2 * linux/fs/binfmt_elf.c
4 * These are the functions used to load ELF format executables as used
5 * on SVr4 machines. Information on the format may be found in the book
6 * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
9 * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
12 #include <linux/module.h>
13 #include <linux/kernel.h>
16 #include <linux/mman.h>
17 #include <linux/errno.h>
18 #include <linux/signal.h>
19 #include <linux/binfmts.h>
20 #include <linux/string.h>
21 #include <linux/file.h>
22 #include <linux/slab.h>
23 #include <linux/personality.h>
24 #include <linux/elfcore.h>
25 #include <linux/init.h>
26 #include <linux/highuid.h>
27 #include <linux/compiler.h>
28 #include <linux/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/vmalloc.h>
31 #include <linux/security.h>
32 #include <linux/random.h>
33 #include <linux/elf.h>
34 #include <linux/elf-randomize.h>
35 #include <linux/utsname.h>
36 #include <linux/coredump.h>
37 #include <linux/sched.h>
38 #include <linux/sched/coredump.h>
39 #include <linux/sched/task_stack.h>
40 #include <linux/sched/cputime.h>
41 #include <linux/cred.h>
42 #include <linux/dax.h>
43 #include <linux/uaccess.h>
44 #include <asm/param.h>
48 #define user_long_t long
50 #ifndef user_siginfo_t
51 #define user_siginfo_t siginfo_t
54 /* That's for binfmt_elf_fdpic to deal with */
55 #ifndef elf_check_fdpic
56 #define elf_check_fdpic(ex) false
59 static int load_elf_binary(struct linux_binprm *bprm);
60 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
61 int, int, unsigned long);
64 static int load_elf_library(struct file *);
66 #define load_elf_library NULL
70 * If we don't support core dumping, then supply a NULL so we
73 #ifdef CONFIG_ELF_CORE
74 static int elf_core_dump(struct coredump_params *cprm);
76 #define elf_core_dump NULL
79 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
80 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
82 #define ELF_MIN_ALIGN PAGE_SIZE
85 #ifndef ELF_CORE_EFLAGS
86 #define ELF_CORE_EFLAGS 0
89 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
90 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
91 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
93 static struct linux_binfmt elf_format = {
94 .module = THIS_MODULE,
95 .load_binary = load_elf_binary,
96 .load_shlib = load_elf_library,
97 .core_dump = elf_core_dump,
98 .min_coredump = ELF_EXEC_PAGESIZE,
101 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
103 static int set_brk(unsigned long start, unsigned long end, int prot)
105 start = ELF_PAGEALIGN(start);
106 end = ELF_PAGEALIGN(end);
109 * Map the last of the bss segment.
110 * If the header is requesting these pages to be
111 * executable, honour that (ppc32 needs this).
113 int error = vm_brk_flags(start, end - start,
114 prot & PROT_EXEC ? VM_EXEC : 0);
118 current->mm->start_brk = current->mm->brk = end;
122 /* We need to explicitly zero any fractional pages
123 after the data section (i.e. bss). This would
124 contain the junk from the file that should not
127 static int padzero(unsigned long elf_bss)
131 nbyte = ELF_PAGEOFFSET(elf_bss);
133 nbyte = ELF_MIN_ALIGN - nbyte;
134 if (clear_user((void __user *) elf_bss, nbyte))
140 /* Let's use some macros to make this stack manipulation a little clearer */
141 #ifdef CONFIG_STACK_GROWSUP
142 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
143 #define STACK_ROUND(sp, items) \
144 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
145 #define STACK_ALLOC(sp, len) ({ \
146 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
149 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
150 #define STACK_ROUND(sp, items) \
151 (((unsigned long) (sp - items)) &~ 15UL)
152 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
155 #ifndef ELF_BASE_PLATFORM
157 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
158 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
159 * will be copied to the user stack in the same manner as AT_PLATFORM.
161 #define ELF_BASE_PLATFORM NULL
165 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
166 unsigned long load_addr, unsigned long interp_load_addr)
168 unsigned long p = bprm->p;
169 int argc = bprm->argc;
170 int envc = bprm->envc;
171 elf_addr_t __user *sp;
172 elf_addr_t __user *u_platform;
173 elf_addr_t __user *u_base_platform;
174 elf_addr_t __user *u_rand_bytes;
175 const char *k_platform = ELF_PLATFORM;
176 const char *k_base_platform = ELF_BASE_PLATFORM;
177 unsigned char k_rand_bytes[16];
179 elf_addr_t *elf_info;
181 const struct cred *cred = current_cred();
182 struct vm_area_struct *vma;
185 * In some cases (e.g. Hyper-Threading), we want to avoid L1
186 * evictions by the processes running on the same package. One
187 * thing we can do is to shuffle the initial stack for them.
190 p = arch_align_stack(p);
193 * If this architecture has a platform capability string, copy it
194 * to userspace. In some cases (Sparc), this info is impossible
195 * for userspace to get any other way, in others (i386) it is
200 size_t len = strlen(k_platform) + 1;
202 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
203 if (__copy_to_user(u_platform, k_platform, len))
208 * If this architecture has a "base" platform capability
209 * string, copy it to userspace.
211 u_base_platform = NULL;
212 if (k_base_platform) {
213 size_t len = strlen(k_base_platform) + 1;
215 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
216 if (__copy_to_user(u_base_platform, k_base_platform, len))
221 * Generate 16 random bytes for userspace PRNG seeding.
223 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
224 u_rand_bytes = (elf_addr_t __user *)
225 STACK_ALLOC(p, sizeof(k_rand_bytes));
226 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
229 /* Create the ELF interpreter info */
230 elf_info = (elf_addr_t *)current->mm->saved_auxv;
231 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
232 #define NEW_AUX_ENT(id, val) \
234 elf_info[ei_index++] = id; \
235 elf_info[ei_index++] = val; \
240 * ARCH_DLINFO must come first so PPC can do its special alignment of
242 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
243 * ARCH_DLINFO changes
247 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
248 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
249 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
250 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
251 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
252 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
253 NEW_AUX_ENT(AT_BASE, interp_load_addr);
254 NEW_AUX_ENT(AT_FLAGS, 0);
255 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
256 NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
257 NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
258 NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
259 NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
260 NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
261 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
263 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
265 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
267 NEW_AUX_ENT(AT_PLATFORM,
268 (elf_addr_t)(unsigned long)u_platform);
270 if (k_base_platform) {
271 NEW_AUX_ENT(AT_BASE_PLATFORM,
272 (elf_addr_t)(unsigned long)u_base_platform);
274 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
275 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
278 /* AT_NULL is zero; clear the rest too */
279 memset(&elf_info[ei_index], 0,
280 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
282 /* And advance past the AT_NULL entry. */
285 sp = STACK_ADD(p, ei_index);
287 items = (argc + 1) + (envc + 1) + 1;
288 bprm->p = STACK_ROUND(sp, items);
290 /* Point sp at the lowest address on the stack */
291 #ifdef CONFIG_STACK_GROWSUP
292 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
293 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
295 sp = (elf_addr_t __user *)bprm->p;
300 * Grow the stack manually; some architectures have a limit on how
301 * far ahead a user-space access may be in order to grow the stack.
