• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/fs/binfmt_elf.c
4  *
5  * These are the functions used to load ELF format executables as used
6  * on SVr4 machines.  Information on the format may be found in the book
7  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
8  * Tools".
9  *
10  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
11  */
12 
13 #include <linux/module.h>
14 #include <linux/kernel.h>
15 #include <linux/fs.h>
16 #include <linux/log2.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/errno.h>
20 #include <linux/signal.h>
21 #include <linux/binfmts.h>
22 #include <linux/string.h>
23 #include <linux/file.h>
24 #include <linux/slab.h>
25 #include <linux/personality.h>
26 #include <linux/elfcore.h>
27 #include <linux/init.h>
28 #include <linux/highuid.h>
29 #include <linux/compiler.h>
30 #include <linux/highmem.h>
31 #include <linux/hugetlb.h>
32 #include <linux/pagemap.h>
33 #include <linux/vmalloc.h>
34 #include <linux/security.h>
35 #include <linux/random.h>
36 #include <linux/elf.h>
37 #include <linux/elf-randomize.h>
38 #include <linux/utsname.h>
39 #include <linux/coredump.h>
40 #include <linux/sched.h>
41 #include <linux/sched/coredump.h>
42 #include <linux/sched/task_stack.h>
43 #include <linux/sched/cputime.h>
44 #include <linux/sizes.h>
45 #include <linux/types.h>
46 #include <linux/cred.h>
47 #include <linux/dax.h>
48 #include <linux/uaccess.h>
49 #include <asm/param.h>
50 #include <asm/page.h>
51 
52 #ifndef ELF_COMPAT
53 #define ELF_COMPAT 0
54 #endif
55 
56 #ifndef user_long_t
57 #define user_long_t long
58 #endif
59 #ifndef user_siginfo_t
60 #define user_siginfo_t siginfo_t
61 #endif
62 
63 /* That's for binfmt_elf_fdpic to deal with */
64 #ifndef elf_check_fdpic
65 #define elf_check_fdpic(ex) false
66 #endif
67 
68 static int load_elf_binary(struct linux_binprm *bprm);
69 
70 #ifdef CONFIG_USELIB
71 static int load_elf_library(struct file *);
72 #else
73 #define load_elf_library NULL
74 #endif
75 
76 /*
77  * If we don't support core dumping, then supply a NULL so we
78  * don't even try.
79  */
80 #ifdef CONFIG_ELF_CORE
81 static int elf_core_dump(struct coredump_params *cprm);
82 #else
83 #define elf_core_dump	NULL
84 #endif
85 
86 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
87 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
88 #else
89 #define ELF_MIN_ALIGN	PAGE_SIZE
90 #endif
91 
92 #ifndef ELF_CORE_EFLAGS
93 #define ELF_CORE_EFLAGS	0
94 #endif
95 
96 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
97 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
98 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
99 
100 static struct linux_binfmt elf_format = {
101 	.module		= THIS_MODULE,
102 	.load_binary	= load_elf_binary,
103 	.load_shlib	= load_elf_library,
104 	.core_dump	= elf_core_dump,
105 	.min_coredump	= ELF_EXEC_PAGESIZE,
106 };
107 
108 #define BAD_ADDR(x) (unlikely((unsigned long)(x) >= TASK_SIZE))
109 
set_brk(unsigned long start,unsigned long end,int prot)110 static int set_brk(unsigned long start, unsigned long end, int prot)
111 {
112 	start = ELF_PAGEALIGN(start);
113 	end = ELF_PAGEALIGN(end);
114 	if (end > start) {
115 		/*
116 		 * Map the last of the bss segment.
117 		 * If the header is requesting these pages to be
118 		 * executable, honour that (ppc32 needs this).
119 		 */
120 		int error = vm_brk_flags(start, end - start,
121 				prot & PROT_EXEC ? VM_EXEC : 0);
122 		if (error)
123 			return error;
124 	}
125 	current->mm->start_brk = current->mm->brk = end;
126 	return 0;
127 }
128 
129 /* We need to explicitly zero any fractional pages
130    after the data section (i.e. bss).  This would
131    contain the junk from the file that should not
132    be in memory
133  */
padzero(unsigned long elf_bss)134 static int padzero(unsigned long elf_bss)
135 {
136 	unsigned long nbyte;
137 
138 	nbyte = ELF_PAGEOFFSET(elf_bss);
139 	if (nbyte) {
140 		nbyte = ELF_MIN_ALIGN - nbyte;
141 		if (clear_user((void __user *) elf_bss, nbyte))
142 			return -EFAULT;
143 	}
144 	return 0;
145 }
146 
147 /* Let's use some macros to make this stack manipulation a little clearer */
148 #ifdef CONFIG_STACK_GROWSUP
149 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
150 #define STACK_ROUND(sp, items) \
151 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
152 #define STACK_ALLOC(sp, len) ({ \
153 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
154 	old_sp; })
155 #else
156 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
157 #define STACK_ROUND(sp, items) \
158 	(((unsigned long) (sp - items)) &~ 15UL)
159 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
160 #endif
161 
162 #ifndef ELF_BASE_PLATFORM
163 /*
164  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
165  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
166  * will be copied to the user stack in the same manner as AT_PLATFORM.
167  */
168 #define ELF_BASE_PLATFORM NULL
169 #endif
170 
171 static int
create_elf_tables(struct linux_binprm * bprm,const struct elfhdr * exec,unsigned long load_addr,unsigned long interp_load_addr,unsigned long e_entry)172 create_elf_tables(struct linux_binprm *bprm, const struct elfhdr *exec,
173 		unsigned long load_addr, unsigned long interp_load_addr,
174 		unsigned long e_entry)
175 {
176 	struct mm_struct *mm = current->mm;
177 	unsigned long p = bprm->p;
178 	int argc = bprm->argc;
179 	int envc = bprm->envc;
180 	elf_addr_t __user *sp;
181 	elf_addr_t __user *u_platform;
182 	elf_addr_t __user *u_base_platform;
183 	elf_addr_t __user *u_rand_bytes;
184 	const char *k_platform = ELF_PLATFORM;
185 	const char *k_base_platform = ELF_BASE_PLATFORM;
186 	unsigned char k_rand_bytes[16];
187 	int items;
188 	elf_addr_t *elf_info;
189 	int ei_index;
190 	const struct cred *cred = current_cred();
191 	struct vm_area_struct *vma;
192 
193 	/*
194 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
195 	 * evictions by the processes running on the same package. One
196 	 * thing we can do is to shuffle the initial stack for them.
197 	 */
198 
199 	p = arch_align_stack(p);
200 
201 	/*
202 	 * If this architecture has a platform capability string, copy it
203 	 * to userspace.  In some cases (Sparc), this info is impossible
204 	 * for userspace to get any other way, in others (i386) it is
205 	 * merely difficult.
206 	 */
207 	u_platform = NULL;
208 	if (k_platform) {
209 		size_t len = strlen(k_platform) + 1;
210 
211 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
212 		if (copy_to_user(u_platform, k_platform, len))
213 			return -EFAULT;
214 	}
215 
216 	/*
217 	 * If this architecture has a "base" platform capability
218 	 * string, copy it to userspace.
219 	 */
220 	u_base_platform = NULL;
221 	if (k_base_platform) {
222 		size_t len = strlen(k_base_platform) + 1;
223 
224 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
225 		if (copy_to_user(u_base_platform, k_base_platform, len))
226 			return -EFAULT;
227 	}
228 
229 	/*
230 	 * Generate 16 random bytes for userspace PRNG seeding.
231 	 */
232 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
233 	u_rand_bytes = (elf_addr_t __user *)
234 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
235 	if (copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
236 		return -EFAULT;
237 
238 	/* Create the ELF interpreter info */
239 	elf_info = (elf_addr_t *)mm->saved_auxv;
240 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
241 #define NEW_AUX_ENT(id, val) \
242 	do { \
243 		*elf_info++ = id; \
244 		*elf_info++ = val; \
245 	} while (0)
246 
247 #ifdef ARCH_DLINFO
248 	/*
249 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
250 	 * AUXV.
251 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
252 	 * ARCH_DLINFO changes
253 	 */
254 	ARCH_DLINFO;
255 #endif
256 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
257 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
258 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
259 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
260 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
261 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
262 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
263 	NEW_AUX_ENT(AT_FLAGS, 0);
264 	NEW_AUX_ENT(AT_ENTRY, e_entry);
265 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
266 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
267 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
268 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
269 	NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
270 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
271 #ifdef ELF_HWCAP2
272 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
273 #endif
274 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
275 	if (k_platform) {
276 		NEW_AUX_ENT(AT_PLATFORM,
277 			    (elf_addr_t)(unsigned long)u_platform);
278 	}
279 	if (k_base_platform) {
280 		NEW_AUX_ENT(AT_BASE_PLATFORM,
281 			    (elf_addr_t)(unsigned long)u_base_platform);
282 	}
283 	if (bprm->have_execfd) {
284 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
285 	}
286 #undef NEW_AUX_ENT
287 	/* AT_NULL is zero; clear the rest too */
288 	memset(elf_info, 0, (char *)mm->saved_auxv +
289 			sizeof(mm->saved_auxv) - (char *)elf_info);
290 
291 	/* And advance past the AT_NULL entry.  */
292 	elf_info += 2;
293 
294 	ei_index = elf_info - (elf_addr_t *)mm->saved_auxv;
295 	sp = STACK_ADD(p, ei_index);
296 
297 	items = (argc + 1) + (envc + 1) + 1;
298 	bprm->p = STACK_ROUND(sp, items);
299 
300 	/* Point sp at the lowest address on the stack */
301 #ifdef CONFIG_STACK_GROWSUP
302 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
303 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
304 #else
305 	sp = (elf_addr_t __user *)bprm->p;
306 #endif
307 
308 
309 	/*
310 	 * Grow the stack manually; some architectures have a limit on how
311 	 * far ahead a user-space access may be in order to grow the stack.
