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#include <linux/dax.h>
39
40#include <asm/uaccess.h>
41#include <asm/param.h>
42#include <asm/pgalloc.h>
43
44typedef char *elf_caddr_t;
45
46#if 0
47#define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
48#else
49#define kdebug(fmt, ...) do {} while(0)
50#endif
51
52#if 0
53#define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
54#else
55#define kdcore(fmt, ...) do {} while(0)
56#endif
57
58MODULE_LICENSE("GPL");
59
60static int load_elf_fdpic_binary(struct linux_binprm *);
61static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
62static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
63			      struct mm_struct *, const char *);
64
65static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
66				   struct elf_fdpic_params *,
67				   struct elf_fdpic_params *);
68
69#ifndef CONFIG_MMU
70static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
71					    unsigned long *);
72static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
73						   struct file *,
74						   struct mm_struct *);
75#endif
76
77static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
78					     struct file *, struct mm_struct *);
79
80#ifdef CONFIG_ELF_CORE
81static int elf_fdpic_core_dump(struct coredump_params *cprm);
82#endif
83
84static struct linux_binfmt elf_fdpic_format = {
85	.module		= THIS_MODULE,
86	.load_binary	= load_elf_fdpic_binary,
87#ifdef CONFIG_ELF_CORE
88	.core_dump	= elf_fdpic_core_dump,
89#endif
90	.min_coredump	= ELF_EXEC_PAGESIZE,
91};
92
93static int __init init_elf_fdpic_binfmt(void)
94{
95	register_binfmt(&elf_fdpic_format);
96	return 0;
97}
98
99static void __exit exit_elf_fdpic_binfmt(void)
100{
101	unregister_binfmt(&elf_fdpic_format);
102}
103
104core_initcall(init_elf_fdpic_binfmt);
105module_exit(exit_elf_fdpic_binfmt);
106
107static int is_elf(struct elfhdr *hdr, struct file *file)
108{
109	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
110		return 0;
111	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
112		return 0;
113	if (!elf_check_arch(hdr))
114		return 0;
115	if (!file->f_op->mmap)
116		return 0;
117	return 1;
118}
119
120#ifndef elf_check_fdpic
121#define elf_check_fdpic(x) 0
122#endif
123
124#ifndef elf_check_const_displacement
125#define elf_check_const_displacement(x) 0
126#endif
127
128static int is_constdisp(struct elfhdr *hdr)
129{
130	if (!elf_check_fdpic(hdr))
131		return 1;
132	if (elf_check_const_displacement(hdr))
133		return 1;
134	return 0;
135}
136
137/*****************************************************************************/
138/*
139 * read the program headers table into memory
140 */
141static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
142				 struct file *file)
143{
144	struct elf32_phdr *phdr;
145	unsigned long size;
146	int retval, loop;
147
148	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
149		return -ENOMEM;
150	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
151		return -ENOMEM;
152
153	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
154	params->phdrs = kmalloc(size, GFP_KERNEL);
155	if (!params->phdrs)
156		return -ENOMEM;
157
158	retval = kernel_read(file, params->hdr.e_phoff,
159			     (char *) params->phdrs, size);
160	if (unlikely(retval != size))
161		return retval < 0 ? retval : -ENOEXEC;
162
163	/* determine stack size for this binary */
164	phdr = params->phdrs;
165	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
166		if (phdr->p_type != PT_GNU_STACK)
167			continue;
168
169		if (phdr->p_flags & PF_X)
170			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
171		else
172			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
173
174		params->stack_size = phdr->p_memsz;
175		break;
176	}
177
178	return 0;
179}
180
181/*****************************************************************************/
182/*
183 * load an fdpic binary into various bits of memory
184 */
185static int load_elf_fdpic_binary(struct linux_binprm *bprm)
186{
187	struct elf_fdpic_params exec_params, interp_params;
188	struct pt_regs *regs = current_pt_regs();
189	struct elf_phdr *phdr;
190	unsigned long stack_size, entryaddr;
191#ifdef ELF_FDPIC_PLAT_INIT
192	unsigned long dynaddr;
193#endif
194#ifndef CONFIG_MMU
195	unsigned long stack_prot;
196#endif
197	struct file *interpreter = NULL; /* to shut gcc up */
198	char *interpreter_name = NULL;
199	int executable_stack;
200	int retval, i;
201
202	kdebug("____ LOAD %d ____", current->pid);
203
204	memset(&exec_params, 0, sizeof(exec_params));
205	memset(&interp_params, 0, sizeof(interp_params));
206
207	exec_params.hdr = *(struct elfhdr *) bprm->buf;
208	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
209
210	/* check that this is a binary we know how to deal with */
211	retval = -ENOEXEC;
212	if (!is_elf(&exec_params.hdr, bprm->file))
213		goto error;
214	if (!elf_check_fdpic(&exec_params.hdr)) {
215#ifdef CONFIG_MMU
216		/* binfmt_elf handles non-fdpic elf except on nommu */
217		goto error;
218#else
219		/* nommu can only load ET_DYN (PIE) ELF */
220		if (exec_params.hdr.e_type != ET_DYN)
221			goto error;
222#endif
223	}
224
225	/* read the program header table */
226	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
227	if (retval < 0)
228		goto error;
229
230	/* scan for a program header that specifies an interpreter */
231	phdr = exec_params.phdrs;
232
233	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
234		switch (phdr->p_type) {
235		case PT_INTERP:
236			retval = -ENOMEM;
237			if (phdr->p_filesz > PATH_MAX)
238				goto error;
239			retval = -ENOENT;
240			if (phdr->p_filesz < 2)
241				goto error;
242
243			/* read the name of the interpreter into memory */
244			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
245			if (!interpreter_name)
246				goto error;
247
248			retval = kernel_read(bprm->file,
249					     phdr->p_offset,
250					     interpreter_name,
251					     phdr->p_filesz);
252			if (unlikely(retval != phdr->p_filesz)) {
253				if (retval >= 0)
254					retval = -ENOEXEC;
255				goto error;
256			}
257
258			retval = -ENOENT;
259			if (interpreter_name[phdr->p_filesz - 1] != '\0')
260				goto error;
261
262			kdebug("Using ELF interpreter %s", interpreter_name);
263
264			/* replace the program with the interpreter */
265			interpreter = open_exec(interpreter_name);
266			retval = PTR_ERR(interpreter);
267			if (IS_ERR(interpreter)) {
268				interpreter = NULL;
269				goto error;
270			}
271
272			/*
273			 * If the binary is not readable then enforce
274			 * mm->dumpable = 0 regardless of the interpreter's
275			 * permissions.
