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>
15 #include <linux/stat.h>
16 #include <linux/time.h>
18 #include <linux/mman.h>
19 #include <linux/errno.h>
20 #include <linux/signal.h>
21 #include <linux/binfmts.h>
22 #include <linux/string.h>
23 #include <linux/file.h>
24 #include <linux/fcntl.h>
25 #include <linux/ptrace.h>
26 #include <linux/slab.h>
27 #include <linux/shm.h>
28 #include <linux/personality.h>
29 #include <linux/elfcore.h>
30 #include <linux/init.h>
31 #include <linux/highuid.h>
32 #include <linux/smp.h>
33 #include <linux/compiler.h>
34 #include <linux/highmem.h>
35 #include <linux/pagemap.h>
36 #include <linux/security.h>
37 #include <linux/syscalls.h>
38 #include <linux/random.h>
39 #include <linux/elf.h>
40 #include <linux/utsname.h>
41 #include <asm/uaccess.h>
42 #include <asm/param.h>
45 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
46 static int load_elf_library(struct file *);
47 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
48 int, int, unsigned long);
51 * If we don't support core dumping, then supply a NULL so we
54 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
55 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit);
57 #define elf_core_dump NULL
60 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
61 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
63 #define ELF_MIN_ALIGN PAGE_SIZE
66 #ifndef ELF_CORE_EFLAGS
67 #define ELF_CORE_EFLAGS 0
70 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
71 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
72 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
74 static struct linux_binfmt elf_format = {
75 .module = THIS_MODULE,
76 .load_binary = load_elf_binary,
77 .load_shlib = load_elf_library,
78 .core_dump = elf_core_dump,
79 .min_coredump = ELF_EXEC_PAGESIZE,
83 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
85 static int set_brk(unsigned long start, unsigned long end)
87 start = ELF_PAGEALIGN(start);
88 end = ELF_PAGEALIGN(end);
91 down_write(¤t->mm->mmap_sem);
92 addr = do_brk(start, end - start);
93 up_write(¤t->mm->mmap_sem);
97 current->mm->start_brk = current->mm->brk = end;
101 /* We need to explicitly zero any fractional pages
102 after the data section (i.e. bss). This would
103 contain the junk from the file that should not
106 static int padzero(unsigned long elf_bss)
110 nbyte = ELF_PAGEOFFSET(elf_bss);
112 nbyte = ELF_MIN_ALIGN - nbyte;
113 if (clear_user((void __user *) elf_bss, nbyte))
119 /* Let's use some macros to make this stack manipulation a little clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 (((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
134 #ifndef ELF_BASE_PLATFORM
136 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
137 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
138 * will be copied to the user stack in the same manner as AT_PLATFORM.
140 #define ELF_BASE_PLATFORM NULL
144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
145 unsigned long load_addr, unsigned long interp_load_addr)
147 unsigned long p = bprm->p;
148 int argc = bprm->argc;
149 int envc = bprm->envc;
150 elf_addr_t __user *argv;
151 elf_addr_t __user *envp;
152 elf_addr_t __user *sp;
153 elf_addr_t __user *u_platform;
154 elf_addr_t __user *u_base_platform;
155 elf_addr_t __user *u_rand_bytes;
156 const char *k_platform = ELF_PLATFORM;
157 const char *k_base_platform = ELF_BASE_PLATFORM;
158 unsigned char k_rand_bytes[16];
160 elf_addr_t *elf_info;
162 const struct cred *cred = current_cred();
163 struct vm_area_struct *vma;
166 * In some cases (e.g. Hyper-Threading), we want to avoid L1
167 * evictions by the processes running on the same package. One
168 * thing we can do is to shuffle the initial stack for them.
171 p = arch_align_stack(p);
174 * If this architecture has a platform capability string, copy it
175 * to userspace. In some cases (Sparc), this info is impossible
176 * for userspace to get any other way, in others (i386) it is
181 size_t len = strlen(k_platform) + 1;
183 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
184 if (__copy_to_user(u_platform, k_platform, len))
189 * If this architecture has a "base" platform capability
190 * string, copy it to userspace.
192 u_base_platform = NULL;
193 if (k_base_platform) {
194 size_t len = strlen(k_base_platform) + 1;
196 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 if (__copy_to_user(u_base_platform, k_base_platform, len))
202 * Generate 16 random bytes for userspace PRNG seeding.
204 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
205 u_rand_bytes = (elf_addr_t __user *)
206 STACK_ALLOC(p, sizeof(k_rand_bytes));
207 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
210 /* Create the ELF interpreter info */
211 elf_info = (elf_addr_t *)current->mm->saved_auxv;
212 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
213 #define NEW_AUX_ENT(id, val) \
215 elf_info[ei_index++] = id; \
216 elf_info[ei_index++] = val; \
221 * ARCH_DLINFO must come first so PPC can do its special alignment of
223 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
224 * ARCH_DLINFO changes
228 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
229 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
230 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
231 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
232 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
233 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
234 NEW_AUX_ENT(AT_BASE, interp_load_addr);
235 NEW_AUX_ENT(AT_FLAGS, 0);
236 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
237 NEW_AUX_ENT(AT_UID, cred->uid);
238 NEW_AUX_ENT(AT_EUID, cred->euid);
239 NEW_AUX_ENT(AT_GID, cred->gid);
240 NEW_AUX_ENT(AT_EGID, cred->egid);
241 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
242 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
243 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
245 NEW_AUX_ENT(AT_PLATFORM,
246 (elf_addr_t)(unsigned long)u_platform);
248 if (k_base_platform) {
249 NEW_AUX_ENT(AT_BASE_PLATFORM,
250 (elf_addr_t)(unsigned long)u_base_platform);
252 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
253 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
256 /* AT_NULL is zero; clear the rest too */
257 memset(&elf_info[ei_index], 0,
258 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
260 /* And advance past the AT_NULL entry. */
263 sp = STACK_ADD(p, ei_index);
265 items = (argc + 1) + (envc + 1) + 1;
266 bprm->p = STACK_ROUND(sp, items);
268 /* Point sp at the lowest address on the stack */
269 #ifdef CONFIG_STACK_GROWSUP
270 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
271 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
273 sp = (elf_addr_t __user *)bprm->p;
278 * Grow the stack manually; some architectures have a limit on how
279 * far ahead a user-space access may be in order to grow the stack.
