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 ; sp; })
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 * If we are loading ET_EXEC or we have already performed
1078 * the ET_DYN load_addr calculations, proceed normally.
1080 if (elf_ex->e_type == ET_EXEC || load_addr_set) {
1081 elf_flags |= MAP_FIXED;
1082 } else if (elf_ex->e_type == ET_DYN) {
1084 * This logic is run once for the first LOAD Program
1085 * Header for ET_DYN binaries to calculate the
1086 * randomization (load_bias) for all the LOAD
1087 * Program Headers, and to calculate the entire
1088 * size of the ELF mapping (total_size). (Note that
1089 * load_addr_set is set to true later once the
1090 * initial mapping is performed.)
1092 * There are effectively two types of ET_DYN
1093 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
1094 * and loaders (ET_DYN without INTERP, since they
1095 * _are_ the ELF interpreter). The loaders must
1096 * be loaded away from programs since the program
1097 * may otherwise collide with the loader (especially
1098 * for ET_EXEC which does not have a randomized
1099 * position). For example to handle invocations of
1100 * "./ld.so someprog" to test out a new version of
1101 * the loader, the subsequent program that the
1102 * loader loads must avoid the loader itself, so
1103 * they cannot share the same load range. Sufficient
1104 * room for the brk must be allocated with the
1105 * loader as well, since brk must be available with
1108 * Therefore, programs are loaded offset from
1109 * ELF_ET_DYN_BASE and loaders are loaded into the
1110 * independently randomized mmap region (0 load_bias
1111 * without MAP_FIXED).
1114 load_bias = ELF_ET_DYN_BASE;
1115 if (current->flags & PF_RANDOMIZE)
1116 load_bias += arch_mmap_rnd();
1117 alignment = maximum_alignment(elf_phdata, elf_ex->e_phnum);
1119 load_bias &= ~(alignment - 1);
1120 elf_flags |= MAP_FIXED;
1125 * Since load_bias is used for all subsequent loading
1126 * calculations, we must lower it by the first vaddr
1127 * so that the remaining calculations based on the
1128 * ELF vaddrs will be correctly offset. The result
1129 * is then page aligned.
1131 load_bias = ELF_PAGESTART(load_bias - vaddr);
1133 total_size = total_mapping_size(elf_phdata,
1137 goto out_free_dentry;
1141 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1142 elf_prot, elf_flags, total_size);
1143 if (BAD_ADDR(error)) {
1144 retval = IS_ERR((void *)error) ?
1145 PTR_ERR((void*)error) : -EINVAL;
1146 goto out_free_dentry;
1149 if (!load_addr_set) {
1151 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1152 if (elf_ex->e_type == ET_DYN) {
1153 load_bias += error -
1154 ELF_PAGESTART(load_bias + vaddr);
1155 load_addr += load_bias;
1156 reloc_func_desc = load_bias;
1159 k = elf_ppnt->p_vaddr;
1160 if ((elf_ppnt->p_flags & PF_X) && k < start_code)
1166 * Check to see if the section's size will overflow the
1167 * allowed task size. Note that p_filesz must always be
1168 * <= p_memsz so it is only necessary to check p_memsz.
1170 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1171 elf_ppnt->p_memsz > TASK_SIZE ||
1172 TASK_SIZE - elf_ppnt->p_memsz < k) {
1173 /* set_brk can never work. Avoid overflows. */
1175 goto out_free_dentry;
1178 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1182 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1186 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1188 bss_prot = elf_prot;
1193 e_entry = elf_ex->e_entry + load_bias;
1194 elf_bss += load_bias;
1195 elf_brk += load_bias;
1196 start_code += load_bias;
1197 end_code += load_bias;
1198 start_data += load_bias;
1199 end_data += load_bias;
1201 /* Calling set_brk effectively mmaps the pages that we need
1202 * for the bss and break sections. We must do this before
1203 * mapping in the interpreter, to make sure it doesn't wind
1204 * up getting placed where the bss needs to go.