312 	 */
313 	if (mmap_read_lock_killable(mm))
314 		return -EINTR;
315 	vma = find_extend_vma(mm, bprm->p);
316 	mmap_read_unlock(mm);
317 	if (!vma)
318 		return -EFAULT;
319 
320 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
321 	if (put_user(argc, sp++))
322 		return -EFAULT;
323 
324 	/* Populate list of argv pointers back to argv strings. */
325 	p = mm->arg_end = mm->arg_start;
326 	while (argc-- > 0) {
327 		size_t len;
328 		if (put_user((elf_addr_t)p, sp++))
329 			return -EFAULT;
330 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
331 		if (!len || len > MAX_ARG_STRLEN)
332 			return -EINVAL;
333 		p += len;
334 	}
335 	if (put_user(0, sp++))
336 		return -EFAULT;
337 	mm->arg_end = p;
338 
339 	/* Populate list of envp pointers back to envp strings. */
340 	mm->env_end = mm->env_start = p;
341 	while (envc-- > 0) {
342 		size_t len;
343 		if (put_user((elf_addr_t)p, sp++))
344 			return -EFAULT;
345 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
346 		if (!len || len > MAX_ARG_STRLEN)
347 			return -EINVAL;
348 		p += len;
349 	}
350 	if (put_user(0, sp++))
351 		return -EFAULT;
352 	mm->env_end = p;
353 
354 	/* Put the elf_info on the stack in the right place.  */
355 	if (copy_to_user(sp, mm->saved_auxv, ei_index * sizeof(elf_addr_t)))
356 		return -EFAULT;
357 	return 0;
358 }
359 
elf_map(struct file * filep,unsigned long addr,const struct elf_phdr * eppnt,int prot,int type,unsigned long total_size)360 static unsigned long elf_map(struct file *filep, unsigned long addr,
361 		const struct elf_phdr *eppnt, int prot, int type,
362 		unsigned long total_size)
363 {
364 	unsigned long map_addr;
365 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
366 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
367 	addr = ELF_PAGESTART(addr);
368 	size = ELF_PAGEALIGN(size);
369 
370 	/* mmap() will return -EINVAL if given a zero size, but a
371 	 * segment with zero filesize is perfectly valid */
372 	if (!size)
373 		return addr;
374 
375 	/*
376 	* total_size is the size of the ELF (interpreter) image.
377 	* The _first_ mmap needs to know the full size, otherwise
378 	* randomization might put this image into an overlapping
379 	* position with the ELF binary image. (since size < total_size)
380 	* So we first map the 'big' image - and unmap the remainder at
381 	* the end. (which unmap is needed for ELF images with holes.)
382 	*/
383 	if (total_size) {
384 		total_size = ELF_PAGEALIGN(total_size);
385 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
386 		if (!BAD_ADDR(map_addr))
387 			vm_munmap(map_addr+size, total_size-size);
388 	} else
389 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
390 
391 	if ((type & MAP_FIXED_NOREPLACE) &&
392 	    PTR_ERR((void *)map_addr) == -EEXIST)
393 		pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
394 			task_pid_nr(current), current->comm, (void *)addr);
395 
396 	return(map_addr);
397 }
398 
total_mapping_size(const struct elf_phdr * cmds,int nr)399 static unsigned long total_mapping_size(const struct elf_phdr *cmds, int nr)
400 {
401 	int i, first_idx = -1, last_idx = -1;
402 
403 	for (i = 0; i < nr; i++) {
404 		if (cmds[i].p_type == PT_LOAD) {
405 			last_idx = i;
406 			if (first_idx == -1)
407 				first_idx = i;
408 		}
409 	}
410 	if (first_idx == -1)
411 		return 0;
412 
413 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
414 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
415 }
416 
elf_read(struct file * file,void * buf,size_t len,loff_t pos)417 static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
418 {
419 	ssize_t rv;
420 
421 	rv = kernel_read(file, buf, len, &pos);
422 	if (unlikely(rv != len)) {
423 		return (rv < 0) ? rv : -EIO;
424 	}
425 	return 0;
426 }
427 
maximum_alignment(struct elf_phdr * cmds,int nr)428 static unsigned long maximum_alignment(struct elf_phdr *cmds, int nr)
429 {
430 	unsigned long alignment = 0;
431 	int i;
432 
433 	for (i = 0; i < nr; i++) {
434 		if (cmds[i].p_type == PT_LOAD) {
435 			unsigned long p_align = cmds[i].p_align;
436 
437 			/* skip non-power of two alignments as invalid */
438 			if (!is_power_of_2(p_align))
439 				continue;
440 			alignment = max(alignment, p_align);
441 		}
442 	}
443 
444 	/* ensure we align to at least one page */
445 	return ELF_PAGEALIGN(alignment);
446 }
447 
448 /**
449  * load_elf_phdrs() - load ELF program headers
450  * @elf_ex:   ELF header of the binary whose program headers should be loaded
451  * @elf_file: the opened ELF binary file
452  *
453  * Loads ELF program headers from the binary file elf_file, which has the ELF
454  * header pointed to by elf_ex, into a newly allocated array. The caller is
455  * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
456  */
load_elf_phdrs(const struct elfhdr * elf_ex,struct file * elf_file)457 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
458 				       struct file *elf_file)
459 {
460 	struct elf_phdr *elf_phdata = NULL;
461 	int retval, err = -1;
462 	unsigned int size;
463 
464 	/*
465 	 * If the size of this structure has changed, then punt, since
466 	 * we will be doing the wrong thing.
467 	 */
468 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
469 		goto out;
470 
471 	/* Sanity check the number of program headers... */
472 	/* ...and their total size. */
473 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
474 	if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
475 		goto out;
476 
477 	elf_phdata = kmalloc(size, GFP_KERNEL);
478 	if (!elf_phdata)
479 		goto out;
480 
481 	/* Read in the program headers */
482 	retval = elf_read(elf_file, elf_phdata, size, elf_ex->e_phoff);
483 	if (retval < 0) {
484 		err = retval;
485 		goto out;
486 	}
487 
488 	/* Success! */
489 	err = 0;
490 out:
491 	if (err) {
492 		kfree(elf_phdata);
493 		elf_phdata = NULL;
494 	}
495 	return elf_phdata;
496 }
497 
498 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
499 
500 /**
501  * struct arch_elf_state - arch-specific ELF loading state
502  *
503  * This structure is used to preserve architecture specific data during
504  * the loading of an ELF file, throughout the checking of architecture
505  * specific ELF headers & through to the point where the ELF load is
506  * known to be proceeding (ie. SET_PERSONALITY).
507  *
508  * This implementation is a dummy for architectures which require no
509  * specific state.
510  */
511 struct arch_elf_state {
512 };
513 
514 #define INIT_ARCH_ELF_STATE {}
515 
516 /**
517  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
518  * @ehdr:	The main ELF header
519  * @phdr:	The program header to check
520  * @elf:	The open ELF file
521  * @is_interp:	True if the phdr is from the interpreter of the ELF being
522  *		loaded, else false.
523  * @state:	Architecture-specific state preserved throughout the process
524  *		of loading the ELF.
525  *
526  * Inspects the program header phdr to validate its correctness and/or
527  * suitability for the system. Called once per ELF program header in the
528  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
529  * interpreter.
530  *
531  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
532  *         with that return code.
533  */
arch_elf_pt_proc(struct elfhdr * ehdr,struct elf_phdr * phdr,struct file * elf,bool is_interp,struct arch_elf_state * state)534 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
535 				   struct elf_phdr *phdr,
536 				   struct file *elf, bool is_interp,
537 				   struct arch_elf_state *state)
538 {
539 	/* Dummy implementation, always proceed */
540 	return 0;
541 }
542 
543 /**
544  * arch_check_elf() - check an ELF executable
545  * @ehdr:	The main ELF header
546  * @has_interp:	True if the ELF has an interpreter, else false.
547  * @interp_ehdr: The interpreter's ELF header
548  * @state:	Architecture-specific state preserved throughout the process
549  *		of loading the ELF.
550  *
551  * Provides a final opportunity for architecture code to reject the loading
552  * of the ELF & cause an exec syscall to return an error. This is called after
553  * all program headers to be checked by arch_elf_pt_proc have been.
554  *
555  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
556  *         with that return code.
557  */
arch_check_elf(struct elfhdr * ehdr,bool has_interp,struct elfhdr * interp_ehdr,struct arch_elf_state * state)558 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
559 				 struct elfhdr *interp_ehdr,
560 				 struct arch_elf_state *state)
561 {
562 	/* Dummy implementation, always proceed */
563 	return 0;
564 }
565 
566 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
567 
make_prot(u32 p_flags,struct arch_elf_state * arch_state,bool has_interp,bool is_interp)568 static inline int make_prot(u32 p_flags, struct arch_elf_state *arch_state,
569 			    bool has_interp, bool is_interp)
570 {
571 	int prot = 0;
572 
573 	if (p_flags & PF_R)
574 		prot |= PROT_READ;
575 	if (p_flags & PF_W)
576 		prot |= PROT_WRITE;
577 	if (p_flags & PF_X)
578 		prot |= PROT_EXEC;
579 
580 	return arch_elf_adjust_prot(prot, arch_state, has_interp, is_interp);
581 }
582 
583 /* This is much more generalized than the library routine read function,
584    so we keep this separate.  Technically the library read function
585    is only provided so that we can read a.out libraries that have
586    an ELF header */
587 
load_elf_interp(struct elfhdr * interp_elf_ex,struct file * interpreter,unsigned long no_base,struct elf_phdr * interp_elf_phdata,struct arch_elf_state * arch_state)588 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
589 		struct file *interpreter,
590 		unsigned long no_base, struct elf_phdr *interp_elf_phdata,
591 		struct arch_elf_state *arch_state)
592 {
593 	struct elf_phdr *eppnt;
594 	unsigned long load_addr = 0;
595 	int load_addr_set = 0;
596 	unsigned long last_bss = 0, elf_bss = 0;
597 	int bss_prot = 0;
598 	unsigned long error = ~0UL;
599 	unsigned long total_size;
600 	int i;
601 
602 	/* First of all, some simple consistency checks */
603 	if (interp_elf_ex->e_type != ET_EXEC &&
604 	    interp_elf_ex->e_type != ET_DYN)
605 		goto out;
606 	if (!elf_check_arch(interp_elf_ex) ||
607 	    elf_check_fdpic(interp_elf_ex))
608 		goto out;
609 	if (!interpreter->f_op->mmap)
610 		goto out;
611 
612 	total_size = total_mapping_size(interp_elf_phdata,
613 					interp_elf_ex->e_phnum);
614 	if (!total_size) {
615 		error = -EINVAL;
616 		goto out;
617 	}
618 
619 	eppnt = interp_elf_phdata;
620 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
621 		if (eppnt->p_type == PT_LOAD) {
622 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
623 			int elf_prot = make_prot(eppnt->p_flags, arch_state,
624 						 true, true);
625 			unsigned long vaddr = 0;
626 			unsigned long k, map_addr;
627 
628 			vaddr = eppnt->p_vaddr;
629 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
630 				elf_type |= MAP_FIXED;
631 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
632 				load_addr = -vaddr;
633 
634 			map_addr = elf_map(interpreter, load_addr + vaddr,
635 					eppnt, elf_prot, elf_type, total_size);
636 			total_size = 0;
637 			error = map_addr;
638 			if (BAD_ADDR(map_addr))
639 				goto out;
640 
641 			if (!load_addr_set &&
642 			    interp_elf_ex->e_type == ET_DYN) {
643 				load_addr = map_addr - ELF_PAGESTART(vaddr);
644 				load_addr_set = 1;
645 			}
646 
647 			/*
648 			 * Check to see if the section's size will overflow the
649 			 * allowed task size. Note that p_filesz must always be
650 			 * <= p_memsize so it's only necessary to check p_memsz.
