xref: /linux/fs/binfmt_elf_fdpic.c (revision 7025bec9125b0a02edcaf22c2dce753bf2c95480)
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/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37 
38 #include <asm/uaccess.h>
39 #include <asm/param.h>
40 #include <asm/pgalloc.h>
41 
42 typedef char *elf_caddr_t;
43 
44 #if 0
45 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
46 #else
47 #define kdebug(fmt, ...) do {} while(0)
48 #endif
49 
50 #if 0
51 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
52 #else
53 #define kdcore(fmt, ...) do {} while(0)
54 #endif
55 
56 MODULE_LICENSE("GPL");
57 
58 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
59 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
60 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
61 			      struct mm_struct *, const char *);
62 
63 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
64 				   struct elf_fdpic_params *,
65 				   struct elf_fdpic_params *);
66 
67 #ifndef CONFIG_MMU
68 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
69 					    unsigned long *);
70 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
71 						   struct file *,
72 						   struct mm_struct *);
73 #endif
74 
75 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
76 					     struct file *, struct mm_struct *);
77 
78 #ifdef CONFIG_ELF_CORE
79 static int elf_fdpic_core_dump(struct coredump_params *cprm);
80 #endif
81 
82 static struct linux_binfmt elf_fdpic_format = {
83 	.module		= THIS_MODULE,
84 	.load_binary	= load_elf_fdpic_binary,
85 #ifdef CONFIG_ELF_CORE
86 	.core_dump	= elf_fdpic_core_dump,
87 #endif
88 	.min_coredump	= ELF_EXEC_PAGESIZE,
89 };
90 
91 static int __init init_elf_fdpic_binfmt(void)
92 {
93 	return register_binfmt(&elf_fdpic_format);
94 }
95 
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 	unregister_binfmt(&elf_fdpic_format);
99 }
100 
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103 
104 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
105 {
106 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 		return 0;
108 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 		return 0;
110 	if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
111 		return 0;
112 	if (!file->f_op || !file->f_op->mmap)
113 		return 0;
114 	return 1;
115 }
116 
117 /*****************************************************************************/
118 /*
119  * read the program headers table into memory
120  */
121 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
122 				 struct file *file)
123 {
124 	struct elf32_phdr *phdr;
125 	unsigned long size;
126 	int retval, loop;
127 
128 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
129 		return -ENOMEM;
130 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
131 		return -ENOMEM;
132 
133 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
134 	params->phdrs = kmalloc(size, GFP_KERNEL);
135 	if (!params->phdrs)
136 		return -ENOMEM;
137 
138 	retval = kernel_read(file, params->hdr.e_phoff,
139 			     (char *) params->phdrs, size);
140 	if (unlikely(retval != size))
141 		return retval < 0 ? retval : -ENOEXEC;
142 
143 	/* determine stack size for this binary */
144 	phdr = params->phdrs;
145 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
146 		if (phdr->p_type != PT_GNU_STACK)
147 			continue;
148 
149 		if (phdr->p_flags & PF_X)
150 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
151 		else
152 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
153 
154 		params->stack_size = phdr->p_memsz;
155 		break;
156 	}
157 
158 	return 0;
159 }
160 
161 /*****************************************************************************/
162 /*
163  * load an fdpic binary into various bits of memory
164  */
165 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
166 				 struct pt_regs *regs)
167 {
168 	struct elf_fdpic_params exec_params, interp_params;
169 	struct elf_phdr *phdr;
170 	unsigned long stack_size, entryaddr;
171 #ifdef ELF_FDPIC_PLAT_INIT
172 	unsigned long dynaddr;
173 #endif
174 	struct file *interpreter = NULL; /* to shut gcc up */
175 	char *interpreter_name = NULL;
176 	int executable_stack;
177 	int retval, i;
178 
179 	kdebug("____ LOAD %d ____", current->pid);
180 
181 	memset(&exec_params, 0, sizeof(exec_params));
182 	memset(&interp_params, 0, sizeof(interp_params));
183 
184 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
185 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
186 
187 	/* check that this is a binary we know how to deal with */
188 	retval = -ENOEXEC;
189 	if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
190 		goto error;
191 
192 	/* read the program header table */
193 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
194 	if (retval < 0)
195 		goto error;
196 
197 	/* scan for a program header that specifies an interpreter */
198 	phdr = exec_params.phdrs;
199 
200 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
201 		switch (phdr->p_type) {
202 		case PT_INTERP:
203 			retval = -ENOMEM;
204 			if (phdr->p_filesz > PATH_MAX)
205 				goto error;
206 			retval = -ENOENT;
207 			if (phdr->p_filesz < 2)
208 				goto error;
209 
210 			/* read the name of the interpreter into memory */
211 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
212 			if (!interpreter_name)
213 				goto error;
214 
215 			retval = kernel_read(bprm->file,
216 					     phdr->p_offset,
217 					     interpreter_name,
218 					     phdr->p_filesz);
219 			if (unlikely(retval != phdr->p_filesz)) {
220 				if (retval >= 0)
221 					retval = -ENOEXEC;
222 				goto error;
223 			}
224 
225 			retval = -ENOENT;
226 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
227 				goto error;
228 
229 			kdebug("Using ELF interpreter %s", interpreter_name);
230 
231 			/* replace the program with the interpreter */
232 			interpreter = open_exec(interpreter_name);
233 			retval = PTR_ERR(interpreter);
234 			if (IS_ERR(interpreter)) {
235 				interpreter = NULL;
236 				goto error;
237 			}
238 
239 			/*
240 			 * If the binary is not readable then enforce
241 			 * mm->dumpable = 0 regardless of the interpreter's
242 			 * permissions.
