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>
17 #include <linux/mman.h>
18 #include <linux/errno.h>
19 #include <linux/signal.h>
20 #include <linux/binfmts.h>
21 #include <linux/string.h>
22 #include <linux/file.h>
23 #include <linux/slab.h>
24 #include <linux/personality.h>
25 #include <linux/elfcore.h>
26 #include <linux/init.h>
27 #include <linux/highuid.h>
28 #include <linux/compiler.h>
29 #include <linux/highmem.h>
30 #include <linux/hugetlb.h>
31 #include <linux/pagemap.h>
32 #include <linux/vmalloc.h>
33 #include <linux/security.h>
34 #include <linux/random.h>
35 #include <linux/elf.h>
36 #include <linux/elf-randomize.h>
37 #include <linux/utsname.h>
38 #include <linux/coredump.h>
39 #include <linux/sched.h>
40 #include <linux/sched/coredump.h>
41 #include <linux/sched/task_stack.h>
42 #include <linux/sched/cputime.h>
43 #include <linux/cred.h>
44 #include <linux/dax.h>
45 #include <linux/uaccess.h>
46 #include <asm/param.h>
50 #define user_long_t long
52 #ifndef user_siginfo_t
53 #define user_siginfo_t siginfo_t
56 /* That's for binfmt_elf_fdpic to deal with */
57 #ifndef elf_check_fdpic
58 #define elf_check_fdpic(ex) false
61 static int load_elf_binary(struct linux_binprm *bprm);
64 static int load_elf_library(struct file *);
66 #define load_elf_library NULL
70 * If we don't support core dumping, then supply a NULL so we
73 #ifdef CONFIG_ELF_CORE
74 static int elf_core_dump(struct coredump_params *cprm);
76 #define elf_core_dump NULL
79 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
80 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
82 #define ELF_MIN_ALIGN PAGE_SIZE
85 #ifndef ELF_CORE_EFLAGS
86 #define ELF_CORE_EFLAGS 0
89 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
90 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
91 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
93 static struct linux_binfmt elf_format = {
94 .module = THIS_MODULE,
95 .load_binary = load_elf_binary,
96 .load_shlib = load_elf_library,
97 .core_dump = elf_core_dump,
98 .min_coredump = ELF_EXEC_PAGESIZE,
101 #define BAD_ADDR(x) (unlikely((unsigned long)(x) >= TASK_SIZE))
103 static int set_brk(unsigned long start, unsigned long end, int prot)
105 start = ELF_PAGEALIGN(start);
106 end = ELF_PAGEALIGN(end);
109 * Map the last of the bss segment.
110 * If the header is requesting these pages to be
111 * executable, honour that (ppc32 needs this).
113 int error = vm_brk_flags(start, end - start,
114 prot & PROT_EXEC ? VM_EXEC : 0);
118 current->mm->start_brk = current->mm->brk = end;
122 /* We need to explicitly zero any fractional pages
123 after the data section (i.e. bss). This would
124 contain the junk from the file that should not
127 static int padzero(unsigned long elf_bss)
131 nbyte = ELF_PAGEOFFSET(elf_bss);
133 nbyte = ELF_MIN_ALIGN - nbyte;
134 if (clear_user((void __user *) elf_bss, nbyte))
140 /* Let's use some macros to make this stack manipulation a little clearer */
141 #ifdef CONFIG_STACK_GROWSUP
142 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
143 #define STACK_ROUND(sp, items) \
144 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
145 #define STACK_ALLOC(sp, len) ({ \
146 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
149 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
150 #define STACK_ROUND(sp, items) \
151 (((unsigned long) (sp - items)) &~ 15UL)
152 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
155 #ifndef ELF_BASE_PLATFORM
157 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
158 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
159 * will be copied to the user stack in the same manner as AT_PLATFORM.
161 #define ELF_BASE_PLATFORM NULL
165 create_elf_tables(struct linux_binprm *bprm, const struct elfhdr *exec,
166 unsigned long load_addr, unsigned long interp_load_addr,
167 unsigned long e_entry)
169 struct mm_struct *mm = current->mm;
170 unsigned long p = bprm->p;
171 int argc = bprm->argc;
172 int envc = bprm->envc;
173 elf_addr_t __user *sp;
174 elf_addr_t __user *u_platform;
175 elf_addr_t __user *u_base_platform;
176 elf_addr_t __user *u_rand_bytes;
177 const char *k_platform = ELF_PLATFORM;
178 const char *k_base_platform = ELF_BASE_PLATFORM;
179 unsigned char k_rand_bytes[16];
181 elf_addr_t *elf_info;
183 const struct cred *cred = current_cred();
184 struct vm_area_struct *vma;
187 * In some cases (e.g. Hyper-Threading), we want to avoid L1
188 * evictions by the processes running on the same package. One
189 * thing we can do is to shuffle the initial stack for them.
192 p = arch_align_stack(p);
195 * If this architecture has a platform capability string, copy it
196 * to userspace. In some cases (Sparc), this info is impossible
197 * for userspace to get any other way, in others (i386) it is
202 size_t len = strlen(k_platform) + 1;
204 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
205 if (__copy_to_user(u_platform, k_platform, len))
210 * If this architecture has a "base" platform capability
211 * string, copy it to userspace.
213 u_base_platform = NULL;
214 if (k_base_platform) {
215 size_t len = strlen(k_base_platform) + 1;
217 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
218 if (__copy_to_user(u_base_platform, k_base_platform, len))
223 * Generate 16 random bytes for userspace PRNG seeding.
