xref: /openbmc/linux/fs/binfmt_elf_fdpic.c (revision f15cbe6f1a4b4d9df59142fc8e4abb973302cf44)
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2  *
3  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  * Derived from binfmt_elf.c
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version
10  * 2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/module.h>
14 
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/elf.h>
34 #include <linux/elf-fdpic.h>
35 #include <linux/elfcore.h>
36 
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/pgalloc.h>
40 
41 typedef char *elf_caddr_t;
42 
43 #if 0
44 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
45 #else
46 #define kdebug(fmt, ...) do {} while(0)
47 #endif
48 
49 #if 0
50 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
51 #else
52 #define kdcore(fmt, ...) do {} while(0)
53 #endif
54 
55 MODULE_LICENSE("GPL");
56 
57 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
58 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
59 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
60 			      struct mm_struct *, const char *);
61 
62 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
63 				   struct elf_fdpic_params *,
64 				   struct elf_fdpic_params *);
65 
66 #ifndef CONFIG_MMU
67 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
68 					    unsigned long *);
69 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
70 						   struct file *,
71 						   struct mm_struct *);
72 #endif
73 
74 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
75 					     struct file *, struct mm_struct *);
76 
77 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
78 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *, unsigned long limit);
79 #endif
80 
81 static struct linux_binfmt elf_fdpic_format = {
82 	.module		= THIS_MODULE,
83 	.load_binary	= load_elf_fdpic_binary,
84 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
85 	.core_dump	= elf_fdpic_core_dump,
86 #endif
87 	.min_coredump	= ELF_EXEC_PAGESIZE,
88 };
89 
90 static int __init init_elf_fdpic_binfmt(void)
91 {
92 	return register_binfmt(&elf_fdpic_format);
93 }
94 
95 static void __exit exit_elf_fdpic_binfmt(void)
96 {
97 	unregister_binfmt(&elf_fdpic_format);
98 }
99 
100 core_initcall(init_elf_fdpic_binfmt);
101 module_exit(exit_elf_fdpic_binfmt);
102 
103 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
104 {
105 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
106 		return 0;
107 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
108 		return 0;
109 	if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
110 		return 0;
111 	if (!file->f_op || !file->f_op->mmap)
112 		return 0;
113 	return 1;
114 }
115 
116 /*****************************************************************************/
117 /*
118  * read the program headers table into memory
119  */
120 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
121 				 struct file *file)
122 {
123 	struct elf32_phdr *phdr;
124 	unsigned long size;
125 	int retval, loop;
126 
127 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
128 		return -ENOMEM;
129 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
130 		return -ENOMEM;
131 
132 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
133 	params->phdrs = kmalloc(size, GFP_KERNEL);
134 	if (!params->phdrs)
135 		return -ENOMEM;
136 
137 	retval = kernel_read(file, params->hdr.e_phoff,
138 			     (char *) params->phdrs, size);
139 	if (unlikely(retval != size))
140 		return retval < 0 ? retval : -ENOEXEC;
141 
142 	/* determine stack size for this binary */
143 	phdr = params->phdrs;
144 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
145 		if (phdr->p_type != PT_GNU_STACK)
146 			continue;
147 
148 		if (phdr->p_flags & PF_X)
149 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
150 		else
151 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
152 
153 		params->stack_size = phdr->p_memsz;
154 		break;
155 	}
156 
157 	return 0;
158 }
159 
160 /*****************************************************************************/
161 /*
162  * load an fdpic binary into various bits of memory
163  */
164 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
165 				 struct pt_regs *regs)
166 {
167 	struct elf_fdpic_params exec_params, interp_params;
168 	struct elf_phdr *phdr;
169 	unsigned long stack_size, entryaddr;
170 #ifndef CONFIG_MMU
171 	unsigned long fullsize;
172 #endif
173 #ifdef ELF_FDPIC_PLAT_INIT
174 	unsigned long dynaddr;
175 #endif
176 	struct file *interpreter = NULL; /* to shut gcc up */
177 	char *interpreter_name = NULL;
178 	int executable_stack;
179 	int retval, i;
180 
181 	kdebug("____ LOAD %d ____", current->pid);
182 
183 	memset(&exec_params, 0, sizeof(exec_params));
184 	memset(&interp_params, 0, sizeof(interp_params));
185 
186 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
187 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
188 
189 	/* check that this is a binary we know how to deal with */
190 	retval = -ENOEXEC;
191 	if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
192 		goto error;
193 
194 	/* read the program header table */
195 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
196 	if (retval < 0)
197 		goto error;
198 
199 	/* scan for a program header that specifies an interpreter */
200 	phdr = exec_params.phdrs;
201 
202 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
203 		switch (phdr->p_type) {
204 		case PT_INTERP:
205 			retval = -ENOMEM;
206 			if (phdr->p_filesz > PATH_MAX)
207 				goto error;
208 			retval = -ENOENT;
209 			if (phdr->p_filesz < 2)
210 				goto error;
211 
212 			/* read the name of the interpreter into memory */
213 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
214 			if (!interpreter_name)
215 				goto error;
216 
217 			retval = kernel_read(bprm->file,
218 					     phdr->p_offset,
219 					     interpreter_name,
220 					     phdr->p_filesz);
221 			if (unlikely(retval != phdr->p_filesz)) {
222 				if (retval >= 0)
223 					retval = -ENOEXEC;
224 				goto error;
225 			}
226 
227 			retval = -ENOENT;
228 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
229 				goto error;
230 
231 			kdebug("Using ELF interpreter %s", interpreter_name);
232 
233 			/* replace the program with the interpreter */
234 			interpreter = open_exec(interpreter_name);
235 			retval = PTR_ERR(interpreter);
236 			if (IS_ERR(interpreter)) {
237 				interpreter = NULL;
238 				goto error;
239 			}
240 
241 			/*
242 			 * If the binary is not readable then enforce
243 			 * mm->dumpable = 0 regardless of the interpreter's
244 			 * permissions.
