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/utsname.h>
35 #include <linux/coredump.h>
36 #include <linux/sched.h>
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
42 #define user_long_t long
44 #ifndef user_siginfo_t
45 #define user_siginfo_t siginfo_t
48 static int load_elf_binary(struct linux_binprm *bprm);
49 static int load_elf_library(struct file *);
50 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
51 int, int, unsigned long);
54 * If we don't support core dumping, then supply a NULL so we
57 #ifdef CONFIG_ELF_CORE
58 static int elf_core_dump(struct coredump_params *cprm);
60 #define elf_core_dump NULL
63 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
64 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
66 #define ELF_MIN_ALIGN PAGE_SIZE
69 #ifndef ELF_CORE_EFLAGS
70 #define ELF_CORE_EFLAGS 0
73 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
74 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
75 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
77 static struct linux_binfmt elf_format = {
78 .module = THIS_MODULE,
79 .load_binary = load_elf_binary,
80 .load_shlib = load_elf_library,
81 .core_dump = elf_core_dump,
82 .min_coredump = ELF_EXEC_PAGESIZE,
85 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
87 static int set_brk(unsigned long start, unsigned long end)
89 start = ELF_PAGEALIGN(start);
90 end = ELF_PAGEALIGN(end);
93 addr = vm_brk(start, end - start);
97 current->mm->start_brk = current->mm->brk = end;
101 /* We need to explicitly zero any fractional pages
102 after the data section (i.e. bss). This would
103 contain the junk from the file that should not
106 static int padzero(unsigned long elf_bss)
110 nbyte = ELF_PAGEOFFSET(elf_bss);
112 nbyte = ELF_MIN_ALIGN - nbyte;
113 if (clear_user((void __user *) elf_bss, nbyte))
119 /* Let's use some macros to make this stack manipulation a little clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 (((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
134 #ifndef ELF_BASE_PLATFORM
136 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
137 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
138 * will be copied to the user stack in the same manner as AT_PLATFORM.
140 #define ELF_BASE_PLATFORM NULL
144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
145 unsigned long load_addr, unsigned long interp_load_addr)
147 unsigned long p = bprm->p;
148 int argc = bprm->argc;
149 int envc = bprm->envc;
150 elf_addr_t __user *argv;
151 elf_addr_t __user *envp;
152 elf_addr_t __user *sp;
153 elf_addr_t __user *u_platform;
154 elf_addr_t __user *u_base_platform;
155 elf_addr_t __user *u_rand_bytes;
156 const char *k_platform = ELF_PLATFORM;
157 const char *k_base_platform = ELF_BASE_PLATFORM;
158 unsigned char k_rand_bytes[16];
160 elf_addr_t *elf_info;
162 const struct cred *cred = current_cred();
163 struct vm_area_struct *vma;
166 * In some cases (e.g. Hyper-Threading), we want to avoid L1
167 * evictions by the processes running on the same package. One
168 * thing we can do is to shuffle the initial stack for them.
171 p = arch_align_stack(p);
174 * If this architecture has a platform capability string, copy it
175 * to userspace. In some cases (Sparc), this info is impossible
176 * for userspace to get any other way, in others (i386) it is
181 size_t len = strlen(k_platform) + 1;
183 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
184 if (__copy_to_user(u_platform, k_platform, len))
189 * If this architecture has a "base" platform capability
190 * string, copy it to userspace.
192 u_base_platform = NULL;
193 if (k_base_platform) {
194 size_t len = strlen(k_base_platform) + 1;
196 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 if (__copy_to_user(u_base_platform, k_base_platform, len))
202 * Generate 16 random bytes for userspace PRNG seeding.
204 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
205 u_rand_bytes = (elf_addr_t __user *)
206 STACK_ALLOC(p, sizeof(k_rand_bytes));
207 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
210 /* Create the ELF interpreter info */
211 elf_info = (elf_addr_t *)current->mm->saved_auxv;
212 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
213 #define NEW_AUX_ENT(id, val) \
215 elf_info[ei_index++] = id; \
216 elf_info[ei_index++] = val; \
221 * ARCH_DLINFO must come first so PPC can do its special alignment of
223 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
224 * ARCH_DLINFO changes
228 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
229 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
230 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
231 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
232 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
233 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
234 NEW_AUX_ENT(AT_BASE, interp_load_addr);
235 NEW_AUX_ENT(AT_FLAGS, 0);
236 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
237 NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
238 NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
239 NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
240 NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
241 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
242 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
244 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
246 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
248 NEW_AUX_ENT(AT_PLATFORM,
249 (elf_addr_t)(unsigned long)u_platform);
251 if (k_base_platform) {
252 NEW_AUX_ENT(AT_BASE_PLATFORM,
253 (elf_addr_t)(unsigned long)u_base_platform);
255 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
256 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
259 /* AT_NULL is zero; clear the rest too */
260 memset(&elf_info[ei_index], 0,
261 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
263 /* And advance past the AT_NULL entry. */
266 sp = STACK_ADD(p, ei_index);
268 items = (argc + 1) + (envc + 1) + 1;
269 bprm->p = STACK_ROUND(sp, items);
271 /* Point sp at the lowest address on the stack */
272 #ifdef CONFIG_STACK_GROWSUP
273 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
274 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
276 sp = (elf_addr_t __user *)bprm->p;
281 * Grow the stack manually; some architectures have a limit on how
282 * far ahead a user-space access may be in order to grow the stack.
284 vma = find_extend_vma(current->mm, bprm->p);
288 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
289 if (__put_user(argc, sp++))
292 envp = argv + argc + 1;
294 /* Populate argv and envp */
295 p = current->mm->arg_end = current->mm->arg_start;
298 if (__put_user((elf_addr_t)p, argv++))
300 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
301 if (!len || len > MAX_ARG_STRLEN)
305 if (__put_user(0, argv))
307 current->mm->arg_end = current->mm->env_start = p;
310 if (__put_user((elf_addr_t)p, envp++))
312 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
313 if (!len || len > MAX_ARG_STRLEN)
317 if (__put_user(0, envp))
319 current->mm->env_end = p;
321 /* Put the elf_info on the stack in the right place. */
322 sp = (elf_addr_t __user *)envp + 1;
323 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
330 static unsigned long elf_map(struct file *filep, unsigned long addr,
331 struct elf_phdr *eppnt, int prot, int type,
332 unsigned long total_size)
334 unsigned long map_addr;
335 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
336 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
337 addr = ELF_PAGESTART(addr);
338 size = ELF_PAGEALIGN(size);
340 /* mmap() will return -EINVAL if given a zero size, but a
341 * segment with zero filesize is perfectly valid */
346 * total_size is the size of the ELF (interpreter) image.
