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#include <linux/coredump.h>
38
39#include <asm/uaccess.h>
40#include <asm/param.h>
41#include <asm/pgalloc.h>
42
43typedef char *elf_caddr_t;
44
45#if 0
46#define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
47#else
48#define kdebug(fmt, ...) do {} while(0)
49#endif
50
51#if 0
52#define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
53#else
54#define kdcore(fmt, ...) do {} while(0)
55#endif
56
57MODULE_LICENSE("GPL");
58
59static int load_elf_fdpic_binary(struct linux_binprm *);
60static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
61static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
62			      struct mm_struct *, const char *);
63
64static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
65				   struct elf_fdpic_params *,
66				   struct elf_fdpic_params *);
67
68#ifndef CONFIG_MMU
69static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
70					    unsigned long *);
71static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
72						   struct file *,
73						   struct mm_struct *);
74#endif
75
76static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
77					     struct file *, struct mm_struct *);
78
79#ifdef CONFIG_ELF_CORE
80static int elf_fdpic_core_dump(struct coredump_params *cprm);
81#endif
82
83static struct linux_binfmt elf_fdpic_format = {
84	.module		= THIS_MODULE,
85	.load_binary	= load_elf_fdpic_binary,
86#ifdef CONFIG_ELF_CORE
87	.core_dump	= elf_fdpic_core_dump,
88#endif
89	.min_coredump	= ELF_EXEC_PAGESIZE,
90};
91
92static int __init init_elf_fdpic_binfmt(void)
93{
94	register_binfmt(&elf_fdpic_format);
95	return 0;
96}
97
98static void __exit exit_elf_fdpic_binfmt(void)
99{
100	unregister_binfmt(&elf_fdpic_format);
101}
102
103core_initcall(init_elf_fdpic_binfmt);
104module_exit(exit_elf_fdpic_binfmt);
105
106static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
107{
108	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
109		return 0;
110	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
111		return 0;
112	if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
113		return 0;
114	if (!file->f_op->mmap)
115		return 0;
116	return 1;
117}
118
119/*****************************************************************************/
120/*
121 * read the program headers table into memory
122 */
123static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
124				 struct file *file)
125{
126	struct elf32_phdr *phdr;
127	unsigned long size;
128	int retval, loop;
129
130	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
131		return -ENOMEM;
132	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
133		return -ENOMEM;
134
135	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
136	params->phdrs = kmalloc(size, GFP_KERNEL);
137	if (!params->phdrs)
138		return -ENOMEM;
139
140	retval = kernel_read(file, params->hdr.e_phoff,
141			     (char *) params->phdrs, size);
142	if (unlikely(retval != size))
143		return retval < 0 ? retval : -ENOEXEC;
144
145	/* determine stack size for this binary */
146	phdr = params->phdrs;
147	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
148		if (phdr->p_type != PT_GNU_STACK)
149			continue;
150
151		if (phdr->p_flags & PF_X)
152			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
153		else
154			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
155
156		params->stack_size = phdr->p_memsz;
157		break;
158	}
159
160	return 0;
161}
162
163/*****************************************************************************/
164/*
165 * load an fdpic binary into various bits of memory
166 */
167static int load_elf_fdpic_binary(struct linux_binprm *bprm)
168{
169	struct elf_fdpic_params exec_params, interp_params;
170	struct pt_regs *regs = current_pt_regs();
171	struct elf_phdr *phdr;
172	unsigned long stack_size, entryaddr;
173#ifdef ELF_FDPIC_PLAT_INIT
174	unsigned long dynaddr;
175#endif
176#ifndef CONFIG_MMU
177	unsigned long stack_prot;
178#endif
179	struct file *interpreter = NULL; /* to shut gcc up */
180	char *interpreter_name = NULL;
181	int executable_stack;
182	int retval, i;
183
184	kdebug("____ LOAD %d ____", current->pid);
185
186	memset(&exec_params, 0, sizeof(exec_params));
187	memset(&interp_params, 0, sizeof(interp_params));
188
189	exec_params.hdr = *(struct elfhdr *) bprm->buf;
190	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
191
192	/* check that this is a binary we know how to deal with */
193	retval = -ENOEXEC;
194	if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
195		goto error;
196
197	/* read the program header table */
198	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
199	if (retval < 0)
200		goto error;
201
202	/* scan for a program header that specifies an interpreter */
203	phdr = exec_params.phdrs;
204
205	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
206		switch (phdr->p_type) {
207		case PT_INTERP:
208			retval = -ENOMEM;
209			if (phdr->p_filesz > PATH_MAX)
210				goto error;
211			retval = -ENOENT;
212			if (phdr->p_filesz < 2)
213				goto error;
214
215			/* read the name of the interpreter into memory */
216			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
217			if (!interpreter_name)
218				goto error;
219
220			retval = kernel_read(bprm->file,
221					     phdr->p_offset,
222					     interpreter_name,
223					     phdr->p_filesz);
224			if (unlikely(retval != phdr->p_filesz)) {
225				if (retval >= 0)
226					retval = -ENOEXEC;
227				goto error;
228			}
229
230			retval = -ENOENT;
231			if (interpreter_name[phdr->p_filesz - 1] != '\0')
232				goto error;
233
234			kdebug("Using ELF interpreter %s", interpreter_name);
235
236			/* replace the program with the interpreter */
237			interpreter = open_exec(interpreter_name);
238			retval = PTR_ERR(interpreter);
239			if (IS_ERR(interpreter)) {
240				interpreter = NULL;
241				goto error;
242			}
243
244			/*
245			 * If the binary is not readable then enforce
246			 * mm->dumpable = 0 regardless of the interpreter's
247			 * permissions.
