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/security.h>
31 #include <linux/random.h>
32 #include <linux/elf.h>
33 #include <linux/utsname.h>
34 #include <linux/coredump.h>
35 #include <asm/uaccess.h>
36 #include <asm/param.h>
39 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
40 static int load_elf_library(struct file *);
41 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
42 int, int, unsigned long);
45 * If we don't support core dumping, then supply a NULL so we
48 #ifdef CONFIG_ELF_CORE
49 static int elf_core_dump(struct coredump_params *cprm);
51 #define elf_core_dump NULL
54 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
55 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
57 #define ELF_MIN_ALIGN PAGE_SIZE
60 #ifndef ELF_CORE_EFLAGS
61 #define ELF_CORE_EFLAGS 0
64 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
65 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
66 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
68 static struct linux_binfmt elf_format = {
69 .module = THIS_MODULE,
70 .load_binary = load_elf_binary,
71 .load_shlib = load_elf_library,
72 .core_dump = elf_core_dump,
73 .min_coredump = ELF_EXEC_PAGESIZE,
76 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
78 static int set_brk(unsigned long start, unsigned long end)
80 start = ELF_PAGEALIGN(start);
81 end = ELF_PAGEALIGN(end);
84 down_write(¤t->mm->mmap_sem);
85 addr = do_brk(start, end - start);
86 up_write(¤t->mm->mmap_sem);
90 current->mm->start_brk = current->mm->brk = end;
94 /* We need to explicitly zero any fractional pages
95 after the data section (i.e. bss). This would
96 contain the junk from the file that should not
99 static int padzero(unsigned long elf_bss)
103 nbyte = ELF_PAGEOFFSET(elf_bss);
105 nbyte = ELF_MIN_ALIGN - nbyte;
106 if (clear_user((void __user *) elf_bss, nbyte))
112 /* Let's use some macros to make this stack manipulation a little clearer */
113 #ifdef CONFIG_STACK_GROWSUP
114 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
115 #define STACK_ROUND(sp, items) \
116 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
117 #define STACK_ALLOC(sp, len) ({ \
118 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
122 #define STACK_ROUND(sp, items) \
123 (((unsigned long) (sp - items)) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
127 #ifndef ELF_BASE_PLATFORM
129 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
130 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
131 * will be copied to the user stack in the same manner as AT_PLATFORM.
133 #define ELF_BASE_PLATFORM NULL
137 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
138 unsigned long load_addr, unsigned long interp_load_addr)
140 unsigned long p = bprm->p;
141 int argc = bprm->argc;
142 int envc = bprm->envc;
143 elf_addr_t __user *argv;
144 elf_addr_t __user *envp;
145 elf_addr_t __user *sp;
146 elf_addr_t __user *u_platform;
147 elf_addr_t __user *u_base_platform;
148 elf_addr_t __user *u_rand_bytes;
149 const char *k_platform = ELF_PLATFORM;
150 const char *k_base_platform = ELF_BASE_PLATFORM;
151 unsigned char k_rand_bytes[16];
153 elf_addr_t *elf_info;
155 const struct cred *cred = current_cred();
156 struct vm_area_struct *vma;
159 * In some cases (e.g. Hyper-Threading), we want to avoid L1
160 * evictions by the processes running on the same package. One
161 * thing we can do is to shuffle the initial stack for them.
164 p = arch_align_stack(p);
167 * If this architecture has a platform capability string, copy it
168 * to userspace. In some cases (Sparc), this info is impossible
169 * for userspace to get any other way, in others (i386) it is
174 size_t len = strlen(k_platform) + 1;
176 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
177 if (__copy_to_user(u_platform, k_platform, len))
182 * If this architecture has a "base" platform capability
183 * string, copy it to userspace.
185 u_base_platform = NULL;
186 if (k_base_platform) {
187 size_t len = strlen(k_base_platform) + 1;
189 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
190 if (__copy_to_user(u_base_platform, k_base_platform, len))
195 * Generate 16 random bytes for userspace PRNG seeding.
197 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
198 u_rand_bytes = (elf_addr_t __user *)
199 STACK_ALLOC(p, sizeof(k_rand_bytes));
200 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
203 /* Create the ELF interpreter info */
204 elf_info = (elf_addr_t *)current->mm->saved_auxv;
205 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
206 #define NEW_AUX_ENT(id, val) \
208 elf_info[ei_index++] = id; \
209 elf_info[ei_index++] = val; \
214 * ARCH_DLINFO must come first so PPC can do its special alignment of
216 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
217 * ARCH_DLINFO changes
221 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
222 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
223 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
224 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
225 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
226 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
227 NEW_AUX_ENT(AT_BASE, interp_load_addr);
228 NEW_AUX_ENT(AT_FLAGS, 0);
229 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
230 NEW_AUX_ENT(AT_UID, cred->uid);
231 NEW_AUX_ENT(AT_EUID, cred->euid);
232 NEW_AUX_ENT(AT_GID, cred->gid);
233 NEW_AUX_ENT(AT_EGID, cred->egid);
234 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
235 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
236 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
238 NEW_AUX_ENT(AT_PLATFORM,
239 (elf_addr_t)(unsigned long)u_platform);
241 if (k_base_platform) {
242 NEW_AUX_ENT(AT_BASE_PLATFORM,
243 (elf_addr_t)(unsigned long)u_base_platform);
245 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
246 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
249 /* AT_NULL is zero; clear the rest too */
250 memset(&elf_info[ei_index], 0,
251 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
253 /* And advance past the AT_NULL entry. */
256 sp = STACK_ADD(p, ei_index);
258 items = (argc + 1) + (envc + 1) + 1;
259 bprm->p = STACK_ROUND(sp, items);
261 /* Point sp at the lowest address on the stack */
262 #ifdef CONFIG_STACK_GROWSUP
263 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
264 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
266 sp = (elf_addr_t __user *)bprm->p;
271 * Grow the stack manually; some architectures have a limit on how
272 * far ahead a user-space access may be in order to grow the stack.
