321 	phdr->p_offset = offset;
1322 	phdr->p_vaddr = 0;
1323 	phdr->p_paddr = 0;
1324 	phdr->p_filesz = sz;
1325 	phdr->p_memsz = 0;
1326 	phdr->p_flags = 0;
1327 	phdr->p_align = 0;
1328 	return;
1329 }
1330 
1331 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1332 		unsigned int sz, void *data)
1333 {
1334 	note->name = name;
1335 	note->type = type;
1336 	note->datasz = sz;
1337 	note->data = data;
1338 	return;
1339 }
1340 
1341 /*
1342  * fill up all the fields in prstatus from the given task struct, except
1343  * registers which need to be filled up seperately.
1344  */
1345 static void fill_prstatus(struct elf_prstatus *prstatus,
1346 			  struct task_struct *p, long signr)
1347 {
1348 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1349 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1350 	prstatus->pr_sighold = p->blocked.sig[0];
1351 	prstatus->pr_pid = task_pid_vnr(p);
1352 	prstatus->pr_ppid = task_pid_vnr(p->parent);
1353 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1354 	prstatus->pr_sid = task_session_vnr(p);
1355 	if (thread_group_leader(p)) {
1356 		/*
1357 		 * This is the record for the group leader.  Add in the
1358 		 * cumulative times of previous dead threads.  This total
1359 		 * won't include the time of each live thread whose state
1360 		 * is included in the core dump.  The final total reported
1361 		 * to our parent process when it calls wait4 will include
1362 		 * those sums as well as the little bit more time it takes
1363 		 * this and each other thread to finish dying after the
1364 		 * core dump synchronization phase.
1365 		 */
1366 		cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1367 				   &prstatus->pr_utime);
1368 		cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1369 				   &prstatus->pr_stime);
1370 	} else {
1371 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1372 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1373 	}
1374 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1375 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1376 
1377 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1378 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1379 }
1380 
1381 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1382 		       struct mm_struct *mm)
1383 {
1384 	unsigned int i, len;
1385 
1386 	/* first copy the parameters from user space */
1387 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1388 
1389 	len = mm->arg_end - mm->arg_start;
1390 	if (len >= ELF_PRARGSZ)
1391 		len = ELF_PRARGSZ - 1;
1392 	if (copy_from_user(&psinfo->pr_psargs,
1393 		           (const char __user *) mm->arg_start, len))
1394 		return -EFAULT;
1395 	for (i = 0; i < len; i++)
1396 		if (psinfo->pr_psargs[i] == 0)
1397 			psinfo->pr_psargs[i] = ' ';
1398 	psinfo->pr_psargs[len] = 0;
1399 
1400 	psinfo->pr_pid = task_pid_vnr(p);
1401 	psinfo->pr_ppid = task_pid_vnr(p->parent);
1402 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1403 	psinfo->pr_sid = task_session_vnr(p);
1404 
1405 	i = p->state ? ffz(~p->state) + 1 : 0;
1406 	psinfo->pr_state = i;
1407 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1408 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1409 	psinfo->pr_nice = task_nice(p);
1410 	psinfo->pr_flag = p->flags;
1411 	SET_UID(psinfo->pr_uid, p->uid);
1412 	SET_GID(psinfo->pr_gid, p->gid);
1413 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1414 
1415 	return 0;
1416 }
1417 
1418 /* Here is the structure in which status of each thread is captured. */
1419 struct elf_thread_status
1420 {
1421 	struct list_head list;
1422 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1423 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1424 	struct task_struct *thread;
1425 #ifdef ELF_CORE_COPY_XFPREGS
1426 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1427 #endif
1428 	struct memelfnote notes[3];
1429 	int num_notes;
1430 };
1431 
1432 /*
1433  * In order to add the specific thread information for the elf file format,
1434  * we need to keep a linked list of every thread's pr_status and then create
1435  * a single section for them in the final core file.