225 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
226 u_rand_bytes = (elf_addr_t __user *)
227 STACK_ALLOC(p, sizeof(k_rand_bytes));
228 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
231 /* Create the ELF interpreter info */
232 elf_info = (elf_addr_t *)mm->saved_auxv;
233 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
234 #define NEW_AUX_ENT(id, val) \
242 * ARCH_DLINFO must come first so PPC can do its special alignment of
244 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
245 * ARCH_DLINFO changes
249 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
250 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
251 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
252 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
253 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
254 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
255 NEW_AUX_ENT(AT_BASE, interp_load_addr);
256 NEW_AUX_ENT(AT_FLAGS, 0);
257 NEW_AUX_ENT(AT_ENTRY, e_entry);
258 NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
259 NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
260 NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
261 NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
262 NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
263 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
265 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
267 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
269 NEW_AUX_ENT(AT_PLATFORM,
270 (elf_addr_t)(unsigned long)u_platform);
272 if (k_base_platform) {
273 NEW_AUX_ENT(AT_BASE_PLATFORM,
274 (elf_addr_t)(unsigned long)u_base_platform);
276 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
277 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
280 /* AT_NULL is zero; clear the rest too */
281 memset(elf_info, 0, (char *)mm->saved_auxv +
282 sizeof(mm->saved_auxv) - (char *)elf_info);
284 /* And advance past the AT_NULL entry. */
287 ei_index = elf_info - (elf_addr_t *)mm->saved_auxv;
288 sp = STACK_ADD(p, ei_index);
290 items = (argc + 1) + (envc + 1) + 1;
291 bprm->p = STACK_ROUND(sp, items);
293 /* Point sp at the lowest address on the stack */
294 #ifdef CONFIG_STACK_GROWSUP
295 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
296 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
298 sp = (elf_addr_t __user *)bprm->p;
303 * Grow the stack manually; some architectures have a limit on how
304 * far ahead a user-space access may be in order to grow the stack.
306 vma = find_extend_vma(mm, bprm->p);
310 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
311 if (__put_user(argc, sp++))
314 /* Populate list of argv pointers back to argv strings. */
315 p = mm->arg_end = mm->arg_start;
318 if (__put_user((elf_addr_t)p, sp++))
320 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
321 if (!len || len > MAX_ARG_STRLEN)
325 if (__put_user(0, sp++))
329 /* Populate list of envp pointers back to envp strings. */
330 mm->env_end = mm->env_start = p;
333 if (__put_user((elf_addr_t)p, sp++))
335 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
336 if (!len || len > MAX_ARG_STRLEN)
340 if (__put_user(0, sp++))
344 /* Put the elf_info on the stack in the right place. */
345 if (copy_to_user(sp, mm->saved_auxv, ei_index * sizeof(elf_addr_t)))
352 static unsigned long elf_map(struct file *filep, unsigned long addr,
353 const struct elf_phdr *eppnt, int prot, int type,
354 unsigned long total_size)
356 unsigned long map_addr;
357 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
358 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
359 addr = ELF_PAGESTART(addr);
360 size = ELF_PAGEALIGN(size);
362 /* mmap() will return -EINVAL if given a zero size, but a
363 * segment with zero filesize is perfectly valid */
368 * total_size is the size of the ELF (interpreter) image.
369 * The _first_ mmap needs to know the full size, otherwise
370 * randomization might put this image into an overlapping
371 * position with the ELF binary image. (since size < total_size)
372 * So we first map the 'big' image - and unmap the remainder at
373 * the end. (which unmap is needed for ELF images with holes.)
376 total_size = ELF_PAGEALIGN(total_size);
377 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
378 if (!BAD_ADDR(map_addr))
379 vm_munmap(map_addr+size, total_size-size);
381 map_addr = vm_mmap(filep, addr, size, prot, type, off);
383 if ((type & MAP_FIXED_NOREPLACE) &&
384 PTR_ERR((void *)map_addr) == -EEXIST)
385 pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
386 task_pid_nr(current), current->comm, (void *)addr);
391 #endif /* !elf_map */
393 static unsigned long total_mapping_size(const struct elf_phdr *cmds, int nr)
395 int i, first_idx = -1, last_idx = -1;
397 for (i = 0; i < nr; i++) {
398 if (cmds[i].p_type == PT_LOAD) {
407 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
408 ELF_PAGESTART(cmds[first_idx].p_vaddr);
411 static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
415 rv = kernel_read(file, buf, len, &pos);
416 if (unlikely(rv != len)) {
417 return (rv < 0) ? rv : -EIO;
423 * load_elf_phdrs() - load ELF program headers
424 * @elf_ex: ELF header of the binary whose program headers should be loaded
425 * @elf_file: the opened ELF binary file
427 * Loads ELF program headers from the binary file elf_file, which has the ELF
428 * header pointed to by elf_ex, into a newly allocated array. The caller is
429 * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
431 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
432 struct file *elf_file)
434 struct elf_phdr *elf_phdata = NULL;
435 int retval, err = -1;
439 * If the size of this structure has changed, then punt, since
440 * we will be doing the wrong thing.
442 if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
445 /* Sanity check the number of program headers... */
446 /* ...and their total size. */
447 size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
448 if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
451 elf_phdata = kmalloc(size, GFP_KERNEL);
455 /* Read in the program headers */
456 retval = elf_read(elf_file, elf_phdata, size, elf_ex->e_phoff);
472 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
475 * struct arch_elf_state - arch-specific ELF loading state
477 * This structure is used to preserve architecture specific data during
478 * the loading of an ELF file, throughout the checking of architecture
479 * specific ELF headers & through to the point where the ELF load is
480 * known to be proceeding (ie. SET_PERSONALITY).
482 * This implementation is a dummy for architectures which require no
485 struct arch_elf_state {
488 #define INIT_ARCH_ELF_STATE {}
491 * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
492 * @ehdr: The main ELF header
493 * @phdr: The program header to check
494 * @elf: The open ELF file
495 * @is_interp: True if the phdr is from the interpreter of the ELF being
496 * loaded, else false.
497 * @state: Architecture-specific state preserved throughout the process
498 * of loading the ELF.