276			 */
277			would_dump(bprm, interpreter);
278
279			retval = kernel_read(interpreter, 0, bprm->buf,
280					     BINPRM_BUF_SIZE);
281			if (unlikely(retval != BINPRM_BUF_SIZE)) {
282				if (retval >= 0)
283					retval = -ENOEXEC;
284				goto error;
285			}
286
287			interp_params.hdr = *((struct elfhdr *) bprm->buf);
288			break;
289
290		case PT_LOAD:
291#ifdef CONFIG_MMU
292			if (exec_params.load_addr == 0)
293				exec_params.load_addr = phdr->p_vaddr;
294#endif
295			break;
296		}
297
298	}
299
300	if (is_constdisp(&exec_params.hdr))
301		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
302
303	/* perform insanity checks on the interpreter */
304	if (interpreter_name) {
305		retval = -ELIBBAD;
306		if (!is_elf(&interp_params.hdr, interpreter))
307			goto error;
308
309		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
310
311		/* read the interpreter's program header table */
312		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
313		if (retval < 0)
314			goto error;
315	}
316
317	stack_size = exec_params.stack_size;
318	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
319		executable_stack = EXSTACK_ENABLE_X;
320	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
321		executable_stack = EXSTACK_DISABLE_X;
322	else
323		executable_stack = EXSTACK_DEFAULT;
324
325	if (stack_size == 0) {
326		stack_size = interp_params.stack_size;
327		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
328			executable_stack = EXSTACK_ENABLE_X;
329		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
330			executable_stack = EXSTACK_DISABLE_X;
331		else
332			executable_stack = EXSTACK_DEFAULT;
333	}
334
335	retval = -ENOEXEC;
336	if (stack_size == 0)
337		stack_size = 131072UL; /* same as exec.c's default commit */
338
339	if (is_constdisp(&interp_params.hdr))
340		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
341
342	/* flush all traces of the currently running executable */
343	retval = flush_old_exec(bprm);
344	if (retval)
345		goto error;
346
347	/* there's now no turning back... the old userspace image is dead,
348	 * defunct, deceased, etc.
349	 */
350	if (elf_check_fdpic(&exec_params.hdr))
351		set_personality(PER_LINUX_FDPIC);
352	else
353		set_personality(PER_LINUX);
354	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
355		current->personality |= READ_IMPLIES_EXEC;
356
357	setup_new_exec(bprm);
358
359	set_binfmt(&elf_fdpic_format);
360
361	current->mm->start_code = 0;
362	current->mm->end_code = 0;
363	current->mm->start_stack = 0;
364	current->mm->start_data = 0;
365	current->mm->end_data = 0;
366	current->mm->context.exec_fdpic_loadmap = 0;
367	current->mm->context.interp_fdpic_loadmap = 0;
368
369#ifdef CONFIG_MMU
370	elf_fdpic_arch_lay_out_mm(&exec_params,
371				  &interp_params,
372				  &current->mm->start_stack,
373				  &current->mm->start_brk);
374
375	retval = setup_arg_pages(bprm, current->mm->start_stack,
376				 executable_stack);
377	if (retval < 0)
378		goto error;
379#endif
380
381	/* load the executable and interpreter into memory */
382	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
383				    "executable");
384	if (retval < 0)
385		goto error;
386
387	if (interpreter_name) {
388		retval = elf_fdpic_map_file(&interp_params, interpreter,
389					    current->mm, "interpreter");
390		if (retval < 0) {
391			printk(KERN_ERR "Unable to load interpreter\n");
392			goto error;
393		}
394
395		allow_write_access(interpreter);
396		fput(interpreter);
397		interpreter = NULL;
398	}
399
400#ifdef CONFIG_MMU
401	if (!current->mm->start_brk)
402		current->mm->start_brk = current->mm->end_data;
403
404	current->mm->brk = current->mm->start_brk =
405		PAGE_ALIGN(current->mm->start_brk);
406
407#else
408	/* create a stack area and zero-size brk area */
409	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
410	if (stack_size < PAGE_SIZE * 2)
411		stack_size = PAGE_SIZE * 2;
412
413	stack_prot = PROT_READ | PROT_WRITE;
414	if (executable_stack == EXSTACK_ENABLE_X ||
415	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
416		stack_prot |= PROT_EXEC;
417
418	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
419					 MAP_PRIVATE | MAP_ANONYMOUS |
420					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
421					 0);
422
423	if (IS_ERR_VALUE(current->mm->start_brk)) {
424		retval = current->mm->start_brk;
425		current->mm->start_brk = 0;
426		goto error;
427	}
428
429	current->mm->brk = current->mm->start_brk;
430	current->mm->context.end_brk = current->mm->start_brk;
431	current->mm->start_stack = current->mm->start_brk + stack_size;
432#endif
433
434	install_exec_creds(bprm);
435	if (create_elf_fdpic_tables(bprm, current->mm,
436				    &exec_params, &interp_params) < 0)
437		goto error;
438
439	kdebug("- start_code  %lx", current->mm->start_code);
440	kdebug("- end_code    %lx", current->mm->end_code);
441	kdebug("- start_data  %lx", current->mm->start_data);
442	kdebug("- end_data    %lx", current->mm->end_data);
443	kdebug("- start_brk   %lx", current->mm->start_brk);
444	kdebug("- brk         %lx", current->mm->brk);
445	kdebug("- start_stack %lx", current->mm->start_stack);
446
447#ifdef ELF_FDPIC_PLAT_INIT
448	/*
449	 * The ABI may specify that certain registers be set up in special
450	 * ways (on i386 %edx is the address of a DT_FINI function, for
451	 * example.  This macro performs whatever initialization to
452	 * the regs structure is required.
453	 */
454	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
455	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
456			    dynaddr);
457#endif
458
459	/* everything is now ready... get the userspace context ready to roll */
460	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
461	start_thread(regs, entryaddr, current->mm->start_stack);
462
463	retval = 0;
464
465error:
466	if (interpreter) {
467		allow_write_access(interpreter);
468		fput(interpreter);
469	}
470	kfree(interpreter_name);
471	kfree(exec_params.phdrs);
472	kfree(exec_params.loadmap);
473	kfree(interp_params.phdrs);
474	kfree(interp_params.loadmap);
475	return retval;
476}
477
478/*****************************************************************************/
479
480#ifndef ELF_BASE_PLATFORM
481/*
482 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
483 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
484 * will be copied to the user stack in the same manner as AT_PLATFORM.