245 			 */
246 			if (file_permission(interpreter, MAY_READ) < 0)
247 				bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
248 
249 			retval = kernel_read(interpreter, 0, bprm->buf,
250 					     BINPRM_BUF_SIZE);
251 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
252 				if (retval >= 0)
253 					retval = -ENOEXEC;
254 				goto error;
255 			}
256 
257 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
258 			break;
259 
260 		case PT_LOAD:
261 #ifdef CONFIG_MMU
262 			if (exec_params.load_addr == 0)
263 				exec_params.load_addr = phdr->p_vaddr;
264 #endif
265 			break;
266 		}
267 
268 	}
269 
270 	if (elf_check_const_displacement(&exec_params.hdr))
271 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
272 
273 	/* perform insanity checks on the interpreter */
274 	if (interpreter_name) {
275 		retval = -ELIBBAD;
276 		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
277 			goto error;
278 
279 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
280 
281 		/* read the interpreter's program header table */
282 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
283 		if (retval < 0)
284 			goto error;
285 	}
286 
287 	stack_size = exec_params.stack_size;
288 	if (stack_size < interp_params.stack_size)
289 		stack_size = interp_params.stack_size;
290 
291 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
292 		executable_stack = EXSTACK_ENABLE_X;
293 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
294 		executable_stack = EXSTACK_DISABLE_X;
295 	else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
296 		executable_stack = EXSTACK_ENABLE_X;
297 	else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
298 		executable_stack = EXSTACK_DISABLE_X;
299 	else
300 		executable_stack = EXSTACK_DEFAULT;
301 
302 	retval = -ENOEXEC;
303 	if (stack_size == 0)
304 		goto error;
305 
306 	if (elf_check_const_displacement(&interp_params.hdr))
307 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
308 
309 	/* flush all traces of the currently running executable */
310 	retval = flush_old_exec(bprm);
311 	if (retval)
312 		goto error;
313 
314 	/* there's now no turning back... the old userspace image is dead,
315 	 * defunct, deceased, etc. after this point we have to exit via
316 	 * error_kill */
317 	set_personality(PER_LINUX_FDPIC);
318 	set_binfmt(&elf_fdpic_format);
319 
320 	current->mm->start_code = 0;
321 	current->mm->end_code = 0;
322 	current->mm->start_stack = 0;
323 	current->mm->start_data = 0;
324 	current->mm->end_data = 0;
325 	current->mm->context.exec_fdpic_loadmap = 0;
326 	current->mm->context.interp_fdpic_loadmap = 0;
327 
328 	current->flags &= ~PF_FORKNOEXEC;
329 
330 #ifdef CONFIG_MMU
331 	elf_fdpic_arch_lay_out_mm(&exec_params,
332 				  &interp_params,
333 				  &current->mm->start_stack,
334 				  &current->mm->start_brk);
335 
336 	retval = setup_arg_pages(bprm, current->mm->start_stack,
337 				 executable_stack);
338 	if (retval < 0) {
339 		send_sig(SIGKILL, current, 0);
340 		goto error_kill;
341 	}
342 #endif
343 
344 	/* load the executable and interpreter into memory */
345 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
346 				    "executable");
347 	if (retval < 0)
348 		goto error_kill;
349 
350 	if (interpreter_name) {
351 		retval = elf_fdpic_map_file(&interp_params, interpreter,
352 					    current->mm, "interpreter");
353 		if (retval < 0) {
354 			printk(KERN_ERR "Unable to load interpreter\n");
355 			goto error_kill;
356 		}
357 
358 		allow_write_access(interpreter);
359 		fput(interpreter);
360 		interpreter = NULL;
361 	}
362 
363 #ifdef CONFIG_MMU
364 	if (!current->mm->start_brk)
365 		current->mm->start_brk = current->mm->end_data;
366 
367 	current->mm->brk = current->mm->start_brk =
368 		PAGE_ALIGN(current->mm->start_brk);
369 
370 #else
371 	/* create a stack and brk area big enough for everyone
372 	 * - the brk heap starts at the bottom and works up
373 	 * - the stack starts at the top and works down
374 	 */
375 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
376 	if (stack_size < PAGE_SIZE * 2)
377 		stack_size = PAGE_SIZE * 2;
378 
379 	down_write(&current->mm->mmap_sem);
380 	current->mm->start_brk = do_mmap(NULL, 0, stack_size,
381 					 PROT_READ | PROT_WRITE | PROT_EXEC,
382 					 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
383 					 0);
384 
385 	if (IS_ERR_VALUE(current->mm->start_brk)) {
386 		up_write(&current->mm->mmap_sem);
387 		retval = current->mm->start_brk;
388 		current->mm->start_brk = 0;
389 		goto error_kill;
390 	}
391 
392 	/* expand the stack mapping to use up the entire allocation granule */
393 	fullsize = kobjsize((char *) current->mm->start_brk);
394 	if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
395 				    fullsize, 0, 0)))
396 		stack_size = fullsize;
397 	up_write(&current->mm->mmap_sem);
398 
399 	current->mm->brk = current->mm->start_brk;
400 	current->mm->context.end_brk = current->mm->start_brk;
401 	current->mm->context.end_brk +=
402 		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
403 	current->mm->start_stack = current->mm->start_brk + stack_size;
404 #endif
405 
406 	compute_creds(bprm);
407 	current->flags &= ~PF_FORKNOEXEC;
408 	if (create_elf_fdpic_tables(bprm, current->mm,
409 				    &exec_params, &interp_params) < 0)
410 		goto error_kill;
411 
412 	kdebug("- start_code  %lx", current->mm->start_code);
413 	kdebug("- end_code    %lx", current->mm->end_code);
414 	kdebug("- start_data  %lx", current->mm->start_data);
415 	kdebug("- end_data    %lx", current->mm->end_data);
416 	kdebug("- start_brk   %lx", current->mm->start_brk);
417 	kdebug("- brk         %lx", current->mm->brk);
418 	kdebug("- start_stack %lx", current->mm->start_stack);
419 
420 #ifdef ELF_FDPIC_PLAT_INIT
421 	/*
422 	 * The ABI may specify that certain registers be set up in special
423 	 * ways (on i386 %edx is the address of a DT_FINI function, for
424 	 * example.  This macro performs whatever initialization to
425 	 * the regs structure is required.
426 	 */
427 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
428 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
429 			    dynaddr);
430 #endif
431 
432 	/* everything is now ready... get the userspace context ready to roll */
433 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
434 	start_thread(regs, entryaddr, current->mm->start_stack);
435 
436 	retval = 0;
437 
438 error:
439 	if (interpreter) {
440 		allow_write_access(interpreter);
441 		fput(interpreter);
442 	}
443 	kfree(interpreter_name);
444 	kfree(exec_params.phdrs);
445 	kfree(exec_params.loadmap);
446 	kfree(interp_params.phdrs);
447 	kfree(interp_params.loadmap);
448 	return retval;
449 
450 	/* unrecoverable error - kill the process */
451 error_kill:
452 	send_sig(SIGSEGV, current, 0);
453 	goto error;
454 
455 }
456 
457 /*****************************************************************************/
458 /*
459  * present useful information to the program
460  */
461 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
462 				   struct mm_struct *mm,
463 				   struct elf_fdpic_params *exec_params,
464 				   struct elf_fdpic_params *interp_params)
465 {
466 	unsigned long sp, csp, nitems;
467 	elf_caddr_t __user *argv, *envp;
468 	size_t platform_len = 0, len;
469 	char *k_platform;
470 	char __user *u_platform, *p;
471 	long hwcap;
472 	int loop;
473 	int nr;	/* reset for each csp adjustment */
474 
475 	/* we're going to shovel a whole load of stuff onto the stack */
476 #ifdef CONFIG_MMU
477 	sp = bprm->p;
478 #else
479 	sp = mm->start_stack;
480 
481 	/* stack the program arguments and environment */
482 	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
483 		return -EFAULT;
484 #endif
485 
486 	/* get hold of platform and hardware capabilities masks for the machine
487 	 * we are running on.  In some cases (Sparc), this info is impossible
488 	 * to get, in others (i386) it is merely difficult.