274 vma = find_extend_vma(current->mm, bprm->p);
278 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
279 if (__put_user(argc, sp++))
282 envp = argv + argc + 1;
284 /* Populate argv and envp */
285 p = current->mm->arg_end = current->mm->arg_start;
288 if (__put_user((elf_addr_t)p, argv++))
290 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
291 if (!len || len > MAX_ARG_STRLEN)
295 if (__put_user(0, argv))
297 current->mm->arg_end = current->mm->env_start = p;
300 if (__put_user((elf_addr_t)p, envp++))
302 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
303 if (!len || len > MAX_ARG_STRLEN)
307 if (__put_user(0, envp))
309 current->mm->env_end = p;
311 /* Put the elf_info on the stack in the right place. */
312 sp = (elf_addr_t __user *)envp + 1;
313 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
318 static unsigned long elf_map(struct file *filep, unsigned long addr,
319 struct elf_phdr *eppnt, int prot, int type,
320 unsigned long total_size)
322 unsigned long map_addr;
323 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
324 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
325 addr = ELF_PAGESTART(addr);
326 size = ELF_PAGEALIGN(size);
328 /* mmap() will return -EINVAL if given a zero size, but a
329 * segment with zero filesize is perfectly valid */
333 down_write(¤t->mm->mmap_sem);
335 * total_size is the size of the ELF (interpreter) image.
336 * The _first_ mmap needs to know the full size, otherwise
337 * randomization might put this image into an overlapping
338 * position with the ELF binary image. (since size < total_size)
339 * So we first map the 'big' image - and unmap the remainder at
340 * the end. (which unmap is needed for ELF images with holes.)
343 total_size = ELF_PAGEALIGN(total_size);
344 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
345 if (!BAD_ADDR(map_addr))
346 do_munmap(current->mm, map_addr+size, total_size-size);
348 map_addr = do_mmap(filep, addr, size, prot, type, off);
350 up_write(¤t->mm->mmap_sem);
354 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
356 int i, first_idx = -1, last_idx = -1;
358 for (i = 0; i < nr; i++) {
359 if (cmds[i].p_type == PT_LOAD) {
368 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
369 ELF_PAGESTART(cmds[first_idx].p_vaddr);
373 /* This is much more generalized than the library routine read function,
374 so we keep this separate. Technically the library read function
375 is only provided so that we can read a.out libraries that have
378 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
379 struct file *interpreter, unsigned long *interp_map_addr,
380 unsigned long no_base)
382 struct elf_phdr *elf_phdata;
383 struct elf_phdr *eppnt;
384 unsigned long load_addr = 0;
385 int load_addr_set = 0;
386 unsigned long last_bss = 0, elf_bss = 0;
387 unsigned long error = ~0UL;
388 unsigned long total_size;
391 /* First of all, some simple consistency checks */
392 if (interp_elf_ex->e_type != ET_EXEC &&
393 interp_elf_ex->e_type != ET_DYN)
395 if (!elf_check_arch(interp_elf_ex))
397 if (!interpreter->f_op || !interpreter->f_op->mmap)
401 * If the size of this structure has changed, then punt, since
402 * we will be doing the wrong thing.
404 if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
406 if (interp_elf_ex->e_phnum < 1 ||
407 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
410 /* Now read in all of the header information */
411 size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
412 if (size > ELF_MIN_ALIGN)
414 elf_phdata = kmalloc(size, GFP_KERNEL);
418 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
419 (char *)elf_phdata, size);
421 if (retval != size) {
427 total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
434 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
435 if (eppnt->p_type == PT_LOAD) {
436 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
438 unsigned long vaddr = 0;
439 unsigned long k, map_addr;
441 if (eppnt->p_flags & PF_R)
442 elf_prot = PROT_READ;
443 if (eppnt->p_flags & PF_W)
444 elf_prot |= PROT_WRITE;
445 if (eppnt->p_flags & PF_X)
446 elf_prot |= PROT_EXEC;
447 vaddr = eppnt->p_vaddr;
448 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
449 elf_type |= MAP_FIXED;
450 else if (no_base && interp_elf_ex->e_type == ET_DYN)
453 map_addr = elf_map(interpreter, load_addr + vaddr,
454 eppnt, elf_prot, elf_type, total_size);
456 if (!*interp_map_addr)
457 *interp_map_addr = map_addr;
459 if (BAD_ADDR(map_addr))
462 if (!load_addr_set &&
463 interp_elf_ex->e_type == ET_DYN) {
464 load_addr = map_addr - ELF_PAGESTART(vaddr);
469 * Check to see if the section's size will overflow the
470 * allowed task size. Note that p_filesz must always be
471 * <= p_memsize so it's only necessary to check p_memsz.
473 k = load_addr + eppnt->p_vaddr;
475 eppnt->p_filesz > eppnt->p_memsz ||
476 eppnt->p_memsz > TASK_SIZE ||
477 TASK_SIZE - eppnt->p_memsz < k) {
483 * Find the end of the file mapping for this phdr, and
484 * keep track of the largest address we see for this.
486 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
491 * Do the same thing for the memory mapping - between
492 * elf_bss and last_bss is the bss section.
494 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
500 if (last_bss > elf_bss) {
502 * Now fill out the bss section. First pad the last page up
503 * to the page boundary, and then perform a mmap to make sure
504 * that there are zero-mapped pages up to and including the
507 if (padzero(elf_bss)) {
512 /* What we have mapped so far */
513 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
515 /* Map the last of the bss segment */
516 down_write(¤t->mm->mmap_sem);
517 error = do_brk(elf_bss, last_bss - elf_bss);
518 up_write(¤t->mm->mmap_sem);
532 * These are the functions used to load ELF style executables and shared
533 * libraries. There is no binary dependent code anywhere else.