347 * The _first_ mmap needs to know the full size, otherwise
348 * randomization might put this image into an overlapping
349 * position with the ELF binary image. (since size < total_size)
350 * So we first map the 'big' image - and unmap the remainder at
351 * the end. (which unmap is needed for ELF images with holes.)
354 total_size = ELF_PAGEALIGN(total_size);
355 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
356 if (!BAD_ADDR(map_addr))
357 vm_munmap(map_addr+size, total_size-size);
359 map_addr = vm_mmap(filep, addr, size, prot, type, off);
364 #endif /* !elf_map */
366 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
368 int i, first_idx = -1, last_idx = -1;
370 for (i = 0; i < nr; i++) {
371 if (cmds[i].p_type == PT_LOAD) {
380 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
381 ELF_PAGESTART(cmds[first_idx].p_vaddr);
385 /* This is much more generalized than the library routine read function,
386 so we keep this separate. Technically the library read function
387 is only provided so that we can read a.out libraries that have
390 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
391 struct file *interpreter, unsigned long *interp_map_addr,
392 unsigned long no_base)
394 struct elf_phdr *elf_phdata;
395 struct elf_phdr *eppnt;
396 unsigned long load_addr = 0;
397 int load_addr_set = 0;
398 unsigned long last_bss = 0, elf_bss = 0;
399 unsigned long error = ~0UL;
400 unsigned long total_size;
403 /* First of all, some simple consistency checks */
404 if (interp_elf_ex->e_type != ET_EXEC &&
405 interp_elf_ex->e_type != ET_DYN)
407 if (!elf_check_arch(interp_elf_ex))
409 if (!interpreter->f_op->mmap)
413 * If the size of this structure has changed, then punt, since
414 * we will be doing the wrong thing.
416 if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
418 if (interp_elf_ex->e_phnum < 1 ||
419 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
422 /* Now read in all of the header information */
423 size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
424 if (size > ELF_MIN_ALIGN)
426 elf_phdata = kmalloc(size, GFP_KERNEL);
430 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
431 (char *)elf_phdata, size);
433 if (retval != size) {
439 total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
446 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
447 if (eppnt->p_type == PT_LOAD) {
448 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
450 unsigned long vaddr = 0;
451 unsigned long k, map_addr;
453 if (eppnt->p_flags & PF_R)
454 elf_prot = PROT_READ;
455 if (eppnt->p_flags & PF_W)
456 elf_prot |= PROT_WRITE;
457 if (eppnt->p_flags & PF_X)
458 elf_prot |= PROT_EXEC;
459 vaddr = eppnt->p_vaddr;
460 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
461 elf_type |= MAP_FIXED;
462 else if (no_base && interp_elf_ex->e_type == ET_DYN)
465 map_addr = elf_map(interpreter, load_addr + vaddr,
466 eppnt, elf_prot, elf_type, total_size);
468 if (!*interp_map_addr)
469 *interp_map_addr = map_addr;
471 if (BAD_ADDR(map_addr))
474 if (!load_addr_set &&
475 interp_elf_ex->e_type == ET_DYN) {
476 load_addr = map_addr - ELF_PAGESTART(vaddr);
481 * Check to see if the section's size will overflow the
482 * allowed task size. Note that p_filesz must always be
483 * <= p_memsize so it's only necessary to check p_memsz.
485 k = load_addr + eppnt->p_vaddr;
487 eppnt->p_filesz > eppnt->p_memsz ||
488 eppnt->p_memsz > TASK_SIZE ||
489 TASK_SIZE - eppnt->p_memsz < k) {
495 * Find the end of the file mapping for this phdr, and
496 * keep track of the largest address we see for this.
498 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
503 * Do the same thing for the memory mapping - between
504 * elf_bss and last_bss is the bss section.
506 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
512 if (last_bss > elf_bss) {
514 * Now fill out the bss section. First pad the last page up
515 * to the page boundary, and then perform a mmap to make sure
516 * that there are zero-mapped pages up to and including the
519 if (padzero(elf_bss)) {
524 /* What we have mapped so far */
525 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
527 /* Map the last of the bss segment */
528 error = vm_brk(elf_bss, last_bss - elf_bss);
542 * These are the functions used to load ELF style executables and shared
543 * libraries. There is no binary dependent code anywhere else.