303 vma = find_extend_vma(current->mm, bprm->p);
307 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
308 if (__put_user(argc, sp++))
311 /* Populate list of argv pointers back to argv strings. */
312 p = current->mm->arg_end = current->mm->arg_start;
315 if (__put_user((elf_addr_t)p, sp++))
317 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
318 if (!len || len > MAX_ARG_STRLEN)
322 if (__put_user(0, sp++))
324 current->mm->arg_end = p;
326 /* Populate list of envp pointers back to envp strings. */
327 current->mm->env_end = current->mm->env_start = p;
330 if (__put_user((elf_addr_t)p, sp++))
332 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
333 if (!len || len > MAX_ARG_STRLEN)
337 if (__put_user(0, sp++))
339 current->mm->env_end = p;
341 /* Put the elf_info on the stack in the right place. */
342 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
349 static unsigned long elf_map(struct file *filep, unsigned long addr,
350 struct elf_phdr *eppnt, int prot, int type,
351 unsigned long total_size)
353 unsigned long map_addr;
354 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
355 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
356 addr = ELF_PAGESTART(addr);
357 size = ELF_PAGEALIGN(size);
359 /* mmap() will return -EINVAL if given a zero size, but a
360 * segment with zero filesize is perfectly valid */
365 * total_size is the size of the ELF (interpreter) image.
366 * The _first_ mmap needs to know the full size, otherwise
367 * randomization might put this image into an overlapping
368 * position with the ELF binary image. (since size < total_size)
369 * So we first map the 'big' image - and unmap the remainder at
370 * the end. (which unmap is needed for ELF images with holes.)
373 total_size = ELF_PAGEALIGN(total_size);
374 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
375 if (!BAD_ADDR(map_addr))
376 vm_munmap(map_addr+size, total_size-size);
378 map_addr = vm_mmap(filep, addr, size, prot, type, off);
380 if ((type & MAP_FIXED_NOREPLACE) && BAD_ADDR(map_addr))
381 pr_info("%d (%s): Uhuuh, elf segment at %p requested but the memory is mapped already\n",
382 task_pid_nr(current), current->comm,
388 #endif /* !elf_map */
390 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
392 int i, first_idx = -1, last_idx = -1;
394 for (i = 0; i < nr; i++) {
395 if (cmds[i].p_type == PT_LOAD) {
404 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
405 ELF_PAGESTART(cmds[first_idx].p_vaddr);
409 * load_elf_phdrs() - load ELF program headers
410 * @elf_ex: ELF header of the binary whose program headers should be loaded
411 * @elf_file: the opened ELF binary file
413 * Loads ELF program headers from the binary file elf_file, which has the ELF
414 * header pointed to by elf_ex, into a newly allocated array. The caller is
415 * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
417 static struct elf_phdr *load_elf_phdrs(struct elfhdr *elf_ex,
418 struct file *elf_file)
420 struct elf_phdr *elf_phdata = NULL;
421 int retval, size, err = -1;
422 loff_t pos = elf_ex->e_phoff;
425 * If the size of this structure has changed, then punt, since
426 * we will be doing the wrong thing.
428 if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
431 /* Sanity check the number of program headers... */
432 if (elf_ex->e_phnum < 1 ||
433 elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
436 /* ...and their total size. */
437 size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
438 if (size > ELF_MIN_ALIGN)
441 elf_phdata = kmalloc(size, GFP_KERNEL);
445 /* Read in the program headers */
446 retval = kernel_read(elf_file, elf_phdata, size, &pos);
447 if (retval != size) {
448 err = (retval < 0) ? retval : -EIO;
462 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
465 * struct arch_elf_state - arch-specific ELF loading state
467 * This structure is used to preserve architecture specific data during
468 * the loading of an ELF file, throughout the checking of architecture
469 * specific ELF headers & through to the point where the ELF load is
470 * known to be proceeding (ie. SET_PERSONALITY).
472 * This implementation is a dummy for architectures which require no
475 struct arch_elf_state {
478 #define INIT_ARCH_ELF_STATE {}
481 * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
482 * @ehdr: The main ELF header
483 * @phdr: The program header to check
484 * @elf: The open ELF file
485 * @is_interp: True if the phdr is from the interpreter of the ELF being
486 * loaded, else false.
487 * @state: Architecture-specific state preserved throughout the process
488 * of loading the ELF.
490 * Inspects the program header phdr to validate its correctness and/or
491 * suitability for the system. Called once per ELF program header in the
492 * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
495 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
496 * with that return code.
498 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
499 struct elf_phdr *phdr,
500 struct file *elf, bool is_interp,
501 struct arch_elf_state *state)
503 /* Dummy implementation, always proceed */
508 * arch_check_elf() - check an ELF executable
509 * @ehdr: The main ELF header
510 * @has_interp: True if the ELF has an interpreter, else false.
511 * @interp_ehdr: The interpreter's ELF header
512 * @state: Architecture-specific state preserved throughout the process
513 * of loading the ELF.
515 * Provides a final opportunity for architecture code to reject the loading
516 * of the ELF & cause an exec syscall to return an error. This is called after
517 * all program headers to be checked by arch_elf_pt_proc have been.
519 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
520 * with that return code.
522 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
523 struct elfhdr *interp_ehdr,
524 struct arch_elf_state *state)
526 /* Dummy implementation, always proceed */
530 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
532 /* This is much more generalized than the library routine read function,
533 so we keep this separate. Technically the library read function
534 is only provided so that we can read a.out libraries that have
537 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
538 struct file *interpreter, unsigned long *interp_map_addr,
539 unsigned long no_base, struct elf_phdr *interp_elf_phdata)
541 struct elf_phdr *eppnt;
542 unsigned long load_addr = 0;
543 int load_addr_set = 0;
544 unsigned long last_bss = 0, elf_bss = 0;
546 unsigned long error = ~0UL;
547 unsigned long total_size;
550 /* First of all, some simple consistency checks */
551 if (interp_elf_ex->e_type != ET_EXEC &&
552 interp_elf_ex->e_type != ET_DYN)
554 if (!elf_check_arch(interp_elf_ex) ||
555 elf_check_fdpic(interp_elf_ex))
557 if (!interpreter->f_op->mmap)
560 total_size = total_mapping_size(interp_elf_phdata,
561 interp_elf_ex->e_phnum);
567 eppnt = interp_elf_phdata;
568 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
569 if (eppnt->p_type == PT_LOAD) {
570 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
572 unsigned long vaddr = 0;
573 unsigned long k, map_addr;
575 if (eppnt->p_flags & PF_R)
576 elf_prot = PROT_READ;
577 if (eppnt->p_flags & PF_W)
578 elf_prot |= PROT_WRITE;
579 if (eppnt->p_flags & PF_X)
580 elf_prot |= PROT_EXEC;
581 vaddr = eppnt->p_vaddr;
582 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
583 elf_type |= MAP_FIXED_NOREPLACE;
584 else if (no_base && interp_elf_ex->e_type == ET_DYN)
587 map_addr = elf_map(interpreter, load_addr + vaddr,
588 eppnt, elf_prot, elf_type, total_size);
590 if (!*interp_map_addr)
591 *interp_map_addr = map_addr;
593 if (BAD_ADDR(map_addr))
596 if (!load_addr_set &&
597 interp_elf_ex->e_type == ET_DYN) {
598 load_addr = map_addr - ELF_PAGESTART(vaddr);
603 * Check to see if the section's size will overflow the
604 * allowed task size. Note that p_filesz must always be
605 * <= p_memsize so it's only necessary to check p_memsz.