489 	 */
490 	hwcap = ELF_HWCAP;
491 	k_platform = ELF_PLATFORM;
492 	u_platform = NULL;
493 
494 	if (k_platform) {
495 		platform_len = strlen(k_platform) + 1;
496 		sp -= platform_len;
497 		u_platform = (char __user *) sp;
498 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
499 			return -EFAULT;
500 	}
501 
502 #if defined(__i386__) && defined(CONFIG_SMP)
503 	/* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
504 	 * by the processes running on the same package. One thing we can do is
505 	 * to shuffle the initial stack for them.
506 	 *
507 	 * the conditionals here are unneeded, but kept in to make the code
508 	 * behaviour the same as pre change unless we have hyperthreaded
509 	 * processors. This keeps Mr Marcelo Person happier but should be
510 	 * removed for 2.5
511 	 */
512 	if (smp_num_siblings > 1)
513 		sp = sp - ((current->pid % 64) << 7);
514 #endif
515 
516 	sp &= ~7UL;
517 
518 	/* stack the load map(s) */
519 	len = sizeof(struct elf32_fdpic_loadmap);
520 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
521 	sp = (sp - len) & ~7UL;
522 	exec_params->map_addr = sp;
523 
524 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
525 		return -EFAULT;
526 
527 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
528 
529 	if (interp_params->loadmap) {
530 		len = sizeof(struct elf32_fdpic_loadmap);
531 		len += sizeof(struct elf32_fdpic_loadseg) *
532 			interp_params->loadmap->nsegs;
533 		sp = (sp - len) & ~7UL;
534 		interp_params->map_addr = sp;
535 
536 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
537 				 len) != 0)
538 			return -EFAULT;
539 
540 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
541 	}
542 
543 	/* force 16 byte _final_ alignment here for generality */
544 #define DLINFO_ITEMS 13
545 
546 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
547 
548 	csp = sp;
549 	sp -= nitems * 2 * sizeof(unsigned long);
550 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
551 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
552 	sp -= 1 * sizeof(unsigned long);		/* argc */
553 
554 	csp -= sp & 15UL;
555 	sp -= sp & 15UL;
556 
557 	/* put the ELF interpreter info on the stack */
558 #define NEW_AUX_ENT(id, val)						\
559 	do {								\
560 		struct { unsigned long _id, _val; } __user *ent;	\
561 									\
562 		ent = (void __user *) csp;				\
563 		__put_user((id), &ent[nr]._id);				\
564 		__put_user((val), &ent[nr]._val);			\
565 		nr++;							\
566 	} while (0)
567 
568 	nr = 0;
569 	csp -= 2 * sizeof(unsigned long);
570 	NEW_AUX_ENT(AT_NULL, 0);
571 	if (k_platform) {
572 		nr = 0;
573 		csp -= 2 * sizeof(unsigned long);
574 		NEW_AUX_ENT(AT_PLATFORM,
575 			    (elf_addr_t) (unsigned long) u_platform);
576 	}
577 
578 	nr = 0;
579 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
580 	NEW_AUX_ENT(AT_HWCAP,	hwcap);
581 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
582 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
583 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
584 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
585 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
586 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
587 	NEW_AUX_ENT(AT_FLAGS,	0);
588 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
589 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) current->uid);
590 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) current->euid);
591 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) current->gid);
592 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) current->egid);
593 
594 #ifdef ARCH_DLINFO
595 	nr = 0;
596 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
597 
598 	/* ARCH_DLINFO must come last so platform specific code can enforce
599 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
600 	 */
601 	ARCH_DLINFO;
602 #endif
603 #undef NEW_AUX_ENT
604 
605 	/* allocate room for argv[] and envv[] */
606 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
607 	envp = (elf_caddr_t __user *) csp;
608 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
609 	argv = (elf_caddr_t __user *) csp;
610 
611 	/* stack argc */
612 	csp -= sizeof(unsigned long);
613 	__put_user(bprm->argc, (unsigned long __user *) csp);
614 
615 	BUG_ON(csp != sp);
616 
617 	/* fill in the argv[] array */
618 #ifdef CONFIG_MMU
619 	current->mm->arg_start = bprm->p;
620 #else
621 	current->mm->arg_start = current->mm->start_stack -
622 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
623 #endif
624 
625 	p = (char __user *) current->mm->arg_start;
626 	for (loop = bprm->argc; loop > 0; loop--) {
627 		__put_user((elf_caddr_t) p, argv++);
628 		len = strnlen_user(p, MAX_ARG_STRLEN);
629 		if (!len || len > MAX_ARG_STRLEN)
630 			return -EINVAL;
631 		p += len;
632 	}
633 	__put_user(NULL, argv);
634 	current->mm->arg_end = (unsigned long) p;
635 
636 	/* fill in the envv[] array */
637 	current->mm->env_start = (unsigned long) p;
638 	for (loop = bprm->envc; loop > 0; loop--) {
639 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
640 		len = strnlen_user(p, MAX_ARG_STRLEN);
641 		if (!len || len > MAX_ARG_STRLEN)
642 			return -EINVAL;
643 		p += len;
644 	}
645 	__put_user(NULL, envp);
646 	current->mm->env_end = (unsigned long) p;
647 
648 	mm->start_stack = (unsigned long) sp;
649 	return 0;
650 }
651 
652 /*****************************************************************************/
653 /*
654  * transfer the program arguments and environment from the holding pages onto
655  * the stack
656  */
657 #ifndef CONFIG_MMU
658 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
659 					    unsigned long *_sp)
660 {
661 	unsigned long index, stop, sp;
662 	char *src;
663 	int ret = 0;
664 
665 	stop = bprm->p >> PAGE_SHIFT;
666 	sp = *_sp;
667 
668 	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
669 		src = kmap(bprm->page[index]);
670 		sp -= PAGE_SIZE;
671 		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
672 			ret = -EFAULT;
673 		kunmap(bprm->page[index]);
674 		if (ret < 0)
675 			goto out;
676 	}
677 
678 	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
679 
680 out:
681 	return ret;
682 }
683 #endif
684 
685 /*****************************************************************************/
686 /*
687  * load the appropriate binary image (executable or interpreter) into memory
688  * - we assume no MMU is available
689  * - if no other PIC bits are set in params->hdr->e_flags
690  *   - we assume that the LOADable segments in the binary are independently relocatable
691  *   - we assume R/O executable segments are shareable
692  * - else
693  *   - we assume the loadable parts of the image to require fixed displacement
694  *   - the image is not shareable
695  */
696 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
697 			      struct file *file,
698 			      struct mm_struct *mm,
699 			      const char *what)
700 {
701 	struct elf32_fdpic_loadmap *loadmap;