281 vma = find_extend_vma(current->mm, bprm->p);
285 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
286 if (__put_user(argc, sp++))
289 envp = argv + argc + 1;
291 /* Populate argv and envp */
292 p = current->mm->arg_end = current->mm->arg_start;
295 if (__put_user((elf_addr_t)p, argv++))
297 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
298 if (!len || len > MAX_ARG_STRLEN)
302 if (__put_user(0, argv))
304 current->mm->arg_end = current->mm->env_start = p;
307 if (__put_user((elf_addr_t)p, envp++))
309 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
310 if (!len || len > MAX_ARG_STRLEN)
314 if (__put_user(0, envp))
316 current->mm->env_end = p;
318 /* Put the elf_info on the stack in the right place. */
319 sp = (elf_addr_t __user *)envp + 1;
320 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
327 static unsigned long elf_map(struct file *filep, unsigned long addr,
328 struct elf_phdr *eppnt, int prot, int type,
329 unsigned long total_size)
331 unsigned long map_addr;
332 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
333 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
334 addr = ELF_PAGESTART(addr);
335 size = ELF_PAGEALIGN(size);
337 /* mmap() will return -EINVAL if given a zero size, but a
338 * segment with zero filesize is perfectly valid */
342 down_write(¤t->mm->mmap_sem);
344 * total_size is the size of the ELF (interpreter) image.
345 * The _first_ mmap needs to know the full size, otherwise
346 * randomization might put this image into an overlapping
347 * position with the ELF binary image. (since size < total_size)
348 * So we first map the 'big' image - and unmap the remainder at
349 * the end. (which unmap is needed for ELF images with holes.)
352 total_size = ELF_PAGEALIGN(total_size);
353 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
354 if (!BAD_ADDR(map_addr))
355 do_munmap(current->mm, map_addr+size, total_size-size);
357 map_addr = do_mmap(filep, addr, size, prot, type, off);
359 up_write(¤t->mm->mmap_sem);
363 #endif /* !elf_map */
365 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
367 int i, first_idx = -1, last_idx = -1;
369 for (i = 0; i < nr; i++) {
370 if (cmds[i].p_type == PT_LOAD) {
379 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
380 ELF_PAGESTART(cmds[first_idx].p_vaddr);
384 /* This is much more generalized than the library routine read function,
385 so we keep this separate. Technically the library read function
386 is only provided so that we can read a.out libraries that have
389 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
390 struct file *interpreter, unsigned long *interp_map_addr,
391 unsigned long no_base)
393 struct elf_phdr *elf_phdata;
394 struct elf_phdr *eppnt;
395 unsigned long load_addr = 0;
396 int load_addr_set = 0;
397 unsigned long last_bss = 0, elf_bss = 0;
398 unsigned long error = ~0UL;
399 unsigned long total_size;
402 /* First of all, some simple consistency checks */
403 if (interp_elf_ex->e_type != ET_EXEC &&
404 interp_elf_ex->e_type != ET_DYN)
406 if (!elf_check_arch(interp_elf_ex))
408 if (!interpreter->f_op || !interpreter->f_op->mmap)
412 * If the size of this structure has changed, then punt, since
413 * we will be doing the wrong thing.
415 if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
417 if (interp_elf_ex->e_phnum < 1 ||
418 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
421 /* Now read in all of the header information */
422 size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
423 if (size > ELF_MIN_ALIGN)
425 elf_phdata = kmalloc(size, GFP_KERNEL);
429 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
430 (char *)elf_phdata,size);
432 if (retval != size) {
438 total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
445 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
446 if (eppnt->p_type == PT_LOAD) {
447 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
449 unsigned long vaddr = 0;
450 unsigned long k, map_addr;
452 if (eppnt->p_flags & PF_R)
453 elf_prot = PROT_READ;
454 if (eppnt->p_flags & PF_W)
455 elf_prot |= PROT_WRITE;
456 if (eppnt->p_flags & PF_X)
457 elf_prot |= PROT_EXEC;
458 vaddr = eppnt->p_vaddr;
459 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
460 elf_type |= MAP_FIXED;
461 else if (no_base && interp_elf_ex->e_type == ET_DYN)
464 map_addr = elf_map(interpreter, load_addr + vaddr,
465 eppnt, elf_prot, elf_type, total_size);
467 if (!*interp_map_addr)
468 *interp_map_addr = map_addr;
470 if (BAD_ADDR(map_addr))
473 if (!load_addr_set &&
474 interp_elf_ex->e_type == ET_DYN) {
475 load_addr = map_addr - ELF_PAGESTART(vaddr);
480 * Check to see if the section's size will overflow the
481 * allowed task size. Note that p_filesz must always be
482 * <= p_memsize so it's only necessary to check p_memsz.
484 k = load_addr + eppnt->p_vaddr;
486 eppnt->p_filesz > eppnt->p_memsz ||
487 eppnt->p_memsz > TASK_SIZE ||
488 TASK_SIZE - eppnt->p_memsz < k) {
494 * Find the end of the file mapping for this phdr, and
495 * keep track of the largest address we see for this.
497 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
502 * Do the same thing for the memory mapping - between
503 * elf_bss and last_bss is the bss section.