243 			 */
244 			if (file_permission(interpreter, MAY_READ) < 0)
245 				bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
246 
247 			retval = kernel_read(interpreter, 0, bprm->buf,
248 					     BINPRM_BUF_SIZE);
249 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
250 				if (retval >= 0)
251 					retval = -ENOEXEC;
252 				goto error;
253 			}
254 
255 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
256 			break;
257 
258 		case PT_LOAD:
259 #ifdef CONFIG_MMU
260 			if (exec_params.load_addr == 0)
261 				exec_params.load_addr = phdr->p_vaddr;
262 #endif
263 			break;
264 		}
265 
266 	}
267 
268 	if (elf_check_const_displacement(&exec_params.hdr))
269 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
270 
271 	/* perform insanity checks on the interpreter */
272 	if (interpreter_name) {
273 		retval = -ELIBBAD;
274 		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
275 			goto error;
276 
277 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
278 
279 		/* read the interpreter's program header table */
280 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
281 		if (retval < 0)
282 			goto error;
283 	}
284 
285 	stack_size = exec_params.stack_size;
286 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
287 		executable_stack = EXSTACK_ENABLE_X;
288 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
289 		executable_stack = EXSTACK_DISABLE_X;
290 	else
291 		executable_stack = EXSTACK_DEFAULT;
292 
293 	if (stack_size == 0) {
294 		stack_size = interp_params.stack_size;
295 		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 
303 	retval = -ENOEXEC;
304 	if (stack_size == 0)
305 		goto error;
306 
307 	if (elf_check_const_displacement(&interp_params.hdr))
308 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
309 
310 	/* flush all traces of the currently running executable */
311 	retval = flush_old_exec(bprm);
312 	if (retval)
313 		goto error;
314 
315 	/* there's now no turning back... the old userspace image is dead,
316 	 * defunct, deceased, etc. after this point we have to exit via
317 	 * error_kill */
318 	set_personality(PER_LINUX_FDPIC);
319 	set_binfmt(&elf_fdpic_format);
320 
321 	current->mm->start_code = 0;
322 	current->mm->end_code = 0;
323 	current->mm->start_stack = 0;
324 	current->mm->start_data = 0;
325 	current->mm->end_data = 0;
326 	current->mm->context.exec_fdpic_loadmap = 0;
327 	current->mm->context.interp_fdpic_loadmap = 0;
328 
329 	current->flags &= ~PF_FORKNOEXEC;
330 
331 #ifdef CONFIG_MMU
332 	elf_fdpic_arch_lay_out_mm(&exec_params,
333 				  &interp_params,
334 				  &current->mm->start_stack,
335 				  &current->mm->start_brk);
336 
337 	retval = setup_arg_pages(bprm, current->mm->start_stack,
338 				 executable_stack);
339 	if (retval < 0) {
340 		send_sig(SIGKILL, current, 0);
341 		goto error_kill;
342 	}
343 #endif
344 
345 	/* load the executable and interpreter into memory */
346 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
347 				    "executable");
348 	if (retval < 0)
349 		goto error_kill;
350 
351 	if (interpreter_name) {
352 		retval = elf_fdpic_map_file(&interp_params, interpreter,
353 					    current->mm, "interpreter");
354 		if (retval < 0) {
355 			printk(KERN_ERR "Unable to load interpreter\n");
356 			goto error_kill;
357 		}
358 
359 		allow_write_access(interpreter);
360 		fput(interpreter);
361 		interpreter = NULL;
362 	}
363 
364 #ifdef CONFIG_MMU
365 	if (!current->mm->start_brk)
366 		current->mm->start_brk = current->mm->end_data;
367 
368 	current->mm->brk = current->mm->start_brk =
369 		PAGE_ALIGN(current->mm->start_brk);
370 
371 #else
372 	/* create a stack and brk area big enough for everyone
373 	 * - the brk heap starts at the bottom and works up
374 	 * - the stack starts at the top and works down
375 	 */
376 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
377 	if (stack_size < PAGE_SIZE * 2)
378 		stack_size = PAGE_SIZE * 2;
379 
380 	down_write(&current->mm->mmap_sem);
381 	current->mm->start_brk = do_mmap(NULL, 0, stack_size,
382 					 PROT_READ | PROT_WRITE | PROT_EXEC,
383 					 MAP_PRIVATE | MAP_ANONYMOUS |
384 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
385 					 0);
386 
387 	if (IS_ERR_VALUE(current->mm->start_brk)) {
388 		up_write(&current->mm->mmap_sem);
389 		retval = current->mm->start_brk;
390 		current->mm->start_brk = 0;
391 		goto error_kill;
392 	}
393 
394 	up_write(&current->mm->mmap_sem);
395 
396 	current->mm->brk = current->mm->start_brk;
397 	current->mm->context.end_brk = current->mm->start_brk;
398 	current->mm->context.end_brk +=
399 		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
400 	current->mm->start_stack = current->mm->start_brk + stack_size;
401 #endif
402 
403 	install_exec_creds(bprm);
404 	current->flags &= ~PF_FORKNOEXEC;
405 	if (create_elf_fdpic_tables(bprm, current->mm,
406 				    &exec_params, &interp_params) < 0)
407 		goto error_kill;
408 
409 	kdebug("- start_code  %lx", current->mm->start_code);
410 	kdebug("- end_code    %lx", current->mm->end_code);
411 	kdebug("- start_data  %lx", current->mm->start_data);
412 	kdebug("- end_data    %lx", current->mm->end_data);
413 	kdebug("- start_brk   %lx", current->mm->start_brk);
414 	kdebug("- brk         %lx", current->mm->brk);
415 	kdebug("- start_stack %lx", current->mm->start_stack);
416 
417 #ifdef ELF_FDPIC_PLAT_INIT
418 	/*
419 	 * The ABI may specify that certain registers be set up in special
420 	 * ways (on i386 %edx is the address of a DT_FINI function, for
421 	 * example.  This macro performs whatever initialization to
422 	 * the regs structure is required.
423 	 */
424 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
425 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
426 			    dynaddr);
427 #endif
428 
429 	/* everything is now ready... get the userspace context ready to roll */
430 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
431 	start_thread(regs, entryaddr, current->mm->start_stack);
432 
433 	retval = 0;
434 
435 error:
436 	if (interpreter) {
437 		allow_write_access(interpreter);
438 		fput(interpreter);
439 	}
440 	kfree(interpreter_name);
441 	kfree(exec_params.phdrs);
442 	kfree(exec_params.loadmap);
443 	kfree(interp_params.phdrs);
444 	kfree(interp_params.loadmap);
445 	return retval;
446 
447 	/* unrecoverable error - kill the process */
448 error_kill:
449 	send_sig(SIGSEGV, current, 0);
450 	goto error;
451 
452 }
453 
454 /*****************************************************************************/
455 
456 #ifndef ELF_BASE_PLATFORM
457 /*
458  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
459  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
460  * will be copied to the user stack in the same manner as AT_PLATFORM.
461  */
462 #define ELF_BASE_PLATFORM NULL
463 #endif
464 
465 /*
466  * present useful information to the program by shovelling it onto the new
467  * process's stack
468  */
469 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
470 				   struct mm_struct *mm,
471 				   struct elf_fdpic_params *exec_params,
472 				   struct elf_fdpic_params *interp_params)
473 {
474 	const struct cred *cred = current_cred();
475 	unsigned long sp, csp, nitems;
476 	elf_caddr_t __user *argv, *envp;
477 	size_t platform_len = 0, len;
478 	char *k_platform, *k_base_platform;
479 	char __user *u_platform, *u_base_platform, *p;
480 	long hwcap;
481 	int loop;
482 	int nr;	/* reset for each csp adjustment */
483 
484 #ifdef CONFIG_MMU
485 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
486 	 * by the processes running on the same package. One thing we can do is
487 	 * to shuffle the initial stack for them, so we give the architecture
488 	 * an opportunity to do so here.
489 	 */
490 	sp = arch_align_stack(bprm->p);
491 #else
492 	sp = mm->start_stack;
493 
494 	/* stack the program arguments and environment */
495 	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
496 		return -EFAULT;
497 #endif
498 
499 	hwcap = ELF_HWCAP;
500 
501 	/*
502 	 * If this architecture has a platform capability string, copy it
503 	 * to userspace.  In some cases (Sparc), this info is impossible
504 	 * for userspace to get any other way, in others (i386) it is
505 	 * merely difficult.