536 #define INTERPRETER_NONE 0
537 #define INTERPRETER_ELF 2
539 #ifndef STACK_RND_MASK
540 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
543 static unsigned long randomize_stack_top(unsigned long stack_top)
545 unsigned int random_variable = 0;
547 if ((current->flags & PF_RANDOMIZE) &&
548 !(current->personality & ADDR_NO_RANDOMIZE)) {
549 random_variable = get_random_int() & STACK_RND_MASK;
550 random_variable <<= PAGE_SHIFT;
552 #ifdef CONFIG_STACK_GROWSUP
553 return PAGE_ALIGN(stack_top) + random_variable;
555 return PAGE_ALIGN(stack_top) - random_variable;
559 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
561 struct file *interpreter = NULL; /* to shut gcc up */
562 unsigned long load_addr = 0, load_bias = 0;
563 int load_addr_set = 0;
564 char * elf_interpreter = NULL;
566 struct elf_phdr *elf_ppnt, *elf_phdata;
567 unsigned long elf_bss, elf_brk;
570 unsigned long elf_entry;
571 unsigned long interp_load_addr = 0;
572 unsigned long start_code, end_code, start_data, end_data;
573 unsigned long reloc_func_desc __maybe_unused = 0;
574 int executable_stack = EXSTACK_DEFAULT;
575 unsigned long def_flags = 0;
577 struct elfhdr elf_ex;
578 struct elfhdr interp_elf_ex;
581 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
587 /* Get the exec-header */
588 loc->elf_ex = *((struct elfhdr *)bprm->buf);
591 /* First of all, some simple consistency checks */
592 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
595 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
597 if (!elf_check_arch(&loc->elf_ex))
599 if (!bprm->file->f_op || !bprm->file->f_op->mmap)
602 /* Now read in all of the header information */
603 if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
605 if (loc->elf_ex.e_phnum < 1 ||
606 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
608 size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
610 elf_phdata = kmalloc(size, GFP_KERNEL);
614 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
615 (char *)elf_phdata, size);
616 if (retval != size) {
622 elf_ppnt = elf_phdata;
631 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
632 if (elf_ppnt->p_type == PT_INTERP) {
633 /* This is the program interpreter used for
634 * shared libraries - for now assume that this
635 * is an a.out format binary
638 if (elf_ppnt->p_filesz > PATH_MAX ||
639 elf_ppnt->p_filesz < 2)
643 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
645 if (!elf_interpreter)
648 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
651 if (retval != elf_ppnt->p_filesz) {
654 goto out_free_interp;
656 /* make sure path is NULL terminated */
658 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
659 goto out_free_interp;
661 interpreter = open_exec(elf_interpreter);
662 retval = PTR_ERR(interpreter);
663 if (IS_ERR(interpreter))
664 goto out_free_interp;
667 * If the binary is not readable then enforce
668 * mm->dumpable = 0 regardless of the interpreter's
671 would_dump(bprm, interpreter);
673 retval = kernel_read(interpreter, 0, bprm->buf,
675 if (retval != BINPRM_BUF_SIZE) {
678 goto out_free_dentry;
681 /* Get the exec headers */
682 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
688 elf_ppnt = elf_phdata;
689 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
690 if (elf_ppnt->p_type == PT_GNU_STACK) {
691 if (elf_ppnt->p_flags & PF_X)
692 executable_stack = EXSTACK_ENABLE_X;
694 executable_stack = EXSTACK_DISABLE_X;
698 /* Some simple consistency checks for the interpreter */
699 if (elf_interpreter) {
701 /* Not an ELF interpreter */
702 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
703 goto out_free_dentry;
704 /* Verify the interpreter has a valid arch */
705 if (!elf_check_arch(&loc->interp_elf_ex))
706 goto out_free_dentry;
709 /* Flush all traces of the currently running executable */
710 retval = flush_old_exec(bprm);
712 goto out_free_dentry;
714 /* OK, This is the point of no return */
715 current->mm->def_flags = def_flags;
717 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
718 may depend on the personality. */
719 SET_PERSONALITY(loc->elf_ex);
720 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
721 current->personality |= READ_IMPLIES_EXEC;
723 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
724 current->flags |= PF_RANDOMIZE;
726 setup_new_exec(bprm);
728 /* Do this so that we can load the interpreter, if need be. We will
729 change some of these later */
730 current->mm->free_area_cache = current->mm->mmap_base;
731 current->mm->cached_hole_size = 0;
732 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
735 send_sig(SIGKILL, current, 0);
736 goto out_free_dentry;
739 current->mm->start_stack = bprm->p;
741 /* Now we do a little grungy work by mmapping the ELF image into
742 the correct location in memory. */
743 for(i = 0, elf_ppnt = elf_phdata;
744 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
745 int elf_prot = 0, elf_flags;
746 unsigned long k, vaddr;
748 if (elf_ppnt->p_type != PT_LOAD)
751 if (unlikely (elf_brk > elf_bss)) {
754 /* There was a PT_LOAD segment with p_memsz > p_filesz
755 before this one. Map anonymous pages, if needed,
756 and clear the area. */
757 retval = set_brk(elf_bss + load_bias,
758 elf_brk + load_bias);
760 send_sig(SIGKILL, current, 0);
761 goto out_free_dentry;
763 nbyte = ELF_PAGEOFFSET(elf_bss);
765 nbyte = ELF_MIN_ALIGN - nbyte;
766 if (nbyte > elf_brk - elf_bss)
767 nbyte = elf_brk - elf_bss;
768 if (clear_user((void __user *)elf_bss +
771 * This bss-zeroing can fail if the ELF
772 * file specifies odd protections. So
773 * we don't check the return value
779 if (elf_ppnt->p_flags & PF_R)
780 elf_prot |= PROT_READ;
781 if (elf_ppnt->p_flags & PF_W)
782 elf_prot |= PROT_WRITE;
783 if (elf_ppnt->p_flags & PF_X)
784 elf_prot |= PROT_EXEC;
786 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
788 vaddr = elf_ppnt->p_vaddr;
789 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
790 elf_flags |= MAP_FIXED;
791 } else if (loc->elf_ex.e_type == ET_DYN) {
792 /* Try and get dynamic programs out of the way of the
793 * default mmap base, as well as whatever program they
794 * might try to exec. This is because the brk will
795 * follow the loader, and is not movable. */
796 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
797 /* Memory randomization might have been switched off
798 * in runtime via sysctl.
