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 = 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 if (file_permission(interpreter, MAY_READ) < 0)
672 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
674 retval = kernel_read(interpreter, 0, bprm->buf,
676 if (retval != BINPRM_BUF_SIZE) {
679 goto out_free_dentry;
682 /* Get the exec headers */
683 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
689 elf_ppnt = elf_phdata;
690 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
691 if (elf_ppnt->p_type == PT_GNU_STACK) {
692 if (elf_ppnt->p_flags & PF_X)
693 executable_stack = EXSTACK_ENABLE_X;
695 executable_stack = EXSTACK_DISABLE_X;
699 /* Some simple consistency checks for the interpreter */
700 if (elf_interpreter) {
702 /* Not an ELF interpreter */
703 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
704 goto out_free_dentry;
705 /* Verify the interpreter has a valid arch */
706 if (!elf_check_arch(&loc->interp_elf_ex))
707 goto out_free_dentry;
710 /* Flush all traces of the currently running executable */
711 retval = flush_old_exec(bprm);
713 goto out_free_dentry;
717 * Turn off the CS limit completely if exec-shield disabled or
720 if (!exec_shield || executable_stack != EXSTACK_DISABLE_X || (__supported_pte_mask & _PAGE_NX))
721 arch_add_exec_range(current->mm, -1);
724 /* OK, This is the point of no return */
725 current->flags &= ~PF_FORKNOEXEC;
726 current->mm->def_flags = def_flags;
728 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
729 may depend on the personality. */
730 SET_PERSONALITY(loc->elf_ex);
731 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
732 current->personality |= READ_IMPLIES_EXEC;
734 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
735 current->flags |= PF_RANDOMIZE;
737 setup_new_exec(bprm);
739 /* Do this so that we can load the interpreter, if need be. We will
740 change some of these later */
741 current->mm->free_area_cache = current->mm->mmap_base;
742 current->mm->cached_hole_size = 0;
743 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
746 send_sig(SIGKILL, current, 0);
747 goto out_free_dentry;
750 current->mm->start_stack = bprm->p;
752 /* Now we do a little grungy work by mmapping the ELF image into
753 the correct location in memory. */
754 for(i = 0, elf_ppnt = elf_phdata;
755 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
756 int elf_prot = 0, elf_flags;
757 unsigned long k, vaddr;
759 if (elf_ppnt->p_type != PT_LOAD)
762 if (unlikely (elf_brk > elf_bss)) {
765 /* There was a PT_LOAD segment with p_memsz > p_filesz
766 before this one. Map anonymous pages, if needed,
767 and clear the area. */
768 retval = set_brk(elf_bss + load_bias,
769 elf_brk + load_bias);
771 send_sig(SIGKILL, current, 0);
772 goto out_free_dentry;
774 nbyte = ELF_PAGEOFFSET(elf_bss);
776 nbyte = ELF_MIN_ALIGN - nbyte;
777 if (nbyte > elf_brk - elf_bss)
778 nbyte = elf_brk - elf_bss;
779 if (clear_user((void __user *)elf_bss +
782 * This bss-zeroing can fail if the ELF
783 * file specifies odd protections. So
784 * we don't check the return value
790 if (elf_ppnt->p_flags & PF_R)
791 elf_prot |= PROT_READ;
792 if (elf_ppnt->p_flags & PF_W)
793 elf_prot |= PROT_WRITE;
794 if (elf_ppnt->p_flags & PF_X)
795 elf_prot |= PROT_EXEC;
797 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
799 vaddr = elf_ppnt->p_vaddr;
800 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
801 elf_flags |= MAP_FIXED;
802 } else if (loc->elf_ex.e_type == ET_DYN) {
803 /* Try and get dynamic programs out of the way of the
804 * default mmap base, as well as whatever program they
805 * might try to exec. This is because the brk will
806 * follow the loader, and is not movable. */
807 #if defined(CONFIG_X86) || defined(CONFIG_ARM)
810 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
814 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
815 elf_prot, elf_flags, 0);
816 if (BAD_ADDR(error)) {
817 send_sig(SIGKILL, current, 0);
818 retval = IS_ERR((void *)error) ?