505 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
512 * Now fill out the bss section. First pad the last page up
513 * to the page boundary, and then perform a mmap to make sure
514 * that there are zero-mapped pages up to and including the
517 if (padzero(elf_bss)) {
522 /* What we have mapped so far */
523 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
525 /* Map the last of the bss segment */
526 if (last_bss > elf_bss) {
527 down_write(¤t->mm->mmap_sem);
528 error = do_brk(elf_bss, last_bss - elf_bss);
529 up_write(¤t->mm->mmap_sem);
543 * These are the functions used to load ELF style executables and shared
544 * libraries. There is no binary dependent code anywhere else.
547 #define INTERPRETER_NONE 0
548 #define INTERPRETER_ELF 2
550 #ifndef STACK_RND_MASK
551 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
554 static unsigned long randomize_stack_top(unsigned long stack_top)
556 unsigned int random_variable = 0;
558 if ((current->flags & PF_RANDOMIZE) &&
559 !(current->personality & ADDR_NO_RANDOMIZE)) {
560 random_variable = get_random_int() & STACK_RND_MASK;
561 random_variable <<= PAGE_SHIFT;
563 #ifdef CONFIG_STACK_GROWSUP
564 return PAGE_ALIGN(stack_top) + random_variable;
566 return PAGE_ALIGN(stack_top) - random_variable;
570 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
572 struct file *interpreter = NULL; /* to shut gcc up */
573 unsigned long load_addr = 0, load_bias = 0;
574 int load_addr_set = 0;
575 char * elf_interpreter = NULL;
577 struct elf_phdr *elf_ppnt, *elf_phdata;
578 unsigned long elf_bss, elf_brk;
581 unsigned long elf_entry;
582 unsigned long interp_load_addr = 0;
583 unsigned long start_code, end_code, start_data, end_data;
584 unsigned long reloc_func_desc = 0;
585 int executable_stack = EXSTACK_DEFAULT;
586 unsigned long def_flags = 0;
588 struct elfhdr elf_ex;
589 struct elfhdr interp_elf_ex;
592 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
598 /* Get the exec-header */
599 loc->elf_ex = *((struct elfhdr *)bprm->buf);
602 /* First of all, some simple consistency checks */
603 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
606 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
608 if (!elf_check_arch(&loc->elf_ex))
610 if (!bprm->file->f_op||!bprm->file->f_op->mmap)
613 /* Now read in all of the header information */
614 if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
616 if (loc->elf_ex.e_phnum < 1 ||
617 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
619 size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
621 elf_phdata = kmalloc(size, GFP_KERNEL);
625 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
626 (char *)elf_phdata, size);
627 if (retval != size) {
633 elf_ppnt = elf_phdata;
642 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
643 if (elf_ppnt->p_type == PT_INTERP) {
644 /* This is the program interpreter used for
645 * shared libraries - for now assume that this
646 * is an a.out format binary
649 if (elf_ppnt->p_filesz > PATH_MAX ||
650 elf_ppnt->p_filesz < 2)
654 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
656 if (!elf_interpreter)
659 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
662 if (retval != elf_ppnt->p_filesz) {
665 goto out_free_interp;
667 /* make sure path is NULL terminated */
669 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
670 goto out_free_interp;
673 * The early SET_PERSONALITY here is so that the lookup
674 * for the interpreter happens in the namespace of the
675 * to-be-execed image. SET_PERSONALITY can select an
678 * However, SET_PERSONALITY is NOT allowed to switch
679 * this task into the new images's memory mapping
680 * policy - that is, TASK_SIZE must still evaluate to
681 * that which is appropriate to the execing application.
682 * This is because exit_mmap() needs to have TASK_SIZE
683 * evaluate to the size of the old image.
685 * So if (say) a 64-bit application is execing a 32-bit
686 * application it is the architecture's responsibility
687 * to defer changing the value of TASK_SIZE until the
688 * switch really is going to happen - do this in
689 * flush_thread(). - akpm
691 SET_PERSONALITY(loc->elf_ex);
693 interpreter = open_exec(elf_interpreter);
694 retval = PTR_ERR(interpreter);
695 if (IS_ERR(interpreter))
696 goto out_free_interp;
699 * If the binary is not readable then enforce
700 * mm->dumpable = 0 regardless of the interpreter's
703 if (file_permission(interpreter, MAY_READ) < 0)
704 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
706 retval = kernel_read(interpreter, 0, bprm->buf,
708 if (retval != BINPRM_BUF_SIZE) {
711 goto out_free_dentry;
714 /* Get the exec headers */
715 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
721 elf_ppnt = elf_phdata;
722 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
723 if (elf_ppnt->p_type == PT_GNU_STACK) {
724 if (elf_ppnt->p_flags & PF_X)
725 executable_stack = EXSTACK_ENABLE_X;
727 executable_stack = EXSTACK_DISABLE_X;
731 /* Some simple consistency checks for the interpreter */
732 if (elf_interpreter) {
734 /* Not an ELF interpreter */
735 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
736 goto out_free_dentry;
737 /* Verify the interpreter has a valid arch */
738 if (!elf_check_arch(&loc->interp_elf_ex))
739 goto out_free_dentry;
741 /* Executables without an interpreter also need a personality */
742 SET_PERSONALITY(loc->elf_ex);
745 /* Flush all traces of the currently running executable */
746 retval = flush_old_exec(bprm);
748 goto out_free_dentry;
750 /* OK, This is the point of no return */
751 current->flags &= ~PF_FORKNOEXEC;
752 current->mm->def_flags = def_flags;
754 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
755 may depend on the personality. */