500 * Inspects the program header phdr to validate its correctness and/or
501 * suitability for the system. Called once per ELF program header in the
502 * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
505 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
506 * with that return code.
508 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
509 struct elf_phdr *phdr,
510 struct file *elf, bool is_interp,
511 struct arch_elf_state *state)
513 /* Dummy implementation, always proceed */
518 * arch_check_elf() - check an ELF executable
519 * @ehdr: The main ELF header
520 * @has_interp: True if the ELF has an interpreter, else false.
521 * @interp_ehdr: The interpreter's ELF header
522 * @state: Architecture-specific state preserved throughout the process
523 * of loading the ELF.
525 * Provides a final opportunity for architecture code to reject the loading
526 * of the ELF & cause an exec syscall to return an error. This is called after
527 * all program headers to be checked by arch_elf_pt_proc have been.
529 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
530 * with that return code.
532 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
533 struct elfhdr *interp_ehdr,
534 struct arch_elf_state *state)
536 /* Dummy implementation, always proceed */
540 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
542 static inline int make_prot(u32 p_flags)
555 /* This is much more generalized than the library routine read function,
556 so we keep this separate. Technically the library read function
557 is only provided so that we can read a.out libraries that have
560 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
561 struct file *interpreter,
562 unsigned long no_base, struct elf_phdr *interp_elf_phdata)
564 struct elf_phdr *eppnt;
565 unsigned long load_addr = 0;
566 int load_addr_set = 0;
567 unsigned long last_bss = 0, elf_bss = 0;
569 unsigned long error = ~0UL;
570 unsigned long total_size;
573 /* First of all, some simple consistency checks */
574 if (interp_elf_ex->e_type != ET_EXEC &&
575 interp_elf_ex->e_type != ET_DYN)
577 if (!elf_check_arch(interp_elf_ex) ||
578 elf_check_fdpic(interp_elf_ex))
580 if (!interpreter->f_op->mmap)
583 total_size = total_mapping_size(interp_elf_phdata,
584 interp_elf_ex->e_phnum);
590 eppnt = interp_elf_phdata;
591 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
592 if (eppnt->p_type == PT_LOAD) {
593 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
594 int elf_prot = make_prot(eppnt->p_flags);
595 unsigned long vaddr = 0;
596 unsigned long k, map_addr;
598 vaddr = eppnt->p_vaddr;
599 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
600 elf_type |= MAP_FIXED_NOREPLACE;
601 else if (no_base && interp_elf_ex->e_type == ET_DYN)
604 map_addr = elf_map(interpreter, load_addr + vaddr,
605 eppnt, elf_prot, elf_type, total_size);
608 if (BAD_ADDR(map_addr))
611 if (!load_addr_set &&
612 interp_elf_ex->e_type == ET_DYN) {
613 load_addr = map_addr - ELF_PAGESTART(vaddr);
618 * Check to see if the section's size will overflow the
619 * allowed task size. Note that p_filesz must always be
620 * <= p_memsize so it's only necessary to check p_memsz.
622 k = load_addr + eppnt->p_vaddr;
624 eppnt->p_filesz > eppnt->p_memsz ||
625 eppnt->p_memsz > TASK_SIZE ||
626 TASK_SIZE - eppnt->p_memsz < k) {
632 * Find the end of the file mapping for this phdr, and
633 * keep track of the largest address we see for this.
635 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
640 * Do the same thing for the memory mapping - between
641 * elf_bss and last_bss is the bss section.
643 k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
652 * Now fill out the bss section: first pad the last page from
653 * the file up to the page boundary, and zero it from elf_bss
654 * up to the end of the page.
656 if (padzero(elf_bss)) {
661 * Next, align both the file and mem bss up to the page size,
662 * since this is where elf_bss was just zeroed up to, and where
663 * last_bss will end after the vm_brk_flags() below.
665 elf_bss = ELF_PAGEALIGN(elf_bss);
666 last_bss = ELF_PAGEALIGN(last_bss);
667 /* Finally, if there is still more bss to allocate, do it. */
668 if (last_bss > elf_bss) {
669 error = vm_brk_flags(elf_bss, last_bss - elf_bss,
670 bss_prot & PROT_EXEC ? VM_EXEC : 0);
681 * These are the functions used to load ELF style executables and shared
682 * libraries. There is no binary dependent code anywhere else.
685 static int load_elf_binary(struct linux_binprm *bprm)
687 struct file *interpreter = NULL; /* to shut gcc up */
688 unsigned long load_addr = 0, load_bias = 0;
689 int load_addr_set = 0;
691 struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
692 unsigned long elf_bss, elf_brk;
695 unsigned long elf_entry;
696 unsigned long e_entry;
697 unsigned long interp_load_addr = 0;
698 unsigned long start_code, end_code, start_data, end_data;
699 unsigned long reloc_func_desc __maybe_unused = 0;
700 int executable_stack = EXSTACK_DEFAULT;
701 struct elfhdr *elf_ex = (struct elfhdr *)bprm->buf;
703 struct elfhdr interp_elf_ex;
705 struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
706 struct mm_struct *mm;
707 struct pt_regs *regs;
709 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
716 /* First of all, some simple consistency checks */
717 if (memcmp(elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
720 if (elf_ex->e_type != ET_EXEC && elf_ex->e_type != ET_DYN)
722 if (!elf_check_arch(elf_ex))
724 if (elf_check_fdpic(elf_ex))
726 if (!bprm->file->f_op->mmap)
729 elf_phdata = load_elf_phdrs(elf_ex, bprm->file);
733 elf_ppnt = elf_phdata;
734 for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++) {
735 char *elf_interpreter;
737 if (elf_ppnt->p_type != PT_INTERP)
741 * This is the program interpreter used for shared libraries -
742 * for now assume that this is an a.out format binary.
