1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/fs/binfmt_elf.c
5 * These are the functions used to load ELF format executables as used
6 * on SVr4 machines. Information on the format may be found in the book
7 * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
10 * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
13 #include <linux/module.h>
14 #include <linux/kernel.h>
16 #include <linux/log2.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/slab.h>
25 #include <linux/personality.h>
26 #include <linux/elfcore.h>
27 #include <linux/init.h>
28 #include <linux/highuid.h>
29 #include <linux/compiler.h>
30 #include <linux/highmem.h>
31 #include <linux/hugetlb.h>
32 #include <linux/pagemap.h>
33 #include <linux/vmalloc.h>
34 #include <linux/security.h>
35 #include <linux/random.h>
36 #include <linux/elf.h>
37 #include <linux/elf-randomize.h>
38 #include <linux/utsname.h>
39 #include <linux/coredump.h>
40 #include <linux/sched.h>
41 #include <linux/sched/coredump.h>
42 #include <linux/sched/task_stack.h>
43 #include <linux/sched/cputime.h>
44 #include <linux/sizes.h>
45 #include <linux/types.h>
46 #include <linux/cred.h>
47 #include <linux/dax.h>
48 #include <linux/uaccess.h>
49 #include <asm/param.h>
57 #define user_long_t long
59 #ifndef user_siginfo_t
60 #define user_siginfo_t siginfo_t
63 /* That's for binfmt_elf_fdpic to deal with */
64 #ifndef elf_check_fdpic
65 #define elf_check_fdpic(ex) false
68 static int load_elf_binary(struct linux_binprm *bprm);
71 static int load_elf_library(struct file *);
73 #define load_elf_library NULL
77 * If we don't support core dumping, then supply a NULL so we
80 #ifdef CONFIG_ELF_CORE
81 static int elf_core_dump(struct coredump_params *cprm);
83 #define elf_core_dump NULL
86 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
87 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
89 #define ELF_MIN_ALIGN PAGE_SIZE
92 #ifndef ELF_CORE_EFLAGS
93 #define ELF_CORE_EFLAGS 0
96 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
97 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
98 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
100 static struct linux_binfmt elf_format = {
101 .module = THIS_MODULE,
102 .load_binary = load_elf_binary,
103 .load_shlib = load_elf_library,
104 .core_dump = elf_core_dump,
105 .min_coredump = ELF_EXEC_PAGESIZE,
108 #define BAD_ADDR(x) (unlikely((unsigned long)(x) >= TASK_SIZE))
110 static int set_brk(unsigned long start, unsigned long end, int prot)
112 start = ELF_PAGEALIGN(start);
113 end = ELF_PAGEALIGN(end);
116 * Map the last of the bss segment.
117 * If the header is requesting these pages to be
118 * executable, honour that (ppc32 needs this).
120 int error = vm_brk_flags(start, end - start,
121 prot & PROT_EXEC ? VM_EXEC : 0);
125 current->mm->start_brk = current->mm->brk = end;
129 /* We need to explicitly zero any fractional pages
130 after the data section (i.e. bss). This would
131 contain the junk from the file that should not
134 static int padzero(unsigned long elf_bss)
138 nbyte = ELF_PAGEOFFSET(elf_bss);
140 nbyte = ELF_MIN_ALIGN - nbyte;
141 if (clear_user((void __user *) elf_bss, nbyte))
147 /* Let's use some macros to make this stack manipulation a little clearer */
148 #ifdef CONFIG_STACK_GROWSUP
149 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
150 #define STACK_ROUND(sp, items) \
151 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
152 #define STACK_ALLOC(sp, len) ({ \
153 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
156 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
157 #define STACK_ROUND(sp, items) \
158 (((unsigned long) (sp - items)) &~ 15UL)
159 #define STACK_ALLOC(sp, len) (sp -= len)
162 #ifndef ELF_BASE_PLATFORM
164 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
165 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
166 * will be copied to the user stack in the same manner as AT_PLATFORM.
168 #define ELF_BASE_PLATFORM NULL
172 create_elf_tables(struct linux_binprm *bprm, const struct elfhdr *exec,
173 unsigned long load_addr, unsigned long interp_load_addr,
174 unsigned long e_entry)
176 struct mm_struct *mm = current->mm;
177 unsigned long p = bprm->p;
178 int argc = bprm->argc;
179 int envc = bprm->envc;
180 elf_addr_t __user *sp;
181 elf_addr_t __user *u_platform;
182 elf_addr_t __user *u_base_platform;
183 elf_addr_t __user *u_rand_bytes;
184 const char *k_platform = ELF_PLATFORM;
185 const char *k_base_platform = ELF_BASE_PLATFORM;
186 unsigned char k_rand_bytes[16];
188 elf_addr_t *elf_info;
189 elf_addr_t flags = 0;
191 const struct cred *cred = current_cred();
192 struct vm_area_struct *vma;
195 * In some cases (e.g. Hyper-Threading), we want to avoid L1
196 * evictions by the processes running on the same package. One
197 * thing we can do is to shuffle the initial stack for them.
200 p = arch_align_stack(p);
203 * If this architecture has a platform capability string, copy it
204 * to userspace. In some cases (Sparc), this info is impossible
205 * for userspace to get any other way, in others (i386) it is
210 size_t len = strlen(k_platform) + 1;
212 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
213 if (copy_to_user(u_platform, k_platform, len))
218 * If this architecture has a "base" platform capability
219 * string, copy it to userspace.
221 u_base_platform = NULL;
222 if (k_base_platform) {
223 size_t len = strlen(k_base_platform) + 1;
225 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
226 if (copy_to_user(u_base_platform, k_base_platform, len))
231 * Generate 16 random bytes for userspace PRNG seeding.
233 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
234 u_rand_bytes = (elf_addr_t __user *)
235 STACK_ALLOC(p, sizeof(k_rand_bytes));
236 if (copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
239 /* Create the ELF interpreter info */
240 elf_info = (elf_addr_t *)mm->saved_auxv;
241 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
242 #define NEW_AUX_ENT(id, val) \
250 * ARCH_DLINFO must come first so PPC can do its special alignment of
252 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
253 * ARCH_DLINFO changes
257 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
258 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
259 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
260 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
261 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
262 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
263 NEW_AUX_ENT(AT_BASE, interp_load_addr);
264 if (bprm->interp_flags & BINPRM_FLAGS_PRESERVE_ARGV0)
265 flags |= AT_FLAGS_PRESERVE_ARGV0;
266 NEW_AUX_ENT(AT_FLAGS, flags);
267 NEW_AUX_ENT(AT_ENTRY, e_entry);
268 NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
269 NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
270 NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
271 NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
272 NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
273 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
275 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
277 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
279 NEW_AUX_ENT(AT_PLATFORM,
280 (elf_addr_t)(unsigned long)u_platform);
282 if (k_base_platform) {
283 NEW_AUX_ENT(AT_BASE_PLATFORM,
284 (elf_addr_t)(unsigned long)u_base_platform);
286 if (bprm->have_execfd) {
287 NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
290 /* AT_NULL is zero; clear the rest too */
291 memset(elf_info, 0, (char *)mm->saved_auxv +
292 sizeof(mm->saved_auxv) - (char *)elf_info);
294 /* And advance past the AT_NULL entry. */
297 ei_index = elf_info - (elf_addr_t *)mm->saved_auxv;
298 sp = STACK_ADD(p, ei_index);
300 items = (argc + 1) + (envc + 1) + 1;
301 bprm->p = STACK_ROUND(sp, items);
303 /* Point sp at the lowest address on the stack */
304 #ifdef CONFIG_STACK_GROWSUP
305 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
306 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
308 sp = (elf_addr_t __user *)bprm->p;
313 * Grow the stack manually; some architectures have a limit on how
314 * far ahead a user-space access may be in order to grow the stack.