819 PTR_ERR((void*)error) : -EINVAL;
820 goto out_free_dentry;
823 if (!load_addr_set) {
825 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
826 if (loc->elf_ex.e_type == ET_DYN) {
828 ELF_PAGESTART(load_bias + vaddr);
829 load_addr += load_bias;
830 reloc_func_desc = load_bias;
833 k = elf_ppnt->p_vaddr;
840 * Check to see if the section's size will overflow the
841 * allowed task size. Note that p_filesz must always be
842 * <= p_memsz so it is only necessary to check p_memsz.
844 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
845 elf_ppnt->p_memsz > TASK_SIZE ||
846 TASK_SIZE - elf_ppnt->p_memsz < k) {
847 /* set_brk can never work. Avoid overflows. */
848 send_sig(SIGKILL, current, 0);
850 goto out_free_dentry;
853 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
857 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
861 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
866 loc->elf_ex.e_entry += load_bias;
867 elf_bss += load_bias;
868 elf_brk += load_bias;
869 start_code += load_bias;
870 end_code += load_bias;
871 start_data += load_bias;
872 end_data += load_bias;
874 /* Calling set_brk effectively mmaps the pages that we need
875 * for the bss and break sections. We must do this before
876 * mapping in the interpreter, to make sure it doesn't wind
877 * up getting placed where the bss needs to go.
879 retval = set_brk(elf_bss, elf_brk);
881 send_sig(SIGKILL, current, 0);
882 goto out_free_dentry;
884 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
885 send_sig(SIGSEGV, current, 0);
886 retval = -EFAULT; /* Nobody gets to see this, but.. */
887 goto out_free_dentry;
890 if (elf_interpreter) {
891 unsigned long uninitialized_var(interp_map_addr);
893 elf_entry = load_elf_interp(&loc->interp_elf_ex,
897 if (!IS_ERR((void *)elf_entry)) {
899 * load_elf_interp() returns relocation
902 interp_load_addr = elf_entry;
903 elf_entry += loc->interp_elf_ex.e_entry;
905 if (BAD_ADDR(elf_entry)) {
906 force_sig(SIGSEGV, current);
907 retval = IS_ERR((void *)elf_entry) ?
908 (int)elf_entry : -EINVAL;
909 goto out_free_dentry;
911 reloc_func_desc = interp_load_addr;
913 allow_write_access(interpreter);
915 kfree(elf_interpreter);
917 elf_entry = loc->elf_ex.e_entry;
918 if (BAD_ADDR(elf_entry)) {
919 force_sig(SIGSEGV, current);
921 goto out_free_dentry;
927 set_binfmt(&elf_format);
929 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
930 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
932 send_sig(SIGKILL, current, 0);
935 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
937 install_exec_creds(bprm);
938 current->flags &= ~PF_FORKNOEXEC;
939 retval = create_elf_tables(bprm, &loc->elf_ex,
940 load_addr, interp_load_addr);
942 send_sig(SIGKILL, current, 0);
945 /* N.B. passed_fileno might not be initialized? */
946 current->mm->end_code = end_code;
947 current->mm->start_code = start_code;
948 current->mm->start_data = start_data;
949 current->mm->end_data = end_data;
950 current->mm->start_stack = bprm->p;
952 #ifdef arch_randomize_brk
953 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
954 current->mm->brk = current->mm->start_brk =
955 arch_randomize_brk(current->mm);
958 if (current->personality & MMAP_PAGE_ZERO) {
959 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
960 and some applications "depend" upon this behavior.
961 Since we do not have the power to recompile these, we
962 emulate the SVr4 behavior. Sigh. */
963 down_write(¤t->mm->mmap_sem);
964 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
965 MAP_FIXED | MAP_PRIVATE, 0);
966 up_write(¤t->mm->mmap_sem);
971 * The ABI may specify that certain registers be set up in special
972 * ways (on i386 %edx is the address of a DT_FINI function, for
973 * example. In addition, it may also specify (eg, PowerPC64 ELF)
974 * that the e_entry field is the address of the function descriptor
975 * for the startup routine, rather than the address of the startup
976 * routine itself. This macro performs whatever initialization to
977 * the regs structure is required as well as any relocations to the
978 * function descriptor entries when executing dynamically links apps.