651 			 */
652 			k = load_addr + eppnt->p_vaddr;
653 			if (BAD_ADDR(k) ||
654 			    eppnt->p_filesz > eppnt->p_memsz ||
655 			    eppnt->p_memsz > TASK_SIZE ||
656 			    TASK_SIZE - eppnt->p_memsz < k) {
657 				error = -ENOMEM;
658 				goto out;
659 			}
660 
661 			/*
662 			 * Find the end of the file mapping for this phdr, and
663 			 * keep track of the largest address we see for this.
664 			 */
665 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
666 			if (k > elf_bss)
667 				elf_bss = k;
668 
669 			/*
670 			 * Do the same thing for the memory mapping - between
671 			 * elf_bss and last_bss is the bss section.
672 			 */
673 			k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
674 			if (k > last_bss) {
675 				last_bss = k;
676 				bss_prot = elf_prot;
677 			}
678 		}
679 	}
680 
681 	/*
682 	 * Now fill out the bss section: first pad the last page from
683 	 * the file up to the page boundary, and zero it from elf_bss
684 	 * up to the end of the page.
685 	 */
686 	if (padzero(elf_bss)) {
687 		error = -EFAULT;
688 		goto out;
689 	}
690 	/*
691 	 * Next, align both the file and mem bss up to the page size,
692 	 * since this is where elf_bss was just zeroed up to, and where
693 	 * last_bss will end after the vm_brk_flags() below.
694 	 */
695 	elf_bss = ELF_PAGEALIGN(elf_bss);
696 	last_bss = ELF_PAGEALIGN(last_bss);
697 	/* Finally, if there is still more bss to allocate, do it. */
698 	if (last_bss > elf_bss) {
699 		error = vm_brk_flags(elf_bss, last_bss - elf_bss,
700 				bss_prot & PROT_EXEC ? VM_EXEC : 0);
701 		if (error)
702 			goto out;
703 	}
704 
705 	error = load_addr;
706 out:
707 	return error;
708 }
709 
710 /*
711  * These are the functions used to load ELF style executables and shared
712  * libraries.  There is no binary dependent code anywhere else.
713  */
714 
parse_elf_property(const char * data,size_t * off,size_t datasz,struct arch_elf_state * arch,bool have_prev_type,u32 * prev_type)715 static int parse_elf_property(const char *data, size_t *off, size_t datasz,
716 			      struct arch_elf_state *arch,
717 			      bool have_prev_type, u32 *prev_type)
718 {
719 	size_t o, step;
720 	const struct gnu_property *pr;
721 	int ret;
722 
723 	if (*off == datasz)
724 		return -ENOENT;
725 
726 	if (WARN_ON_ONCE(*off > datasz || *off % ELF_GNU_PROPERTY_ALIGN))
727 		return -EIO;
728 	o = *off;
729 	datasz -= *off;
730 
731 	if (datasz < sizeof(*pr))
732 		return -ENOEXEC;
733 	pr = (const struct gnu_property *)(data + o);
734 	o += sizeof(*pr);
735 	datasz -= sizeof(*pr);
736 
737 	if (pr->pr_datasz > datasz)
738 		return -ENOEXEC;
739 
740 	WARN_ON_ONCE(o % ELF_GNU_PROPERTY_ALIGN);
741 	step = round_up(pr->pr_datasz, ELF_GNU_PROPERTY_ALIGN);
742 	if (step > datasz)
743 		return -ENOEXEC;
744 
745 	/* Properties are supposed to be unique and sorted on pr_type: */
746 	if (have_prev_type && pr->pr_type <= *prev_type)
747 		return -ENOEXEC;
748 	*prev_type = pr->pr_type;
749 
750 	ret = arch_parse_elf_property(pr->pr_type, data + o,
751 				      pr->pr_datasz, ELF_COMPAT, arch);
752 	if (ret)
753 		return ret;
754 
755 	*off = o + step;
756 	return 0;
757 }
758 
759 #define NOTE_DATA_SZ SZ_1K
760 #define GNU_PROPERTY_TYPE_0_NAME "GNU"
761 #define NOTE_NAME_SZ (sizeof(GNU_PROPERTY_TYPE_0_NAME))
762 
parse_elf_properties(struct file * f,const struct elf_phdr * phdr,struct arch_elf_state * arch)763 static int parse_elf_properties(struct file *f, const struct elf_phdr *phdr,
764 				struct arch_elf_state *arch)
765 {
766 	union {
767 		struct elf_note nhdr;
768 		char data[NOTE_DATA_SZ];
769 	} note;
770 	loff_t pos;
771 	ssize_t n;
772 	size_t off, datasz;
773 	int ret;
774 	bool have_prev_type;
775 	u32 prev_type;
776 
777 	if (!IS_ENABLED(CONFIG_ARCH_USE_GNU_PROPERTY) || !phdr)
778 		return 0;
779 
780 	/* load_elf_binary() shouldn't call us unless this is true... */
781 	if (WARN_ON_ONCE(phdr->p_type != PT_GNU_PROPERTY))
782 		return -ENOEXEC;
783 
784 	/* If the properties are crazy large, that's too bad (for now): */
785 	if (phdr->p_filesz > sizeof(note))
786 		return -ENOEXEC;
787 
788 	pos = phdr->p_offset;
789 	n = kernel_read(f, &note, phdr->p_filesz, &pos);
790 
791 	BUILD_BUG_ON(sizeof(note) < sizeof(note.nhdr) + NOTE_NAME_SZ);
792 	if (n < 0 || n < sizeof(note.nhdr) + NOTE_NAME_SZ)
793 		return -EIO;
794 
795 	if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
796 	    note.nhdr.n_namesz != NOTE_NAME_SZ ||
797 	    strncmp(note.data + sizeof(note.nhdr),
798 		    GNU_PROPERTY_TYPE_0_NAME, n - sizeof(note.nhdr)))
799 		return -ENOEXEC;
800 
801 	off = round_up(sizeof(note.nhdr) + NOTE_NAME_SZ,
802 		       ELF_GNU_PROPERTY_ALIGN);
803 	if (off > n)
804 		return -ENOEXEC;
805 
806 	if (note.nhdr.n_descsz > n - off)
807 		return -ENOEXEC;
808 	datasz = off + note.nhdr.n_descsz;
809 
810 	have_prev_type = false;
811 	do {
812 		ret = parse_elf_property(note.data, &off, datasz, arch,
813 					 have_prev_type, &prev_type);
814 		have_prev_type = true;
815 	} while (!ret);
816 
817 	return ret == -ENOENT ? 0 : ret;
818 }
819 
load_elf_binary(struct linux_binprm * bprm)820 static int load_elf_binary(struct linux_binprm *bprm)
821 {
822 	struct file *interpreter = NULL; /* to shut gcc up */
823  	unsigned long load_addr = 0, load_bias = 0;
824 	int load_addr_set = 0;
825 	unsigned long error;
826 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
827 	struct elf_phdr *elf_property_phdata = NULL;
828 	unsigned long elf_bss, elf_brk;
829 	int bss_prot = 0;
830 	int retval, i;
831 	unsigned long elf_entry;
832 	unsigned long e_entry;
833 	unsigned long interp_load_addr = 0;
834 	unsigned long start_code, end_code, start_data, end_data;
835 	unsigned long reloc_func_desc __maybe_unused = 0;
836 	int executable_stack = EXSTACK_DEFAULT;
837 	struct elfhdr *elf_ex = (struct elfhdr *)bprm->buf;
838 	struct elfhdr *interp_elf_ex = NULL;
839 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
840 	struct mm_struct *mm;
841 	struct pt_regs *regs;
842 
843 	retval = -ENOEXEC;
844 	/* First of all, some simple consistency checks */
845 	if (memcmp(elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
846 		goto out;
847 
848 	if (elf_ex->e_type != ET_EXEC && elf_ex->e_type != ET_DYN)
849 		goto out;
850 	if (!elf_check_arch(elf_ex))
851 		goto out;
852 	if (elf_check_fdpic(elf_ex))
853 		goto out;
854 	if (!bprm->file->f_op->mmap)
855 		goto out;
856 
857 	elf_phdata = load_elf_phdrs(elf_ex, bprm->file);
858 	if (!elf_phdata)
859 		goto out;
860 
861 	elf_ppnt = elf_phdata;
862 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++) {
863 		char *elf_interpreter;
864 
865 		if (elf_ppnt->p_type == PT_GNU_PROPERTY) {
866 			elf_property_phdata = elf_ppnt;
867 			continue;
868 		}
869 
870 		if (elf_ppnt->p_type != PT_INTERP)
871 			continue;
872 
873 		/*
874 		 * This is the program interpreter used for shared libraries -
875 		 * for now assume that this is an a.out format binary.