316 if (mmap_read_lock_killable(mm))
318 vma = find_extend_vma(mm, bprm->p);
319 mmap_read_unlock(mm);
323 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
324 if (put_user(argc, sp++))
327 /* Populate list of argv pointers back to argv strings. */
328 p = mm->arg_end = mm->arg_start;
331 if (put_user((elf_addr_t)p, sp++))
333 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
334 if (!len || len > MAX_ARG_STRLEN)
338 if (put_user(0, sp++))
342 /* Populate list of envp pointers back to envp strings. */
343 mm->env_end = mm->env_start = p;
346 if (put_user((elf_addr_t)p, sp++))
348 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
349 if (!len || len > MAX_ARG_STRLEN)
353 if (put_user(0, sp++))
357 /* Put the elf_info on the stack in the right place. */
358 if (copy_to_user(sp, mm->saved_auxv, ei_index * sizeof(elf_addr_t)))
363 static unsigned long elf_map(struct file *filep, unsigned long addr,
364 const struct elf_phdr *eppnt, int prot, int type,
365 unsigned long total_size)
367 unsigned long map_addr;
368 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
369 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
370 addr = ELF_PAGESTART(addr);
371 size = ELF_PAGEALIGN(size);
373 /* mmap() will return -EINVAL if given a zero size, but a
374 * segment with zero filesize is perfectly valid */
379 * total_size is the size of the ELF (interpreter) image.
380 * The _first_ mmap needs to know the full size, otherwise
381 * randomization might put this image into an overlapping
382 * position with the ELF binary image. (since size < total_size)
383 * So we first map the 'big' image - and unmap the remainder at
384 * the end. (which unmap is needed for ELF images with holes.)
387 total_size = ELF_PAGEALIGN(total_size);
388 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
389 if (!BAD_ADDR(map_addr))
390 vm_munmap(map_addr+size, total_size-size);
392 map_addr = vm_mmap(filep, addr, size, prot, type, off);
394 if ((type & MAP_FIXED_NOREPLACE) &&
395 PTR_ERR((void *)map_addr) == -EEXIST)
396 pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
397 task_pid_nr(current), current->comm, (void *)addr);
402 static unsigned long total_mapping_size(const struct elf_phdr *cmds, int nr)
404 int i, first_idx = -1, last_idx = -1;
406 for (i = 0; i < nr; i++) {
407 if (cmds[i].p_type == PT_LOAD) {
416 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
417 ELF_PAGESTART(cmds[first_idx].p_vaddr);
420 static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
424 rv = kernel_read(file, buf, len, &pos);
425 if (unlikely(rv != len)) {
426 return (rv < 0) ? rv : -EIO;
431 static unsigned long maximum_alignment(struct elf_phdr *cmds, int nr)
433 unsigned long alignment = 0;
436 for (i = 0; i < nr; i++) {
437 if (cmds[i].p_type == PT_LOAD) {
438 unsigned long p_align = cmds[i].p_align;
440 /* skip non-power of two alignments as invalid */
441 if (!is_power_of_2(p_align))
443 alignment = max(alignment, p_align);
447 /* ensure we align to at least one page */
448 return ELF_PAGEALIGN(alignment);
452 * load_elf_phdrs() - load ELF program headers
453 * @elf_ex: ELF header of the binary whose program headers should be loaded
454 * @elf_file: the opened ELF binary file
456 * Loads ELF program headers from the binary file elf_file, which has the ELF
457 * header pointed to by elf_ex, into a newly allocated array. The caller is
458 * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
460 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
461 struct file *elf_file)
463 struct elf_phdr *elf_phdata = NULL;
464 int retval, err = -1;
468 * If the size of this structure has changed, then punt, since
469 * we will be doing the wrong thing.
471 if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
474 /* Sanity check the number of program headers... */
475 /* ...and their total size. */
476 size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
477 if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
480 elf_phdata = kmalloc(size, GFP_KERNEL);
484 /* Read in the program headers */
485 retval = elf_read(elf_file, elf_phdata, size, elf_ex->e_phoff);
501 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
504 * struct arch_elf_state - arch-specific ELF loading state
506 * This structure is used to preserve architecture specific data during
507 * the loading of an ELF file, throughout the checking of architecture
508 * specific ELF headers & through to the point where the ELF load is
509 * known to be proceeding (ie. SET_PERSONALITY).
511 * This implementation is a dummy for architectures which require no
514 struct arch_elf_state {
517 #define INIT_ARCH_ELF_STATE {}
520 * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
521 * @ehdr: The main ELF header
522 * @phdr: The program header to check
523 * @elf: The open ELF file
524 * @is_interp: True if the phdr is from the interpreter of the ELF being
525 * loaded, else false.
526 * @state: Architecture-specific state preserved throughout the process
527 * of loading the ELF.
529 * Inspects the program header phdr to validate its correctness and/or
530 * suitability for the system. Called once per ELF program header in the
531 * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
534 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
535 * with that return code.
537 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
538 struct elf_phdr *phdr,
539 struct file *elf, bool is_interp,
540 struct arch_elf_state *state)
542 /* Dummy implementation, always proceed */
547 * arch_check_elf() - check an ELF executable
548 * @ehdr: The main ELF header
549 * @has_interp: True if the ELF has an interpreter, else false.
550 * @interp_ehdr: The interpreter's ELF header
551 * @state: Architecture-specific state preserved throughout the process
552 * of loading the ELF.
554 * Provides a final opportunity for architecture code to reject the loading
555 * of the ELF & cause an exec syscall to return an error. This is called after
556 * all program headers to be checked by arch_elf_pt_proc have been.
558 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
559 * with that return code.