980 ELF_PLAT_INIT(regs, reloc_func_desc);
983 start_thread(regs, elf_entry, bprm->p);
992 allow_write_access(interpreter);
996 kfree(elf_interpreter);
1002 /* This is really simpleminded and specialized - we are loading an
1003 a.out library that is given an ELF header. */
1004 static int load_elf_library(struct file *file)
1006 struct elf_phdr *elf_phdata;
1007 struct elf_phdr *eppnt;
1008 unsigned long elf_bss, bss, len;
1009 int retval, error, i, j;
1010 struct elfhdr elf_ex;
1013 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1014 if (retval != sizeof(elf_ex))
1017 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1020 /* First of all, some simple consistency checks */
1021 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1022 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1025 /* Now read in all of the header information */
1027 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1028 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1031 elf_phdata = kmalloc(j, GFP_KERNEL);
1037 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1041 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1042 if ((eppnt + i)->p_type == PT_LOAD)
1047 while (eppnt->p_type != PT_LOAD)
1050 /* Now use mmap to map the library into memory. */
1051 down_write(¤t->mm->mmap_sem);
1052 error = do_mmap(file,
1053 ELF_PAGESTART(eppnt->p_vaddr),
1055 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1056 PROT_READ | PROT_WRITE | PROT_EXEC,
1057 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1059 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1060 up_write(¤t->mm->mmap_sem);
1061 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1064 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1065 if (padzero(elf_bss)) {
1070 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1072 bss = eppnt->p_memsz + eppnt->p_vaddr;
1074 down_write(¤t->mm->mmap_sem);
1075 do_brk(len, bss - len);
1076 up_write(¤t->mm->mmap_sem);
1086 #ifdef CONFIG_ELF_CORE
1090 * Modelled on fs/exec.c:aout_core_dump()
1091 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1095 * Decide what to dump of a segment, part, all or none.
1097 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1098 unsigned long mm_flags)
1100 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1102 /* The vma can be set up to tell us the answer directly. */
1103 if (vma->vm_flags & VM_ALWAYSDUMP)
1106 /* Hugetlb memory check */
1107 if (vma->vm_flags & VM_HUGETLB) {
1108 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1110 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1114 /* Do not dump I/O mapped devices or special mappings */
1115 if (vma->vm_flags & (VM_IO | VM_RESERVED))
1118 /* By default, dump shared memory if mapped from an anonymous file. */
1119 if (vma->vm_flags & VM_SHARED) {
1120 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1121 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1126 /* Dump segments that have been written to. */
1127 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1129 if (vma->vm_file == NULL)
1132 if (FILTER(MAPPED_PRIVATE))
1136 * If this looks like the beginning of a DSO or executable mapping,
1137 * check for an ELF header. If we find one, dump the first page to
1138 * aid in determining what was mapped here.
1140 if (FILTER(ELF_HEADERS) &&
1141 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1142 u32 __user *header = (u32 __user *) vma->vm_start;
1144 mm_segment_t fs = get_fs();
1146 * Doing it this way gets the constant folded by GCC.
1150 char elfmag[SELFMAG];
1152 BUILD_BUG_ON(SELFMAG != sizeof word);
1153 magic.elfmag[EI_MAG0] = ELFMAG0;
1154 magic.elfmag[EI_MAG1] = ELFMAG1;
1155 magic.elfmag[EI_MAG2] = ELFMAG2;
1156 magic.elfmag[EI_MAG3] = ELFMAG3;
1158 * Switch to the user "segment" for get_user(),
1159 * then put back what elf_core_dump() had in place.