702 #ifdef CONFIG_MMU
703 	struct elf32_fdpic_loadseg *mseg;
704 #endif
705 	struct elf32_fdpic_loadseg *seg;
706 	struct elf32_phdr *phdr;
707 	unsigned long load_addr, stop;
708 	unsigned nloads, tmp;
709 	size_t size;
710 	int loop, ret;
711 
712 	/* allocate a load map table */
713 	nloads = 0;
714 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
715 		if (params->phdrs[loop].p_type == PT_LOAD)
716 			nloads++;
717 
718 	if (nloads == 0)
719 		return -ELIBBAD;
720 
721 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
722 	loadmap = kzalloc(size, GFP_KERNEL);
723 	if (!loadmap)
724 		return -ENOMEM;
725 
726 	params->loadmap = loadmap;
727 
728 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
729 	loadmap->nsegs = nloads;
730 
731 	load_addr = params->load_addr;
732 	seg = loadmap->segs;
733 
734 	/* map the requested LOADs into the memory space */
735 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
736 	case ELF_FDPIC_FLAG_CONSTDISP:
737 	case ELF_FDPIC_FLAG_CONTIGUOUS:
738 #ifndef CONFIG_MMU
739 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
740 		if (ret < 0)
741 			return ret;
742 		break;
743 #endif
744 	default:
745 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
746 		if (ret < 0)
747 			return ret;
748 		break;
749 	}
750 
751 	/* map the entry point */
752 	if (params->hdr.e_entry) {
753 		seg = loadmap->segs;
754 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
755 			if (params->hdr.e_entry >= seg->p_vaddr &&
756 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
757 				params->entry_addr =
758 					(params->hdr.e_entry - seg->p_vaddr) +
759 					seg->addr;
760 				break;
761 			}
762 		}
763 	}
764 
765 	/* determine where the program header table has wound up if mapped */
766 	stop = params->hdr.e_phoff;
767 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
768 	phdr = params->phdrs;
769 
770 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
771 		if (phdr->p_type != PT_LOAD)
772 			continue;
773 
774 		if (phdr->p_offset > params->hdr.e_phoff ||
775 		    phdr->p_offset + phdr->p_filesz < stop)
776 			continue;
777 
778 		seg = loadmap->segs;
779 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
780 			if (phdr->p_vaddr >= seg->p_vaddr &&
781 			    phdr->p_vaddr + phdr->p_filesz <=
782 			    seg->p_vaddr + seg->p_memsz) {
783 				params->ph_addr =
784 					(phdr->p_vaddr - seg->p_vaddr) +
785 					seg->addr +
786 					params->hdr.e_phoff - phdr->p_offset;
787 				break;
788 			}
789 		}
790 		break;
791 	}
792 
793 	/* determine where the dynamic section has wound up if there is one */
794 	phdr = params->phdrs;
795 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
796 		if (phdr->p_type != PT_DYNAMIC)
797 			continue;
798 
799 		seg = loadmap->segs;
800 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
801 			if (phdr->p_vaddr >= seg->p_vaddr &&
802 			    phdr->p_vaddr + phdr->p_memsz <=
803 			    seg->p_vaddr + seg->p_memsz) {
804 				params->dynamic_addr =
805 					(phdr->p_vaddr - seg->p_vaddr) +
806 					seg->addr;
807 
808 				/* check the dynamic section contains at least
809 				 * one item, and that the last item is a NULL
810 				 * entry */
811 				if (phdr->p_memsz == 0 ||
812 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
813 					goto dynamic_error;
814 
815 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
816 				if (((Elf32_Dyn *)
817 				     params->dynamic_addr)[tmp - 1].d_tag != 0)
818 					goto dynamic_error;
819 				break;
820 			}
821 		}
822 		break;
823 	}
824 
825 	/* now elide adjacent segments in the load map on MMU linux
826 	 * - on uClinux the holes between may actually be filled with system
827 	 *   stuff or stuff from other processes
828 	 */
829 #ifdef CONFIG_MMU
830 	nloads = loadmap->nsegs;
831 	mseg = loadmap->segs;
832 	seg = mseg + 1;
833 	for (loop = 1; loop < nloads; loop++) {
834 		/* see if we have a candidate for merging */
835 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
836 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
837 			if (load_addr == (seg->addr & PAGE_MASK)) {
838 				mseg->p_memsz +=
839 					load_addr -
840 					(mseg->addr + mseg->p_memsz);
841 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
842 				mseg->p_memsz += seg->p_memsz;
843 				loadmap->nsegs--;
844 				continue;
845 			}
846 		}
847 
848 		mseg++;
849 		if (mseg != seg)
850 			*mseg = *seg;
851 	}
852 #endif
853 
854 	kdebug("Mapped Object [%s]:", what);
855 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
856 	kdebug("- entry    : %lx", params->entry_addr);
857 	kdebug("- PHDR[]   : %lx", params->ph_addr);
858 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
859 	seg = loadmap->segs;
860 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
861 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
862 		       loop,
863 		       seg->addr, seg->addr + seg->p_memsz - 1,
864 		       seg->p_vaddr, seg->p_memsz);
865 
866 	return 0;
867 
868 dynamic_error:
869 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
870 	       what, file->f_path.dentry->d_inode->i_ino);
871 	return -ELIBBAD;
872 }
873 
874 /*****************************************************************************/
875 /*
876  * map a file with constant displacement under uClinux
877  */
878 #ifndef CONFIG_MMU
879 static int elf_fdpic_map_file_constdisp_on_uclinux(
880 	struct elf_fdpic_params *params,
881 	struct file *file,
882 	struct mm_struct *mm)
883 {
884 	struct elf32_fdpic_loadseg *seg;
885 	struct elf32_phdr *phdr;
886 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
887 	loff_t fpos;
888 	int loop, ret;
889 
890 	load_addr = params->load_addr;
891 	seg = params->loadmap->segs;
892 
893 	/* determine the bounds of the contiguous overall allocation we must
894 	 * make */
895 	phdr = params->phdrs;
896 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
897 		if (params->phdrs[loop].p_type != PT_LOAD)
898 			continue;
899 
900 		if (base > phdr->p_vaddr)
901 			base = phdr->p_vaddr;
902 		if (top < phdr->p_vaddr + phdr->p_memsz)
903 			top = phdr->p_vaddr + phdr->p_memsz;
904 	}
905 
906 	/* allocate one big anon block for everything */
907 	mflags = MAP_PRIVATE;
908 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
909 		mflags |= MAP_EXECUTABLE;
910 
911 	down_write(&mm->mmap_sem);
912 	maddr = do_mmap(NULL, load_addr, top - base,
913 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
914 	up_write(&mm->mmap_sem);
915 	if (IS_ERR_VALUE(maddr))
916 		return (int) maddr;
917 
918 	if (load_addr != 0)
919 		load_addr += PAGE_ALIGN(top - base);
920 
921 	/* and then load the file segments into it */
922 	phdr = params->phdrs;
923 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
924 		if (params->phdrs[loop].p_type != PT_LOAD)
925 			continue;
926 
927 		fpos = phdr->p_offset;
928 
929 		seg->addr = maddr + (phdr->p_vaddr - base);
930 		seg->p_vaddr = phdr->p_vaddr;
931 		seg->p_memsz = phdr->p_memsz;