607 k = load_addr + eppnt->p_vaddr;
609 eppnt->p_filesz > eppnt->p_memsz ||
610 eppnt->p_memsz > TASK_SIZE ||
611 TASK_SIZE - eppnt->p_memsz < k) {
617 * Find the end of the file mapping for this phdr, and
618 * keep track of the largest address we see for this.
620 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
625 * Do the same thing for the memory mapping - between
626 * elf_bss and last_bss is the bss section.
628 k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
637 * Now fill out the bss section: first pad the last page from
638 * the file up to the page boundary, and zero it from elf_bss
639 * up to the end of the page.
641 if (padzero(elf_bss)) {
646 * Next, align both the file and mem bss up to the page size,
647 * since this is where elf_bss was just zeroed up to, and where
648 * last_bss will end after the vm_brk_flags() below.
650 elf_bss = ELF_PAGEALIGN(elf_bss);
651 last_bss = ELF_PAGEALIGN(last_bss);
652 /* Finally, if there is still more bss to allocate, do it. */
653 if (last_bss > elf_bss) {
654 error = vm_brk_flags(elf_bss, last_bss - elf_bss,
655 bss_prot & PROT_EXEC ? VM_EXEC : 0);
666 * These are the functions used to load ELF style executables and shared
667 * libraries. There is no binary dependent code anywhere else.
670 #ifndef STACK_RND_MASK
671 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
674 static unsigned long randomize_stack_top(unsigned long stack_top)
676 unsigned long random_variable = 0;
678 if (current->flags & PF_RANDOMIZE) {
679 random_variable = get_random_long();
680 random_variable &= STACK_RND_MASK;
681 random_variable <<= PAGE_SHIFT;
683 #ifdef CONFIG_STACK_GROWSUP
684 return PAGE_ALIGN(stack_top) + random_variable;
686 return PAGE_ALIGN(stack_top) - random_variable;
690 static int load_elf_binary(struct linux_binprm *bprm)
692 struct file *interpreter = NULL; /* to shut gcc up */
693 unsigned long load_addr = 0, load_bias = 0;
694 int load_addr_set = 0;
695 char * elf_interpreter = NULL;
697 struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
698 unsigned long elf_bss, elf_brk;
701 unsigned long elf_entry;
702 unsigned long interp_load_addr = 0;
703 unsigned long start_code, end_code, start_data, end_data;
704 unsigned long reloc_func_desc __maybe_unused = 0;
705 int executable_stack = EXSTACK_DEFAULT;
706 struct pt_regs *regs = current_pt_regs();
708 struct elfhdr elf_ex;
709 struct elfhdr interp_elf_ex;
711 struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
714 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
720 /* Get the exec-header */
721 loc->elf_ex = *((struct elfhdr *)bprm->buf);
724 /* First of all, some simple consistency checks */
725 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
728 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
730 if (!elf_check_arch(&loc->elf_ex))
732 if (elf_check_fdpic(&loc->elf_ex))
734 if (!bprm->file->f_op->mmap)
737 elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
741 elf_ppnt = elf_phdata;
750 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
751 if (elf_ppnt->p_type == PT_INTERP) {
752 /* This is the program interpreter used for
753 * shared libraries - for now assume that this
754 * is an a.out format binary
757 if (elf_ppnt->p_filesz > PATH_MAX ||
758 elf_ppnt->p_filesz < 2)
762 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
764 if (!elf_interpreter)
767 pos = elf_ppnt->p_offset;
768 retval = kernel_read(bprm->file, elf_interpreter,
769 elf_ppnt->p_filesz, &pos);
770 if (retval != elf_ppnt->p_filesz) {
773 goto out_free_interp;
775 /* make sure path is NULL terminated */
777 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
778 goto out_free_interp;
780 interpreter = open_exec(elf_interpreter);
781 retval = PTR_ERR(interpreter);
782 if (IS_ERR(interpreter))
783 goto out_free_interp;
786 * If the binary is not readable then enforce
787 * mm->dumpable = 0 regardless of the interpreter's
790 would_dump(bprm, interpreter);
792 /* Get the exec headers */
794 retval = kernel_read(interpreter, &loc->interp_elf_ex,
795 sizeof(loc->interp_elf_ex), &pos);
796 if (retval != sizeof(loc->interp_elf_ex)) {
799 goto out_free_dentry;
807 elf_ppnt = elf_phdata;
808 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
809 switch (elf_ppnt->p_type) {
811 if (elf_ppnt->p_flags & PF_X)
812 executable_stack = EXSTACK_ENABLE_X;
814 executable_stack = EXSTACK_DISABLE_X;
817 case PT_LOPROC ... PT_HIPROC:
818 retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
822 goto out_free_dentry;
826 /* Some simple consistency checks for the interpreter */
827 if (elf_interpreter) {
829 /* Not an ELF interpreter */
830 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
831 goto out_free_dentry;
832 /* Verify the interpreter has a valid arch */
833 if (!elf_check_arch(&loc->interp_elf_ex) ||
834 elf_check_fdpic(&loc->interp_elf_ex))
835 goto out_free_dentry;
837 /* Load the interpreter program headers */
838 interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
840 if (!interp_elf_phdata)
841 goto out_free_dentry;
843 /* Pass PT_LOPROC..PT_HIPROC headers to arch code */
844 elf_ppnt = interp_elf_phdata;
845 for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
846 switch (elf_ppnt->p_type) {
847 case PT_LOPROC ... PT_HIPROC:
848 retval = arch_elf_pt_proc(&loc->interp_elf_ex,
849 elf_ppnt, interpreter,
852 goto out_free_dentry;
858 * Allow arch code to reject the ELF at this point, whilst it's
859 * still possible to return an error to the code that invoked
862 retval = arch_check_elf(&loc->elf_ex,
863 !!interpreter, &loc->interp_elf_ex,
866 goto out_free_dentry;
868 /* Flush all traces of the currently running executable */
869 retval = flush_old_exec(bprm);
871 goto out_free_dentry;
873 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
874 may depend on the personality. */
875 SET_PERSONALITY2(loc->elf_ex, &arch_state);
876 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
877 current->personality |= READ_IMPLIES_EXEC;
879 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
880 current->flags |= PF_RANDOMIZE;
882 setup_new_exec(bprm);
883 install_exec_creds(bprm);
885 /* Do this so that we can load the interpreter, if need be. We will
886 change some of these later */
887 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
890 goto out_free_dentry;
892 current->mm->start_stack = bprm->p;
894 /* Now we do a little grungy work by mmapping the ELF image into
895 the correct location in memory. */
896 for(i = 0, elf_ppnt = elf_phdata;
897 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
898 int elf_prot = 0, elf_flags, elf_fixed = MAP_FIXED_NOREPLACE;
899 unsigned long k, vaddr;
900 unsigned long total_size = 0;
902 if (elf_ppnt->p_type != PT_LOAD)
905 if (unlikely (elf_brk > elf_bss)) {
908 /* There was a PT_LOAD segment with p_memsz > p_filesz
909 before this one. Map anonymous pages, if needed,
910 and clear the area. */
911 retval = set_brk(elf_bss + load_bias,
915 goto out_free_dentry;
916 nbyte = ELF_PAGEOFFSET(elf_bss);
918 nbyte = ELF_MIN_ALIGN - nbyte;
919 if (nbyte > elf_brk - elf_bss)
920 nbyte = elf_brk - elf_bss;
921 if (clear_user((void __user *)elf_bss +
924 * This bss-zeroing can fail if the ELF
925 * file specifies odd protections. So
926 * we don't check the return value
932 * Some binaries have overlapping elf segments and then
933 * we have to forcefully map over an existing mapping
934 * e.g. over this newly established brk mapping.