1436  */
1437 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1438 {
1439 	struct task_struct *p = t->thread;
1440 	int sz = 0;
1441 
1442 	t->num_notes = 0;
1443 
1444 	fill_prstatus(&t->prstatus, p, signr);
1445 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1446 
1447 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1448 		  &t->prstatus);
1449 	t->num_notes++;
1450 	sz += notesize(&t->notes[0]);
1451 
1452 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1453 	if (t->prstatus.pr_fpvalid) {
1454 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1455 			  &t->fpu);
1456 		t->num_notes++;
1457 		sz += notesize(&t->notes[1]);
1458 	}
1459 
1460 #ifdef ELF_CORE_COPY_XFPREGS
1461 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1462 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1463 			  sizeof(t->xfpu), &t->xfpu);
1464 		t->num_notes++;
1465 		sz += notesize(&t->notes[2]);
1466 	}
1467 #endif
1468 	return sz;
1469 }
1470 
1471 /*
1472  * dump the segments for an MMU process
1473  */
1474 #ifdef CONFIG_MMU
1475 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1476 			   unsigned long *limit, unsigned long mm_flags)
1477 {
1478 	struct vm_area_struct *vma;
1479 
1480 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1481 		unsigned long addr;
1482 
1483 		if (!maydump(vma, mm_flags))
1484 			continue;
1485 
1486 		for (addr = vma->vm_start;
1487 		     addr < vma->vm_end;
1488 		     addr += PAGE_SIZE
1489 		     ) {
1490 			struct vm_area_struct *vma;
1491 			struct page *page;
1492 
1493 			if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1494 					   &page, &vma) <= 0) {
1495 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1496 			}
1497 			else if (page == ZERO_PAGE(0)) {
1498 				page_cache_release(page);
1499 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1500 			}
1501 			else {
1502 				void *kaddr;
1503 
1504 				flush_cache_page(vma, addr, page_to_pfn(page));
1505 				kaddr = kmap(page);
1506 				if ((*size += PAGE_SIZE) > *limit ||
1507 				    !dump_write(file, kaddr, PAGE_SIZE)
1508 				    ) {
1509 					kunmap(page);
1510 					page_cache_release(page);
1511 					return -EIO;
1512 				}
1513 				kunmap(page);
1514 				page_cache_release(page);
1515 			}
1516 		}
1517 	}
1518 
1519 	return 0;
1520 
1521 end_coredump:
1522 	return -EFBIG;
1523 }
1524 #endif
1525 
1526 /*
1527  * dump the segments for a NOMMU process
1528  */
1529 #ifndef CONFIG_MMU
1530 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1531 			   unsigned long *limit, unsigned long mm_flags)
1532 {
1533 	struct vm_list_struct *vml;
1534 
1535 	for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1536 	struct vm_area_struct *vma = vml->vma;
1537 
1538 		if (!maydump(vma, mm_flags))
1539 			continue;
1540 
1541 		if ((*size += PAGE_SIZE) > *limit)
1542 			return -EFBIG;
1543 
1544 		if (!dump_write(file, (void *) vma->vm_start,
1545 				vma->vm_end - vma->vm_start))
1546 			return -EIO;
1547 	}
1548 
1549 	return 0;
1550 }
1551 #endif
1552 
1553 /*
1554  * Actual dumper
1555  *
1556  * This is a two-pass process; first we find the offsets of the bits,
1557  * and then they are actually written out.  If we run out of core limit
1558  * we just truncate.
1559  */
1560 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1561 			       struct file *file, unsigned long limit)
1562 {
1563 #define	NUM_NOTES	6
1564 	int has_dumped = 0;
1565 	mm_segment_t fs;
1566 	int segs;
1567 	size_t size = 0;
1568 	int i;
1569 	struct vm_area_struct *vma;
1570 	struct elfhdr *elf = NULL;
1571 	loff_t offset = 0, dataoff;
1572 	int numnote;
1573 	struct memelfnote *notes = NULL;
1574 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1575 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1576  	struct task_struct *g, *p;
1577  	LIST_HEAD(thread_list);
1578  	struct list_head *t;
1579 	elf_fpregset_t *fpu = NULL;
1580 #ifdef ELF_CORE_COPY_XFPREGS
1581 	elf_fpxregset_t *xfpu = NULL;
1582 #endif
1583 	int thread_status_size = 0;
1584 #ifndef CONFIG_MMU
1585 	struct vm_list_struct *vml;
1586 #endif
1587 	elf_addr_t *auxv;
1588 	unsigned long mm_flags;
1589 
1590 	/*
1591 	 * We no longer stop all VM operations.