561 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
562 struct elfhdr *interp_ehdr,
563 struct arch_elf_state *state)
565 /* Dummy implementation, always proceed */
569 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
571 static inline int make_prot(u32 p_flags, struct arch_elf_state *arch_state,
572 bool has_interp, bool is_interp)
583 return arch_elf_adjust_prot(prot, arch_state, has_interp, is_interp);
586 /* This is much more generalized than the library routine read function,
587 so we keep this separate. Technically the library read function
588 is only provided so that we can read a.out libraries that have
591 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
592 struct file *interpreter,
593 unsigned long no_base, struct elf_phdr *interp_elf_phdata,
594 struct arch_elf_state *arch_state)
596 struct elf_phdr *eppnt;
597 unsigned long load_addr = 0;
598 int load_addr_set = 0;
599 unsigned long last_bss = 0, elf_bss = 0;
601 unsigned long error = ~0UL;
602 unsigned long total_size;
605 /* First of all, some simple consistency checks */
606 if (interp_elf_ex->e_type != ET_EXEC &&
607 interp_elf_ex->e_type != ET_DYN)
609 if (!elf_check_arch(interp_elf_ex) ||
610 elf_check_fdpic(interp_elf_ex))
612 if (!interpreter->f_op->mmap)
615 total_size = total_mapping_size(interp_elf_phdata,
616 interp_elf_ex->e_phnum);
622 eppnt = interp_elf_phdata;
623 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
624 if (eppnt->p_type == PT_LOAD) {
625 int elf_type = MAP_PRIVATE;
626 int elf_prot = make_prot(eppnt->p_flags, arch_state,
628 unsigned long vaddr = 0;
629 unsigned long k, map_addr;
631 vaddr = eppnt->p_vaddr;
632 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
633 elf_type |= MAP_FIXED;
634 else if (no_base && interp_elf_ex->e_type == ET_DYN)
637 map_addr = elf_map(interpreter, load_addr + vaddr,
638 eppnt, elf_prot, elf_type, total_size);
641 if (BAD_ADDR(map_addr))
644 if (!load_addr_set &&
645 interp_elf_ex->e_type == ET_DYN) {
646 load_addr = map_addr - ELF_PAGESTART(vaddr);
651 * Check to see if the section's size will overflow the
652 * allowed task size. Note that p_filesz must always be
653 * <= p_memsize so it's only necessary to check p_memsz.
655 k = load_addr + eppnt->p_vaddr;
657 eppnt->p_filesz > eppnt->p_memsz ||
658 eppnt->p_memsz > TASK_SIZE ||
659 TASK_SIZE - eppnt->p_memsz < k) {
665 * Find the end of the file mapping for this phdr, and
666 * keep track of the largest address we see for this.
668 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
673 * Do the same thing for the memory mapping - between
674 * elf_bss and last_bss is the bss section.
676 k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
685 * Now fill out the bss section: first pad the last page from
686 * the file up to the page boundary, and zero it from elf_bss
687 * up to the end of the page.
689 if (padzero(elf_bss)) {
694 * Next, align both the file and mem bss up to the page size,
695 * since this is where elf_bss was just zeroed up to, and where
696 * last_bss will end after the vm_brk_flags() below.
698 elf_bss = ELF_PAGEALIGN(elf_bss);
699 last_bss = ELF_PAGEALIGN(last_bss);
700 /* Finally, if there is still more bss to allocate, do it. */
701 if (last_bss > elf_bss) {
702 error = vm_brk_flags(elf_bss, last_bss - elf_bss,
703 bss_prot & PROT_EXEC ? VM_EXEC : 0);
714 * These are the functions used to load ELF style executables and shared
715 * libraries. There is no binary dependent code anywhere else.
718 static int parse_elf_property(const char *data, size_t *off, size_t datasz,
719 struct arch_elf_state *arch,
720 bool have_prev_type, u32 *prev_type)
723 const struct gnu_property *pr;
729 if (WARN_ON_ONCE(*off > datasz || *off % ELF_GNU_PROPERTY_ALIGN))
734 if (datasz < sizeof(*pr))
736 pr = (const struct gnu_property *)(data + o);
738 datasz -= sizeof(*pr);
740 if (pr->pr_datasz > datasz)
743 WARN_ON_ONCE(o % ELF_GNU_PROPERTY_ALIGN);
744 step = round_up(pr->pr_datasz, ELF_GNU_PROPERTY_ALIGN);
748 /* Properties are supposed to be unique and sorted on pr_type: */
749 if (have_prev_type && pr->pr_type <= *prev_type)
751 *prev_type = pr->pr_type;
753 ret = arch_parse_elf_property(pr->pr_type, data + o,
754 pr->pr_datasz, ELF_COMPAT, arch);
762 #define NOTE_DATA_SZ SZ_1K
763 #define GNU_PROPERTY_TYPE_0_NAME "GNU"
764 #define NOTE_NAME_SZ (sizeof(GNU_PROPERTY_TYPE_0_NAME))
766 static int parse_elf_properties(struct file *f, const struct elf_phdr *phdr,
767 struct arch_elf_state *arch)
770 struct elf_note nhdr;
771 char data[NOTE_DATA_SZ];
780 if (!IS_ENABLED(CONFIG_ARCH_USE_GNU_PROPERTY) || !phdr)
783 /* load_elf_binary() shouldn't call us unless this is true... */
784 if (WARN_ON_ONCE(phdr->p_type != PT_GNU_PROPERTY))
787 /* If the properties are crazy large, that's too bad (for now): */
788 if (phdr->p_filesz > sizeof(note))
791 pos = phdr->p_offset;
792 n = kernel_read(f, ¬e, phdr->p_filesz, &pos);
794 BUILD_BUG_ON(sizeof(note) < sizeof(note.nhdr) + NOTE_NAME_SZ);
795 if (n < 0 || n < sizeof(note.nhdr) + NOTE_NAME_SZ)
798 if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
799 note.nhdr.n_namesz != NOTE_NAME_SZ ||
800 strncmp(note.data + sizeof(note.nhdr),
801 GNU_PROPERTY_TYPE_0_NAME, n - sizeof(note.nhdr)))
804 off = round_up(sizeof(note.nhdr) + NOTE_NAME_SZ,
805 ELF_GNU_PROPERTY_ALIGN);
809 if (note.nhdr.n_descsz > n - off)
811 datasz = off + note.nhdr.n_descsz;
813 have_prev_type = false;
815 ret = parse_elf_property(note.data, &off, datasz, arch,
816 have_prev_type, &prev_type);
817 have_prev_type = true;
820 return ret == -ENOENT ? 0 : ret;
823 static int load_elf_binary(struct linux_binprm *bprm)
825 struct file *interpreter = NULL; /* to shut gcc up */
826 unsigned long load_addr = 0, load_bias = 0;
827 int load_addr_set = 0;
829 struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
830 struct elf_phdr *elf_property_phdata = NULL;
831 unsigned long elf_bss, elf_brk;
834 unsigned long elf_entry;
835 unsigned long e_entry;
836 unsigned long interp_load_addr = 0;
837 unsigned long start_code, end_code, start_data, end_data;
838 unsigned long reloc_func_desc __maybe_unused = 0;
839 int executable_stack = EXSTACK_DEFAULT;
840 struct elfhdr *elf_ex = (struct elfhdr *)bprm->buf;
841 struct elfhdr *interp_elf_ex = NULL;
842 struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
843 struct mm_struct *mm;
844 struct pt_regs *regs;
847 /* First of all, some simple consistency checks */
848 if (memcmp(elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
851 if (elf_ex->e_type != ET_EXEC && elf_ex->e_type != ET_DYN)
853 if (!elf_check_arch(elf_ex))
855 if (elf_check_fdpic(elf_ex))
857 if (!bprm->file->f_op->mmap)
860 elf_phdata = load_elf_phdrs(elf_ex, bprm->file);
864 elf_ppnt = elf_phdata;
865 for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++) {
866 char *elf_interpreter;
868 if (elf_ppnt->p_type == PT_GNU_PROPERTY) {
869 elf_property_phdata = elf_ppnt;
873 if (elf_ppnt->p_type != PT_INTERP)
877 * This is the program interpreter used for shared libraries -
878 * for now assume that this is an a.out format binary.