936 elf_fixed = MAP_FIXED;
939 if (elf_ppnt->p_flags & PF_R)
940 elf_prot |= PROT_READ;
941 if (elf_ppnt->p_flags & PF_W)
942 elf_prot |= PROT_WRITE;
943 if (elf_ppnt->p_flags & PF_X)
944 elf_prot |= PROT_EXEC;
946 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
948 vaddr = elf_ppnt->p_vaddr;
950 * If we are loading ET_EXEC or we have already performed
951 * the ET_DYN load_addr calculations, proceed normally.
953 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
954 elf_flags |= elf_fixed;
955 } else if (loc->elf_ex.e_type == ET_DYN) {
957 * This logic is run once for the first LOAD Program
958 * Header for ET_DYN binaries to calculate the
959 * randomization (load_bias) for all the LOAD
960 * Program Headers, and to calculate the entire
961 * size of the ELF mapping (total_size). (Note that
962 * load_addr_set is set to true later once the
963 * initial mapping is performed.)
965 * There are effectively two types of ET_DYN
966 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
967 * and loaders (ET_DYN without INTERP, since they
968 * _are_ the ELF interpreter). The loaders must
969 * be loaded away from programs since the program
970 * may otherwise collide with the loader (especially
971 * for ET_EXEC which does not have a randomized
972 * position). For example to handle invocations of
973 * "./ld.so someprog" to test out a new version of
974 * the loader, the subsequent program that the
975 * loader loads must avoid the loader itself, so
976 * they cannot share the same load range. Sufficient
977 * room for the brk must be allocated with the
978 * loader as well, since brk must be available with
981 * Therefore, programs are loaded offset from
982 * ELF_ET_DYN_BASE and loaders are loaded into the
983 * independently randomized mmap region (0 load_bias
984 * without MAP_FIXED).
986 if (elf_interpreter) {
987 load_bias = ELF_ET_DYN_BASE;
988 if (current->flags & PF_RANDOMIZE)
989 load_bias += arch_mmap_rnd();
990 elf_flags |= elf_fixed;
995 * Since load_bias is used for all subsequent loading
996 * calculations, we must lower it by the first vaddr
997 * so that the remaining calculations based on the
998 * ELF vaddrs will be correctly offset. The result
999 * is then page aligned.
1001 load_bias = ELF_PAGESTART(load_bias - vaddr);
1003 total_size = total_mapping_size(elf_phdata,
1004 loc->elf_ex.e_phnum);
1007 goto out_free_dentry;
1011 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1012 elf_prot, elf_flags, total_size);
1013 if (BAD_ADDR(error)) {
1014 retval = IS_ERR((void *)error) ?
1015 PTR_ERR((void*)error) : -EINVAL;
1016 goto out_free_dentry;
1019 if (!load_addr_set) {
1021 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1022 if (loc->elf_ex.e_type == ET_DYN) {
1023 load_bias += error -
1024 ELF_PAGESTART(load_bias + vaddr);
1025 load_addr += load_bias;
1026 reloc_func_desc = load_bias;
1029 k = elf_ppnt->p_vaddr;
1036 * Check to see if the section's size will overflow the
1037 * allowed task size. Note that p_filesz must always be
1038 * <= p_memsz so it is only necessary to check p_memsz.
1040 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1041 elf_ppnt->p_memsz > TASK_SIZE ||
1042 TASK_SIZE - elf_ppnt->p_memsz < k) {
1043 /* set_brk can never work. Avoid overflows. */
1045 goto out_free_dentry;
1048 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1052 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1056 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1058 bss_prot = elf_prot;
1063 loc->elf_ex.e_entry += load_bias;
1064 elf_bss += load_bias;
1065 elf_brk += load_bias;
1066 start_code += load_bias;
1067 end_code += load_bias;
1068 start_data += load_bias;
1069 end_data += load_bias;
1071 /* Calling set_brk effectively mmaps the pages that we need
1072 * for the bss and break sections. We must do this before
1073 * mapping in the interpreter, to make sure it doesn't wind
1074 * up getting placed where the bss needs to go.
1076 retval = set_brk(elf_bss, elf_brk, bss_prot);
1078 goto out_free_dentry;
1079 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1080 retval = -EFAULT; /* Nobody gets to see this, but.. */
1081 goto out_free_dentry;
1084 if (elf_interpreter) {
1085 unsigned long interp_map_addr = 0;
1087 elf_entry = load_elf_interp(&loc->interp_elf_ex,
1090 load_bias, interp_elf_phdata);
1091 if (!IS_ERR((void *)elf_entry)) {
1093 * load_elf_interp() returns relocation
1096 interp_load_addr = elf_entry;
1097 elf_entry += loc->interp_elf_ex.e_entry;
1099 if (BAD_ADDR(elf_entry)) {
1100 retval = IS_ERR((void *)elf_entry) ?
1101 (int)elf_entry : -EINVAL;
1102 goto out_free_dentry;
1104 reloc_func_desc = interp_load_addr;
1106 allow_write_access(interpreter);
1108 kfree(elf_interpreter);
1110 elf_entry = loc->elf_ex.e_entry;
1111 if (BAD_ADDR(elf_entry)) {
1113 goto out_free_dentry;
1117 kfree(interp_elf_phdata);
1120 set_binfmt(&elf_format);
1122 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1123 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1126 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1128 retval = create_elf_tables(bprm, &loc->elf_ex,
1129 load_addr, interp_load_addr);
1132 /* N.B. passed_fileno might not be initialized? */
1133 current->mm->end_code = end_code;
1134 current->mm->start_code = start_code;
1135 current->mm->start_data = start_data;
1136 current->mm->end_data = end_data;
1137 current->mm->start_stack = bprm->p;
1139 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1140 current->mm->brk = current->mm->start_brk =
1141 arch_randomize_brk(current->mm);
1142 #ifdef compat_brk_randomized
1143 current->brk_randomized = 1;
1147 if (current->personality & MMAP_PAGE_ZERO) {
1148 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1149 and some applications "depend" upon this behavior.
