xref: /linux/fs/binfmt_elf_fdpic.c (revision 606d099cdd1080bbb50ea50dc52d98252f8f10a1)
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2  *
3  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  * Derived from binfmt_elf.c
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version
10  * 2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/module.h>
14 
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/smp_lock.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37 
38 #include <asm/uaccess.h>
39 #include <asm/param.h>
40 #include <asm/pgalloc.h>
41 
42 typedef char *elf_caddr_t;
43 
44 #if 0
45 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
46 #else
47 #define kdebug(fmt, ...) do {} while(0)
48 #endif
49 
50 #if 0
51 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
52 #else
53 #define kdcore(fmt, ...) do {} while(0)
54 #endif
55 
56 MODULE_LICENSE("GPL");
57 
58 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
59 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
60 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
61 			      struct mm_struct *, const char *);
62 
63 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
64 				   struct elf_fdpic_params *,
65 				   struct elf_fdpic_params *);
66 
67 #ifndef CONFIG_MMU
68 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
69 					    unsigned long *);
70 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
71 						   struct file *,
72 						   struct mm_struct *);
73 #endif
74 
75 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
76 					     struct file *, struct mm_struct *);
77 
78 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
79 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *);
80 #endif
81 
82 static struct linux_binfmt elf_fdpic_format = {
83 	.module		= THIS_MODULE,
84 	.load_binary	= load_elf_fdpic_binary,
85 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
86 	.core_dump	= elf_fdpic_core_dump,
87 #endif
88 	.min_coredump	= ELF_EXEC_PAGESIZE,
89 };
90 
91 static int __init init_elf_fdpic_binfmt(void)
92 {
93 	return register_binfmt(&elf_fdpic_format);
94 }
95 
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 	unregister_binfmt(&elf_fdpic_format);
99 }
100 
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103 
104 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
105 {
106 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 		return 0;
108 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 		return 0;
110 	if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
111 		return 0;
112 	if (!file->f_op || !file->f_op->mmap)
113 		return 0;
114 	return 1;
115 }
116 
117 /*****************************************************************************/
118 /*
119  * read the program headers table into memory
120  */
121 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
122 				 struct file *file)
123 {
124 	struct elf32_phdr *phdr;
125 	unsigned long size;
126 	int retval, loop;
127 
128 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
129 		return -ENOMEM;
130 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
131 		return -ENOMEM;
132 
133 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
134 	params->phdrs = kmalloc(size, GFP_KERNEL);
135 	if (!params->phdrs)
136 		return -ENOMEM;
137 
138 	retval = kernel_read(file, params->hdr.e_phoff,
139 			     (char *) params->phdrs, size);
140 	if (retval < 0)
141 		return retval;
142 
143 	/* determine stack size for this binary */
144 	phdr = params->phdrs;
145 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
146 		if (phdr->p_type != PT_GNU_STACK)
147 			continue;
148 
149 		if (phdr->p_flags & PF_X)
150 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
151 		else
152 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
153 
154 		params->stack_size = phdr->p_memsz;
155 		break;
156 	}
157 
158 	return 0;
159 }
160 
161 /*****************************************************************************/
162 /*
163  * load an fdpic binary into various bits of memory
164  */
165 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
166 				 struct pt_regs *regs)
167 {
168 	struct elf_fdpic_params exec_params, interp_params;
169 	struct elf_phdr *phdr;
170 	unsigned long stack_size, entryaddr;
171 #ifndef CONFIG_MMU
172 	unsigned long fullsize;
173 #endif
174 #ifdef ELF_FDPIC_PLAT_INIT
175 	unsigned long dynaddr;
176 #endif
177 	struct file *interpreter = NULL; /* to shut gcc up */
178 	char *interpreter_name = NULL;
179 	int executable_stack;
180 	int retval, i;
181 
182 	memset(&exec_params, 0, sizeof(exec_params));
183 	memset(&interp_params, 0, sizeof(interp_params));
184 
185 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
186 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
187 
188 	/* check that this is a binary we know how to deal with */
189 	retval = -ENOEXEC;
190 	if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
191 		goto error;
192 
193 	/* read the program header table */
194 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
195 	if (retval < 0)
196 		goto error;
197 
198 	/* scan for a program header that specifies an interpreter */
199 	phdr = exec_params.phdrs;
200 
201 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
202 		switch (phdr->p_type) {
203 		case PT_INTERP:
204 			retval = -ENOMEM;
205 			if (phdr->p_filesz > PATH_MAX)
206 				goto error;
207 			retval = -ENOENT;
208 			if (phdr->p_filesz < 2)
209 				goto error;
210 
211 			/* read the name of the interpreter into memory */
212 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
213 			if (!interpreter_name)
214 				goto error;
215 
216 			retval = kernel_read(bprm->file,
217 					     phdr->p_offset,
218 					     interpreter_name,
219 					     phdr->p_filesz);
220 			if (retval < 0)
221 				goto error;
222 
223 			retval = -ENOENT;
224 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
225 				goto error;
226 
227 			kdebug("Using ELF interpreter %s", interpreter_name);
228 
229 			/* replace the program with the interpreter */
230 			interpreter = open_exec(interpreter_name);
231 			retval = PTR_ERR(interpreter);
232 			if (IS_ERR(interpreter)) {
233 				interpreter = NULL;
234 				goto error;
235 			}
236 
237 			retval = kernel_read(interpreter, 0, bprm->buf,
238 					     BINPRM_BUF_SIZE);
239 			if (retval < 0)
240 				goto error;
241 
242 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
243 			break;
244 
245 		case PT_LOAD:
246 #ifdef CONFIG_MMU
247 			if (exec_params.load_addr == 0)
248 				exec_params.load_addr = phdr->p_vaddr;
249 #endif
250 			break;
251 		}
252 
253 	}
254 
255 	if (elf_check_const_displacement(&exec_params.hdr))
256 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
257 
258 	/* perform insanity checks on the interpreter */
259 	if (interpreter_name) {
260 		retval = -ELIBBAD;
261 		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
262 			goto error;
263 
264 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
265 
266 		/* read the interpreter's program header table */
267 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
268 		if (retval < 0)
269 			goto error;
270 	}
271 
272 	stack_size = exec_params.stack_size;
273 	if (stack_size < interp_params.stack_size)
274 		stack_size = interp_params.stack_size;
275 