881 if (elf_ppnt->p_filesz > PATH_MAX || elf_ppnt->p_filesz < 2)
885 elf_interpreter = kmalloc(elf_ppnt->p_filesz, GFP_KERNEL);
886 if (!elf_interpreter)
889 retval = elf_read(bprm->file, elf_interpreter, elf_ppnt->p_filesz,
892 goto out_free_interp;
893 /* make sure path is NULL terminated */
895 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
896 goto out_free_interp;
898 interpreter = open_exec(elf_interpreter);
899 kfree(elf_interpreter);
900 retval = PTR_ERR(interpreter);
901 if (IS_ERR(interpreter))
905 * If the binary is not readable then enforce mm->dumpable = 0
906 * regardless of the interpreter's permissions.
908 would_dump(bprm, interpreter);
910 interp_elf_ex = kmalloc(sizeof(*interp_elf_ex), GFP_KERNEL);
911 if (!interp_elf_ex) {
916 /* Get the exec headers */
917 retval = elf_read(interpreter, interp_elf_ex,
918 sizeof(*interp_elf_ex), 0);
920 goto out_free_dentry;
925 kfree(elf_interpreter);
929 elf_ppnt = elf_phdata;
930 for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++)
931 switch (elf_ppnt->p_type) {
933 if (elf_ppnt->p_flags & PF_X)
934 executable_stack = EXSTACK_ENABLE_X;
936 executable_stack = EXSTACK_DISABLE_X;
939 case PT_LOPROC ... PT_HIPROC:
940 retval = arch_elf_pt_proc(elf_ex, elf_ppnt,
944 goto out_free_dentry;
948 /* Some simple consistency checks for the interpreter */
951 /* Not an ELF interpreter */
952 if (memcmp(interp_elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
953 goto out_free_dentry;
954 /* Verify the interpreter has a valid arch */
955 if (!elf_check_arch(interp_elf_ex) ||
956 elf_check_fdpic(interp_elf_ex))
957 goto out_free_dentry;
959 /* Load the interpreter program headers */
960 interp_elf_phdata = load_elf_phdrs(interp_elf_ex,
962 if (!interp_elf_phdata)
963 goto out_free_dentry;
965 /* Pass PT_LOPROC..PT_HIPROC headers to arch code */
966 elf_property_phdata = NULL;
967 elf_ppnt = interp_elf_phdata;
968 for (i = 0; i < interp_elf_ex->e_phnum; i++, elf_ppnt++)
969 switch (elf_ppnt->p_type) {
970 case PT_GNU_PROPERTY:
971 elf_property_phdata = elf_ppnt;
974 case PT_LOPROC ... PT_HIPROC:
975 retval = arch_elf_pt_proc(interp_elf_ex,
976 elf_ppnt, interpreter,
979 goto out_free_dentry;
984 retval = parse_elf_properties(interpreter ?: bprm->file,
985 elf_property_phdata, &arch_state);
987 goto out_free_dentry;
990 * Allow arch code to reject the ELF at this point, whilst it's
991 * still possible to return an error to the code that invoked
994 retval = arch_check_elf(elf_ex,
995 !!interpreter, interp_elf_ex,
998 goto out_free_dentry;
1000 /* Flush all traces of the currently running executable */
1001 retval = begin_new_exec(bprm);
1003 goto out_free_dentry;
1005 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
1006 may depend on the personality. */
1007 SET_PERSONALITY2(*elf_ex, &arch_state);
1008 if (elf_read_implies_exec(*elf_ex, executable_stack))
1009 current->personality |= READ_IMPLIES_EXEC;
1011 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1012 current->flags |= PF_RANDOMIZE;
1014 setup_new_exec(bprm);
1016 /* Do this so that we can load the interpreter, if need be. We will
1017 change some of these later */
1018 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
1021 goto out_free_dentry;
1031 /* Now we do a little grungy work by mmapping the ELF image into
1032 the correct location in memory. */
1033 for(i = 0, elf_ppnt = elf_phdata;
1034 i < elf_ex->e_phnum; i++, elf_ppnt++) {
1035 int elf_prot, elf_flags;
1036 unsigned long k, vaddr;
1037 unsigned long total_size = 0;
1038 unsigned long alignment;
1040 if (elf_ppnt->p_type != PT_LOAD)
1043 if (unlikely (elf_brk > elf_bss)) {
1044 unsigned long nbyte;
1046 /* There was a PT_LOAD segment with p_memsz > p_filesz
1047 before this one. Map anonymous pages, if needed,
1048 and clear the area. */
1049 retval = set_brk(elf_bss + load_bias,
1050 elf_brk + load_bias,
1053 goto out_free_dentry;
1054 nbyte = ELF_PAGEOFFSET(elf_bss);
1056 nbyte = ELF_MIN_ALIGN - nbyte;
1057 if (nbyte > elf_brk - elf_bss)
1058 nbyte = elf_brk - elf_bss;
1059 if (clear_user((void __user *)elf_bss +
1060 load_bias, nbyte)) {
1062 * This bss-zeroing can fail if the ELF
1063 * file specifies odd protections. So
1064 * we don't check the return value
1070 elf_prot = make_prot(elf_ppnt->p_flags, &arch_state,
1071 !!interpreter, false);
1073 elf_flags = MAP_PRIVATE;
1075 vaddr = elf_ppnt->p_vaddr;
1077 * The first time through the loop, load_addr_set is false:
1078 * layout will be calculated. Once set, use MAP_FIXED since
1079 * we know we've already safely mapped the entire region with
1080 * MAP_FIXED_NOREPLACE in the once-per-binary logic following.
1082 if (load_addr_set) {
1083 elf_flags |= MAP_FIXED;
1084 } else if (elf_ex->e_type == ET_EXEC) {
1086 * This logic is run once for the first LOAD Program
1087 * Header for ET_EXEC binaries. No special handling
1090 elf_flags |= MAP_FIXED_NOREPLACE;
1091 } else if (elf_ex->e_type == ET_DYN) {
1093 * This logic is run once for the first LOAD Program
1094 * Header for ET_DYN binaries to calculate the
1095 * randomization (load_bias) for all the LOAD
1098 * There are effectively two types of ET_DYN
1099 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
1100 * and loaders (ET_DYN without INTERP, since they
1101 * _are_ the ELF interpreter). The loaders must
1102 * be loaded away from programs since the program
1103 * may otherwise collide with the loader (especially
1104 * for ET_EXEC which does not have a randomized
1105 * position). For example to handle invocations of
1106 * "./ld.so someprog" to test out a new version of
1107 * the loader, the subsequent program that the
1108 * loader loads must avoid the loader itself, so
1109 * they cannot share the same load range. Sufficient
1110 * room for the brk must be allocated with the
1111 * loader as well, since brk must be available with
1114 * Therefore, programs are loaded offset from
1115 * ELF_ET_DYN_BASE and loaders are loaded into the
1116 * independently randomized mmap region (0 load_bias
1117 * without MAP_FIXED nor MAP_FIXED_NOREPLACE).