799 * If that is the case, retain the original non-zero
800 * load_bias value in order to establish proper
801 * non-randomized mappings.
803 if (current->flags & PF_RANDOMIZE)
806 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
808 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
812 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
813 elf_prot, elf_flags, 0);
814 if (BAD_ADDR(error)) {
815 send_sig(SIGKILL, current, 0);
816 retval = IS_ERR((void *)error) ?
817 PTR_ERR((void*)error) : -EINVAL;
818 goto out_free_dentry;
821 if (!load_addr_set) {
823 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
824 if (loc->elf_ex.e_type == ET_DYN) {
826 ELF_PAGESTART(load_bias + vaddr);
827 load_addr += load_bias;
828 reloc_func_desc = load_bias;
831 k = elf_ppnt->p_vaddr;
838 * Check to see if the section's size will overflow the
839 * allowed task size. Note that p_filesz must always be
840 * <= p_memsz so it is only necessary to check p_memsz.
842 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
843 elf_ppnt->p_memsz > TASK_SIZE ||
844 TASK_SIZE - elf_ppnt->p_memsz < k) {
845 /* set_brk can never work. Avoid overflows. */
846 send_sig(SIGKILL, current, 0);
848 goto out_free_dentry;
851 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
855 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
859 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
864 loc->elf_ex.e_entry += load_bias;
865 elf_bss += load_bias;
866 elf_brk += load_bias;
867 start_code += load_bias;
868 end_code += load_bias;
869 start_data += load_bias;
870 end_data += load_bias;
872 /* Calling set_brk effectively mmaps the pages that we need
873 * for the bss and break sections. We must do this before
874 * mapping in the interpreter, to make sure it doesn't wind
875 * up getting placed where the bss needs to go.
877 retval = set_brk(elf_bss, elf_brk);
879 send_sig(SIGKILL, current, 0);
880 goto out_free_dentry;
882 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
883 send_sig(SIGSEGV, current, 0);
884 retval = -EFAULT; /* Nobody gets to see this, but.. */
885 goto out_free_dentry;
888 if (elf_interpreter) {
889 unsigned long uninitialized_var(interp_map_addr);
891 elf_entry = load_elf_interp(&loc->interp_elf_ex,
895 if (!IS_ERR((void *)elf_entry)) {
897 * load_elf_interp() returns relocation
900 interp_load_addr = elf_entry;
901 elf_entry += loc->interp_elf_ex.e_entry;
903 if (BAD_ADDR(elf_entry)) {
904 force_sig(SIGSEGV, current);
905 retval = IS_ERR((void *)elf_entry) ?
906 (int)elf_entry : -EINVAL;
907 goto out_free_dentry;
909 reloc_func_desc = interp_load_addr;
911 allow_write_access(interpreter);
913 kfree(elf_interpreter);
915 elf_entry = loc->elf_ex.e_entry;
916 if (BAD_ADDR(elf_entry)) {
917 force_sig(SIGSEGV, current);
919 goto out_free_dentry;
925 set_binfmt(&elf_format);
927 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
928 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
930 send_sig(SIGKILL, current, 0);
933 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
935 install_exec_creds(bprm);
936 retval = create_elf_tables(bprm, &loc->elf_ex,
937 load_addr, interp_load_addr);
939 send_sig(SIGKILL, current, 0);
942 /* N.B. passed_fileno might not be initialized? */
943 current->mm->end_code = end_code;
944 current->mm->start_code = start_code;
945 current->mm->start_data = start_data;
946 current->mm->end_data = end_data;
947 current->mm->start_stack = bprm->p;
949 #ifdef arch_randomize_brk
950 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
951 current->mm->brk = current->mm->start_brk =
952 arch_randomize_brk(current->mm);
953 #ifdef CONFIG_COMPAT_BRK
954 current->brk_randomized = 1;
959 if (current->personality & MMAP_PAGE_ZERO) {
960 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
961 and some applications "depend" upon this behavior.
962 Since we do not have the power to recompile these, we
963 emulate the SVr4 behavior. Sigh. */
964 down_write(¤t->mm->mmap_sem);
965 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
966 MAP_FIXED | MAP_PRIVATE, 0);
967 up_write(¤t->mm->mmap_sem);
972 * The ABI may specify that certain registers be set up in special
973 * ways (on i386 %edx is the address of a DT_FINI function, for
974 * example. In addition, it may also specify (eg, PowerPC64 ELF)
975 * that the e_entry field is the address of the function descriptor
976 * for the startup routine, rather than the address of the startup
977 * routine itself. This macro performs whatever initialization to
978 * the regs structure is required as well as any relocations to the
979 * function descriptor entries when executing dynamically links apps.