248			 */
249			would_dump(bprm, interpreter);
250
251			retval = kernel_read(interpreter, 0, bprm->buf,
252					     BINPRM_BUF_SIZE);
253			if (unlikely(retval != BINPRM_BUF_SIZE)) {
254				if (retval >= 0)
255					retval = -ENOEXEC;
256				goto error;
257			}
258
259			interp_params.hdr = *((struct elfhdr *) bprm->buf);
260			break;
261
262		case PT_LOAD:
263#ifdef CONFIG_MMU
264			if (exec_params.load_addr == 0)
265				exec_params.load_addr = phdr->p_vaddr;
266#endif
267			break;
268		}
269
270	}
271
272	if (elf_check_const_displacement(&exec_params.hdr))
273		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
274
275	/* perform insanity checks on the interpreter */
276	if (interpreter_name) {
277		retval = -ELIBBAD;
278		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
279			goto error;
280
281		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
282
283		/* read the interpreter's program header table */
284		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
285		if (retval < 0)
286			goto error;
287	}
288
289	stack_size = exec_params.stack_size;
290	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
291		executable_stack = EXSTACK_ENABLE_X;
292	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
293		executable_stack = EXSTACK_DISABLE_X;
294	else
295		executable_stack = EXSTACK_DEFAULT;
296
297	if (stack_size == 0) {
298		stack_size = interp_params.stack_size;
299		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
300			executable_stack = EXSTACK_ENABLE_X;
301		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
302			executable_stack = EXSTACK_DISABLE_X;
303		else
304			executable_stack = EXSTACK_DEFAULT;
305	}
306
307	retval = -ENOEXEC;
308	if (stack_size == 0)
309		goto error;
310
311	if (elf_check_const_displacement(&interp_params.hdr))
312		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
313
314	/* flush all traces of the currently running executable */
315	retval = flush_old_exec(bprm);
316	if (retval)
317		goto error;
318
319	/* there's now no turning back... the old userspace image is dead,
320	 * defunct, deceased, etc.
321	 */
322	set_personality(PER_LINUX_FDPIC);
323	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
324		current->personality |= READ_IMPLIES_EXEC;
325
326	setup_new_exec(bprm);
327
328	set_binfmt(&elf_fdpic_format);
329
330	current->mm->start_code = 0;
331	current->mm->end_code = 0;
332	current->mm->start_stack = 0;
333	current->mm->start_data = 0;
334	current->mm->end_data = 0;
335	current->mm->context.exec_fdpic_loadmap = 0;
336	current->mm->context.interp_fdpic_loadmap = 0;
337
338#ifdef CONFIG_MMU
339	elf_fdpic_arch_lay_out_mm(&exec_params,
340				  &interp_params,
341				  &current->mm->start_stack,
342				  &current->mm->start_brk);
343
344	retval = setup_arg_pages(bprm, current->mm->start_stack,
345				 executable_stack);
346	if (retval < 0)
347		goto error;
348#endif
349
350	/* load the executable and interpreter into memory */
351	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
352				    "executable");
353	if (retval < 0)
354		goto error;
355
356	if (interpreter_name) {
357		retval = elf_fdpic_map_file(&interp_params, interpreter,
358					    current->mm, "interpreter");
359		if (retval < 0) {
360			printk(KERN_ERR "Unable to load interpreter\n");
361			goto error;
362		}
363
364		allow_write_access(interpreter);
365		fput(interpreter);
366		interpreter = NULL;
367	}
368
369#ifdef CONFIG_MMU
370	if (!current->mm->start_brk)
371		current->mm->start_brk = current->mm->end_data;
372
373	current->mm->brk = current->mm->start_brk =
374		PAGE_ALIGN(current->mm->start_brk);
375
376#else
377	/* create a stack and brk area big enough for everyone
378	 * - the brk heap starts at the bottom and works up
379	 * - the stack starts at the top and works down
380	 */
381	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
382	if (stack_size < PAGE_SIZE * 2)
383		stack_size = PAGE_SIZE * 2;
384
385	stack_prot = PROT_READ | PROT_WRITE;
386	if (executable_stack == EXSTACK_ENABLE_X ||
387	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
388		stack_prot |= PROT_EXEC;
389
390	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
391					 MAP_PRIVATE | MAP_ANONYMOUS |
392					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
393					 0);
394
395	if (IS_ERR_VALUE(current->mm->start_brk)) {
396		retval = current->mm->start_brk;
397		current->mm->start_brk = 0;
398		goto error;
399	}
400
401	current->mm->brk = current->mm->start_brk;
402	current->mm->context.end_brk = current->mm->start_brk;
403	current->mm->context.end_brk +=
404		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
405	current->mm->start_stack = current->mm->start_brk + stack_size;
406#endif
407
408	install_exec_creds(bprm);
409	if (create_elf_fdpic_tables(bprm, current->mm,
410				    &exec_params, &interp_params) < 0)
411		goto error;
412
413	kdebug("- start_code  %lx", current->mm->start_code);
414	kdebug("- end_code    %lx", current->mm->end_code);
415	kdebug("- start_data  %lx", current->mm->start_data);
416	kdebug("- end_data    %lx", current->mm->end_data);
417	kdebug("- start_brk   %lx", current->mm->start_brk);
418	kdebug("- brk         %lx", current->mm->brk);
419	kdebug("- start_stack %lx", current->mm->start_stack);
420
421#ifdef ELF_FDPIC_PLAT_INIT
422	/*
423	 * The ABI may specify that certain registers be set up in special
424	 * ways (on i386 %edx is the address of a DT_FINI function, for
425	 * example.  This macro performs whatever initialization to
426	 * the regs structure is required.
427	 */
428	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
429	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
430			    dynaddr);
431#endif
432
433	/* everything is now ready... get the userspace context ready to roll */
434	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
435	start_thread(regs, entryaddr, current->mm->start_stack);
436
437	retval = 0;
438
439error:
440	if (interpreter) {
441		allow_write_access(interpreter);
442		fput(interpreter);
443	}
444	kfree(interpreter_name);
445	kfree(exec_params.phdrs);
446	kfree(exec_params.loadmap);
447	kfree(interp_params.phdrs);
448	kfree(interp_params.loadmap);
449	return retval;
450}
451
452/*****************************************************************************/
453
454#ifndef ELF_BASE_PLATFORM
455/*
456 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
457 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
458 * will be copied to the user stack in the same manner as AT_PLATFORM.
459 */
460#define ELF_BASE_PLATFORM NULL
461#endif
462
463/*
464 * present useful information to the program by shovelling it onto the new
465 * process's stack
466 */
467static int create_elf_fdpic_tables(struct linux_binprm *bprm,
468				   struct mm_struct *mm,
469				   struct elf_fdpic_params *exec_params,
470				   struct elf_fdpic_params *interp_params)
471{
472	const struct cred *cred = current_cred();
473	unsigned long sp, csp, nitems;
474	elf_caddr_t __user *argv, *envp;
475	size_t platform_len = 0, len;
476	char *k_platform, *k_base_platform;
477	char __user *u_platform, *u_base_platform, *p;
478	int loop;
479	int nr;	/* reset for each csp adjustment */
480
481#ifdef CONFIG_MMU
482	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
483	 * by the processes running on the same package. One thing we can do is
484	 * to shuffle the initial stack for them, so we give the architecture
485	 * an opportunity to do so here.