756 SET_PERSONALITY(loc->elf_ex);
757 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
758 current->personality |= READ_IMPLIES_EXEC;
760 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
761 current->flags |= PF_RANDOMIZE;
762 arch_pick_mmap_layout(current->mm);
764 /* Do this so that we can load the interpreter, if need be. We will
765 change some of these later */
766 current->mm->free_area_cache = current->mm->mmap_base;
767 current->mm->cached_hole_size = 0;
768 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
771 send_sig(SIGKILL, current, 0);
772 goto out_free_dentry;
775 current->mm->start_stack = bprm->p;
777 /* Now we do a little grungy work by mmaping the ELF image into
778 the correct location in memory. */
779 for(i = 0, elf_ppnt = elf_phdata;
780 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
781 int elf_prot = 0, elf_flags;
782 unsigned long k, vaddr;
784 if (elf_ppnt->p_type != PT_LOAD)
787 if (unlikely (elf_brk > elf_bss)) {
790 /* There was a PT_LOAD segment with p_memsz > p_filesz
791 before this one. Map anonymous pages, if needed,
792 and clear the area. */
793 retval = set_brk (elf_bss + load_bias,
794 elf_brk + load_bias);
796 send_sig(SIGKILL, current, 0);
797 goto out_free_dentry;
799 nbyte = ELF_PAGEOFFSET(elf_bss);
801 nbyte = ELF_MIN_ALIGN - nbyte;
802 if (nbyte > elf_brk - elf_bss)
803 nbyte = elf_brk - elf_bss;
804 if (clear_user((void __user *)elf_bss +
807 * This bss-zeroing can fail if the ELF
808 * file specifies odd protections. So
809 * we don't check the return value
815 if (elf_ppnt->p_flags & PF_R)
816 elf_prot |= PROT_READ;
817 if (elf_ppnt->p_flags & PF_W)
818 elf_prot |= PROT_WRITE;
819 if (elf_ppnt->p_flags & PF_X)
820 elf_prot |= PROT_EXEC;
822 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
824 vaddr = elf_ppnt->p_vaddr;
825 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
826 elf_flags |= MAP_FIXED;
827 } else if (loc->elf_ex.e_type == ET_DYN) {
828 /* Try and get dynamic programs out of the way of the
829 * default mmap base, as well as whatever program they
830 * might try to exec. This is because the brk will
831 * follow the loader, and is not movable. */
835 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
839 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
840 elf_prot, elf_flags, 0);
841 if (BAD_ADDR(error)) {
842 send_sig(SIGKILL, current, 0);
843 retval = IS_ERR((void *)error) ?
844 PTR_ERR((void*)error) : -EINVAL;
845 goto out_free_dentry;
848 if (!load_addr_set) {
850 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
851 if (loc->elf_ex.e_type == ET_DYN) {
853 ELF_PAGESTART(load_bias + vaddr);
854 load_addr += load_bias;
855 reloc_func_desc = load_bias;
858 k = elf_ppnt->p_vaddr;
865 * Check to see if the section's size will overflow the
866 * allowed task size. Note that p_filesz must always be
867 * <= p_memsz so it is only necessary to check p_memsz.
869 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
870 elf_ppnt->p_memsz > TASK_SIZE ||
871 TASK_SIZE - elf_ppnt->p_memsz < k) {
872 /* set_brk can never work. Avoid overflows. */
873 send_sig(SIGKILL, current, 0);
875 goto out_free_dentry;
878 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
882 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
886 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
891 loc->elf_ex.e_entry += load_bias;
892 elf_bss += load_bias;
893 elf_brk += load_bias;
894 start_code += load_bias;
895 end_code += load_bias;
896 start_data += load_bias;
897 end_data += load_bias;
899 /* Calling set_brk effectively mmaps the pages that we need
900 * for the bss and break sections. We must do this before
901 * mapping in the interpreter, to make sure it doesn't wind
902 * up getting placed where the bss needs to go.
904 retval = set_brk(elf_bss, elf_brk);
906 send_sig(SIGKILL, current, 0);
907 goto out_free_dentry;
909 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
910 send_sig(SIGSEGV, current, 0);
911 retval = -EFAULT; /* Nobody gets to see this, but.. */
912 goto out_free_dentry;
915 if (elf_interpreter) {
916 unsigned long uninitialized_var(interp_map_addr);
918 elf_entry = load_elf_interp(&loc->interp_elf_ex,
922 if (!IS_ERR((void *)elf_entry)) {
924 * load_elf_interp() returns relocation
927 interp_load_addr = elf_entry;
928 elf_entry += loc->interp_elf_ex.e_entry;
930 if (BAD_ADDR(elf_entry)) {
931 force_sig(SIGSEGV, current);
932 retval = IS_ERR((void *)elf_entry) ?
933 (int)elf_entry : -EINVAL;
934 goto out_free_dentry;
936 reloc_func_desc = interp_load_addr;
938 allow_write_access(interpreter);
940 kfree(elf_interpreter);
942 elf_entry = loc->elf_ex.e_entry;
943 if (BAD_ADDR(elf_entry)) {
944 force_sig(SIGSEGV, current);
946 goto out_free_dentry;
952 set_binfmt(&elf_format);
954 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
955 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
957 send_sig(SIGKILL, current, 0);
960 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
962 install_exec_creds(bprm);
963 current->flags &= ~PF_FORKNOEXEC;
964 retval = create_elf_tables(bprm, &loc->elf_ex,
965 load_addr, interp_load_addr);
967 send_sig(SIGKILL, current, 0);
970 /* N.B. passed_fileno might not be initialized? */
971 current->mm->end_code = end_code;
972 current->mm->start_code = start_code;
973 current->mm->start_data = start_data;
974 current->mm->end_data = end_data;
975 current->mm->start_stack = bprm->p;
977 #ifdef arch_randomize_brk
978 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
979 current->mm->brk = current->mm->start_brk =
980 arch_randomize_brk(current->mm);
983 if (current->personality & MMAP_PAGE_ZERO) {
984 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
985 and some applications "depend" upon this behavior.
