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 || !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 long random_variable = 0;
557 if ((current->flags & PF_RANDOMIZE) &&
558 !(current->personality & ADDR_NO_RANDOMIZE)) {
559 random_variable = (unsigned long) get_random_int();
560 random_variable &= STACK_RND_MASK;
561 random_variable <<= PAGE_SHIFT;
563 #ifdef CONFIG_STACK_GROWSUP
564 return PAGE_ALIGN(stack_top) + random_variable;
566 return PAGE_ALIGN(stack_top) - random_variable;
570 static int load_elf_binary(struct linux_binprm *bprm)
572 struct file *interpreter = NULL; /* to shut gcc up */
573 unsigned long load_addr = 0, load_bias = 0;
574 int load_addr_set = 0;
575 char * elf_interpreter = NULL;
577 struct elf_phdr *elf_ppnt, *elf_phdata;
578 unsigned long elf_bss, elf_brk;
581 unsigned long elf_entry;
582 unsigned long interp_load_addr = 0;
583 unsigned long start_code, end_code, start_data, end_data;
584 unsigned long reloc_func_desc __maybe_unused = 0;
585 int executable_stack = EXSTACK_DEFAULT;
586 unsigned long def_flags = 0;
587 struct pt_regs *regs = current_pt_regs();
589 struct elfhdr elf_ex;
590 struct elfhdr interp_elf_ex;
593 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
599 /* Get the exec-header */
600 loc->elf_ex = *((struct elfhdr *)bprm->buf);
603 /* First of all, some simple consistency checks */
604 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
607 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
609 if (!elf_check_arch(&loc->elf_ex))
611 if (!bprm->file->f_op || !bprm->file->f_op->mmap)
614 /* Now read in all of the header information */
615 if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
617 if (loc->elf_ex.e_phnum < 1 ||
618 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
620 size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
622 elf_phdata = kmalloc(size, GFP_KERNEL);
626 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
627 (char *)elf_phdata, size);
628 if (retval != size) {
634 elf_ppnt = elf_phdata;
643 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
644 if (elf_ppnt->p_type == PT_INTERP) {
645 /* This is the program interpreter used for
646 * shared libraries - for now assume that this
647 * is an a.out format binary
650 if (elf_ppnt->p_filesz > PATH_MAX ||
651 elf_ppnt->p_filesz < 2)
655 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
657 if (!elf_interpreter)
660 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
663 if (retval != elf_ppnt->p_filesz) {
666 goto out_free_interp;
668 /* make sure path is NULL terminated */
670 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
671 goto out_free_interp;
673 interpreter = open_exec(elf_interpreter);
674 retval = PTR_ERR(interpreter);
675 if (IS_ERR(interpreter))
676 goto out_free_interp;
679 * If the binary is not readable then enforce
680 * mm->dumpable = 0 regardless of the interpreter's
683 would_dump(bprm, interpreter);
685 retval = kernel_read(interpreter, 0, bprm->buf,
687 if (retval != BINPRM_BUF_SIZE) {
690 goto out_free_dentry;
693 /* Get the exec headers */
694 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
700 elf_ppnt = elf_phdata;
701 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
702 if (elf_ppnt->p_type == PT_GNU_STACK) {
703 if (elf_ppnt->p_flags & PF_X)
704 executable_stack = EXSTACK_ENABLE_X;
706 executable_stack = EXSTACK_DISABLE_X;
710 /* Some simple consistency checks for the interpreter */
711 if (elf_interpreter) {
713 /* Not an ELF interpreter */
714 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
715 goto out_free_dentry;
716 /* Verify the interpreter has a valid arch */
717 if (!elf_check_arch(&loc->interp_elf_ex))
718 goto out_free_dentry;
721 /* Flush all traces of the currently running executable */
722 retval = flush_old_exec(bprm);
724 goto out_free_dentry;
726 /* OK, This is the point of no return */
727 current->mm->def_flags = def_flags;
729 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
730 may depend on the personality. */
731 SET_PERSONALITY(loc->elf_ex);
732 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
733 current->personality |= READ_IMPLIES_EXEC;
735 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
736 current->flags |= PF_RANDOMIZE;
738 setup_new_exec(bprm);
740 /* Do this so that we can load the interpreter, if need be. We will
741 change some of these later */
742 current->mm->free_area_cache = current->mm->mmap_base;
743 current->mm->cached_hole_size = 0;