486	 */
487	sp = arch_align_stack(bprm->p);
488#else
489	sp = mm->start_stack;
490
491	/* stack the program arguments and environment */
492	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
493		return -EFAULT;
494#endif
495
496	/*
497	 * If this architecture has a platform capability string, copy it
498	 * to userspace.  In some cases (Sparc), this info is impossible
499	 * for userspace to get any other way, in others (i386) it is
500	 * merely difficult.
501	 */
502	k_platform = ELF_PLATFORM;
503	u_platform = NULL;
504
505	if (k_platform) {
506		platform_len = strlen(k_platform) + 1;
507		sp -= platform_len;
508		u_platform = (char __user *) sp;
509		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
510			return -EFAULT;
511	}
512
513	/*
514	 * If this architecture has a "base" platform capability
515	 * string, copy it to userspace.
516	 */
517	k_base_platform = ELF_BASE_PLATFORM;
518	u_base_platform = NULL;
519
520	if (k_base_platform) {
521		platform_len = strlen(k_base_platform) + 1;
522		sp -= platform_len;
523		u_base_platform = (char __user *) sp;
524		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
525			return -EFAULT;
526	}
527
528	sp &= ~7UL;
529
530	/* stack the load map(s) */
531	len = sizeof(struct elf32_fdpic_loadmap);
532	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
533	sp = (sp - len) & ~7UL;
534	exec_params->map_addr = sp;
535
536	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
537		return -EFAULT;
538
539	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
540
541	if (interp_params->loadmap) {
542		len = sizeof(struct elf32_fdpic_loadmap);
543		len += sizeof(struct elf32_fdpic_loadseg) *
544			interp_params->loadmap->nsegs;
545		sp = (sp - len) & ~7UL;
546		interp_params->map_addr = sp;
547
548		if (copy_to_user((void __user *) sp, interp_params->loadmap,
549				 len) != 0)
550			return -EFAULT;
551
552		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
553	}
554
555	/* force 16 byte _final_ alignment here for generality */
556#define DLINFO_ITEMS 15
557
558	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
559		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
560
561	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
562		nitems++;
563
564	csp = sp;
565	sp -= nitems * 2 * sizeof(unsigned long);
566	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
567	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
568	sp -= 1 * sizeof(unsigned long);		/* argc */
569
570	csp -= sp & 15UL;
571	sp -= sp & 15UL;
572
573	/* put the ELF interpreter info on the stack */
574#define NEW_AUX_ENT(id, val)						\
575	do {								\
576		struct { unsigned long _id, _val; } __user *ent;	\
577									\
578		ent = (void __user *) csp;				\
579		__put_user((id), &ent[nr]._id);				\
580		__put_user((val), &ent[nr]._val);			\
581		nr++;							\
582	} while (0)
583
584	nr = 0;
585	csp -= 2 * sizeof(unsigned long);
586	NEW_AUX_ENT(AT_NULL, 0);
587	if (k_platform) {
588		nr = 0;
589		csp -= 2 * sizeof(unsigned long);
590		NEW_AUX_ENT(AT_PLATFORM,
591			    (elf_addr_t) (unsigned long) u_platform);
592	}
593
594	if (k_base_platform) {
595		nr = 0;
596		csp -= 2 * sizeof(unsigned long);
597		NEW_AUX_ENT(AT_BASE_PLATFORM,
598			    (elf_addr_t) (unsigned long) u_base_platform);
599	}
600
601	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
602		nr = 0;
603		csp -= 2 * sizeof(unsigned long);
604		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
605	}
606
607	nr = 0;
608	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
609	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
610#ifdef ELF_HWCAP2
611	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
612#endif
613	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
614	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
615	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
616	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
617	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
618	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
619	NEW_AUX_ENT(AT_FLAGS,	0);
620	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
621	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
622	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
623	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
624	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
625	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
626	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
627
628#ifdef ARCH_DLINFO
629	nr = 0;
630	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
631
632	/* ARCH_DLINFO must come last so platform specific code can enforce
633	 * special alignment requirements on the AUXV if necessary (eg. PPC).
634	 */
635	ARCH_DLINFO;
636#endif
637#undef NEW_AUX_ENT
638
639	/* allocate room for argv[] and envv[] */
640	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
641	envp = (elf_caddr_t __user *) csp;
642	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
643	argv = (elf_caddr_t __user *) csp;
644
645	/* stack argc */
646	csp -= sizeof(unsigned long);
647	__put_user(bprm->argc, (unsigned long __user *) csp);
648
649	BUG_ON(csp != sp);
650
651	/* fill in the argv[] array */
652#ifdef CONFIG_MMU
653	current->mm->arg_start = bprm->p;
654#else
655	current->mm->arg_start = current->mm->start_stack -
656		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
657#endif
658
659	p = (char __user *) current->mm->arg_start;
660	for (loop = bprm->argc; loop > 0; loop--) {
661		__put_user((elf_caddr_t) p, argv++);
662		len = strnlen_user(p, MAX_ARG_STRLEN);
663		if (!len || len > MAX_ARG_STRLEN)
664			return -EINVAL;
665		p += len;
666	}
667	__put_user(NULL, argv);
668	current->mm->arg_end = (unsigned long) p;
669
670	/* fill in the envv[] array */
671	current->mm->env_start = (unsigned long) p;
672	for (loop = bprm->envc; loop > 0; loop--) {
673		__put_user((elf_caddr_t)(unsigned long) p, envp++);
674		len = strnlen_user(p, MAX_ARG_STRLEN);
675		if (!len || len > MAX_ARG_STRLEN)
676			return -EINVAL;
677		p += len;
678	}
679	__put_user(NULL, envp);
680	current->mm->env_end = (unsigned long) p;
681
682	mm->start_stack = (unsigned long) sp;
683	return 0;
684}
685
686/*****************************************************************************/
687/*
688 * transfer the program arguments and environment from the holding pages onto
689 * the stack
690 */
691#ifndef CONFIG_MMU
692static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
693					    unsigned long *_sp)
694{
695	unsigned long index, stop, sp;
696	char *src;
697	int ret = 0;
698
699	stop = bprm->p >> PAGE_SHIFT;
700	sp = *_sp;
701
702	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
703		src = kmap(bprm->page[index]);
704		sp -= PAGE_SIZE;
705		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
706			ret = -EFAULT;
707		kunmap(bprm->page[index]);
708		if (ret < 0)
709			goto out;
710	}
711
712	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
713
714out:
715	return ret;
716}
717#endif
718
719/*****************************************************************************/
720/*
721 * load the appropriate binary image (executable or interpreter) into memory
722 * - we assume no MMU is available
723 * - if no other PIC bits are set in params->hdr->e_flags
724 *   - we assume that the LOADable segments in the binary are independently relocatable
725 *   - we assume R/O executable segments are shareable
726 * - else
727 *   - we assume the loadable parts of the image to require fixed displacement
728 *   - the image is not shareable
729 */
730static int elf_fdpic_map_file(struct elf_fdpic_params *params,
731			      struct file *file,
732			      struct mm_struct *mm,
733			      const char *what)
734{
735	struct elf32_fdpic_loadmap *loadmap;
736#ifdef CONFIG_MMU
737	struct elf32_fdpic_loadseg *mseg;
738#endif
739	struct elf32_fdpic_loadseg *seg;
740	struct elf32_phdr *phdr;
741	unsigned long load_addr, stop;
742	unsigned nloads, tmp;
743	size_t size;
744	int loop, ret;
745
746	/* allocate a load map table */
747	nloads = 0;
748	for (loop = 0; loop < params->hdr.e_phnum; loop++)
749		if (params->phdrs[loop].p_type == PT_LOAD)
750			nloads++;
751
752	if (nloads == 0)
753		return -ELIBBAD;
754
755	size = sizeof(*loadmap) + nloads * sizeof(*seg);
756	loadmap = kzalloc(size, GFP_KERNEL);
757	if (!loadmap)
758		return -ENOMEM;
759
760	params->loadmap = loadmap;
761
762	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
763	loadmap->nsegs = nloads;
764
765	load_addr = params->load_addr;
766	seg = loadmap->segs;
767
768	/* map the requested LOADs into the memory space */
769	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
770	case ELF_FDPIC_FLAG_CONSTDISP:
771	case ELF_FDPIC_FLAG_CONTIGUOUS:
772#ifndef CONFIG_MMU
773		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
774		if (ret < 0)
775			return ret;
776		break;
777#endif
778	default:
779		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
780		if (ret < 0)
781			return ret;
782		break;
783	}
784
785	/* map the entry point */
786	if (params->hdr.e_entry) {
787		seg = loadmap->segs;
788		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
789			if (params->hdr.e_entry >= seg->p_vaddr &&
790			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
791				params->entry_addr =
792					(params->hdr.e_entry - seg->p_vaddr) +
793					seg->addr;
794				break;
795			}
796		}
797	}
798
799	/* determine where the program header table has wound up if mapped */
800	stop = params->hdr.e_phoff;
801	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
802	phdr = params->phdrs;
803
804	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
805		if (phdr->p_type != PT_LOAD)
806			continue;
807
808		if (phdr->p_offset > params->hdr.e_phoff ||
809		    phdr->p_offset + phdr->p_filesz < stop)
810			continue;
811
812		seg = loadmap->segs;
813		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
814			if (phdr->p_vaddr >= seg->p_vaddr &&
815			    phdr->p_vaddr + phdr->p_filesz <=
816			    seg->p_vaddr + seg->p_memsz) {
817				params->ph_addr =
818					(phdr->p_vaddr - seg->p_vaddr) +
819					seg->addr +
820					params->hdr.e_phoff - phdr->p_offset;
821				break;
822			}
823		}
824		break;
825	}
826
827	/* determine where the dynamic section has wound up if there is one */
828	phdr = params->phdrs;
829	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
830		if (phdr->p_type != PT_DYNAMIC)
831			continue;
832
833		seg = loadmap->segs;
834		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
835			if (phdr->p_vaddr >= seg->p_vaddr &&
836			    phdr->p_vaddr + phdr->p_memsz <=
837			    seg->p_vaddr + seg->p_memsz) {
838				params->dynamic_addr =
839					(phdr->p_vaddr - seg->p_vaddr) +
840					seg->addr;
841
842				/* check the dynamic section contains at least
843				 * one item, and that the last item is a NULL
844				 * entry */
845				if (phdr->p_memsz == 0 ||
846				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
847					goto dynamic_error;
848
849				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
850				if (((Elf32_Dyn *)
851				     params->dynamic_addr)[tmp - 1].d_tag != 0)
852					goto dynamic_error;
853				break;
854			}
855		}
856		break;
857	}
858
859	/* now elide adjacent segments in the load map on MMU linux
860	 * - on uClinux the holes between may actually be filled with system
861	 *   stuff or stuff from other processes
862	 */
863#ifdef CONFIG_MMU
864	nloads = loadmap->nsegs;
865	mseg = loadmap->segs;
866	seg = mseg + 1;
867	for (loop = 1; loop < nloads; loop++) {
868		/* see if we have a candidate for merging */
869		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
870			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
871			if (load_addr == (seg->addr & PAGE_MASK)) {
872				mseg->p_memsz +=
873					load_addr -
874					(mseg->addr + mseg->p_memsz);
875				mseg->p_memsz += seg->addr & ~PAGE_MASK;
876				mseg->p_memsz += seg->p_memsz;
877				loadmap->nsegs--;
878				continue;
879			}
880		}
881
882		mseg++;
883		if (mseg != seg)
884			*mseg = *seg;
885	}
886#endif
887
888	kdebug("Mapped Object [%s]:", what);
889	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
890	kdebug("- entry    : %lx", params->entry_addr);
891	kdebug("- PHDR[]   : %lx", params->ph_addr);
892	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
893	seg = loadmap->segs;
894	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
895		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
896		       loop,
897		       seg->addr, seg->addr + seg->p_memsz - 1,
898		       seg->p_vaddr, seg->p_memsz);
899
900	return 0;
901
902dynamic_error:
903	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
904	       what, file_inode(file)->i_ino);
905	return -ELIBBAD;
906}
907
908/*****************************************************************************/
909/*
910 * map a file with constant displacement under uClinux
911 */
912#ifndef CONFIG_MMU
913static int elf_fdpic_map_file_constdisp_on_uclinux(
914	struct elf_fdpic_params *params,
915	struct file *file,
916	struct mm_struct *mm)
917{
918	struct elf32_fdpic_loadseg *seg;
919	struct elf32_phdr *phdr;
920	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
921	int loop, ret;
922
923	load_addr = params->load_addr;
924	seg = params->loadmap->segs;
925
926	/* determine the bounds of the contiguous overall allocation we must
927	 * make */
928	phdr = params->phdrs;
929	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
930		if (params->phdrs[loop].p_type != PT_LOAD)
931			continue;
932
933		if (base > phdr->p_vaddr)
934			base = phdr->p_vaddr;
935		if (top < phdr->p_vaddr + phdr->p_memsz)
936			top = phdr->p_vaddr + phdr->p_memsz;
937	}
938
939	/* allocate one big anon block for everything */
940	mflags = MAP_PRIVATE;
941	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
942		mflags |= MAP_EXECUTABLE;
943
944	maddr = vm_mmap(NULL, load_addr, top - base,
945			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