981 ELF_PLAT_INIT(regs, reloc_func_desc);
984 start_thread(regs, elf_entry, bprm->p);
993 allow_write_access(interpreter);
997 kfree(elf_interpreter);
1003 /* This is really simpleminded and specialized - we are loading an
1004 a.out library that is given an ELF header. */
1005 static int load_elf_library(struct file *file)
1007 struct elf_phdr *elf_phdata;
1008 struct elf_phdr *eppnt;
1009 unsigned long elf_bss, bss, len;
1010 int retval, error, i, j;
1011 struct elfhdr elf_ex;
1014 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1015 if (retval != sizeof(elf_ex))
1018 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1021 /* First of all, some simple consistency checks */
1022 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1023 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1026 /* Now read in all of the header information */
1028 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1029 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1032 elf_phdata = kmalloc(j, GFP_KERNEL);
1038 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1042 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1043 if ((eppnt + i)->p_type == PT_LOAD)
1048 while (eppnt->p_type != PT_LOAD)
1051 /* Now use mmap to map the library into memory. */
1052 down_write(¤t->mm->mmap_sem);
1053 error = do_mmap(file,
1054 ELF_PAGESTART(eppnt->p_vaddr),
1056 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1057 PROT_READ | PROT_WRITE | PROT_EXEC,
1058 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1060 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1061 up_write(¤t->mm->mmap_sem);
1062 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1065 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1066 if (padzero(elf_bss)) {
1071 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1073 bss = eppnt->p_memsz + eppnt->p_vaddr;
1075 down_write(¤t->mm->mmap_sem);
1076 do_brk(len, bss - len);
1077 up_write(¤t->mm->mmap_sem);
1087 #ifdef CONFIG_ELF_CORE
1091 * Modelled on fs/exec.c:aout_core_dump()
1092 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1096 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1097 * that are useful for post-mortem analysis are included in every core dump.
1098 * In that way we ensure that the core dump is fully interpretable later
1099 * without matching up the same kernel and hardware config to see what PC values
1100 * meant. These special mappings include - vDSO, vsyscall, and other
1101 * architecture specific mappings
1103 static bool always_dump_vma(struct vm_area_struct *vma)
1105 /* Any vsyscall mappings? */
1106 if (vma == get_gate_vma(vma->vm_mm))
1109 * arch_vma_name() returns non-NULL for special architecture mappings,
1110 * such as vDSO sections.
1112 if (arch_vma_name(vma))
1119 * Decide what to dump of a segment, part, all or none.
1121 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1122 unsigned long mm_flags)
1124 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1126 /* always dump the vdso and vsyscall sections */
1127 if (always_dump_vma(vma))
1130 if (vma->vm_flags & VM_NODUMP)
1133 /* Hugetlb memory check */
1134 if (vma->vm_flags & VM_HUGETLB) {
1135 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1137 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1141 /* Do not dump I/O mapped devices or special mappings */
1142 if (vma->vm_flags & (VM_IO | VM_RESERVED))
1145 /* By default, dump shared memory if mapped from an anonymous file. */
1146 if (vma->vm_flags & VM_SHARED) {
1147 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1148 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1153 /* Dump segments that have been written to. */
1154 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1156 if (vma->vm_file == NULL)
1159 if (FILTER(MAPPED_PRIVATE))
1163 * If this looks like the beginning of a DSO or executable mapping,
1164 * check for an ELF header. If we find one, dump the first page to
1165 * aid in determining what was mapped here.
1167 if (FILTER(ELF_HEADERS) &&
1168 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1169 u32 __user *header = (u32 __user *) vma->vm_start;
1171 mm_segment_t fs = get_fs();
1173 * Doing it this way gets the constant folded by GCC.
1177 char elfmag[SELFMAG];
1179 BUILD_BUG_ON(SELFMAG != sizeof word);
1180 magic.elfmag[EI_MAG0] = ELFMAG0;
1181 magic.elfmag[EI_MAG1] = ELFMAG1;
1182 magic.elfmag[EI_MAG2] = ELFMAG2;
1183 magic.elfmag[EI_MAG3] = ELFMAG3;
1185 * Switch to the user "segment" for get_user(),
1186 * then put back what elf_core_dump() had in place.
1189 if (unlikely(get_user(word, header)))
1192 if (word == magic.cmp)
1201 return vma->vm_end - vma->vm_start;
1204 /* An ELF note in memory */
1209 unsigned int datasz;
1213 static int notesize(struct memelfnote *en)
1217 sz = sizeof(struct elf_note);
1218 sz += roundup(strlen(en->name) + 1, 4);
1219 sz += roundup(en->datasz, 4);
1224 #define DUMP_WRITE(addr, nr, foffset) \
1225 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1227 static int alignfile(struct file *file, loff_t *foffset)
1229 static const char buf[4] = { 0, };
1230 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1234 static int writenote(struct memelfnote *men, struct file *file,
1238 en.n_namesz = strlen(men->name) + 1;
1239 en.n_descsz = men->datasz;
1240 en.n_type = men->type;
1242 DUMP_WRITE(&en, sizeof(en), foffset);
1243 DUMP_WRITE(men->name, en.n_namesz, foffset);
1244 if (!alignfile(file, foffset))
1246 DUMP_WRITE(men->data, men->datasz, foffset);
1247 if (!alignfile(file, foffset))
1254 static void fill_elf_header(struct elfhdr *elf, int segs,
1255 u16 machine, u32 flags, u8 osabi)
1257 memset(elf, 0, sizeof(*elf));
1259 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1260 elf->e_ident[EI_CLASS] = ELF_CLASS;
1261 elf->e_ident[EI_DATA] = ELF_DATA;
1262 elf->e_ident[EI_VERSION] = EV_CURRENT;
1263 elf->e_ident[EI_OSABI] = ELF_OSABI;
1265 elf->e_type = ET_CORE;
1266 elf->e_machine = machine;
1267 elf->e_version = EV_CURRENT;
1268 elf->e_phoff = sizeof(struct elfhdr);
1269 elf->e_flags = flags;
1270 elf->e_ehsize = sizeof(struct elfhdr);
1271 elf->e_phentsize = sizeof(struct elf_phdr);
1272 elf->e_phnum = segs;
1277 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1279 phdr->p_type = PT_NOTE;
1280 phdr->p_offset = offset;
1283 phdr->p_filesz = sz;
1290 static void fill_note(struct memelfnote *note, const char *name, int type,
1291 unsigned int sz, void *data)
1301 * fill up all the fields in prstatus from the given task struct, except
1302 * registers which need to be filled up separately.