276 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
277 		executable_stack = EXSTACK_ENABLE_X;
278 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
279 		executable_stack = EXSTACK_DISABLE_X;
280 	else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
281 		executable_stack = EXSTACK_ENABLE_X;
282 	else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
283 		executable_stack = EXSTACK_DISABLE_X;
284 	else
285 		executable_stack = EXSTACK_DEFAULT;
286 
287 	retval = -ENOEXEC;
288 	if (stack_size == 0)
289 		goto error;
290 
291 	if (elf_check_const_displacement(&interp_params.hdr))
292 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
293 
294 	/* flush all traces of the currently running executable */
295 	retval = flush_old_exec(bprm);
296 	if (retval)
297 		goto error;
298 
299 	/* there's now no turning back... the old userspace image is dead,
300 	 * defunct, deceased, etc. after this point we have to exit via
301 	 * error_kill */
302 	set_personality(PER_LINUX_FDPIC);
303 	set_binfmt(&elf_fdpic_format);
304 
305 	current->mm->start_code = 0;
306 	current->mm->end_code = 0;
307 	current->mm->start_stack = 0;
308 	current->mm->start_data = 0;
309 	current->mm->end_data = 0;
310 	current->mm->context.exec_fdpic_loadmap = 0;
311 	current->mm->context.interp_fdpic_loadmap = 0;
312 
313 	current->flags &= ~PF_FORKNOEXEC;
314 
315 #ifdef CONFIG_MMU
316 	elf_fdpic_arch_lay_out_mm(&exec_params,
317 				  &interp_params,
318 				  &current->mm->start_stack,
319 				  &current->mm->start_brk);
320 
321 	retval = setup_arg_pages(bprm, current->mm->start_stack,
322 				 executable_stack);
323 	if (retval < 0) {
324 		send_sig(SIGKILL, current, 0);
325 		goto error_kill;
326 	}
327 #endif
328 
329 	/* load the executable and interpreter into memory */
330 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
331 				    "executable");
332 	if (retval < 0)
333 		goto error_kill;
334 
335 	if (interpreter_name) {
336 		retval = elf_fdpic_map_file(&interp_params, interpreter,
337 					    current->mm, "interpreter");
338 		if (retval < 0) {
339 			printk(KERN_ERR "Unable to load interpreter\n");
340 			goto error_kill;
341 		}
342 
343 		allow_write_access(interpreter);
344 		fput(interpreter);
345 		interpreter = NULL;
346 	}
347 
348 #ifdef CONFIG_MMU
349 	if (!current->mm->start_brk)
350 		current->mm->start_brk = current->mm->end_data;
351 
352 	current->mm->brk = current->mm->start_brk =
353 		PAGE_ALIGN(current->mm->start_brk);
354 
355 #else
356 	/* create a stack and brk area big enough for everyone
357 	 * - the brk heap starts at the bottom and works up
358 	 * - the stack starts at the top and works down
359 	 */
360 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
361 	if (stack_size < PAGE_SIZE * 2)
362 		stack_size = PAGE_SIZE * 2;
363 
364 	down_write(&current->mm->mmap_sem);
365 	current->mm->start_brk = do_mmap(NULL, 0, stack_size,
366 					 PROT_READ | PROT_WRITE | PROT_EXEC,
367 					 MAP_PRIVATE | MAP_ANON | MAP_GROWSDOWN,
368 					 0);
369 
370 	if (IS_ERR_VALUE(current->mm->start_brk)) {
371 		up_write(&current->mm->mmap_sem);
372 		retval = current->mm->start_brk;
373 		current->mm->start_brk = 0;
374 		goto error_kill;
375 	}
376 
377 	/* expand the stack mapping to use up the entire allocation granule */
378 	fullsize = ksize((char *) current->mm->start_brk);
379 	if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
380 				    fullsize, 0, 0)))
381 		stack_size = fullsize;
382 	up_write(&current->mm->mmap_sem);
383 
384 	current->mm->brk = current->mm->start_brk;
385 	current->mm->context.end_brk = current->mm->start_brk;
386 	current->mm->context.end_brk +=
387 		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
388 	current->mm->start_stack = current->mm->start_brk + stack_size;
389 #endif
390 
391 	compute_creds(bprm);
392 	current->flags &= ~PF_FORKNOEXEC;
393 	if (create_elf_fdpic_tables(bprm, current->mm,
394 				    &exec_params, &interp_params) < 0)
395 		goto error_kill;
396 
397 	kdebug("- start_code  %lx", current->mm->start_code);
398 	kdebug("- end_code    %lx", current->mm->end_code);
399 	kdebug("- start_data  %lx", current->mm->start_data);
400 	kdebug("- end_data    %lx", current->mm->end_data);
401 	kdebug("- start_brk   %lx", current->mm->start_brk);
402 	kdebug("- brk         %lx", current->mm->brk);
403 	kdebug("- start_stack %lx", current->mm->start_stack);
404 
405 #ifdef ELF_FDPIC_PLAT_INIT
406 	/*
407 	 * The ABI may specify that certain registers be set up in special
408 	 * ways (on i386 %edx is the address of a DT_FINI function, for
409 	 * example.  This macro performs whatever initialization to
410 	 * the regs structure is required.
411 	 */
412 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
413 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
414 			    dynaddr);
415 #endif
416 
417 	/* everything is now ready... get the userspace context ready to roll */
418 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
419 	start_thread(regs, entryaddr, current->mm->start_stack);
420 
421 	if (unlikely(current->ptrace & PT_PTRACED)) {
422 		if (current->ptrace & PT_TRACE_EXEC)
423 			ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
424 		else
425 			send_sig(SIGTRAP, current, 0);
426 	}
427 
428 	retval = 0;
429 
430 error:
431 	if (interpreter) {
432 		allow_write_access(interpreter);
433 		fput(interpreter);
434 	}
435 	kfree(interpreter_name);
436 	kfree(exec_params.phdrs);
437 	kfree(exec_params.loadmap);
438 	kfree(interp_params.phdrs);
439 	kfree(interp_params.loadmap);
440 	return retval;
441 
442 	/* unrecoverable error - kill the process */
443 error_kill:
444 	send_sig(SIGSEGV, current, 0);
445 	goto error;
446 
447 }
448 
449 /*****************************************************************************/
450 /*
451  * present useful information to the program
452  */
453 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
454 				   struct mm_struct *mm,
455 				   struct elf_fdpic_params *exec_params,
456 				   struct elf_fdpic_params *interp_params)
457 {
458 	unsigned long sp, csp, nitems;
459 	elf_caddr_t __user *argv, *envp;
460 	size_t platform_len = 0, len;
461 	char *k_platform;
462 	char __user *u_platform, *p;
463 	long hwcap;
464 	int loop;
465 
466 	/* we're going to shovel a whole load of stuff onto the stack */
467 #ifdef CONFIG_MMU
468 	sp = bprm->p;
469 #else
470 	sp = mm->start_stack;
471 
472 	/* stack the program arguments and environment */
473 	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
474 		return -EFAULT;
475 #endif
476 
477 	/* get hold of platform and hardware capabilities masks for the machine
478 	 * we are running on.  In some cases (Sparc), this info is impossible
479 	 * to get, in others (i386) it is merely difficult.
480 	 */
481 	hwcap = ELF_HWCAP;
482 	k_platform = ELF_PLATFORM;
483 	u_platform = NULL;
484 
485 	if (k_platform) {
486 		platform_len = strlen(k_platform) + 1;
487 		sp -= platform_len;
488 		u_platform = (char __user *) sp;
489 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
490 			return -EFAULT;
491 	}
492 
493 #if defined(__i386__) && defined(CONFIG_SMP)
494 	/* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
495 	 * by the processes running on the same package. One thing we can do is
496 	 * to shuffle the initial stack for them.