506 	 */
507 	k_platform = ELF_PLATFORM;
508 	u_platform = NULL;
509 
510 	if (k_platform) {
511 		platform_len = strlen(k_platform) + 1;
512 		sp -= platform_len;
513 		u_platform = (char __user *) sp;
514 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
515 			return -EFAULT;
516 	}
517 
518 	/*
519 	 * If this architecture has a "base" platform capability
520 	 * string, copy it to userspace.
521 	 */
522 	k_base_platform = ELF_BASE_PLATFORM;
523 	u_base_platform = NULL;
524 
525 	if (k_base_platform) {
526 		platform_len = strlen(k_base_platform) + 1;
527 		sp -= platform_len;
528 		u_base_platform = (char __user *) sp;
529 		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
530 			return -EFAULT;
531 	}
532 
533 	sp &= ~7UL;
534 
535 	/* stack the load map(s) */
536 	len = sizeof(struct elf32_fdpic_loadmap);
537 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
538 	sp = (sp - len) & ~7UL;
539 	exec_params->map_addr = sp;
540 
541 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
542 		return -EFAULT;
543 
544 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
545 
546 	if (interp_params->loadmap) {
547 		len = sizeof(struct elf32_fdpic_loadmap);
548 		len += sizeof(struct elf32_fdpic_loadseg) *
549 			interp_params->loadmap->nsegs;
550 		sp = (sp - len) & ~7UL;
551 		interp_params->map_addr = sp;
552 
553 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
554 				 len) != 0)
555 			return -EFAULT;
556 
557 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
558 	}
559 
560 	/* force 16 byte _final_ alignment here for generality */
561 #define DLINFO_ITEMS 15
562 
563 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
564 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
565 
566 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
567 		nitems++;
568 
569 	csp = sp;
570 	sp -= nitems * 2 * sizeof(unsigned long);
571 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
572 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
573 	sp -= 1 * sizeof(unsigned long);		/* argc */
574 
575 	csp -= sp & 15UL;
576 	sp -= sp & 15UL;
577 
578 	/* put the ELF interpreter info on the stack */
579 #define NEW_AUX_ENT(id, val)						\
580 	do {								\
581 		struct { unsigned long _id, _val; } __user *ent;	\
582 									\
583 		ent = (void __user *) csp;				\
584 		__put_user((id), &ent[nr]._id);				\
585 		__put_user((val), &ent[nr]._val);			\
586 		nr++;							\
587 	} while (0)
588 
589 	nr = 0;
590 	csp -= 2 * sizeof(unsigned long);
591 	NEW_AUX_ENT(AT_NULL, 0);
592 	if (k_platform) {
593 		nr = 0;
594 		csp -= 2 * sizeof(unsigned long);
595 		NEW_AUX_ENT(AT_PLATFORM,
596 			    (elf_addr_t) (unsigned long) u_platform);
597 	}
598 
599 	if (k_base_platform) {
600 		nr = 0;
601 		csp -= 2 * sizeof(unsigned long);
602 		NEW_AUX_ENT(AT_BASE_PLATFORM,
603 			    (elf_addr_t) (unsigned long) u_base_platform);
604 	}
605 
606 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
607 		nr = 0;
608 		csp -= 2 * sizeof(unsigned long);
609 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
610 	}
611 
612 	nr = 0;
613 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
614 	NEW_AUX_ENT(AT_HWCAP,	hwcap);
615 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
616 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
617 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
618 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
619 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
620 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
621 	NEW_AUX_ENT(AT_FLAGS,	0);
622 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
623 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) cred->uid);
624 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) cred->euid);
625 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) cred->gid);
626 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) cred->egid);
627 	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
628 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
629 
630 #ifdef ARCH_DLINFO
631 	nr = 0;
632 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
633 
634 	/* ARCH_DLINFO must come last so platform specific code can enforce
635 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
636 	 */
637 	ARCH_DLINFO;
638 #endif
639 #undef NEW_AUX_ENT
640 
641 	/* allocate room for argv[] and envv[] */
642 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
643 	envp = (elf_caddr_t __user *) csp;
644 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
645 	argv = (elf_caddr_t __user *) csp;
646 
647 	/* stack argc */
648 	csp -= sizeof(unsigned long);
649 	__put_user(bprm->argc, (unsigned long __user *) csp);
650 
651 	BUG_ON(csp != sp);
652 
653 	/* fill in the argv[] array */
654 #ifdef CONFIG_MMU
655 	current->mm->arg_start = bprm->p;
656 #else
657 	current->mm->arg_start = current->mm->start_stack -
658 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
659 #endif
660 
661 	p = (char __user *) current->mm->arg_start;
662 	for (loop = bprm->argc; loop > 0; loop--) {
663 		__put_user((elf_caddr_t) p, argv++);
664 		len = strnlen_user(p, MAX_ARG_STRLEN);
665 		if (!len || len > MAX_ARG_STRLEN)
666 			return -EINVAL;
667 		p += len;
668 	}
669 	__put_user(NULL, argv);
670 	current->mm->arg_end = (unsigned long) p;
671 
672 	/* fill in the envv[] array */
673 	current->mm->env_start = (unsigned long) p;
674 	for (loop = bprm->envc; loop > 0; loop--) {
675 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
676 		len = strnlen_user(p, MAX_ARG_STRLEN);
677 		if (!len || len > MAX_ARG_STRLEN)
678 			return -EINVAL;
679 		p += len;
680 	}
681 	__put_user(NULL, envp);
682 	current->mm->env_end = (unsigned long) p;
683 
684 	mm->start_stack = (unsigned long) sp;
685 	return 0;
686 }
687 
688 /*****************************************************************************/
689 /*
690  * transfer the program arguments and environment from the holding pages onto
691  * the stack
692  */
693 #ifndef CONFIG_MMU
694 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
695 					    unsigned long *_sp)
696 {
697 	unsigned long index, stop, sp;
698 	char *src;
699 	int ret = 0;
700 
701 	stop = bprm->p >> PAGE_SHIFT;
702 	sp = *_sp;
703 
704 	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
705 		src = kmap(bprm->page[index]);
706 		sp -= PAGE_SIZE;
707 		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
708 			ret = -EFAULT;
709 		kunmap(bprm->page[index]);
710 		if (ret < 0)
711 			goto out;
712 	}
713 
714 	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
715 
716 out:
717 	return ret;
718 }
719 #endif
720 
721 /*****************************************************************************/
722 /*
723  * load the appropriate binary image (executable or interpreter) into memory
724  * - we assume no MMU is available
725  * - if no other PIC bits are set in params->hdr->e_flags
726  *   - we assume that the LOADable segments in the binary are independently relocatable
727  *   - we assume R/O executable segments are shareable
728  * - else
729  *   - we assume the loadable parts of the image to require fixed displacement