1150 Since we do not have the power to recompile these, we
1151 emulate the SVr4 behavior. Sigh. */
1152 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1153 MAP_FIXED | MAP_PRIVATE, 0);
1156 #ifdef ELF_PLAT_INIT
1158 * The ABI may specify that certain registers be set up in special
1159 * ways (on i386 %edx is the address of a DT_FINI function, for
1160 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1161 * that the e_entry field is the address of the function descriptor
1162 * for the startup routine, rather than the address of the startup
1163 * routine itself. This macro performs whatever initialization to
1164 * the regs structure is required as well as any relocations to the
1165 * function descriptor entries when executing dynamically links apps.
1167 ELF_PLAT_INIT(regs, reloc_func_desc);
1170 finalize_exec(bprm);
1171 start_thread(regs, elf_entry, bprm->p);
1180 kfree(interp_elf_phdata);
1181 allow_write_access(interpreter);
1185 kfree(elf_interpreter);
1191 #ifdef CONFIG_USELIB
1192 /* This is really simpleminded and specialized - we are loading an
1193 a.out library that is given an ELF header. */
1194 static int load_elf_library(struct file *file)
1196 struct elf_phdr *elf_phdata;
1197 struct elf_phdr *eppnt;
1198 unsigned long elf_bss, bss, len;
1199 int retval, error, i, j;
1200 struct elfhdr elf_ex;
1204 retval = kernel_read(file, &elf_ex, sizeof(elf_ex), &pos);
1205 if (retval != sizeof(elf_ex))
1208 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1211 /* First of all, some simple consistency checks */
1212 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1213 !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1215 if (elf_check_fdpic(&elf_ex))
1218 /* Now read in all of the header information */
1220 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1221 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1224 elf_phdata = kmalloc(j, GFP_KERNEL);
1230 pos = elf_ex.e_phoff;
1231 retval = kernel_read(file, eppnt, j, &pos);
1235 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1236 if ((eppnt + i)->p_type == PT_LOAD)
1241 while (eppnt->p_type != PT_LOAD)
1244 /* Now use mmap to map the library into memory. */
1245 error = vm_mmap(file,
1246 ELF_PAGESTART(eppnt->p_vaddr),
1248 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1249 PROT_READ | PROT_WRITE | PROT_EXEC,
1250 MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_DENYWRITE,
1252 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1253 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1256 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1257 if (padzero(elf_bss)) {
1262 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1264 bss = eppnt->p_memsz + eppnt->p_vaddr;
1266 error = vm_brk(len, bss - len);
1277 #endif /* #ifdef CONFIG_USELIB */
1279 #ifdef CONFIG_ELF_CORE
1283 * Modelled on fs/exec.c:aout_core_dump()
1284 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1288 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1289 * that are useful for post-mortem analysis are included in every core dump.
1290 * In that way we ensure that the core dump is fully interpretable later
1291 * without matching up the same kernel and hardware config to see what PC values
1292 * meant. These special mappings include - vDSO, vsyscall, and other
1293 * architecture specific mappings
1295 static bool always_dump_vma(struct vm_area_struct *vma)
1297 /* Any vsyscall mappings? */
1298 if (vma == get_gate_vma(vma->vm_mm))
1302 * Assume that all vmas with a .name op should always be dumped.
1303 * If this changes, a new vm_ops field can easily be added.
1305 if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1309 * arch_vma_name() returns non-NULL for special architecture mappings,
1310 * such as vDSO sections.
1312 if (arch_vma_name(vma))
1319 * Decide what to dump of a segment, part, all or none.
1321 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1322 unsigned long mm_flags)
1324 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1326 /* always dump the vdso and vsyscall sections */
1327 if (always_dump_vma(vma))
1330 if (vma->vm_flags & VM_DONTDUMP)
1333 /* support for DAX */
1334 if (vma_is_dax(vma)) {
1335 if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1337 if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1342 /* Hugetlb memory check */
1343 if (vma->vm_flags & VM_HUGETLB) {
1344 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1346 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1351 /* Do not dump I/O mapped devices or special mappings */
1352 if (vma->vm_flags & VM_IO)
1355 /* By default, dump shared memory if mapped from an anonymous file. */
1356 if (vma->vm_flags & VM_SHARED) {
1357 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1358 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1363 /* Dump segments that have been written to. */
1364 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1366 if (vma->vm_file == NULL)
1369 if (FILTER(MAPPED_PRIVATE))
1373 * If this looks like the beginning of a DSO or executable mapping,
1374 * check for an ELF header. If we find one, dump the first page to
1375 * aid in determining what was mapped here.
1377 if (FILTER(ELF_HEADERS) &&
1378 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1379 u32 __user *header = (u32 __user *) vma->vm_start;
1381 mm_segment_t fs = get_fs();
1383 * Doing it this way gets the constant folded by GCC.
1387 char elfmag[SELFMAG];
1389 BUILD_BUG_ON(SELFMAG != sizeof word);
1390 magic.elfmag[EI_MAG0] = ELFMAG0;
1391 magic.elfmag[EI_MAG1] = ELFMAG1;
1392 magic.elfmag[EI_MAG2] = ELFMAG2;
1393 magic.elfmag[EI_MAG3] = ELFMAG3;
1395 * Switch to the user "segment" for get_user(),
1396 * then put back what elf_core_dump() had in place.
1399 if (unlikely(get_user(word, header)))
1402 if (word == magic.cmp)
1411 return vma->vm_end - vma->vm_start;
1414 /* An ELF note in memory */
1419 unsigned int datasz;
1423 static int notesize(struct memelfnote *en)
1427 sz = sizeof(struct elf_note);
1428 sz += roundup(strlen(en->name) + 1, 4);
1429 sz += roundup(en->datasz, 4);
1434 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1437 en.n_namesz = strlen(men->name) + 1;
1438 en.n_descsz = men->datasz;
1439 en.n_type = men->type;
1441 return dump_emit(cprm, &en, sizeof(en)) &&
1442 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1443 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1446 static void fill_elf_header(struct elfhdr *elf, int segs,
1447 u16 machine, u32 flags)
1449 memset(elf, 0, sizeof(*elf));
1451 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1452 elf->e_ident[EI_CLASS] = ELF_CLASS;
1453 elf->e_ident[EI_DATA] = ELF_DATA;
1454 elf->e_ident[EI_VERSION] = EV_CURRENT;
1455 elf->e_ident[EI_OSABI] = ELF_OSABI;
1457 elf->e_type = ET_CORE;
1458 elf->e_machine = machine;
1459 elf->e_version = EV_CURRENT;
1460 elf->e_phoff = sizeof(struct elfhdr);
1461 elf->e_flags = flags;
1462 elf->e_ehsize = sizeof(struct elfhdr);
1463 elf->e_phentsize = sizeof(struct elf_phdr);
1464 elf->e_phnum = segs;
1469 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1471 phdr->p_type = PT_NOTE;
1472 phdr->p_offset = offset;
1475 phdr->p_filesz = sz;
1482 static void fill_note(struct memelfnote *note, const char *name, int type,
1483 unsigned int sz, void *data)
1493 * fill up all the fields in prstatus from the given task struct, except
1494 * registers which need to be filled up separately.