932 
933 		ret = file->f_op->read(file, (void *) seg->addr,
934 				       phdr->p_filesz, &fpos);
935 		if (ret < 0)
936 			return ret;
937 
938 		/* map the ELF header address if in this segment */
939 		if (phdr->p_offset == 0)
940 			params->elfhdr_addr = seg->addr;
941 
942 		/* clear any space allocated but not loaded */
943 		if (phdr->p_filesz < phdr->p_memsz)
944 			clear_user((void *) (seg->addr + phdr->p_filesz),
945 				   phdr->p_memsz - phdr->p_filesz);
946 
947 		if (mm) {
948 			if (phdr->p_flags & PF_X) {
949 				if (!mm->start_code) {
950 					mm->start_code = seg->addr;
951 					mm->end_code = seg->addr +
952 						phdr->p_memsz;
953 				}
954 			} else if (!mm->start_data) {
955 				mm->start_data = seg->addr;
956 #ifndef CONFIG_MMU
957 				mm->end_data = seg->addr + phdr->p_memsz;
958 #endif
959 			}
960 
961 #ifdef CONFIG_MMU
962 			if (seg->addr + phdr->p_memsz > mm->end_data)
963 				mm->end_data = seg->addr + phdr->p_memsz;
964 #endif
965 		}
966 
967 		seg++;
968 	}
969 
970 	return 0;
971 }
972 #endif
973 
974 /*****************************************************************************/
975 /*
976  * map a binary by direct mmap() of the individual PT_LOAD segments
977  */
978 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
979 					     struct file *file,
980 					     struct mm_struct *mm)
981 {
982 	struct elf32_fdpic_loadseg *seg;
983 	struct elf32_phdr *phdr;
984 	unsigned long load_addr, delta_vaddr;
985 	int loop, dvset;
986 
987 	load_addr = params->load_addr;
988 	delta_vaddr = 0;
989 	dvset = 0;
990 
991 	seg = params->loadmap->segs;
992 
993 	/* deal with each load segment separately */
994 	phdr = params->phdrs;
995 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
996 		unsigned long maddr, disp, excess, excess1;
997 		int prot = 0, flags;
998 
999 		if (phdr->p_type != PT_LOAD)
1000 			continue;
1001 
1002 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1003 		       (unsigned long) phdr->p_vaddr,
1004 		       (unsigned long) phdr->p_offset,
1005 		       (unsigned long) phdr->p_filesz,
1006 		       (unsigned long) phdr->p_memsz);
1007 
1008 		/* determine the mapping parameters */
1009 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1010 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1011 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1012 
1013 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1014 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1015 			flags |= MAP_EXECUTABLE;
1016 
1017 		maddr = 0;
1018 
1019 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1020 		case ELF_FDPIC_FLAG_INDEPENDENT:
1021 			/* PT_LOADs are independently locatable */
1022 			break;
1023 
1024 		case ELF_FDPIC_FLAG_HONOURVADDR:
1025 			/* the specified virtual address must be honoured */
1026 			maddr = phdr->p_vaddr;
1027 			flags |= MAP_FIXED;
1028 			break;
1029 
1030 		case ELF_FDPIC_FLAG_CONSTDISP:
1031 			/* constant displacement
1032 			 * - can be mapped anywhere, but must be mapped as a
1033 			 *   unit
1034 			 */
1035 			if (!dvset) {
1036 				maddr = load_addr;
1037 				delta_vaddr = phdr->p_vaddr;
1038 				dvset = 1;
1039 			} else {
1040 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1041 				flags |= MAP_FIXED;
1042 			}
1043 			break;
1044 
1045 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1046 			/* contiguity handled later */
1047 			break;
1048 
1049 		default:
1050 			BUG();
1051 		}
1052 
1053 		maddr &= PAGE_MASK;
1054 
1055 		/* create the mapping */
1056 		disp = phdr->p_vaddr & ~PAGE_MASK;
1057 		down_write(&mm->mmap_sem);
1058 		maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1059 				phdr->p_offset - disp);
1060 		up_write(&mm->mmap_sem);
1061 
1062 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1063 		       loop, phdr->p_memsz + disp, prot, flags,
1064 		       phdr->p_offset - disp, maddr);
1065 
1066 		if (IS_ERR_VALUE(maddr))
1067 			return (int) maddr;
1068 
1069 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1070 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1071 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1072 
1073 		seg->addr = maddr + disp;
1074 		seg->p_vaddr = phdr->p_vaddr;
1075 		seg->p_memsz = phdr->p_memsz;
1076 
1077 		/* map the ELF header address if in this segment */
1078 		if (phdr->p_offset == 0)
1079 			params->elfhdr_addr = seg->addr;
1080 
1081 		/* clear the bit between beginning of mapping and beginning of
1082 		 * PT_LOAD */
1083 		if (prot & PROT_WRITE && disp > 0) {
1084 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1085 			clear_user((void __user *) maddr, disp);
1086 			maddr += disp;
1087 		}
1088 
1089 		/* clear any space allocated but not loaded
1090 		 * - on uClinux we can just clear the lot
1091 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1092 		 *   extant in the file
1093 		 */
1094 		excess = phdr->p_memsz - phdr->p_filesz;
1095 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1096 
1097 #ifdef CONFIG_MMU
1098 		if (excess > excess1) {
1099 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1100 			unsigned long xmaddr;
1101 
1102 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1103 			down_write(&mm->mmap_sem);
1104 			xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1105 					 prot, flags, 0);
1106 			up_write(&mm->mmap_sem);
1107 
1108 			kdebug("mmap[%d] <anon>"
1109 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1110 			       loop, xaddr, excess - excess1, prot, flags,
1111 			       xmaddr);
1112 
1113 			if (xmaddr != xaddr)
1114 				return -ENOMEM;
1115 		}
1116 
1117 		if (prot & PROT_WRITE && excess1 > 0) {
1118 			kdebug("clear[%d] ad=%lx sz=%lx",
1119 			       loop, maddr + phdr->p_filesz, excess1);
1120 			clear_user((void __user *) maddr + phdr->p_filesz,
1121 				   excess1);
1122 		}
1123 
1124 #else
1125 		if (excess > 0) {
1126 			kdebug("clear[%d] ad=%lx sz=%lx",
1127 			       loop, maddr + phdr->p_filesz, excess);
1128 			clear_user((void *) maddr + phdr->p_filesz, excess);
1129 		}
1130 #endif
1131 
1132 		if (mm) {
1133 			if (phdr->p_flags & PF_X) {
1134 				if (!mm->start_code) {
1135 					mm->start_code = maddr;
1136 					mm->end_code = maddr + phdr->p_memsz;
1137 				}
1138 			} else if (!mm->start_data) {
1139 				mm->start_data = maddr;
1140 				mm->end_data = maddr + phdr->p_memsz;
1141 			}
1142 		}
1143 
1144 		seg++;
1145 	}
1146 
1147 	return 0;
1148 }
1149 
1150 /*****************************************************************************/
1151 /*
1152  * ELF-FDPIC core dumper
1153  *
1154  * Modelled on fs/exec.c:aout_core_dump()
1155  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1156  *
1157  * Modelled on fs/binfmt_elf.c core dumper
1158  */
1159 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1160 
1161 /*
1162  * These are the only things you should do on a core-file: use only these
1163  * functions to write out all the necessary info.