730  *   - the image is not shareable
731  */
732 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
733 			      struct file *file,
734 			      struct mm_struct *mm,
735 			      const char *what)
736 {
737 	struct elf32_fdpic_loadmap *loadmap;
738 #ifdef CONFIG_MMU
739 	struct elf32_fdpic_loadseg *mseg;
740 #endif
741 	struct elf32_fdpic_loadseg *seg;
742 	struct elf32_phdr *phdr;
743 	unsigned long load_addr, stop;
744 	unsigned nloads, tmp;
745 	size_t size;
746 	int loop, ret;
747 
748 	/* allocate a load map table */
749 	nloads = 0;
750 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
751 		if (params->phdrs[loop].p_type == PT_LOAD)
752 			nloads++;
753 
754 	if (nloads == 0)
755 		return -ELIBBAD;
756 
757 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
758 	loadmap = kzalloc(size, GFP_KERNEL);
759 	if (!loadmap)
760 		return -ENOMEM;
761 
762 	params->loadmap = loadmap;
763 
764 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
765 	loadmap->nsegs = nloads;
766 
767 	load_addr = params->load_addr;
768 	seg = loadmap->segs;
769 
770 	/* map the requested LOADs into the memory space */
771 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
772 	case ELF_FDPIC_FLAG_CONSTDISP:
773 	case ELF_FDPIC_FLAG_CONTIGUOUS:
774 #ifndef CONFIG_MMU
775 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
776 		if (ret < 0)
777 			return ret;
778 		break;
779 #endif
780 	default:
781 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
782 		if (ret < 0)
783 			return ret;
784 		break;
785 	}
786 
787 	/* map the entry point */
788 	if (params->hdr.e_entry) {
789 		seg = loadmap->segs;
790 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
791 			if (params->hdr.e_entry >= seg->p_vaddr &&
792 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
793 				params->entry_addr =
794 					(params->hdr.e_entry - seg->p_vaddr) +
795 					seg->addr;
796 				break;
797 			}
798 		}
799 	}
800 
801 	/* determine where the program header table has wound up if mapped */
802 	stop = params->hdr.e_phoff;
803 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
804 	phdr = params->phdrs;
805 
806 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
807 		if (phdr->p_type != PT_LOAD)
808 			continue;
809 
810 		if (phdr->p_offset > params->hdr.e_phoff ||
811 		    phdr->p_offset + phdr->p_filesz < stop)
812 			continue;
813 
814 		seg = loadmap->segs;
815 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
816 			if (phdr->p_vaddr >= seg->p_vaddr &&
817 			    phdr->p_vaddr + phdr->p_filesz <=
818 			    seg->p_vaddr + seg->p_memsz) {
819 				params->ph_addr =
820 					(phdr->p_vaddr - seg->p_vaddr) +
821 					seg->addr +
822 					params->hdr.e_phoff - phdr->p_offset;
823 				break;
824 			}
825 		}
826 		break;
827 	}
828 
829 	/* determine where the dynamic section has wound up if there is one */
830 	phdr = params->phdrs;
831 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
832 		if (phdr->p_type != PT_DYNAMIC)
833 			continue;
834 
835 		seg = loadmap->segs;
836 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
837 			if (phdr->p_vaddr >= seg->p_vaddr &&
838 			    phdr->p_vaddr + phdr->p_memsz <=
839 			    seg->p_vaddr + seg->p_memsz) {
840 				params->dynamic_addr =
841 					(phdr->p_vaddr - seg->p_vaddr) +
842 					seg->addr;
843 
844 				/* check the dynamic section contains at least
845 				 * one item, and that the last item is a NULL
846 				 * entry */
847 				if (phdr->p_memsz == 0 ||
848 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
849 					goto dynamic_error;
850 
851 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
852 				if (((Elf32_Dyn *)
853 				     params->dynamic_addr)[tmp - 1].d_tag != 0)
854 					goto dynamic_error;
855 				break;
856 			}
857 		}
858 		break;
859 	}
860 
861 	/* now elide adjacent segments in the load map on MMU linux
862 	 * - on uClinux the holes between may actually be filled with system
863 	 *   stuff or stuff from other processes
864 	 */
865 #ifdef CONFIG_MMU
866 	nloads = loadmap->nsegs;
867 	mseg = loadmap->segs;
868 	seg = mseg + 1;
869 	for (loop = 1; loop < nloads; loop++) {
870 		/* see if we have a candidate for merging */
871 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
872 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
873 			if (load_addr == (seg->addr & PAGE_MASK)) {
874 				mseg->p_memsz +=
875 					load_addr -
876 					(mseg->addr + mseg->p_memsz);
877 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
878 				mseg->p_memsz += seg->p_memsz;
879 				loadmap->nsegs--;
880 				continue;
881 			}
882 		}
883 
884 		mseg++;
885 		if (mseg != seg)
886 			*mseg = *seg;
887 	}
888 #endif
889 
890 	kdebug("Mapped Object [%s]:", what);
891 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
892 	kdebug("- entry    : %lx", params->entry_addr);
893 	kdebug("- PHDR[]   : %lx", params->ph_addr);
894 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
895 	seg = loadmap->segs;
896 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
897 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
898 		       loop,
899 		       seg->addr, seg->addr + seg->p_memsz - 1,
900 		       seg->p_vaddr, seg->p_memsz);
901 
902 	return 0;
903 
904 dynamic_error:
905 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
906 	       what, file->f_path.dentry->d_inode->i_ino);
907 	return -ELIBBAD;
908 }
909 
910 /*****************************************************************************/
911 /*
912  * map a file with constant displacement under uClinux
913  */
914 #ifndef CONFIG_MMU
915 static int elf_fdpic_map_file_constdisp_on_uclinux(
916 	struct elf_fdpic_params *params,
917 	struct file *file,
918 	struct mm_struct *mm)
919 {
920 	struct elf32_fdpic_loadseg *seg;
921 	struct elf32_phdr *phdr;
922 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
923 	loff_t fpos;
924 	int loop, ret;
925 
926 	load_addr = params->load_addr;
927 	seg = params->loadmap->segs;
928 
929 	/* determine the bounds of the contiguous overall allocation we must
930 	 * make */
931 	phdr = params->phdrs;
932 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
933 		if (params->phdrs[loop].p_type != PT_LOAD)
934 			continue;
935 
936 		if (base > phdr->p_vaddr)
937 			base = phdr->p_vaddr;
938 		if (top < phdr->p_vaddr + phdr->p_memsz)
939 			top = phdr->p_vaddr + phdr->p_memsz;
940 	}
941 
942 	/* allocate one big anon block for everything */
943 	mflags = MAP_PRIVATE;
944 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
945 		mflags |= MAP_EXECUTABLE;
946 
947 	down_write(&mm->mmap_sem);
948 	maddr = do_mmap(NULL, load_addr, top - base,
949 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
950 	up_write(&mm->mmap_sem);
951 	if (IS_ERR_VALUE(maddr))
952 		return (int) maddr;
953 
954 	if (load_addr != 0)
955 		load_addr += PAGE_ALIGN(top - base);
956 
957 	/* and then load the file segments into it */
958 	phdr = params->phdrs;
959 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
960 		if (params->phdrs[loop].p_type != PT_LOAD)
961 			continue;
962 
963 		fpos = phdr->p_offset;
964 
965 		seg->addr = maddr + (phdr->p_vaddr - base);
966 		seg->p_vaddr = phdr->p_vaddr;