546 #define INTERPRETER_NONE 0
547 #define INTERPRETER_ELF 2
549 #ifndef STACK_RND_MASK
550 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
553 static unsigned long randomize_stack_top(unsigned long stack_top)
555 unsigned int random_variable = 0;
557 if ((current->flags & PF_RANDOMIZE) &&
558 !(current->personality & ADDR_NO_RANDOMIZE)) {
559 random_variable = get_random_int() & STACK_RND_MASK;
560 random_variable <<= PAGE_SHIFT;
562 #ifdef CONFIG_STACK_GROWSUP
563 return PAGE_ALIGN(stack_top) + random_variable;
565 return PAGE_ALIGN(stack_top) - random_variable;
569 static int load_elf_binary(struct linux_binprm *bprm)
571 struct file *interpreter = NULL; /* to shut gcc up */
572 unsigned long load_addr = 0, load_bias = 0;
573 int load_addr_set = 0;
574 char * elf_interpreter = NULL;
576 struct elf_phdr *elf_ppnt, *elf_phdata;
577 unsigned long elf_bss, elf_brk;
580 unsigned long elf_entry;
581 unsigned long interp_load_addr = 0;
582 unsigned long start_code, end_code, start_data, end_data;
583 unsigned long reloc_func_desc __maybe_unused = 0;
584 int executable_stack = EXSTACK_DEFAULT;
585 unsigned long def_flags = 0;
586 struct pt_regs *regs = current_pt_regs();
588 struct elfhdr elf_ex;
589 struct elfhdr interp_elf_ex;
592 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
598 /* Get the exec-header */
599 loc->elf_ex = *((struct elfhdr *)bprm->buf);
602 /* First of all, some simple consistency checks */
603 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
606 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
608 if (!elf_check_arch(&loc->elf_ex))
610 if (!bprm->file->f_op->mmap)
613 /* Now read in all of the header information */
614 if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
616 if (loc->elf_ex.e_phnum < 1 ||
617 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
619 size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
621 elf_phdata = kmalloc(size, GFP_KERNEL);
625 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
626 (char *)elf_phdata, size);
627 if (retval != size) {
633 elf_ppnt = elf_phdata;
642 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
643 if (elf_ppnt->p_type == PT_INTERP) {
644 /* This is the program interpreter used for
645 * shared libraries - for now assume that this
646 * is an a.out format binary
649 if (elf_ppnt->p_filesz > PATH_MAX ||
650 elf_ppnt->p_filesz < 2)
654 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
656 if (!elf_interpreter)
659 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
662 if (retval != elf_ppnt->p_filesz) {
665 goto out_free_interp;
667 /* make sure path is NULL terminated */
669 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
670 goto out_free_interp;
672 interpreter = open_exec(elf_interpreter);
673 retval = PTR_ERR(interpreter);
674 if (IS_ERR(interpreter))
675 goto out_free_interp;
678 * If the binary is not readable then enforce
679 * mm->dumpable = 0 regardless of the interpreter's
682 would_dump(bprm, interpreter);
684 retval = kernel_read(interpreter, 0, bprm->buf,
686 if (retval != BINPRM_BUF_SIZE) {
689 goto out_free_dentry;
692 /* Get the exec headers */
693 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
699 elf_ppnt = elf_phdata;
700 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
701 if (elf_ppnt->p_type == PT_GNU_STACK) {
702 if (elf_ppnt->p_flags & PF_X)
703 executable_stack = EXSTACK_ENABLE_X;
705 executable_stack = EXSTACK_DISABLE_X;
709 /* Some simple consistency checks for the interpreter */
710 if (elf_interpreter) {
712 /* Not an ELF interpreter */
713 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
714 goto out_free_dentry;
715 /* Verify the interpreter has a valid arch */
716 if (!elf_check_arch(&loc->interp_elf_ex))
717 goto out_free_dentry;
720 /* Flush all traces of the currently running executable */
721 retval = flush_old_exec(bprm);
723 goto out_free_dentry;
725 /* OK, This is the point of no return */
726 current->mm->def_flags = def_flags;
728 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
729 may depend on the personality. */
730 SET_PERSONALITY(loc->elf_ex);
731 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
732 current->personality |= READ_IMPLIES_EXEC;
734 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
735 current->flags |= PF_RANDOMIZE;
737 setup_new_exec(bprm);
739 /* Do this so that we can load the interpreter, if need be. We will
740 change some of these later */
741 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
744 send_sig(SIGKILL, current, 0);
745 goto out_free_dentry;
748 current->mm->start_stack = bprm->p;
750 /* Now we do a little grungy work by mmapping the ELF image into
751 the correct location in memory. */
752 for(i = 0, elf_ppnt = elf_phdata;
753 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
754 int elf_prot = 0, elf_flags;
755 unsigned long k, vaddr;
757 if (elf_ppnt->p_type != PT_LOAD)
760 if (unlikely (elf_brk > elf_bss)) {
763 /* There was a PT_LOAD segment with p_memsz > p_filesz
764 before this one. Map anonymous pages, if needed,
765 and clear the area. */
766 retval = set_brk(elf_bss + load_bias,
767 elf_brk + load_bias);
769 send_sig(SIGKILL, current, 0);
770 goto out_free_dentry;
772 nbyte = ELF_PAGEOFFSET(elf_bss);
774 nbyte = ELF_MIN_ALIGN - nbyte;
775 if (nbyte > elf_brk - elf_bss)
776 nbyte = elf_brk - elf_bss;
777 if (clear_user((void __user *)elf_bss +
780 * This bss-zeroing can fail if the ELF
781 * file specifies odd protections. So
782 * we don't check the return value
788 if (elf_ppnt->p_flags & PF_R)
789 elf_prot |= PROT_READ;
790 if (elf_ppnt->p_flags & PF_W)
791 elf_prot |= PROT_WRITE;
792 if (elf_ppnt->p_flags & PF_X)
793 elf_prot |= PROT_EXEC;
795 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
797 vaddr = elf_ppnt->p_vaddr;
798 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
799 elf_flags |= MAP_FIXED;
800 } else if (loc->elf_ex.e_type == ET_DYN) {
801 /* Try and get dynamic programs out of the way of the
802 * default mmap base, as well as whatever program they
803 * might try to exec. This is because the brk will
804 * follow the loader, and is not movable. */
805 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
806 /* Memory randomization might have been switched off
807 * in runtime via sysctl or explicit setting of
809 * If that is the case, retain the original non-zero
810 * load_bias value in order to establish proper
811 * non-randomized mappings.
813 if (current->flags & PF_RANDOMIZE)
816 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
818 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
822 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
823 elf_prot, elf_flags, 0);
824 if (BAD_ADDR(error)) {
825 send_sig(SIGKILL, current, 0);
826 retval = IS_ERR((void *)error) ?
827 PTR_ERR((void*)error) : -EINVAL;
828 goto out_free_dentry;
831 if (!load_addr_set) {
833 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
834 if (loc->elf_ex.e_type == ET_DYN) {
836 ELF_PAGESTART(load_bias + vaddr);
837 load_addr += load_bias;
838 reloc_func_desc = load_bias;
841 k = elf_ppnt->p_vaddr;
848 * Check to see if the section's size will overflow the
849 * allowed task size. Note that p_filesz must always be
850 * <= p_memsz so it is only necessary to check p_memsz.