497 	 *
498 	 * the conditionals here are unneeded, but kept in to make the code
499 	 * behaviour the same as pre change unless we have hyperthreaded
500 	 * processors. This keeps Mr Marcelo Person happier but should be
501 	 * removed for 2.5
502 	 */
503 	if (smp_num_siblings > 1)
504 		sp = sp - ((current->pid % 64) << 7);
505 #endif
506 
507 	sp &= ~7UL;
508 
509 	/* stack the load map(s) */
510 	len = sizeof(struct elf32_fdpic_loadmap);
511 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
512 	sp = (sp - len) & ~7UL;
513 	exec_params->map_addr = sp;
514 
515 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
516 		return -EFAULT;
517 
518 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
519 
520 	if (interp_params->loadmap) {
521 		len = sizeof(struct elf32_fdpic_loadmap);
522 		len += sizeof(struct elf32_fdpic_loadseg) *
523 			interp_params->loadmap->nsegs;
524 		sp = (sp - len) & ~7UL;
525 		interp_params->map_addr = sp;
526 
527 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
528 				 len) != 0)
529 			return -EFAULT;
530 
531 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
532 	}
533 
534 	/* force 16 byte _final_ alignment here for generality */
535 #define DLINFO_ITEMS 13
536 
537 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0);
538 #ifdef DLINFO_ARCH_ITEMS
539 	nitems += DLINFO_ARCH_ITEMS;
540 #endif
541 
542 	csp = sp;
543 	sp -= nitems * 2 * sizeof(unsigned long);
544 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
545 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
546 	sp -= 1 * sizeof(unsigned long);		/* argc */
547 
548 	csp -= sp & 15UL;
549 	sp -= sp & 15UL;
550 
551 	/* put the ELF interpreter info on the stack */
552 #define NEW_AUX_ENT(nr, id, val)					\
553 	do {								\
554 		struct { unsigned long _id, _val; } __user *ent;	\
555 									\
556 		ent = (void __user *) csp;				\
557 		__put_user((id), &ent[nr]._id);				\
558 		__put_user((val), &ent[nr]._val);			\
559 	} while (0)
560 
561 	csp -= 2 * sizeof(unsigned long);
562 	NEW_AUX_ENT(0, AT_NULL, 0);
563 	if (k_platform) {
564 		csp -= 2 * sizeof(unsigned long);
565 		NEW_AUX_ENT(0, AT_PLATFORM,
566 			    (elf_addr_t) (unsigned long) u_platform);
567 	}
568 
569 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
570 	NEW_AUX_ENT( 0, AT_HWCAP,	hwcap);
571 	NEW_AUX_ENT( 1, AT_PAGESZ,	PAGE_SIZE);
572 	NEW_AUX_ENT( 2, AT_CLKTCK,	CLOCKS_PER_SEC);
573 	NEW_AUX_ENT( 3, AT_PHDR,	exec_params->ph_addr);
574 	NEW_AUX_ENT( 4, AT_PHENT,	sizeof(struct elf_phdr));
575 	NEW_AUX_ENT( 5, AT_PHNUM,	exec_params->hdr.e_phnum);
576 	NEW_AUX_ENT( 6,	AT_BASE,	interp_params->elfhdr_addr);
577 	NEW_AUX_ENT( 7, AT_FLAGS,	0);
578 	NEW_AUX_ENT( 8, AT_ENTRY,	exec_params->entry_addr);
579 	NEW_AUX_ENT( 9, AT_UID,		(elf_addr_t) current->uid);
580 	NEW_AUX_ENT(10, AT_EUID,	(elf_addr_t) current->euid);
581 	NEW_AUX_ENT(11, AT_GID,		(elf_addr_t) current->gid);
582 	NEW_AUX_ENT(12, AT_EGID,	(elf_addr_t) current->egid);
583 
584 #ifdef ARCH_DLINFO
585 	/* ARCH_DLINFO must come last so platform specific code can enforce
586 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
587 	 */
588 	ARCH_DLINFO;
589 #endif
590 #undef NEW_AUX_ENT
591 
592 	/* allocate room for argv[] and envv[] */
593 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
594 	envp = (elf_caddr_t __user *) csp;
595 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
596 	argv = (elf_caddr_t __user *) csp;
597 
598 	/* stack argc */
599 	csp -= sizeof(unsigned long);
600 	__put_user(bprm->argc, (unsigned long __user *) csp);
601 
602 	BUG_ON(csp != sp);
603 
604 	/* fill in the argv[] array */
605 #ifdef CONFIG_MMU
606 	current->mm->arg_start = bprm->p;
607 #else
608 	current->mm->arg_start = current->mm->start_stack -
609 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
610 #endif
611 
612 	p = (char __user *) current->mm->arg_start;
613 	for (loop = bprm->argc; loop > 0; loop--) {
614 		__put_user((elf_caddr_t) p, argv++);
615 		len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES);
616 		if (!len || len > PAGE_SIZE * MAX_ARG_PAGES)
617 			return -EINVAL;
618 		p += len;
619 	}
620 	__put_user(NULL, argv);
621 	current->mm->arg_end = (unsigned long) p;
622 
623 	/* fill in the envv[] array */
624 	current->mm->env_start = (unsigned long) p;
625 	for (loop = bprm->envc; loop > 0; loop--) {
626 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
627 		len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES);
628 		if (!len || len > PAGE_SIZE * MAX_ARG_PAGES)
629 			return -EINVAL;
630 		p += len;
631 	}
632 	__put_user(NULL, envp);
633 	current->mm->env_end = (unsigned long) p;
634 
635 	mm->start_stack = (unsigned long) sp;
636 	return 0;
637 }
638 
639 /*****************************************************************************/
640 /*
641  * transfer the program arguments and environment from the holding pages onto
642  * the stack
643  */
644 #ifndef CONFIG_MMU
645 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
646 					    unsigned long *_sp)
647 {
648 	unsigned long index, stop, sp;
649 	char *src;
650 	int ret = 0;
651 
652 	stop = bprm->p >> PAGE_SHIFT;
653 	sp = *_sp;
654 
655 	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
656 		src = kmap(bprm->page[index]);
657 		sp -= PAGE_SIZE;
658 		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
659 			ret = -EFAULT;
660 		kunmap(bprm->page[index]);
661 		if (ret < 0)
662 			goto out;
663 	}
664 
665 	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
666 
667 out:
668 	return ret;
669 }
670 #endif
671 
672 /*****************************************************************************/
673 /*
674  * load the appropriate binary image (executable or interpreter) into memory
675  * - we assume no MMU is available
676  * - if no other PIC bits are set in params->hdr->e_flags
677  *   - we assume that the LOADable segments in the binary are independently relocatable
678  *   - we assume R/O executable segments are shareable
679  * - else
680  *   - we assume the loadable parts of the image to require fixed displacement
681  *   - the image is not shareable
682  */
683 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
684 			      struct file *file,
685 			      struct mm_struct *mm,
686 			      const char *what)
687 {
688 	struct elf32_fdpic_loadmap *loadmap;
689 #ifdef CONFIG_MMU
690 	struct elf32_fdpic_loadseg *mseg;
691 #endif
692 	struct elf32_fdpic_loadseg *seg;
693 	struct elf32_phdr *phdr;
694 	unsigned long load_addr, stop;
695 	unsigned nloads, tmp;
696 	size_t size;
697 	int loop, ret;
698 
699 	/* allocate a load map table */
700 	nloads = 0;
701 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
702 		if (params->phdrs[loop].p_type == PT_LOAD)