1206 retval = set_brk(elf_bss, elf_brk, bss_prot);
1208 goto out_free_dentry;
1209 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1210 retval = -EFAULT; /* Nobody gets to see this, but.. */
1211 goto out_free_dentry;
1215 elf_entry = load_elf_interp(interp_elf_ex,
1217 load_bias, interp_elf_phdata,
1219 if (!IS_ERR((void *)elf_entry)) {
1221 * load_elf_interp() returns relocation
1224 interp_load_addr = elf_entry;
1225 elf_entry += interp_elf_ex->e_entry;
1227 if (BAD_ADDR(elf_entry)) {
1228 retval = IS_ERR((void *)elf_entry) ?
1229 (int)elf_entry : -EINVAL;
1230 goto out_free_dentry;
1232 reloc_func_desc = interp_load_addr;
1234 allow_write_access(interpreter);
1237 kfree(interp_elf_ex);
1238 kfree(interp_elf_phdata);
1240 elf_entry = e_entry;
1241 if (BAD_ADDR(elf_entry)) {
1243 goto out_free_dentry;
1249 set_binfmt(&elf_format);
1251 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1252 retval = ARCH_SETUP_ADDITIONAL_PAGES(bprm, elf_ex, !!interpreter);
1255 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1257 retval = create_elf_tables(bprm, elf_ex,
1258 load_addr, interp_load_addr, e_entry);
1263 mm->end_code = end_code;
1264 mm->start_code = start_code;
1265 mm->start_data = start_data;
1266 mm->end_data = end_data;
1267 mm->start_stack = bprm->p;
1269 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1271 * For architectures with ELF randomization, when executing
1272 * a loader directly (i.e. no interpreter listed in ELF
1273 * headers), move the brk area out of the mmap region
1274 * (since it grows up, and may collide early with the stack
1275 * growing down), and into the unused ELF_ET_DYN_BASE region.
1277 if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) &&
1278 elf_ex->e_type == ET_DYN && !interpreter) {
1279 mm->brk = mm->start_brk = ELF_ET_DYN_BASE;
1282 mm->brk = mm->start_brk = arch_randomize_brk(mm);
1283 #ifdef compat_brk_randomized
1284 current->brk_randomized = 1;
1288 if (current->personality & MMAP_PAGE_ZERO) {
1289 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1290 and some applications "depend" upon this behavior.
1291 Since we do not have the power to recompile these, we
1292 emulate the SVr4 behavior. Sigh. */
1293 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1294 MAP_FIXED | MAP_PRIVATE, 0);
1297 regs = current_pt_regs();
1298 #ifdef ELF_PLAT_INIT
1300 * The ABI may specify that certain registers be set up in special
1301 * ways (on i386 %edx is the address of a DT_FINI function, for
1302 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1303 * that the e_entry field is the address of the function descriptor
1304 * for the startup routine, rather than the address of the startup
1305 * routine itself. This macro performs whatever initialization to
1306 * the regs structure is required as well as any relocations to the
1307 * function descriptor entries when executing dynamically links apps.
1309 ELF_PLAT_INIT(regs, reloc_func_desc);
1312 finalize_exec(bprm);
1313 START_THREAD(elf_ex, regs, elf_entry, bprm->p);
1320 kfree(interp_elf_ex);
1321 kfree(interp_elf_phdata);
1322 allow_write_access(interpreter);
1330 #ifdef CONFIG_USELIB
1331 /* This is really simpleminded and specialized - we are loading an
1332 a.out library that is given an ELF header. */
1333 static int load_elf_library(struct file *file)
1335 struct elf_phdr *elf_phdata;
1336 struct elf_phdr *eppnt;
1337 unsigned long elf_bss, bss, len;
1338 int retval, error, i, j;
1339 struct elfhdr elf_ex;
1342 retval = elf_read(file, &elf_ex, sizeof(elf_ex), 0);
1346 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1349 /* First of all, some simple consistency checks */
1350 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1351 !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1353 if (elf_check_fdpic(&elf_ex))
1356 /* Now read in all of the header information */
1358 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1359 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1362 elf_phdata = kmalloc(j, GFP_KERNEL);
1368 retval = elf_read(file, eppnt, j, elf_ex.e_phoff);
1372 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1373 if ((eppnt + i)->p_type == PT_LOAD)
1378 while (eppnt->p_type != PT_LOAD)
1381 /* Now use mmap to map the library into memory. */
1382 error = vm_mmap(file,
1383 ELF_PAGESTART(eppnt->p_vaddr),
1385 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1386 PROT_READ | PROT_WRITE | PROT_EXEC,
1387 MAP_FIXED_NOREPLACE | MAP_PRIVATE,
1389 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1390 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1393 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1394 if (padzero(elf_bss)) {
1399 len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1400 bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1402 error = vm_brk(len, bss - len);
1413 #endif /* #ifdef CONFIG_USELIB */
1415 #ifdef CONFIG_ELF_CORE
1419 * Modelled on fs/exec.c:aout_core_dump()
1420 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1423 /* An ELF note in memory */
1428 unsigned int datasz;