1162 if (unlikely(get_user(word, header)))
1165 if (word == magic.cmp)
1174 return vma->vm_end - vma->vm_start;
1177 /* An ELF note in memory */
1182 unsigned int datasz;
1186 static int notesize(struct memelfnote *en)
1190 sz = sizeof(struct elf_note);
1191 sz += roundup(strlen(en->name) + 1, 4);
1192 sz += roundup(en->datasz, 4);
1197 #define DUMP_WRITE(addr, nr, foffset) \
1198 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1200 static int alignfile(struct file *file, loff_t *foffset)
1202 static const char buf[4] = { 0, };
1203 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1207 static int writenote(struct memelfnote *men, struct file *file,
1211 en.n_namesz = strlen(men->name) + 1;
1212 en.n_descsz = men->datasz;
1213 en.n_type = men->type;
1215 DUMP_WRITE(&en, sizeof(en), foffset);
1216 DUMP_WRITE(men->name, en.n_namesz, foffset);
1217 if (!alignfile(file, foffset))
1219 DUMP_WRITE(men->data, men->datasz, foffset);
1220 if (!alignfile(file, foffset))
1227 static void fill_elf_header(struct elfhdr *elf, int segs,
1228 u16 machine, u32 flags, u8 osabi)
1230 memset(elf, 0, sizeof(*elf));
1232 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1233 elf->e_ident[EI_CLASS] = ELF_CLASS;
1234 elf->e_ident[EI_DATA] = ELF_DATA;
1235 elf->e_ident[EI_VERSION] = EV_CURRENT;
1236 elf->e_ident[EI_OSABI] = ELF_OSABI;
1238 elf->e_type = ET_CORE;
1239 elf->e_machine = machine;
1240 elf->e_version = EV_CURRENT;
1241 elf->e_phoff = sizeof(struct elfhdr);
1242 elf->e_flags = flags;
1243 elf->e_ehsize = sizeof(struct elfhdr);
1244 elf->e_phentsize = sizeof(struct elf_phdr);
1245 elf->e_phnum = segs;
1250 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1252 phdr->p_type = PT_NOTE;
1253 phdr->p_offset = offset;
1256 phdr->p_filesz = sz;
1263 static void fill_note(struct memelfnote *note, const char *name, int type,
1264 unsigned int sz, void *data)
1274 * fill up all the fields in prstatus from the given task struct, except
1275 * registers which need to be filled up separately.
1277 static void fill_prstatus(struct elf_prstatus *prstatus,
1278 struct task_struct *p, long signr)
1280 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1281 prstatus->pr_sigpend = p->pending.signal.sig[0];
1282 prstatus->pr_sighold = p->blocked.sig[0];
1284 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1286 prstatus->pr_pid = task_pid_vnr(p);
1287 prstatus->pr_pgrp = task_pgrp_vnr(p);
1288 prstatus->pr_sid = task_session_vnr(p);
1289 if (thread_group_leader(p)) {
1290 struct task_cputime cputime;
1293 * This is the record for the group leader. It shows the
1294 * group-wide total, not its individual thread total.
1296 thread_group_cputime(p, &cputime);
1297 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1298 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1300 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1301 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1303 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1304 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1307 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1308 struct mm_struct *mm)
1310 const struct cred *cred;
1311 unsigned int i, len;
1313 /* first copy the parameters from user space */
1314 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1316 len = mm->arg_end - mm->arg_start;
1317 if (len >= ELF_PRARGSZ)
1318 len = ELF_PRARGSZ-1;
1319 if (copy_from_user(&psinfo->pr_psargs,
1320 (const char __user *)mm->arg_start, len))
1322 for(i = 0; i < len; i++)
1323 if (psinfo->pr_psargs[i] == 0)
1324 psinfo->pr_psargs[i] = ' ';
1325 psinfo->pr_psargs[len] = 0;
1328 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1330 psinfo->pr_pid = task_pid_vnr(p);
1331 psinfo->pr_pgrp = task_pgrp_vnr(p);
1332 psinfo->pr_sid = task_session_vnr(p);
1334 i = p->state ? ffz(~p->state) + 1 : 0;
1335 psinfo->pr_state = i;
1336 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1337 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1338 psinfo->pr_nice = task_nice(p);
1339 psinfo->pr_flag = p->flags;
1341 cred = __task_cred(p);
1342 SET_UID(psinfo->pr_uid, cred->uid);
1343 SET_GID(psinfo->pr_gid, cred->gid);
1345 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1350 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1352 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1356 while (auxv[i - 2] != AT_NULL);
1357 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1360 #ifdef CORE_DUMP_USE_REGSET
1361 #include <linux/regset.h>
1363 struct elf_thread_core_info {
1364 struct elf_thread_core_info *next;
1365 struct task_struct *task;
1366 struct elf_prstatus prstatus;
1367 struct memelfnote notes[0];
1370 struct elf_note_info {
1371 struct elf_thread_core_info *thread;
1372 struct memelfnote psinfo;
1373 struct memelfnote auxv;
1379 * When a regset has a writeback hook, we call it on each thread before
1380 * dumping user memory. On register window machines, this makes sure the
1381 * user memory backing the register data is up to date before we read it.