876 		 */
877 		retval = -ENOEXEC;
878 		if (elf_ppnt->p_filesz > PATH_MAX || elf_ppnt->p_filesz < 2)
879 			goto out_free_ph;
880 
881 		retval = -ENOMEM;
882 		elf_interpreter = kmalloc(elf_ppnt->p_filesz, GFP_KERNEL);
883 		if (!elf_interpreter)
884 			goto out_free_ph;
885 
886 		retval = elf_read(bprm->file, elf_interpreter, elf_ppnt->p_filesz,
887 				  elf_ppnt->p_offset);
888 		if (retval < 0)
889 			goto out_free_interp;
890 		/* make sure path is NULL terminated */
891 		retval = -ENOEXEC;
892 		if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
893 			goto out_free_interp;
894 
895 		interpreter = open_exec(elf_interpreter);
896 		kfree(elf_interpreter);
897 		retval = PTR_ERR(interpreter);
898 		if (IS_ERR(interpreter))
899 			goto out_free_ph;
900 
901 		/*
902 		 * If the binary is not readable then enforce mm->dumpable = 0
903 		 * regardless of the interpreter's permissions.
904 		 */
905 		would_dump(bprm, interpreter);
906 
907 		interp_elf_ex = kmalloc(sizeof(*interp_elf_ex), GFP_KERNEL);
908 		if (!interp_elf_ex) {
909 			retval = -ENOMEM;
910 			goto out_free_ph;
911 		}
912 
913 		/* Get the exec headers */
914 		retval = elf_read(interpreter, interp_elf_ex,
915 				  sizeof(*interp_elf_ex), 0);
916 		if (retval < 0)
917 			goto out_free_dentry;
918 
919 		break;
920 
921 out_free_interp:
922 		kfree(elf_interpreter);
923 		goto out_free_ph;
924 	}
925 
926 	elf_ppnt = elf_phdata;
927 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++)
928 		switch (elf_ppnt->p_type) {
929 		case PT_GNU_STACK:
930 			if (elf_ppnt->p_flags & PF_X)
931 				executable_stack = EXSTACK_ENABLE_X;
932 			else
933 				executable_stack = EXSTACK_DISABLE_X;
934 			break;
935 
936 		case PT_LOPROC ... PT_HIPROC:
937 			retval = arch_elf_pt_proc(elf_ex, elf_ppnt,
938 						  bprm->file, false,
939 						  &arch_state);
940 			if (retval)
941 				goto out_free_dentry;
942 			break;
943 		}
944 
945 	/* Some simple consistency checks for the interpreter */
946 	if (interpreter) {
947 		retval = -ELIBBAD;
948 		/* Not an ELF interpreter */
949 		if (memcmp(interp_elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
950 			goto out_free_dentry;
951 		/* Verify the interpreter has a valid arch */
952 		if (!elf_check_arch(interp_elf_ex) ||
953 		    elf_check_fdpic(interp_elf_ex))
954 			goto out_free_dentry;
955 
956 		/* Load the interpreter program headers */
957 		interp_elf_phdata = load_elf_phdrs(interp_elf_ex,
958 						   interpreter);
959 		if (!interp_elf_phdata)
960 			goto out_free_dentry;
961 
962 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
963 		elf_property_phdata = NULL;
964 		elf_ppnt = interp_elf_phdata;
965 		for (i = 0; i < interp_elf_ex->e_phnum; i++, elf_ppnt++)
966 			switch (elf_ppnt->p_type) {
967 			case PT_GNU_PROPERTY:
968 				elf_property_phdata = elf_ppnt;
969 				break;
970 
971 			case PT_LOPROC ... PT_HIPROC:
972 				retval = arch_elf_pt_proc(interp_elf_ex,
973 							  elf_ppnt, interpreter,
974 							  true, &arch_state);
975 				if (retval)
976 					goto out_free_dentry;
977 				break;
978 			}
979 	}
980 
981 	retval = parse_elf_properties(interpreter ?: bprm->file,
982 				      elf_property_phdata, &arch_state);
983 	if (retval)
984 		goto out_free_dentry;
985 
986 	/*
987 	 * Allow arch code to reject the ELF at this point, whilst it's
988 	 * still possible to return an error to the code that invoked
989 	 * the exec syscall.
990 	 */
991 	retval = arch_check_elf(elf_ex,
992 				!!interpreter, interp_elf_ex,
993 				&arch_state);
994 	if (retval)
995 		goto out_free_dentry;
996 
997 	/* Flush all traces of the currently running executable */
998 	retval = begin_new_exec(bprm);
999 	if (retval)
1000 		goto out_free_dentry;
1001 
1002 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
1003 	   may depend on the personality.  */
1004 	SET_PERSONALITY2(*elf_ex, &arch_state);
1005 	if (elf_read_implies_exec(*elf_ex, executable_stack))
1006 		current->personality |= READ_IMPLIES_EXEC;
1007 
1008 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1009 		current->flags |= PF_RANDOMIZE;
1010 
1011 	setup_new_exec(bprm);
1012 
1013 	/* Do this so that we can load the interpreter, if need be.  We will
1014 	   change some of these later */
1015 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
1016 				 executable_stack);
1017 	if (retval < 0)
1018 		goto out_free_dentry;
1019 
1020 	elf_bss = 0;
1021 	elf_brk = 0;
1022 
1023 	start_code = ~0UL;
1024 	end_code = 0;
1025 	start_data = 0;
1026 	end_data = 0;
1027 
1028 	/* Now we do a little grungy work by mmapping the ELF image into
1029 	   the correct location in memory. */
1030 	for(i = 0, elf_ppnt = elf_phdata;
1031 	    i < elf_ex->e_phnum; i++, elf_ppnt++) {
1032 		int elf_prot, elf_flags;
1033 		unsigned long k, vaddr;
1034 		unsigned long total_size = 0;
1035 		unsigned long alignment;
1036 
1037 		if (elf_ppnt->p_type == PT_OHOS_RANDOMDATA) {
1038 			get_random_bytes((void *)(elf_ppnt->p_vaddr + load_bias), (int)elf_ppnt->p_memsz);
1039 			continue;
1040 		}
1041 
1042 		if (elf_ppnt->p_type != PT_LOAD)
1043 			continue;
1044 
1045 		if (unlikely (elf_brk > elf_bss)) {
1046 			unsigned long nbyte;
1047 
1048 			/* There was a PT_LOAD segment with p_memsz > p_filesz
1049 			   before this one. Map anonymous pages, if needed,
1050 			   and clear the area.  */
1051 			retval = set_brk(elf_bss + load_bias,
1052 					 elf_brk + load_bias,
1053 					 bss_prot);
1054 			if (retval)
1055 				goto out_free_dentry;
1056 			nbyte = ELF_PAGEOFFSET(elf_bss);
1057 			if (nbyte) {
1058 				nbyte = ELF_MIN_ALIGN - nbyte;
1059 				if (nbyte > elf_brk - elf_bss)
1060 					nbyte = elf_brk - elf_bss;
1061 				if (clear_user((void __user *)elf_bss +
1062 							load_bias, nbyte)) {
1063 					/*
1064 					 * This bss-zeroing can fail if the ELF
1065 					 * file specifies odd protections. So
1066 					 * we don't check the return value
1067 					 */
1068 				}
1069 			}
1070 		}
1071 
1072 		elf_prot = make_prot(elf_ppnt->p_flags, &arch_state,
1073 				     !!interpreter, false);
1074 
1075 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
1076 
1077 		vaddr = elf_ppnt->p_vaddr;
1078 		/*
1079 		 * If we are loading ET_EXEC or we have already performed
1080 		 * the ET_DYN load_addr calculations, proceed normally.
1081 		 */
1082 		if (elf_ex->e_type == ET_EXEC || load_addr_set) {
1083 			elf_flags |= MAP_FIXED;
1084 		} else if (elf_ex->e_type == ET_DYN) {
1085 			/*
1086 			 * This logic is run once for the first LOAD Program
1087 			 * Header for ET_DYN binaries to calculate the
1088 			 * randomization (load_bias) for all the LOAD
1089 			 * Program Headers, and to calculate the entire
1090 			 * size of the ELF mapping (total_size). (Note that
1091 			 * load_addr_set is set to true later once the
1092 			 * initial mapping is performed.)
1093 			 *
1094 			 * There are effectively two types of ET_DYN
1095 			 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
1096 			 * and loaders (ET_DYN without INTERP, since they
1097 			 * _are_ the ELF interpreter). The loaders must
1098 			 * be loaded away from programs since the program
1099 			 * may otherwise collide with the loader (especially
1100 			 * for ET_EXEC which does not have a randomized
1101 			 * position). For example to handle invocations of
1102 			 * "./ld.so someprog" to test out a new version of
1103 			 * the loader, the subsequent program that the
1104 			 * loader loads must avoid the loader itself, so
1105 			 * they cannot share the same load range. Sufficient
1106 			 * room for the brk must be allocated with the
1107 			 * loader as well, since brk must be available with
1108 			 * the loader.
1109 			 *
1110 			 * Therefore, programs are loaded offset from
1111 			 * ELF_ET_DYN_BASE and loaders are loaded into the
1112 			 * independently randomized mmap region (0 load_bias
1113 			 * without MAP_FIXED).
1114 			 */
1115 			if (interpreter) {
1116 				load_bias = ELF_ET_DYN_BASE;
1117 				if (current->flags & PF_RANDOMIZE)
1118 					load_bias += arch_mmap_rnd();
1119 				alignment = maximum_alignment(elf_phdata, elf_ex->e_phnum);
1120 				if (alignment)
1121 					load_bias &= ~(alignment - 1);
1122 				elf_flags |= MAP_FIXED;
1123 			} else
1124 				load_bias = 0;
1125 
1126 			/*
1127 			 * Since load_bias is used for all subsequent loading
1128 			 * calculations, we must lower it by the first vaddr
1129 			 * so that the remaining calculations based on the
1130 			 * ELF vaddrs will be correctly offset. The result
1131 			 * is then page aligned.