485 */
486#define ELF_BASE_PLATFORM NULL
487#endif
488
489/*
490 * present useful information to the program by shovelling it onto the new
491 * process's stack
492 */
493static int create_elf_fdpic_tables(struct linux_binprm *bprm,
494				   struct mm_struct *mm,
495				   struct elf_fdpic_params *exec_params,
496				   struct elf_fdpic_params *interp_params)
497{
498	const struct cred *cred = current_cred();
499	unsigned long sp, csp, nitems;
500	elf_caddr_t __user *argv, *envp;
501	size_t platform_len = 0, len;
502	char *k_platform, *k_base_platform;
503	char __user *u_platform, *u_base_platform, *p;
504	int loop;
505	int nr;	/* reset for each csp adjustment */
506
507#ifdef CONFIG_MMU
508	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
509	 * by the processes running on the same package. One thing we can do is
510	 * to shuffle the initial stack for them, so we give the architecture
511	 * an opportunity to do so here.
512	 */
513	sp = arch_align_stack(bprm->p);
514#else
515	sp = mm->start_stack;
516
517	/* stack the program arguments and environment */
518	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
519		return -EFAULT;
520#endif
521
522	/*
523	 * If this architecture has a platform capability string, copy it
524	 * to userspace.  In some cases (Sparc), this info is impossible
525	 * for userspace to get any other way, in others (i386) it is
526	 * merely difficult.
527	 */
528	k_platform = ELF_PLATFORM;
529	u_platform = NULL;
530
531	if (k_platform) {
532		platform_len = strlen(k_platform) + 1;
533		sp -= platform_len;
534		u_platform = (char __user *) sp;
535		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
536			return -EFAULT;
537	}
538
539	/*
540	 * If this architecture has a "base" platform capability
541	 * string, copy it to userspace.
542	 */
543	k_base_platform = ELF_BASE_PLATFORM;
544	u_base_platform = NULL;
545
546	if (k_base_platform) {
547		platform_len = strlen(k_base_platform) + 1;
548		sp -= platform_len;
549		u_base_platform = (char __user *) sp;
550		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
551			return -EFAULT;
552	}
553
554	sp &= ~7UL;
555
556	/* stack the load map(s) */
557	len = sizeof(struct elf32_fdpic_loadmap);
558	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
559	sp = (sp - len) & ~7UL;
560	exec_params->map_addr = sp;
561
562	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
563		return -EFAULT;
564
565	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
566
567	if (interp_params->loadmap) {
568		len = sizeof(struct elf32_fdpic_loadmap);
569		len += sizeof(struct elf32_fdpic_loadseg) *
570			interp_params->loadmap->nsegs;
571		sp = (sp - len) & ~7UL;
572		interp_params->map_addr = sp;
573
574		if (copy_to_user((void __user *) sp, interp_params->loadmap,
575				 len) != 0)
576			return -EFAULT;
577
578		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
579	}
580
581	/* force 16 byte _final_ alignment here for generality */
582#define DLINFO_ITEMS 15
583
584	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
585		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
586
587	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
588		nitems++;
589
590	csp = sp;
591	sp -= nitems * 2 * sizeof(unsigned long);
592	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
593	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
594	sp -= 1 * sizeof(unsigned long);		/* argc */
595
596	csp -= sp & 15UL;
597	sp -= sp & 15UL;
598
599	/* put the ELF interpreter info on the stack */
600#define NEW_AUX_ENT(id, val)						\
601	do {								\
602		struct { unsigned long _id, _val; } __user *ent;	\
603									\
604		ent = (void __user *) csp;				\
605		__put_user((id), &ent[nr]._id);				\
606		__put_user((val), &ent[nr]._val);			\
607		nr++;							\
608	} while (0)
609
610	nr = 0;
611	csp -= 2 * sizeof(unsigned long);
612	NEW_AUX_ENT(AT_NULL, 0);
613	if (k_platform) {
614		nr = 0;
615		csp -= 2 * sizeof(unsigned long);
616		NEW_AUX_ENT(AT_PLATFORM,
617			    (elf_addr_t) (unsigned long) u_platform);
618	}
619
620	if (k_base_platform) {
621		nr = 0;
622		csp -= 2 * sizeof(unsigned long);
623		NEW_AUX_ENT(AT_BASE_PLATFORM,
624			    (elf_addr_t) (unsigned long) u_base_platform);
625	}
626
627	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
628		nr = 0;
629		csp -= 2 * sizeof(unsigned long);
630		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
631	}
632
633	nr = 0;
634	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
635	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
636#ifdef ELF_HWCAP2
637	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
638#endif
639	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
640	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
641	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
642	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
643	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
644	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
645	NEW_AUX_ENT(AT_FLAGS,	0);
646	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
647	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
648	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
649	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
650	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
651	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
652	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
653
654#ifdef ARCH_DLINFO
655	nr = 0;
656	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
657
658	/* ARCH_DLINFO must come last so platform specific code can enforce
659	 * special alignment requirements on the AUXV if necessary (eg. PPC).