745 if (elf_ppnt->p_filesz > PATH_MAX || elf_ppnt->p_filesz < 2)
749 elf_interpreter = kmalloc(elf_ppnt->p_filesz, GFP_KERNEL);
750 if (!elf_interpreter)
753 retval = elf_read(bprm->file, elf_interpreter, elf_ppnt->p_filesz,
756 goto out_free_interp;
757 /* make sure path is NULL terminated */
759 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
760 goto out_free_interp;
762 interpreter = open_exec(elf_interpreter);
763 kfree(elf_interpreter);
764 retval = PTR_ERR(interpreter);
765 if (IS_ERR(interpreter))
769 * If the binary is not readable then enforce mm->dumpable = 0
770 * regardless of the interpreter's permissions.
772 would_dump(bprm, interpreter);
774 /* Get the exec headers */
775 retval = elf_read(interpreter, &loc->interp_elf_ex,
776 sizeof(loc->interp_elf_ex), 0);
778 goto out_free_dentry;
783 kfree(elf_interpreter);
787 elf_ppnt = elf_phdata;
788 for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++)
789 switch (elf_ppnt->p_type) {
791 if (elf_ppnt->p_flags & PF_X)
792 executable_stack = EXSTACK_ENABLE_X;
794 executable_stack = EXSTACK_DISABLE_X;
797 case PT_LOPROC ... PT_HIPROC:
798 retval = arch_elf_pt_proc(elf_ex, elf_ppnt,
802 goto out_free_dentry;
806 /* Some simple consistency checks for the interpreter */
809 /* Not an ELF interpreter */
810 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
811 goto out_free_dentry;
812 /* Verify the interpreter has a valid arch */
813 if (!elf_check_arch(&loc->interp_elf_ex) ||
814 elf_check_fdpic(&loc->interp_elf_ex))
815 goto out_free_dentry;
817 /* Load the interpreter program headers */
818 interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
820 if (!interp_elf_phdata)
821 goto out_free_dentry;
823 /* Pass PT_LOPROC..PT_HIPROC headers to arch code */
824 elf_ppnt = interp_elf_phdata;
825 for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
826 switch (elf_ppnt->p_type) {
827 case PT_LOPROC ... PT_HIPROC:
828 retval = arch_elf_pt_proc(&loc->interp_elf_ex,
829 elf_ppnt, interpreter,
832 goto out_free_dentry;
838 * Allow arch code to reject the ELF at this point, whilst it's
839 * still possible to return an error to the code that invoked
842 retval = arch_check_elf(elf_ex,
843 !!interpreter, &loc->interp_elf_ex,
846 goto out_free_dentry;
848 /* Flush all traces of the currently running executable */
849 retval = flush_old_exec(bprm);
851 goto out_free_dentry;
853 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
854 may depend on the personality. */
855 SET_PERSONALITY2(*elf_ex, &arch_state);
856 if (elf_read_implies_exec(*elf_ex, executable_stack))
857 current->personality |= READ_IMPLIES_EXEC;
859 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
860 current->flags |= PF_RANDOMIZE;
862 setup_new_exec(bprm);
863 install_exec_creds(bprm);
865 /* Do this so that we can load the interpreter, if need be. We will
866 change some of these later */
867 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
870 goto out_free_dentry;
880 /* Now we do a little grungy work by mmapping the ELF image into
881 the correct location in memory. */
882 for(i = 0, elf_ppnt = elf_phdata;
883 i < elf_ex->e_phnum; i++, elf_ppnt++) {
884 int elf_prot, elf_flags;
885 unsigned long k, vaddr;
886 unsigned long total_size = 0;
888 if (elf_ppnt->p_type != PT_LOAD)
891 if (unlikely (elf_brk > elf_bss)) {
894 /* There was a PT_LOAD segment with p_memsz > p_filesz
895 before this one. Map anonymous pages, if needed,
896 and clear the area. */
897 retval = set_brk(elf_bss + load_bias,
901 goto out_free_dentry;
902 nbyte = ELF_PAGEOFFSET(elf_bss);
904 nbyte = ELF_MIN_ALIGN - nbyte;
905 if (nbyte > elf_brk - elf_bss)
906 nbyte = elf_brk - elf_bss;
907 if (clear_user((void __user *)elf_bss +
910 * This bss-zeroing can fail if the ELF
911 * file specifies odd protections. So
912 * we don't check the return value
918 elf_prot = make_prot(elf_ppnt->p_flags);
920 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
922 vaddr = elf_ppnt->p_vaddr;
924 * If we are loading ET_EXEC or we have already performed
925 * the ET_DYN load_addr calculations, proceed normally.
927 if (elf_ex->e_type == ET_EXEC || load_addr_set) {
928 elf_flags |= MAP_FIXED;
929 } else if (elf_ex->e_type == ET_DYN) {
931 * This logic is run once for the first LOAD Program
932 * Header for ET_DYN binaries to calculate the
933 * randomization (load_bias) for all the LOAD
934 * Program Headers, and to calculate the entire
935 * size of the ELF mapping (total_size). (Note that
936 * load_addr_set is set to true later once the
937 * initial mapping is performed.)
939 * There are effectively two types of ET_DYN
940 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
941 * and loaders (ET_DYN without INTERP, since they
942 * _are_ the ELF interpreter). The loaders must
943 * be loaded away from programs since the program
944 * may otherwise collide with the loader (especially
945 * for ET_EXEC which does not have a randomized
946 * position). For example to handle invocations of
947 * "./ld.so someprog" to test out a new version of
948 * the loader, the subsequent program that the
949 * loader loads must avoid the loader itself, so
950 * they cannot share the same load range. Sufficient
951 * room for the brk must be allocated with the
952 * loader as well, since brk must be available with
955 * Therefore, programs are loaded offset from
956 * ELF_ET_DYN_BASE and loaders are loaded into the
957 * independently randomized mmap region (0 load_bias
958 * without MAP_FIXED).