1119 alignment = maximum_alignment(elf_phdata, elf_ex->e_phnum);
1120 if (alignment > ELF_MIN_ALIGN) {
1121 load_bias = ELF_ET_DYN_BASE;
1122 if (current->flags & PF_RANDOMIZE)
1123 load_bias += arch_mmap_rnd();
1125 load_bias &= ~(alignment - 1);
1126 elf_flags |= MAP_FIXED_NOREPLACE;
1131 * Since load_bias is used for all subsequent loading
1132 * calculations, we must lower it by the first vaddr
1133 * so that the remaining calculations based on the
1134 * ELF vaddrs will be correctly offset. The result
1135 * is then page aligned.
1137 load_bias = ELF_PAGESTART(load_bias - vaddr);
1141 * Calculate the entire size of the ELF mapping (total_size).
1142 * (Note that load_addr_set is set to true later once the
1143 * initial mapping is performed.)
1145 if (!load_addr_set) {
1146 total_size = total_mapping_size(elf_phdata,
1150 goto out_free_dentry;
1154 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1155 elf_prot, elf_flags, total_size);
1156 if (BAD_ADDR(error)) {
1157 retval = IS_ERR((void *)error) ?
1158 PTR_ERR((void*)error) : -EINVAL;
1159 goto out_free_dentry;
1162 if (!load_addr_set) {
1164 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1165 if (elf_ex->e_type == ET_DYN) {
1166 load_bias += error -
1167 ELF_PAGESTART(load_bias + vaddr);
1168 load_addr += load_bias;
1169 reloc_func_desc = load_bias;
1172 k = elf_ppnt->p_vaddr;
1173 if ((elf_ppnt->p_flags & PF_X) && k < start_code)
1179 * Check to see if the section's size will overflow the
1180 * allowed task size. Note that p_filesz must always be
1181 * <= p_memsz so it is only necessary to check p_memsz.
1183 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1184 elf_ppnt->p_memsz > TASK_SIZE ||
1185 TASK_SIZE - elf_ppnt->p_memsz < k) {
1186 /* set_brk can never work. Avoid overflows. */
1188 goto out_free_dentry;
1191 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1195 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1199 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1201 bss_prot = elf_prot;
1206 e_entry = elf_ex->e_entry + load_bias;
1207 elf_bss += load_bias;
1208 elf_brk += load_bias;
1209 start_code += load_bias;
1210 end_code += load_bias;
1211 start_data += load_bias;
1212 end_data += load_bias;
1214 /* Calling set_brk effectively mmaps the pages that we need
1215 * for the bss and break sections. We must do this before
1216 * mapping in the interpreter, to make sure it doesn't wind
1217 * up getting placed where the bss needs to go.
1219 retval = set_brk(elf_bss, elf_brk, bss_prot);
1221 goto out_free_dentry;
1222 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1223 retval = -EFAULT; /* Nobody gets to see this, but.. */
1224 goto out_free_dentry;
1228 elf_entry = load_elf_interp(interp_elf_ex,
1230 load_bias, interp_elf_phdata,
1232 if (!IS_ERR((void *)elf_entry)) {
1234 * load_elf_interp() returns relocation
1237 interp_load_addr = elf_entry;
1238 elf_entry += interp_elf_ex->e_entry;
1240 if (BAD_ADDR(elf_entry)) {
1241 retval = IS_ERR((void *)elf_entry) ?
1242 (int)elf_entry : -EINVAL;
1243 goto out_free_dentry;
1245 reloc_func_desc = interp_load_addr;
1247 allow_write_access(interpreter);
1250 kfree(interp_elf_ex);
1251 kfree(interp_elf_phdata);
1253 elf_entry = e_entry;
1254 if (BAD_ADDR(elf_entry)) {
1256 goto out_free_dentry;
1262 set_binfmt(&elf_format);
1264 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1265 retval = ARCH_SETUP_ADDITIONAL_PAGES(bprm, elf_ex, !!interpreter);
1268 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1270 retval = create_elf_tables(bprm, elf_ex,
1271 load_addr, interp_load_addr, e_entry);
1276 mm->end_code = end_code;
1277 mm->start_code = start_code;
1278 mm->start_data = start_data;
1279 mm->end_data = end_data;
1280 mm->start_stack = bprm->p;
1282 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1284 * For architectures with ELF randomization, when executing
1285 * a loader directly (i.e. no interpreter listed in ELF
1286 * headers), move the brk area out of the mmap region
1287 * (since it grows up, and may collide early with the stack
1288 * growing down), and into the unused ELF_ET_DYN_BASE region.
1290 if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) &&
1291 elf_ex->e_type == ET_DYN && !interpreter) {
1292 mm->brk = mm->start_brk = ELF_ET_DYN_BASE;
1295 mm->brk = mm->start_brk = arch_randomize_brk(mm);
1296 #ifdef compat_brk_randomized
1297 current->brk_randomized = 1;
1301 if (current->personality & MMAP_PAGE_ZERO) {
1302 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1303 and some applications "depend" upon this behavior.
1304 Since we do not have the power to recompile these, we
1305 emulate the SVr4 behavior. Sigh. */
1306 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1307 MAP_FIXED | MAP_PRIVATE, 0);
1310 regs = current_pt_regs();
1311 #ifdef ELF_PLAT_INIT
1313 * The ABI may specify that certain registers be set up in special
1314 * ways (on i386 %edx is the address of a DT_FINI function, for
1315 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1316 * that the e_entry field is the address of the function descriptor
1317 * for the startup routine, rather than the address of the startup
1318 * routine itself. This macro performs whatever initialization to
1319 * the regs structure is required as well as any relocations to the
1320 * function descriptor entries when executing dynamically links apps.