660	 */
661	ARCH_DLINFO;
662#endif
663#undef NEW_AUX_ENT
664
665	/* allocate room for argv[] and envv[] */
666	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
667	envp = (elf_caddr_t __user *) csp;
668	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
669	argv = (elf_caddr_t __user *) csp;
670
671	/* stack argc */
672	csp -= sizeof(unsigned long);
673	__put_user(bprm->argc, (unsigned long __user *) csp);
674
675	BUG_ON(csp != sp);
676
677	/* fill in the argv[] array */
678#ifdef CONFIG_MMU
679	current->mm->arg_start = bprm->p;
680#else
681	current->mm->arg_start = current->mm->start_stack -
682		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
683#endif
684
685	p = (char __user *) current->mm->arg_start;
686	for (loop = bprm->argc; loop > 0; loop--) {
687		__put_user((elf_caddr_t) p, argv++);
688		len = strnlen_user(p, MAX_ARG_STRLEN);
689		if (!len || len > MAX_ARG_STRLEN)
690			return -EINVAL;
691		p += len;
692	}
693	__put_user(NULL, argv);
694	current->mm->arg_end = (unsigned long) p;
695
696	/* fill in the envv[] array */
697	current->mm->env_start = (unsigned long) p;
698	for (loop = bprm->envc; loop > 0; loop--) {
699		__put_user((elf_caddr_t)(unsigned long) p, envp++);
700		len = strnlen_user(p, MAX_ARG_STRLEN);
701		if (!len || len > MAX_ARG_STRLEN)
702			return -EINVAL;
703		p += len;
704	}
705	__put_user(NULL, envp);
706	current->mm->env_end = (unsigned long) p;
707
708	mm->start_stack = (unsigned long) sp;
709	return 0;
710}
711
712/*****************************************************************************/
713/*
714 * transfer the program arguments and environment from the holding pages onto
715 * the stack
716 */
717#ifndef CONFIG_MMU
718static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
719					    unsigned long *_sp)
720{
721	unsigned long index, stop, sp;
722	char *src;
723	int ret = 0;
724
725	stop = bprm->p >> PAGE_SHIFT;
726	sp = *_sp;
727
728	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
729		src = kmap(bprm->page[index]);
730		sp -= PAGE_SIZE;
731		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
732			ret = -EFAULT;
733		kunmap(bprm->page[index]);
734		if (ret < 0)
735			goto out;
736	}
737
738	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
739
740out:
741	return ret;
742}
743#endif
744
745/*****************************************************************************/
746/*
747 * load the appropriate binary image (executable or interpreter) into memory
748 * - we assume no MMU is available
749 * - if no other PIC bits are set in params->hdr->e_flags
750 *   - we assume that the LOADable segments in the binary are independently relocatable
751 *   - we assume R/O executable segments are shareable
752 * - else
753 *   - we assume the loadable parts of the image to require fixed displacement
754 *   - the image is not shareable
755 */
756static int elf_fdpic_map_file(struct elf_fdpic_params *params,
757			      struct file *file,
758			      struct mm_struct *mm,
759			      const char *what)
760{
761	struct elf32_fdpic_loadmap *loadmap;
762#ifdef CONFIG_MMU
763	struct elf32_fdpic_loadseg *mseg;
764#endif
765	struct elf32_fdpic_loadseg *seg;
766	struct elf32_phdr *phdr;
767	unsigned long load_addr, stop;
768	unsigned nloads, tmp;
769	size_t size;
770	int loop, ret;
771
772	/* allocate a load map table */
773	nloads = 0;
774	for (loop = 0; loop < params->hdr.e_phnum; loop++)
775		if (params->phdrs[loop].p_type == PT_LOAD)
776			nloads++;
777
778	if (nloads == 0)
779		return -ELIBBAD;
780
781	size = sizeof(*loadmap) + nloads * sizeof(*seg);
782	loadmap = kzalloc(size, GFP_KERNEL);
783	if (!loadmap)
784		return -ENOMEM;
785
786	params->loadmap = loadmap;
787
788	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
789	loadmap->nsegs = nloads;
790
791	load_addr = params->load_addr;
792	seg = loadmap->segs;
793
794	/* map the requested LOADs into the memory space */
795	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
796	case ELF_FDPIC_FLAG_CONSTDISP:
797	case ELF_FDPIC_FLAG_CONTIGUOUS:
798#ifndef CONFIG_MMU
799		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
800		if (ret < 0)
801			return ret;
802		break;
803#endif
804	default:
805		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
806		if (ret < 0)
807			return ret;
808		break;
809	}
810
811	/* map the entry point */
812	if (params->hdr.e_entry) {
813		seg = loadmap->segs;
814		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
815			if (params->hdr.e_entry >= seg->p_vaddr &&
816			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
817				params->entry_addr =
818					(params->hdr.e_entry - seg->p_vaddr) +
819					seg->addr;
820				break;
821			}
822		}
823	}
824
825	/* determine where the program header table has wound up if mapped */
826	stop = params->hdr.e_phoff;
827	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
828	phdr = params->phdrs;
829
830	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
831		if (phdr->p_type != PT_LOAD)
832			continue;
833
834		if (phdr->p_offset > params->hdr.e_phoff ||
835		    phdr->p_offset + phdr->p_filesz < stop)
836			continue;
837
838		seg = loadmap->segs;
839		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
840			if (phdr->p_vaddr >= seg->p_vaddr &&
841			    phdr->p_vaddr + phdr->p_filesz <=
842			    seg->p_vaddr + seg->p_memsz) {
843				params->ph_addr =
844					(phdr->p_vaddr - seg->p_vaddr) +
845					seg->addr +
846					params->hdr.e_phoff - phdr->p_offset;
847				break;
848			}
849		}
850		break;
851	}
852
853	/* determine where the dynamic section has wound up if there is one */
854	phdr = params->phdrs;
855	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
856		if (phdr->p_type != PT_DYNAMIC)
857			continue;
858
859		seg = loadmap->segs;
860		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
861			if (phdr->p_vaddr >= seg->p_vaddr &&
862			    phdr->p_vaddr + phdr->p_memsz <=
863			    seg->p_vaddr + seg->p_memsz) {
864				params->dynamic_addr =
865					(phdr->p_vaddr - seg->p_vaddr) +
866					seg->addr;
867
868				/* check the dynamic section contains at least
869				 * one item, and that the last item is a NULL
870				 * entry */
871				if (phdr->p_memsz == 0 ||
872				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
873					goto dynamic_error;
874
875				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
876				if (((Elf32_Dyn *)
877				     params->dynamic_addr)[tmp - 1].d_tag != 0)
878					goto dynamic_error;
879				break;
880			}
881		}
882		break;
883	}
884
885	/* now elide adjacent segments in the load map on MMU linux
886	 * - on uClinux the holes between may actually be filled with system
887	 *   stuff or stuff from other processes
888	 */
889#ifdef CONFIG_MMU
890	nloads = loadmap->nsegs;
891	mseg = loadmap->segs;
892	seg = mseg + 1;
893	for (loop = 1; loop < nloads; loop++) {
894		/* see if we have a candidate for merging */
895		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