852 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
853 elf_ppnt->p_memsz > TASK_SIZE ||
854 TASK_SIZE - elf_ppnt->p_memsz < k) {
855 /* set_brk can never work. Avoid overflows. */
856 send_sig(SIGKILL, current, 0);
858 goto out_free_dentry;
861 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
865 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
869 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
874 loc->elf_ex.e_entry += load_bias;
875 elf_bss += load_bias;
876 elf_brk += load_bias;
877 start_code += load_bias;
878 end_code += load_bias;
879 start_data += load_bias;
880 end_data += load_bias;
882 /* Calling set_brk effectively mmaps the pages that we need
883 * for the bss and break sections. We must do this before
884 * mapping in the interpreter, to make sure it doesn't wind
885 * up getting placed where the bss needs to go.
887 retval = set_brk(elf_bss, elf_brk);
889 send_sig(SIGKILL, current, 0);
890 goto out_free_dentry;
892 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
893 send_sig(SIGSEGV, current, 0);
894 retval = -EFAULT; /* Nobody gets to see this, but.. */
895 goto out_free_dentry;
898 if (elf_interpreter) {
899 unsigned long interp_map_addr = 0;
901 elf_entry = load_elf_interp(&loc->interp_elf_ex,
905 if (!IS_ERR((void *)elf_entry)) {
907 * load_elf_interp() returns relocation
910 interp_load_addr = elf_entry;
911 elf_entry += loc->interp_elf_ex.e_entry;
913 if (BAD_ADDR(elf_entry)) {
914 force_sig(SIGSEGV, current);
915 retval = IS_ERR((void *)elf_entry) ?
916 (int)elf_entry : -EINVAL;
917 goto out_free_dentry;
919 reloc_func_desc = interp_load_addr;
921 allow_write_access(interpreter);
923 kfree(elf_interpreter);
925 elf_entry = loc->elf_ex.e_entry;
926 if (BAD_ADDR(elf_entry)) {
927 force_sig(SIGSEGV, current);
929 goto out_free_dentry;
935 set_binfmt(&elf_format);
937 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
938 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
940 send_sig(SIGKILL, current, 0);
943 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
945 install_exec_creds(bprm);
946 retval = create_elf_tables(bprm, &loc->elf_ex,
947 load_addr, interp_load_addr);
949 send_sig(SIGKILL, current, 0);
952 /* N.B. passed_fileno might not be initialized? */
953 current->mm->end_code = end_code;
954 current->mm->start_code = start_code;
955 current->mm->start_data = start_data;
956 current->mm->end_data = end_data;
957 current->mm->start_stack = bprm->p;
959 #ifdef arch_randomize_brk
960 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
961 current->mm->brk = current->mm->start_brk =
962 arch_randomize_brk(current->mm);
963 #ifdef CONFIG_COMPAT_BRK
964 current->brk_randomized = 1;
969 if (current->personality & MMAP_PAGE_ZERO) {
970 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
971 and some applications "depend" upon this behavior.
972 Since we do not have the power to recompile these, we
973 emulate the SVr4 behavior. Sigh. */
974 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
975 MAP_FIXED | MAP_PRIVATE, 0);
980 * The ABI may specify that certain registers be set up in special
981 * ways (on i386 %edx is the address of a DT_FINI function, for
982 * example. In addition, it may also specify (eg, PowerPC64 ELF)
983 * that the e_entry field is the address of the function descriptor
984 * for the startup routine, rather than the address of the startup
985 * routine itself. This macro performs whatever initialization to
986 * the regs structure is required as well as any relocations to the
987 * function descriptor entries when executing dynamically links apps.
989 ELF_PLAT_INIT(regs, reloc_func_desc);
992 start_thread(regs, elf_entry, bprm->p);
1001 allow_write_access(interpreter);
1005 kfree(elf_interpreter);
1011 /* This is really simpleminded and specialized - we are loading an
1012 a.out library that is given an ELF header. */
1013 static int load_elf_library(struct file *file)
1015 struct elf_phdr *elf_phdata;
1016 struct elf_phdr *eppnt;
1017 unsigned long elf_bss, bss, len;
1018 int retval, error, i, j;
1019 struct elfhdr elf_ex;
1022 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1023 if (retval != sizeof(elf_ex))
1026 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1029 /* First of all, some simple consistency checks */
1030 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1031 !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1034 /* Now read in all of the header information */
1036 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1037 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1040 elf_phdata = kmalloc(j, GFP_KERNEL);
1046 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1050 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1051 if ((eppnt + i)->p_type == PT_LOAD)
1056 while (eppnt->p_type != PT_LOAD)
1059 /* Now use mmap to map the library into memory. */
1060 error = vm_mmap(file,
1061 ELF_PAGESTART(eppnt->p_vaddr),
1063 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1064 PROT_READ | PROT_WRITE | PROT_EXEC,
1065 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1067 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1068 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1071 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1072 if (padzero(elf_bss)) {
1077 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1079 bss = eppnt->p_memsz + eppnt->p_vaddr;
1081 vm_brk(len, bss - len);
1090 #ifdef CONFIG_ELF_CORE
1094 * Modelled on fs/exec.c:aout_core_dump()
1095 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1099 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1100 * that are useful for post-mortem analysis are included in every core dump.
1101 * In that way we ensure that the core dump is fully interpretable later
1102 * without matching up the same kernel and hardware config to see what PC values
1103 * meant. These special mappings include - vDSO, vsyscall, and other
1104 * architecture specific mappings
1106 static bool always_dump_vma(struct vm_area_struct *vma)
1108 /* Any vsyscall mappings? */
1109 if (vma == get_gate_vma(vma->vm_mm))
1112 * arch_vma_name() returns non-NULL for special architecture mappings,
1113 * such as vDSO sections.
1115 if (arch_vma_name(vma))
1122 * Decide what to dump of a segment, part, all or none.