744 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
747 send_sig(SIGKILL, current, 0);
748 goto out_free_dentry;
751 current->mm->start_stack = bprm->p;
753 /* Now we do a little grungy work by mmapping the ELF image into
754 the correct location in memory. */
755 for(i = 0, elf_ppnt = elf_phdata;
756 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
757 int elf_prot = 0, elf_flags;
758 unsigned long k, vaddr;
759 unsigned long total_size = 0;
761 if (elf_ppnt->p_type != PT_LOAD)
764 if (unlikely (elf_brk > elf_bss)) {
767 /* There was a PT_LOAD segment with p_memsz > p_filesz
768 before this one. Map anonymous pages, if needed,
769 and clear the area. */
770 retval = set_brk(elf_bss + load_bias,
771 elf_brk + load_bias);
773 send_sig(SIGKILL, current, 0);
774 goto out_free_dentry;
776 nbyte = ELF_PAGEOFFSET(elf_bss);
778 nbyte = ELF_MIN_ALIGN - nbyte;
779 if (nbyte > elf_brk - elf_bss)
780 nbyte = elf_brk - elf_bss;
781 if (clear_user((void __user *)elf_bss +
784 * This bss-zeroing can fail if the ELF
785 * file specifies odd protections. So
786 * we don't check the return value
792 if (elf_ppnt->p_flags & PF_R)
793 elf_prot |= PROT_READ;
794 if (elf_ppnt->p_flags & PF_W)
795 elf_prot |= PROT_WRITE;
796 if (elf_ppnt->p_flags & PF_X)
797 elf_prot |= PROT_EXEC;
799 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
801 vaddr = elf_ppnt->p_vaddr;
802 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
803 elf_flags |= MAP_FIXED;
804 } else if (loc->elf_ex.e_type == ET_DYN) {
805 /* Try and get dynamic programs out of the way of the
806 * default mmap base, as well as whatever program they
807 * might try to exec. This is because the brk will
808 * follow the loader, and is not movable. */
809 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
810 /* Memory randomization might have been switched off
811 * in runtime via sysctl or explicit setting of
813 * If that is the case, retain the original non-zero
814 * load_bias value in order to establish proper
815 * non-randomized mappings.
817 if (current->flags & PF_RANDOMIZE)
820 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
822 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
824 total_size = total_mapping_size(elf_phdata,
825 loc->elf_ex.e_phnum);
828 goto out_free_dentry;
832 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
833 elf_prot, elf_flags, total_size);
834 if (BAD_ADDR(error)) {
835 send_sig(SIGKILL, current, 0);
836 retval = IS_ERR((void *)error) ?
837 PTR_ERR((void*)error) : -EINVAL;
838 goto out_free_dentry;
841 if (!load_addr_set) {
843 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
844 if (loc->elf_ex.e_type == ET_DYN) {
846 ELF_PAGESTART(load_bias + vaddr);
847 load_addr += load_bias;
848 reloc_func_desc = load_bias;
851 k = elf_ppnt->p_vaddr;
858 * Check to see if the section's size will overflow the
859 * allowed task size. Note that p_filesz must always be
860 * <= p_memsz so it is only necessary to check p_memsz.
862 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
863 elf_ppnt->p_memsz > TASK_SIZE ||
864 TASK_SIZE - elf_ppnt->p_memsz < k) {
865 /* set_brk can never work. Avoid overflows. */
866 send_sig(SIGKILL, current, 0);
868 goto out_free_dentry;
871 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
875 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
879 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
884 loc->elf_ex.e_entry += load_bias;
885 elf_bss += load_bias;
886 elf_brk += load_bias;
887 start_code += load_bias;
888 end_code += load_bias;
889 start_data += load_bias;
890 end_data += load_bias;
892 /* Calling set_brk effectively mmaps the pages that we need
893 * for the bss and break sections. We must do this before
894 * mapping in the interpreter, to make sure it doesn't wind
895 * up getting placed where the bss needs to go.
897 retval = set_brk(elf_bss, elf_brk);
899 send_sig(SIGKILL, current, 0);
900 goto out_free_dentry;
902 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
903 send_sig(SIGSEGV, current, 0);
904 retval = -EFAULT; /* Nobody gets to see this, but.. */
905 goto out_free_dentry;
908 if (elf_interpreter) {
909 unsigned long interp_map_addr = 0;
911 elf_entry = load_elf_interp(&loc->interp_elf_ex,
915 if (!IS_ERR((void *)elf_entry)) {
917 * load_elf_interp() returns relocation
920 interp_load_addr = elf_entry;
921 elf_entry += loc->interp_elf_ex.e_entry;
923 if (BAD_ADDR(elf_entry)) {
924 force_sig(SIGSEGV, current);
925 retval = IS_ERR((void *)elf_entry) ?