986 Since we do not have the power to recompile these, we
987 emulate the SVr4 behavior. Sigh. */
988 down_write(¤t->mm->mmap_sem);
989 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
990 MAP_FIXED | MAP_PRIVATE, 0);
991 up_write(¤t->mm->mmap_sem);
996 * The ABI may specify that certain registers be set up in special
997 * ways (on i386 %edx is the address of a DT_FINI function, for
998 * example. In addition, it may also specify (eg, PowerPC64 ELF)
999 * that the e_entry field is the address of the function descriptor
1000 * for the startup routine, rather than the address of the startup
1001 * routine itself. This macro performs whatever initialization to
1002 * the regs structure is required as well as any relocations to the
1003 * function descriptor entries when executing dynamically links apps.
1005 ELF_PLAT_INIT(regs, reloc_func_desc);
1008 start_thread(regs, elf_entry, bprm->p);
1017 allow_write_access(interpreter);
1021 kfree(elf_interpreter);
1027 /* This is really simpleminded and specialized - we are loading an
1028 a.out library that is given an ELF header. */
1029 static int load_elf_library(struct file *file)
1031 struct elf_phdr *elf_phdata;
1032 struct elf_phdr *eppnt;
1033 unsigned long elf_bss, bss, len;
1034 int retval, error, i, j;
1035 struct elfhdr elf_ex;
1038 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1039 if (retval != sizeof(elf_ex))
1042 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1045 /* First of all, some simple consistency checks */
1046 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1047 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1050 /* Now read in all of the header information */
1052 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1053 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1056 elf_phdata = kmalloc(j, GFP_KERNEL);
1062 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1066 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1067 if ((eppnt + i)->p_type == PT_LOAD)
1072 while (eppnt->p_type != PT_LOAD)
1075 /* Now use mmap to map the library into memory. */
1076 down_write(¤t->mm->mmap_sem);
1077 error = do_mmap(file,
1078 ELF_PAGESTART(eppnt->p_vaddr),
1080 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1081 PROT_READ | PROT_WRITE | PROT_EXEC,
1082 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1084 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1085 up_write(¤t->mm->mmap_sem);
1086 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1089 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1090 if (padzero(elf_bss)) {
1095 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1097 bss = eppnt->p_memsz + eppnt->p_vaddr;
1099 down_write(¤t->mm->mmap_sem);
1100 do_brk(len, bss - len);
1101 up_write(¤t->mm->mmap_sem);
1112 * Note that some platforms still use traditional core dumps and not
1113 * the ELF core dump. Each platform can select it as appropriate.
1115 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1120 * Modelled on fs/exec.c:aout_core_dump()
1121 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1124 * These are the only things you should do on a core-file: use only these
1125 * functions to write out all the necessary info.
1127 static int dump_write(struct file *file, const void *addr, int nr)
1129 return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1132 static int dump_seek(struct file *file, loff_t off)
1134 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
1135 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
1138 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
1142 unsigned long n = off;
1145 if (!dump_write(file, buf, n))
1149 free_page((unsigned long)buf);
1155 * Decide what to dump of a segment, part, all or none.
1157 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1158 unsigned long mm_flags)
1160 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1162 /* The vma can be set up to tell us the answer directly. */
1163 if (vma->vm_flags & VM_ALWAYSDUMP)
1166 /* Hugetlb memory check */
1167 if (vma->vm_flags & VM_HUGETLB) {
1168 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1170 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1174 /* Do not dump I/O mapped devices or special mappings */
1175 if (vma->vm_flags & (VM_IO | VM_RESERVED))
1178 /* By default, dump shared memory if mapped from an anonymous file. */
1179 if (vma->vm_flags & VM_SHARED) {
1180 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1181 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1186 /* Dump segments that have been written to. */
1187 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1189 if (vma->vm_file == NULL)
1192 if (FILTER(MAPPED_PRIVATE))
1196 * If this looks like the beginning of a DSO or executable mapping,
1197 * check for an ELF header. If we find one, dump the first page to
1198 * aid in determining what was mapped here.
1200 if (FILTER(ELF_HEADERS) &&
1201 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1202 u32 __user *header = (u32 __user *) vma->vm_start;
1204 mm_segment_t fs = get_fs();
1206 * Doing it this way gets the constant folded by GCC.
1210 char elfmag[SELFMAG];
1212 BUILD_BUG_ON(SELFMAG != sizeof word);
1213 magic.elfmag[EI_MAG0] = ELFMAG0;
1214 magic.elfmag[EI_MAG1] = ELFMAG1;
1215 magic.elfmag[EI_MAG2] = ELFMAG2;
1216 magic.elfmag[EI_MAG3] = ELFMAG3;
1218 * Switch to the user "segment" for get_user(),
1219 * then put back what elf_core_dump() had in place.
1222 if (unlikely(get_user(word, header)))
1225 if (word == magic.cmp)
1234 return vma->vm_end - vma->vm_start;
1237 /* An ELF note in memory */
1242 unsigned int datasz;
1246 static int notesize(struct memelfnote *en)
1250 sz = sizeof(struct elf_note);
1251 sz += roundup(strlen(en->name) + 1, 4);
1252 sz += roundup(en->datasz, 4);
1257 #define DUMP_WRITE(addr, nr, foffset) \
1258 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1260 static int alignfile(struct file *file, loff_t *foffset)
1262 static const char buf[4] = { 0, };
1263 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1267 static int writenote(struct memelfnote *men, struct file *file,
1271 en.n_namesz = strlen(men->name) + 1;
1272 en.n_descsz = men->datasz;
1273 en.n_type = men->type;
1275 DUMP_WRITE(&en, sizeof(en), foffset);
1276 DUMP_WRITE(men->name, en.n_namesz, foffset);
1277 if (!alignfile(file, foffset))
1279 DUMP_WRITE(men->data, men->datasz, foffset);
1280 if (!alignfile(file, foffset))
1287 #define DUMP_WRITE(addr, nr) \
1288 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1290 #define DUMP_SEEK(off) \
1291 if (!dump_seek(file, (off))) \
1294 static void fill_elf_header(struct elfhdr *elf, int segs,
1295 u16 machine, u32 flags, u8 osabi)
1297 memset(elf, 0, sizeof(*elf));
1299 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1300 elf->e_ident[EI_CLASS] = ELF_CLASS;
1301 elf->e_ident[EI_DATA] = ELF_DATA;
1302 elf->e_ident[EI_VERSION] = EV_CURRENT;
1303 elf->e_ident[EI_OSABI] = ELF_OSABI;
1305 elf->e_type = ET_CORE;
1306 elf->e_machine = machine;
1307 elf->e_version = EV_CURRENT;
1308 elf->e_phoff = sizeof(struct elfhdr);
1309 elf->e_flags = flags;
1310 elf->e_ehsize = sizeof(struct elfhdr);
1311 elf->e_phentsize = sizeof(struct elf_phdr);
1312 elf->e_phnum = segs;
1317 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1319 phdr->p_type = PT_NOTE;
1320 phdr->p_offset = offset;
1323 phdr->p_filesz = sz;
1330 static void fill_note(struct memelfnote *note, const char *name, int type,
1331 unsigned int sz, void *data)
1341 * fill up all the fields in prstatus from the given task struct, except
1342 * registers which need to be filled up separately.