1124 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1125 unsigned long mm_flags)
1127 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1129 /* always dump the vdso and vsyscall sections */
1130 if (always_dump_vma(vma))
1133 if (vma->vm_flags & VM_DONTDUMP)
1136 /* Hugetlb memory check */
1137 if (vma->vm_flags & VM_HUGETLB) {
1138 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1140 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1145 /* Do not dump I/O mapped devices or special mappings */
1146 if (vma->vm_flags & VM_IO)
1149 /* By default, dump shared memory if mapped from an anonymous file. */
1150 if (vma->vm_flags & VM_SHARED) {
1151 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1152 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1157 /* Dump segments that have been written to. */
1158 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1160 if (vma->vm_file == NULL)
1163 if (FILTER(MAPPED_PRIVATE))
1167 * If this looks like the beginning of a DSO or executable mapping,
1168 * check for an ELF header. If we find one, dump the first page to
1169 * aid in determining what was mapped here.
1171 if (FILTER(ELF_HEADERS) &&
1172 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1173 u32 __user *header = (u32 __user *) vma->vm_start;
1175 mm_segment_t fs = get_fs();
1177 * Doing it this way gets the constant folded by GCC.
1181 char elfmag[SELFMAG];
1183 BUILD_BUG_ON(SELFMAG != sizeof word);
1184 magic.elfmag[EI_MAG0] = ELFMAG0;
1185 magic.elfmag[EI_MAG1] = ELFMAG1;
1186 magic.elfmag[EI_MAG2] = ELFMAG2;
1187 magic.elfmag[EI_MAG3] = ELFMAG3;
1189 * Switch to the user "segment" for get_user(),
1190 * then put back what elf_core_dump() had in place.
1193 if (unlikely(get_user(word, header)))
1196 if (word == magic.cmp)
1205 return vma->vm_end - vma->vm_start;
1208 /* An ELF note in memory */
1213 unsigned int datasz;
1217 static int notesize(struct memelfnote *en)
1221 sz = sizeof(struct elf_note);
1222 sz += roundup(strlen(en->name) + 1, 4);
1223 sz += roundup(en->datasz, 4);
1228 static int alignfile(struct coredump_params *cprm)
1230 static const char buf[4] = { 0, };
1231 return dump_emit(cprm, buf, roundup(cprm->written, 4) - cprm->written);
1234 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1237 en.n_namesz = strlen(men->name) + 1;
1238 en.n_descsz = men->datasz;
1239 en.n_type = men->type;
1241 return dump_emit(cprm, &en, sizeof(en)) &&
1242 dump_emit(cprm, men->name, en.n_namesz) && alignfile(cprm) &&
1243 dump_emit(cprm, men->data, men->datasz) && alignfile(cprm);
1246 static void fill_elf_header(struct elfhdr *elf, int segs,
1247 u16 machine, u32 flags)
1249 memset(elf, 0, sizeof(*elf));
1251 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1252 elf->e_ident[EI_CLASS] = ELF_CLASS;
1253 elf->e_ident[EI_DATA] = ELF_DATA;
1254 elf->e_ident[EI_VERSION] = EV_CURRENT;
1255 elf->e_ident[EI_OSABI] = ELF_OSABI;
1257 elf->e_type = ET_CORE;
1258 elf->e_machine = machine;
1259 elf->e_version = EV_CURRENT;
1260 elf->e_phoff = sizeof(struct elfhdr);
1261 elf->e_flags = flags;
1262 elf->e_ehsize = sizeof(struct elfhdr);
1263 elf->e_phentsize = sizeof(struct elf_phdr);
1264 elf->e_phnum = segs;
1269 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1271 phdr->p_type = PT_NOTE;
1272 phdr->p_offset = offset;
1275 phdr->p_filesz = sz;
1282 static void fill_note(struct memelfnote *note, const char *name, int type,
1283 unsigned int sz, void *data)
1293 * fill up all the fields in prstatus from the given task struct, except
1294 * registers which need to be filled up separately.
1296 static void fill_prstatus(struct elf_prstatus *prstatus,
1297 struct task_struct *p, long signr)
1299 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1300 prstatus->pr_sigpend = p->pending.signal.sig[0];
1301 prstatus->pr_sighold = p->blocked.sig[0];
1303 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1305 prstatus->pr_pid = task_pid_vnr(p);
1306 prstatus->pr_pgrp = task_pgrp_vnr(p);
1307 prstatus->pr_sid = task_session_vnr(p);
1308 if (thread_group_leader(p)) {
1309 struct task_cputime cputime;
1312 * This is the record for the group leader. It shows the
1313 * group-wide total, not its individual thread total.
1315 thread_group_cputime(p, &cputime);
1316 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1317 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1319 cputime_t utime, stime;
1321 task_cputime(p, &utime, &stime);
1322 cputime_to_timeval(utime, &prstatus->pr_utime);
1323 cputime_to_timeval(stime, &prstatus->pr_stime);
1325 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1326 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1329 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1330 struct mm_struct *mm)
1332 const struct cred *cred;
1333 unsigned int i, len;
1335 /* first copy the parameters from user space */
1336 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1338 len = mm->arg_end - mm->arg_start;
1339 if (len >= ELF_PRARGSZ)
1340 len = ELF_PRARGSZ-1;
1341 if (copy_from_user(&psinfo->pr_psargs,
1342 (const char __user *)mm->arg_start, len))
1344 for(i = 0; i < len; i++)
1345 if (psinfo->pr_psargs[i] == 0)
1346 psinfo->pr_psargs[i] = ' ';
1347 psinfo->pr_psargs[len] = 0;
1350 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1352 psinfo->pr_pid = task_pid_vnr(p);
1353 psinfo->pr_pgrp = task_pgrp_vnr(p);
1354 psinfo->pr_sid = task_session_vnr(p);
1356 i = p->state ? ffz(~p->state) + 1 : 0;
1357 psinfo->pr_state = i;
1358 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1359 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1360 psinfo->pr_nice = task_nice(p);
1361 psinfo->pr_flag = p->flags;
1363 cred = __task_cred(p);
1364 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1365 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1367 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1372 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1374 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1378 while (auxv[i - 2] != AT_NULL);
1379 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1382 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1385 mm_segment_t old_fs = get_fs();
1387 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1389 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1392 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1394 * Format of NT_FILE note:
1396 * long count -- how many files are mapped
1397 * long page_size -- units for file_ofs
1398 * array of [COUNT] elements of
1402 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1404 static int fill_files_note(struct memelfnote *note)
1406 struct vm_area_struct *vma;
1407 unsigned count, size, names_ofs, remaining, n;
1409 user_long_t *start_end_ofs;
1410 char *name_base, *name_curpos;
1412 /* *Estimated* file count and total data size needed */
1413 count = current->mm->map_count;
1416 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1418 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1420 size = round_up(size, PAGE_SIZE);
1421 data = vmalloc(size);
1425 start_end_ofs = data + 2;
1426 name_base = name_curpos = ((char *)data) + names_ofs;
1427 remaining = size - names_ofs;
1429 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1431 const char *filename;
1433 file = vma->vm_file;
1436 filename = d_path(&file->f_path, name_curpos, remaining);
1437 if (IS_ERR(filename)) {
1438 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1440 size = size * 5 / 4;
1446 /* d_path() fills at the end, move name down */
1447 /* n = strlen(filename) + 1: */
1448 n = (name_curpos + remaining) - filename;
1449 remaining = filename - name_curpos;
1450 memmove(name_curpos, filename, n);
1453 *start_end_ofs++ = vma->vm_start;
1454 *start_end_ofs++ = vma->vm_end;
1455 *start_end_ofs++ = vma->vm_pgoff;
1459 /* Now we know exact count of files, can store it */
1461 data[1] = PAGE_SIZE;
1463 * Count usually is less than current->mm->map_count,
1464 * we need to move filenames down.