1322 ELF_PLAT_INIT(regs, reloc_func_desc);
1325 finalize_exec(bprm);
1326 START_THREAD(elf_ex, regs, elf_entry, bprm->p);
1333 kfree(interp_elf_ex);
1334 kfree(interp_elf_phdata);
1335 allow_write_access(interpreter);
1343 #ifdef CONFIG_USELIB
1344 /* This is really simpleminded and specialized - we are loading an
1345 a.out library that is given an ELF header. */
1346 static int load_elf_library(struct file *file)
1348 struct elf_phdr *elf_phdata;
1349 struct elf_phdr *eppnt;
1350 unsigned long elf_bss, bss, len;
1351 int retval, error, i, j;
1352 struct elfhdr elf_ex;
1355 retval = elf_read(file, &elf_ex, sizeof(elf_ex), 0);
1359 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1362 /* First of all, some simple consistency checks */
1363 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1364 !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1366 if (elf_check_fdpic(&elf_ex))
1369 /* Now read in all of the header information */
1371 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1372 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1375 elf_phdata = kmalloc(j, GFP_KERNEL);
1381 retval = elf_read(file, eppnt, j, elf_ex.e_phoff);
1385 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1386 if ((eppnt + i)->p_type == PT_LOAD)
1391 while (eppnt->p_type != PT_LOAD)
1394 /* Now use mmap to map the library into memory. */
1395 error = vm_mmap(file,
1396 ELF_PAGESTART(eppnt->p_vaddr),
1398 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1399 PROT_READ | PROT_WRITE | PROT_EXEC,
1400 MAP_FIXED_NOREPLACE | MAP_PRIVATE,
1402 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1403 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1406 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1407 if (padzero(elf_bss)) {
1412 len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1413 bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1415 error = vm_brk(len, bss - len);
1426 #endif /* #ifdef CONFIG_USELIB */
1428 #ifdef CONFIG_ELF_CORE
1432 * Modelled on fs/exec.c:aout_core_dump()
1433 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1436 /* An ELF note in memory */
1441 unsigned int datasz;
1445 static int notesize(struct memelfnote *en)
1449 sz = sizeof(struct elf_note);
1450 sz += roundup(strlen(en->name) + 1, 4);
1451 sz += roundup(en->datasz, 4);
1456 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1459 en.n_namesz = strlen(men->name) + 1;
1460 en.n_descsz = men->datasz;
1461 en.n_type = men->type;
1463 return dump_emit(cprm, &en, sizeof(en)) &&
1464 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1465 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1468 static void fill_elf_header(struct elfhdr *elf, int segs,
1469 u16 machine, u32 flags)
1471 memset(elf, 0, sizeof(*elf));
1473 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1474 elf->e_ident[EI_CLASS] = ELF_CLASS;
1475 elf->e_ident[EI_DATA] = ELF_DATA;
1476 elf->e_ident[EI_VERSION] = EV_CURRENT;
1477 elf->e_ident[EI_OSABI] = ELF_OSABI;
1479 elf->e_type = ET_CORE;
1480 elf->e_machine = machine;
1481 elf->e_version = EV_CURRENT;
1482 elf->e_phoff = sizeof(struct elfhdr);
1483 elf->e_flags = flags;
1484 elf->e_ehsize = sizeof(struct elfhdr);
1485 elf->e_phentsize = sizeof(struct elf_phdr);
1486 elf->e_phnum = segs;
1489 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1491 phdr->p_type = PT_NOTE;
1492 phdr->p_offset = offset;
1495 phdr->p_filesz = sz;
1501 static void fill_note(struct memelfnote *note, const char *name, int type,
1502 unsigned int sz, void *data)
1511 * fill up all the fields in prstatus from the given task struct, except
1512 * registers which need to be filled up separately.
1514 static void fill_prstatus(struct elf_prstatus_common *prstatus,
1515 struct task_struct *p, long signr)
1517 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1518 prstatus->pr_sigpend = p->pending.signal.sig[0];
1519 prstatus->pr_sighold = p->blocked.sig[0];
1521 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1523 prstatus->pr_pid = task_pid_vnr(p);
1524 prstatus->pr_pgrp = task_pgrp_vnr(p);
1525 prstatus->pr_sid = task_session_vnr(p);
1526 if (thread_group_leader(p)) {
1527 struct task_cputime cputime;
1530 * This is the record for the group leader. It shows the
1531 * group-wide total, not its individual thread total.
1533 thread_group_cputime(p, &cputime);
1534 prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1535 prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1539 task_cputime(p, &utime, &stime);
1540 prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1541 prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1544 prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1545 prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1548 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1549 struct mm_struct *mm)
1551 const struct cred *cred;
1552 unsigned int i, len;
1555 /* first copy the parameters from user space */
1556 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1558 len = mm->arg_end - mm->arg_start;
1559 if (len >= ELF_PRARGSZ)
1560 len = ELF_PRARGSZ-1;
1561 if (copy_from_user(&psinfo->pr_psargs,
1562 (const char __user *)mm->arg_start, len))
1564 for(i = 0; i < len; i++)
1565 if (psinfo->pr_psargs[i] == 0)
1566 psinfo->pr_psargs[i] = ' ';
1567 psinfo->pr_psargs[len] = 0;
1570 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1572 psinfo->pr_pid = task_pid_vnr(p);
1573 psinfo->pr_pgrp = task_pgrp_vnr(p);
1574 psinfo->pr_sid = task_session_vnr(p);
1576 state = READ_ONCE(p->__state);
1577 i = state ? ffz(~state) + 1 : 0;
1578 psinfo->pr_state = i;
1579 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1580 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1581 psinfo->pr_nice = task_nice(p);
1582 psinfo->pr_flag = p->flags;
1584 cred = __task_cred(p);
1585 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1586 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1588 get_task_comm(psinfo->pr_fname, p);
1593 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1595 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1599 while (auxv[i - 2] != AT_NULL);
1600 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1603 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1604 const kernel_siginfo_t *siginfo)
1606 copy_siginfo_to_external(csigdata, siginfo);
1607 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1610 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1612 * Format of NT_FILE note:
1614 * long count -- how many files are mapped
1615 * long page_size -- units for file_ofs
1616 * array of [COUNT] elements of
1620 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1622 static int fill_files_note(struct memelfnote *note)
1624 struct mm_struct *mm = current->mm;
1625 struct vm_area_struct *vma;
1626 unsigned count, size, names_ofs, remaining, n;
1628 user_long_t *start_end_ofs;
1629 char *name_base, *name_curpos;
1631 /* *Estimated* file count and total data size needed */
1632 count = mm->map_count;
1633 if (count > UINT_MAX / 64)
1637 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1639 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1641 size = round_up(size, PAGE_SIZE);
1643 * "size" can be 0 here legitimately.
1644 * Let it ENOMEM and omit NT_FILE section which will be empty anyway.
1646 data = kvmalloc(size, GFP_KERNEL);
1647 if (ZERO_OR_NULL_PTR(data))
1650 start_end_ofs = data + 2;
1651 name_base = name_curpos = ((char *)data) + names_ofs;
1652 remaining = size - names_ofs;
1654 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
1656 const char *filename;
1658 file = vma->vm_file;
1661 filename = file_path(file, name_curpos, remaining);
1662 if (IS_ERR(filename)) {
1663 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1665 size = size * 5 / 4;
1671 /* file_path() fills at the end, move name down */
1672 /* n = strlen(filename) + 1: */
1673 n = (name_curpos + remaining) - filename;
1674 remaining = filename - name_curpos;
1675 memmove(name_curpos, filename, n);
1678 *start_end_ofs++ = vma->vm_start;
1679 *start_end_ofs++ = vma->vm_end;
1680 *start_end_ofs++ = vma->vm_pgoff;
1684 /* Now we know exact count of files, can store it */
1686 data[1] = PAGE_SIZE;
1688 * Count usually is less than mm->map_count,
1689 * we need to move filenames down.