1496 static void fill_prstatus(struct elf_prstatus *prstatus,
1497 struct task_struct *p, long signr)
1499 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1500 prstatus->pr_sigpend = p->pending.signal.sig[0];
1501 prstatus->pr_sighold = p->blocked.sig[0];
1503 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1505 prstatus->pr_pid = task_pid_vnr(p);
1506 prstatus->pr_pgrp = task_pgrp_vnr(p);
1507 prstatus->pr_sid = task_session_vnr(p);
1508 if (thread_group_leader(p)) {
1509 struct task_cputime cputime;
1512 * This is the record for the group leader. It shows the
1513 * group-wide total, not its individual thread total.
1515 thread_group_cputime(p, &cputime);
1516 prstatus->pr_utime = ns_to_timeval(cputime.utime);
1517 prstatus->pr_stime = ns_to_timeval(cputime.stime);
1521 task_cputime(p, &utime, &stime);
1522 prstatus->pr_utime = ns_to_timeval(utime);
1523 prstatus->pr_stime = ns_to_timeval(stime);
1526 prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1527 prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1530 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1531 struct mm_struct *mm)
1533 const struct cred *cred;
1534 unsigned int i, len;
1536 /* first copy the parameters from user space */
1537 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1539 len = mm->arg_end - mm->arg_start;
1540 if (len >= ELF_PRARGSZ)
1541 len = ELF_PRARGSZ-1;
1542 if (copy_from_user(&psinfo->pr_psargs,
1543 (const char __user *)mm->arg_start, len))
1545 for(i = 0; i < len; i++)
1546 if (psinfo->pr_psargs[i] == 0)
1547 psinfo->pr_psargs[i] = ' ';
1548 psinfo->pr_psargs[len] = 0;
1551 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1553 psinfo->pr_pid = task_pid_vnr(p);
1554 psinfo->pr_pgrp = task_pgrp_vnr(p);
1555 psinfo->pr_sid = task_session_vnr(p);
1557 i = p->state ? ffz(~p->state) + 1 : 0;
1558 psinfo->pr_state = i;
1559 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1560 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1561 psinfo->pr_nice = task_nice(p);
1562 psinfo->pr_flag = p->flags;
1564 cred = __task_cred(p);
1565 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1566 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1568 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1573 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1575 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1579 while (auxv[i - 2] != AT_NULL);
1580 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1583 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1584 const siginfo_t *siginfo)
1586 mm_segment_t old_fs = get_fs();
1588 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1590 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1593 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1595 * Format of NT_FILE note:
1597 * long count -- how many files are mapped
1598 * long page_size -- units for file_ofs
1599 * array of [COUNT] elements of
1603 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1605 static int fill_files_note(struct memelfnote *note)
1607 struct vm_area_struct *vma;
1608 unsigned count, size, names_ofs, remaining, n;
1610 user_long_t *start_end_ofs;
1611 char *name_base, *name_curpos;
1613 /* *Estimated* file count and total data size needed */
1614 count = current->mm->map_count;
1615 if (count > UINT_MAX / 64)
1619 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1621 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1623 size = round_up(size, PAGE_SIZE);
1624 data = vmalloc(size);
1628 start_end_ofs = data + 2;
1629 name_base = name_curpos = ((char *)data) + names_ofs;
1630 remaining = size - names_ofs;
1632 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1634 const char *filename;
1636 file = vma->vm_file;
1639 filename = file_path(file, name_curpos, remaining);
1640 if (IS_ERR(filename)) {
1641 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1643 size = size * 5 / 4;
1649 /* file_path() fills at the end, move name down */
1650 /* n = strlen(filename) + 1: */
1651 n = (name_curpos + remaining) - filename;
1652 remaining = filename - name_curpos;
1653 memmove(name_curpos, filename, n);
1656 *start_end_ofs++ = vma->vm_start;
1657 *start_end_ofs++ = vma->vm_end;
1658 *start_end_ofs++ = vma->vm_pgoff;
1662 /* Now we know exact count of files, can store it */
1664 data[1] = PAGE_SIZE;
1666 * Count usually is less than current->mm->map_count,
1667 * we need to move filenames down.
1669 n = current->mm->map_count - count;
1671 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1672 memmove(name_base - shift_bytes, name_base,
1673 name_curpos - name_base);
1674 name_curpos -= shift_bytes;
1677 size = name_curpos - (char *)data;
1678 fill_note(note, "CORE", NT_FILE, size, data);
1682 #ifdef CORE_DUMP_USE_REGSET
1683 #include <linux/regset.h>
1685 struct elf_thread_core_info {
1686 struct elf_thread_core_info *next;
1687 struct task_struct *task;
1688 struct elf_prstatus prstatus;
1689 struct memelfnote notes[0];
1692 struct elf_note_info {
1693 struct elf_thread_core_info *thread;
1694 struct memelfnote psinfo;
1695 struct memelfnote signote;
1696 struct memelfnote auxv;
1697 struct memelfnote files;
1698 user_siginfo_t csigdata;
1704 * When a regset has a writeback hook, we call it on each thread before
1705 * dumping user memory. On register window machines, this makes sure the
1706 * user memory backing the register data is up to date before we read it.
1708 static void do_thread_regset_writeback(struct task_struct *task,
1709 const struct user_regset *regset)
1711 if (regset->writeback)
1712 regset->writeback(task, regset, 1);
1715 #ifndef PRSTATUS_SIZE
1716 #define PRSTATUS_SIZE(S, R) sizeof(S)
1719 #ifndef SET_PR_FPVALID
1720 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1723 static int fill_thread_core_info(struct elf_thread_core_info *t,
1724 const struct user_regset_view *view,
1725 long signr, size_t *total)
1728 unsigned int regset0_size = regset_size(t->task, &view->regsets[0]);
1731 * NT_PRSTATUS is the one special case, because the regset data
1732 * goes into the pr_reg field inside the note contents, rather
1733 * than being the whole note contents. We fill the reset in here.
1734 * We assume that regset 0 is NT_PRSTATUS.
1736 fill_prstatus(&t->prstatus, t->task, signr);
1737 (void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset0_size,
1738 &t->prstatus.pr_reg, NULL);
1740 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1741 PRSTATUS_SIZE(t->prstatus, regset0_size), &t->prstatus);
1742 *total += notesize(&t->notes[0]);
1744 do_thread_regset_writeback(t->task, &view->regsets[0]);
1747 * Each other regset might generate a note too. For each regset
1748 * that has no core_note_type or is inactive, we leave t->notes[i]
1749 * all zero and we'll know to skip writing it later.