1592 	 *
1593 	 * This is because those proceses that could possibly change map_count
1594 	 * or the mmap / vma pages are now blocked in do_exit on current
1595 	 * finishing this core dump.
1596 	 *
1597 	 * Only ptrace can touch these memory addresses, but it doesn't change
1598 	 * the map_count or the pages allocated. So no possibility of crashing
1599 	 * exists while dumping the mm->vm_next areas to the core file.
1600 	 */
1601 
1602 	/* alloc memory for large data structures: too large to be on stack */
1603 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1604 	if (!elf)
1605 		goto cleanup;
1606 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1607 	if (!prstatus)
1608 		goto cleanup;
1609 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1610 	if (!psinfo)
1611 		goto cleanup;
1612 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1613 	if (!notes)
1614 		goto cleanup;
1615 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1616 	if (!fpu)
1617 		goto cleanup;
1618 #ifdef ELF_CORE_COPY_XFPREGS
1619 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1620 	if (!xfpu)
1621 		goto cleanup;
1622 #endif
1623 
1624 	if (signr) {
1625 		struct elf_thread_status *tmp;
1626 		rcu_read_lock();
1627 		do_each_thread(g,p)
1628 			if (current->mm == p->mm && current != p) {
1629 				tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1630 				if (!tmp) {
1631 					rcu_read_unlock();
1632 					goto cleanup;
1633 				}
1634 				tmp->thread = p;
1635 				list_add(&tmp->list, &thread_list);
1636 			}
1637 		while_each_thread(g,p);
1638 		rcu_read_unlock();
1639 		list_for_each(t, &thread_list) {
1640 			struct elf_thread_status *tmp;
1641 			int sz;
1642 
1643 			tmp = list_entry(t, struct elf_thread_status, list);
1644 			sz = elf_dump_thread_status(signr, tmp);
1645 			thread_status_size += sz;
1646 		}
1647 	}
1648 
1649 	/* now collect the dump for the current */
1650 	fill_prstatus(prstatus, current, signr);
1651 	elf_core_copy_regs(&prstatus->pr_reg, regs);
1652 
1653 #ifdef CONFIG_MMU
1654 	segs = current->mm->map_count;
1655 #else
1656 	segs = 0;
1657 	for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1658 	    segs++;
1659 #endif
1660 #ifdef ELF_CORE_EXTRA_PHDRS
1661 	segs += ELF_CORE_EXTRA_PHDRS;
1662 #endif
1663 
1664 	/* Set up header */
1665 	fill_elf_fdpic_header(elf, segs + 1);	/* including notes section */
1666 
1667 	has_dumped = 1;
1668 	current->flags |= PF_DUMPCORE;
1669 
1670 	/*
1671 	 * Set up the notes in similar form to SVR4 core dumps made
1672 	 * with info from their /proc.
1673 	 */
1674 
1675 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1676 	fill_psinfo(psinfo, current->group_leader, current->mm);
1677 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1678 
1679 	numnote = 2;
1680 
1681 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1682 
1683 	i = 0;
1684 	do
1685 		i += 2;
1686 	while (auxv[i - 2] != AT_NULL);
1687 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1688 		  i * sizeof(elf_addr_t), auxv);
1689 
1690   	/* Try to dump the FPU. */
1691 	if ((prstatus->pr_fpvalid =
1692 	     elf_core_copy_task_fpregs(current, regs, fpu)))
1693 		fill_note(notes + numnote++,
1694 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1695 #ifdef ELF_CORE_COPY_XFPREGS
1696 	if (elf_core_copy_task_xfpregs(current, xfpu))
1697 		fill_note(notes + numnote++,
1698 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1699 #endif
1700 
1701 	fs = get_fs();
1702 	set_fs(KERNEL_DS);
1703 
1704 	DUMP_WRITE(elf, sizeof(*elf));
1705 	offset += sizeof(*elf);				/* Elf header */
1706 	offset += (segs+1) * sizeof(struct elf_phdr);	/* Program headers */
1707 
1708 	/* Write notes phdr entry */
1709 	{
1710 		struct elf_phdr phdr;
1711 		int sz = 0;
1712 
1713 		for (i = 0; i < numnote; i++)
1714 			sz += notesize(notes + i);
1715 
1716 		sz += thread_status_size;
1717 
1718 		fill_elf_note_phdr(&phdr, sz, offset);
1719 		offset += sz;
1720 		DUMP_WRITE(&phdr, sizeof(phdr));
1721 	}
1722 
1723 	/* Page-align dumped data */
1724 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1725 
1726 	/*
1727 	 * We must use the same mm->flags while dumping core to avoid
1728 	 * inconsistency between the program headers and bodies, otherwise an
1729 	 * unusable core file can be generated.