967 		seg->p_memsz = phdr->p_memsz;
968 
969 		ret = file->f_op->read(file, (void *) seg->addr,
970 				       phdr->p_filesz, &fpos);
971 		if (ret < 0)
972 			return ret;
973 
974 		/* map the ELF header address if in this segment */
975 		if (phdr->p_offset == 0)
976 			params->elfhdr_addr = seg->addr;
977 
978 		/* clear any space allocated but not loaded */
979 		if (phdr->p_filesz < phdr->p_memsz) {
980 			ret = clear_user((void *) (seg->addr + phdr->p_filesz),
981 					 phdr->p_memsz - phdr->p_filesz);
982 			if (ret)
983 				return ret;
984 		}
985 
986 		if (mm) {
987 			if (phdr->p_flags & PF_X) {
988 				if (!mm->start_code) {
989 					mm->start_code = seg->addr;
990 					mm->end_code = seg->addr +
991 						phdr->p_memsz;
992 				}
993 			} else if (!mm->start_data) {
994 				mm->start_data = seg->addr;
995 #ifndef CONFIG_MMU
996 				mm->end_data = seg->addr + phdr->p_memsz;
997 #endif
998 			}
999 
1000 #ifdef CONFIG_MMU
1001 			if (seg->addr + phdr->p_memsz > mm->end_data)
1002 				mm->end_data = seg->addr + phdr->p_memsz;
1003 #endif
1004 		}
1005 
1006 		seg++;
1007 	}
1008 
1009 	return 0;
1010 }
1011 #endif
1012 
1013 /*****************************************************************************/
1014 /*
1015  * map a binary by direct mmap() of the individual PT_LOAD segments
1016  */
1017 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1018 					     struct file *file,
1019 					     struct mm_struct *mm)
1020 {
1021 	struct elf32_fdpic_loadseg *seg;
1022 	struct elf32_phdr *phdr;
1023 	unsigned long load_addr, delta_vaddr;
1024 	int loop, dvset, ret;
1025 
1026 	load_addr = params->load_addr;
1027 	delta_vaddr = 0;
1028 	dvset = 0;
1029 
1030 	seg = params->loadmap->segs;
1031 
1032 	/* deal with each load segment separately */
1033 	phdr = params->phdrs;
1034 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1035 		unsigned long maddr, disp, excess, excess1;
1036 		int prot = 0, flags;
1037 
1038 		if (phdr->p_type != PT_LOAD)
1039 			continue;
1040 
1041 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1042 		       (unsigned long) phdr->p_vaddr,
1043 		       (unsigned long) phdr->p_offset,
1044 		       (unsigned long) phdr->p_filesz,
1045 		       (unsigned long) phdr->p_memsz);
1046 
1047 		/* determine the mapping parameters */
1048 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1049 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1050 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1051 
1052 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1053 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1054 			flags |= MAP_EXECUTABLE;
1055 
1056 		maddr = 0;
1057 
1058 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1059 		case ELF_FDPIC_FLAG_INDEPENDENT:
1060 			/* PT_LOADs are independently locatable */
1061 			break;
1062 
1063 		case ELF_FDPIC_FLAG_HONOURVADDR:
1064 			/* the specified virtual address must be honoured */
1065 			maddr = phdr->p_vaddr;
1066 			flags |= MAP_FIXED;
1067 			break;
1068 
1069 		case ELF_FDPIC_FLAG_CONSTDISP:
1070 			/* constant displacement
1071 			 * - can be mapped anywhere, but must be mapped as a
1072 			 *   unit
1073 			 */
1074 			if (!dvset) {
1075 				maddr = load_addr;
1076 				delta_vaddr = phdr->p_vaddr;
1077 				dvset = 1;
1078 			} else {
1079 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1080 				flags |= MAP_FIXED;
1081 			}
1082 			break;
1083 
1084 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1085 			/* contiguity handled later */
1086 			break;
1087 
1088 		default:
1089 			BUG();
1090 		}
1091 
1092 		maddr &= PAGE_MASK;
1093 
1094 		/* create the mapping */
1095 		disp = phdr->p_vaddr & ~PAGE_MASK;
1096 		down_write(&mm->mmap_sem);
1097 		maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1098 				phdr->p_offset - disp);
1099 		up_write(&mm->mmap_sem);
1100 
1101 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1102 		       loop, phdr->p_memsz + disp, prot, flags,
1103 		       phdr->p_offset - disp, maddr);
1104 
1105 		if (IS_ERR_VALUE(maddr))
1106 			return (int) maddr;
1107 
1108 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1109 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1110 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1111 
1112 		seg->addr = maddr + disp;
1113 		seg->p_vaddr = phdr->p_vaddr;
1114 		seg->p_memsz = phdr->p_memsz;
1115 
1116 		/* map the ELF header address if in this segment */
1117 		if (phdr->p_offset == 0)
1118 			params->elfhdr_addr = seg->addr;
1119 
1120 		/* clear the bit between beginning of mapping and beginning of
1121 		 * PT_LOAD */
1122 		if (prot & PROT_WRITE && disp > 0) {
1123 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1124 			ret = clear_user((void __user *) maddr, disp);
1125 			if (ret)
1126 				return ret;
1127 			maddr += disp;
1128 		}
1129 
1130 		/* clear any space allocated but not loaded
1131 		 * - on uClinux we can just clear the lot
1132 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1133 		 *   extant in the file
1134 		 */
1135 		excess = phdr->p_memsz - phdr->p_filesz;
1136 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1137 
1138 #ifdef CONFIG_MMU
1139 		if (excess > excess1) {
1140 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1141 			unsigned long xmaddr;
1142 
1143 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1144 			down_write(&mm->mmap_sem);
1145 			xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1146 					 prot, flags, 0);
1147 			up_write(&mm->mmap_sem);
1148 
1149 			kdebug("mmap[%d] <anon>"
1150 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1151 			       loop, xaddr, excess - excess1, prot, flags,
1152 			       xmaddr);
1153 
1154 			if (xmaddr != xaddr)
1155 				return -ENOMEM;
1156 		}
1157 
1158 		if (prot & PROT_WRITE && excess1 > 0) {
1159 			kdebug("clear[%d] ad=%lx sz=%lx",
1160 			       loop, maddr + phdr->p_filesz, excess1);
1161 			ret = clear_user((void __user *) maddr + phdr->p_filesz,
1162 					 excess1);
1163 			if (ret)
1164 				return ret;
1165 		}
1166 
1167 #else
1168 		if (excess > 0) {
1169 			kdebug("clear[%d] ad=%lx sz=%lx",
1170 			       loop, maddr + phdr->p_filesz, excess);
1171 			ret = clear_user((void *) maddr + phdr->p_filesz, excess);
1172 			if (ret)
1173 				return ret;
1174 		}
1175 #endif
1176 
1177 		if (mm) {
1178 			if (phdr->p_flags & PF_X) {
1179 				if (!mm->start_code) {
1180 					mm->start_code = maddr;
1181 					mm->end_code = maddr + phdr->p_memsz;
1182 				}
1183 			} else if (!mm->start_data) {
1184 				mm->start_data = maddr;
1185 				mm->end_data = maddr + phdr->p_memsz;
1186 			}
1187 		}
1188 
1189 		seg++;
1190 	}
1191 
1192 	return 0;
1193 }
1194 
1195 /*****************************************************************************/
1196 /*
1197  * ELF-FDPIC core dumper
1198  *
1199  * Modelled on fs/exec.c:aout_core_dump()
1200  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1201  *
1202  * Modelled on fs/binfmt_elf.c core dumper
1203  */
1204 #ifdef CONFIG_ELF_CORE
1205 
1206 /*
1207  * These are the only things you should do on a core-file: use only these
1208  * functions to write out all the necessary info.