1132 			 */
1133 			load_bias = ELF_PAGESTART(load_bias - vaddr);
1134 
1135 			total_size = total_mapping_size(elf_phdata,
1136 							elf_ex->e_phnum);
1137 			if (!total_size) {
1138 				retval = -EINVAL;
1139 				goto out_free_dentry;
1140 			}
1141 		}
1142 
1143 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1144 				elf_prot, elf_flags, total_size);
1145 		if (BAD_ADDR(error)) {
1146 			retval = IS_ERR((void *)error) ?
1147 				PTR_ERR((void*)error) : -EINVAL;
1148 			goto out_free_dentry;
1149 		}
1150 
1151 		if (!load_addr_set) {
1152 			load_addr_set = 1;
1153 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1154 			if (elf_ex->e_type == ET_DYN) {
1155 				load_bias += error -
1156 				             ELF_PAGESTART(load_bias + vaddr);
1157 				load_addr += load_bias;
1158 				reloc_func_desc = load_bias;
1159 			}
1160 		}
1161 		k = elf_ppnt->p_vaddr;
1162 		if ((elf_ppnt->p_flags & PF_X) && k < start_code)
1163 			start_code = k;
1164 		if (start_data < k)
1165 			start_data = k;
1166 
1167 		/*
1168 		 * Check to see if the section's size will overflow the
1169 		 * allowed task size. Note that p_filesz must always be
1170 		 * <= p_memsz so it is only necessary to check p_memsz.
1171 		 */
1172 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1173 		    elf_ppnt->p_memsz > TASK_SIZE ||
1174 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
1175 			/* set_brk can never work. Avoid overflows. */
1176 			retval = -EINVAL;
1177 			goto out_free_dentry;
1178 		}
1179 
1180 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1181 
1182 		if (k > elf_bss)
1183 			elf_bss = k;
1184 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1185 			end_code = k;
1186 		if (end_data < k)
1187 			end_data = k;
1188 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1189 		if (k > elf_brk) {
1190 			bss_prot = elf_prot;
1191 			elf_brk = k;
1192 		}
1193 	}
1194 
1195 	e_entry = elf_ex->e_entry + load_bias;
1196 	elf_bss += load_bias;
1197 	elf_brk += load_bias;
1198 	start_code += load_bias;
1199 	end_code += load_bias;
1200 	start_data += load_bias;
1201 	end_data += load_bias;
1202 
1203 	/* Calling set_brk effectively mmaps the pages that we need
1204 	 * for the bss and break sections.  We must do this before
1205 	 * mapping in the interpreter, to make sure it doesn't wind
1206 	 * up getting placed where the bss needs to go.
1207 	 */
1208 	retval = set_brk(elf_bss, elf_brk, bss_prot);
1209 	if (retval)
1210 		goto out_free_dentry;
1211 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1212 		retval = -EFAULT; /* Nobody gets to see this, but.. */
1213 		goto out_free_dentry;
1214 	}
1215 
1216 	if (interpreter) {
1217 		elf_entry = load_elf_interp(interp_elf_ex,
1218 					    interpreter,
1219 					    load_bias, interp_elf_phdata,
1220 					    &arch_state);
1221 		if (!IS_ERR((void *)elf_entry)) {
1222 			/*
1223 			 * load_elf_interp() returns relocation
1224 			 * adjustment
1225 			 */
1226 			interp_load_addr = elf_entry;
1227 			elf_entry += interp_elf_ex->e_entry;
1228 		}
1229 		if (BAD_ADDR(elf_entry)) {
1230 			retval = IS_ERR((void *)elf_entry) ?
1231 					(int)elf_entry : -EINVAL;
1232 			goto out_free_dentry;
1233 		}
1234 		reloc_func_desc = interp_load_addr;
1235 
1236 		allow_write_access(interpreter);
1237 		fput(interpreter);
1238 
1239 		kfree(interp_elf_ex);
1240 		kfree(interp_elf_phdata);
1241 	} else {
1242 		elf_entry = e_entry;
1243 		if (BAD_ADDR(elf_entry)) {
1244 			retval = -EINVAL;
1245 			goto out_free_dentry;
1246 		}
1247 	}
1248 
1249 	kfree(elf_phdata);
1250 
1251 	set_binfmt(&elf_format);
1252 
1253 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1254 	retval = arch_setup_additional_pages(bprm, !!interpreter);
1255 	if (retval < 0)
1256 		goto out;
1257 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1258 
1259 	retval = create_elf_tables(bprm, elf_ex,
1260 			  load_addr, interp_load_addr, e_entry);
1261 	if (retval < 0)
1262 		goto out;
1263 
1264 	mm = current->mm;
1265 	mm->end_code = end_code;
1266 	mm->start_code = start_code;
1267 	mm->start_data = start_data;
1268 	mm->end_data = end_data;
1269 	mm->start_stack = bprm->p;
1270 
1271 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1272 		/*
1273 		 * For architectures with ELF randomization, when executing
1274 		 * a loader directly (i.e. no interpreter listed in ELF
1275 		 * headers), move the brk area out of the mmap region
1276 		 * (since it grows up, and may collide early with the stack
1277 		 * growing down), and into the unused ELF_ET_DYN_BASE region.
1278 		 */
1279 		if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) &&
1280 		    elf_ex->e_type == ET_DYN && !interpreter) {
1281 			mm->brk = mm->start_brk = ELF_ET_DYN_BASE;
1282 		}
1283 
1284 		mm->brk = mm->start_brk = arch_randomize_brk(mm);
1285 #ifdef compat_brk_randomized
1286 		current->brk_randomized = 1;
1287 #endif
1288 	}
1289 
1290 	if (current->personality & MMAP_PAGE_ZERO) {
1291 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1292 		   and some applications "depend" upon this behavior.
1293 		   Since we do not have the power to recompile these, we
1294 		   emulate the SVr4 behavior. Sigh. */
1295 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1296 				MAP_FIXED | MAP_PRIVATE, 0);
1297 	}
1298 
1299 	regs = current_pt_regs();
1300 #ifdef ELF_PLAT_INIT
1301 	/*
1302 	 * The ABI may specify that certain registers be set up in special
1303 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1304 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1305 	 * that the e_entry field is the address of the function descriptor
1306 	 * for the startup routine, rather than the address of the startup
1307 	 * routine itself.  This macro performs whatever initialization to
1308 	 * the regs structure is required as well as any relocations to the
1309 	 * function descriptor entries when executing dynamically links apps.
1310 	 */
1311 	ELF_PLAT_INIT(regs, reloc_func_desc);
1312 #endif
1313 
1314 	finalize_exec(bprm);
1315 	start_thread(regs, elf_entry, bprm->p);
1316 	retval = 0;
1317 out:
1318 	return retval;
1319 
1320 	/* error cleanup */
1321 out_free_dentry:
1322 	kfree(interp_elf_ex);
1323 	kfree(interp_elf_phdata);
1324 	allow_write_access(interpreter);
1325 	if (interpreter)
1326 		fput(interpreter);
1327 out_free_ph:
1328 	kfree(elf_phdata);
1329 	goto out;
1330 }
1331 
1332 #ifdef CONFIG_USELIB
1333 /* This is really simpleminded and specialized - we are loading an
1334    a.out library that is given an ELF header. */
load_elf_library(struct file * file)1335 static int load_elf_library(struct file *file)
1336 {
1337 	struct elf_phdr *elf_phdata;
1338 	struct elf_phdr *eppnt;
1339 	unsigned long elf_bss, bss, len;
1340 	int retval, error, i, j;
1341 	struct elfhdr elf_ex;
1342 
1343 	error = -ENOEXEC;
1344 	retval = elf_read(file, &elf_ex, sizeof(elf_ex), 0);
1345 	if (retval < 0)
1346 		goto out;
1347 
1348 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1349 		goto out;
1350 
1351 	/* First of all, some simple consistency checks */
1352 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1353 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1354 		goto out;
1355 	if (elf_check_fdpic(&elf_ex))
1356 		goto out;
1357 
1358 	/* Now read in all of the header information */
1359 
1360 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1361 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1362 
1363 	error = -ENOMEM;
1364 	elf_phdata = kmalloc(j, GFP_KERNEL);
1365 	if (!elf_phdata)
1366 		goto out;
1367 
1368 	eppnt = elf_phdata;
1369 	error = -ENOEXEC;
1370 	retval = elf_read(file, eppnt, j, elf_ex.e_phoff);
1371 	if (retval < 0)
1372 		goto out_free_ph;
1373 
1374 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1375 		if ((eppnt + i)->p_type == PT_LOAD)
1376 			j++;
1377 	if (j != 1)
1378 		goto out_free_ph;
1379 
1380 	while (eppnt->p_type != PT_LOAD)
1381 		eppnt++;
1382 
1383 	/* Now use mmap to map the library into memory. */
1384 	error = vm_mmap(file,
1385 			ELF_PAGESTART(eppnt->p_vaddr),
1386 			(eppnt->p_filesz +
1387 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1388 			PROT_READ | PROT_WRITE | PROT_EXEC,
1389 			MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_DENYWRITE,
1390 			(eppnt->p_offset -
1391 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1392 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1393 		goto out_free_ph;
1394 
1395 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1396 	if (padzero(elf_bss)) {
1397 		error = -EFAULT;
1398 		goto out_free_ph;
1399 	}
1400 
1401 	len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1402 	bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1403 	if (bss > len) {
1404 		error = vm_brk(len, bss - len);
1405 		if (error)
1406 			goto out_free_ph;
1407 	}
1408 	error = 0;
1409 
1410 out_free_ph:
1411 	kfree(elf_phdata);
1412 out:
1413 	return error;
1414 }
1415 #endif /* #ifdef CONFIG_USELIB */
1416 
1417 #ifdef CONFIG_ELF_CORE
1418 /*
1419  * ELF core dumper
1420  *
1421  * Modelled on fs/exec.c:aout_core_dump()
1422  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1423  */
1424 
1425 /* An ELF note in memory */
1426 struct memelfnote
1427 {
1428 	const char *name;
1429 	int type;
1430 	unsigned int datasz;
1431 	void *data;
1432 };
1433 
notesize(struct memelfnote * en)1434 static int notesize(struct memelfnote *en)
1435 {
1436 	int sz;
1437 
1438 	sz = sizeof(struct elf_note);
1439 	sz += roundup(strlen(en->name) + 1, 4);
1440 	sz += roundup(en->datasz, 4);
1441 
1442 	return sz;
1443 }
1444 
writenote(struct memelfnote * men,struct coredump_params * cprm)1445 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1446 {
1447 	struct elf_note en;
1448 	en.n_namesz = strlen(men->name) + 1;
1449 	en.n_descsz = men->datasz;
1450 	en.n_type = men->type;
1451 
1452 	return dump_emit(cprm, &en, sizeof(en)) &&
1453 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1454 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1455 }
1456 
fill_elf_header(struct elfhdr * elf,int segs,u16 machine,u32 flags)1457 static void fill_elf_header(struct elfhdr *elf, int segs,
1458 			    u16 machine, u32 flags)
1459 {
1460 	memset(elf, 0, sizeof(*elf));
1461 
1462 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1463 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1464 	elf->e_ident[EI_DATA] = ELF_DATA;
1465 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1466 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1467 
1468 	elf->e_type = ET_CORE;
1469 	elf->e_machine = machine;
1470 	elf->e_version = EV_CURRENT;
1471 	elf->e_phoff = sizeof(struct elfhdr);
1472 	elf->e_flags = flags;
1473 	elf->e_ehsize = sizeof(struct elfhdr);
1474 	elf->e_phentsize = sizeof(struct elf_phdr);
1475 	elf->e_phnum = segs;
1476 }
1477 
fill_elf_note_phdr(struct elf_phdr * phdr,int sz,loff_t offset)1478 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1479 {
1480 	phdr->p_type = PT_NOTE;
1481 	phdr->p_offset = offset;
1482 	phdr->p_vaddr = 0;
1483 	phdr->p_paddr = 0;
1484 	phdr->p_filesz = sz;
1485 	phdr->p_memsz = 0;
1486 	phdr->p_flags = 0;
1487 	phdr->p_align = 0;
1488 }
1489 
fill_note(struct memelfnote * note,const char * name,int type,unsigned int sz,void * data)1490 static void fill_note(struct memelfnote *note, const char *name, int type,
1491 		unsigned int sz, void *data)
1492 {
1493 	note->name = name;
1494 	note->type = type;
1495 	note->datasz = sz;
1496 	note->data = data;
1497 }
1498 
1499 /*
1500  * fill up all the fields in prstatus from the given task struct, except
1501  * registers which need to be filled up separately.