1730 	 */
1731 	mm_flags = current->mm->flags;
1732 
1733 	/* write program headers for segments dump */
1734 	for (
1735 #ifdef CONFIG_MMU
1736 		vma = current->mm->mmap; vma; vma = vma->vm_next
1737 #else
1738 			vml = current->mm->context.vmlist; vml; vml = vml->next
1739 #endif
1740 	     ) {
1741 		struct elf_phdr phdr;
1742 		size_t sz;
1743 
1744 #ifndef CONFIG_MMU
1745 		vma = vml->vma;
1746 #endif
1747 
1748 		sz = vma->vm_end - vma->vm_start;
1749 
1750 		phdr.p_type = PT_LOAD;
1751 		phdr.p_offset = offset;
1752 		phdr.p_vaddr = vma->vm_start;
1753 		phdr.p_paddr = 0;
1754 		phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1755 		phdr.p_memsz = sz;
1756 		offset += phdr.p_filesz;
1757 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1758 		if (vma->vm_flags & VM_WRITE)
1759 			phdr.p_flags |= PF_W;
1760 		if (vma->vm_flags & VM_EXEC)
1761 			phdr.p_flags |= PF_X;
1762 		phdr.p_align = ELF_EXEC_PAGESIZE;
1763 
1764 		DUMP_WRITE(&phdr, sizeof(phdr));
1765 	}
1766 
1767 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1768 	ELF_CORE_WRITE_EXTRA_PHDRS;
1769 #endif
1770 
1771  	/* write out the notes section */
1772 	for (i = 0; i < numnote; i++)
1773 		if (!writenote(notes + i, file))
1774 			goto end_coredump;
1775 
1776 	/* write out the thread status notes section */
1777 	list_for_each(t, &thread_list) {
1778 		struct elf_thread_status *tmp =
1779 				list_entry(t, struct elf_thread_status, list);
1780 
1781 		for (i = 0; i < tmp->num_notes; i++)
1782 			if (!writenote(&tmp->notes[i], file))
1783 				goto end_coredump;
1784 	}
1785 
1786 	DUMP_SEEK(dataoff);
1787 
1788 	if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1789 		goto end_coredump;
1790 
1791 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1792 	ELF_CORE_WRITE_EXTRA_DATA;
1793 #endif
1794 
1795 	if (file->f_pos != offset) {
1796 		/* Sanity check */
1797 		printk(KERN_WARNING
1798 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1799 		       file->f_pos, offset);
1800 	}
1801 
1802 end_coredump:
1803 	set_fs(fs);
1804 
1805 cleanup:
1806 	while (!list_empty(&thread_list)) {
1807 		struct list_head *tmp = thread_list.next;
1808 		list_del(tmp);
1809 		kfree(list_entry(tmp, struct elf_thread_status, list));
1810 	}
1811 
1812 	kfree(elf);
1813 	kfree(prstatus);
1814 	kfree(psinfo);
1815 	kfree(notes);
1816 	kfree(fpu);
1817 #ifdef ELF_CORE_COPY_XFPREGS
1818 	kfree(xfpu);
1819 #endif
1820 	return has_dumped;
1821 #undef NUM_NOTES
1822 }
1823 
1824 #endif		/* USE_ELF_CORE_DUMP */
1825