1691 n = mm->map_count - count;
1693 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1694 memmove(name_base - shift_bytes, name_base,
1695 name_curpos - name_base);
1696 name_curpos -= shift_bytes;
1699 size = name_curpos - (char *)data;
1700 fill_note(note, "CORE", NT_FILE, size, data);
1704 #ifdef CORE_DUMP_USE_REGSET
1705 #include <linux/regset.h>
1707 struct elf_thread_core_info {
1708 struct elf_thread_core_info *next;
1709 struct task_struct *task;
1710 struct elf_prstatus prstatus;
1711 struct memelfnote notes[];
1714 struct elf_note_info {
1715 struct elf_thread_core_info *thread;
1716 struct memelfnote psinfo;
1717 struct memelfnote signote;
1718 struct memelfnote auxv;
1719 struct memelfnote files;
1720 user_siginfo_t csigdata;
1726 * When a regset has a writeback hook, we call it on each thread before
1727 * dumping user memory. On register window machines, this makes sure the
1728 * user memory backing the register data is up to date before we read it.
1730 static void do_thread_regset_writeback(struct task_struct *task,
1731 const struct user_regset *regset)
1733 if (regset->writeback)
1734 regset->writeback(task, regset, 1);
1737 #ifndef PRSTATUS_SIZE
1738 #define PRSTATUS_SIZE sizeof(struct elf_prstatus)
1741 #ifndef SET_PR_FPVALID
1742 #define SET_PR_FPVALID(S) ((S)->pr_fpvalid = 1)
1745 static int fill_thread_core_info(struct elf_thread_core_info *t,
1746 const struct user_regset_view *view,
1747 long signr, size_t *total)
1752 * NT_PRSTATUS is the one special case, because the regset data
1753 * goes into the pr_reg field inside the note contents, rather
1754 * than being the whole note contents. We fill the reset in here.
1755 * We assume that regset 0 is NT_PRSTATUS.
1757 fill_prstatus(&t->prstatus.common, t->task, signr);
1758 regset_get(t->task, &view->regsets[0],
1759 sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1761 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1762 PRSTATUS_SIZE, &t->prstatus);
1763 *total += notesize(&t->notes[0]);
1765 do_thread_regset_writeback(t->task, &view->regsets[0]);
1768 * Each other regset might generate a note too. For each regset
1769 * that has no core_note_type or is inactive, we leave t->notes[i]
1770 * all zero and we'll know to skip writing it later.
1772 for (i = 1; i < view->n; ++i) {
1773 const struct user_regset *regset = &view->regsets[i];
1774 int note_type = regset->core_note_type;
1775 bool is_fpreg = note_type == NT_PRFPREG;
1779 do_thread_regset_writeback(t->task, regset);
1780 if (!note_type) // not for coredumps
1782 if (regset->active && regset->active(t->task, regset) <= 0)
1785 ret = regset_get_alloc(t->task, regset, ~0U, &data);
1790 SET_PR_FPVALID(&t->prstatus);
1792 fill_note(&t->notes[i], is_fpreg ? "CORE" : "LINUX",
1793 note_type, ret, data);
1795 *total += notesize(&t->notes[i]);
1801 static int fill_note_info(struct elfhdr *elf, int phdrs,
1802 struct elf_note_info *info,
1803 const kernel_siginfo_t *siginfo, struct pt_regs *regs)
1805 struct task_struct *dump_task = current;
1806 const struct user_regset_view *view = task_user_regset_view(dump_task);
1807 struct elf_thread_core_info *t;
1808 struct elf_prpsinfo *psinfo;
1809 struct core_thread *ct;
1813 info->thread = NULL;
1815 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1816 if (psinfo == NULL) {
1817 info->psinfo.data = NULL; /* So we don't free this wrongly */
1821 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1824 * Figure out how many notes we're going to need for each thread.
1826 info->thread_notes = 0;
1827 for (i = 0; i < view->n; ++i)
1828 if (view->regsets[i].core_note_type != 0)
1829 ++info->thread_notes;
1832 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1833 * since it is our one special case.
1835 if (unlikely(info->thread_notes == 0) ||
1836 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1842 * Initialize the ELF file header.
1844 fill_elf_header(elf, phdrs,
1845 view->e_machine, view->e_flags);
1848 * Allocate a structure for each thread.
1850 for (ct = &dump_task->signal->core_state->dumper; ct; ct = ct->next) {
1851 t = kzalloc(offsetof(struct elf_thread_core_info,
1852 notes[info->thread_notes]),
1858 if (ct->task == dump_task || !info->thread) {
1859 t->next = info->thread;
1863 * Make sure to keep the original task at
1864 * the head of the list.
1866 t->next = info->thread->next;
1867 info->thread->next = t;
1872 * Now fill in each thread's information.
1874 for (t = info->thread; t != NULL; t = t->next)
1875 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1879 * Fill in the two process-wide notes.
1881 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1882 info->size += notesize(&info->psinfo);
1884 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1885 info->size += notesize(&info->signote);
1887 fill_auxv_note(&info->auxv, current->mm);
1888 info->size += notesize(&info->auxv);
1890 if (fill_files_note(&info->files) == 0)
1891 info->size += notesize(&info->files);
1896 static size_t get_note_info_size(struct elf_note_info *info)
1902 * Write all the notes for each thread. When writing the first thread, the
1903 * process-wide notes are interleaved after the first thread-specific note.
1905 static int write_note_info(struct elf_note_info *info,
1906 struct coredump_params *cprm)
1909 struct elf_thread_core_info *t = info->thread;
1914 if (!writenote(&t->notes[0], cprm))
1917 if (first && !writenote(&info->psinfo, cprm))
1919 if (first && !writenote(&info->signote, cprm))
1921 if (first && !writenote(&info->auxv, cprm))
1923 if (first && info->files.data &&
1924 !writenote(&info->files, cprm))
1927 for (i = 1; i < info->thread_notes; ++i)
1928 if (t->notes[i].data &&
1929 !writenote(&t->notes[i], cprm))
1939 static void free_note_info(struct elf_note_info *info)
1941 struct elf_thread_core_info *threads = info->thread;
1944 struct elf_thread_core_info *t = threads;
1946 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1947 for (i = 1; i < info->thread_notes; ++i)
1948 kfree(t->notes[i].data);
1951 kfree(info->psinfo.data);
1952 kvfree(info->files.data);
1957 /* Here is the structure in which status of each thread is captured. */
1958 struct elf_thread_status
1960 struct list_head list;
1961 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1962 elf_fpregset_t fpu; /* NT_PRFPREG */
1963 struct task_struct *thread;
1964 struct memelfnote notes[3];
1969 * In order to add the specific thread information for the elf file format,
1970 * we need to keep a linked list of every threads pr_status and then create
1971 * a single section for them in the final core file.