1502  */
fill_prstatus(struct elf_prstatus * prstatus,struct task_struct * p,long signr)1503 static void fill_prstatus(struct elf_prstatus *prstatus,
1504 		struct task_struct *p, long signr)
1505 {
1506 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1507 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1508 	prstatus->pr_sighold = p->blocked.sig[0];
1509 	rcu_read_lock();
1510 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1511 	rcu_read_unlock();
1512 	prstatus->pr_pid = task_pid_vnr(p);
1513 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1514 	prstatus->pr_sid = task_session_vnr(p);
1515 	if (thread_group_leader(p)) {
1516 		struct task_cputime cputime;
1517 
1518 		/*
1519 		 * This is the record for the group leader.  It shows the
1520 		 * group-wide total, not its individual thread total.
1521 		 */
1522 		thread_group_cputime(p, &cputime);
1523 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1524 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1525 	} else {
1526 		u64 utime, stime;
1527 
1528 		task_cputime(p, &utime, &stime);
1529 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1530 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1531 	}
1532 
1533 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1534 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1535 }
1536 
fill_psinfo(struct elf_prpsinfo * psinfo,struct task_struct * p,struct mm_struct * mm)1537 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1538 		       struct mm_struct *mm)
1539 {
1540 	const struct cred *cred;
1541 	unsigned int i, len;
1542 
1543 	/* first copy the parameters from user space */
1544 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1545 
1546 	len = mm->arg_end - mm->arg_start;
1547 	if (len >= ELF_PRARGSZ)
1548 		len = ELF_PRARGSZ-1;
1549 	if (copy_from_user(&psinfo->pr_psargs,
1550 		           (const char __user *)mm->arg_start, len))
1551 		return -EFAULT;
1552 	for(i = 0; i < len; i++)
1553 		if (psinfo->pr_psargs[i] == 0)
1554 			psinfo->pr_psargs[i] = ' ';
1555 	psinfo->pr_psargs[len] = 0;
1556 
1557 	rcu_read_lock();
1558 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1559 	rcu_read_unlock();
1560 	psinfo->pr_pid = task_pid_vnr(p);
1561 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1562 	psinfo->pr_sid = task_session_vnr(p);
1563 
1564 	i = p->state ? ffz(~p->state) + 1 : 0;
1565 	psinfo->pr_state = i;
1566 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1567 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1568 	psinfo->pr_nice = task_nice(p);
1569 	psinfo->pr_flag = p->flags;
1570 	rcu_read_lock();
1571 	cred = __task_cred(p);
1572 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1573 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1574 	rcu_read_unlock();
1575 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1576 
1577 	return 0;
1578 }
1579 
fill_auxv_note(struct memelfnote * note,struct mm_struct * mm)1580 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1581 {
1582 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1583 	int i = 0;
1584 	do
1585 		i += 2;
1586 	while (auxv[i - 2] != AT_NULL);
1587 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1588 }
1589 
fill_siginfo_note(struct memelfnote * note,user_siginfo_t * csigdata,const kernel_siginfo_t * siginfo)1590 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1591 		const kernel_siginfo_t *siginfo)
1592 {
1593 	copy_siginfo_to_external(csigdata, siginfo);
1594 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1595 }
1596 
1597 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1598 /*
1599  * Format of NT_FILE note:
1600  *
1601  * long count     -- how many files are mapped
1602  * long page_size -- units for file_ofs
1603  * array of [COUNT] elements of
1604  *   long start
1605  *   long end
1606  *   long file_ofs
1607  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1608  */
fill_files_note(struct memelfnote * note)1609 static int fill_files_note(struct memelfnote *note)
1610 {
1611 	struct mm_struct *mm = current->mm;
1612 	struct vm_area_struct *vma;
1613 	unsigned count, size, names_ofs, remaining, n;
1614 	user_long_t *data;
1615 	user_long_t *start_end_ofs;
1616 	char *name_base, *name_curpos;
1617 
1618 	/* *Estimated* file count and total data size needed */
1619 	count = mm->map_count;
1620 	if (count > UINT_MAX / 64)
1621 		return -EINVAL;
1622 	size = count * 64;
1623 
1624 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1625  alloc:
1626 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1627 		return -EINVAL;
1628 	size = round_up(size, PAGE_SIZE);
1629 	/*
1630 	 * "size" can be 0 here legitimately.
1631 	 * Let it ENOMEM and omit NT_FILE section which will be empty anyway.
1632 	 */
1633 	data = kvmalloc(size, GFP_KERNEL);
1634 	if (ZERO_OR_NULL_PTR(data))
1635 		return -ENOMEM;
1636 
1637 	start_end_ofs = data + 2;
1638 	name_base = name_curpos = ((char *)data) + names_ofs;
1639 	remaining = size - names_ofs;
1640 	count = 0;
1641 	for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
1642 		struct file *file;
1643 		const char *filename;
1644 
1645 		file = vma->vm_file;
1646 		if (!file)
1647 			continue;
1648 		filename = file_path(file, name_curpos, remaining);
1649 		if (IS_ERR(filename)) {
1650 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1651 				kvfree(data);
1652 				size = size * 5 / 4;
1653 				goto alloc;
1654 			}
1655 			continue;
1656 		}
1657 
1658 		/* file_path() fills at the end, move name down */
1659 		/* n = strlen(filename) + 1: */
1660 		n = (name_curpos + remaining) - filename;
1661 		remaining = filename - name_curpos;
1662 		memmove(name_curpos, filename, n);
1663 		name_curpos += n;
1664 
1665 		*start_end_ofs++ = vma->vm_start;
1666 		*start_end_ofs++ = vma->vm_end;
1667 		*start_end_ofs++ = vma->vm_pgoff;
1668 		count++;
1669 	}
1670 
1671 	/* Now we know exact count of files, can store it */
1672 	data[0] = count;
1673 	data[1] = PAGE_SIZE;
1674 	/*
1675 	 * Count usually is less than mm->map_count,
1676 	 * we need to move filenames down.
1677 	 */
1678 	n = mm->map_count - count;
1679 	if (n != 0) {
1680 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1681 		memmove(name_base - shift_bytes, name_base,
1682 			name_curpos - name_base);
1683 		name_curpos -= shift_bytes;
1684 	}
1685 
1686 	size = name_curpos - (char *)data;
1687 	fill_note(note, "CORE", NT_FILE, size, data);
1688 	return 0;
1689 }
1690 
1691 #ifdef CORE_DUMP_USE_REGSET
1692 #include <linux/regset.h>
1693 
1694 struct elf_thread_core_info {
1695 	struct elf_thread_core_info *next;
1696 	struct task_struct *task;
1697 	struct elf_prstatus prstatus;
1698 	struct memelfnote notes[];
1699 };
1700 
1701 struct elf_note_info {
1702 	struct elf_thread_core_info *thread;
1703 	struct memelfnote psinfo;
1704 	struct memelfnote signote;
1705 	struct memelfnote auxv;
1706 	struct memelfnote files;
1707 	user_siginfo_t csigdata;
1708 	size_t size;
1709 	int thread_notes;
1710 };
1711 
1712 /*
1713  * When a regset has a writeback hook, we call it on each thread before
1714  * dumping user memory.  On register window machines, this makes sure the
1715  * user memory backing the register data is up to date before we read it.