961 load_bias = ELF_ET_DYN_BASE;
962 if (current->flags & PF_RANDOMIZE)
963 load_bias += arch_mmap_rnd();
964 elf_flags |= MAP_FIXED;
969 * Since load_bias is used for all subsequent loading
970 * calculations, we must lower it by the first vaddr
971 * so that the remaining calculations based on the
972 * ELF vaddrs will be correctly offset. The result
973 * is then page aligned.
975 load_bias = ELF_PAGESTART(load_bias - vaddr);
977 total_size = total_mapping_size(elf_phdata,
981 goto out_free_dentry;
985 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
986 elf_prot, elf_flags, total_size);
987 if (BAD_ADDR(error)) {
988 retval = IS_ERR((void *)error) ?
989 PTR_ERR((void*)error) : -EINVAL;
990 goto out_free_dentry;
993 if (!load_addr_set) {
995 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
996 if (elf_ex->e_type == ET_DYN) {
998 ELF_PAGESTART(load_bias + vaddr);
999 load_addr += load_bias;
1000 reloc_func_desc = load_bias;
1003 k = elf_ppnt->p_vaddr;
1004 if ((elf_ppnt->p_flags & PF_X) && k < start_code)
1010 * Check to see if the section's size will overflow the
1011 * allowed task size. Note that p_filesz must always be
1012 * <= p_memsz so it is only necessary to check p_memsz.
1014 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1015 elf_ppnt->p_memsz > TASK_SIZE ||
1016 TASK_SIZE - elf_ppnt->p_memsz < k) {
1017 /* set_brk can never work. Avoid overflows. */
1019 goto out_free_dentry;
1022 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1026 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1030 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1032 bss_prot = elf_prot;
1037 e_entry = elf_ex->e_entry + load_bias;
1038 elf_bss += load_bias;
1039 elf_brk += load_bias;
1040 start_code += load_bias;
1041 end_code += load_bias;
1042 start_data += load_bias;
1043 end_data += load_bias;
1045 /* Calling set_brk effectively mmaps the pages that we need
1046 * for the bss and break sections. We must do this before
1047 * mapping in the interpreter, to make sure it doesn't wind
1048 * up getting placed where the bss needs to go.
1050 retval = set_brk(elf_bss, elf_brk, bss_prot);
1052 goto out_free_dentry;
1053 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1054 retval = -EFAULT; /* Nobody gets to see this, but.. */
1055 goto out_free_dentry;
1059 elf_entry = load_elf_interp(&loc->interp_elf_ex,
1061 load_bias, interp_elf_phdata);
1062 if (!IS_ERR((void *)elf_entry)) {
1064 * load_elf_interp() returns relocation
1067 interp_load_addr = elf_entry;
1068 elf_entry += loc->interp_elf_ex.e_entry;
1070 if (BAD_ADDR(elf_entry)) {
1071 retval = IS_ERR((void *)elf_entry) ?
1072 (int)elf_entry : -EINVAL;
1073 goto out_free_dentry;
1075 reloc_func_desc = interp_load_addr;
1077 allow_write_access(interpreter);
1080 elf_entry = e_entry;
1081 if (BAD_ADDR(elf_entry)) {
1083 goto out_free_dentry;
1087 kfree(interp_elf_phdata);
1090 set_binfmt(&elf_format);
1092 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1093 retval = arch_setup_additional_pages(bprm, !!interpreter);
1096 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1098 retval = create_elf_tables(bprm, elf_ex,
1099 load_addr, interp_load_addr, e_entry);
1104 mm->end_code = end_code;
1105 mm->start_code = start_code;
1106 mm->start_data = start_data;
1107 mm->end_data = end_data;
1108 mm->start_stack = bprm->p;
1110 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1112 * For architectures with ELF randomization, when executing
1113 * a loader directly (i.e. no interpreter listed in ELF
1114 * headers), move the brk area out of the mmap region
1115 * (since it grows up, and may collide early with the stack
1116 * growing down), and into the unused ELF_ET_DYN_BASE region.
1118 if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) &&
1119 elf_ex->e_type == ET_DYN && !interpreter) {
1120 mm->brk = mm->start_brk = ELF_ET_DYN_BASE;
1123 mm->brk = mm->start_brk = arch_randomize_brk(mm);
1124 #ifdef compat_brk_randomized
1125 current->brk_randomized = 1;
1129 if (current->personality & MMAP_PAGE_ZERO) {
1130 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1131 and some applications "depend" upon this behavior.
1132 Since we do not have the power to recompile these, we
1133 emulate the SVr4 behavior. Sigh. */
1134 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1135 MAP_FIXED | MAP_PRIVATE, 0);
1138 regs = current_pt_regs();
1139 #ifdef ELF_PLAT_INIT
1141 * The ABI may specify that certain registers be set up in special
1142 * ways (on i386 %edx is the address of a DT_FINI function, for
1143 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1144 * that the e_entry field is the address of the function descriptor
1145 * for the startup routine, rather than the address of the startup
1146 * routine itself. This macro performs whatever initialization to
1147 * the regs structure is required as well as any relocations to the
1148 * function descriptor entries when executing dynamically links apps.