1432 static int notesize(struct memelfnote *en)
1436 sz = sizeof(struct elf_note);
1437 sz += roundup(strlen(en->name) + 1, 4);
1438 sz += roundup(en->datasz, 4);
1443 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1446 en.n_namesz = strlen(men->name) + 1;
1447 en.n_descsz = men->datasz;
1448 en.n_type = men->type;
1450 return dump_emit(cprm, &en, sizeof(en)) &&
1451 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1452 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1455 static void fill_elf_header(struct elfhdr *elf, int segs,
1456 u16 machine, u32 flags)
1458 memset(elf, 0, sizeof(*elf));
1460 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1461 elf->e_ident[EI_CLASS] = ELF_CLASS;
1462 elf->e_ident[EI_DATA] = ELF_DATA;
1463 elf->e_ident[EI_VERSION] = EV_CURRENT;
1464 elf->e_ident[EI_OSABI] = ELF_OSABI;
1466 elf->e_type = ET_CORE;
1467 elf->e_machine = machine;
1468 elf->e_version = EV_CURRENT;
1469 elf->e_phoff = sizeof(struct elfhdr);
1470 elf->e_flags = flags;
1471 elf->e_ehsize = sizeof(struct elfhdr);
1472 elf->e_phentsize = sizeof(struct elf_phdr);
1473 elf->e_phnum = segs;
1476 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1478 phdr->p_type = PT_NOTE;
1479 phdr->p_offset = offset;
1482 phdr->p_filesz = sz;
1488 static void fill_note(struct memelfnote *note, const char *name, int type,
1489 unsigned int sz, void *data)
1498 * fill up all the fields in prstatus from the given task struct, except
1499 * registers which need to be filled up separately.
1501 static void fill_prstatus(struct elf_prstatus_common *prstatus,
1502 struct task_struct *p, long signr)
1504 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1505 prstatus->pr_sigpend = p->pending.signal.sig[0];
1506 prstatus->pr_sighold = p->blocked.sig[0];
1508 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1510 prstatus->pr_pid = task_pid_vnr(p);
1511 prstatus->pr_pgrp = task_pgrp_vnr(p);
1512 prstatus->pr_sid = task_session_vnr(p);
1513 if (thread_group_leader(p)) {
1514 struct task_cputime cputime;
1517 * This is the record for the group leader. It shows the
1518 * group-wide total, not its individual thread total.
1520 thread_group_cputime(p, &cputime);
1521 prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1522 prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1526 task_cputime(p, &utime, &stime);
1527 prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1528 prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1531 prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1532 prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1535 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1536 struct mm_struct *mm)
1538 const struct cred *cred;
1539 unsigned int i, len;
1542 /* first copy the parameters from user space */
1543 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1545 len = mm->arg_end - mm->arg_start;
1546 if (len >= ELF_PRARGSZ)
1547 len = ELF_PRARGSZ-1;
1548 if (copy_from_user(&psinfo->pr_psargs,
1549 (const char __user *)mm->arg_start, len))
1551 for(i = 0; i < len; i++)
1552 if (psinfo->pr_psargs[i] == 0)
1553 psinfo->pr_psargs[i] = ' ';
1554 psinfo->pr_psargs[len] = 0;
1557 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1559 psinfo->pr_pid = task_pid_vnr(p);
1560 psinfo->pr_pgrp = task_pgrp_vnr(p);
1561 psinfo->pr_sid = task_session_vnr(p);
1563 state = READ_ONCE(p->__state);
1564 i = state ? ffz(~state) + 1 : 0;
1565 psinfo->pr_state = i;
1566 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1567 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1568 psinfo->pr_nice = task_nice(p);
1569 psinfo->pr_flag = p->flags;
1571 cred = __task_cred(p);
1572 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1573 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1575 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1580 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1582 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1586 while (auxv[i - 2] != AT_NULL);
1587 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1590 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1591 const kernel_siginfo_t *siginfo)
1593 copy_siginfo_to_external(csigdata, siginfo);
1594 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1597 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1599 * Format of NT_FILE note:
1601 * long count -- how many files are mapped
1602 * long page_size -- units for file_ofs
1603 * array of [COUNT] elements of
1607 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1609 static int fill_files_note(struct memelfnote *note)
1611 struct mm_struct *mm = current->mm;
1612 struct vm_area_struct *vma;
1613 unsigned count, size, names_ofs, remaining, n;
1615 user_long_t *start_end_ofs;
1616 char *name_base, *name_curpos;
1618 /* *Estimated* file count and total data size needed */
1619 count = mm->map_count;
1620 if (count > UINT_MAX / 64)
1624 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1626 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1628 size = round_up(size, PAGE_SIZE);
1630 * "size" can be 0 here legitimately.