1751 for (i = 1; i < view->n; ++i) {
1752 const struct user_regset *regset = &view->regsets[i];
1753 do_thread_regset_writeback(t->task, regset);
1754 if (regset->core_note_type && regset->get &&
1755 (!regset->active || regset->active(t->task, regset))) {
1757 size_t size = regset_size(t->task, regset);
1758 void *data = kmalloc(size, GFP_KERNEL);
1759 if (unlikely(!data))
1761 ret = regset->get(t->task, regset,
1762 0, size, data, NULL);
1766 if (regset->core_note_type != NT_PRFPREG)
1767 fill_note(&t->notes[i], "LINUX",
1768 regset->core_note_type,
1771 SET_PR_FPVALID(&t->prstatus,
1773 fill_note(&t->notes[i], "CORE",
1774 NT_PRFPREG, size, data);
1776 *total += notesize(&t->notes[i]);
1784 static int fill_note_info(struct elfhdr *elf, int phdrs,
1785 struct elf_note_info *info,
1786 const siginfo_t *siginfo, struct pt_regs *regs)
1788 struct task_struct *dump_task = current;
1789 const struct user_regset_view *view = task_user_regset_view(dump_task);
1790 struct elf_thread_core_info *t;
1791 struct elf_prpsinfo *psinfo;
1792 struct core_thread *ct;
1796 info->thread = NULL;
1798 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1799 if (psinfo == NULL) {
1800 info->psinfo.data = NULL; /* So we don't free this wrongly */
1804 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1807 * Figure out how many notes we're going to need for each thread.
1809 info->thread_notes = 0;
1810 for (i = 0; i < view->n; ++i)
1811 if (view->regsets[i].core_note_type != 0)
1812 ++info->thread_notes;
1815 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1816 * since it is our one special case.
1818 if (unlikely(info->thread_notes == 0) ||
1819 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1825 * Initialize the ELF file header.
1827 fill_elf_header(elf, phdrs,
1828 view->e_machine, view->e_flags);
1831 * Allocate a structure for each thread.
1833 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1834 t = kzalloc(offsetof(struct elf_thread_core_info,
1835 notes[info->thread_notes]),
1841 if (ct->task == dump_task || !info->thread) {
1842 t->next = info->thread;
1846 * Make sure to keep the original task at
1847 * the head of the list.
1849 t->next = info->thread->next;
1850 info->thread->next = t;
1855 * Now fill in each thread's information.
1857 for (t = info->thread; t != NULL; t = t->next)
1858 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1862 * Fill in the two process-wide notes.
1864 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1865 info->size += notesize(&info->psinfo);
1867 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1868 info->size += notesize(&info->signote);
1870 fill_auxv_note(&info->auxv, current->mm);
1871 info->size += notesize(&info->auxv);
1873 if (fill_files_note(&info->files) == 0)
1874 info->size += notesize(&info->files);
1879 static size_t get_note_info_size(struct elf_note_info *info)
1885 * Write all the notes for each thread. When writing the first thread, the
1886 * process-wide notes are interleaved after the first thread-specific note.
1888 static int write_note_info(struct elf_note_info *info,
1889 struct coredump_params *cprm)
1892 struct elf_thread_core_info *t = info->thread;
1897 if (!writenote(&t->notes[0], cprm))
1900 if (first && !writenote(&info->psinfo, cprm))
1902 if (first && !writenote(&info->signote, cprm))
1904 if (first && !writenote(&info->auxv, cprm))
1906 if (first && info->files.data &&
1907 !writenote(&info->files, cprm))
1910 for (i = 1; i < info->thread_notes; ++i)
1911 if (t->notes[i].data &&
1912 !writenote(&t->notes[i], cprm))
1922 static void free_note_info(struct elf_note_info *info)
1924 struct elf_thread_core_info *threads = info->thread;
1927 struct elf_thread_core_info *t = threads;
1929 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1930 for (i = 1; i < info->thread_notes; ++i)
1931 kfree(t->notes[i].data);
1934 kfree(info->psinfo.data);
1935 vfree(info->files.data);
1940 /* Here is the structure in which status of each thread is captured. */
1941 struct elf_thread_status
1943 struct list_head list;
1944 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1945 elf_fpregset_t fpu; /* NT_PRFPREG */
1946 struct task_struct *thread;
1947 #ifdef ELF_CORE_COPY_XFPREGS
1948 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1950 struct memelfnote notes[3];
1955 * In order to add the specific thread information for the elf file format,
1956 * we need to keep a linked list of every threads pr_status and then create
1957 * a single section for them in the final core file.
1959 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1962 struct task_struct *p = t->thread;
1965 fill_prstatus(&t->prstatus, p, signr);
1966 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1968 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1971 sz += notesize(&t->notes[0]);
1973 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1975 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1978 sz += notesize(&t->notes[1]);
1981 #ifdef ELF_CORE_COPY_XFPREGS
1982 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1983 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1984 sizeof(t->xfpu), &t->xfpu);
1986 sz += notesize(&t->notes[2]);
1992 struct elf_note_info {
1993 struct memelfnote *notes;
1994 struct memelfnote *notes_files;
1995 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1996 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1997 struct list_head thread_list;
1998 elf_fpregset_t *fpu;
1999 #ifdef ELF_CORE_COPY_XFPREGS
2000 elf_fpxregset_t *xfpu;
2002 user_siginfo_t csigdata;
2003 int thread_status_size;
2007 static int elf_note_info_init(struct elf_note_info *info)
2009 memset(info, 0, sizeof(*info));
2010 INIT_LIST_HEAD(&info->thread_list);
2012 /* Allocate space for ELF notes */
2013 info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
2016 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2019 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2020 if (!info->prstatus)
2022 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2025 #ifdef ELF_CORE_COPY_XFPREGS
2026 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
2033 static int fill_note_info(struct elfhdr *elf, int phdrs,
2034 struct elf_note_info *info,
2035 const siginfo_t *siginfo, struct pt_regs *regs)
2037 struct list_head *t;
2038 struct core_thread *ct;
2039 struct elf_thread_status *ets;
2041 if (!elf_note_info_init(info))
2044 for (ct = current->mm->core_state->dumper.next;
2045 ct; ct = ct->next) {
2046 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2050 ets->thread = ct->task;
2051 list_add(&ets->list, &info->thread_list);
2054 list_for_each(t, &info->thread_list) {
2057 ets = list_entry(t, struct elf_thread_status, list);
2058 sz = elf_dump_thread_status(siginfo->si_signo, ets);
2059 info->thread_status_size += sz;
2061 /* now collect the dump for the current */
2062 memset(info->prstatus, 0, sizeof(*info->prstatus));
2063 fill_prstatus(info->prstatus, current, siginfo->si_signo);
2064 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2067 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2070 * Set up the notes in similar form to SVR4 core dumps made
2071 * with info from their /proc.