1383 static void do_thread_regset_writeback(struct task_struct *task,
1384 const struct user_regset *regset)
1386 if (regset->writeback)
1387 regset->writeback(task, regset, 1);
1390 static int fill_thread_core_info(struct elf_thread_core_info *t,
1391 const struct user_regset_view *view,
1392 long signr, size_t *total)
1397 * NT_PRSTATUS is the one special case, because the regset data
1398 * goes into the pr_reg field inside the note contents, rather
1399 * than being the whole note contents. We fill the reset in here.
1400 * We assume that regset 0 is NT_PRSTATUS.
1402 fill_prstatus(&t->prstatus, t->task, signr);
1403 (void) view->regsets[0].get(t->task, &view->regsets[0],
1404 0, sizeof(t->prstatus.pr_reg),
1405 &t->prstatus.pr_reg, NULL);
1407 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1408 sizeof(t->prstatus), &t->prstatus);
1409 *total += notesize(&t->notes[0]);
1411 do_thread_regset_writeback(t->task, &view->regsets[0]);
1414 * Each other regset might generate a note too. For each regset
1415 * that has no core_note_type or is inactive, we leave t->notes[i]
1416 * all zero and we'll know to skip writing it later.
1418 for (i = 1; i < view->n; ++i) {
1419 const struct user_regset *regset = &view->regsets[i];
1420 do_thread_regset_writeback(t->task, regset);
1421 if (regset->core_note_type &&
1422 (!regset->active || regset->active(t->task, regset))) {
1424 size_t size = regset->n * regset->size;
1425 void *data = kmalloc(size, GFP_KERNEL);
1426 if (unlikely(!data))
1428 ret = regset->get(t->task, regset,
1429 0, size, data, NULL);
1433 if (regset->core_note_type != NT_PRFPREG)
1434 fill_note(&t->notes[i], "LINUX",
1435 regset->core_note_type,
1438 t->prstatus.pr_fpvalid = 1;
1439 fill_note(&t->notes[i], "CORE",
1440 NT_PRFPREG, size, data);
1442 *total += notesize(&t->notes[i]);
1450 static int fill_note_info(struct elfhdr *elf, int phdrs,
1451 struct elf_note_info *info,
1452 long signr, struct pt_regs *regs)
1454 struct task_struct *dump_task = current;
1455 const struct user_regset_view *view = task_user_regset_view(dump_task);
1456 struct elf_thread_core_info *t;
1457 struct elf_prpsinfo *psinfo;
1458 struct core_thread *ct;
1462 info->thread = NULL;
1464 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1468 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1471 * Figure out how many notes we're going to need for each thread.
1473 info->thread_notes = 0;
1474 for (i = 0; i < view->n; ++i)
1475 if (view->regsets[i].core_note_type != 0)
1476 ++info->thread_notes;
1479 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1480 * since it is our one special case.
1482 if (unlikely(info->thread_notes == 0) ||
1483 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1489 * Initialize the ELF file header.
1491 fill_elf_header(elf, phdrs,
1492 view->e_machine, view->e_flags, view->ei_osabi);
1495 * Allocate a structure for each thread.