1631 * Let it ENOMEM and omit NT_FILE section which will be empty anyway.
1633 data = kvmalloc(size, GFP_KERNEL);
1634 if (ZERO_OR_NULL_PTR(data))
1637 start_end_ofs = data + 2;
1638 name_base = name_curpos = ((char *)data) + names_ofs;
1639 remaining = size - names_ofs;
1641 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
1643 const char *filename;
1645 file = vma->vm_file;
1648 filename = file_path(file, name_curpos, remaining);
1649 if (IS_ERR(filename)) {
1650 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1652 size = size * 5 / 4;
1658 /* file_path() fills at the end, move name down */
1659 /* n = strlen(filename) + 1: */
1660 n = (name_curpos + remaining) - filename;
1661 remaining = filename - name_curpos;
1662 memmove(name_curpos, filename, n);
1665 *start_end_ofs++ = vma->vm_start;
1666 *start_end_ofs++ = vma->vm_end;
1667 *start_end_ofs++ = vma->vm_pgoff;
1671 /* Now we know exact count of files, can store it */
1673 data[1] = PAGE_SIZE;
1675 * Count usually is less than mm->map_count,
1676 * we need to move filenames down.
1678 n = mm->map_count - count;
1680 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1681 memmove(name_base - shift_bytes, name_base,
1682 name_curpos - name_base);
1683 name_curpos -= shift_bytes;
1686 size = name_curpos - (char *)data;
1687 fill_note(note, "CORE", NT_FILE, size, data);
1691 #ifdef CORE_DUMP_USE_REGSET
1692 #include <linux/regset.h>
1694 struct elf_thread_core_info {
1695 struct elf_thread_core_info *next;
1696 struct task_struct *task;
1697 struct elf_prstatus prstatus;
1698 struct memelfnote notes[];
1701 struct elf_note_info {
1702 struct elf_thread_core_info *thread;
1703 struct memelfnote psinfo;
1704 struct memelfnote signote;
1705 struct memelfnote auxv;
1706 struct memelfnote files;
1707 user_siginfo_t csigdata;
1713 * When a regset has a writeback hook, we call it on each thread before
1714 * dumping user memory. On register window machines, this makes sure the
1715 * user memory backing the register data is up to date before we read it.
1717 static void do_thread_regset_writeback(struct task_struct *task,
1718 const struct user_regset *regset)
1720 if (regset->writeback)
1721 regset->writeback(task, regset, 1);
1724 #ifndef PRSTATUS_SIZE
1725 #define PRSTATUS_SIZE sizeof(struct elf_prstatus)
1728 #ifndef SET_PR_FPVALID
1729 #define SET_PR_FPVALID(S) ((S)->pr_fpvalid = 1)
1732 static int fill_thread_core_info(struct elf_thread_core_info *t,
1733 const struct user_regset_view *view,
1734 long signr, size_t *total)
1739 * NT_PRSTATUS is the one special case, because the regset data
1740 * goes into the pr_reg field inside the note contents, rather
1741 * than being the whole note contents. We fill the reset in here.
1742 * We assume that regset 0 is NT_PRSTATUS.