1150 ELF_PLAT_INIT(regs, reloc_func_desc);
1153 finalize_exec(bprm);
1154 start_thread(regs, elf_entry, bprm->p);
1163 kfree(interp_elf_phdata);
1164 allow_write_access(interpreter);
1172 #ifdef CONFIG_USELIB
1173 /* This is really simpleminded and specialized - we are loading an
1174 a.out library that is given an ELF header. */
1175 static int load_elf_library(struct file *file)
1177 struct elf_phdr *elf_phdata;
1178 struct elf_phdr *eppnt;
1179 unsigned long elf_bss, bss, len;
1180 int retval, error, i, j;
1181 struct elfhdr elf_ex;
1184 retval = elf_read(file, &elf_ex, sizeof(elf_ex), 0);
1188 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1191 /* First of all, some simple consistency checks */
1192 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1193 !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1195 if (elf_check_fdpic(&elf_ex))
1198 /* Now read in all of the header information */
1200 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1201 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1204 elf_phdata = kmalloc(j, GFP_KERNEL);
1210 retval = elf_read(file, eppnt, j, elf_ex.e_phoff);
1214 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1215 if ((eppnt + i)->p_type == PT_LOAD)
1220 while (eppnt->p_type != PT_LOAD)
1223 /* Now use mmap to map the library into memory. */
1224 error = vm_mmap(file,
1225 ELF_PAGESTART(eppnt->p_vaddr),
1227 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1228 PROT_READ | PROT_WRITE | PROT_EXEC,
1229 MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_DENYWRITE,
1231 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1232 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1235 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1236 if (padzero(elf_bss)) {
1241 len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1242 bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1244 error = vm_brk(len, bss - len);
1255 #endif /* #ifdef CONFIG_USELIB */
1257 #ifdef CONFIG_ELF_CORE
1261 * Modelled on fs/exec.c:aout_core_dump()
1262 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1266 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1267 * that are useful for post-mortem analysis are included in every core dump.
1268 * In that way we ensure that the core dump is fully interpretable later
1269 * without matching up the same kernel and hardware config to see what PC values
1270 * meant. These special mappings include - vDSO, vsyscall, and other
1271 * architecture specific mappings
1273 static bool always_dump_vma(struct vm_area_struct *vma)
1275 /* Any vsyscall mappings? */
1276 if (vma == get_gate_vma(vma->vm_mm))
1280 * Assume that all vmas with a .name op should always be dumped.
1281 * If this changes, a new vm_ops field can easily be added.
1283 if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1287 * arch_vma_name() returns non-NULL for special architecture mappings,
1288 * such as vDSO sections.
1290 if (arch_vma_name(vma))
1297 * Decide what to dump of a segment, part, all or none.
1299 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1300 unsigned long mm_flags)
1302 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1304 /* always dump the vdso and vsyscall sections */
1305 if (always_dump_vma(vma))
1308 if (vma->vm_flags & VM_DONTDUMP)
1311 /* support for DAX */
1312 if (vma_is_dax(vma)) {
1313 if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1315 if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1320 /* Hugetlb memory check */
1321 if (is_vm_hugetlb_page(vma)) {
1322 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1324 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1329 /* Do not dump I/O mapped devices or special mappings */
1330 if (vma->vm_flags & VM_IO)
1333 /* By default, dump shared memory if mapped from an anonymous file. */
1334 if (vma->vm_flags & VM_SHARED) {
1335 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1336 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1341 /* Dump segments that have been written to. */
1342 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1344 if (vma->vm_file == NULL)
1347 if (FILTER(MAPPED_PRIVATE))
1351 * If this looks like the beginning of a DSO or executable mapping,
1352 * check for an ELF header. If we find one, dump the first page to
1353 * aid in determining what was mapped here.
1355 if (FILTER(ELF_HEADERS) &&
1356 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1357 u32 __user *header = (u32 __user *) vma->vm_start;
1359 mm_segment_t fs = get_fs();
1361 * Doing it this way gets the constant folded by GCC.
1365 char elfmag[SELFMAG];
1367 BUILD_BUG_ON(SELFMAG != sizeof word);
1368 magic.elfmag[EI_MAG0] = ELFMAG0;
1369 magic.elfmag[EI_MAG1] = ELFMAG1;
1370 magic.elfmag[EI_MAG2] = ELFMAG2;
1371 magic.elfmag[EI_MAG3] = ELFMAG3;
1373 * Switch to the user "segment" for get_user(),
1374 * then put back what elf_core_dump() had in place.
1377 if (unlikely(get_user(word, header)))
1380 if (word == magic.cmp)
1389 return vma->vm_end - vma->vm_start;
1392 /* An ELF note in memory */
1397 unsigned int datasz;
1401 static int notesize(struct memelfnote *en)
1405 sz = sizeof(struct elf_note);
1406 sz += roundup(strlen(en->name) + 1, 4);
1407 sz += roundup(en->datasz, 4);
1412 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1415 en.n_namesz = strlen(men->name) + 1;
1416 en.n_descsz = men->datasz;
1417 en.n_type = men->type;
1419 return dump_emit(cprm, &en, sizeof(en)) &&
1420 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1421 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1424 static void fill_elf_header(struct elfhdr *elf, int segs,
1425 u16 machine, u32 flags)
1427 memset(elf, 0, sizeof(*elf));
1429 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1430 elf->e_ident[EI_CLASS] = ELF_CLASS;
1431 elf->e_ident[EI_DATA] = ELF_DATA;
1432 elf->e_ident[EI_VERSION] = EV_CURRENT;
1433 elf->e_ident[EI_OSABI] = ELF_OSABI;
1435 elf->e_type = ET_CORE;
1436 elf->e_machine = machine;
1437 elf->e_version = EV_CURRENT;
1438 elf->e_phoff = sizeof(struct elfhdr);
1439 elf->e_flags = flags;
1440 elf->e_ehsize = sizeof(struct elfhdr);
1441 elf->e_phentsize = sizeof(struct elf_phdr);
1442 elf->e_phnum = segs;
1445 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1447 phdr->p_type = PT_NOTE;
1448 phdr->p_offset = offset;
1451 phdr->p_filesz = sz;
1457 static void fill_note(struct memelfnote *note, const char *name, int type,
1458 unsigned int sz, void *data)
1467 * fill up all the fields in prstatus from the given task struct, except
1468 * registers which need to be filled up separately.