1164  */
1165 static int dump_write(struct file *file, const void *addr, int nr)
1166 {
1167 	return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1168 }
1169 
1170 static int dump_seek(struct file *file, loff_t off)
1171 {
1172 	if (file->f_op->llseek) {
1173 		if (file->f_op->llseek(file, off, SEEK_SET) != off)
1174 			return 0;
1175 	} else {
1176 		file->f_pos = off;
1177 	}
1178 	return 1;
1179 }
1180 
1181 /*
1182  * Decide whether a segment is worth dumping; default is yes to be
1183  * sure (missing info is worse than too much; etc).
1184  * Personally I'd include everything, and use the coredump limit...
1185  *
1186  * I think we should skip something. But I am not sure how. H.J.
1187  */
1188 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1189 {
1190 	int dump_ok;
1191 
1192 	/* Do not dump I/O mapped devices or special mappings */
1193 	if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1194 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1195 		return 0;
1196 	}
1197 
1198 	/* If we may not read the contents, don't allow us to dump
1199 	 * them either. "dump_write()" can't handle it anyway.
1200 	 */
1201 	if (!(vma->vm_flags & VM_READ)) {
1202 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1203 		return 0;
1204 	}
1205 
1206 	/* By default, dump shared memory if mapped from an anonymous file. */
1207 	if (vma->vm_flags & VM_SHARED) {
1208 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1209 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1210 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1211 			       vma->vm_flags, dump_ok ? "yes" : "no");
1212 			return dump_ok;
1213 		}
1214 
1215 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1216 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1217 		       vma->vm_flags, dump_ok ? "yes" : "no");
1218 		return dump_ok;
1219 	}
1220 
1221 #ifdef CONFIG_MMU
1222 	/* By default, if it hasn't been written to, don't write it out */
1223 	if (!vma->anon_vma) {
1224 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1225 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1226 		       vma->vm_flags, dump_ok ? "yes" : "no");
1227 		return dump_ok;
1228 	}
1229 #endif
1230 
1231 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1232 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1233 	       dump_ok ? "yes" : "no");
1234 	return dump_ok;
1235 }
1236 
1237 /* An ELF note in memory */
1238 struct memelfnote
1239 {
1240 	const char *name;
1241 	int type;
1242 	unsigned int datasz;
1243 	void *data;
1244 };
1245 
1246 static int notesize(struct memelfnote *en)
1247 {
1248 	int sz;
1249 
1250 	sz = sizeof(struct elf_note);
1251 	sz += roundup(strlen(en->name) + 1, 4);
1252 	sz += roundup(en->datasz, 4);
1253 
1254 	return sz;
1255 }
1256 
1257 /* #define DEBUG */
1258 
1259 #define DUMP_WRITE(addr, nr)	\
1260 	do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1261 #define DUMP_SEEK(off)	\
1262 	do { if (!dump_seek(file, (off))) return 0; } while(0)
1263 
1264 static int writenote(struct memelfnote *men, struct file *file)
1265 {
1266 	struct elf_note en;
1267 
1268 	en.n_namesz = strlen(men->name) + 1;
1269 	en.n_descsz = men->datasz;
1270 	en.n_type = men->type;
1271 
1272 	DUMP_WRITE(&en, sizeof(en));
1273 	DUMP_WRITE(men->name, en.n_namesz);
1274 	/* XXX - cast from long long to long to avoid need for libgcc.a */
1275 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1276 	DUMP_WRITE(men->data, men->datasz);
1277 	DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));	/* XXX */
1278 
1279 	return 1;
1280 }
1281 #undef DUMP_WRITE
1282 #undef DUMP_SEEK
1283 
1284 #define DUMP_WRITE(addr, nr)	\
1285 	if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1286 		goto end_coredump;
1287 #define DUMP_SEEK(off)	\
1288 	if (!dump_seek(file, (off))) \
1289 		goto end_coredump;
1290 
1291 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1292 {
1293 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1294 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1295 	elf->e_ident[EI_DATA] = ELF_DATA;
1296 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1297 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1298 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1299 
1300 	elf->e_type = ET_CORE;
1301 	elf->e_machine = ELF_ARCH;
1302 	elf->e_version = EV_CURRENT;
1303 	elf->e_entry = 0;
1304 	elf->e_phoff = sizeof(struct elfhdr);
1305 	elf->e_shoff = 0;
1306 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1307 	elf->e_ehsize = sizeof(struct elfhdr);
1308 	elf->e_phentsize = sizeof(struct elf_phdr);
1309 	elf->e_phnum = segs;
1310 	elf->e_shentsize = 0;
1311 	elf->e_shnum = 0;
1312 	elf->e_shstrndx = 0;
1313 	return;
1314 }
1315 
1316 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1317 {
1318 	phdr->p_type = PT_NOTE;
1319 	phdr->p_offset = offset;
1320 	phdr->p_vaddr = 0;
1321 	phdr->p_paddr = 0;
1322 	phdr->p_filesz = sz;
1323 	phdr->p_memsz = 0;
1324 	phdr->p_flags = 0;
1325 	phdr->p_align = 0;
1326 	return;
1327 }
1328 
1329 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1330 		unsigned int sz, void *data)
1331 {
1332 	note->name = name;
1333 	note->type = type;
1334 	note->datasz = sz;
1335 	note->data = data;
1336 	return;
1337 }
1338 
1339 /*
1340  * fill up all the fields in prstatus from the given task struct, except
1341  * registers which need to be filled up seperately.
1342  */
1343 static void fill_prstatus(struct elf_prstatus *prstatus,
1344 			  struct task_struct *p, long signr)
1345 {
1346 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1347 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1348 	prstatus->pr_sighold = p->blocked.sig[0];
1349 	prstatus->pr_pid = task_pid_vnr(p);
1350 	prstatus->pr_ppid = task_pid_vnr(p->parent);
1351 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1352 	prstatus->pr_sid = task_session_vnr(p);
1353 	if (thread_group_leader(p)) {
1354 		/*
1355 		 * This is the record for the group leader.  Add in the
1356 		 * cumulative times of previous dead threads.  This total
1357 		 * won't include the time of each live thread whose state
1358 		 * is included in the core dump.  The final total reported
1359 		 * to our parent process when it calls wait4 will include
1360 		 * those sums as well as the little bit more time it takes
1361 		 * this and each other thread to finish dying after the
1362 		 * core dump synchronization phase.