1466 n = current->mm->map_count - count;
1468 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1469 memmove(name_base - shift_bytes, name_base,
1470 name_curpos - name_base);
1471 name_curpos -= shift_bytes;
1474 size = name_curpos - (char *)data;
1475 fill_note(note, "CORE", NT_FILE, size, data);
1479 #ifdef CORE_DUMP_USE_REGSET
1480 #include <linux/regset.h>
1482 struct elf_thread_core_info {
1483 struct elf_thread_core_info *next;
1484 struct task_struct *task;
1485 struct elf_prstatus prstatus;
1486 struct memelfnote notes[0];
1489 struct elf_note_info {
1490 struct elf_thread_core_info *thread;
1491 struct memelfnote psinfo;
1492 struct memelfnote signote;
1493 struct memelfnote auxv;
1494 struct memelfnote files;
1495 user_siginfo_t csigdata;
1501 * When a regset has a writeback hook, we call it on each thread before
1502 * dumping user memory. On register window machines, this makes sure the
1503 * user memory backing the register data is up to date before we read it.
1505 static void do_thread_regset_writeback(struct task_struct *task,
1506 const struct user_regset *regset)
1508 if (regset->writeback)
1509 regset->writeback(task, regset, 1);
1513 #define PR_REG_SIZE(S) sizeof(S)
1516 #ifndef PRSTATUS_SIZE
1517 #define PRSTATUS_SIZE(S) sizeof(S)
1521 #define PR_REG_PTR(S) (&((S)->pr_reg))
1524 #ifndef SET_PR_FPVALID
1525 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1528 static int fill_thread_core_info(struct elf_thread_core_info *t,
1529 const struct user_regset_view *view,
1530 long signr, size_t *total)
1535 * NT_PRSTATUS is the one special case, because the regset data
1536 * goes into the pr_reg field inside the note contents, rather
1537 * than being the whole note contents. We fill the reset in here.
1538 * We assume that regset 0 is NT_PRSTATUS.
1540 fill_prstatus(&t->prstatus, t->task, signr);
1541 (void) view->regsets[0].get(t->task, &view->regsets[0],
1542 0, PR_REG_SIZE(t->prstatus.pr_reg),
1543 PR_REG_PTR(&t->prstatus), NULL);
1545 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1546 PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1547 *total += notesize(&t->notes[0]);
1549 do_thread_regset_writeback(t->task, &view->regsets[0]);
1552 * Each other regset might generate a note too. For each regset
1553 * that has no core_note_type or is inactive, we leave t->notes[i]
1554 * all zero and we'll know to skip writing it later.
1556 for (i = 1; i < view->n; ++i) {
1557 const struct user_regset *regset = &view->regsets[i];
1558 do_thread_regset_writeback(t->task, regset);
1559 if (regset->core_note_type && regset->get &&
1560 (!regset->active || regset->active(t->task, regset))) {
1562 size_t size = regset->n * regset->size;
1563 void *data = kmalloc(size, GFP_KERNEL);
1564 if (unlikely(!data))
1566 ret = regset->get(t->task, regset,
1567 0, size, data, NULL);
1571 if (regset->core_note_type != NT_PRFPREG)
1572 fill_note(&t->notes[i], "LINUX",
1573 regset->core_note_type,
1576 SET_PR_FPVALID(&t->prstatus, 1);
1577 fill_note(&t->notes[i], "CORE",
1578 NT_PRFPREG, size, data);
1580 *total += notesize(&t->notes[i]);
1588 static int fill_note_info(struct elfhdr *elf, int phdrs,
1589 struct elf_note_info *info,
1590 siginfo_t *siginfo, struct pt_regs *regs)
1592 struct task_struct *dump_task = current;
1593 const struct user_regset_view *view = task_user_regset_view(dump_task);
1594 struct elf_thread_core_info *t;
1595 struct elf_prpsinfo *psinfo;
1596 struct core_thread *ct;
1600 info->thread = NULL;
1602 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1603 if (psinfo == NULL) {
1604 info->psinfo.data = NULL; /* So we don't free this wrongly */
1608 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1611 * Figure out how many notes we're going to need for each thread.
1613 info->thread_notes = 0;
1614 for (i = 0; i < view->n; ++i)
1615 if (view->regsets[i].core_note_type != 0)
1616 ++info->thread_notes;
1619 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1620 * since it is our one special case.
1622 if (unlikely(info->thread_notes == 0) ||
1623 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1629 * Initialize the ELF file header.
1631 fill_elf_header(elf, phdrs,
1632 view->e_machine, view->e_flags);
1635 * Allocate a structure for each thread.
1637 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1638 t = kzalloc(offsetof(struct elf_thread_core_info,
1639 notes[info->thread_notes]),
1645 if (ct->task == dump_task || !info->thread) {
1646 t->next = info->thread;
1650 * Make sure to keep the original task at
1651 * the head of the list.