896			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
897			if (load_addr == (seg->addr & PAGE_MASK)) {
898				mseg->p_memsz +=
899					load_addr -
900					(mseg->addr + mseg->p_memsz);
901				mseg->p_memsz += seg->addr & ~PAGE_MASK;
902				mseg->p_memsz += seg->p_memsz;
903				loadmap->nsegs--;
904				continue;
905			}
906		}
907
908		mseg++;
909		if (mseg != seg)
910			*mseg = *seg;
911	}
912#endif
913
914	kdebug("Mapped Object [%s]:", what);
915	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
916	kdebug("- entry    : %lx", params->entry_addr);
917	kdebug("- PHDR[]   : %lx", params->ph_addr);
918	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
919	seg = loadmap->segs;
920	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
921		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
922		       loop,
923		       seg->addr, seg->addr + seg->p_memsz - 1,
924		       seg->p_vaddr, seg->p_memsz);
925
926	return 0;
927
928dynamic_error:
929	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
930	       what, file_inode(file)->i_ino);
931	return -ELIBBAD;
932}
933
934/*****************************************************************************/
935/*
936 * map a file with constant displacement under uClinux
937 */
938#ifndef CONFIG_MMU
939static int elf_fdpic_map_file_constdisp_on_uclinux(
940	struct elf_fdpic_params *params,
941	struct file *file,
942	struct mm_struct *mm)
943{
944	struct elf32_fdpic_loadseg *seg;
945	struct elf32_phdr *phdr;
946	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
947	int loop, ret;
948
949	load_addr = params->load_addr;
950	seg = params->loadmap->segs;
951
952	/* determine the bounds of the contiguous overall allocation we must
953	 * make */
954	phdr = params->phdrs;
955	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
956		if (params->phdrs[loop].p_type != PT_LOAD)
957			continue;
958
959		if (base > phdr->p_vaddr)
960			base = phdr->p_vaddr;
961		if (top < phdr->p_vaddr + phdr->p_memsz)
962			top = phdr->p_vaddr + phdr->p_memsz;
963	}
964
965	/* allocate one big anon block for everything */
966	mflags = MAP_PRIVATE;
967	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
968		mflags |= MAP_EXECUTABLE;
969
970	maddr = vm_mmap(NULL, load_addr, top - base,
971			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
972	if (IS_ERR_VALUE(maddr))
973		return (int) maddr;
974
975	if (load_addr != 0)
976		load_addr += PAGE_ALIGN(top - base);
977
978	/* and then load the file segments into it */
979	phdr = params->phdrs;
980	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
981		if (params->phdrs[loop].p_type != PT_LOAD)
982			continue;
983
984		seg->addr = maddr + (phdr->p_vaddr - base);
985		seg->p_vaddr = phdr->p_vaddr;
986		seg->p_memsz = phdr->p_memsz;
987
988		ret = read_code(file, seg->addr, phdr->p_offset,
989				       phdr->p_filesz);
990		if (ret < 0)
991			return ret;
992
993		/* map the ELF header address if in this segment */
994		if (phdr->p_offset == 0)
995			params->elfhdr_addr = seg->addr;
996
997		/* clear any space allocated but not loaded */
998		if (phdr->p_filesz < phdr->p_memsz) {
999			if (clear_user((void *) (seg->addr + phdr->p_filesz),
1000				       phdr->p_memsz - phdr->p_filesz))
1001				return -EFAULT;
1002		}
1003
1004		if (mm) {
1005			if (phdr->p_flags & PF_X) {
1006				if (!mm->start_code) {
1007					mm->start_code = seg->addr;
1008					mm->end_code = seg->addr +
1009						phdr->p_memsz;
1010				}
1011			} else if (!mm->start_data) {
1012				mm->start_data = seg->addr;
1013				mm->end_data = seg->addr + phdr->p_memsz;
1014			}
1015		}
1016
1017		seg++;
1018	}
1019
1020	return 0;
1021}
1022#endif
1023
1024/*****************************************************************************/
1025/*
1026 * map a binary by direct mmap() of the individual PT_LOAD segments
1027 */
1028static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1029					     struct file *file,
1030					     struct mm_struct *mm)
1031{
1032	struct elf32_fdpic_loadseg *seg;
1033	struct elf32_phdr *phdr;
1034	unsigned long load_addr, delta_vaddr;
1035	int loop, dvset;
1036
1037	load_addr = params->load_addr;
1038	delta_vaddr = 0;
1039	dvset = 0;
1040
1041	seg = params->loadmap->segs;
1042
1043	/* deal with each load segment separately */
1044	phdr = params->phdrs;
1045	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1046		unsigned long maddr, disp, excess, excess1;
1047		int prot = 0, flags;
1048
1049		if (phdr->p_type != PT_LOAD)
1050			continue;
1051
1052		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1053		       (unsigned long) phdr->p_vaddr,
1054		       (unsigned long) phdr->p_offset,
1055		       (unsigned long) phdr->p_filesz,
1056		       (unsigned long) phdr->p_memsz);
1057
1058		/* determine the mapping parameters */
1059		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1060		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1061		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1062
1063		flags = MAP_PRIVATE | MAP_DENYWRITE;
1064		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1065			flags |= MAP_EXECUTABLE;
1066
1067		maddr = 0;
1068
1069		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1070		case ELF_FDPIC_FLAG_INDEPENDENT:
1071			/* PT_LOADs are independently locatable */
1072			break;
1073
1074		case ELF_FDPIC_FLAG_HONOURVADDR:
1075			/* the specified virtual address must be honoured */
1076			maddr = phdr->p_vaddr;
1077			flags |= MAP_FIXED;
1078			break;
1079
1080		case ELF_FDPIC_FLAG_CONSTDISP:
1081			/* constant displacement
1082			 * - can be mapped anywhere, but must be mapped as a
1083			 *   unit
1084			 */
1085			if (!dvset) {
1086				maddr = load_addr;
1087				delta_vaddr = phdr->p_vaddr;
1088				dvset = 1;
1089			} else {
1090				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1091				flags |= MAP_FIXED;
1092			}
1093			break;
1094
1095		case ELF_FDPIC_FLAG_CONTIGUOUS:
1096			/* contiguity handled later */
1097			break;
1098
1099		default:
1100			BUG();
1101		}
1102
1103		maddr &= PAGE_MASK;
1104
1105		/* create the mapping */
1106		disp = phdr->p_vaddr & ~PAGE_MASK;
1107		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1108				phdr->p_offset - disp);
1109
1110		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1111		       loop, phdr->p_memsz + disp, prot, flags,
1112		       phdr->p_offset - disp, maddr);
1113
1114		if (IS_ERR_VALUE(maddr))
1115			return (int) maddr;
1116
1117		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1118		    ELF_FDPIC_FLAG_CONTIGUOUS)
1119			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1120
1121		seg->addr = maddr + disp;
1122		seg->p_vaddr = phdr->p_vaddr;
1123		seg->p_memsz = phdr->p_memsz;
1124
1125		/* map the ELF header address if in this segment */
1126		if (phdr->p_offset == 0)
1127			params->elfhdr_addr = seg->addr;
1128
1129		/* clear the bit between beginning of mapping and beginning of
1130		 * PT_LOAD */