1344 static void fill_prstatus(struct elf_prstatus *prstatus,
1345 struct task_struct *p, long signr)
1347 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1348 prstatus->pr_sigpend = p->pending.signal.sig[0];
1349 prstatus->pr_sighold = p->blocked.sig[0];
1350 prstatus->pr_pid = task_pid_vnr(p);
1351 prstatus->pr_ppid = task_pid_vnr(p->real_parent);
1352 prstatus->pr_pgrp = task_pgrp_vnr(p);
1353 prstatus->pr_sid = task_session_vnr(p);
1354 if (thread_group_leader(p)) {
1355 struct task_cputime cputime;
1358 * This is the record for the group leader. It shows the
1359 * group-wide total, not its individual thread total.
1361 thread_group_cputime(p, &cputime);
1362 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1363 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1365 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1366 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1368 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1369 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1372 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1373 struct mm_struct *mm)
1375 const struct cred *cred;
1376 unsigned int i, len;
1378 /* first copy the parameters from user space */
1379 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1381 len = mm->arg_end - mm->arg_start;
1382 if (len >= ELF_PRARGSZ)
1383 len = ELF_PRARGSZ-1;
1384 if (copy_from_user(&psinfo->pr_psargs,
1385 (const char __user *)mm->arg_start, len))
1387 for(i = 0; i < len; i++)
1388 if (psinfo->pr_psargs[i] == 0)
1389 psinfo->pr_psargs[i] = ' ';
1390 psinfo->pr_psargs[len] = 0;
1392 psinfo->pr_pid = task_pid_vnr(p);
1393 psinfo->pr_ppid = task_pid_vnr(p->real_parent);
1394 psinfo->pr_pgrp = task_pgrp_vnr(p);
1395 psinfo->pr_sid = task_session_vnr(p);
1397 i = p->state ? ffz(~p->state) + 1 : 0;
1398 psinfo->pr_state = i;
1399 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1400 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1401 psinfo->pr_nice = task_nice(p);
1402 psinfo->pr_flag = p->flags;
1404 cred = __task_cred(p);
1405 SET_UID(psinfo->pr_uid, cred->uid);
1406 SET_GID(psinfo->pr_gid, cred->gid);
1408 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1413 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1415 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1419 while (auxv[i - 2] != AT_NULL);
1420 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1423 #ifdef CORE_DUMP_USE_REGSET
1424 #include <linux/regset.h>
1426 struct elf_thread_core_info {
1427 struct elf_thread_core_info *next;
1428 struct task_struct *task;
1429 struct elf_prstatus prstatus;
1430 struct memelfnote notes[0];
1433 struct elf_note_info {
1434 struct elf_thread_core_info *thread;
1435 struct memelfnote psinfo;
1436 struct memelfnote auxv;
1442 * When a regset has a writeback hook, we call it on each thread before
1443 * dumping user memory. On register window machines, this makes sure the
1444 * user memory backing the register data is up to date before we read it.
1446 static void do_thread_regset_writeback(struct task_struct *task,
1447 const struct user_regset *regset)
1449 if (regset->writeback)
1450 regset->writeback(task, regset, 1);
1453 static int fill_thread_core_info(struct elf_thread_core_info *t,
1454 const struct user_regset_view *view,
1455 long signr, size_t *total)
1460 * NT_PRSTATUS is the one special case, because the regset data
1461 * goes into the pr_reg field inside the note contents, rather
1462 * than being the whole note contents. We fill the reset in here.
1463 * We assume that regset 0 is NT_PRSTATUS.
1465 fill_prstatus(&t->prstatus, t->task, signr);
1466 (void) view->regsets[0].get(t->task, &view->regsets[0],
1467 0, sizeof(t->prstatus.pr_reg),
1468 &t->prstatus.pr_reg, NULL);
1470 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1471 sizeof(t->prstatus), &t->prstatus);
1472 *total += notesize(&t->notes[0]);
1474 do_thread_regset_writeback(t->task, &view->regsets[0]);
1477 * Each other regset might generate a note too. For each regset
1478 * that has no core_note_type or is inactive, we leave t->notes[i]
1479 * all zero and we'll know to skip writing it later.