1304 static void fill_prstatus(struct elf_prstatus *prstatus,
1305 struct task_struct *p, long signr)
1307 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1308 prstatus->pr_sigpend = p->pending.signal.sig[0];
1309 prstatus->pr_sighold = p->blocked.sig[0];
1311 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1313 prstatus->pr_pid = task_pid_vnr(p);
1314 prstatus->pr_pgrp = task_pgrp_vnr(p);
1315 prstatus->pr_sid = task_session_vnr(p);
1316 if (thread_group_leader(p)) {
1317 struct task_cputime cputime;
1320 * This is the record for the group leader. It shows the
1321 * group-wide total, not its individual thread total.
1323 thread_group_cputime(p, &cputime);
1324 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1325 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1327 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1328 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1330 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1331 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1334 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1335 struct mm_struct *mm)
1337 const struct cred *cred;
1338 unsigned int i, len;
1340 /* first copy the parameters from user space */
1341 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1343 len = mm->arg_end - mm->arg_start;
1344 if (len >= ELF_PRARGSZ)
1345 len = ELF_PRARGSZ-1;
1346 if (copy_from_user(&psinfo->pr_psargs,
1347 (const char __user *)mm->arg_start, len))
1349 for(i = 0; i < len; i++)
1350 if (psinfo->pr_psargs[i] == 0)
1351 psinfo->pr_psargs[i] = ' ';
1352 psinfo->pr_psargs[len] = 0;
1355 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1357 psinfo->pr_pid = task_pid_vnr(p);
1358 psinfo->pr_pgrp = task_pgrp_vnr(p);
1359 psinfo->pr_sid = task_session_vnr(p);
1361 i = p->state ? ffz(~p->state) + 1 : 0;
1362 psinfo->pr_state = i;
1363 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1364 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1365 psinfo->pr_nice = task_nice(p);
1366 psinfo->pr_flag = p->flags;
1368 cred = __task_cred(p);
1369 SET_UID(psinfo->pr_uid, cred->uid);
1370 SET_GID(psinfo->pr_gid, cred->gid);
1372 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1377 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1379 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1383 while (auxv[i - 2] != AT_NULL);
1384 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1387 #ifdef CORE_DUMP_USE_REGSET
1388 #include <linux/regset.h>
1390 struct elf_thread_core_info {
1391 struct elf_thread_core_info *next;
1392 struct task_struct *task;
1393 struct elf_prstatus prstatus;
1394 struct memelfnote notes[0];
1397 struct elf_note_info {
1398 struct elf_thread_core_info *thread;
1399 struct memelfnote psinfo;
1400 struct memelfnote auxv;
1406 * When a regset has a writeback hook, we call it on each thread before
1407 * dumping user memory. On register window machines, this makes sure the
1408 * user memory backing the register data is up to date before we read it.
1410 static void do_thread_regset_writeback(struct task_struct *task,
1411 const struct user_regset *regset)
1413 if (regset->writeback)
1414 regset->writeback(task, regset, 1);
1417 static int fill_thread_core_info(struct elf_thread_core_info *t,
1418 const struct user_regset_view *view,
1419 long signr, size_t *total)
1424 * NT_PRSTATUS is the one special case, because the regset data
1425 * goes into the pr_reg field inside the note contents, rather
1426 * than being the whole note contents. We fill the reset in here.
1427 * We assume that regset 0 is NT_PRSTATUS.
1429 fill_prstatus(&t->prstatus, t->task, signr);
1430 (void) view->regsets[0].get(t->task, &view->regsets[0],
1431 0, sizeof(t->prstatus.pr_reg),
1432 &t->prstatus.pr_reg, NULL);
1434 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1435 sizeof(t->prstatus), &t->prstatus);
1436 *total += notesize(&t->notes[0]);
1438 do_thread_regset_writeback(t->task, &view->regsets[0]);
1441 * Each other regset might generate a note too. For each regset
1442 * that has no core_note_type or is inactive, we leave t->notes[i]
1443 * all zero and we'll know to skip writing it later.
1445 for (i = 1; i < view->n; ++i) {
1446 const struct user_regset *regset = &view->regsets[i];
1447 do_thread_regset_writeback(t->task, regset);
1448 if (regset->core_note_type && regset->get &&
1449 (!regset->active || regset->active(t->task, regset))) {
1451 size_t size = regset->n * regset->size;
1452 void *data = kmalloc(size, GFP_KERNEL);
1453 if (unlikely(!data))
1455 ret = regset->get(t->task, regset,
1456 0, size, data, NULL);
1460 if (regset->core_note_type != NT_PRFPREG)
1461 fill_note(&t->notes[i], "LINUX",
1462 regset->core_note_type,
1465 t->prstatus.pr_fpvalid = 1;
1466 fill_note(&t->notes[i], "CORE",
1467 NT_PRFPREG, size, data);
1469 *total += notesize(&t->notes[i]);
1477 static int fill_note_info(struct elfhdr *elf, int phdrs,
1478 struct elf_note_info *info,
1479 long signr, struct pt_regs *regs)
1481 struct task_struct *dump_task = current;
1482 const struct user_regset_view *view = task_user_regset_view(dump_task);
1483 struct elf_thread_core_info *t;
1484 struct elf_prpsinfo *psinfo;
1485 struct core_thread *ct;
1489 info->thread = NULL;
1491 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1495 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1498 * Figure out how many notes we're going to need for each thread.
1500 info->thread_notes = 0;
1501 for (i = 0; i < view->n; ++i)
1502 if (view->regsets[i].core_note_type != 0)
1503 ++info->thread_notes;
1506 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1507 * since it is our one special case.
1509 if (unlikely(info->thread_notes == 0) ||
1510 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1516 * Initialize the ELF file header.
1518 fill_elf_header(elf, phdrs,
1519 view->e_machine, view->e_flags, view->ei_osabi);
1522 * Allocate a structure for each thread.
1524 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1525 t = kzalloc(offsetof(struct elf_thread_core_info,
1526 notes[info->thread_notes]),
1532 if (ct->task == dump_task || !info->thread) {
1533 t->next = info->thread;
1537 * Make sure to keep the original task at
1538 * the head of the list.