926 (int)elf_entry : -EINVAL;
927 goto out_free_dentry;
929 reloc_func_desc = interp_load_addr;
931 allow_write_access(interpreter);
933 kfree(elf_interpreter);
935 elf_entry = loc->elf_ex.e_entry;
936 if (BAD_ADDR(elf_entry)) {
937 force_sig(SIGSEGV, current);
939 goto out_free_dentry;
945 set_binfmt(&elf_format);
947 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
948 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
950 send_sig(SIGKILL, current, 0);
953 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
955 install_exec_creds(bprm);
956 retval = create_elf_tables(bprm, &loc->elf_ex,
957 load_addr, interp_load_addr);
959 send_sig(SIGKILL, current, 0);
962 /* N.B. passed_fileno might not be initialized? */
963 current->mm->end_code = end_code;
964 current->mm->start_code = start_code;
965 current->mm->start_data = start_data;
966 current->mm->end_data = end_data;
967 current->mm->start_stack = bprm->p;
969 #ifdef arch_randomize_brk
970 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
971 current->mm->brk = current->mm->start_brk =
972 arch_randomize_brk(current->mm);
973 #ifdef CONFIG_COMPAT_BRK
974 current->brk_randomized = 1;
979 if (current->personality & MMAP_PAGE_ZERO) {
980 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
981 and some applications "depend" upon this behavior.
982 Since we do not have the power to recompile these, we
983 emulate the SVr4 behavior. Sigh. */
984 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
985 MAP_FIXED | MAP_PRIVATE, 0);
990 * The ABI may specify that certain registers be set up in special
991 * ways (on i386 %edx is the address of a DT_FINI function, for
992 * example. In addition, it may also specify (eg, PowerPC64 ELF)
993 * that the e_entry field is the address of the function descriptor
994 * for the startup routine, rather than the address of the startup
995 * routine itself. This macro performs whatever initialization to
996 * the regs structure is required as well as any relocations to the
997 * function descriptor entries when executing dynamically links apps.
999 ELF_PLAT_INIT(regs, reloc_func_desc);
1002 start_thread(regs, elf_entry, bprm->p);
1011 allow_write_access(interpreter);
1015 kfree(elf_interpreter);
1021 /* This is really simpleminded and specialized - we are loading an
1022 a.out library that is given an ELF header. */
1023 static int load_elf_library(struct file *file)
1025 struct elf_phdr *elf_phdata;
1026 struct elf_phdr *eppnt;
1027 unsigned long elf_bss, bss, len;
1028 int retval, error, i, j;
1029 struct elfhdr elf_ex;
1032 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1033 if (retval != sizeof(elf_ex))
1036 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1039 /* First of all, some simple consistency checks */
1040 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1041 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1044 /* Now read in all of the header information */
1046 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1047 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1050 elf_phdata = kmalloc(j, GFP_KERNEL);
1056 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1060 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1061 if ((eppnt + i)->p_type == PT_LOAD)
1066 while (eppnt->p_type != PT_LOAD)
1069 /* Now use mmap to map the library into memory. */
1070 error = vm_mmap(file,
1071 ELF_PAGESTART(eppnt->p_vaddr),
1073 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1074 PROT_READ | PROT_WRITE | PROT_EXEC,
1075 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1077 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1078 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1081 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1082 if (padzero(elf_bss)) {
1087 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1089 bss = eppnt->p_memsz + eppnt->p_vaddr;
1091 vm_brk(len, bss - len);
1100 #ifdef CONFIG_ELF_CORE
1104 * Modelled on fs/exec.c:aout_core_dump()
1105 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1109 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1110 * that are useful for post-mortem analysis are included in every core dump.
1111 * In that way we ensure that the core dump is fully interpretable later
1112 * without matching up the same kernel and hardware config to see what PC values
1113 * meant. These special mappings include - vDSO, vsyscall, and other
1114 * architecture specific mappings
1116 static bool always_dump_vma(struct vm_area_struct *vma)
1118 /* Any vsyscall mappings? */
1119 if (vma == get_gate_vma(vma->vm_mm))
1122 * arch_vma_name() returns non-NULL for special architecture mappings,
1123 * such as vDSO sections.
1125 if (arch_vma_name(vma))
1132 * Decide what to dump of a segment, part, all or none.
1134 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1135 unsigned long mm_flags)
1137 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1139 /* always dump the vdso and vsyscall sections */
1140 if (always_dump_vma(vma))
1143 if (vma->vm_flags & VM_DONTDUMP)
1146 /* Hugetlb memory check */
1147 if (vma->vm_flags & VM_HUGETLB) {
1148 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1150 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1155 /* Do not dump I/O mapped devices or special mappings */
1156 if (vma->vm_flags & VM_IO)
1159 /* By default, dump shared memory if mapped from an anonymous file. */
1160 if (vma->vm_flags & VM_SHARED) {
1161 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1162 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1167 /* Dump segments that have been written to. */
1168 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1170 if (vma->vm_file == NULL)
1173 if (FILTER(MAPPED_PRIVATE))
1177 * If this looks like the beginning of a DSO or executable mapping,
1178 * check for an ELF header. If we find one, dump the first page to
1179 * aid in determining what was mapped here.