703 			nloads++;
704 
705 	if (nloads == 0)
706 		return -ELIBBAD;
707 
708 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
709 	loadmap = kmalloc(size, GFP_KERNEL);
710 	if (!loadmap)
711 		return -ENOMEM;
712 
713 	params->loadmap = loadmap;
714 	memset(loadmap, 0, size);
715 
716 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
717 	loadmap->nsegs = nloads;
718 
719 	load_addr = params->load_addr;
720 	seg = loadmap->segs;
721 
722 	/* map the requested LOADs into the memory space */
723 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
724 	case ELF_FDPIC_FLAG_CONSTDISP:
725 	case ELF_FDPIC_FLAG_CONTIGUOUS:
726 #ifndef CONFIG_MMU
727 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
728 		if (ret < 0)
729 			return ret;
730 		break;
731 #endif
732 	default:
733 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
734 		if (ret < 0)
735 			return ret;
736 		break;
737 	}
738 
739 	/* map the entry point */
740 	if (params->hdr.e_entry) {
741 		seg = loadmap->segs;
742 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
743 			if (params->hdr.e_entry >= seg->p_vaddr &&
744 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
745 				params->entry_addr =
746 					(params->hdr.e_entry - seg->p_vaddr) +
747 					seg->addr;
748 				break;
749 			}
750 		}
751 	}
752 
753 	/* determine where the program header table has wound up if mapped */
754 	stop = params->hdr.e_phoff;
755 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
756 	phdr = params->phdrs;
757 
758 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
759 		if (phdr->p_type != PT_LOAD)
760 			continue;
761 
762 		if (phdr->p_offset > params->hdr.e_phoff ||
763 		    phdr->p_offset + phdr->p_filesz < stop)
764 			continue;
765 
766 		seg = loadmap->segs;
767 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
768 			if (phdr->p_vaddr >= seg->p_vaddr &&
769 			    phdr->p_vaddr + phdr->p_filesz <=
770 			    seg->p_vaddr + seg->p_memsz) {
771 				params->ph_addr =
772 					(phdr->p_vaddr - seg->p_vaddr) +
773 					seg->addr +
774 					params->hdr.e_phoff - phdr->p_offset;
775 				break;
776 			}
777 		}
778 		break;
779 	}
780 
781 	/* determine where the dynamic section has wound up if there is one */
782 	phdr = params->phdrs;
783 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
784 		if (phdr->p_type != PT_DYNAMIC)
785 			continue;
786 
787 		seg = loadmap->segs;
788 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
789 			if (phdr->p_vaddr >= seg->p_vaddr &&
790 			    phdr->p_vaddr + phdr->p_memsz <=
791 			    seg->p_vaddr + seg->p_memsz) {
792 				params->dynamic_addr =
793 					(phdr->p_vaddr - seg->p_vaddr) +
794 					seg->addr;
795 
796 				/* check the dynamic section contains at least
797 				 * one item, and that the last item is a NULL
798 				 * entry */
799 				if (phdr->p_memsz == 0 ||
800 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
801 					goto dynamic_error;
802 
803 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
804 				if (((Elf32_Dyn *)
805 				     params->dynamic_addr)[tmp - 1].d_tag != 0)
806 					goto dynamic_error;
807 				break;
808 			}
809 		}
810 		break;
811 	}
812 
813 	/* now elide adjacent segments in the load map on MMU linux
814 	 * - on uClinux the holes between may actually be filled with system
815 	 *   stuff or stuff from other processes
816 	 */
817 #ifdef CONFIG_MMU
818 	nloads = loadmap->nsegs;
819 	mseg = loadmap->segs;
820 	seg = mseg + 1;
821 	for (loop = 1; loop < nloads; loop++) {
822 		/* see if we have a candidate for merging */
823 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
824 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
825 			if (load_addr == (seg->addr & PAGE_MASK)) {
826 				mseg->p_memsz +=
827 					load_addr -
828 					(mseg->addr + mseg->p_memsz);
829 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
830 				mseg->p_memsz += seg->p_memsz;
831 				loadmap->nsegs--;
832 				continue;
833 			}
834 		}
835 
836 		mseg++;
837 		if (mseg != seg)
838 			*mseg = *seg;
839 	}
840 #endif
841 
842 	kdebug("Mapped Object [%s]:", what);
843 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
844 	kdebug("- entry    : %lx", params->entry_addr);
845 	kdebug("- PHDR[]   : %lx", params->ph_addr);
846 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
847 	seg = loadmap->segs;
848 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
849 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
850 		       loop,
851 		       seg->addr, seg->addr + seg->p_memsz - 1,
852 		       seg->p_vaddr, seg->p_memsz);
853 
854 	return 0;
855 
856 dynamic_error:
857 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
858 	       what, file->f_path.dentry->d_inode->i_ino);
859 	return -ELIBBAD;
860 }
861 
862 /*****************************************************************************/
863 /*
864  * map a file with constant displacement under uClinux
865  */
866 #ifndef CONFIG_MMU
867 static int elf_fdpic_map_file_constdisp_on_uclinux(
868 	struct elf_fdpic_params *params,
869 	struct file *file,
870 	struct mm_struct *mm)
871 {
872 	struct elf32_fdpic_loadseg *seg;
873 	struct elf32_phdr *phdr;
874 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
875 	loff_t fpos;
876 	int loop, ret;
877 
878 	load_addr = params->load_addr;
879 	seg = params->loadmap->segs;
880 
881 	/* determine the bounds of the contiguous overall allocation we must
882 	 * make */
883 	phdr = params->phdrs;
884 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
885 		if (params->phdrs[loop].p_type != PT_LOAD)
886 			continue;
887 
888 		if (base > phdr->p_vaddr)
889 			base = phdr->p_vaddr;
890 		if (top < phdr->p_vaddr + phdr->p_memsz)
891 			top = phdr->p_vaddr + phdr->p_memsz;
892 	}
893 
894 	/* allocate one big anon block for everything */
895 	mflags = MAP_PRIVATE;
896 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
897 		mflags |= MAP_EXECUTABLE;
898 
899 	down_write(&mm->mmap_sem);
900 	maddr = do_mmap(NULL, load_addr, top - base,
901 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
902 	up_write(&mm->mmap_sem);
903 	if (IS_ERR_VALUE(maddr))
904 		return (int) maddr;
905 
906 	if (load_addr != 0)
907 		load_addr += PAGE_ALIGN(top - base);
908 
909 	/* and then load the file segments into it */
910 	phdr = params->phdrs;
911 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
912 		if (params->phdrs[loop].p_type != PT_LOAD)
913 			continue;
914 
915 		fpos = phdr->p_offset;
916 
917 		seg->addr = maddr + (phdr->p_vaddr - base);
918 		seg->p_vaddr = phdr->p_vaddr;
919 		seg->p_memsz = phdr->p_memsz;
920 
921 		ret = file->f_op->read(file, (void *) seg->addr,
922 				       phdr->p_filesz, &fpos);
923 		if (ret < 0)