946	if (IS_ERR_VALUE(maddr))
947		return (int) maddr;
948
949	if (load_addr != 0)
950		load_addr += PAGE_ALIGN(top - base);
951
952	/* and then load the file segments into it */
953	phdr = params->phdrs;
954	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
955		if (params->phdrs[loop].p_type != PT_LOAD)
956			continue;
957
958		seg->addr = maddr + (phdr->p_vaddr - base);
959		seg->p_vaddr = phdr->p_vaddr;
960		seg->p_memsz = phdr->p_memsz;
961
962		ret = read_code(file, seg->addr, phdr->p_offset,
963				       phdr->p_filesz);
964		if (ret < 0)
965			return ret;
966
967		/* map the ELF header address if in this segment */
968		if (phdr->p_offset == 0)
969			params->elfhdr_addr = seg->addr;
970
971		/* clear any space allocated but not loaded */
972		if (phdr->p_filesz < phdr->p_memsz) {
973			if (clear_user((void *) (seg->addr + phdr->p_filesz),
974				       phdr->p_memsz - phdr->p_filesz))
975				return -EFAULT;
976		}
977
978		if (mm) {
979			if (phdr->p_flags & PF_X) {
980				if (!mm->start_code) {
981					mm->start_code = seg->addr;
982					mm->end_code = seg->addr +
983						phdr->p_memsz;
984				}
985			} else if (!mm->start_data) {
986				mm->start_data = seg->addr;
987				mm->end_data = seg->addr + phdr->p_memsz;
988			}
989		}
990
991		seg++;
992	}
993
994	return 0;
995}
996#endif
997
998/*****************************************************************************/
999/*
1000 * map a binary by direct mmap() of the individual PT_LOAD segments
1001 */
1002static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1003					     struct file *file,
1004					     struct mm_struct *mm)
1005{
1006	struct elf32_fdpic_loadseg *seg;
1007	struct elf32_phdr *phdr;
1008	unsigned long load_addr, delta_vaddr;
1009	int loop, dvset;
1010
1011	load_addr = params->load_addr;
1012	delta_vaddr = 0;
1013	dvset = 0;
1014
1015	seg = params->loadmap->segs;
1016
1017	/* deal with each load segment separately */
1018	phdr = params->phdrs;
1019	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1020		unsigned long maddr, disp, excess, excess1;
1021		int prot = 0, flags;
1022
1023		if (phdr->p_type != PT_LOAD)
1024			continue;
1025
1026		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1027		       (unsigned long) phdr->p_vaddr,
1028		       (unsigned long) phdr->p_offset,
1029		       (unsigned long) phdr->p_filesz,
1030		       (unsigned long) phdr->p_memsz);
1031
1032		/* determine the mapping parameters */
1033		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1034		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1035		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1036
1037		flags = MAP_PRIVATE | MAP_DENYWRITE;
1038		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1039			flags |= MAP_EXECUTABLE;
1040
1041		maddr = 0;
1042
1043		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1044		case ELF_FDPIC_FLAG_INDEPENDENT:
1045			/* PT_LOADs are independently locatable */
1046			break;
1047
1048		case ELF_FDPIC_FLAG_HONOURVADDR:
1049			/* the specified virtual address must be honoured */
1050			maddr = phdr->p_vaddr;
1051			flags |= MAP_FIXED;
1052			break;
1053
1054		case ELF_FDPIC_FLAG_CONSTDISP:
1055			/* constant displacement
1056			 * - can be mapped anywhere, but must be mapped as a
1057			 *   unit
1058			 */
1059			if (!dvset) {
1060				maddr = load_addr;
1061				delta_vaddr = phdr->p_vaddr;
1062				dvset = 1;
1063			} else {
1064				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1065				flags |= MAP_FIXED;
1066			}
1067			break;
1068
1069		case ELF_FDPIC_FLAG_CONTIGUOUS:
1070			/* contiguity handled later */
1071			break;
1072
1073		default:
1074			BUG();
1075		}
1076
1077		maddr &= PAGE_MASK;
1078
1079		/* create the mapping */
1080		disp = phdr->p_vaddr & ~PAGE_MASK;
1081		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1082				phdr->p_offset - disp);
1083
1084		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1085		       loop, phdr->p_memsz + disp, prot, flags,
1086		       phdr->p_offset - disp, maddr);
1087
1088		if (IS_ERR_VALUE(maddr))
1089			return (int) maddr;
1090
1091		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1092		    ELF_FDPIC_FLAG_CONTIGUOUS)
1093			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1094
1095		seg->addr = maddr + disp;
1096		seg->p_vaddr = phdr->p_vaddr;
1097		seg->p_memsz = phdr->p_memsz;
1098
1099		/* map the ELF header address if in this segment */
1100		if (phdr->p_offset == 0)
1101			params->elfhdr_addr = seg->addr;
1102
1103		/* clear the bit between beginning of mapping and beginning of
1104		 * PT_LOAD */
1105		if (prot & PROT_WRITE && disp > 0) {
1106			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1107			if (clear_user((void __user *) maddr, disp))
1108				return -EFAULT;
1109			maddr += disp;
1110		}
1111
1112		/* clear any space allocated but not loaded
1113		 * - on uClinux we can just clear the lot
1114		 * - on MMU linux we'll get a SIGBUS beyond the last page
1115		 *   extant in the file
1116		 */
1117		excess = phdr->p_memsz - phdr->p_filesz;
1118		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1119
1120#ifdef CONFIG_MMU
1121		if (excess > excess1) {
1122			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1123			unsigned long xmaddr;
1124
1125			flags |= MAP_FIXED | MAP_ANONYMOUS;
1126			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1127					 prot, flags, 0);
1128
1129			kdebug("mmap[%d] <anon>"
1130			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1131			       loop, xaddr, excess - excess1, prot, flags,
1132			       xmaddr);
1133
1134			if (xmaddr != xaddr)
1135				return -ENOMEM;
1136		}
1137
1138		if (prot & PROT_WRITE && excess1 > 0) {
1139			kdebug("clear[%d] ad=%lx sz=%lx",
1140			       loop, maddr + phdr->p_filesz, excess1);
1141			if (clear_user((void __user *) maddr + phdr->p_filesz,
1142				       excess1))
1143				return -EFAULT;
1144		}
1145
1146#else
1147		if (excess > 0) {
1148			kdebug("clear[%d] ad=%lx sz=%lx",
1149			       loop, maddr + phdr->p_filesz, excess);
1150			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1151				return -EFAULT;
1152		}
1153#endif
1154
1155		if (mm) {
1156			if (phdr->p_flags & PF_X) {
1157				if (!mm->start_code) {
1158					mm->start_code = maddr;
1159					mm->end_code = maddr + phdr->p_memsz;
1160				}
1161			} else if (!mm->start_data) {
1162				mm->start_data = maddr;
1163				mm->end_data = maddr + phdr->p_memsz;
1164			}
1165		}
1166
1167		seg++;
1168	}
1169
1170	return 0;
1171}
1172
1173/*****************************************************************************/
1174/*
1175 * ELF-FDPIC core dumper
1176 *
1177 * Modelled on fs/exec.c:aout_core_dump()
1178 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1179 *
1180 * Modelled on fs/binfmt_elf.c core dumper
1181 */
1182#ifdef CONFIG_ELF_CORE
1183
1184/*
1185 * Decide whether a segment is worth dumping; default is yes to be
1186 * sure (missing info is worse than too much; etc).