1181 if (FILTER(ELF_HEADERS) &&
1182 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1183 u32 __user *header = (u32 __user *) vma->vm_start;
1185 mm_segment_t fs = get_fs();
1187 * Doing it this way gets the constant folded by GCC.
1191 char elfmag[SELFMAG];
1193 BUILD_BUG_ON(SELFMAG != sizeof word);
1194 magic.elfmag[EI_MAG0] = ELFMAG0;
1195 magic.elfmag[EI_MAG1] = ELFMAG1;
1196 magic.elfmag[EI_MAG2] = ELFMAG2;
1197 magic.elfmag[EI_MAG3] = ELFMAG3;
1199 * Switch to the user "segment" for get_user(),
1200 * then put back what elf_core_dump() had in place.
1203 if (unlikely(get_user(word, header)))
1206 if (word == magic.cmp)
1215 return vma->vm_end - vma->vm_start;
1218 /* An ELF note in memory */
1223 unsigned int datasz;
1227 static int notesize(struct memelfnote *en)
1231 sz = sizeof(struct elf_note);
1232 sz += roundup(strlen(en->name) + 1, 4);
1233 sz += roundup(en->datasz, 4);
1238 #define DUMP_WRITE(addr, nr, foffset) \
1239 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1241 static int alignfile(struct file *file, loff_t *foffset)
1243 static const char buf[4] = { 0, };
1244 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1248 static int writenote(struct memelfnote *men, struct file *file,
1252 en.n_namesz = strlen(men->name) + 1;
1253 en.n_descsz = men->datasz;
1254 en.n_type = men->type;
1256 DUMP_WRITE(&en, sizeof(en), foffset);
1257 DUMP_WRITE(men->name, en.n_namesz, foffset);
1258 if (!alignfile(file, foffset))
1260 DUMP_WRITE(men->data, men->datasz, foffset);
1261 if (!alignfile(file, foffset))
1268 static void fill_elf_header(struct elfhdr *elf, int segs,
1269 u16 machine, u32 flags)
1271 memset(elf, 0, sizeof(*elf));
1273 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1274 elf->e_ident[EI_CLASS] = ELF_CLASS;
1275 elf->e_ident[EI_DATA] = ELF_DATA;
1276 elf->e_ident[EI_VERSION] = EV_CURRENT;
1277 elf->e_ident[EI_OSABI] = ELF_OSABI;
1279 elf->e_type = ET_CORE;
1280 elf->e_machine = machine;
1281 elf->e_version = EV_CURRENT;
1282 elf->e_phoff = sizeof(struct elfhdr);
1283 elf->e_flags = flags;
1284 elf->e_ehsize = sizeof(struct elfhdr);
1285 elf->e_phentsize = sizeof(struct elf_phdr);
1286 elf->e_phnum = segs;
1291 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1293 phdr->p_type = PT_NOTE;
1294 phdr->p_offset = offset;
1297 phdr->p_filesz = sz;
1304 static void fill_note(struct memelfnote *note, const char *name, int type,
1305 unsigned int sz, void *data)
1315 * fill up all the fields in prstatus from the given task struct, except
1316 * registers which need to be filled up separately.
1318 static void fill_prstatus(struct elf_prstatus *prstatus,
1319 struct task_struct *p, long signr)
1321 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1322 prstatus->pr_sigpend = p->pending.signal.sig[0];
1323 prstatus->pr_sighold = p->blocked.sig[0];
1325 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1327 prstatus->pr_pid = task_pid_vnr(p);
1328 prstatus->pr_pgrp = task_pgrp_vnr(p);
1329 prstatus->pr_sid = task_session_vnr(p);
1330 if (thread_group_leader(p)) {
1331 struct task_cputime cputime;
1334 * This is the record for the group leader. It shows the
1335 * group-wide total, not its individual thread total.