1131		if (prot & PROT_WRITE && disp > 0) {
1132			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1133			if (clear_user((void __user *) maddr, disp))
1134				return -EFAULT;
1135			maddr += disp;
1136		}
1137
1138		/* clear any space allocated but not loaded
1139		 * - on uClinux we can just clear the lot
1140		 * - on MMU linux we'll get a SIGBUS beyond the last page
1141		 *   extant in the file
1142		 */
1143		excess = phdr->p_memsz - phdr->p_filesz;
1144		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1145
1146#ifdef CONFIG_MMU
1147		if (excess > excess1) {
1148			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1149			unsigned long xmaddr;
1150
1151			flags |= MAP_FIXED | MAP_ANONYMOUS;
1152			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1153					 prot, flags, 0);
1154
1155			kdebug("mmap[%d] <anon>"
1156			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1157			       loop, xaddr, excess - excess1, prot, flags,
1158			       xmaddr);
1159
1160			if (xmaddr != xaddr)
1161				return -ENOMEM;
1162		}
1163
1164		if (prot & PROT_WRITE && excess1 > 0) {
1165			kdebug("clear[%d] ad=%lx sz=%lx",
1166			       loop, maddr + phdr->p_filesz, excess1);
1167			if (clear_user((void __user *) maddr + phdr->p_filesz,
1168				       excess1))
1169				return -EFAULT;
1170		}
1171
1172#else
1173		if (excess > 0) {
1174			kdebug("clear[%d] ad=%lx sz=%lx",
1175			       loop, maddr + phdr->p_filesz, excess);
1176			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1177				return -EFAULT;
1178		}
1179#endif
1180
1181		if (mm) {
1182			if (phdr->p_flags & PF_X) {
1183				if (!mm->start_code) {
1184					mm->start_code = maddr;
1185					mm->end_code = maddr + phdr->p_memsz;
1186				}
1187			} else if (!mm->start_data) {
1188				mm->start_data = maddr;
1189				mm->end_data = maddr + phdr->p_memsz;
1190			}
1191		}
1192
1193		seg++;
1194	}
1195
1196	return 0;
1197}
1198
1199/*****************************************************************************/
1200/*
1201 * ELF-FDPIC core dumper
1202 *
1203 * Modelled on fs/exec.c:aout_core_dump()
1204 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1205 *
1206 * Modelled on fs/binfmt_elf.c core dumper
1207 */
1208#ifdef CONFIG_ELF_CORE
1209
1210/*
1211 * Decide whether a segment is worth dumping; default is yes to be
1212 * sure (missing info is worse than too much; etc).
1213 * Personally I'd include everything, and use the coredump limit...
1214 *
1215 * I think we should skip something. But I am not sure how. H.J.
1216 */
1217static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1218{
1219	int dump_ok;
1220
1221	/* Do not dump I/O mapped devices or special mappings */
1222	if (vma->vm_flags & VM_IO) {
1223		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1224		return 0;
1225	}
1226
1227	/* If we may not read the contents, don't allow us to dump
1228	 * them either. "dump_write()" can't handle it anyway.
1229	 */
1230	if (!(vma->vm_flags & VM_READ)) {
1231		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1232		return 0;
1233	}
1234
1235	/* support for DAX */
1236	if (vma_is_dax(vma)) {
1237		if (vma->vm_flags & VM_SHARED) {
1238			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1239			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1240			       vma->vm_flags, dump_ok ? "yes" : "no");
1241		} else {
1242			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1243			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1244			       vma->vm_flags, dump_ok ? "yes" : "no");
1245		}
1246		return dump_ok;
1247	}
1248
1249	/* By default, dump shared memory if mapped from an anonymous file. */
1250	if (vma->vm_flags & VM_SHARED) {
1251		if (file_inode(vma->vm_file)->i_nlink == 0) {
1252			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1253			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1254			       vma->vm_flags, dump_ok ? "yes" : "no");
1255			return dump_ok;
1256		}
1257
1258		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1259		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1260		       vma->vm_flags, dump_ok ? "yes" : "no");
1261		return dump_ok;
1262	}
1263
1264#ifdef CONFIG_MMU
1265	/* By default, if it hasn't been written to, don't write it out */
1266	if (!vma->anon_vma) {
1267		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1268		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1269		       vma->vm_flags, dump_ok ? "yes" : "no");
1270		return dump_ok;
1271	}
1272#endif
1273
1274	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1275	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1276	       dump_ok ? "yes" : "no");
1277	return dump_ok;
1278}
1279
1280/* An ELF note in memory */
1281struct memelfnote
1282{
1283	const char *name;
1284	int type;
1285	unsigned int datasz;
1286	void *data;
1287};
1288
1289static int notesize(struct memelfnote *en)
1290{
1291	int sz;
1292
1293	sz = sizeof(struct elf_note);
1294	sz += roundup(strlen(en->name) + 1, 4);
1295	sz += roundup(en->datasz, 4);
1296
1297	return sz;
1298}
1299
1300/* #define DEBUG */
1301
1302static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1303{
1304	struct elf_note en;
1305	en.n_namesz = strlen(men->name) + 1;
1306	en.n_descsz = men->datasz;
1307	en.n_type = men->type;
1308
1309	return dump_emit(cprm, &en, sizeof(en)) &&
1310		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1311		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1312}
1313
1314static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1315{
1316	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1317	elf->e_ident[EI_CLASS] = ELF_CLASS;
1318	elf->e_ident[EI_DATA] = ELF_DATA;
1319	elf->e_ident[EI_VERSION] = EV_CURRENT;
1320	elf->e_ident[EI_OSABI] = ELF_OSABI;
1321	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1322
1323	elf->e_type = ET_CORE;
1324	elf->e_machine = ELF_ARCH;
1325	elf->e_version = EV_CURRENT;
1326	elf->e_entry = 0;
1327	elf->e_phoff = sizeof(struct elfhdr);
1328	elf->e_shoff = 0;
1329	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1330	elf->e_ehsize = sizeof(struct elfhdr);
1331	elf->e_phentsize = sizeof(struct elf_phdr);
1332	elf->e_phnum = segs;
1333	elf->e_shentsize = 0;
1334	elf->e_shnum = 0;
1335	elf->e_shstrndx = 0;
1336	return;
1337}
1338
1339static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1340{
1341	phdr->p_type = PT_NOTE;
1342	phdr->p_offset = offset;
1343	phdr->p_vaddr = 0;
1344	phdr->p_paddr = 0;
1345	phdr->p_filesz = sz;
1346	phdr->p_memsz = 0;
1347	phdr->p_flags = 0;
1348	phdr->p_align = 0;
1349	return;
1350}
1351
1352static inline void fill_note(struct memelfnote *note, const char *name, int type,
1353		unsigned int sz, void *data)
1354{
1355	note->name = name;
1356	note->type = type;
1357	note->datasz = sz;
1358	note->data = data;
1359	return;
1360}
1361
1362/*
1363 * fill up all the fields in prstatus from the given task struct, except
1364 * registers which need to be filled up separately.