1470 static void fill_prstatus(struct elf_prstatus *prstatus,
1471 struct task_struct *p, long signr)
1473 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1474 prstatus->pr_sigpend = p->pending.signal.sig[0];
1475 prstatus->pr_sighold = p->blocked.sig[0];
1477 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1479 prstatus->pr_pid = task_pid_vnr(p);
1480 prstatus->pr_pgrp = task_pgrp_vnr(p);
1481 prstatus->pr_sid = task_session_vnr(p);
1482 if (thread_group_leader(p)) {
1483 struct task_cputime cputime;
1486 * This is the record for the group leader. It shows the
1487 * group-wide total, not its individual thread total.
1489 thread_group_cputime(p, &cputime);
1490 prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1491 prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1495 task_cputime(p, &utime, &stime);
1496 prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1497 prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1500 prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1501 prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1504 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1505 struct mm_struct *mm)
1507 const struct cred *cred;
1508 unsigned int i, len;
1510 /* first copy the parameters from user space */
1511 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1513 len = mm->arg_end - mm->arg_start;
1514 if (len >= ELF_PRARGSZ)
1515 len = ELF_PRARGSZ-1;
1516 if (copy_from_user(&psinfo->pr_psargs,
1517 (const char __user *)mm->arg_start, len))
1519 for(i = 0; i < len; i++)
1520 if (psinfo->pr_psargs[i] == 0)
1521 psinfo->pr_psargs[i] = ' ';
1522 psinfo->pr_psargs[len] = 0;
1525 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1527 psinfo->pr_pid = task_pid_vnr(p);
1528 psinfo->pr_pgrp = task_pgrp_vnr(p);
1529 psinfo->pr_sid = task_session_vnr(p);
1531 i = p->state ? ffz(~p->state) + 1 : 0;
1532 psinfo->pr_state = i;
1533 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1534 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1535 psinfo->pr_nice = task_nice(p);
1536 psinfo->pr_flag = p->flags;
1538 cred = __task_cred(p);
1539 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1540 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1542 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1547 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1549 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1553 while (auxv[i - 2] != AT_NULL);
1554 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1557 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1558 const kernel_siginfo_t *siginfo)
1560 mm_segment_t old_fs = get_fs();
1562 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1564 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1567 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1569 * Format of NT_FILE note:
1571 * long count -- how many files are mapped
1572 * long page_size -- units for file_ofs
1573 * array of [COUNT] elements of
1577 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1579 static int fill_files_note(struct memelfnote *note)
1581 struct mm_struct *mm = current->mm;
1582 struct vm_area_struct *vma;
1583 unsigned count, size, names_ofs, remaining, n;
1585 user_long_t *start_end_ofs;
1586 char *name_base, *name_curpos;
1588 /* *Estimated* file count and total data size needed */
1589 count = mm->map_count;
1590 if (count > UINT_MAX / 64)
1594 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1596 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1598 size = round_up(size, PAGE_SIZE);
1600 * "size" can be 0 here legitimately.
1601 * Let it ENOMEM and omit NT_FILE section which will be empty anyway.
1603 data = kvmalloc(size, GFP_KERNEL);
1604 if (ZERO_OR_NULL_PTR(data))
1607 start_end_ofs = data + 2;
1608 name_base = name_curpos = ((char *)data) + names_ofs;
1609 remaining = size - names_ofs;
1611 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
1613 const char *filename;
1615 file = vma->vm_file;
1618 filename = file_path(file, name_curpos, remaining);
1619 if (IS_ERR(filename)) {
1620 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1622 size = size * 5 / 4;
1628 /* file_path() fills at the end, move name down */
1629 /* n = strlen(filename) + 1: */
1630 n = (name_curpos + remaining) - filename;
1631 remaining = filename - name_curpos;
1632 memmove(name_curpos, filename, n);
1635 *start_end_ofs++ = vma->vm_start;
1636 *start_end_ofs++ = vma->vm_end;
1637 *start_end_ofs++ = vma->vm_pgoff;
1641 /* Now we know exact count of files, can store it */
1643 data[1] = PAGE_SIZE;
1645 * Count usually is less than mm->map_count,
1646 * we need to move filenames down.
1648 n = mm->map_count - count;
1650 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1651 memmove(name_base - shift_bytes, name_base,
1652 name_curpos - name_base);
1653 name_curpos -= shift_bytes;
1656 size = name_curpos - (char *)data;
1657 fill_note(note, "CORE", NT_FILE, size, data);
1661 #ifdef CORE_DUMP_USE_REGSET
1662 #include <linux/regset.h>
1664 struct elf_thread_core_info {
1665 struct elf_thread_core_info *next;
1666 struct task_struct *task;
1667 struct elf_prstatus prstatus;
1668 struct memelfnote notes[0];
1671 struct elf_note_info {
1672 struct elf_thread_core_info *thread;
1673 struct memelfnote psinfo;
1674 struct memelfnote signote;
1675 struct memelfnote auxv;
1676 struct memelfnote files;
1677 user_siginfo_t csigdata;
1683 * When a regset has a writeback hook, we call it on each thread before
1684 * dumping user memory. On register window machines, this makes sure the
1685 * user memory backing the register data is up to date before we read it.
1687 static void do_thread_regset_writeback(struct task_struct *task,
1688 const struct user_regset *regset)
1690 if (regset->writeback)
1691 regset->writeback(task, regset, 1);
1694 #ifndef PRSTATUS_SIZE
1695 #define PRSTATUS_SIZE(S, R) sizeof(S)
1698 #ifndef SET_PR_FPVALID
1699 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1702 static int fill_thread_core_info(struct elf_thread_core_info *t,
1703 const struct user_regset_view *view,
1704 long signr, size_t *total)
1707 unsigned int regset0_size = regset_size(t->task, &view->regsets[0]);
1710 * NT_PRSTATUS is the one special case, because the regset data
1711 * goes into the pr_reg field inside the note contents, rather
1712 * than being the whole note contents. We fill the reset in here.