1337 thread_group_cputime(p, &cputime);
1338 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1339 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1341 cputime_t utime, stime;
1343 task_cputime(p, &utime, &stime);
1344 cputime_to_timeval(utime, &prstatus->pr_utime);
1345 cputime_to_timeval(stime, &prstatus->pr_stime);
1347 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1348 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1351 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1352 struct mm_struct *mm)
1354 const struct cred *cred;
1355 unsigned int i, len;
1357 /* first copy the parameters from user space */
1358 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1360 len = mm->arg_end - mm->arg_start;
1361 if (len >= ELF_PRARGSZ)
1362 len = ELF_PRARGSZ-1;
1363 if (copy_from_user(&psinfo->pr_psargs,
1364 (const char __user *)mm->arg_start, len))
1366 for(i = 0; i < len; i++)
1367 if (psinfo->pr_psargs[i] == 0)
1368 psinfo->pr_psargs[i] = ' ';
1369 psinfo->pr_psargs[len] = 0;
1372 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1374 psinfo->pr_pid = task_pid_vnr(p);
1375 psinfo->pr_pgrp = task_pgrp_vnr(p);
1376 psinfo->pr_sid = task_session_vnr(p);
1378 i = p->state ? ffz(~p->state) + 1 : 0;
1379 psinfo->pr_state = i;
1380 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1381 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1382 psinfo->pr_nice = task_nice(p);
1383 psinfo->pr_flag = p->flags;
1385 cred = __task_cred(p);
1386 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1387 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1389 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1394 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1396 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1400 while (auxv[i - 2] != AT_NULL);
1401 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1404 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1407 mm_segment_t old_fs = get_fs();
1409 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1411 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1414 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1416 * Format of NT_FILE note:
1418 * long count -- how many files are mapped
1419 * long page_size -- units for file_ofs
1420 * array of [COUNT] elements of
1424 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1426 static int fill_files_note(struct memelfnote *note)
1428 struct vm_area_struct *vma;
1429 unsigned count, size, names_ofs, remaining, n;
1431 user_long_t *start_end_ofs;
1432 char *name_base, *name_curpos;
1434 /* *Estimated* file count and total data size needed */
1435 count = current->mm->map_count;
1438 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1440 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1442 size = round_up(size, PAGE_SIZE);
1443 data = vmalloc(size);
1447 start_end_ofs = data + 2;
1448 name_base = name_curpos = ((char *)data) + names_ofs;
1449 remaining = size - names_ofs;
1451 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1453 const char *filename;
1455 file = vma->vm_file;
1458 filename = d_path(&file->f_path, name_curpos, remaining);
1459 if (IS_ERR(filename)) {
1460 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1462 size = size * 5 / 4;
1468 /* d_path() fills at the end, move name down */
1469 /* n = strlen(filename) + 1: */
1470 n = (name_curpos + remaining) - filename;
1471 remaining = filename - name_curpos;
1472 memmove(name_curpos, filename, n);
1475 *start_end_ofs++ = vma->vm_start;
1476 *start_end_ofs++ = vma->vm_end;
1477 *start_end_ofs++ = vma->vm_pgoff;
1481 /* Now we know exact count of files, can store it */
1483 data[1] = PAGE_SIZE;
1485 * Count usually is less than current->mm->map_count,
1486 * we need to move filenames down.
1488 n = current->mm->map_count - count;
1490 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1491 memmove(name_base - shift_bytes, name_base,
1492 name_curpos - name_base);
1493 name_curpos -= shift_bytes;
1496 size = name_curpos - (char *)data;
1497 fill_note(note, "CORE", NT_FILE, size, data);
1501 #ifdef CORE_DUMP_USE_REGSET
1502 #include <linux/regset.h>
1504 struct elf_thread_core_info {
1505 struct elf_thread_core_info *next;
1506 struct task_struct *task;
1507 struct elf_prstatus prstatus;
1508 struct memelfnote notes[0];
1511 struct elf_note_info {
1512 struct elf_thread_core_info *thread;
1513 struct memelfnote psinfo;
1514 struct memelfnote signote;
1515 struct memelfnote auxv;
1516 struct memelfnote files;
1517 user_siginfo_t csigdata;
1523 * When a regset has a writeback hook, we call it on each thread before
1524 * dumping user memory. On register window machines, this makes sure the
1525 * user memory backing the register data is up to date before we read it.
1527 static void do_thread_regset_writeback(struct task_struct *task,
1528 const struct user_regset *regset)
1530 if (regset->writeback)
1531 regset->writeback(task, regset, 1);
1535 #define PR_REG_SIZE(S) sizeof(S)
1538 #ifndef PRSTATUS_SIZE
1539 #define PRSTATUS_SIZE(S) sizeof(S)
1543 #define PR_REG_PTR(S) (&((S)->pr_reg))
1546 #ifndef SET_PR_FPVALID
1547 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1550 static int fill_thread_core_info(struct elf_thread_core_info *t,
1551 const struct user_regset_view *view,
1552 long signr, size_t *total)
1557 * NT_PRSTATUS is the one special case, because the regset data
1558 * goes into the pr_reg field inside the note contents, rather
1559 * than being the whole note contents. We fill the reset in here.
1560 * We assume that regset 0 is NT_PRSTATUS.
1562 fill_prstatus(&t->prstatus, t->task, signr);
1563 (void) view->regsets[0].get(t->task, &view->regsets[0],
1564 0, PR_REG_SIZE(t->prstatus.pr_reg),
1565 PR_REG_PTR(&t->prstatus), NULL);
1567 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1568 PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1569 *total += notesize(&t->notes[0]);
1571 do_thread_regset_writeback(t->task, &view->regsets[0]);
1574 * Each other regset might generate a note too. For each regset
1575 * that has no core_note_type or is inactive, we leave t->notes[i]
1576 * all zero and we'll know to skip writing it later.