1481 for (i = 1; i < view->n; ++i) {
1482 const struct user_regset *regset = &view->regsets[i];
1483 do_thread_regset_writeback(t->task, regset);
1484 if (regset->core_note_type &&
1485 (!regset->active || regset->active(t->task, regset))) {
1487 size_t size = regset->n * regset->size;
1488 void *data = kmalloc(size, GFP_KERNEL);
1489 if (unlikely(!data))
1491 ret = regset->get(t->task, regset,
1492 0, size, data, NULL);
1496 if (regset->core_note_type != NT_PRFPREG)
1497 fill_note(&t->notes[i], "LINUX",
1498 regset->core_note_type,
1501 t->prstatus.pr_fpvalid = 1;
1502 fill_note(&t->notes[i], "CORE",
1503 NT_PRFPREG, size, data);
1505 *total += notesize(&t->notes[i]);
1513 static int fill_note_info(struct elfhdr *elf, int phdrs,
1514 struct elf_note_info *info,
1515 long signr, struct pt_regs *regs)
1517 struct task_struct *dump_task = current;
1518 const struct user_regset_view *view = task_user_regset_view(dump_task);
1519 struct elf_thread_core_info *t;
1520 struct elf_prpsinfo *psinfo;
1521 struct core_thread *ct;
1525 info->thread = NULL;
1527 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1528 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1534 * Figure out how many notes we're going to need for each thread.
1536 info->thread_notes = 0;
1537 for (i = 0; i < view->n; ++i)
1538 if (view->regsets[i].core_note_type != 0)
1539 ++info->thread_notes;
1542 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1543 * since it is our one special case.
1545 if (unlikely(info->thread_notes == 0) ||
1546 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1552 * Initialize the ELF file header.
1554 fill_elf_header(elf, phdrs,
1555 view->e_machine, view->e_flags, view->ei_osabi);
1558 * Allocate a structure for each thread.
1560 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1561 t = kzalloc(offsetof(struct elf_thread_core_info,
1562 notes[info->thread_notes]),
1568 if (ct->task == dump_task || !info->thread) {
1569 t->next = info->thread;
1573 * Make sure to keep the original task at
1574 * the head of the list.
1576 t->next = info->thread->next;
1577 info->thread->next = t;
1582 * Now fill in each thread's information.
1584 for (t = info->thread; t != NULL; t = t->next)
1585 if (!fill_thread_core_info(t, view, signr, &info->size))
1589 * Fill in the two process-wide notes.
1591 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1592 info->size += notesize(&info->psinfo);
1594 fill_auxv_note(&info->auxv, current->mm);
1595 info->size += notesize(&info->auxv);
1600 static size_t get_note_info_size(struct elf_note_info *info)
1606 * Write all the notes for each thread. When writing the first thread, the
1607 * process-wide notes are interleaved after the first thread-specific note.
1609 static int write_note_info(struct elf_note_info *info,
1610 struct file *file, loff_t *foffset)
1613 struct elf_thread_core_info *t = info->thread;
1618 if (!writenote(&t->notes[0], file, foffset))
1621 if (first && !writenote(&info->psinfo, file, foffset))
1623 if (first && !writenote(&info->auxv, file, foffset))
1626 for (i = 1; i < info->thread_notes; ++i)
1627 if (t->notes[i].data &&
1628 !writenote(&t->notes[i], file, foffset))
1638 static void free_note_info(struct elf_note_info *info)
1640 struct elf_thread_core_info *threads = info->thread;
1643 struct elf_thread_core_info *t = threads;
1645 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1646 for (i = 1; i < info->thread_notes; ++i)
1647 kfree(t->notes[i].data);
1650 kfree(info->psinfo.data);
1655 /* Here is the structure in which status of each thread is captured. */
1656 struct elf_thread_status
1658 struct list_head list;
1659 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1660 elf_fpregset_t fpu; /* NT_PRFPREG */
1661 struct task_struct *thread;
1662 #ifdef ELF_CORE_COPY_XFPREGS
1663 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1665 struct memelfnote notes[3];
1670 * In order to add the specific thread information for the elf file format,
1671 * we need to keep a linked list of every threads pr_status and then create
1672 * a single section for them in the final core file.
1674 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1677 struct task_struct *p = t->thread;
1680 fill_prstatus(&t->prstatus, p, signr);
1681 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1683 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1686 sz += notesize(&t->notes[0]);
1688 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1690 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1693 sz += notesize(&t->notes[1]);
1696 #ifdef ELF_CORE_COPY_XFPREGS
1697 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1698 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1699 sizeof(t->xfpu), &t->xfpu);
1701 sz += notesize(&t->notes[2]);
1707 struct elf_note_info {
1708 struct memelfnote *notes;
1709 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1710 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1711 struct list_head thread_list;
1712 elf_fpregset_t *fpu;
1713 #ifdef ELF_CORE_COPY_XFPREGS
1714 elf_fpxregset_t *xfpu;
1716 int thread_status_size;
1720 static int fill_note_info(struct elfhdr *elf, int phdrs,
1721 struct elf_note_info *info,
1722 long signr, struct pt_regs *regs)
1725 struct list_head *t;
1728 info->prstatus = NULL;
1729 info->psinfo = NULL;
1731 #ifdef ELF_CORE_COPY_XFPREGS
1734 INIT_LIST_HEAD(&info->thread_list);
1736 info->notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote),
1740 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1743 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1744 if (!info->prstatus)
1746 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1749 #ifdef ELF_CORE_COPY_XFPREGS
1750 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1755 info->thread_status_size = 0;
1757 struct core_thread *ct;
1758 struct elf_thread_status *ets;
1760 for (ct = current->mm->core_state->dumper.next;
1761 ct; ct = ct->next) {
1762 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1766 ets->thread = ct->task;
1767 list_add(&ets->list, &info->thread_list);
1770 list_for_each(t, &info->thread_list) {
1773 ets = list_entry(t, struct elf_thread_status, list);
1774 sz = elf_dump_thread_status(signr, ets);
1775 info->thread_status_size += sz;
1778 /* now collect the dump for the current */
1779 memset(info->prstatus, 0, sizeof(*info->prstatus));
1780 fill_prstatus(info->prstatus, current, signr);
1781 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1784 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1787 * Set up the notes in similar form to SVR4 core dumps made
1788 * with info from their /proc.