924 			return ret;
925 
926 		/* map the ELF header address if in this segment */
927 		if (phdr->p_offset == 0)
928 			params->elfhdr_addr = seg->addr;
929 
930 		/* clear any space allocated but not loaded */
931 		if (phdr->p_filesz < phdr->p_memsz)
932 			clear_user((void *) (seg->addr + phdr->p_filesz),
933 				   phdr->p_memsz - phdr->p_filesz);
934 
935 		if (mm) {
936 			if (phdr->p_flags & PF_X) {
937 				mm->start_code = seg->addr;
938 				mm->end_code = seg->addr + phdr->p_memsz;
939 			} else if (!mm->start_data) {
940 				mm->start_data = seg->addr;
941 #ifndef CONFIG_MMU
942 				mm->end_data = seg->addr + phdr->p_memsz;
943 #endif
944 			}
945 
946 #ifdef CONFIG_MMU
947 			if (seg->addr + phdr->p_memsz > mm->end_data)
948 				mm->end_data = seg->addr + phdr->p_memsz;
949 #endif
950 		}
951 
952 		seg++;
953 	}
954 
955 	return 0;
956 }
957 #endif
958 
959 /*****************************************************************************/
960 /*
961  * map a binary by direct mmap() of the individual PT_LOAD segments
962  */
963 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
964 					     struct file *file,
965 					     struct mm_struct *mm)
966 {
967 	struct elf32_fdpic_loadseg *seg;
968 	struct elf32_phdr *phdr;
969 	unsigned long load_addr, delta_vaddr;
970 	int loop, dvset;
971 
972 	load_addr = params->load_addr;
973 	delta_vaddr = 0;
974 	dvset = 0;
975 
976 	seg = params->loadmap->segs;
977 
978 	/* deal with each load segment separately */
979 	phdr = params->phdrs;
980 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
981 		unsigned long maddr, disp, excess, excess1;
982 		int prot = 0, flags;
983 
984 		if (phdr->p_type != PT_LOAD)
985 			continue;
986 
987 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
988 		       (unsigned long) phdr->p_vaddr,
989 		       (unsigned long) phdr->p_offset,
990 		       (unsigned long) phdr->p_filesz,
991 		       (unsigned long) phdr->p_memsz);
992 
993 		/* determine the mapping parameters */
994 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
995 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
996 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
997 
998 		flags = MAP_PRIVATE | MAP_DENYWRITE;
999 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1000 			flags |= MAP_EXECUTABLE;
1001 
1002 		maddr = 0;
1003 
1004 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1005 		case ELF_FDPIC_FLAG_INDEPENDENT:
1006 			/* PT_LOADs are independently locatable */
1007 			break;
1008 
1009 		case ELF_FDPIC_FLAG_HONOURVADDR:
1010 			/* the specified virtual address must be honoured */
1011 			maddr = phdr->p_vaddr;
1012 			flags |= MAP_FIXED;
1013 			break;
1014 
1015 		case ELF_FDPIC_FLAG_CONSTDISP:
1016 			/* constant displacement
1017 			 * - can be mapped anywhere, but must be mapped as a
1018 			 *   unit
1019 			 */
1020 			if (!dvset) {
1021 				maddr = load_addr;
1022 				delta_vaddr = phdr->p_vaddr;
1023 				dvset = 1;
1024 			} else {
1025 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1026 				flags |= MAP_FIXED;
1027 			}
1028 			break;
1029 
1030 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1031 			/* contiguity handled later */
1032 			break;
1033 
1034 		default:
1035 			BUG();
1036 		}
1037 
1038 		maddr &= PAGE_MASK;
1039 
1040 		/* create the mapping */
1041 		disp = phdr->p_vaddr & ~PAGE_MASK;
1042 		down_write(&mm->mmap_sem);
1043 		maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1044 				phdr->p_offset - disp);
1045 		up_write(&mm->mmap_sem);
1046 
1047 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1048 		       loop, phdr->p_memsz + disp, prot, flags,
1049 		       phdr->p_offset - disp, maddr);
1050 
1051 		if (IS_ERR_VALUE(maddr))
1052 			return (int) maddr;
1053 
1054 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1055 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1056 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1057 
1058 		seg->addr = maddr + disp;
1059 		seg->p_vaddr = phdr->p_vaddr;
1060 		seg->p_memsz = phdr->p_memsz;
1061 
1062 		/* map the ELF header address if in this segment */
1063 		if (phdr->p_offset == 0)
1064 			params->elfhdr_addr = seg->addr;
1065 
1066 		/* clear the bit between beginning of mapping and beginning of
1067 		 * PT_LOAD */
1068 		if (prot & PROT_WRITE && disp > 0) {
1069 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1070 			clear_user((void __user *) maddr, disp);
1071 			maddr += disp;
1072 		}
1073 
1074 		/* clear any space allocated but not loaded
1075 		 * - on uClinux we can just clear the lot
1076 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1077 		 *   extant in the file
1078 		 */
1079 		excess = phdr->p_memsz - phdr->p_filesz;
1080 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1081 
1082 #ifdef CONFIG_MMU
1083 		if (excess > excess1) {
1084 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1085 			unsigned long xmaddr;
1086 
1087 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1088 			down_write(&mm->mmap_sem);
1089 			xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1090 					 prot, flags, 0);
1091 			up_write(&mm->mmap_sem);
1092 
1093 			kdebug("mmap[%d] <anon>"
1094 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1095 			       loop, xaddr, excess - excess1, prot, flags,
1096 			       xmaddr);
1097 
1098 			if (xmaddr != xaddr)
1099 				return -ENOMEM;
1100 		}
1101 
1102 		if (prot & PROT_WRITE && excess1 > 0) {
1103 			kdebug("clear[%d] ad=%lx sz=%lx",
1104 			       loop, maddr + phdr->p_filesz, excess1);
1105 			clear_user((void __user *) maddr + phdr->p_filesz,
1106 				   excess1);
1107 		}
1108 
1109 #else
1110 		if (excess > 0) {
1111 			kdebug("clear[%d] ad=%lx sz=%lx",
1112 			       loop, maddr + phdr->p_filesz, excess);
1113 			clear_user((void *) maddr + phdr->p_filesz, excess);
1114 		}
1115 #endif
1116 
1117 		if (mm) {
1118 			if (phdr->p_flags & PF_X) {
1119 				mm->start_code = maddr;
1120 				mm->end_code = maddr + phdr->p_memsz;
1121 			} else if (!mm->start_data) {
1122 				mm->start_data = maddr;
1123 				mm->end_data = maddr + phdr->p_memsz;
1124 			}
1125 		}
1126 
1127 		seg++;
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 /*****************************************************************************/
1134 /*
1135  * ELF-FDPIC core dumper
1136  *
1137  * Modelled on fs/exec.c:aout_core_dump()
1138  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1139  *
1140  * Modelled on fs/binfmt_elf.c core dumper
1141  */
1142 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1143 
1144 /*
1145  * These are the only things you should do on a core-file: use only these
1146  * functions to write out all the necessary info.