1578 for (i = 1; i < view->n; ++i) {
1579 const struct user_regset *regset = &view->regsets[i];
1580 do_thread_regset_writeback(t->task, regset);
1581 if (regset->core_note_type && regset->get &&
1582 (!regset->active || regset->active(t->task, regset))) {
1584 size_t size = regset->n * regset->size;
1585 void *data = kmalloc(size, GFP_KERNEL);
1586 if (unlikely(!data))
1588 ret = regset->get(t->task, regset,
1589 0, size, data, NULL);
1593 if (regset->core_note_type != NT_PRFPREG)
1594 fill_note(&t->notes[i], "LINUX",
1595 regset->core_note_type,
1598 SET_PR_FPVALID(&t->prstatus, 1);
1599 fill_note(&t->notes[i], "CORE",
1600 NT_PRFPREG, size, data);
1602 *total += notesize(&t->notes[i]);
1610 static int fill_note_info(struct elfhdr *elf, int phdrs,
1611 struct elf_note_info *info,
1612 siginfo_t *siginfo, struct pt_regs *regs)
1614 struct task_struct *dump_task = current;
1615 const struct user_regset_view *view = task_user_regset_view(dump_task);
1616 struct elf_thread_core_info *t;
1617 struct elf_prpsinfo *psinfo;
1618 struct core_thread *ct;
1622 info->thread = NULL;
1624 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1625 if (psinfo == NULL) {
1626 info->psinfo.data = NULL; /* So we don't free this wrongly */
1630 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1633 * Figure out how many notes we're going to need for each thread.
1635 info->thread_notes = 0;
1636 for (i = 0; i < view->n; ++i)
1637 if (view->regsets[i].core_note_type != 0)
1638 ++info->thread_notes;
1641 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1642 * since it is our one special case.
1644 if (unlikely(info->thread_notes == 0) ||
1645 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1651 * Initialize the ELF file header.
1653 fill_elf_header(elf, phdrs,
1654 view->e_machine, view->e_flags);
1657 * Allocate a structure for each thread.
1659 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1660 t = kzalloc(offsetof(struct elf_thread_core_info,
1661 notes[info->thread_notes]),
1667 if (ct->task == dump_task || !info->thread) {
1668 t->next = info->thread;
1672 * Make sure to keep the original task at
1673 * the head of the list.
1675 t->next = info->thread->next;
1676 info->thread->next = t;
1681 * Now fill in each thread's information.
1683 for (t = info->thread; t != NULL; t = t->next)
1684 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1688 * Fill in the two process-wide notes.
1690 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1691 info->size += notesize(&info->psinfo);
1693 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1694 info->size += notesize(&info->signote);
1696 fill_auxv_note(&info->auxv, current->mm);
1697 info->size += notesize(&info->auxv);
1699 if (fill_files_note(&info->files) == 0)
1700 info->size += notesize(&info->files);
1705 static size_t get_note_info_size(struct elf_note_info *info)
1711 * Write all the notes for each thread. When writing the first thread, the
1712 * process-wide notes are interleaved after the first thread-specific note.
1714 static int write_note_info(struct elf_note_info *info,
1715 struct file *file, loff_t *foffset)
1718 struct elf_thread_core_info *t = info->thread;
1723 if (!writenote(&t->notes[0], file, foffset))
1726 if (first && !writenote(&info->psinfo, file, foffset))
1728 if (first && !writenote(&info->signote, file, foffset))
1730 if (first && !writenote(&info->auxv, file, foffset))
1732 if (first && info->files.data &&
1733 !writenote(&info->files, file, foffset))
1736 for (i = 1; i < info->thread_notes; ++i)
1737 if (t->notes[i].data &&
1738 !writenote(&t->notes[i], file, foffset))
1748 static void free_note_info(struct elf_note_info *info)
1750 struct elf_thread_core_info *threads = info->thread;
1753 struct elf_thread_core_info *t = threads;
1755 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1756 for (i = 1; i < info->thread_notes; ++i)
1757 kfree(t->notes[i].data);
1760 kfree(info->psinfo.data);
1761 vfree(info->files.data);
1766 /* Here is the structure in which status of each thread is captured. */
1767 struct elf_thread_status
1769 struct list_head list;
1770 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1771 elf_fpregset_t fpu; /* NT_PRFPREG */
1772 struct task_struct *thread;
1773 #ifdef ELF_CORE_COPY_XFPREGS
1774 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1776 struct memelfnote notes[3];
1781 * In order to add the specific thread information for the elf file format,
1782 * we need to keep a linked list of every threads pr_status and then create
1783 * a single section for them in the final core file.