1973 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1976 struct task_struct *p = t->thread;
1979 fill_prstatus(&t->prstatus.common, p, signr);
1980 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1982 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1985 sz += notesize(&t->notes[0]);
1987 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1989 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1992 sz += notesize(&t->notes[1]);
1997 struct elf_note_info {
1998 struct memelfnote *notes;
1999 struct memelfnote *notes_files;
2000 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
2001 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
2002 struct list_head thread_list;
2003 elf_fpregset_t *fpu;
2004 user_siginfo_t csigdata;
2005 int thread_status_size;
2009 static int elf_note_info_init(struct elf_note_info *info)
2011 memset(info, 0, sizeof(*info));
2012 INIT_LIST_HEAD(&info->thread_list);
2014 /* Allocate space for ELF notes */
2015 info->notes = kmalloc_array(8, sizeof(struct memelfnote), GFP_KERNEL);
2018 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2021 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2022 if (!info->prstatus)
2024 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2030 static int fill_note_info(struct elfhdr *elf, int phdrs,
2031 struct elf_note_info *info,
2032 const kernel_siginfo_t *siginfo, struct pt_regs *regs)
2034 struct core_thread *ct;
2035 struct elf_thread_status *ets;
2037 if (!elf_note_info_init(info))
2040 for (ct = current->signal->core_state->dumper.next;
2041 ct; ct = ct->next) {
2042 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2046 ets->thread = ct->task;
2047 list_add(&ets->list, &info->thread_list);
2050 list_for_each_entry(ets, &info->thread_list, list) {
2053 sz = elf_dump_thread_status(siginfo->si_signo, ets);
2054 info->thread_status_size += sz;
2056 /* now collect the dump for the current */
2057 memset(info->prstatus, 0, sizeof(*info->prstatus));
2058 fill_prstatus(&info->prstatus->common, current, siginfo->si_signo);
2059 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2062 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2065 * Set up the notes in similar form to SVR4 core dumps made
2066 * with info from their /proc.
2069 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2070 sizeof(*info->prstatus), info->prstatus);
2071 fill_psinfo(info->psinfo, current->group_leader, current->mm);
2072 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2073 sizeof(*info->psinfo), info->psinfo);
2075 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2076 fill_auxv_note(info->notes + 3, current->mm);
2079 if (fill_files_note(info->notes + info->numnote) == 0) {
2080 info->notes_files = info->notes + info->numnote;
2084 /* Try to dump the FPU. */
2085 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2087 if (info->prstatus->pr_fpvalid)
2088 fill_note(info->notes + info->numnote++,
2089 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2093 static size_t get_note_info_size(struct elf_note_info *info)
2098 for (i = 0; i < info->numnote; i++)
2099 sz += notesize(info->notes + i);
2101 sz += info->thread_status_size;
2106 static int write_note_info(struct elf_note_info *info,
2107 struct coredump_params *cprm)
2109 struct elf_thread_status *ets;
2112 for (i = 0; i < info->numnote; i++)
2113 if (!writenote(info->notes + i, cprm))
2116 /* write out the thread status notes section */
2117 list_for_each_entry(ets, &info->thread_list, list) {
2118 for (i = 0; i < ets->num_notes; i++)
2119 if (!writenote(&ets->notes[i], cprm))
2126 static void free_note_info(struct elf_note_info *info)
2128 while (!list_empty(&info->thread_list)) {
2129 struct list_head *tmp = info->thread_list.next;
2131 kfree(list_entry(tmp, struct elf_thread_status, list));
2134 /* Free data possibly allocated by fill_files_note(): */
2135 if (info->notes_files)
2136 kvfree(info->notes_files->data);
2138 kfree(info->prstatus);
2139 kfree(info->psinfo);
2146 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2147 elf_addr_t e_shoff, int segs)
2149 elf->e_shoff = e_shoff;
2150 elf->e_shentsize = sizeof(*shdr4extnum);
2152 elf->e_shstrndx = SHN_UNDEF;
2154 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2156 shdr4extnum->sh_type = SHT_NULL;
2157 shdr4extnum->sh_size = elf->e_shnum;
2158 shdr4extnum->sh_link = elf->e_shstrndx;
2159 shdr4extnum->sh_info = segs;
2165 * This is a two-pass process; first we find the offsets of the bits,
2166 * and then they are actually written out. If we run out of core limit
2169 static int elf_core_dump(struct coredump_params *cprm)
2172 int vma_count, segs, i;
2173 size_t vma_data_size;
2175 loff_t offset = 0, dataoff;
2176 struct elf_note_info info = { };
2177 struct elf_phdr *phdr4note = NULL;
2178 struct elf_shdr *shdr4extnum = NULL;
2181 struct core_vma_metadata *vma_meta;
2183 if (dump_vma_snapshot(cprm, &vma_count, &vma_meta, &vma_data_size))
2187 * The number of segs are recored into ELF header as 16bit value.
2188 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2190 segs = vma_count + elf_core_extra_phdrs();
2192 /* for notes section */
2195 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2196 * this, kernel supports extended numbering. Have a look at
2197 * include/linux/elf.h for further information. */
2198 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2201 * Collect all the non-memory information about the process for the
2202 * notes. This also sets up the file header.
2204 if (!fill_note_info(&elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2209 offset += sizeof(elf); /* Elf header */
2210 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2212 /* Write notes phdr entry */
2214 size_t sz = get_note_info_size(&info);
2216 /* For cell spufs */
2217 sz += elf_coredump_extra_notes_size();
2219 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2223 fill_elf_note_phdr(phdr4note, sz, offset);
2227 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2229 offset += vma_data_size;
2230 offset += elf_core_extra_data_size();
2233 if (e_phnum == PN_XNUM) {
2234 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2237 fill_extnum_info(&elf, shdr4extnum, e_shoff, segs);
2242 if (!dump_emit(cprm, &elf, sizeof(elf)))
2245 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2248 /* Write program headers for segments dump */
2249 for (i = 0; i < vma_count; i++) {
2250 struct core_vma_metadata *meta = vma_meta + i;
2251 struct elf_phdr phdr;
2253 phdr.p_type = PT_LOAD;
2254 phdr.p_offset = offset;
2255 phdr.p_vaddr = meta->start;
2257 phdr.p_filesz = meta->dump_size;
2258 phdr.p_memsz = meta->end - meta->start;
2259 offset += phdr.p_filesz;
2261 if (meta->flags & VM_READ)
2262 phdr.p_flags |= PF_R;
2263 if (meta->flags & VM_WRITE)
2264 phdr.p_flags |= PF_W;
2265 if (meta->flags & VM_EXEC)
2266 phdr.p_flags |= PF_X;
2267 phdr.p_align = ELF_EXEC_PAGESIZE;
2269 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2273 if (!elf_core_write_extra_phdrs(cprm, offset))
2276 /* write out the notes section */
2277 if (!write_note_info(&info, cprm))
2280 /* For cell spufs */
2281 if (elf_coredump_extra_notes_write(cprm))
2285 dump_skip_to(cprm, dataoff);
2287 for (i = 0; i < vma_count; i++) {
2288 struct core_vma_metadata *meta = vma_meta + i;
2290 if (!dump_user_range(cprm, meta->start, meta->dump_size))
2294 if (!elf_core_write_extra_data(cprm))
2297 if (e_phnum == PN_XNUM) {
2298 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2303 free_note_info(&info);
2310 #endif /* CONFIG_ELF_CORE */
2312 static int __init init_elf_binfmt(void)
2314 register_binfmt(&elf_format);
2318 static void __exit exit_elf_binfmt(void)
2320 /* Remove the COFF and ELF loaders. */
2321 unregister_binfmt(&elf_format);
2324 core_initcall(init_elf_binfmt);
2325 module_exit(exit_elf_binfmt);
2326 MODULE_LICENSE("GPL");