1363 		 */
1364 		cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1365 				   &prstatus->pr_utime);
1366 		cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1367 				   &prstatus->pr_stime);
1368 	} else {
1369 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1370 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1371 	}
1372 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1373 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1374 
1375 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1376 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1377 }
1378 
1379 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1380 		       struct mm_struct *mm)
1381 {
1382 	unsigned int i, len;
1383 
1384 	/* first copy the parameters from user space */
1385 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1386 
1387 	len = mm->arg_end - mm->arg_start;
1388 	if (len >= ELF_PRARGSZ)
1389 		len = ELF_PRARGSZ - 1;
1390 	if (copy_from_user(&psinfo->pr_psargs,
1391 		           (const char __user *) mm->arg_start, len))
1392 		return -EFAULT;
1393 	for (i = 0; i < len; i++)
1394 		if (psinfo->pr_psargs[i] == 0)
1395 			psinfo->pr_psargs[i] = ' ';
1396 	psinfo->pr_psargs[len] = 0;
1397 
1398 	psinfo->pr_pid = task_pid_vnr(p);
1399 	psinfo->pr_ppid = task_pid_vnr(p->parent);
1400 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1401 	psinfo->pr_sid = task_session_vnr(p);
1402 
1403 	i = p->state ? ffz(~p->state) + 1 : 0;
1404 	psinfo->pr_state = i;
1405 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1406 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1407 	psinfo->pr_nice = task_nice(p);
1408 	psinfo->pr_flag = p->flags;
1409 	SET_UID(psinfo->pr_uid, p->uid);
1410 	SET_GID(psinfo->pr_gid, p->gid);
1411 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1412 
1413 	return 0;
1414 }
1415 
1416 /* Here is the structure in which status of each thread is captured. */
1417 struct elf_thread_status
1418 {
1419 	struct list_head list;
1420 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1421 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1422 	struct task_struct *thread;
1423 #ifdef ELF_CORE_COPY_XFPREGS
1424 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1425 #endif
1426 	struct memelfnote notes[3];
1427 	int num_notes;
1428 };
1429 
1430 /*
1431  * In order to add the specific thread information for the elf file format,
1432  * we need to keep a linked list of every thread's pr_status and then create
1433  * a single section for them in the final core file.
1434  */
1435 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1436 {
1437 	struct task_struct *p = t->thread;
1438 	int sz = 0;
1439 
1440 	t->num_notes = 0;
1441 
1442 	fill_prstatus(&t->prstatus, p, signr);
1443 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1444 
1445 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1446 		  &t->prstatus);
1447 	t->num_notes++;
1448 	sz += notesize(&t->notes[0]);
1449 
1450 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1451 	if (t->prstatus.pr_fpvalid) {
1452 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1453 			  &t->fpu);
1454 		t->num_notes++;
1455 		sz += notesize(&t->notes[1]);
1456 	}
1457 
1458 #ifdef ELF_CORE_COPY_XFPREGS
1459 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1460 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1461 			  sizeof(t->xfpu), &t->xfpu);
1462 		t->num_notes++;
1463 		sz += notesize(&t->notes[2]);
1464 	}
1465 #endif
1466 	return sz;
1467 }
1468 
1469 /*
1470  * dump the segments for an MMU process
1471  */
1472 #ifdef CONFIG_MMU
1473 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1474 			   unsigned long *limit, unsigned long mm_flags)
1475 {
1476 	struct vm_area_struct *vma;
1477 
1478 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1479 		unsigned long addr;
1480 
1481 		if (!maydump(vma, mm_flags))
1482 			continue;
1483 
1484 		for (addr = vma->vm_start;
1485 		     addr < vma->vm_end;
1486 		     addr += PAGE_SIZE
1487 		     ) {
1488 			struct vm_area_struct *vma;
1489 			struct page *page;
1490 
1491 			if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1492 					   &page, &vma) <= 0) {
1493 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1494 			}
1495 			else if (page == ZERO_PAGE(0)) {
1496 				page_cache_release(page);
1497 				DUMP_SEEK(file->f_pos + PAGE_SIZE);
1498 			}
1499 			else {
1500 				void *kaddr;
1501 
1502 				flush_cache_page(vma, addr, page_to_pfn(page));
1503 				kaddr = kmap(page);
1504 				if ((*size += PAGE_SIZE) > *limit ||
1505 				    !dump_write(file, kaddr, PAGE_SIZE)
1506 				    ) {
1507 					kunmap(page);
1508 					page_cache_release(page);
1509 					return -EIO;
1510 				}
1511 				kunmap(page);
1512 				page_cache_release(page);
1513 			}
1514 		}
1515 	}
1516 
1517 	return 0;
1518 
1519 end_coredump:
1520 	return -EFBIG;
1521 }
1522 #endif
1523 
1524 /*
1525  * dump the segments for a NOMMU process
1526  */
1527 #ifndef CONFIG_MMU
1528 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1529 			   unsigned long *limit, unsigned long mm_flags)
1530 {
1531 	struct vm_list_struct *vml;
1532 
1533 	for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1534 	struct vm_area_struct *vma = vml->vma;
1535 
1536 		if (!maydump(vma, mm_flags))
1537 			continue;
1538 
1539 		if ((*size += PAGE_SIZE) > *limit)
1540 			return -EFBIG;
1541 
1542 		if (!dump_write(file, (void *) vma->vm_start,
1543 				vma->vm_end - vma->vm_start))
1544 			return -EIO;
1545 	}
1546 
1547 	return 0;
1548 }
1549 #endif
1550 
1551 /*
1552  * Actual dumper
1553  *
1554  * This is a two-pass process; first we find the offsets of the bits,
1555  * and then they are actually written out.  If we run out of core limit
1556  * we just truncate.
1557  */
1558 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1559 			       struct file *file, unsigned long limit)
1560 {
1561 #define	NUM_NOTES	6
1562 	int has_dumped = 0;
1563 	mm_segment_t fs;
1564 	int segs;
1565 	size_t size = 0;
1566 	int i;
1567 	struct vm_area_struct *vma;
1568 	struct elfhdr *elf = NULL;
1569 	loff_t offset = 0, dataoff;
1570 	int numnote;
1571 	struct memelfnote *notes = NULL;
1572 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1573 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1574  	LIST_HEAD(thread_list);
1575  	struct list_head *t;
1576 	elf_fpregset_t *fpu = NULL;
1577 #ifdef ELF_CORE_COPY_XFPREGS
1578 	elf_fpxregset_t *xfpu = NULL;
1579 #endif
1580 	int thread_status_size = 0;
1581 #ifndef CONFIG_MMU
1582 	struct vm_list_struct *vml;
1583 #endif
1584 	elf_addr_t *auxv;
1585 	unsigned long mm_flags;
1586 
1587 	/*
1588 	 * We no longer stop all VM operations.
1589 	 *
1590 	 * This is because those proceses that could possibly change map_count
1591 	 * or the mmap / vma pages are now blocked in do_exit on current
1592 	 * finishing this core dump.
1593 	 *
1594 	 * Only ptrace can touch these memory addresses, but it doesn't change
1595 	 * the map_count or the pages allocated. So no possibility of crashing
1596 	 * exists while dumping the mm->vm_next areas to the core file.