1785 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1788 struct task_struct *p = t->thread;
1791 fill_prstatus(&t->prstatus, p, signr);
1792 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1794 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1797 sz += notesize(&t->notes[0]);
1799 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1801 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1804 sz += notesize(&t->notes[1]);
1807 #ifdef ELF_CORE_COPY_XFPREGS
1808 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1809 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1810 sizeof(t->xfpu), &t->xfpu);
1812 sz += notesize(&t->notes[2]);
1818 struct elf_note_info {
1819 struct memelfnote *notes;
1820 struct memelfnote *notes_files;
1821 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1822 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1823 struct list_head thread_list;
1824 elf_fpregset_t *fpu;
1825 #ifdef ELF_CORE_COPY_XFPREGS
1826 elf_fpxregset_t *xfpu;
1828 user_siginfo_t csigdata;
1829 int thread_status_size;
1833 static int elf_note_info_init(struct elf_note_info *info)
1835 memset(info, 0, sizeof(*info));
1836 INIT_LIST_HEAD(&info->thread_list);
1838 /* Allocate space for ELF notes */
1839 info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1842 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1845 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1846 if (!info->prstatus)
1848 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1851 #ifdef ELF_CORE_COPY_XFPREGS
1852 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1859 static int fill_note_info(struct elfhdr *elf, int phdrs,
1860 struct elf_note_info *info,
1861 siginfo_t *siginfo, struct pt_regs *regs)
1863 struct list_head *t;
1865 if (!elf_note_info_init(info))
1868 if (siginfo->si_signo) {
1869 struct core_thread *ct;
1870 struct elf_thread_status *ets;
1872 for (ct = current->mm->core_state->dumper.next;
1873 ct; ct = ct->next) {
1874 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1878 ets->thread = ct->task;
1879 list_add(&ets->list, &info->thread_list);
1882 list_for_each(t, &info->thread_list) {
1885 ets = list_entry(t, struct elf_thread_status, list);
1886 sz = elf_dump_thread_status(siginfo->si_signo, ets);
1887 info->thread_status_size += sz;
1890 /* now collect the dump for the current */
1891 memset(info->prstatus, 0, sizeof(*info->prstatus));
1892 fill_prstatus(info->prstatus, current, siginfo->si_signo);
1893 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1896 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1899 * Set up the notes in similar form to SVR4 core dumps made
1900 * with info from their /proc.
1903 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1904 sizeof(*info->prstatus), info->prstatus);
1905 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1906 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1907 sizeof(*info->psinfo), info->psinfo);
1909 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1910 fill_auxv_note(info->notes + 3, current->mm);
1913 if (fill_files_note(info->notes + info->numnote) == 0) {
1914 info->notes_files = info->notes + info->numnote;
1918 /* Try to dump the FPU. */
1919 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1921 if (info->prstatus->pr_fpvalid)
1922 fill_note(info->notes + info->numnote++,
1923 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1924 #ifdef ELF_CORE_COPY_XFPREGS
1925 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1926 fill_note(info->notes + info->numnote++,
1927 "LINUX", ELF_CORE_XFPREG_TYPE,
1928 sizeof(*info->xfpu), info->xfpu);
1934 static size_t get_note_info_size(struct elf_note_info *info)
1939 for (i = 0; i < info->numnote; i++)
1940 sz += notesize(info->notes + i);
1942 sz += info->thread_status_size;
1947 static int write_note_info(struct elf_note_info *info,
1948 struct file *file, loff_t *foffset)
1951 struct list_head *t;
1953 for (i = 0; i < info->numnote; i++)
1954 if (!writenote(info->notes + i, file, foffset))
1957 /* write out the thread status notes section */
1958 list_for_each(t, &info->thread_list) {
1959 struct elf_thread_status *tmp =
1960 list_entry(t, struct elf_thread_status, list);
1962 for (i = 0; i < tmp->num_notes; i++)
1963 if (!writenote(&tmp->notes[i], file, foffset))
1970 static void free_note_info(struct elf_note_info *info)
1972 while (!list_empty(&info->thread_list)) {
1973 struct list_head *tmp = info->thread_list.next;
1975 kfree(list_entry(tmp, struct elf_thread_status, list));
1978 /* Free data possibly allocated by fill_files_note(): */
1979 if (info->notes_files)
1980 vfree(info->notes_files->data);
1982 kfree(info->prstatus);
1983 kfree(info->psinfo);
1986 #ifdef ELF_CORE_COPY_XFPREGS
1993 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1994 struct vm_area_struct *gate_vma)
1996 struct vm_area_struct *ret = tsk->mm->mmap;
2003 * Helper function for iterating across a vma list. It ensures that the caller
2004 * will visit `gate_vma' prior to terminating the search.