1791 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1792 sizeof(*info->prstatus), info->prstatus);
1793 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1794 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1795 sizeof(*info->psinfo), info->psinfo);
1799 fill_auxv_note(&info->notes[info->numnote++], current->mm);
1801 /* Try to dump the FPU. */
1802 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1804 if (info->prstatus->pr_fpvalid)
1805 fill_note(info->notes + info->numnote++,
1806 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1807 #ifdef ELF_CORE_COPY_XFPREGS
1808 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1809 fill_note(info->notes + info->numnote++,
1810 "LINUX", ELF_CORE_XFPREG_TYPE,
1811 sizeof(*info->xfpu), info->xfpu);
1819 static size_t get_note_info_size(struct elf_note_info *info)
1824 for (i = 0; i < info->numnote; i++)
1825 sz += notesize(info->notes + i);
1827 sz += info->thread_status_size;
1832 static int write_note_info(struct elf_note_info *info,
1833 struct file *file, loff_t *foffset)
1836 struct list_head *t;
1838 for (i = 0; i < info->numnote; i++)
1839 if (!writenote(info->notes + i, file, foffset))
1842 /* write out the thread status notes section */
1843 list_for_each(t, &info->thread_list) {
1844 struct elf_thread_status *tmp =
1845 list_entry(t, struct elf_thread_status, list);
1847 for (i = 0; i < tmp->num_notes; i++)
1848 if (!writenote(&tmp->notes[i], file, foffset))
1855 static void free_note_info(struct elf_note_info *info)
1857 while (!list_empty(&info->thread_list)) {
1858 struct list_head *tmp = info->thread_list.next;
1860 kfree(list_entry(tmp, struct elf_thread_status, list));
1863 kfree(info->prstatus);
1864 kfree(info->psinfo);
1867 #ifdef ELF_CORE_COPY_XFPREGS
1874 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1875 struct vm_area_struct *gate_vma)
1877 struct vm_area_struct *ret = tsk->mm->mmap;
1884 * Helper function for iterating across a vma list. It ensures that the caller
1885 * will visit `gate_vma' prior to terminating the search.
1887 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1888 struct vm_area_struct *gate_vma)
1890 struct vm_area_struct *ret;
1892 ret = this_vma->vm_next;
1895 if (this_vma == gate_vma)
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(long signr, struct pt_regs *regs, struct file *file, unsigned long limit)
1913 struct vm_area_struct *vma, *gate_vma;
1914 struct elfhdr *elf = NULL;
1915 loff_t offset = 0, dataoff, foffset;
1916 unsigned long mm_flags;
1917 struct elf_note_info info;
1920 * We no longer stop all VM operations.
1922 * This is because those proceses that could possibly change map_count
1923 * or the mmap / vma pages are now blocked in do_exit on current
1924 * finishing this core dump.
1926 * Only ptrace can touch these memory addresses, but it doesn't change
1927 * the map_count or the pages allocated. So no possibility of crashing
1928 * exists while dumping the mm->vm_next areas to the core file.
1931 /* alloc memory for large data structures: too large to be on stack */
1932 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1936 segs = current->mm->map_count;
1937 #ifdef ELF_CORE_EXTRA_PHDRS
1938 segs += ELF_CORE_EXTRA_PHDRS;
1941 gate_vma = get_gate_vma(current);
1942 if (gate_vma != NULL)
1946 * Collect all the non-memory information about the process for the
1947 * notes. This also sets up the file header.
1949 if (!fill_note_info(elf, segs + 1, /* including notes section */
1950 &info, signr, regs))
1954 current->flags |= PF_DUMPCORE;
1959 DUMP_WRITE(elf, sizeof(*elf));
1960 offset += sizeof(*elf); /* Elf header */
1961 offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
1964 /* Write notes phdr entry */
1966 struct elf_phdr phdr;
1967 size_t sz = get_note_info_size(&info);
1969 sz += elf_coredump_extra_notes_size();
1971 fill_elf_note_phdr(&phdr, sz, offset);
1973 DUMP_WRITE(&phdr, sizeof(phdr));
1976 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1979 * We must use the same mm->flags while dumping core to avoid
1980 * inconsistency between the program headers and bodies, otherwise an
1981 * unusable core file can be generated.
1983 mm_flags = current->mm->flags;
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, 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 DUMP_WRITE(&phdr, sizeof(phdr));
2007 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
2008 ELF_CORE_WRITE_EXTRA_PHDRS;
2011 /* write out the notes section */
2012 if (!write_note_info(&info, file, &foffset))
2015 if (elf_coredump_extra_notes_write(file, &foffset))
2019 DUMP_SEEK(dataoff - foffset);
2021 for (vma = first_vma(current, gate_vma); vma != NULL;
2022 vma = next_vma(vma, gate_vma)) {
2026 end = vma->vm_start + vma_dump_size(vma, mm_flags);
2028 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2030 struct vm_area_struct *tmp_vma;
2032 if (get_user_pages(current, current->mm, addr, 1, 0, 1,
2033 &page, &tmp_vma) <= 0) {
2034 DUMP_SEEK(PAGE_SIZE);
2036 if (page == ZERO_PAGE(0)) {
2037 if (!dump_seek(file, PAGE_SIZE)) {
2038 page_cache_release(page);
2043 flush_cache_page(tmp_vma, addr,
2046 if ((size += PAGE_SIZE) > limit ||
2047 !dump_write(file, kaddr,
2050 page_cache_release(page);
2055 page_cache_release(page);
2060 #ifdef ELF_CORE_WRITE_EXTRA_DATA
2061 ELF_CORE_WRITE_EXTRA_DATA;
2068 free_note_info(&info);
2074 #endif /* USE_ELF_CORE_DUMP */
2076 static int __init init_elf_binfmt(void)
2078 return register_binfmt(&elf_format);
2081 static void __exit exit_elf_binfmt(void)
2083 /* Remove the COFF and ELF loaders. */
2084 unregister_binfmt(&elf_format);
2087 core_initcall(init_elf_binfmt);
2088 module_exit(exit_elf_binfmt);
2089 MODULE_LICENSE("GPL");