1497 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1498 t = kzalloc(offsetof(struct elf_thread_core_info,
1499 notes[info->thread_notes]),
1505 if (ct->task == dump_task || !info->thread) {
1506 t->next = info->thread;
1510 * Make sure to keep the original task at
1511 * the head of the list.
1513 t->next = info->thread->next;
1514 info->thread->next = t;
1519 * Now fill in each thread's information.
1521 for (t = info->thread; t != NULL; t = t->next)
1522 if (!fill_thread_core_info(t, view, signr, &info->size))
1526 * Fill in the two process-wide notes.
1528 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1529 info->size += notesize(&info->psinfo);
1531 fill_auxv_note(&info->auxv, current->mm);
1532 info->size += notesize(&info->auxv);
1537 static size_t get_note_info_size(struct elf_note_info *info)
1543 * Write all the notes for each thread. When writing the first thread, the
1544 * process-wide notes are interleaved after the first thread-specific note.
1546 static int write_note_info(struct elf_note_info *info,
1547 struct file *file, loff_t *foffset)
1550 struct elf_thread_core_info *t = info->thread;
1555 if (!writenote(&t->notes[0], file, foffset))
1558 if (first && !writenote(&info->psinfo, file, foffset))
1560 if (first && !writenote(&info->auxv, file, foffset))
1563 for (i = 1; i < info->thread_notes; ++i)
1564 if (t->notes[i].data &&
1565 !writenote(&t->notes[i], file, foffset))
1575 static void free_note_info(struct elf_note_info *info)
1577 struct elf_thread_core_info *threads = info->thread;
1580 struct elf_thread_core_info *t = threads;
1582 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1583 for (i = 1; i < info->thread_notes; ++i)
1584 kfree(t->notes[i].data);
1587 kfree(info->psinfo.data);
1592 /* Here is the structure in which status of each thread is captured. */
1593 struct elf_thread_status
1595 struct list_head list;
1596 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1597 elf_fpregset_t fpu; /* NT_PRFPREG */
1598 struct task_struct *thread;
1599 #ifdef ELF_CORE_COPY_XFPREGS
1600 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1602 struct memelfnote notes[3];
1607 * In order to add the specific thread information for the elf file format,
1608 * we need to keep a linked list of every threads pr_status and then create
1609 * a single section for them in the final core file.
1611 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1614 struct task_struct *p = t->thread;
1617 fill_prstatus(&t->prstatus, p, signr);
1618 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1620 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1623 sz += notesize(&t->notes[0]);
1625 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1627 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1630 sz += notesize(&t->notes[1]);
1633 #ifdef ELF_CORE_COPY_XFPREGS
1634 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1635 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1636 sizeof(t->xfpu), &t->xfpu);
1638 sz += notesize(&t->notes[2]);
1644 struct elf_note_info {
1645 struct memelfnote *notes;
1646 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1647 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1648 struct list_head thread_list;
1649 elf_fpregset_t *fpu;
1650 #ifdef ELF_CORE_COPY_XFPREGS
1651 elf_fpxregset_t *xfpu;
1653 int thread_status_size;
1657 static int elf_note_info_init(struct elf_note_info *info)
1659 memset(info, 0, sizeof(*info));
1660 INIT_LIST_HEAD(&info->thread_list);
1662 /* Allocate space for six ELF notes */
1663 info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
1666 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1669 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1670 if (!info->prstatus)
1672 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1675 #ifdef ELF_CORE_COPY_XFPREGS
1676 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1681 #ifdef ELF_CORE_COPY_XFPREGS
1686 kfree(info->prstatus);
1688 kfree(info->psinfo);
1694 static int fill_note_info(struct elfhdr *elf, int phdrs,
1695 struct elf_note_info *info,
1696 long signr, struct pt_regs *regs)
1698 struct list_head *t;
1700 if (!elf_note_info_init(info))
1704 struct core_thread *ct;
1705 struct elf_thread_status *ets;
1707 for (ct = current->mm->core_state->dumper.next;
1708 ct; ct = ct->next) {
1709 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1713 ets->thread = ct->task;
1714 list_add(&ets->list, &info->thread_list);
1717 list_for_each(t, &info->thread_list) {
1720 ets = list_entry(t, struct elf_thread_status, list);
1721 sz = elf_dump_thread_status(signr, ets);
1722 info->thread_status_size += sz;
1725 /* now collect the dump for the current */
1726 memset(info->prstatus, 0, sizeof(*info->prstatus));
1727 fill_prstatus(info->prstatus, current, signr);
1728 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1731 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1734 * Set up the notes in similar form to SVR4 core dumps made
1735 * with info from their /proc.