1187 * Personally I'd include everything, and use the coredump limit...
1188 *
1189 * I think we should skip something. But I am not sure how. H.J.
1190 */
1191static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1192{
1193	int dump_ok;
1194
1195	/* Do not dump I/O mapped devices or special mappings */
1196	if (vma->vm_flags & VM_IO) {
1197		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1198		return 0;
1199	}
1200
1201	/* If we may not read the contents, don't allow us to dump
1202	 * them either. "dump_write()" can't handle it anyway.
1203	 */
1204	if (!(vma->vm_flags & VM_READ)) {
1205		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1206		return 0;
1207	}
1208
1209	/* By default, dump shared memory if mapped from an anonymous file. */
1210	if (vma->vm_flags & VM_SHARED) {
1211		if (file_inode(vma->vm_file)->i_nlink == 0) {
1212			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1213			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1214			       vma->vm_flags, dump_ok ? "yes" : "no");
1215			return dump_ok;
1216		}
1217
1218		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1219		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1220		       vma->vm_flags, dump_ok ? "yes" : "no");
1221		return dump_ok;
1222	}
1223
1224#ifdef CONFIG_MMU
1225	/* By default, if it hasn't been written to, don't write it out */
1226	if (!vma->anon_vma) {
1227		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1228		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1229		       vma->vm_flags, dump_ok ? "yes" : "no");
1230		return dump_ok;
1231	}
1232#endif
1233
1234	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1235	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1236	       dump_ok ? "yes" : "no");
1237	return dump_ok;
1238}
1239
1240/* An ELF note in memory */
1241struct memelfnote
1242{
1243	const char *name;
1244	int type;
1245	unsigned int datasz;
1246	void *data;
1247};
1248
1249static int notesize(struct memelfnote *en)
1250{
1251	int sz;
1252
1253	sz = sizeof(struct elf_note);
1254	sz += roundup(strlen(en->name) + 1, 4);
1255	sz += roundup(en->datasz, 4);
1256
1257	return sz;
1258}
1259
1260/* #define DEBUG */
1261
1262static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1263{
1264	struct elf_note en;
1265	en.n_namesz = strlen(men->name) + 1;
1266	en.n_descsz = men->datasz;
1267	en.n_type = men->type;
1268
1269	return dump_emit(cprm, &en, sizeof(en)) &&
1270		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1271		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1272}
1273
1274static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1275{
1276	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1277	elf->e_ident[EI_CLASS] = ELF_CLASS;
1278	elf->e_ident[EI_DATA] = ELF_DATA;
1279	elf->e_ident[EI_VERSION] = EV_CURRENT;
1280	elf->e_ident[EI_OSABI] = ELF_OSABI;
1281	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1282
1283	elf->e_type = ET_CORE;
1284	elf->e_machine = ELF_ARCH;
1285	elf->e_version = EV_CURRENT;
1286	elf->e_entry = 0;
1287	elf->e_phoff = sizeof(struct elfhdr);
1288	elf->e_shoff = 0;
1289	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1290	elf->e_ehsize = sizeof(struct elfhdr);
1291	elf->e_phentsize = sizeof(struct elf_phdr);
1292	elf->e_phnum = segs;
1293	elf->e_shentsize = 0;
1294	elf->e_shnum = 0;
1295	elf->e_shstrndx = 0;
1296	return;
1297}
1298
1299static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1300{
1301	phdr->p_type = PT_NOTE;
1302	phdr->p_offset = offset;
1303	phdr->p_vaddr = 0;
1304	phdr->p_paddr = 0;
1305	phdr->p_filesz = sz;
1306	phdr->p_memsz = 0;
1307	phdr->p_flags = 0;
1308	phdr->p_align = 0;
1309	return;
1310}
1311
1312static inline void fill_note(struct memelfnote *note, const char *name, int type,
1313		unsigned int sz, void *data)
1314{
1315	note->name = name;
1316	note->type = type;
1317	note->datasz = sz;
1318	note->data = data;
1319	return;
1320}
1321
1322/*
1323 * fill up all the fields in prstatus from the given task struct, except
1324 * registers which need to be filled up separately.
1325 */
1326static void fill_prstatus(struct elf_prstatus *prstatus,
1327			  struct task_struct *p, long signr)
1328{
1329	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1330	prstatus->pr_sigpend = p->pending.signal.sig[0];
1331	prstatus->pr_sighold = p->blocked.sig[0];
1332	rcu_read_lock();
1333	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1334	rcu_read_unlock();
1335	prstatus->pr_pid = task_pid_vnr(p);
1336	prstatus->pr_pgrp = task_pgrp_vnr(p);
1337	prstatus->pr_sid = task_session_vnr(p);
1338	if (thread_group_leader(p)) {
1339		struct task_cputime cputime;
1340
1341		/*
1342		 * This is the record for the group leader.  It shows the
1343		 * group-wide total, not its individual thread total.