1365 */
1366static void fill_prstatus(struct elf_prstatus *prstatus,
1367			  struct task_struct *p, long signr)
1368{
1369	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1370	prstatus->pr_sigpend = p->pending.signal.sig[0];
1371	prstatus->pr_sighold = p->blocked.sig[0];
1372	rcu_read_lock();
1373	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1374	rcu_read_unlock();
1375	prstatus->pr_pid = task_pid_vnr(p);
1376	prstatus->pr_pgrp = task_pgrp_vnr(p);
1377	prstatus->pr_sid = task_session_vnr(p);
1378	if (thread_group_leader(p)) {
1379		struct task_cputime cputime;
1380
1381		/*
1382		 * This is the record for the group leader.  It shows the
1383		 * group-wide total, not its individual thread total.
1384		 */
1385		thread_group_cputime(p, &cputime);
1386		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1387		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1388	} else {
1389		cputime_t utime, stime;
1390
1391		task_cputime(p, &utime, &stime);
1392		cputime_to_timeval(utime, &prstatus->pr_utime);
1393		cputime_to_timeval(stime, &prstatus->pr_stime);
1394	}
1395	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1396	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1397
1398	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1399	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1400}
1401
1402static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1403		       struct mm_struct *mm)
1404{
1405	const struct cred *cred;
1406	unsigned int i, len;
1407
1408	/* first copy the parameters from user space */
1409	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1410
1411	len = mm->arg_end - mm->arg_start;
1412	if (len >= ELF_PRARGSZ)
1413		len = ELF_PRARGSZ - 1;
1414	if (copy_from_user(&psinfo->pr_psargs,
1415		           (const char __user *) mm->arg_start, len))
1416		return -EFAULT;
1417	for (i = 0; i < len; i++)
1418		if (psinfo->pr_psargs[i] == 0)
1419			psinfo->pr_psargs[i] = ' ';
1420	psinfo->pr_psargs[len] = 0;
1421
1422	rcu_read_lock();
1423	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1424	rcu_read_unlock();
1425	psinfo->pr_pid = task_pid_vnr(p);
1426	psinfo->pr_pgrp = task_pgrp_vnr(p);
1427	psinfo->pr_sid = task_session_vnr(p);
1428
1429	i = p->state ? ffz(~p->state) + 1 : 0;
1430	psinfo->pr_state = i;
1431	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1432	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1433	psinfo->pr_nice = task_nice(p);
1434	psinfo->pr_flag = p->flags;
1435	rcu_read_lock();
1436	cred = __task_cred(p);
1437	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1438	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1439	rcu_read_unlock();
1440	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1441
1442	return 0;
1443}
1444
1445/* Here is the structure in which status of each thread is captured. */
1446struct elf_thread_status
1447{
1448	struct list_head list;
1449	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1450	elf_fpregset_t fpu;		/* NT_PRFPREG */
1451	struct task_struct *thread;
1452#ifdef ELF_CORE_COPY_XFPREGS
1453	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1454#endif
1455	struct memelfnote notes[3];
1456	int num_notes;
1457};
1458
1459/*
1460 * In order to add the specific thread information for the elf file format,
1461 * we need to keep a linked list of every thread's pr_status and then create
1462 * a single section for them in the final core file.
1463 */
1464static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1465{
1466	struct task_struct *p = t->thread;
1467	int sz = 0;
1468
1469	t->num_notes = 0;
1470
1471	fill_prstatus(&t->prstatus, p, signr);
1472	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1473
1474	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1475		  &t->prstatus);
1476	t->num_notes++;
1477	sz += notesize(&t->notes[0]);
1478
1479	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1480	if (t->prstatus.pr_fpvalid) {
1481		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1482			  &t->fpu);
1483		t->num_notes++;
1484		sz += notesize(&t->notes[1]);
1485	}
1486
1487#ifdef ELF_CORE_COPY_XFPREGS
1488	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1489		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1490			  sizeof(t->xfpu), &t->xfpu);
1491		t->num_notes++;
1492		sz += notesize(&t->notes[2]);
1493	}
1494#endif
1495	return sz;
1496}
1497
1498static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1499			     elf_addr_t e_shoff, int segs)
1500{
1501	elf->e_shoff = e_shoff;
1502	elf->e_shentsize = sizeof(*shdr4extnum);
1503	elf->e_shnum = 1;
1504	elf->e_shstrndx = SHN_UNDEF;
1505
1506	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1507
1508	shdr4extnum->sh_type = SHT_NULL;
1509	shdr4extnum->sh_size = elf->e_shnum;
1510	shdr4extnum->sh_link = elf->e_shstrndx;
1511	shdr4extnum->sh_info = segs;
1512}
1513
1514/*
1515 * dump the segments for an MMU process
1516 */
1517static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1518{
1519	struct vm_area_struct *vma;
1520
1521	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1522		unsigned long addr;
1523
1524		if (!maydump(vma, cprm->mm_flags))
1525			continue;
1526
1527#ifdef CONFIG_MMU
1528		for (addr = vma->vm_start; addr < vma->vm_end;
1529							addr += PAGE_SIZE) {
1530			bool res;
1531			struct page *page = get_dump_page(addr);
1532			if (page) {
1533				void *kaddr = kmap(page);
1534				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1535				kunmap(page);
1536				page_cache_release(page);
1537			} else {
1538				res = dump_skip(cprm, PAGE_SIZE);
1539			}
1540			if (!res)
1541				return false;
1542		}
1543#else
1544		if (!dump_emit(cprm, (void *) vma->vm_start,
1545				vma->vm_end - vma->vm_start))
1546			return false;
1547#endif
1548	}
1549	return true;
1550}
1551
1552static size_t elf_core_vma_data_size(unsigned long mm_flags)
1553{
1554	struct vm_area_struct *vma;
1555	size_t size = 0;
1556
1557	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1558		if (maydump(vma, mm_flags))
1559			size += vma->vm_end - vma->vm_start;
1560	return size;
1561}
1562
1563/*
1564 * Actual dumper
1565 *
1566 * This is a two-pass process; first we find the offsets of the bits,
1567 * and then they are actually written out.  If we run out of core limit
1568 * we just truncate.