1716  */
do_thread_regset_writeback(struct task_struct * task,const struct user_regset * regset)1717 static void do_thread_regset_writeback(struct task_struct *task,
1718 				       const struct user_regset *regset)
1719 {
1720 	if (regset->writeback)
1721 		regset->writeback(task, regset, 1);
1722 }
1723 
1724 #ifndef PRSTATUS_SIZE
1725 #define PRSTATUS_SIZE(S, R) sizeof(S)
1726 #endif
1727 
1728 #ifndef SET_PR_FPVALID
1729 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1730 #endif
1731 
fill_thread_core_info(struct elf_thread_core_info * t,const struct user_regset_view * view,long signr,size_t * total)1732 static int fill_thread_core_info(struct elf_thread_core_info *t,
1733 				 const struct user_regset_view *view,
1734 				 long signr, size_t *total)
1735 {
1736 	unsigned int i;
1737 	int regset0_size;
1738 
1739 	/*
1740 	 * NT_PRSTATUS is the one special case, because the regset data
1741 	 * goes into the pr_reg field inside the note contents, rather
1742 	 * than being the whole note contents.  We fill the reset in here.
1743 	 * We assume that regset 0 is NT_PRSTATUS.
1744 	 */
1745 	fill_prstatus(&t->prstatus, t->task, signr);
1746 	regset0_size = regset_get(t->task, &view->regsets[0],
1747 		   sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1748 	if (regset0_size < 0)
1749 		return 0;
1750 
1751 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1752 		  PRSTATUS_SIZE(t->prstatus, regset0_size), &t->prstatus);
1753 	*total += notesize(&t->notes[0]);
1754 
1755 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1756 
1757 	/*
1758 	 * Each other regset might generate a note too.  For each regset
1759 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1760 	 * all zero and we'll know to skip writing it later.
1761 	 */
1762 	for (i = 1; i < view->n; ++i) {
1763 		const struct user_regset *regset = &view->regsets[i];
1764 		int note_type = regset->core_note_type;
1765 		bool is_fpreg = note_type == NT_PRFPREG;
1766 		void *data;
1767 		int ret;
1768 
1769 		do_thread_regset_writeback(t->task, regset);
1770 		if (!note_type) // not for coredumps
1771 			continue;
1772 		if (regset->active && regset->active(t->task, regset) <= 0)
1773 			continue;
1774 
1775 		ret = regset_get_alloc(t->task, regset, ~0U, &data);
1776 		if (ret < 0)
1777 			continue;
1778 
1779 		if (is_fpreg)
1780 			SET_PR_FPVALID(&t->prstatus, 1, regset0_size);
1781 
1782 		fill_note(&t->notes[i], is_fpreg ? "CORE" : "LINUX",
1783 			  note_type, ret, data);
1784 
1785 		*total += notesize(&t->notes[i]);
1786 	}
1787 
1788 	return 1;
1789 }
1790 
fill_note_info(struct elfhdr * elf,int phdrs,struct elf_note_info * info,const kernel_siginfo_t * siginfo,struct pt_regs * regs)1791 static int fill_note_info(struct elfhdr *elf, int phdrs,
1792 			  struct elf_note_info *info,
1793 			  const kernel_siginfo_t *siginfo, struct pt_regs *regs)
1794 {
1795 	struct task_struct *dump_task = current;
1796 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1797 	struct elf_thread_core_info *t;
1798 	struct elf_prpsinfo *psinfo;
1799 	struct core_thread *ct;
1800 	unsigned int i;
1801 
1802 	info->size = 0;
1803 	info->thread = NULL;
1804 
1805 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1806 	if (psinfo == NULL) {
1807 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1808 		return 0;
1809 	}
1810 
1811 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1812 
1813 	/*
1814 	 * Figure out how many notes we're going to need for each thread.
1815 	 */
1816 	info->thread_notes = 0;
1817 	for (i = 0; i < view->n; ++i)
1818 		if (view->regsets[i].core_note_type != 0)
1819 			++info->thread_notes;
1820 
1821 	/*
1822 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1823 	 * since it is our one special case.
1824 	 */
1825 	if (unlikely(info->thread_notes == 0) ||
1826 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1827 		WARN_ON(1);
1828 		return 0;
1829 	}
1830 
1831 	/*
1832 	 * Initialize the ELF file header.
1833 	 */
1834 	fill_elf_header(elf, phdrs,
1835 			view->e_machine, view->e_flags);
1836 
1837 	/*
1838 	 * Allocate a structure for each thread.
1839 	 */
1840 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1841 		t = kzalloc(offsetof(struct elf_thread_core_info,
1842 				     notes[info->thread_notes]),
1843 			    GFP_KERNEL);
1844 		if (unlikely(!t))
1845 			return 0;
1846 
1847 		t->task = ct->task;
1848 		if (ct->task == dump_task || !info->thread) {
1849 			t->next = info->thread;
1850 			info->thread = t;
1851 		} else {
1852 			/*
1853 			 * Make sure to keep the original task at
1854 			 * the head of the list.
1855 			 */
1856 			t->next = info->thread->next;
1857 			info->thread->next = t;
1858 		}
1859 	}
1860 
1861 	/*
1862 	 * Now fill in each thread's information.
1863 	 */
1864 	for (t = info->thread; t != NULL; t = t->next)
1865 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1866 			return 0;
1867 
1868 	/*
1869 	 * Fill in the two process-wide notes.
1870 	 */
1871 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1872 	info->size += notesize(&info->psinfo);
1873 
1874 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1875 	info->size += notesize(&info->signote);
1876 
1877 	fill_auxv_note(&info->auxv, current->mm);
1878 	info->size += notesize(&info->auxv);
1879 
1880 	if (fill_files_note(&info->files) == 0)
1881 		info->size += notesize(&info->files);
1882 
1883 	return 1;
1884 }
1885 
get_note_info_size(struct elf_note_info * info)1886 static size_t get_note_info_size(struct elf_note_info *info)
1887 {
1888 	return info->size;
1889 }
1890 
1891 /*
1892  * Write all the notes for each thread.  When writing the first thread, the
1893  * process-wide notes are interleaved after the first thread-specific note.
1894  */
write_note_info(struct elf_note_info * info,struct coredump_params * cprm)1895 static int write_note_info(struct elf_note_info *info,
1896 			   struct coredump_params *cprm)
1897 {
1898 	bool first = true;
1899 	struct elf_thread_core_info *t = info->thread;
1900 
1901 	do {
1902 		int i;
1903 
1904 		if (!writenote(&t->notes[0], cprm))
1905 			return 0;
1906 
1907 		if (first && !writenote(&info->psinfo, cprm))
1908 			return 0;
1909 		if (first && !writenote(&info->signote, cprm))
1910 			return 0;
1911 		if (first && !writenote(&info->auxv, cprm))
1912 			return 0;
1913 		if (first && info->files.data &&
1914 				!writenote(&info->files, cprm))
1915 			return 0;
1916 
1917 		for (i = 1; i < info->thread_notes; ++i)
1918 			if (t->notes[i].data &&
1919 			    !writenote(&t->notes[i], cprm))
1920 				return 0;
1921 
1922 		first = false;
1923 		t = t->next;
1924 	} while (t);
1925 
1926 	return 1;
1927 }
1928 
free_note_info(struct elf_note_info * info)1929 static void free_note_info(struct elf_note_info *info)
1930 {
1931 	struct elf_thread_core_info *threads = info->thread;
1932 	while (threads) {
1933 		unsigned int i;
1934 		struct elf_thread_core_info *t = threads;
1935 		threads = t->next;
1936 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1937 		for (i = 1; i < info->thread_notes; ++i)
1938 			kfree(t->notes[i].data);
1939 		kfree(t);
1940 	}
1941 	kfree(info->psinfo.data);
1942 	kvfree(info->files.data);
1943 }
1944 
1945 #else
1946 
1947 /* Here is the structure in which status of each thread is captured. */
1948 struct elf_thread_status
1949 {
1950 	struct list_head list;
1951 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1952 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1953 	struct task_struct *thread;
1954 	struct memelfnote notes[3];
1955 	int num_notes;
1956 };
1957 
1958 /*
1959  * In order to add the specific thread information for the elf file format,
1960  * we need to keep a linked list of every threads pr_status and then create
1961  * a single section for them in the final core file.
1962  */
elf_dump_thread_status(long signr,struct elf_thread_status * t)1963 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1964 {
1965 	int sz = 0;
1966 	struct task_struct *p = t->thread;
1967 	t->num_notes = 0;
1968 
1969 	fill_prstatus(&t->prstatus, p, signr);
1970 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1971 
1972 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1973 		  &(t->prstatus));
1974 	t->num_notes++;
1975 	sz += notesize(&t->notes[0]);
1976 
1977 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1978 								&t->fpu))) {
1979 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1980 			  &(t->fpu));
1981 		t->num_notes++;
1982 		sz += notesize(&t->notes[1]);
1983 	}
1984 	return sz;
1985 }
1986 
1987 struct elf_note_info {
1988 	struct memelfnote *notes;
1989 	struct memelfnote *notes_files;
1990 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1991 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1992 	struct list_head thread_list;
1993 	elf_fpregset_t *fpu;
1994 	user_siginfo_t csigdata;
1995 	int thread_status_size;
1996 	int numnote;
1997 };
1998 
elf_note_info_init(struct elf_note_info * info)1999 static int elf_note_info_init(struct elf_note_info *info)
2000 {
2001 	memset(info, 0, sizeof(*info));
2002 	INIT_LIST_HEAD(&info->thread_list);
2003 
2004 	/* Allocate space for ELF notes */
2005 	info->notes = kmalloc_array(8, sizeof(struct memelfnote), GFP_KERNEL);
2006 	if (!info->notes)
2007 		return 0;
2008 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2009 	if (!info->psinfo)
2010 		return 0;
2011 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2012 	if (!info->prstatus)
2013 		return 0;
2014 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2015 	if (!info->fpu)
2016 		return 0;
2017 	return 1;
2018 }
2019 
fill_note_info(struct elfhdr * elf,int phdrs,struct elf_note_info * info,const kernel_siginfo_t * siginfo,struct pt_regs * regs)2020 static int fill_note_info(struct elfhdr *elf, int phdrs,
2021 			  struct elf_note_info *info,
2022 			  const kernel_siginfo_t *siginfo, struct pt_regs *regs)
2023 {
2024 	struct core_thread *ct;
2025 	struct elf_thread_status *ets;
2026 
2027 	if (!elf_note_info_init(info))
2028 		return 0;
2029 
2030 	for (ct = current->mm->core_state->dumper.next;
2031 					ct; ct = ct->next) {
2032 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2033 		if (!ets)
2034 			return 0;
2035 
2036 		ets->thread = ct->task;
2037 		list_add(&ets->list, &info->thread_list);
2038 	}
2039 
2040 	list_for_each_entry(ets, &info->thread_list, list) {
2041 		int sz;
2042 
2043 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
2044 		info->thread_status_size += sz;
2045 	}
2046 	/* now collect the dump for the current */
2047 	memset(info->prstatus, 0, sizeof(*info->prstatus));
2048 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
2049 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2050 
2051 	/* Set up header */
2052 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2053 
2054 	/*
2055 	 * Set up the notes in similar form to SVR4 core dumps made
2056 	 * with info from their /proc.