2074 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2075 sizeof(*info->prstatus), info->prstatus);
2076 fill_psinfo(info->psinfo, current->group_leader, current->mm);
2077 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2078 sizeof(*info->psinfo), info->psinfo);
2080 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2081 fill_auxv_note(info->notes + 3, current->mm);
2084 if (fill_files_note(info->notes + info->numnote) == 0) {
2085 info->notes_files = info->notes + info->numnote;
2089 /* Try to dump the FPU. */
2090 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2092 if (info->prstatus->pr_fpvalid)
2093 fill_note(info->notes + info->numnote++,
2094 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2095 #ifdef ELF_CORE_COPY_XFPREGS
2096 if (elf_core_copy_task_xfpregs(current, info->xfpu))
2097 fill_note(info->notes + info->numnote++,
2098 "LINUX", ELF_CORE_XFPREG_TYPE,
2099 sizeof(*info->xfpu), info->xfpu);
2105 static size_t get_note_info_size(struct elf_note_info *info)
2110 for (i = 0; i < info->numnote; i++)
2111 sz += notesize(info->notes + i);
2113 sz += info->thread_status_size;
2118 static int write_note_info(struct elf_note_info *info,
2119 struct coredump_params *cprm)
2122 struct list_head *t;
2124 for (i = 0; i < info->numnote; i++)
2125 if (!writenote(info->notes + i, cprm))
2128 /* write out the thread status notes section */
2129 list_for_each(t, &info->thread_list) {
2130 struct elf_thread_status *tmp =
2131 list_entry(t, struct elf_thread_status, list);
2133 for (i = 0; i < tmp->num_notes; i++)
2134 if (!writenote(&tmp->notes[i], cprm))
2141 static void free_note_info(struct elf_note_info *info)
2143 while (!list_empty(&info->thread_list)) {
2144 struct list_head *tmp = info->thread_list.next;
2146 kfree(list_entry(tmp, struct elf_thread_status, list));
2149 /* Free data possibly allocated by fill_files_note(): */
2150 if (info->notes_files)
2151 vfree(info->notes_files->data);
2153 kfree(info->prstatus);
2154 kfree(info->psinfo);
2157 #ifdef ELF_CORE_COPY_XFPREGS
2164 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2165 struct vm_area_struct *gate_vma)
2167 struct vm_area_struct *ret = tsk->mm->mmap;
2174 * Helper function for iterating across a vma list. It ensures that the caller
2175 * will visit `gate_vma' prior to terminating the search.
2177 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2178 struct vm_area_struct *gate_vma)
2180 struct vm_area_struct *ret;
2182 ret = this_vma->vm_next;
2185 if (this_vma == gate_vma)
2190 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2191 elf_addr_t e_shoff, int segs)
2193 elf->e_shoff = e_shoff;
2194 elf->e_shentsize = sizeof(*shdr4extnum);
2196 elf->e_shstrndx = SHN_UNDEF;
2198 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2200 shdr4extnum->sh_type = SHT_NULL;
2201 shdr4extnum->sh_size = elf->e_shnum;
2202 shdr4extnum->sh_link = elf->e_shstrndx;
2203 shdr4extnum->sh_info = segs;
2209 * This is a two-pass process; first we find the offsets of the bits,
2210 * and then they are actually written out. If we run out of core limit
2213 static int elf_core_dump(struct coredump_params *cprm)
2218 size_t vma_data_size = 0;
2219 struct vm_area_struct *vma, *gate_vma;
2220 struct elfhdr *elf = NULL;
2221 loff_t offset = 0, dataoff;
2222 struct elf_note_info info = { };
2223 struct elf_phdr *phdr4note = NULL;
2224 struct elf_shdr *shdr4extnum = NULL;
2227 elf_addr_t *vma_filesz = NULL;
2230 * We no longer stop all VM operations.
2232 * This is because those proceses that could possibly change map_count
2233 * or the mmap / vma pages are now blocked in do_exit on current
2234 * finishing this core dump.
2236 * Only ptrace can touch these memory addresses, but it doesn't change
2237 * the map_count or the pages allocated. So no possibility of crashing
2238 * exists while dumping the mm->vm_next areas to the core file.
2241 /* alloc memory for large data structures: too large to be on stack */
2242 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2246 * The number of segs are recored into ELF header as 16bit value.
2247 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2249 segs = current->mm->map_count;
2250 segs += elf_core_extra_phdrs();
2252 gate_vma = get_gate_vma(current->mm);
2253 if (gate_vma != NULL)
2256 /* for notes section */
2259 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2260 * this, kernel supports extended numbering. Have a look at
2261 * include/linux/elf.h for further information. */
2262 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2265 * Collect all the non-memory information about the process for the
2266 * notes. This also sets up the file header.
2268 if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2276 offset += sizeof(*elf); /* Elf header */
2277 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2279 /* Write notes phdr entry */
2281 size_t sz = get_note_info_size(&info);
2283 sz += elf_coredump_extra_notes_size();
2285 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2289 fill_elf_note_phdr(phdr4note, sz, offset);
2293 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2295 if (segs - 1 > ULONG_MAX / sizeof(*vma_filesz))
2297 vma_filesz = vmalloc((segs - 1) * sizeof(*vma_filesz));
2301 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2302 vma = next_vma(vma, gate_vma)) {
2303 unsigned long dump_size;
2305 dump_size = vma_dump_size(vma, cprm->mm_flags);
2306 vma_filesz[i++] = dump_size;
2307 vma_data_size += dump_size;
2310 offset += vma_data_size;
2311 offset += elf_core_extra_data_size();
2314 if (e_phnum == PN_XNUM) {
2315 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2318 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2323 if (!dump_emit(cprm, elf, sizeof(*elf)))
2326 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2329 /* Write program headers for segments dump */
2330 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2331 vma = next_vma(vma, gate_vma)) {
2332 struct elf_phdr phdr;
2334 phdr.p_type = PT_LOAD;
2335 phdr.p_offset = offset;
2336 phdr.p_vaddr = vma->vm_start;
2338 phdr.p_filesz = vma_filesz[i++];
2339 phdr.p_memsz = vma->vm_end - vma->vm_start;
2340 offset += phdr.p_filesz;
2341 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2342 if (vma->vm_flags & VM_WRITE)
2343 phdr.p_flags |= PF_W;
2344 if (vma->vm_flags & VM_EXEC)
2345 phdr.p_flags |= PF_X;
2346 phdr.p_align = ELF_EXEC_PAGESIZE;
2348 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2352 if (!elf_core_write_extra_phdrs(cprm, offset))
2355 /* write out the notes section */
2356 if (!write_note_info(&info, cprm))
2359 if (elf_coredump_extra_notes_write(cprm))
2363 if (!dump_skip(cprm, dataoff - cprm->pos))
2366 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2367 vma = next_vma(vma, gate_vma)) {
2371 end = vma->vm_start + vma_filesz[i++];
2373 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2377 page = get_dump_page(addr);
2379 void *kaddr = kmap(page);
2380 stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2384 stop = !dump_skip(cprm, PAGE_SIZE);
2389 dump_truncate(cprm);
2391 if (!elf_core_write_extra_data(cprm))
2394 if (e_phnum == PN_XNUM) {
2395 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2403 free_note_info(&info);
2412 #endif /* CONFIG_ELF_CORE */
2414 static int __init init_elf_binfmt(void)
2416 register_binfmt(&elf_format);
2420 static void __exit exit_elf_binfmt(void)
2422 /* Remove the COFF and ELF loaders. */
2423 unregister_binfmt(&elf_format);
2426 core_initcall(init_elf_binfmt);
2427 module_exit(exit_elf_binfmt);
2428 MODULE_LICENSE("GPL");