1744 fill_prstatus(&t->prstatus.common, t->task, signr);
1745 regset_get(t->task, &view->regsets[0],
1746 sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1748 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1749 PRSTATUS_SIZE, &t->prstatus);
1750 *total += notesize(&t->notes[0]);
1752 do_thread_regset_writeback(t->task, &view->regsets[0]);
1755 * Each other regset might generate a note too. For each regset
1756 * that has no core_note_type or is inactive, we leave t->notes[i]
1757 * all zero and we'll know to skip writing it later.
1759 for (i = 1; i < view->n; ++i) {
1760 const struct user_regset *regset = &view->regsets[i];
1761 int note_type = regset->core_note_type;
1762 bool is_fpreg = note_type == NT_PRFPREG;
1766 do_thread_regset_writeback(t->task, regset);
1767 if (!note_type) // not for coredumps
1769 if (regset->active && regset->active(t->task, regset) <= 0)
1772 ret = regset_get_alloc(t->task, regset, ~0U, &data);
1777 SET_PR_FPVALID(&t->prstatus);
1779 fill_note(&t->notes[i], is_fpreg ? "CORE" : "LINUX",
1780 note_type, ret, data);
1782 *total += notesize(&t->notes[i]);
1788 static int fill_note_info(struct elfhdr *elf, int phdrs,
1789 struct elf_note_info *info,
1790 const kernel_siginfo_t *siginfo, struct pt_regs *regs)
1792 struct task_struct *dump_task = current;
1793 const struct user_regset_view *view = task_user_regset_view(dump_task);
1794 struct elf_thread_core_info *t;
1795 struct elf_prpsinfo *psinfo;
1796 struct core_thread *ct;
1800 info->thread = NULL;
1802 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1803 if (psinfo == NULL) {
1804 info->psinfo.data = NULL; /* So we don't free this wrongly */
1808 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1811 * Figure out how many notes we're going to need for each thread.
1813 info->thread_notes = 0;
1814 for (i = 0; i < view->n; ++i)
1815 if (view->regsets[i].core_note_type != 0)
1816 ++info->thread_notes;
1819 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1820 * since it is our one special case.
1822 if (unlikely(info->thread_notes == 0) ||
1823 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1829 * Initialize the ELF file header.
1831 fill_elf_header(elf, phdrs,
1832 view->e_machine, view->e_flags);
1835 * Allocate a structure for each thread.
1837 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1838 t = kzalloc(offsetof(struct elf_thread_core_info,
1839 notes[info->thread_notes]),
1845 if (ct->task == dump_task || !info->thread) {
1846 t->next = info->thread;
1850 * Make sure to keep the original task at
1851 * the head of the list.
1853 t->next = info->thread->next;
1854 info->thread->next = t;
1859 * Now fill in each thread's information.
1861 for (t = info->thread; t != NULL; t = t->next)
1862 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1866 * Fill in the two process-wide notes.
1868 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1869 info->size += notesize(&info->psinfo);
1871 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1872 info->size += notesize(&info->signote);
1874 fill_auxv_note(&info->auxv, current->mm);
1875 info->size += notesize(&info->auxv);
1877 if (fill_files_note(&info->files) == 0)
1878 info->size += notesize(&info->files);
1883 static size_t get_note_info_size(struct elf_note_info *info)
1889 * Write all the notes for each thread. When writing the first thread, the
1890 * process-wide notes are interleaved after the first thread-specific note.
1892 static int write_note_info(struct elf_note_info *info,
1893 struct coredump_params *cprm)
1896 struct elf_thread_core_info *t = info->thread;
1901 if (!writenote(&t->notes[0], cprm))
1904 if (first && !writenote(&info->psinfo, cprm))
1906 if (first && !writenote(&info->signote, cprm))
1908 if (first && !writenote(&info->auxv, cprm))
1910 if (first && info->files.data &&
1911 !writenote(&info->files, cprm))
1914 for (i = 1; i < info->thread_notes; ++i)
1915 if (t->notes[i].data &&
1916 !writenote(&t->notes[i], cprm))
1926 static void free_note_info(struct elf_note_info *info)
1928 struct elf_thread_core_info *threads = info->thread;
1931 struct elf_thread_core_info *t = threads;
1933 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1934 for (i = 1; i < info->thread_notes; ++i)
1935 kfree(t->notes[i].data);
1938 kfree(info->psinfo.data);
1939 kvfree(info->files.data);
1944 /* Here is the structure in which status of each thread is captured. */
1945 struct elf_thread_status
1947 struct list_head list;
1948 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1949 elf_fpregset_t fpu; /* NT_PRFPREG */
1950 struct task_struct *thread;
1951 struct memelfnote notes[3];
1956 * In order to add the specific thread information for the elf file format,
1957 * we need to keep a linked list of every threads pr_status and then create
1958 * a single section for them in the final core file.