1653 t->next = info->thread->next;
1654 info->thread->next = t;
1659 * Now fill in each thread's information.
1661 for (t = info->thread; t != NULL; t = t->next)
1662 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1666 * Fill in the two process-wide notes.
1668 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1669 info->size += notesize(&info->psinfo);
1671 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1672 info->size += notesize(&info->signote);
1674 fill_auxv_note(&info->auxv, current->mm);
1675 info->size += notesize(&info->auxv);
1677 if (fill_files_note(&info->files) == 0)
1678 info->size += notesize(&info->files);
1683 static size_t get_note_info_size(struct elf_note_info *info)
1689 * Write all the notes for each thread. When writing the first thread, the
1690 * process-wide notes are interleaved after the first thread-specific note.
1692 static int write_note_info(struct elf_note_info *info,
1693 struct coredump_params *cprm)
1696 struct elf_thread_core_info *t = info->thread;
1701 if (!writenote(&t->notes[0], cprm))
1704 if (first && !writenote(&info->psinfo, cprm))
1706 if (first && !writenote(&info->signote, cprm))
1708 if (first && !writenote(&info->auxv, cprm))
1710 if (first && info->files.data &&
1711 !writenote(&info->files, cprm))
1714 for (i = 1; i < info->thread_notes; ++i)
1715 if (t->notes[i].data &&
1716 !writenote(&t->notes[i], cprm))
1726 static void free_note_info(struct elf_note_info *info)
1728 struct elf_thread_core_info *threads = info->thread;
1731 struct elf_thread_core_info *t = threads;
1733 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1734 for (i = 1; i < info->thread_notes; ++i)
1735 kfree(t->notes[i].data);
1738 kfree(info->psinfo.data);
1739 vfree(info->files.data);
1744 /* Here is the structure in which status of each thread is captured. */
1745 struct elf_thread_status
1747 struct list_head list;
1748 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1749 elf_fpregset_t fpu; /* NT_PRFPREG */
1750 struct task_struct *thread;
1751 #ifdef ELF_CORE_COPY_XFPREGS
1752 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1754 struct memelfnote notes[3];
1759 * In order to add the specific thread information for the elf file format,
1760 * we need to keep a linked list of every threads pr_status and then create
1761 * a single section for them in the final core file.
1763 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1766 struct task_struct *p = t->thread;
1769 fill_prstatus(&t->prstatus, p, signr);
1770 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1772 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1775 sz += notesize(&t->notes[0]);
1777 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1779 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1782 sz += notesize(&t->notes[1]);
1785 #ifdef ELF_CORE_COPY_XFPREGS
1786 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1787 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1788 sizeof(t->xfpu), &t->xfpu);
1790 sz += notesize(&t->notes[2]);
1796 struct elf_note_info {
1797 struct memelfnote *notes;
1798 struct memelfnote *notes_files;
1799 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1800 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1801 struct list_head thread_list;
1802 elf_fpregset_t *fpu;
1803 #ifdef ELF_CORE_COPY_XFPREGS
1804 elf_fpxregset_t *xfpu;
1806 user_siginfo_t csigdata;
1807 int thread_status_size;
1811 static int elf_note_info_init(struct elf_note_info *info)
1813 memset(info, 0, sizeof(*info));
1814 INIT_LIST_HEAD(&info->thread_list);
1816 /* Allocate space for ELF notes */
1817 info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1820 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1823 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1824 if (!info->prstatus)
1826 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1829 #ifdef ELF_CORE_COPY_XFPREGS
1830 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1837 static int fill_note_info(struct elfhdr *elf, int phdrs,
1838 struct elf_note_info *info,
1839 siginfo_t *siginfo, struct pt_regs *regs)
1841 struct list_head *t;
1843 if (!elf_note_info_init(info))
1846 if (siginfo->si_signo) {
1847 struct core_thread *ct;
1848 struct elf_thread_status *ets;
1850 for (ct = current->mm->core_state->dumper.next;
1851 ct; ct = ct->next) {
1852 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1856 ets->thread = ct->task;
1857 list_add(&ets->list, &info->thread_list);
1860 list_for_each(t, &info->thread_list) {
1863 ets = list_entry(t, struct elf_thread_status, list);
1864 sz = elf_dump_thread_status(siginfo->si_signo, ets);
1865 info->thread_status_size += sz;
1868 /* now collect the dump for the current */
1869 memset(info->prstatus, 0, sizeof(*info->prstatus));
1870 fill_prstatus(info->prstatus, current, siginfo->si_signo);
1871 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1874 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1877 * Set up the notes in similar form to SVR4 core dumps made
1878 * with info from their /proc.
1881 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1882 sizeof(*info->prstatus), info->prstatus);
1883 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1884 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1885 sizeof(*info->psinfo), info->psinfo);
1887 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1888 fill_auxv_note(info->notes + 3, current->mm);
1891 if (fill_files_note(info->notes + info->numnote) == 0) {
1892 info->notes_files = info->notes + info->numnote;
1896 /* Try to dump the FPU. */
1897 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1899 if (info->prstatus->pr_fpvalid)
1900 fill_note(info->notes + info->numnote++,
1901 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1902 #ifdef ELF_CORE_COPY_XFPREGS
1903 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1904 fill_note(info->notes + info->numnote++,
1905 "LINUX", ELF_CORE_XFPREG_TYPE,
1906 sizeof(*info->xfpu), info->xfpu);
1912 static size_t get_note_info_size(struct elf_note_info *info)
1917 for (i = 0; i < info->numnote; i++)
1918 sz += notesize(info->notes + i);
1920 sz += info->thread_status_size;
1925 static int write_note_info(struct elf_note_info *info,
1926 struct coredump_params *cprm)
1929 struct list_head *t;
1931 for (i = 0; i < info->numnote; i++)
1932 if (!writenote(info->notes + i, cprm))
1935 /* write out the thread status notes section */
1936 list_for_each(t, &info->thread_list) {
1937 struct elf_thread_status *tmp =
1938 list_entry(t, struct elf_thread_status, list);
1940 for (i = 0; i < tmp->num_notes; i++)
1941 if (!writenote(&tmp->notes[i], cprm))
1948 static void free_note_info(struct elf_note_info *info)
1950 while (!list_empty(&info->thread_list)) {
1951 struct list_head *tmp = info->thread_list.next;
1953 kfree(list_entry(tmp, struct elf_thread_status, list));
1956 /* Free data possibly allocated by fill_files_note(): */
1957 if (info->notes_files)
1958 vfree(info->notes_files->data);
1960 kfree(info->prstatus);
1961 kfree(info->psinfo);
1964 #ifdef ELF_CORE_COPY_XFPREGS
1971 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1972 struct vm_area_struct *gate_vma)
1974 struct vm_area_struct *ret = tsk->mm->mmap;
1981 * Helper function for iterating across a vma list. It ensures that the caller
1982 * will visit `gate_vma' prior to terminating the search.