2006 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2007 struct vm_area_struct *gate_vma)
2009 struct vm_area_struct *ret;
2011 ret = this_vma->vm_next;
2014 if (this_vma == gate_vma)
2019 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2020 elf_addr_t e_shoff, int segs)
2022 elf->e_shoff = e_shoff;
2023 elf->e_shentsize = sizeof(*shdr4extnum);
2025 elf->e_shstrndx = SHN_UNDEF;
2027 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2029 shdr4extnum->sh_type = SHT_NULL;
2030 shdr4extnum->sh_size = elf->e_shnum;
2031 shdr4extnum->sh_link = elf->e_shstrndx;
2032 shdr4extnum->sh_info = segs;
2035 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2036 unsigned long mm_flags)
2038 struct vm_area_struct *vma;
2041 for (vma = first_vma(current, gate_vma); vma != NULL;
2042 vma = next_vma(vma, gate_vma))
2043 size += vma_dump_size(vma, mm_flags);
2050 * This is a two-pass process; first we find the offsets of the bits,
2051 * and then they are actually written out. If we run out of core limit
2054 static int elf_core_dump(struct coredump_params *cprm)
2060 struct vm_area_struct *vma, *gate_vma;
2061 struct elfhdr *elf = NULL;
2062 loff_t offset = 0, dataoff, foffset;
2063 struct elf_note_info info = { };
2064 struct elf_phdr *phdr4note = NULL;
2065 struct elf_shdr *shdr4extnum = NULL;
2070 * We no longer stop all VM operations.
2072 * This is because those proceses that could possibly change map_count
2073 * or the mmap / vma pages are now blocked in do_exit on current
2074 * finishing this core dump.
2076 * Only ptrace can touch these memory addresses, but it doesn't change
2077 * the map_count or the pages allocated. So no possibility of crashing
2078 * exists while dumping the mm->vm_next areas to the core file.
2081 /* alloc memory for large data structures: too large to be on stack */
2082 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2086 * The number of segs are recored into ELF header as 16bit value.
2087 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2089 segs = current->mm->map_count;
2090 segs += elf_core_extra_phdrs();
2092 gate_vma = get_gate_vma(current->mm);
2093 if (gate_vma != NULL)
2096 /* for notes section */
2099 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2100 * this, kernel supports extended numbering. Have a look at
2101 * include/linux/elf.h for further information. */
2102 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2105 * Collect all the non-memory information about the process for the
2106 * notes. This also sets up the file header.
2108 if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2116 offset += sizeof(*elf); /* Elf header */
2117 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2120 /* Write notes phdr entry */
2122 size_t sz = get_note_info_size(&info);
2124 sz += elf_coredump_extra_notes_size();
2126 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2130 fill_elf_note_phdr(phdr4note, sz, offset);
2134 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2136 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2137 offset += elf_core_extra_data_size();
2140 if (e_phnum == PN_XNUM) {
2141 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2144 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2149 size += sizeof(*elf);
2150 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2153 size += sizeof(*phdr4note);
2154 if (size > cprm->limit
2155 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2158 /* Write program headers for segments dump */
2159 for (vma = first_vma(current, gate_vma); vma != NULL;
2160 vma = next_vma(vma, gate_vma)) {
2161 struct elf_phdr phdr;
2163 phdr.p_type = PT_LOAD;
2164 phdr.p_offset = offset;
2165 phdr.p_vaddr = vma->vm_start;
2167 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2168 phdr.p_memsz = vma->vm_end - vma->vm_start;
2169 offset += phdr.p_filesz;
2170 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2171 if (vma->vm_flags & VM_WRITE)
2172 phdr.p_flags |= PF_W;
2173 if (vma->vm_flags & VM_EXEC)
2174 phdr.p_flags |= PF_X;
2175 phdr.p_align = ELF_EXEC_PAGESIZE;
2177 size += sizeof(phdr);
2178 if (size > cprm->limit
2179 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2183 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2186 /* write out the notes section */
2187 if (!write_note_info(&info, cprm->file, &foffset))
2190 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2194 if (!dump_seek(cprm->file, dataoff - foffset))
2197 for (vma = first_vma(current, gate_vma); vma != NULL;
2198 vma = next_vma(vma, gate_vma)) {
2202 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2204 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2208 page = get_dump_page(addr);
2210 void *kaddr = kmap(page);
2211 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2212 !dump_write(cprm->file, kaddr,
2215 page_cache_release(page);
2217 stop = !dump_seek(cprm->file, PAGE_SIZE);
2223 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2226 if (e_phnum == PN_XNUM) {
2227 size += sizeof(*shdr4extnum);
2228 if (size > cprm->limit
2229 || !dump_write(cprm->file, shdr4extnum,
2230 sizeof(*shdr4extnum)))
2238 free_note_info(&info);
2246 #endif /* CONFIG_ELF_CORE */
2248 static int __init init_elf_binfmt(void)
2250 register_binfmt(&elf_format);
2254 static void __exit exit_elf_binfmt(void)
2256 /* Remove the COFF and ELF loaders. */
2257 unregister_binfmt(&elf_format);
2260 core_initcall(init_elf_binfmt);
2261 module_exit(exit_elf_binfmt);
2262 MODULE_LICENSE("GPL");