1960 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1963 struct task_struct *p = t->thread;
1966 fill_prstatus(&t->prstatus.common, p, signr);
1967 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1969 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1972 sz += notesize(&t->notes[0]);
1974 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1976 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1979 sz += notesize(&t->notes[1]);
1984 struct elf_note_info {
1985 struct memelfnote *notes;
1986 struct memelfnote *notes_files;
1987 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1988 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1989 struct list_head thread_list;
1990 elf_fpregset_t *fpu;
1991 user_siginfo_t csigdata;
1992 int thread_status_size;
1996 static int elf_note_info_init(struct elf_note_info *info)
1998 memset(info, 0, sizeof(*info));
1999 INIT_LIST_HEAD(&info->thread_list);
2001 /* Allocate space for ELF notes */
2002 info->notes = kmalloc_array(8, sizeof(struct memelfnote), GFP_KERNEL);
2005 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2008 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2009 if (!info->prstatus)
2011 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2017 static int fill_note_info(struct elfhdr *elf, int phdrs,
2018 struct elf_note_info *info,
2019 const kernel_siginfo_t *siginfo, struct pt_regs *regs)
2021 struct core_thread *ct;
2022 struct elf_thread_status *ets;
2024 if (!elf_note_info_init(info))
2027 for (ct = current->mm->core_state->dumper.next;
2028 ct; ct = ct->next) {
2029 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2033 ets->thread = ct->task;
2034 list_add(&ets->list, &info->thread_list);
2037 list_for_each_entry(ets, &info->thread_list, list) {
2040 sz = elf_dump_thread_status(siginfo->si_signo, ets);
2041 info->thread_status_size += sz;
2043 /* now collect the dump for the current */
2044 memset(info->prstatus, 0, sizeof(*info->prstatus));
2045 fill_prstatus(&info->prstatus->common, current, siginfo->si_signo);
2046 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2049 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2052 * Set up the notes in similar form to SVR4 core dumps made
2053 * with info from their /proc.
2056 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2057 sizeof(*info->prstatus), info->prstatus);
2058 fill_psinfo(info->psinfo, current->group_leader, current->mm);
2059 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2060 sizeof(*info->psinfo), info->psinfo);
2062 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2063 fill_auxv_note(info->notes + 3, current->mm);
2066 if (fill_files_note(info->notes + info->numnote) == 0) {
2067 info->notes_files = info->notes + info->numnote;
2071 /* Try to dump the FPU. */
2072 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2074 if (info->prstatus->pr_fpvalid)
2075 fill_note(info->notes + info->numnote++,
2076 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2080 static size_t get_note_info_size(struct elf_note_info *info)
2085 for (i = 0; i < info->numnote; i++)
2086 sz += notesize(info->notes + i);
2088 sz += info->thread_status_size;
2093 static int write_note_info(struct elf_note_info *info,
2094 struct coredump_params *cprm)
2096 struct elf_thread_status *ets;
2099 for (i = 0; i < info->numnote; i++)
2100 if (!writenote(info->notes + i, cprm))
2103 /* write out the thread status notes section */
2104 list_for_each_entry(ets, &info->thread_list, list) {
2105 for (i = 0; i < ets->num_notes; i++)
2106 if (!writenote(&ets->notes[i], cprm))
2113 static void free_note_info(struct elf_note_info *info)
2115 while (!list_empty(&info->thread_list)) {
2116 struct list_head *tmp = info->thread_list.next;
2118 kfree(list_entry(tmp, struct elf_thread_status, list));
2121 /* Free data possibly allocated by fill_files_note(): */
2122 if (info->notes_files)
2123 kvfree(info->notes_files->data);
2125 kfree(info->prstatus);
2126 kfree(info->psinfo);
2133 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2134 elf_addr_t e_shoff, int segs)
2136 elf->e_shoff = e_shoff;
2137 elf->e_shentsize = sizeof(*shdr4extnum);
2139 elf->e_shstrndx = SHN_UNDEF;
2141 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2143 shdr4extnum->sh_type = SHT_NULL;
2144 shdr4extnum->sh_size = elf->e_shnum;
2145 shdr4extnum->sh_link = elf->e_shstrndx;
2146 shdr4extnum->sh_info = segs;
2152 * This is a two-pass process; first we find the offsets of the bits,
2153 * and then they are actually written out. If we run out of core limit
2156 static int elf_core_dump(struct coredump_params *cprm)
2159 int vma_count, segs, i;
2160 size_t vma_data_size;
2162 loff_t offset = 0, dataoff;
2163 struct elf_note_info info = { };
2164 struct elf_phdr *phdr4note = NULL;
2165 struct elf_shdr *shdr4extnum = NULL;
2168 struct core_vma_metadata *vma_meta;
2170 if (dump_vma_snapshot(cprm, &vma_count, &vma_meta, &vma_data_size))
2174 * The number of segs are recored into ELF header as 16bit value.