1738 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1739 sizeof(*info->prstatus), info->prstatus);
1740 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1741 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1742 sizeof(*info->psinfo), info->psinfo);
1746 fill_auxv_note(&info->notes[info->numnote++], current->mm);
1748 /* Try to dump the FPU. */
1749 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1751 if (info->prstatus->pr_fpvalid)
1752 fill_note(info->notes + info->numnote++,
1753 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1754 #ifdef ELF_CORE_COPY_XFPREGS
1755 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1756 fill_note(info->notes + info->numnote++,
1757 "LINUX", ELF_CORE_XFPREG_TYPE,
1758 sizeof(*info->xfpu), info->xfpu);
1764 static size_t get_note_info_size(struct elf_note_info *info)
1769 for (i = 0; i < info->numnote; i++)
1770 sz += notesize(info->notes + i);
1772 sz += info->thread_status_size;
1777 static int write_note_info(struct elf_note_info *info,
1778 struct file *file, loff_t *foffset)
1781 struct list_head *t;
1783 for (i = 0; i < info->numnote; i++)
1784 if (!writenote(info->notes + i, file, foffset))
1787 /* write out the thread status notes section */
1788 list_for_each(t, &info->thread_list) {
1789 struct elf_thread_status *tmp =
1790 list_entry(t, struct elf_thread_status, list);
1792 for (i = 0; i < tmp->num_notes; i++)
1793 if (!writenote(&tmp->notes[i], file, foffset))
1800 static void free_note_info(struct elf_note_info *info)
1802 while (!list_empty(&info->thread_list)) {
1803 struct list_head *tmp = info->thread_list.next;
1805 kfree(list_entry(tmp, struct elf_thread_status, list));
1808 kfree(info->prstatus);
1809 kfree(info->psinfo);
1812 #ifdef ELF_CORE_COPY_XFPREGS
1819 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1820 struct vm_area_struct *gate_vma)
1822 struct vm_area_struct *ret = tsk->mm->mmap;
1829 * Helper function for iterating across a vma list. It ensures that the caller
1830 * will visit `gate_vma' prior to terminating the search.
1832 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1833 struct vm_area_struct *gate_vma)
1835 struct vm_area_struct *ret;
1837 ret = this_vma->vm_next;
1840 if (this_vma == gate_vma)
1845 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1846 elf_addr_t e_shoff, int segs)
1848 elf->e_shoff = e_shoff;
1849 elf->e_shentsize = sizeof(*shdr4extnum);
1851 elf->e_shstrndx = SHN_UNDEF;
1853 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1855 shdr4extnum->sh_type = SHT_NULL;
1856 shdr4extnum->sh_size = elf->e_shnum;
1857 shdr4extnum->sh_link = elf->e_shstrndx;
1858 shdr4extnum->sh_info = segs;
1861 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1862 unsigned long mm_flags)
1864 struct vm_area_struct *vma;
1867 for (vma = first_vma(current, gate_vma); vma != NULL;
1868 vma = next_vma(vma, gate_vma))
1869 size += vma_dump_size(vma, mm_flags);
1876 * This is a two-pass process; first we find the offsets of the bits,
1877 * and then they are actually written out. If we run out of core limit
1880 static int elf_core_dump(struct coredump_params *cprm)
1886 struct vm_area_struct *vma, *gate_vma;
1887 struct elfhdr *elf = NULL;
1888 loff_t offset = 0, dataoff, foffset;
1889 struct elf_note_info info;
1890 struct elf_phdr *phdr4note = NULL;
1891 struct elf_shdr *shdr4extnum = NULL;
1896 * We no longer stop all VM operations.