1344		 */
1345		thread_group_cputime(p, &cputime);
1346		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1347		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1348	} else {
1349		cputime_t utime, stime;
1350
1351		task_cputime(p, &utime, &stime);
1352		cputime_to_timeval(utime, &prstatus->pr_utime);
1353		cputime_to_timeval(stime, &prstatus->pr_stime);
1354	}
1355	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1356	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1357
1358	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1359	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1360}
1361
1362static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1363		       struct mm_struct *mm)
1364{
1365	const struct cred *cred;
1366	unsigned int i, len;
1367
1368	/* first copy the parameters from user space */
1369	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1370
1371	len = mm->arg_end - mm->arg_start;
1372	if (len >= ELF_PRARGSZ)
1373		len = ELF_PRARGSZ - 1;
1374	if (copy_from_user(&psinfo->pr_psargs,
1375		           (const char __user *) mm->arg_start, len))
1376		return -EFAULT;
1377	for (i = 0; i < len; i++)
1378		if (psinfo->pr_psargs[i] == 0)
1379			psinfo->pr_psargs[i] = ' ';
1380	psinfo->pr_psargs[len] = 0;
1381
1382	rcu_read_lock();
1383	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1384	rcu_read_unlock();
1385	psinfo->pr_pid = task_pid_vnr(p);
1386	psinfo->pr_pgrp = task_pgrp_vnr(p);
1387	psinfo->pr_sid = task_session_vnr(p);
1388
1389	i = p->state ? ffz(~p->state) + 1 : 0;
1390	psinfo->pr_state = i;
1391	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1392	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1393	psinfo->pr_nice = task_nice(p);
1394	psinfo->pr_flag = p->flags;
1395	rcu_read_lock();
1396	cred = __task_cred(p);
1397	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1398	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1399	rcu_read_unlock();
1400	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1401
1402	return 0;
1403}
1404
1405/* Here is the structure in which status of each thread is captured. */
1406struct elf_thread_status
1407{
1408	struct list_head list;
1409	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1410	elf_fpregset_t fpu;		/* NT_PRFPREG */
1411	struct task_struct *thread;
1412#ifdef ELF_CORE_COPY_XFPREGS
1413	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1414#endif
1415	struct memelfnote notes[3];
1416	int num_notes;
1417};
1418
1419/*
1420 * In order to add the specific thread information for the elf file format,
1421 * we need to keep a linked list of every thread's pr_status and then create
1422 * a single section for them in the final core file.
1423 */
1424static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1425{
1426	struct task_struct *p = t->thread;
1427	int sz = 0;
1428
1429	t->num_notes = 0;
1430
1431	fill_prstatus(&t->prstatus, p, signr);
1432	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1433
1434	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1435		  &t->prstatus);
1436	t->num_notes++;
1437	sz += notesize(&t->notes[0]);
1438
1439	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1440	if (t->prstatus.pr_fpvalid) {
1441		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1442			  &t->fpu);
1443		t->num_notes++;
1444		sz += notesize(&t->notes[1]);
1445	}
1446
1447#ifdef ELF_CORE_COPY_XFPREGS
1448	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1449		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1450			  sizeof(t->xfpu), &t->xfpu);
1451		t->num_notes++;
1452		sz += notesize(&t->notes[2]);
1453	}
1454#endif
1455	return sz;
1456}
1457
1458static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1459			     elf_addr_t e_shoff, int segs)
1460{
1461	elf->e_shoff = e_shoff;
1462	elf->e_shentsize = sizeof(*shdr4extnum);
1463	elf->e_shnum = 1;
1464	elf->e_shstrndx = SHN_UNDEF;
1465
1466	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1467
1468	shdr4extnum->sh_type = SHT_NULL;
1469	shdr4extnum->sh_size = elf->e_shnum;
1470	shdr4extnum->sh_link = elf->e_shstrndx;
1471	shdr4extnum->sh_info = segs;
1472}
1473
1474/*
1475 * dump the segments for an MMU process
1476 */
1477static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1478{
1479	struct vm_area_struct *vma;
1480
1481	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1482		unsigned long addr;
1483
1484		if (!maydump(vma, cprm->mm_flags))
1485			continue;
1486
1487#ifdef CONFIG_MMU
1488		for (addr = vma->vm_start; addr < vma->vm_end;
1489							addr += PAGE_SIZE) {
1490			bool res;
1491			struct page *page = get_dump_page(addr);
1492			if (page) {
1493				void *kaddr = kmap(page);
1494				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1495				kunmap(page);
1496				page_cache_release(page);
1497			} else {
1498				res = dump_skip(cprm, PAGE_SIZE);
1499			}
1500			if (!res)
1501				return false;
1502		}
1503#else
1504		if (!dump_emit(cprm, (void *) vma->vm_start,
1505				vma->vm_end - vma->vm_start))
1506			return false;
1507#endif
1508	}
1509	return true;
1510}
1511
1512static size_t elf_core_vma_data_size(unsigned long mm_flags)
1513{
1514	struct vm_area_struct *vma;
1515	size_t size = 0;
1516
1517	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1518		if (maydump(vma, mm_flags))
1519			size += vma->vm_end - vma->vm_start;
1520	return size;
1521}
1522
1523/*
1524 * Actual dumper
1525 *
1526 * This is a two-pass process; first we find the offsets of the bits,
1527 * and then they are actually written out.  If we run out of core limit
1528 * we just truncate.
1529 */
1530static int elf_fdpic_core_dump(struct coredump_params *cprm)
1531{
1532#define	NUM_NOTES	6
1533	int has_dumped = 0;
1534	mm_segment_t fs;
1535	int segs;
1536	int i;
1537	struct vm_area_struct *vma;
1538	struct elfhdr *elf = NULL;
1539	loff_t offset = 0, dataoff;
1540	int numnote;
1541	struct memelfnote *notes = NULL;
1542	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1543	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1544 	LIST_HEAD(thread_list);
1545 	struct list_head *t;
1546	elf_fpregset_t *fpu = NULL;
1547#ifdef ELF_CORE_COPY_XFPREGS
1548	elf_fpxregset_t *xfpu = NULL;
1549#endif
1550	int thread_status_size = 0;
1551	elf_addr_t *auxv;
1552	struct elf_phdr *phdr4note = NULL;
1553	struct elf_shdr *shdr4extnum = NULL;
1554	Elf_Half e_phnum;
1555	elf_addr_t e_shoff;
1556	struct core_thread *ct;
1557	struct elf_thread_status *tmp;
1558
1559	/*
1560	 * We no longer stop all VM operations.
1561	 *
1562	 * This is because those proceses that could possibly change map_count
1563	 * or the mmap / vma pages are now blocked in do_exit on current
1564	 * finishing this core dump.
1565	 *
1566	 * Only ptrace can touch these memory addresses, but it doesn't change
1567	 * the map_count or the pages allocated. So no possibility of crashing
1568	 * exists while dumping the mm->vm_next areas to the core file.