1209  */
1210 static int dump_write(struct file *file, const void *addr, int nr)
1211 {
1212 	return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1213 }
1214 
1215 static int dump_seek(struct file *file, loff_t off)
1216 {
1217 	if (file->f_op->llseek) {
1218 		if (file->f_op->llseek(file, off, SEEK_SET) != off)
1219 			return 0;
1220 	} else {
1221 		file->f_pos = off;
1222 	}
1223 	return 1;
1224 }
1225 
1226 /*
1227  * Decide whether a segment is worth dumping; default is yes to be
1228  * sure (missing info is worse than too much; etc).
1229  * Personally I'd include everything, and use the coredump limit...
1230  *
1231  * I think we should skip something. But I am not sure how. H.J.
1232  */
1233 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1234 {
1235 	int dump_ok;
1236 
1237 	/* Do not dump I/O mapped devices or special mappings */
1238 	if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1239 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1240 		return 0;
1241 	}
1242 
1243 	/* If we may not read the contents, don't allow us to dump
1244 	 * them either. "dump_write()" can't handle it anyway.
1245 	 */
1246 	if (!(vma->vm_flags & VM_READ)) {
1247 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1248 		return 0;
1249 	}
1250 
1251 	/* By default, dump shared memory if mapped from an anonymous file. */
1252 	if (vma->vm_flags & VM_SHARED) {
1253 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1254 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1255 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1256 			       vma->vm_flags, dump_ok ? "yes" : "no");
1257 			return dump_ok;
1258 		}
1259 
1260 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1261 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1262 		       vma->vm_flags, dump_ok ? "yes" : "no");
1263 		return dump_ok;
1264 	}
1265 
1266 #ifdef CONFIG_MMU
1267 	/* By default, if it hasn't been written to, don't write it out */
1268 	if (!vma->anon_vma) {
1269 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1270 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1271 		       vma->vm_flags, dump_ok ? "yes" : "no");
1272 		return dump_ok;
1273 	}
1274 #endif
1275 
1276 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1277 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1278 	       dump_ok ? "yes" : "no");
1279 	return dump_ok;
1280 }
1281 
1282 /* An ELF note in memory */
1283 struct memelfnote
1284 {
1285 	const char *name;
1286 	int type;
1287 	unsigned int datasz;
1288 	void *data;
1289 };
1290 
1291 static int notesize(struct memelfnote *en)
1292 {
1293 	int sz;
1294 
1295 	sz = sizeof(struct elf_note);
1296 	sz += roundup(strlen(en->name) + 1, 4);
1297 	sz += roundup(en->datasz, 4);
1298 
1299 	return sz;
1300 }
1301 
1302 /* #define DEBUG */
1303 
1304 #define DUMP_WRITE(addr, nr)	\
1305 	do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1306 #define DUMP_SEEK(off)	\
1307 	do { if (!dump_seek(file, (off))) return 0; } while(0)
1308 
1309 static int writenote(struct memelfnote *men, struct file *file)
1310 {
1311 	struct elf_note en;
1312 
1313 	en.n_namesz = strlen(men->name) + 1;
1314 	en.n_descsz = men->datasz;
1315 	en.n_type = men->type;
1316 
1317 	DUMP_WRITE(&en, sizeof(en));
1318 	DUMP_WRITE(men->name, en.n_namesz);
1319 	/* XXX - cast from long long to long to avoid need for libgcc.a */
1320 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1321 	DUMP_WRITE(men->data, men->datasz);
1322 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1323 
1324 	return 1;
1325 }
1326 #undef DUMP_WRITE
1327 #undef DUMP_SEEK
1328 
1329 #define DUMP_WRITE(addr, nr)				\
1330 	if ((size += (nr)) > cprm->limit ||		\
1331 	    !dump_write(cprm->file, (addr), (nr)))	\
1332 		goto end_coredump;
1333 
1334 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1335 {
1336 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1337 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1338 	elf->e_ident[EI_DATA] = ELF_DATA;
1339 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1340 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1341 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1342 
1343 	elf->e_type = ET_CORE;
1344 	elf->e_machine = ELF_ARCH;
1345 	elf->e_version = EV_CURRENT;
1346 	elf->e_entry = 0;
1347 	elf->e_phoff = sizeof(struct elfhdr);
1348 	elf->e_shoff = 0;
1349 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1350 	elf->e_ehsize = sizeof(struct elfhdr);
1351 	elf->e_phentsize = sizeof(struct elf_phdr);
1352 	elf->e_phnum = segs;
1353 	elf->e_shentsize = 0;
1354 	elf->e_shnum = 0;
1355 	elf->e_shstrndx = 0;
1356 	return;
1357 }
1358 
1359 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1360 {
1361 	phdr->p_type = PT_NOTE;
1362 	phdr->p_offset = offset;
1363 	phdr->p_vaddr = 0;
1364 	phdr->p_paddr = 0;
1365 	phdr->p_filesz = sz;
1366 	phdr->p_memsz = 0;
1367 	phdr->p_flags = 0;
1368 	phdr->p_align = 0;
1369 	return;
1370 }
1371 
1372 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1373 		unsigned int sz, void *data)
1374 {
1375 	note->name = name;
1376 	note->type = type;
1377 	note->datasz = sz;
1378 	note->data = data;
1379 	return;
1380 }
1381 
1382 /*
1383  * fill up all the fields in prstatus from the given task struct, except
1384  * registers which need to be filled up seperately.
1385  */
1386 static void fill_prstatus(struct elf_prstatus *prstatus,
1387 			  struct task_struct *p, long signr)
1388 {
1389 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1390 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1391 	prstatus->pr_sighold = p->blocked.sig[0];
1392 	rcu_read_lock();
1393 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1394 	rcu_read_unlock();
1395 	prstatus->pr_pid = task_pid_vnr(p);
1396 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1397 	prstatus->pr_sid = task_session_vnr(p);
1398 	if (thread_group_leader(p)) {
1399 		struct task_cputime cputime;
1400 
1401 		/*
1402 		 * This is the record for the group leader.  It shows the
1403 		 * group-wide total, not its individual thread total.