2175 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2177 segs = vma_count + elf_core_extra_phdrs();
2179 /* for notes section */
2182 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2183 * this, kernel supports extended numbering. Have a look at
2184 * include/linux/elf.h for further information. */
2185 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2188 * Collect all the non-memory information about the process for the
2189 * notes. This also sets up the file header.
2191 if (!fill_note_info(&elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2196 offset += sizeof(elf); /* Elf header */
2197 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2199 /* Write notes phdr entry */
2201 size_t sz = get_note_info_size(&info);
2203 /* For cell spufs */
2204 sz += elf_coredump_extra_notes_size();
2206 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2210 fill_elf_note_phdr(phdr4note, sz, offset);
2214 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2216 offset += vma_data_size;
2217 offset += elf_core_extra_data_size();
2220 if (e_phnum == PN_XNUM) {
2221 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2224 fill_extnum_info(&elf, shdr4extnum, e_shoff, segs);
2229 if (!dump_emit(cprm, &elf, sizeof(elf)))
2232 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2235 /* Write program headers for segments dump */
2236 for (i = 0; i < vma_count; i++) {
2237 struct core_vma_metadata *meta = vma_meta + i;
2238 struct elf_phdr phdr;
2240 phdr.p_type = PT_LOAD;
2241 phdr.p_offset = offset;
2242 phdr.p_vaddr = meta->start;
2244 phdr.p_filesz = meta->dump_size;
2245 phdr.p_memsz = meta->end - meta->start;
2246 offset += phdr.p_filesz;
2248 if (meta->flags & VM_READ)
2249 phdr.p_flags |= PF_R;
2250 if (meta->flags & VM_WRITE)
2251 phdr.p_flags |= PF_W;
2252 if (meta->flags & VM_EXEC)
2253 phdr.p_flags |= PF_X;
2254 phdr.p_align = ELF_EXEC_PAGESIZE;
2256 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2260 if (!elf_core_write_extra_phdrs(cprm, offset))
2263 /* write out the notes section */
2264 if (!write_note_info(&info, cprm))
2267 /* For cell spufs */
2268 if (elf_coredump_extra_notes_write(cprm))
2272 dump_skip_to(cprm, dataoff);
2274 for (i = 0; i < vma_count; i++) {
2275 struct core_vma_metadata *meta = vma_meta + i;
2277 if (!dump_user_range(cprm, meta->start, meta->dump_size))
2281 if (!elf_core_write_extra_data(cprm))
2284 if (e_phnum == PN_XNUM) {
2285 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2290 free_note_info(&info);
2297 #endif /* CONFIG_ELF_CORE */
2299 static int __init init_elf_binfmt(void)
2301 register_binfmt(&elf_format);
2305 static void __exit exit_elf_binfmt(void)
2307 /* Remove the COFF and ELF loaders. */
2308 unregister_binfmt(&elf_format);
2311 core_initcall(init_elf_binfmt);
2312 module_exit(exit_elf_binfmt);
2313 MODULE_LICENSE("GPL");