1569	 */
1570
1571	/* alloc memory for large data structures: too large to be on stack */
1572	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1573	if (!elf)
1574		goto cleanup;
1575	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1576	if (!prstatus)
1577		goto cleanup;
1578	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1579	if (!psinfo)
1580		goto cleanup;
1581	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1582	if (!notes)
1583		goto cleanup;
1584	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1585	if (!fpu)
1586		goto cleanup;
1587#ifdef ELF_CORE_COPY_XFPREGS
1588	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1589	if (!xfpu)
1590		goto cleanup;
1591#endif
1592
1593	for (ct = current->mm->core_state->dumper.next;
1594					ct; ct = ct->next) {
1595		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1596		if (!tmp)
1597			goto cleanup;
1598
1599		tmp->thread = ct->task;
1600		list_add(&tmp->list, &thread_list);
1601	}
1602
1603	list_for_each(t, &thread_list) {
1604		struct elf_thread_status *tmp;
1605		int sz;
1606
1607		tmp = list_entry(t, struct elf_thread_status, list);
1608		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1609		thread_status_size += sz;
1610	}
1611
1612	/* now collect the dump for the current */
1613	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1614	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1615
1616	segs = current->mm->map_count;
1617	segs += elf_core_extra_phdrs();
1618
1619	/* for notes section */
1620	segs++;
1621
1622	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1623	 * this, kernel supports extended numbering. Have a look at
1624	 * include/linux/elf.h for further information. */
1625	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1626
1627	/* Set up header */
1628	fill_elf_fdpic_header(elf, e_phnum);
1629
1630	has_dumped = 1;
1631	/*
1632	 * Set up the notes in similar form to SVR4 core dumps made
1633	 * with info from their /proc.
1634	 */
1635
1636	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1637	fill_psinfo(psinfo, current->group_leader, current->mm);
1638	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1639
1640	numnote = 2;
1641
1642	auxv = (elf_addr_t *) current->mm->saved_auxv;
1643
1644	i = 0;
1645	do
1646		i += 2;
1647	while (auxv[i - 2] != AT_NULL);
1648	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1649		  i * sizeof(elf_addr_t), auxv);
1650
1651  	/* Try to dump the FPU. */
1652	if ((prstatus->pr_fpvalid =
1653	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1654		fill_note(notes + numnote++,
1655			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1656#ifdef ELF_CORE_COPY_XFPREGS
1657	if (elf_core_copy_task_xfpregs(current, xfpu))
1658		fill_note(notes + numnote++,
1659			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1660#endif
1661
1662	fs = get_fs();
1663	set_fs(KERNEL_DS);
1664
1665	offset += sizeof(*elf);				/* Elf header */
1666	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1667
1668	/* Write notes phdr entry */
1669	{
1670		int sz = 0;
1671
1672		for (i = 0; i < numnote; i++)
1673			sz += notesize(notes + i);
1674
1675		sz += thread_status_size;
1676
1677		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1678		if (!phdr4note)
1679			goto end_coredump;
1680
1681		fill_elf_note_phdr(phdr4note, sz, offset);
1682		offset += sz;
1683	}
1684
1685	/* Page-align dumped data */
1686	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1687
1688	offset += elf_core_vma_data_size(cprm->mm_flags);
1689	offset += elf_core_extra_data_size();
1690	e_shoff = offset;
1691
1692	if (e_phnum == PN_XNUM) {
1693		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1694		if (!shdr4extnum)
1695			goto end_coredump;
1696		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1697	}
1698
1699	offset = dataoff;
1700
1701	if (!dump_emit(cprm, elf, sizeof(*elf)))
1702		goto end_coredump;
1703
1704	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1705		goto end_coredump;
1706
1707	/* write program headers for segments dump */
1708	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1709		struct elf_phdr phdr;
1710		size_t sz;
1711
1712		sz = vma->vm_end - vma->vm_start;
1713
1714		phdr.p_type = PT_LOAD;
1715		phdr.p_offset = offset;
1716		phdr.p_vaddr = vma->vm_start;
1717		phdr.p_paddr = 0;
1718		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1719		phdr.p_memsz = sz;
1720		offset += phdr.p_filesz;
1721		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1722		if (vma->vm_flags & VM_WRITE)
1723			phdr.p_flags |= PF_W;
1724		if (vma->vm_flags & VM_EXEC)
1725			phdr.p_flags |= PF_X;
1726		phdr.p_align = ELF_EXEC_PAGESIZE;
1727
1728		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1729			goto end_coredump;
1730	}
1731
1732	if (!elf_core_write_extra_phdrs(cprm, offset))
1733		goto end_coredump;
1734
1735 	/* write out the notes section */
1736	for (i = 0; i < numnote; i++)
1737		if (!writenote(notes + i, cprm))
1738			goto end_coredump;
1739
1740	/* write out the thread status notes section */
1741	list_for_each(t, &thread_list) {
1742		struct elf_thread_status *tmp =
1743				list_entry(t, struct elf_thread_status, list);
1744
1745		for (i = 0; i < tmp->num_notes; i++)
1746			if (!writenote(&tmp->notes[i], cprm))
1747				goto end_coredump;
1748	}
1749
1750	if (!dump_skip(cprm, dataoff - cprm->written))
1751		goto end_coredump;
1752
1753	if (!elf_fdpic_dump_segments(cprm))
1754		goto end_coredump;
1755
1756	if (!elf_core_write_extra_data(cprm))
1757		goto end_coredump;
1758
1759	if (e_phnum == PN_XNUM) {
1760		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1761			goto end_coredump;
1762	}
1763
1764	if (cprm->file->f_pos != offset) {
1765		/* Sanity check */
1766		printk(KERN_WARNING
1767		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1768		       cprm->file->f_pos, offset);
1769	}
1770
1771end_coredump:
1772	set_fs(fs);
1773
1774cleanup:
1775	while (!list_empty(&thread_list)) {
1776		struct list_head *tmp = thread_list.next;
1777		list_del(tmp);
1778		kfree(list_entry(tmp, struct elf_thread_status, list));
1779	}
1780	kfree(phdr4note);
1781	kfree(elf);
1782	kfree(prstatus);
1783	kfree(psinfo);
1784	kfree(notes);
1785	kfree(fpu);
1786	kfree(shdr4extnum);
1787#ifdef ELF_CORE_COPY_XFPREGS
1788	kfree(xfpu);
1789#endif
1790	return has_dumped;
1791#undef NUM_NOTES
1792}
1793
1794#endif		/* CONFIG_ELF_CORE */
1795