1404 		 */
1405 		thread_group_cputime(p, &cputime);
1406 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1407 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1408 	} else {
1409 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1410 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1411 	}
1412 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1413 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1414 
1415 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1416 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1417 }
1418 
1419 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1420 		       struct mm_struct *mm)
1421 {
1422 	const struct cred *cred;
1423 	unsigned int i, len;
1424 
1425 	/* first copy the parameters from user space */
1426 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1427 
1428 	len = mm->arg_end - mm->arg_start;
1429 	if (len >= ELF_PRARGSZ)
1430 		len = ELF_PRARGSZ - 1;
1431 	if (copy_from_user(&psinfo->pr_psargs,
1432 		           (const char __user *) mm->arg_start, len))
1433 		return -EFAULT;
1434 	for (i = 0; i < len; i++)
1435 		if (psinfo->pr_psargs[i] == 0)
1436 			psinfo->pr_psargs[i] = ' ';
1437 	psinfo->pr_psargs[len] = 0;
1438 
1439 	rcu_read_lock();
1440 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1441 	rcu_read_unlock();
1442 	psinfo->pr_pid = task_pid_vnr(p);
1443 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1444 	psinfo->pr_sid = task_session_vnr(p);
1445 
1446 	i = p->state ? ffz(~p->state) + 1 : 0;
1447 	psinfo->pr_state = i;
1448 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1449 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1450 	psinfo->pr_nice = task_nice(p);
1451 	psinfo->pr_flag = p->flags;
1452 	rcu_read_lock();
1453 	cred = __task_cred(p);
1454 	SET_UID(psinfo->pr_uid, cred->uid);
1455 	SET_GID(psinfo->pr_gid, cred->gid);
1456 	rcu_read_unlock();
1457 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1458 
1459 	return 0;
1460 }
1461 
1462 /* Here is the structure in which status of each thread is captured. */
1463 struct elf_thread_status
1464 {
1465 	struct list_head list;
1466 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1467 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1468 	struct task_struct *thread;
1469 #ifdef ELF_CORE_COPY_XFPREGS
1470 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1471 #endif
1472 	struct memelfnote notes[3];
1473 	int num_notes;
1474 };
1475 
1476 /*
1477  * In order to add the specific thread information for the elf file format,
1478  * we need to keep a linked list of every thread's pr_status and then create
1479  * a single section for them in the final core file.
1480  */
1481 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1482 {
1483 	struct task_struct *p = t->thread;
1484 	int sz = 0;
1485 
1486 	t->num_notes = 0;
1487 
1488 	fill_prstatus(&t->prstatus, p, signr);
1489 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1490 
1491 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1492 		  &t->prstatus);
1493 	t->num_notes++;
1494 	sz += notesize(&t->notes[0]);
1495 
1496 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1497 	if (t->prstatus.pr_fpvalid) {
1498 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1499 			  &t->fpu);
1500 		t->num_notes++;
1501 		sz += notesize(&t->notes[1]);
1502 	}
1503 
1504 #ifdef ELF_CORE_COPY_XFPREGS
1505 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1506 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1507 			  sizeof(t->xfpu), &t->xfpu);
1508 		t->num_notes++;
1509 		sz += notesize(&t->notes[2]);
1510 	}
1511 #endif
1512 	return sz;
1513 }
1514 
1515 /*
1516  * dump the segments for an MMU process
1517  */
1518 #ifdef CONFIG_MMU
1519 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1520 			   unsigned long *limit, unsigned long mm_flags)
1521 {
1522 	struct vm_area_struct *vma;
1523 	int err = 0;
1524 
1525 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1526 		unsigned long addr;
1527 
1528 		if (!maydump(vma, mm_flags))
1529 			continue;
1530 
1531 		for (addr = vma->vm_start; addr < vma->vm_end;
1532 							addr += PAGE_SIZE) {
1533 			struct page *page = get_dump_page(addr);
1534 			if (page) {
1535 				void *kaddr = kmap(page);
1536 				*size += PAGE_SIZE;
1537 				if (*size > *limit)
1538 					err = -EFBIG;
1539 				else if (!dump_write(file, kaddr, PAGE_SIZE))
1540 					err = -EIO;
1541 				kunmap(page);
1542 				page_cache_release(page);
1543 			} else if (!dump_seek(file, file->f_pos + PAGE_SIZE))
1544 				err = -EFBIG;
1545 			if (err)
1546 				goto out;
1547 		}
1548 	}
1549 out:
1550 	return err;
1551 }
1552 #endif
1553 
1554 /*
1555  * dump the segments for a NOMMU process
1556  */
1557 #ifndef CONFIG_MMU
1558 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1559 			   unsigned long *limit, unsigned long mm_flags)
1560 {
1561 	struct vm_area_struct *vma;
1562 
1563 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1564 		if (!maydump(vma, mm_flags))
1565 			continue;
1566 
1567 		if ((*size += PAGE_SIZE) > *limit)
1568 			return -EFBIG;
1569 
1570 		if (!dump_write(file, (void *) vma->vm_start,
1571 				vma->vm_end - vma->vm_start))
1572 			return -EIO;
1573 	}
1574 
1575 	return 0;
1576 }
1577 #endif
1578 
1579 /*
1580  * Actual dumper
1581  *
1582  * This is a two-pass process; first we find the offsets of the bits,
1583  * and then they are actually written out.  If we run out of core limit
1584  * we just truncate.
1585  */
1586 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1587 {
1588 #define	NUM_NOTES	6
1589 	int has_dumped = 0;
1590 	mm_segment_t fs;
1591 	int segs;
1592 	size_t size = 0;
1593 	int i;
1594 	struct vm_area_struct *vma;
1595 	struct elfhdr *elf = NULL;
1596 	loff_t offset = 0, dataoff;
1597 	int numnote;
1598 	struct memelfnote *notes = NULL;
1599 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1600 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1601  	LIST_HEAD(thread_list);
1602  	struct list_head *t;
1603 	elf_fpregset_t *fpu = NULL;
1604 #ifdef ELF_CORE_COPY_XFPREGS
1605 	elf_fpxregset_t *xfpu = NULL;
1606 #endif
1607 	int thread_status_size = 0;
1608 	elf_addr_t *auxv;
1609 	unsigned long mm_flags;
1610 
1611 	/*
1612 	 * We no longer stop all VM operations.
1613 	 *
1614 	 * This is because those proceses that could possibly change map_count
1615 	 * or the mmap / vma pages are now blocked in do_exit on current
1616 	 * finishing this core dump.
1617 	 *
1618 	 * Only ptrace can touch these memory addresses, but it doesn't change
1619 	 * the map_count or the pages allocated. So no possibility of crashing
1620 	 * exists while dumping the mm->vm_next areas to the core file.