2057 	 */
2058 
2059 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2060 		  sizeof(*info->prstatus), info->prstatus);
2061 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
2062 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2063 		  sizeof(*info->psinfo), info->psinfo);
2064 
2065 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2066 	fill_auxv_note(info->notes + 3, current->mm);
2067 	info->numnote = 4;
2068 
2069 	if (fill_files_note(info->notes + info->numnote) == 0) {
2070 		info->notes_files = info->notes + info->numnote;
2071 		info->numnote++;
2072 	}
2073 
2074 	/* Try to dump the FPU. */
2075 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2076 							       info->fpu);
2077 	if (info->prstatus->pr_fpvalid)
2078 		fill_note(info->notes + info->numnote++,
2079 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2080 	return 1;
2081 }
2082 
get_note_info_size(struct elf_note_info * info)2083 static size_t get_note_info_size(struct elf_note_info *info)
2084 {
2085 	int sz = 0;
2086 	int i;
2087 
2088 	for (i = 0; i < info->numnote; i++)
2089 		sz += notesize(info->notes + i);
2090 
2091 	sz += info->thread_status_size;
2092 
2093 	return sz;
2094 }
2095 
write_note_info(struct elf_note_info * info,struct coredump_params * cprm)2096 static int write_note_info(struct elf_note_info *info,
2097 			   struct coredump_params *cprm)
2098 {
2099 	struct elf_thread_status *ets;
2100 	int i;
2101 
2102 	for (i = 0; i < info->numnote; i++)
2103 		if (!writenote(info->notes + i, cprm))
2104 			return 0;
2105 
2106 	/* write out the thread status notes section */
2107 	list_for_each_entry(ets, &info->thread_list, list) {
2108 		for (i = 0; i < ets->num_notes; i++)
2109 			if (!writenote(&ets->notes[i], cprm))
2110 				return 0;
2111 	}
2112 
2113 	return 1;
2114 }
2115 
free_note_info(struct elf_note_info * info)2116 static void free_note_info(struct elf_note_info *info)
2117 {
2118 	while (!list_empty(&info->thread_list)) {
2119 		struct list_head *tmp = info->thread_list.next;
2120 		list_del(tmp);
2121 		kfree(list_entry(tmp, struct elf_thread_status, list));
2122 	}
2123 
2124 	/* Free data possibly allocated by fill_files_note(): */
2125 	if (info->notes_files)
2126 		kvfree(info->notes_files->data);
2127 
2128 	kfree(info->prstatus);
2129 	kfree(info->psinfo);
2130 	kfree(info->notes);
2131 	kfree(info->fpu);
2132 }
2133 
2134 #endif
2135 
fill_extnum_info(struct elfhdr * elf,struct elf_shdr * shdr4extnum,elf_addr_t e_shoff,int segs)2136 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2137 			     elf_addr_t e_shoff, int segs)
2138 {
2139 	elf->e_shoff = e_shoff;
2140 	elf->e_shentsize = sizeof(*shdr4extnum);
2141 	elf->e_shnum = 1;
2142 	elf->e_shstrndx = SHN_UNDEF;
2143 
2144 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2145 
2146 	shdr4extnum->sh_type = SHT_NULL;
2147 	shdr4extnum->sh_size = elf->e_shnum;
2148 	shdr4extnum->sh_link = elf->e_shstrndx;
2149 	shdr4extnum->sh_info = segs;
2150 }
2151 
2152 /*
2153  * Actual dumper
2154  *
2155  * This is a two-pass process; first we find the offsets of the bits,
2156  * and then they are actually written out.  If we run out of core limit
2157  * we just truncate.
2158  */
elf_core_dump(struct coredump_params * cprm)2159 static int elf_core_dump(struct coredump_params *cprm)
2160 {
2161 	int has_dumped = 0;
2162 	int vma_count, segs, i;
2163 	size_t vma_data_size;
2164 	struct elfhdr elf;
2165 	loff_t offset = 0, dataoff;
2166 	struct elf_note_info info = { };
2167 	struct elf_phdr *phdr4note = NULL;
2168 	struct elf_shdr *shdr4extnum = NULL;
2169 	Elf_Half e_phnum;
2170 	elf_addr_t e_shoff;
2171 	struct core_vma_metadata *vma_meta;
2172 
2173 	if (dump_vma_snapshot(cprm, &vma_count, &vma_meta, &vma_data_size))
2174 		return 0;
2175 
2176 	/*
2177 	 * The number of segs are recored into ELF header as 16bit value.
2178 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2179 	 */
2180 	segs = vma_count + elf_core_extra_phdrs();
2181 
2182 	/* for notes section */
2183 	segs++;
2184 
2185 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2186 	 * this, kernel supports extended numbering. Have a look at
2187 	 * include/linux/elf.h for further information. */
2188 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2189 
2190 	/*
2191 	 * Collect all the non-memory information about the process for the
2192 	 * notes.  This also sets up the file header.
2193 	 */
2194 	if (!fill_note_info(&elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2195 		goto end_coredump;
2196 
2197 	has_dumped = 1;
2198 
2199 	offset += sizeof(elf);				/* Elf header */
2200 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2201 
2202 	/* Write notes phdr entry */
2203 	{
2204 		size_t sz = get_note_info_size(&info);
2205 
2206 		sz += elf_coredump_extra_notes_size();
2207 
2208 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2209 		if (!phdr4note)
2210 			goto end_coredump;
2211 
2212 		fill_elf_note_phdr(phdr4note, sz, offset);
2213 		offset += sz;
2214 	}
2215 
2216 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2217 
2218 	offset += vma_data_size;
2219 	offset += elf_core_extra_data_size();
2220 	e_shoff = offset;
2221 
2222 	if (e_phnum == PN_XNUM) {
2223 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2224 		if (!shdr4extnum)
2225 			goto end_coredump;
2226 		fill_extnum_info(&elf, shdr4extnum, e_shoff, segs);
2227 	}
2228 
2229 	offset = dataoff;
2230 
2231 	if (!dump_emit(cprm, &elf, sizeof(elf)))
2232 		goto end_coredump;
2233 
2234 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2235 		goto end_coredump;
2236 
2237 	/* Write program headers for segments dump */
2238 	for (i = 0; i < vma_count; i++) {
2239 		struct core_vma_metadata *meta = vma_meta + i;
2240 		struct elf_phdr phdr;
2241 
2242 		phdr.p_type = PT_LOAD;
2243 		phdr.p_offset = offset;
2244 		phdr.p_vaddr = meta->start;
2245 		phdr.p_paddr = 0;
2246 		phdr.p_filesz = meta->dump_size;
2247 		phdr.p_memsz = meta->end - meta->start;
2248 		offset += phdr.p_filesz;
2249 		phdr.p_flags = 0;
2250 		if (meta->flags & VM_READ)
2251 			phdr.p_flags |= PF_R;
2252 		if (meta->flags & VM_WRITE)
2253 			phdr.p_flags |= PF_W;
2254 		if (meta->flags & VM_EXEC)
2255 			phdr.p_flags |= PF_X;
2256 		phdr.p_align = ELF_EXEC_PAGESIZE;
2257 
2258 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2259 			goto end_coredump;
2260 	}
2261 
2262 	if (!elf_core_write_extra_phdrs(cprm, offset))
2263 		goto end_coredump;
2264 
2265  	/* write out the notes section */
2266 	if (!write_note_info(&info, cprm))
2267 		goto end_coredump;
2268 
2269 	if (elf_coredump_extra_notes_write(cprm))
2270 		goto end_coredump;
2271 
2272 	/* Align to page */
2273 	if (!dump_skip(cprm, dataoff - cprm->pos))
2274 		goto end_coredump;
2275 
2276 	for (i = 0; i < vma_count; i++) {
2277 		struct core_vma_metadata *meta = vma_meta + i;
2278 
2279 		if (!dump_user_range(cprm, meta->start, meta->dump_size))
2280 			goto end_coredump;
2281 	}
2282 	dump_truncate(cprm);
2283 
2284 	if (!elf_core_write_extra_data(cprm))
2285 		goto end_coredump;
2286 
2287 	if (e_phnum == PN_XNUM) {
2288 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2289 			goto end_coredump;
2290 	}
2291 
2292 end_coredump:
2293 	free_note_info(&info);
2294 	kfree(shdr4extnum);
2295 	kvfree(vma_meta);
2296 	kfree(phdr4note);
2297 	return has_dumped;
2298 }
2299 
2300 #endif		/* CONFIG_ELF_CORE */
2301 
init_elf_binfmt(void)2302 static int __init init_elf_binfmt(void)
2303 {
2304 	register_binfmt(&elf_format);
2305 	return 0;
2306 }
2307 
exit_elf_binfmt(void)2308 static void __exit exit_elf_binfmt(void)
2309 {
2310 	/* Remove the COFF and ELF loaders. */
2311 	unregister_binfmt(&elf_format);
2312 }
2313 
2314 core_initcall(init_elf_binfmt);
2315 module_exit(exit_elf_binfmt);
2316 MODULE_LICENSE("GPL");
2317