1147  */
1148 static int dump_write(struct file *file, const void *addr, int nr)
1149 {
1150 	return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1151 }
1152 
1153 static int dump_seek(struct file *file, loff_t off)
1154 {
1155 	if (file->f_op->llseek) {
1156 		if (file->f_op->llseek(file, off, SEEK_SET) != off)
1157 			return 0;
1158 	} else {
1159 		file->f_pos = off;
1160 	}
1161 	return 1;
1162 }
1163 
1164 /*
1165  * Decide whether a segment is worth dumping; default is yes to be
1166  * sure (missing info is worse than too much; etc).
1167  * Personally I'd include everything, and use the coredump limit...
1168  *
1169  * I think we should skip something. But I am not sure how. H.J.
1170  */
1171 static int maydump(struct vm_area_struct *vma)
1172 {
1173 	/* Do not dump I/O mapped devices or special mappings */
1174 	if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1175 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1176 		return 0;
1177 	}
1178 
1179 	/* If we may not read the contents, don't allow us to dump
1180 	 * them either. "dump_write()" can't handle it anyway.
1181 	 */
1182 	if (!(vma->vm_flags & VM_READ)) {
1183 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1184 		return 0;
1185 	}
1186 
1187 	/* Dump shared memory only if mapped from an anonymous file. */
1188 	if (vma->vm_flags & VM_SHARED) {
1189 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1190 			kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags);
1191 			return 1;
1192 		}
1193 
1194 		kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags);
1195 		return 0;
1196 	}
1197 
1198 #ifdef CONFIG_MMU
1199 	/* If it hasn't been written to, don't write it out */
1200 	if (!vma->anon_vma) {
1201 		kdcore("%08lx: %08lx: no (!anon)", vma->vm_start, vma->vm_flags);
1202 		return 0;
1203 	}
1204 #endif
1205 
1206 	kdcore("%08lx: %08lx: yes", vma->vm_start, vma->vm_flags);
1207 	return 1;
1208 }
1209 
1210 /* An ELF note in memory */
1211 struct memelfnote
1212 {
1213 	const char *name;
1214 	int type;
1215 	unsigned int datasz;
1216 	void *data;
1217 };
1218 
1219 static int notesize(struct memelfnote *en)
1220 {
1221 	int sz;
1222 
1223 	sz = sizeof(struct elf_note);
1224 	sz += roundup(strlen(en->name) + 1, 4);
1225 	sz += roundup(en->datasz, 4);
1226 
1227 	return sz;
1228 }
1229 
1230 /* #define DEBUG */
1231 
1232 #define DUMP_WRITE(addr, nr)	\
1233 	do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1234 #define DUMP_SEEK(off)	\
1235 	do { if (!dump_seek(file, (off))) return 0; } while(0)
1236 
1237 static int writenote(struct memelfnote *men, struct file *file)
1238 {
1239 	struct elf_note en;
1240 
1241 	en.n_namesz = strlen(men->name) + 1;
1242 	en.n_descsz = men->datasz;
1243 	en.n_type = men->type;
1244 
1245 	DUMP_WRITE(&en, sizeof(en));
1246 	DUMP_WRITE(men->name, en.n_namesz);
1247 	/* XXX - cast from long long to long to avoid need for libgcc.a */
1248 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1249 	DUMP_WRITE(men->data, men->datasz);
1250 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1251 
1252 	return 1;
1253 }
1254 #undef DUMP_WRITE
1255 #undef DUMP_SEEK
1256 
1257 #define DUMP_WRITE(addr, nr)	\
1258 	if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1259 		goto end_coredump;
1260 #define DUMP_SEEK(off)	\
1261 	if (!dump_seek(file, (off))) \
1262 		goto end_coredump;
1263 
1264 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1265 {
1266 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1267 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1268 	elf->e_ident[EI_DATA] = ELF_DATA;
1269 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1270 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1271 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1272 
1273 	elf->e_type = ET_CORE;
1274 	elf->e_machine = ELF_ARCH;
1275 	elf->e_version = EV_CURRENT;
1276 	elf->e_entry = 0;
1277 	elf->e_phoff = sizeof(struct elfhdr);
1278 	elf->e_shoff = 0;
1279 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1280 	elf->e_ehsize = sizeof(struct elfhdr);
1281 	elf->e_phentsize = sizeof(struct elf_phdr);
1282 	elf->e_phnum = segs;
1283 	elf->e_shentsize = 0;
1284 	elf->e_shnum = 0;
1285 	elf->e_shstrndx = 0;
1286 	return;
1287 }
1288 
1289 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1290 {
1291 	phdr->p_type = PT_NOTE;
1292 	phdr->p_offset = offset;
1293 	phdr->p_vaddr = 0;
1294 	phdr->p_paddr = 0;
1295 	phdr->p_filesz = sz;
1296 	phdr->p_memsz = 0;
1297 	phdr->p_flags = 0;
1298 	phdr->p_align = 0;
1299 	return;
1300 }
1301 
1302 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1303 		unsigned int sz, void *data)
1304 {
1305 	note->name = name;
1306 	note->type = type;
1307 	note->datasz = sz;
1308 	note->data = data;
1309 	return;
1310 }
1311 
1312 /*
1313  * fill up all the fields in prstatus from the given task struct, except
1314  * registers which need to be filled up seperately.
1315  */
1316 static void fill_prstatus(struct elf_prstatus *prstatus,
1317 			  struct task_struct *p, long signr)
1318 {
1319 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1320 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1321 	prstatus->pr_sighold = p->blocked.sig[0];
1322 	prstatus->pr_pid = p->pid;
1323 	prstatus->pr_ppid = p->parent->pid;
1324 	prstatus->pr_pgrp = process_group(p);
1325 	prstatus->pr_sid = process_session(p);
1326 	if (thread_group_leader(p)) {
1327 		/*
1328 		 * This is the record for the group leader.  Add in the
1329 		 * cumulative times of previous dead threads.  This total
1330 		 * won't include the time of each live thread whose state
1331 		 * is included in the core dump.  The final total reported
1332 		 * to our parent process when it calls wait4 will include
1333 		 * those sums as well as the little bit more time it takes
1334 		 * this and each other thread to finish dying after the
1335 		 * core dump synchronization phase.