1540 t->next = info->thread->next;
1541 info->thread->next = t;
1546 * Now fill in each thread's information.
1548 for (t = info->thread; t != NULL; t = t->next)
1549 if (!fill_thread_core_info(t, view, signr, &info->size))
1553 * Fill in the two process-wide notes.
1555 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1556 info->size += notesize(&info->psinfo);
1558 fill_auxv_note(&info->auxv, current->mm);
1559 info->size += notesize(&info->auxv);
1564 static size_t get_note_info_size(struct elf_note_info *info)
1570 * Write all the notes for each thread. When writing the first thread, the
1571 * process-wide notes are interleaved after the first thread-specific note.
1573 static int write_note_info(struct elf_note_info *info,
1574 struct file *file, loff_t *foffset)
1577 struct elf_thread_core_info *t = info->thread;
1582 if (!writenote(&t->notes[0], file, foffset))
1585 if (first && !writenote(&info->psinfo, file, foffset))
1587 if (first && !writenote(&info->auxv, file, foffset))
1590 for (i = 1; i < info->thread_notes; ++i)
1591 if (t->notes[i].data &&
1592 !writenote(&t->notes[i], file, foffset))
1602 static void free_note_info(struct elf_note_info *info)
1604 struct elf_thread_core_info *threads = info->thread;
1607 struct elf_thread_core_info *t = threads;
1609 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1610 for (i = 1; i < info->thread_notes; ++i)
1611 kfree(t->notes[i].data);
1614 kfree(info->psinfo.data);
1619 /* Here is the structure in which status of each thread is captured. */
1620 struct elf_thread_status
1622 struct list_head list;
1623 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1624 elf_fpregset_t fpu; /* NT_PRFPREG */
1625 struct task_struct *thread;
1626 #ifdef ELF_CORE_COPY_XFPREGS
1627 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1629 struct memelfnote notes[3];
1634 * In order to add the specific thread information for the elf file format,
1635 * we need to keep a linked list of every threads pr_status and then create
1636 * a single section for them in the final core file.
1638 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1641 struct task_struct *p = t->thread;
1644 fill_prstatus(&t->prstatus, p, signr);
1645 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1647 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1650 sz += notesize(&t->notes[0]);
1652 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1654 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1657 sz += notesize(&t->notes[1]);
1660 #ifdef ELF_CORE_COPY_XFPREGS
1661 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1662 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1663 sizeof(t->xfpu), &t->xfpu);
1665 sz += notesize(&t->notes[2]);
1671 struct elf_note_info {
1672 struct memelfnote *notes;
1673 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1674 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1675 struct list_head thread_list;
1676 elf_fpregset_t *fpu;
1677 #ifdef ELF_CORE_COPY_XFPREGS
1678 elf_fpxregset_t *xfpu;
1680 int thread_status_size;
1684 static int elf_note_info_init(struct elf_note_info *info)
1686 memset(info, 0, sizeof(*info));
1687 INIT_LIST_HEAD(&info->thread_list);
1689 /* Allocate space for six ELF notes */
1690 info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
1693 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1696 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1697 if (!info->prstatus)
1699 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1702 #ifdef ELF_CORE_COPY_XFPREGS
1703 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1708 #ifdef ELF_CORE_COPY_XFPREGS
1713 kfree(info->prstatus);
1715 kfree(info->psinfo);
1721 static int fill_note_info(struct elfhdr *elf, int phdrs,
1722 struct elf_note_info *info,
1723 long signr, struct pt_regs *regs)
1725 struct list_head *t;
1727 if (!elf_note_info_init(info))
1731 struct core_thread *ct;
1732 struct elf_thread_status *ets;
1734 for (ct = current->mm->core_state->dumper.next;
1735 ct; ct = ct->next) {
1736 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1740 ets->thread = ct->task;
1741 list_add(&ets->list, &info->thread_list);
1744 list_for_each(t, &info->thread_list) {
1747 ets = list_entry(t, struct elf_thread_status, list);
1748 sz = elf_dump_thread_status(signr, ets);
1749 info->thread_status_size += sz;
1752 /* now collect the dump for the current */
1753 memset(info->prstatus, 0, sizeof(*info->prstatus));
1754 fill_prstatus(info->prstatus, current, signr);
1755 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1758 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1761 * Set up the notes in similar form to SVR4 core dumps made
1762 * with info from their /proc.
1765 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1766 sizeof(*info->prstatus), info->prstatus);
1767 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1768 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1769 sizeof(*info->psinfo), info->psinfo);
1773 fill_auxv_note(&info->notes[info->numnote++], current->mm);
1775 /* Try to dump the FPU. */
1776 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1778 if (info->prstatus->pr_fpvalid)
1779 fill_note(info->notes + info->numnote++,
1780 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1781 #ifdef ELF_CORE_COPY_XFPREGS
1782 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1783 fill_note(info->notes + info->numnote++,
1784 "LINUX", ELF_CORE_XFPREG_TYPE,
1785 sizeof(*info->xfpu), info->xfpu);
1791 static size_t get_note_info_size(struct elf_note_info *info)
1796 for (i = 0; i < info->numnote; i++)
1797 sz += notesize(info->notes + i);
1799 sz += info->thread_status_size;
1804 static int write_note_info(struct elf_note_info *info,
1805 struct file *file, loff_t *foffset)
1808 struct list_head *t;
1810 for (i = 0; i < info->numnote; i++)
1811 if (!writenote(info->notes + i, file, foffset))
1814 /* write out the thread status notes section */
1815 list_for_each(t, &info->thread_list) {
1816 struct elf_thread_status *tmp =
1817 list_entry(t, struct elf_thread_status, list);
1819 for (i = 0; i < tmp->num_notes; i++)
1820 if (!writenote(&tmp->notes[i], file, foffset))
1827 static void free_note_info(struct elf_note_info *info)
1829 while (!list_empty(&info->thread_list)) {
1830 struct list_head *tmp = info->thread_list.next;
1832 kfree(list_entry(tmp, struct elf_thread_status, list));
1835 kfree(info->prstatus);
1836 kfree(info->psinfo);
1839 #ifdef ELF_CORE_COPY_XFPREGS
1846 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1847 struct vm_area_struct *gate_vma)
1849 struct vm_area_struct *ret = tsk->mm->mmap;
1856 * Helper function for iterating across a vma list. It ensures that the caller
1857 * will visit `gate_vma' prior to terminating the search.