1597 	 */
1598 
1599 	/* alloc memory for large data structures: too large to be on stack */
1600 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1601 	if (!elf)
1602 		goto cleanup;
1603 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1604 	if (!prstatus)
1605 		goto cleanup;
1606 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1607 	if (!psinfo)
1608 		goto cleanup;
1609 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1610 	if (!notes)
1611 		goto cleanup;
1612 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1613 	if (!fpu)
1614 		goto cleanup;
1615 #ifdef ELF_CORE_COPY_XFPREGS
1616 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1617 	if (!xfpu)
1618 		goto cleanup;
1619 #endif
1620 
1621 	if (signr) {
1622 		struct core_thread *ct;
1623 		struct elf_thread_status *tmp;
1624 
1625 		for (ct = current->mm->core_state->dumper.next;
1626 						ct; ct = ct->next) {
1627 			tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1628 			if (!tmp)
1629 				goto cleanup;
1630 
1631 			tmp->thread = ct->task;
1632 			list_add(&tmp->list, &thread_list);
1633 		}
1634 
1635 		list_for_each(t, &thread_list) {
1636 			struct elf_thread_status *tmp;
1637 			int sz;
1638 
1639 			tmp = list_entry(t, struct elf_thread_status, list);
1640 			sz = elf_dump_thread_status(signr, tmp);
1641 			thread_status_size += sz;
1642 		}
1643 	}
1644 
1645 	/* now collect the dump for the current */
1646 	fill_prstatus(prstatus, current, signr);
1647 	elf_core_copy_regs(&prstatus->pr_reg, regs);
1648 
1649 #ifdef CONFIG_MMU
1650 	segs = current->mm->map_count;
1651 #else
1652 	segs = 0;
1653 	for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1654 	    segs++;
1655 #endif
1656 #ifdef ELF_CORE_EXTRA_PHDRS
1657 	segs += ELF_CORE_EXTRA_PHDRS;
1658 #endif
1659 
1660 	/* Set up header */
1661 	fill_elf_fdpic_header(elf, segs + 1);	/* including notes section */
1662 
1663 	has_dumped = 1;
1664 	current->flags |= PF_DUMPCORE;
1665 
1666 	/*
1667 	 * Set up the notes in similar form to SVR4 core dumps made
1668 	 * with info from their /proc.
1669 	 */
1670 
1671 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1672 	fill_psinfo(psinfo, current->group_leader, current->mm);
1673 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1674 
1675 	numnote = 2;
1676 
1677 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1678 
1679 	i = 0;
1680 	do
1681 		i += 2;
1682 	while (auxv[i - 2] != AT_NULL);
1683 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1684 		  i * sizeof(elf_addr_t), auxv);
1685 
1686   	/* Try to dump the FPU. */
1687 	if ((prstatus->pr_fpvalid =
1688 	     elf_core_copy_task_fpregs(current, regs, fpu)))
1689 		fill_note(notes + numnote++,
1690 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1691 #ifdef ELF_CORE_COPY_XFPREGS
1692 	if (elf_core_copy_task_xfpregs(current, xfpu))
1693 		fill_note(notes + numnote++,
1694 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1695 #endif
1696 
1697 	fs = get_fs();
1698 	set_fs(KERNEL_DS);
1699 
1700 	DUMP_WRITE(elf, sizeof(*elf));
1701 	offset += sizeof(*elf);				/* Elf header */
1702 	offset += (segs+1) * sizeof(struct elf_phdr);	/* Program headers */
1703 
1704 	/* Write notes phdr entry */
1705 	{
1706 		struct elf_phdr phdr;
1707 		int sz = 0;
1708 
1709 		for (i = 0; i < numnote; i++)
1710 			sz += notesize(notes + i);
1711 
1712 		sz += thread_status_size;
1713 
1714 		fill_elf_note_phdr(&phdr, sz, offset);
1715 		offset += sz;
1716 		DUMP_WRITE(&phdr, sizeof(phdr));
1717 	}
1718 
1719 	/* Page-align dumped data */
1720 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1721 
1722 	/*
1723 	 * We must use the same mm->flags while dumping core to avoid
1724 	 * inconsistency between the program headers and bodies, otherwise an
1725 	 * unusable core file can be generated.
1726 	 */
1727 	mm_flags = current->mm->flags;
1728 
1729 	/* write program headers for segments dump */
1730 	for (
1731 #ifdef CONFIG_MMU
1732 		vma = current->mm->mmap; vma; vma = vma->vm_next
1733 #else
1734 			vml = current->mm->context.vmlist; vml; vml = vml->next
1735 #endif
1736 	     ) {
1737 		struct elf_phdr phdr;
1738 		size_t sz;
1739 
1740 #ifndef CONFIG_MMU
1741 		vma = vml->vma;
1742 #endif
1743 
1744 		sz = vma->vm_end - vma->vm_start;
1745 
1746 		phdr.p_type = PT_LOAD;
1747 		phdr.p_offset = offset;
1748 		phdr.p_vaddr = vma->vm_start;
1749 		phdr.p_paddr = 0;
1750 		phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1751 		phdr.p_memsz = sz;
1752 		offset += phdr.p_filesz;
1753 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1754 		if (vma->vm_flags & VM_WRITE)
1755 			phdr.p_flags |= PF_W;
1756 		if (vma->vm_flags & VM_EXEC)
1757 			phdr.p_flags |= PF_X;
1758 		phdr.p_align = ELF_EXEC_PAGESIZE;
1759 
1760 		DUMP_WRITE(&phdr, sizeof(phdr));
1761 	}
1762 
1763 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1764 	ELF_CORE_WRITE_EXTRA_PHDRS;
1765 #endif
1766 
1767  	/* write out the notes section */
1768 	for (i = 0; i < numnote; i++)
1769 		if (!writenote(notes + i, file))
1770 			goto end_coredump;
1771 
1772 	/* write out the thread status notes section */
1773 	list_for_each(t, &thread_list) {
1774 		struct elf_thread_status *tmp =
1775 				list_entry(t, struct elf_thread_status, list);
1776 
1777 		for (i = 0; i < tmp->num_notes; i++)
1778 			if (!writenote(&tmp->notes[i], file))
1779 				goto end_coredump;
1780 	}
1781 
1782 	DUMP_SEEK(dataoff);
1783 
1784 	if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1785 		goto end_coredump;
1786 
1787 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1788 	ELF_CORE_WRITE_EXTRA_DATA;
1789 #endif
1790 
1791 	if (file->f_pos != offset) {
1792 		/* Sanity check */
1793 		printk(KERN_WARNING
1794 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1795 		       file->f_pos, offset);
1796 	}
1797 
1798 end_coredump:
1799 	set_fs(fs);
1800 
1801 cleanup:
1802 	while (!list_empty(&thread_list)) {
1803 		struct list_head *tmp = thread_list.next;
1804 		list_del(tmp);
1805 		kfree(list_entry(tmp, struct elf_thread_status, list));
1806 	}
1807 
1808 	kfree(elf);
1809 	kfree(prstatus);
1810 	kfree(psinfo);
1811 	kfree(notes);
1812 	kfree(fpu);
1813 #ifdef ELF_CORE_COPY_XFPREGS
1814 	kfree(xfpu);
1815 #endif
1816 	return has_dumped;
1817 #undef NUM_NOTES
1818 }
1819 
1820 #endif		/* USE_ELF_CORE_DUMP */
1821