1336 		 */
1337 		cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1338 				   &prstatus->pr_utime);
1339 		cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1340 				   &prstatus->pr_stime);
1341 	} else {
1342 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1343 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1344 	}
1345 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1346 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1347 
1348 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1349 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1350 }
1351 
1352 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1353 		       struct mm_struct *mm)
1354 {
1355 	unsigned int i, len;
1356 
1357 	/* first copy the parameters from user space */
1358 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1359 
1360 	len = mm->arg_end - mm->arg_start;
1361 	if (len >= ELF_PRARGSZ)
1362 		len = ELF_PRARGSZ - 1;
1363 	if (copy_from_user(&psinfo->pr_psargs,
1364 		           (const char __user *) mm->arg_start, len))
1365 		return -EFAULT;
1366 	for (i = 0; i < len; i++)
1367 		if (psinfo->pr_psargs[i] == 0)
1368 			psinfo->pr_psargs[i] = ' ';
1369 	psinfo->pr_psargs[len] = 0;
1370 
1371 	psinfo->pr_pid = p->pid;
1372 	psinfo->pr_ppid = p->parent->pid;
1373 	psinfo->pr_pgrp = process_group(p);
1374 	psinfo->pr_sid = process_session(p);
1375 
1376 	i = p->state ? ffz(~p->state) + 1 : 0;
1377 	psinfo->pr_state = i;
1378 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1379 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1380 	psinfo->pr_nice = task_nice(p);
1381 	psinfo->pr_flag = p->flags;
1382 	SET_UID(psinfo->pr_uid, p->uid);
1383 	SET_GID(psinfo->pr_gid, p->gid);
1384 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1385 
1386 	return 0;
1387 }
1388 
1389 /* Here is the structure in which status of each thread is captured. */
1390 struct elf_thread_status
1391 {
1392 	struct list_head list;
1393 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1394 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1395 	struct task_struct *thread;
1396 #ifdef ELF_CORE_COPY_XFPREGS
1397 	elf_fpxregset_t xfpu;		/* NT_PRXFPREG */
1398 #endif
1399 	struct memelfnote notes[3];
1400 	int num_notes;
1401 };
1402 
1403 /*
1404  * In order to add the specific thread information for the elf file format,
1405  * we need to keep a linked list of every thread's pr_status and then create
1406  * a single section for them in the final core file.
1407  */
1408 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1409 {
1410 	struct task_struct *p = t->thread;
1411 	int sz = 0;
1412 
1413 	t->num_notes = 0;
1414 
1415 	fill_prstatus(&t->prstatus, p, signr);
1416 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1417 
1418 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1419 		  &t->prstatus);
1420 	t->num_notes++;
1421 	sz += notesize(&t->notes[0]);
1422 
1423 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1424 	if (t->prstatus.pr_fpvalid) {
1425 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1426 			  &t->fpu);
1427 		t->num_notes++;
1428 		sz += notesize(&t->notes[1]);
1429 	}
1430 
1431 #ifdef ELF_CORE_COPY_XFPREGS
1432 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1433 		fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu),
1434 			  &t->xfpu);
1435 		t->num_notes++;
1436 		sz += notesize(&t->notes[2]);
1437 	}
1438 #endif
1439 	return sz;
1440 }
1441 
1442 /*
1443  * dump the segments for an MMU process
1444  */
1445 #ifdef CONFIG_MMU
1446 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm,
1447 				   size_t *size, unsigned long *limit)
1448 {
1449 	struct vm_area_struct *vma;
1450 
1451 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1452 		unsigned long addr;
1453 
1454 		if (!maydump(vma))
1455 			continue;
1456 
1457 		for (addr = vma->vm_start;
1458 		     addr < vma->vm_end;
1459 		     addr += PAGE_SIZE
1460 		     ) {
1461 			struct vm_area_struct *vma;
1462 			struct page *page;
1463 
1464 			if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1465 					   &page, &vma) <= 0) {
1466 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1467 			}
1468 			else if (page == ZERO_PAGE(addr)) {
1469 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1470 				page_cache_release(page);
1471 			}
1472 			else {
1473 				void *kaddr;
1474 
1475 				flush_cache_page(vma, addr, page_to_pfn(page));
1476 				kaddr = kmap(page);
1477 				if ((*size += PAGE_SIZE) > *limit ||
1478 				    !dump_write(file, kaddr, PAGE_SIZE)
1479 				    ) {
1480 					kunmap(page);
1481 					page_cache_release(page);
1482 					return -EIO;
1483 				}
1484 				kunmap(page);
1485 				page_cache_release(page);
1486 			}
1487 		}
1488 	}
1489 
1490 	return 0;
1491 
1492 end_coredump:
1493 	return -EFBIG;
1494 }
1495 #endif
1496 
1497 /*
1498  * dump the segments for a NOMMU process
1499  */
1500 #ifndef CONFIG_MMU
1501 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm,
1502 				   size_t *size, unsigned long *limit)
1503 {
1504 	struct vm_list_struct *vml;
1505 
1506 	for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1507 	struct vm_area_struct *vma = vml->vma;
1508 
1509 		if (!maydump(vma))
1510 			continue;
1511 
1512 		if ((*size += PAGE_SIZE) > *limit)
1513 			return -EFBIG;
1514 
1515 		if (!dump_write(file, (void *) vma->vm_start,
1516 				vma->vm_end - vma->vm_start))
1517 			return -EIO;
1518 	}
1519 
1520 	return 0;
1521 }
1522 #endif
1523 
1524 /*
1525  * Actual dumper
1526  *
1527  * This is a two-pass process; first we find the offsets of the bits,
1528  * and then they are actually written out.  If we run out of core limit
1529  * we just truncate.
1530  */
1531 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1532 			       struct file *file)
1533 {
1534 #define	NUM_NOTES	6
1535 	int has_dumped = 0;
1536 	mm_segment_t fs;
1537 	int segs;
1538 	size_t size = 0;
1539 	int i;
1540 	struct vm_area_struct *vma;
1541 	struct elfhdr *elf = NULL;
1542 	loff_t offset = 0, dataoff;
1543 	unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1544 	int numnote;
1545 	struct memelfnote *notes = NULL;
1546 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1547 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1548  	struct task_struct *g, *p;
1549  	LIST_HEAD(thread_list);
1550  	struct list_head *t;
1551 	elf_fpregset_t *fpu = NULL;
1552 #ifdef ELF_CORE_COPY_XFPREGS
1553 	elf_fpxregset_t *xfpu = NULL;
1554 #endif
1555 	int thread_status_size = 0;
1556 #ifndef CONFIG_MMU
1557 	struct vm_list_struct *vml;
1558 #endif
1559 	elf_addr_t *auxv;
1560 
1561 	/*
1562 	 * We no longer stop all VM operations.