1621 	 */
1622 
1623 	/* alloc memory for large data structures: too large to be on stack */
1624 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1625 	if (!elf)
1626 		goto cleanup;
1627 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1628 	if (!prstatus)
1629 		goto cleanup;
1630 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1631 	if (!psinfo)
1632 		goto cleanup;
1633 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1634 	if (!notes)
1635 		goto cleanup;
1636 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1637 	if (!fpu)
1638 		goto cleanup;
1639 #ifdef ELF_CORE_COPY_XFPREGS
1640 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1641 	if (!xfpu)
1642 		goto cleanup;
1643 #endif
1644 
1645 	if (cprm->signr) {
1646 		struct core_thread *ct;
1647 		struct elf_thread_status *tmp;
1648 
1649 		for (ct = current->mm->core_state->dumper.next;
1650 						ct; ct = ct->next) {
1651 			tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1652 			if (!tmp)
1653 				goto cleanup;
1654 
1655 			tmp->thread = ct->task;
1656 			list_add(&tmp->list, &thread_list);
1657 		}
1658 
1659 		list_for_each(t, &thread_list) {
1660 			struct elf_thread_status *tmp;
1661 			int sz;
1662 
1663 			tmp = list_entry(t, struct elf_thread_status, list);
1664 			sz = elf_dump_thread_status(cprm->signr, tmp);
1665 			thread_status_size += sz;
1666 		}
1667 	}
1668 
1669 	/* now collect the dump for the current */
1670 	fill_prstatus(prstatus, current, cprm->signr);
1671 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1672 
1673 	segs = current->mm->map_count;
1674 #ifdef ELF_CORE_EXTRA_PHDRS
1675 	segs += ELF_CORE_EXTRA_PHDRS;
1676 #endif
1677 
1678 	/* Set up header */
1679 	fill_elf_fdpic_header(elf, segs + 1);	/* including notes section */
1680 
1681 	has_dumped = 1;
1682 	current->flags |= PF_DUMPCORE;
1683 
1684 	/*
1685 	 * Set up the notes in similar form to SVR4 core dumps made
1686 	 * with info from their /proc.
1687 	 */
1688 
1689 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1690 	fill_psinfo(psinfo, current->group_leader, current->mm);
1691 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1692 
1693 	numnote = 2;
1694 
1695 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1696 
1697 	i = 0;
1698 	do
1699 		i += 2;
1700 	while (auxv[i - 2] != AT_NULL);
1701 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1702 		  i * sizeof(elf_addr_t), auxv);
1703 
1704   	/* Try to dump the FPU. */
1705 	if ((prstatus->pr_fpvalid =
1706 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1707 		fill_note(notes + numnote++,
1708 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1709 #ifdef ELF_CORE_COPY_XFPREGS
1710 	if (elf_core_copy_task_xfpregs(current, xfpu))
1711 		fill_note(notes + numnote++,
1712 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1713 #endif
1714 
1715 	fs = get_fs();
1716 	set_fs(KERNEL_DS);
1717 
1718 	DUMP_WRITE(elf, sizeof(*elf));
1719 	offset += sizeof(*elf);				/* Elf header */
1720 	offset += (segs+1) * sizeof(struct elf_phdr);	/* Program headers */
1721 
1722 	/* Write notes phdr entry */
1723 	{
1724 		struct elf_phdr phdr;
1725 		int sz = 0;
1726 
1727 		for (i = 0; i < numnote; i++)
1728 			sz += notesize(notes + i);
1729 
1730 		sz += thread_status_size;
1731 
1732 		fill_elf_note_phdr(&phdr, sz, offset);
1733 		offset += sz;
1734 		DUMP_WRITE(&phdr, sizeof(phdr));
1735 	}
1736 
1737 	/* Page-align dumped data */
1738 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1739 
1740 	/*
1741 	 * We must use the same mm->flags while dumping core to avoid
1742 	 * inconsistency between the program headers and bodies, otherwise an
1743 	 * unusable core file can be generated.
1744 	 */
1745 	mm_flags = current->mm->flags;
1746 
1747 	/* write program headers for segments dump */
1748 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1749 		struct elf_phdr phdr;
1750 		size_t sz;
1751 
1752 		sz = vma->vm_end - vma->vm_start;
1753 
1754 		phdr.p_type = PT_LOAD;
1755 		phdr.p_offset = offset;
1756 		phdr.p_vaddr = vma->vm_start;
1757 		phdr.p_paddr = 0;
1758 		phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1759 		phdr.p_memsz = sz;
1760 		offset += phdr.p_filesz;
1761 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1762 		if (vma->vm_flags & VM_WRITE)
1763 			phdr.p_flags |= PF_W;
1764 		if (vma->vm_flags & VM_EXEC)
1765 			phdr.p_flags |= PF_X;
1766 		phdr.p_align = ELF_EXEC_PAGESIZE;
1767 
1768 		DUMP_WRITE(&phdr, sizeof(phdr));
1769 	}
1770 
1771 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1772 	ELF_CORE_WRITE_EXTRA_PHDRS;
1773 #endif
1774 
1775  	/* write out the notes section */
1776 	for (i = 0; i < numnote; i++)
1777 		if (!writenote(notes + i, cprm->file))
1778 			goto end_coredump;
1779 
1780 	/* write out the thread status notes section */
1781 	list_for_each(t, &thread_list) {
1782 		struct elf_thread_status *tmp =
1783 				list_entry(t, struct elf_thread_status, list);
1784 
1785 		for (i = 0; i < tmp->num_notes; i++)
1786 			if (!writenote(&tmp->notes[i], cprm->file))
1787 				goto end_coredump;
1788 	}
1789 
1790 	if (!dump_seek(cprm->file, dataoff))
1791 		goto end_coredump;
1792 
1793 	if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit,
1794 				    mm_flags) < 0)
1795 		goto end_coredump;
1796 
1797 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1798 	ELF_CORE_WRITE_EXTRA_DATA;
1799 #endif
1800 
1801 	if (file->f_pos != offset) {
1802 		/* Sanity check */
1803 		printk(KERN_WARNING
1804 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1805 		       file->f_pos, offset);
1806 	}
1807 
1808 end_coredump:
1809 	set_fs(fs);
1810 
1811 cleanup:
1812 	while (!list_empty(&thread_list)) {
1813 		struct list_head *tmp = thread_list.next;
1814 		list_del(tmp);
1815 		kfree(list_entry(tmp, struct elf_thread_status, list));
1816 	}
1817 
1818 	kfree(elf);
1819 	kfree(prstatus);
1820 	kfree(psinfo);
1821 	kfree(notes);
1822 	kfree(fpu);
1823 #ifdef ELF_CORE_COPY_XFPREGS
1824 	kfree(xfpu);
1825 #endif
1826 	return has_dumped;
1827 #undef NUM_NOTES
1828 }
1829 
1830 #endif		/* CONFIG_ELF_CORE */
1831