1569 */
1570static int elf_fdpic_core_dump(struct coredump_params *cprm)
1571{
1572#define	NUM_NOTES	6
1573	int has_dumped = 0;
1574	mm_segment_t fs;
1575	int segs;
1576	int i;
1577	struct vm_area_struct *vma;
1578	struct elfhdr *elf = NULL;
1579	loff_t offset = 0, dataoff;
1580	int numnote;
1581	struct memelfnote *notes = NULL;
1582	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1583	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1584 	LIST_HEAD(thread_list);
1585 	struct list_head *t;
1586	elf_fpregset_t *fpu = NULL;
1587#ifdef ELF_CORE_COPY_XFPREGS
1588	elf_fpxregset_t *xfpu = NULL;
1589#endif
1590	int thread_status_size = 0;
1591	elf_addr_t *auxv;
1592	struct elf_phdr *phdr4note = NULL;
1593	struct elf_shdr *shdr4extnum = NULL;
1594	Elf_Half e_phnum;
1595	elf_addr_t e_shoff;
1596	struct core_thread *ct;
1597	struct elf_thread_status *tmp;
1598
1599	/*
1600	 * We no longer stop all VM operations.
1601	 *
1602	 * This is because those proceses that could possibly change map_count
1603	 * or the mmap / vma pages are now blocked in do_exit on current
1604	 * finishing this core dump.
1605	 *
1606	 * Only ptrace can touch these memory addresses, but it doesn't change
1607	 * the map_count or the pages allocated. So no possibility of crashing
1608	 * exists while dumping the mm->vm_next areas to the core file.
1609	 */
1610
1611	/* alloc memory for large data structures: too large to be on stack */
1612	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1613	if (!elf)
1614		goto cleanup;
1615	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1616	if (!prstatus)
1617		goto cleanup;
1618	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1619	if (!psinfo)
1620		goto cleanup;
1621	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1622	if (!notes)
1623		goto cleanup;
1624	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1625	if (!fpu)
1626		goto cleanup;
1627#ifdef ELF_CORE_COPY_XFPREGS
1628	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1629	if (!xfpu)
1630		goto cleanup;
1631#endif
1632
1633	for (ct = current->mm->core_state->dumper.next;
1634					ct; ct = ct->next) {
1635		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1636		if (!tmp)
1637			goto cleanup;
1638
1639		tmp->thread = ct->task;
1640		list_add(&tmp->list, &thread_list);
1641	}
1642
1643	list_for_each(t, &thread_list) {
1644		struct elf_thread_status *tmp;
1645		int sz;
1646
1647		tmp = list_entry(t, struct elf_thread_status, list);
1648		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1649		thread_status_size += sz;
1650	}
1651
1652	/* now collect the dump for the current */
1653	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1654	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1655
1656	segs = current->mm->map_count;
1657	segs += elf_core_extra_phdrs();
1658
1659	/* for notes section */
1660	segs++;
1661
1662	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1663	 * this, kernel supports extended numbering. Have a look at
1664	 * include/linux/elf.h for further information. */
1665	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1666
1667	/* Set up header */
1668	fill_elf_fdpic_header(elf, e_phnum);
1669
1670	has_dumped = 1;
1671	/*
1672	 * Set up the notes in similar form to SVR4 core dumps made
1673	 * with info from their /proc.
1674	 */
1675
1676	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1677	fill_psinfo(psinfo, current->group_leader, current->mm);
1678	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1679
1680	numnote = 2;
1681
1682	auxv = (elf_addr_t *) current->mm->saved_auxv;
1683
1684	i = 0;
1685	do
1686		i += 2;
1687	while (auxv[i - 2] != AT_NULL);
1688	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1689		  i * sizeof(elf_addr_t), auxv);
1690
1691  	/* Try to dump the FPU. */
1692	if ((prstatus->pr_fpvalid =
1693	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1694		fill_note(notes + numnote++,
1695			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1696#ifdef ELF_CORE_COPY_XFPREGS
1697	if (elf_core_copy_task_xfpregs(current, xfpu))
1698		fill_note(notes + numnote++,
1699			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1700#endif
1701
1702	fs = get_fs();
1703	set_fs(KERNEL_DS);
1704
1705	offset += sizeof(*elf);				/* Elf header */
1706	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1707
1708	/* Write notes phdr entry */
1709	{
1710		int sz = 0;
1711
1712		for (i = 0; i < numnote; i++)
1713			sz += notesize(notes + i);
1714
1715		sz += thread_status_size;
1716
1717		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1718		if (!phdr4note)
1719			goto end_coredump;
1720
1721		fill_elf_note_phdr(phdr4note, sz, offset);
1722		offset += sz;
1723	}
1724
1725	/* Page-align dumped data */
1726	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1727
1728	offset += elf_core_vma_data_size(cprm->mm_flags);
1729	offset += elf_core_extra_data_size();
1730	e_shoff = offset;
1731
1732	if (e_phnum == PN_XNUM) {
1733		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1734		if (!shdr4extnum)
1735			goto end_coredump;
1736		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1737	}
1738
1739	offset = dataoff;
1740
1741	if (!dump_emit(cprm, elf, sizeof(*elf)))
1742		goto end_coredump;
1743
1744	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1745		goto end_coredump;
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, cprm->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		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1769			goto end_coredump;
1770	}
1771
1772	if (!elf_core_write_extra_phdrs(cprm, offset))
1773		goto end_coredump;
1774
1775 	/* write out the notes section */
1776	for (i = 0; i < numnote; i++)
1777		if (!writenote(notes + i, cprm))
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))
1787				goto end_coredump;
1788	}
1789
1790	if (!dump_skip(cprm, dataoff - cprm->written))
1791		goto end_coredump;
1792
1793	if (!elf_fdpic_dump_segments(cprm))
1794		goto end_coredump;
1795
1796	if (!elf_core_write_extra_data(cprm))
1797		goto end_coredump;
1798
1799	if (e_phnum == PN_XNUM) {
1800		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1801			goto end_coredump;
1802	}
1803
1804	if (cprm->file->f_pos != offset) {
1805		/* Sanity check */
1806		printk(KERN_WARNING
1807		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1808		       cprm->file->f_pos, offset);
1809	}
1810
1811end_coredump:
1812	set_fs(fs);
1813
1814cleanup:
1815	while (!list_empty(&thread_list)) {
1816		struct list_head *tmp = thread_list.next;
1817		list_del(tmp);
1818		kfree(list_entry(tmp, struct elf_thread_status, list));
1819	}
1820	kfree(phdr4note);
1821	kfree(elf);
1822	kfree(prstatus);
1823	kfree(psinfo);
1824	kfree(notes);
1825	kfree(fpu);
1826	kfree(shdr4extnum);
1827#ifdef ELF_CORE_COPY_XFPREGS
1828	kfree(xfpu);
1829#endif
1830	return has_dumped;
1831#undef NUM_NOTES
1832}
1833
1834#endif		/* CONFIG_ELF_CORE */
1835