1563 	 *
1564 	 * This is because those proceses that could possibly change map_count
1565 	 * or the mmap / vma pages are now blocked in do_exit on current
1566 	 * finishing this core dump.
1567 	 *
1568 	 * Only ptrace can touch these memory addresses, but it doesn't change
1569 	 * the map_count or the pages allocated. So no possibility of crashing
1570 	 * exists while dumping the mm->vm_next areas to the core file.
1571 	 */
1572 
1573 	/* alloc memory for large data structures: too large to be on stack */
1574 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1575 	if (!elf)
1576 		goto cleanup;
1577 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1578 	if (!prstatus)
1579 		goto cleanup;
1580 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1581 	if (!psinfo)
1582 		goto cleanup;
1583 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1584 	if (!notes)
1585 		goto cleanup;
1586 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1587 	if (!fpu)
1588 		goto cleanup;
1589 #ifdef ELF_CORE_COPY_XFPREGS
1590 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1591 	if (!xfpu)
1592 		goto cleanup;
1593 #endif
1594 
1595 	if (signr) {
1596 		struct elf_thread_status *tmp;
1597 		rcu_read_lock();
1598 		do_each_thread(g,p)
1599 			if (current->mm == p->mm && current != p) {
1600 				tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1601 				if (!tmp) {
1602 					rcu_read_unlock();
1603 					goto cleanup;
1604 				}
1605 				tmp->thread = p;
1606 				list_add(&tmp->list, &thread_list);
1607 			}
1608 		while_each_thread(g,p);
1609 		rcu_read_unlock();
1610 		list_for_each(t, &thread_list) {
1611 			struct elf_thread_status *tmp;
1612 			int sz;
1613 
1614 			tmp = list_entry(t, struct elf_thread_status, list);
1615 			sz = elf_dump_thread_status(signr, tmp);
1616 			thread_status_size += sz;
1617 		}
1618 	}
1619 
1620 	/* now collect the dump for the current */
1621 	fill_prstatus(prstatus, current, signr);
1622 	elf_core_copy_regs(&prstatus->pr_reg, regs);
1623 
1624 #ifdef CONFIG_MMU
1625 	segs = current->mm->map_count;
1626 #else
1627 	segs = 0;
1628 	for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1629 	    segs++;
1630 #endif
1631 #ifdef ELF_CORE_EXTRA_PHDRS
1632 	segs += ELF_CORE_EXTRA_PHDRS;
1633 #endif
1634 
1635 	/* Set up header */
1636 	fill_elf_fdpic_header(elf, segs + 1);	/* including notes section */
1637 
1638 	has_dumped = 1;
1639 	current->flags |= PF_DUMPCORE;
1640 
1641 	/*
1642 	 * Set up the notes in similar form to SVR4 core dumps made
1643 	 * with info from their /proc.
1644 	 */
1645 
1646 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1647 	fill_psinfo(psinfo, current->group_leader, current->mm);
1648 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1649 
1650 	numnote = 2;
1651 
1652 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1653 
1654 	i = 0;
1655 	do
1656 		i += 2;
1657 	while (auxv[i - 2] != AT_NULL);
1658 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1659 		  i * sizeof(elf_addr_t), auxv);
1660 
1661   	/* Try to dump the FPU. */
1662 	if ((prstatus->pr_fpvalid =
1663 	     elf_core_copy_task_fpregs(current, regs, fpu)))
1664 		fill_note(notes + numnote++,
1665 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1666 #ifdef ELF_CORE_COPY_XFPREGS
1667 	if (elf_core_copy_task_xfpregs(current, xfpu))
1668 		fill_note(notes + numnote++,
1669 			  "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu);
1670 #endif
1671 
1672 	fs = get_fs();
1673 	set_fs(KERNEL_DS);
1674 
1675 	DUMP_WRITE(elf, sizeof(*elf));
1676 	offset += sizeof(*elf);				/* Elf header */
1677 	offset += (segs+1) * sizeof(struct elf_phdr);	/* Program headers */
1678 
1679 	/* Write notes phdr entry */
1680 	{
1681 		struct elf_phdr phdr;
1682 		int sz = 0;
1683 
1684 		for (i = 0; i < numnote; i++)
1685 			sz += notesize(notes + i);
1686 
1687 		sz += thread_status_size;
1688 
1689 		fill_elf_note_phdr(&phdr, sz, offset);
1690 		offset += sz;
1691 		DUMP_WRITE(&phdr, sizeof(phdr));
1692 	}
1693 
1694 	/* Page-align dumped data */
1695 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1696 
1697 	/* write program headers for segments dump */
1698 	for (
1699 #ifdef CONFIG_MMU
1700 		vma = current->mm->mmap; vma; vma = vma->vm_next
1701 #else
1702 			vml = current->mm->context.vmlist; vml; vml = vml->next
1703 #endif
1704 	     ) {
1705 		struct elf_phdr phdr;
1706 		size_t sz;
1707 
1708 #ifndef CONFIG_MMU
1709 		vma = vml->vma;
1710 #endif
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) ? 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 		DUMP_WRITE(&phdr, sizeof(phdr));
1729 	}
1730 
1731 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1732 	ELF_CORE_WRITE_EXTRA_PHDRS;
1733 #endif
1734 
1735  	/* write out the notes section */
1736 	for (i = 0; i < numnote; i++)
1737 		if (!writenote(notes + i, file))
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], file))
1747 				goto end_coredump;
1748 	}
1749 
1750 	DUMP_SEEK(dataoff);
1751 
1752 	if (elf_fdpic_dump_segments(file, current->mm, &size, &limit) < 0)
1753 		goto end_coredump;
1754 
1755 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1756 	ELF_CORE_WRITE_EXTRA_DATA;
1757 #endif
1758 
1759 	if (file->f_pos != offset) {
1760 		/* Sanity check */
1761 		printk(KERN_WARNING
1762 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1763 		       file->f_pos, offset);
1764 	}
1765 
1766 end_coredump:
1767 	set_fs(fs);
1768 
1769 cleanup:
1770 	while (!list_empty(&thread_list)) {
1771 		struct list_head *tmp = thread_list.next;
1772 		list_del(tmp);
1773 		kfree(list_entry(tmp, struct elf_thread_status, list));
1774 	}
1775 
1776 	kfree(elf);
1777 	kfree(prstatus);
1778 	kfree(psinfo);
1779 	kfree(notes);
1780 	kfree(fpu);
1781 #ifdef ELF_CORE_COPY_XFPREGS
1782 	kfree(xfpu);
1783 #endif
1784 	return has_dumped;
1785 #undef NUM_NOTES
1786 }
1787 
1788 #endif		/* USE_ELF_CORE_DUMP */
1789