1898 * This is because those proceses that could possibly change map_count
1899 * or the mmap / vma pages are now blocked in do_exit on current
1900 * finishing this core dump.
1902 * Only ptrace can touch these memory addresses, but it doesn't change
1903 * the map_count or the pages allocated. So no possibility of crashing
1904 * exists while dumping the mm->vm_next areas to the core file.
1907 /* alloc memory for large data structures: too large to be on stack */
1908 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1912 * The number of segs are recored into ELF header as 16bit value.
1913 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1915 segs = current->mm->map_count;
1916 segs += elf_core_extra_phdrs();
1918 gate_vma = get_gate_vma(current);
1919 if (gate_vma != NULL)
1922 /* for notes section */
1925 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1926 * this, kernel supports extended numbering. Have a look at
1927 * include/linux/elf.h for further information. */
1928 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1931 * Collect all the non-memory information about the process for the
1932 * notes. This also sets up the file header.
1934 if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
1938 current->flags |= PF_DUMPCORE;
1943 offset += sizeof(*elf); /* Elf header */
1944 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1947 /* Write notes phdr entry */
1949 size_t sz = get_note_info_size(&info);
1951 sz += elf_coredump_extra_notes_size();
1953 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1957 fill_elf_note_phdr(phdr4note, sz, offset);
1961 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1963 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
1964 offset += elf_core_extra_data_size();
1967 if (e_phnum == PN_XNUM) {
1968 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1971 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1976 size += sizeof(*elf);
1977 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1980 size += sizeof(*phdr4note);
1981 if (size > cprm->limit
1982 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1985 /* Write program headers for segments dump */
1986 for (vma = first_vma(current, gate_vma); vma != NULL;
1987 vma = next_vma(vma, gate_vma)) {
1988 struct elf_phdr phdr;
1990 phdr.p_type = PT_LOAD;
1991 phdr.p_offset = offset;
1992 phdr.p_vaddr = vma->vm_start;
1994 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
1995 phdr.p_memsz = vma->vm_end - vma->vm_start;
1996 offset += phdr.p_filesz;
1997 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1998 if (vma->vm_flags & VM_WRITE)
1999 phdr.p_flags |= PF_W;
2000 if (vma->vm_flags & VM_EXEC)
2001 phdr.p_flags |= PF_X;
2002 phdr.p_align = ELF_EXEC_PAGESIZE;
2004 size += sizeof(phdr);
2005 if (size > cprm->limit
2006 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2010 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2013 /* write out the notes section */
2014 if (!write_note_info(&info, cprm->file, &foffset))
2017 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2021 if (!dump_seek(cprm->file, dataoff - foffset))
2024 for (vma = first_vma(current, gate_vma); vma != NULL;
2025 vma = next_vma(vma, gate_vma)) {
2029 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2031 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2035 page = get_dump_page(addr);
2037 void *kaddr = kmap(page);
2038 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2039 !dump_write(cprm->file, kaddr,
2042 page_cache_release(page);
2044 stop = !dump_seek(cprm->file, PAGE_SIZE);
2050 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2053 if (e_phnum == PN_XNUM) {
2054 size += sizeof(*shdr4extnum);
2055 if (size > cprm->limit
2056 || !dump_write(cprm->file, shdr4extnum,
2057 sizeof(*shdr4extnum)))
2065 free_note_info(&info);
2073 #endif /* CONFIG_ELF_CORE */
2075 static int __init init_elf_binfmt(void)
2077 return register_binfmt(&elf_format);
2080 static void __exit exit_elf_binfmt(void)
2082 /* Remove the COFF and ELF loaders. */
2083 unregister_binfmt(&elf_format);
2086 core_initcall(init_elf_binfmt);
2087 module_exit(exit_elf_binfmt);
2088 MODULE_LICENSE("GPL");