1859 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1860 struct vm_area_struct *gate_vma)
1862 struct vm_area_struct *ret;
1864 ret = this_vma->vm_next;
1867 if (this_vma == gate_vma)
1872 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1873 elf_addr_t e_shoff, int segs)
1875 elf->e_shoff = e_shoff;
1876 elf->e_shentsize = sizeof(*shdr4extnum);
1878 elf->e_shstrndx = SHN_UNDEF;
1880 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1882 shdr4extnum->sh_type = SHT_NULL;
1883 shdr4extnum->sh_size = elf->e_shnum;
1884 shdr4extnum->sh_link = elf->e_shstrndx;
1885 shdr4extnum->sh_info = segs;
1888 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1889 unsigned long mm_flags)
1891 struct vm_area_struct *vma;
1894 for (vma = first_vma(current, gate_vma); vma != NULL;
1895 vma = next_vma(vma, gate_vma))
1896 size += vma_dump_size(vma, mm_flags);
1903 * This is a two-pass process; first we find the offsets of the bits,
1904 * and then they are actually written out. If we run out of core limit
1907 static int elf_core_dump(struct coredump_params *cprm)
1913 struct vm_area_struct *vma, *gate_vma;
1914 struct elfhdr *elf = NULL;
1915 loff_t offset = 0, dataoff, foffset;
1916 struct elf_note_info info;
1917 struct elf_phdr *phdr4note = NULL;
1918 struct elf_shdr *shdr4extnum = NULL;
1923 * We no longer stop all VM operations.
1925 * This is because those proceses that could possibly change map_count
1926 * or the mmap / vma pages are now blocked in do_exit on current
1927 * finishing this core dump.
1929 * Only ptrace can touch these memory addresses, but it doesn't change
1930 * the map_count or the pages allocated. So no possibility of crashing
1931 * exists while dumping the mm->vm_next areas to the core file.
1934 /* alloc memory for large data structures: too large to be on stack */
1935 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1939 * The number of segs are recored into ELF header as 16bit value.
1940 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1942 segs = current->mm->map_count;
1943 segs += elf_core_extra_phdrs();
1945 gate_vma = get_gate_vma(current->mm);
1946 if (gate_vma != NULL)
1949 /* for notes section */
1952 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1953 * this, kernel supports extended numbering. Have a look at
1954 * include/linux/elf.h for further information. */
1955 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1958 * Collect all the non-memory information about the process for the
1959 * notes. This also sets up the file header.
1961 if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
1965 current->flags |= PF_DUMPCORE;
1970 offset += sizeof(*elf); /* Elf header */
1971 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1974 /* Write notes phdr entry */
1976 size_t sz = get_note_info_size(&info);
1978 sz += elf_coredump_extra_notes_size();
1980 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1984 fill_elf_note_phdr(phdr4note, sz, offset);
1988 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1990 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
1991 offset += elf_core_extra_data_size();
1994 if (e_phnum == PN_XNUM) {
1995 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1998 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2003 size += sizeof(*elf);
2004 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2007 size += sizeof(*phdr4note);
2008 if (size > cprm->limit
2009 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2012 /* Write program headers for segments dump */
2013 for (vma = first_vma(current, gate_vma); vma != NULL;
2014 vma = next_vma(vma, gate_vma)) {
2015 struct elf_phdr phdr;
2017 phdr.p_type = PT_LOAD;
2018 phdr.p_offset = offset;
2019 phdr.p_vaddr = vma->vm_start;
2021 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2022 phdr.p_memsz = vma->vm_end - vma->vm_start;
2023 offset += phdr.p_filesz;
2024 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2025 if (vma->vm_flags & VM_WRITE)
2026 phdr.p_flags |= PF_W;
2027 if (vma->vm_flags & VM_EXEC)
2028 phdr.p_flags |= PF_X;
2029 phdr.p_align = ELF_EXEC_PAGESIZE;
2031 size += sizeof(phdr);
2032 if (size > cprm->limit
2033 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2037 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2040 /* write out the notes section */
2041 if (!write_note_info(&info, cprm->file, &foffset))
2044 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2048 if (!dump_seek(cprm->file, dataoff - foffset))
2051 for (vma = first_vma(current, gate_vma); vma != NULL;
2052 vma = next_vma(vma, gate_vma)) {
2056 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2058 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2062 page = get_dump_page(addr);
2064 void *kaddr = kmap(page);
2065 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2066 !dump_write(cprm->file, kaddr,
2069 page_cache_release(page);
2071 stop = !dump_seek(cprm->file, PAGE_SIZE);
2077 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2080 if (e_phnum == PN_XNUM) {
2081 size += sizeof(*shdr4extnum);
2082 if (size > cprm->limit
2083 || !dump_write(cprm->file, shdr4extnum,
2084 sizeof(*shdr4extnum)))
2092 free_note_info(&info);
2100 #endif /* CONFIG_ELF_CORE */
2102 static int __init init_elf_binfmt(void)
2104 register_binfmt(&elf_format);
2108 static void __exit exit_elf_binfmt(void)
2110 /* Remove the COFF and ELF loaders. */
2111 unregister_binfmt(&elf_format);
2114 core_initcall(init_elf_binfmt);
2115 module_exit(exit_elf_binfmt);
2116 MODULE_LICENSE("GPL");