1984 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1985 struct vm_area_struct *gate_vma)
1987 struct vm_area_struct *ret;
1989 ret = this_vma->vm_next;
1992 if (this_vma == gate_vma)
1997 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1998 elf_addr_t e_shoff, int segs)
2000 elf->e_shoff = e_shoff;
2001 elf->e_shentsize = sizeof(*shdr4extnum);
2003 elf->e_shstrndx = SHN_UNDEF;
2005 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2007 shdr4extnum->sh_type = SHT_NULL;
2008 shdr4extnum->sh_size = elf->e_shnum;
2009 shdr4extnum->sh_link = elf->e_shstrndx;
2010 shdr4extnum->sh_info = segs;
2013 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2014 unsigned long mm_flags)
2016 struct vm_area_struct *vma;
2019 for (vma = first_vma(current, gate_vma); vma != NULL;
2020 vma = next_vma(vma, gate_vma))
2021 size += vma_dump_size(vma, mm_flags);
2028 * This is a two-pass process; first we find the offsets of the bits,
2029 * and then they are actually written out. If we run out of core limit
2032 static int elf_core_dump(struct coredump_params *cprm)
2038 struct vm_area_struct *vma, *gate_vma;
2039 struct elfhdr *elf = NULL;
2040 loff_t offset = 0, dataoff, foffset;
2041 struct elf_note_info info = { };
2042 struct elf_phdr *phdr4note = NULL;
2043 struct elf_shdr *shdr4extnum = NULL;
2048 * We no longer stop all VM operations.
2050 * This is because those proceses that could possibly change map_count
2051 * or the mmap / vma pages are now blocked in do_exit on current
2052 * finishing this core dump.
2054 * Only ptrace can touch these memory addresses, but it doesn't change
2055 * the map_count or the pages allocated. So no possibility of crashing
2056 * exists while dumping the mm->vm_next areas to the core file.
2059 /* alloc memory for large data structures: too large to be on stack */
2060 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2064 * The number of segs are recored into ELF header as 16bit value.
2065 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2067 segs = current->mm->map_count;
2068 segs += elf_core_extra_phdrs();
2070 gate_vma = get_gate_vma(current->mm);
2071 if (gate_vma != NULL)
2074 /* for notes section */
2077 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2078 * this, kernel supports extended numbering. Have a look at
2079 * include/linux/elf.h for further information. */
2080 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2083 * Collect all the non-memory information about the process for the
2084 * notes. This also sets up the file header.
2086 if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2094 offset += sizeof(*elf); /* Elf header */
2095 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2098 /* Write notes phdr entry */
2100 size_t sz = get_note_info_size(&info);
2102 sz += elf_coredump_extra_notes_size();
2104 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2108 fill_elf_note_phdr(phdr4note, sz, offset);
2112 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2114 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2115 offset += elf_core_extra_data_size();
2118 if (e_phnum == PN_XNUM) {
2119 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2122 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2127 if (!dump_emit(cprm, elf, sizeof(*elf)))
2130 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2133 /* Write program headers for segments dump */
2134 for (vma = first_vma(current, gate_vma); vma != NULL;
2135 vma = next_vma(vma, gate_vma)) {
2136 struct elf_phdr phdr;
2138 phdr.p_type = PT_LOAD;
2139 phdr.p_offset = offset;
2140 phdr.p_vaddr = vma->vm_start;
2142 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2143 phdr.p_memsz = vma->vm_end - vma->vm_start;
2144 offset += phdr.p_filesz;
2145 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2146 if (vma->vm_flags & VM_WRITE)
2147 phdr.p_flags |= PF_W;
2148 if (vma->vm_flags & VM_EXEC)
2149 phdr.p_flags |= PF_X;
2150 phdr.p_align = ELF_EXEC_PAGESIZE;
2152 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2156 if (!elf_core_write_extra_phdrs(cprm, offset))
2159 size = cprm->written;
2160 cprm->written = foffset; /* will disappear */
2161 /* write out the notes section */
2162 if (!write_note_info(&info, cprm))
2165 foffset = cprm->written;
2166 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2170 if (!dump_seek(cprm->file, dataoff - foffset))
2173 for (vma = first_vma(current, gate_vma); vma != NULL;
2174 vma = next_vma(vma, gate_vma)) {
2178 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2180 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2184 page = get_dump_page(addr);
2186 void *kaddr = kmap(page);
2187 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2188 !dump_write(cprm->file, kaddr,
2191 page_cache_release(page);
2193 stop = !dump_seek(cprm->file, PAGE_SIZE);
2199 cprm->written = size;
2200 if (!elf_core_write_extra_data(cprm))
2202 size = cprm->written;
2204 if (e_phnum == PN_XNUM) {
2205 size += sizeof(*shdr4extnum);
2206 if (size > cprm->limit
2207 || !dump_write(cprm->file, shdr4extnum,
2208 sizeof(*shdr4extnum)))
2216 free_note_info(&info);
2224 #endif /* CONFIG_ELF_CORE */
2226 static int __init init_elf_binfmt(void)
2228 register_binfmt(&elf_format);
2232 static void __exit exit_elf_binfmt(void)
2234 /* Remove the COFF and ELF loaders. */
2235 unregister_binfmt(&elf_format);
2238 core_initcall(init_elf_binfmt);
2239 module_exit(exit_elf_binfmt);
2240 MODULE_LICENSE("GPL");