1713 * We assume that regset 0 is NT_PRSTATUS.
1715 fill_prstatus(&t->prstatus, t->task, signr);
1716 (void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset0_size,
1717 &t->prstatus.pr_reg, NULL);
1719 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1720 PRSTATUS_SIZE(t->prstatus, regset0_size), &t->prstatus);
1721 *total += notesize(&t->notes[0]);
1723 do_thread_regset_writeback(t->task, &view->regsets[0]);
1726 * Each other regset might generate a note too. For each regset
1727 * that has no core_note_type or is inactive, we leave t->notes[i]
1728 * all zero and we'll know to skip writing it later.
1730 for (i = 1; i < view->n; ++i) {
1731 const struct user_regset *regset = &view->regsets[i];
1732 do_thread_regset_writeback(t->task, regset);
1733 if (regset->core_note_type && regset->get &&
1734 (!regset->active || regset->active(t->task, regset) > 0)) {
1736 size_t size = regset_size(t->task, regset);
1737 void *data = kmalloc(size, GFP_KERNEL);
1738 if (unlikely(!data))
1740 ret = regset->get(t->task, regset,
1741 0, size, data, NULL);
1745 if (regset->core_note_type != NT_PRFPREG)
1746 fill_note(&t->notes[i], "LINUX",
1747 regset->core_note_type,
1750 SET_PR_FPVALID(&t->prstatus,
1752 fill_note(&t->notes[i], "CORE",
1753 NT_PRFPREG, size, data);
1755 *total += notesize(&t->notes[i]);
1763 static int fill_note_info(struct elfhdr *elf, int phdrs,
1764 struct elf_note_info *info,
1765 const kernel_siginfo_t *siginfo, struct pt_regs *regs)
1767 struct task_struct *dump_task = current;
1768 const struct user_regset_view *view = task_user_regset_view(dump_task);
1769 struct elf_thread_core_info *t;
1770 struct elf_prpsinfo *psinfo;
1771 struct core_thread *ct;
1775 info->thread = NULL;
1777 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1778 if (psinfo == NULL) {
1779 info->psinfo.data = NULL; /* So we don't free this wrongly */
1783 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1786 * Figure out how many notes we're going to need for each thread.
1788 info->thread_notes = 0;
1789 for (i = 0; i < view->n; ++i)
1790 if (view->regsets[i].core_note_type != 0)
1791 ++info->thread_notes;
1794 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1795 * since it is our one special case.
1797 if (unlikely(info->thread_notes == 0) ||
1798 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1804 * Initialize the ELF file header.
1806 fill_elf_header(elf, phdrs,
1807 view->e_machine, view->e_flags);
1810 * Allocate a structure for each thread.
1812 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1813 t = kzalloc(offsetof(struct elf_thread_core_info,
1814 notes[info->thread_notes]),
1820 if (ct->task == dump_task || !info->thread) {
1821 t->next = info->thread;
1825 * Make sure to keep the original task at
1826 * the head of the list.
1828 t->next = info->thread->next;
1829 info->thread->next = t;
1834 * Now fill in each thread's information.
1836 for (t = info->thread; t != NULL; t = t->next)
1837 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1841 * Fill in the two process-wide notes.
1843 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1844 info->size += notesize(&info->psinfo);
1846 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1847 info->size += notesize(&info->signote);
1849 fill_auxv_note(&info->auxv, current->mm);
1850 info->size += notesize(&info->auxv);
1852 if (fill_files_note(&info->files) == 0)
1853 info->size += notesize(&info->files);
1858 static size_t get_note_info_size(struct elf_note_info *info)
1864 * Write all the notes for each thread. When writing the first thread, the
1865 * process-wide notes are interleaved after the first thread-specific note.
1867 static int write_note_info(struct elf_note_info *info,
1868 struct coredump_params *cprm)
1871 struct elf_thread_core_info *t = info->thread;
1876 if (!writenote(&t->notes[0], cprm))
1879 if (first && !writenote(&info->psinfo, cprm))
1881 if (first && !writenote(&info->signote, cprm))
1883 if (first && !writenote(&info->auxv, cprm))
1885 if (first && info->files.data &&
1886 !writenote(&info->files, cprm))
1889 for (i = 1; i < info->thread_notes; ++i)
1890 if (t->notes[i].data &&
1891 !writenote(&t->notes[i], cprm))
1901 static void free_note_info(struct elf_note_info *info)
1903 struct elf_thread_core_info *threads = info->thread;
1906 struct elf_thread_core_info *t = threads;
1908 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1909 for (i = 1; i < info->thread_notes; ++i)
1910 kfree(t->notes[i].data);
1913 kfree(info->psinfo.data);
1914 kvfree(info->files.data);
1919 /* Here is the structure in which status of each thread is captured. */
1920 struct elf_thread_status
1922 struct list_head list;
1923 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1924 elf_fpregset_t fpu; /* NT_PRFPREG */
1925 struct task_struct *thread;
1926 #ifdef ELF_CORE_COPY_XFPREGS
1927 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1929 struct memelfnote notes[3];
1934 * In order to add the specific thread information for the elf file format,
1935 * we need to keep a linked list of every threads pr_status and then create
1936 * a single section for them in the final core file.
1938 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1941 struct task_struct *p = t->thread;
1944 fill_prstatus(&t->prstatus, p, signr);
1945 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1947 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1950 sz += notesize(&t->notes[0]);
1952 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1954 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1957 sz += notesize(&t->notes[1]);
1960 #ifdef ELF_CORE_COPY_XFPREGS
1961 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1962 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1963 sizeof(t->xfpu), &t->xfpu);
1965 sz += notesize(&t->notes[2]);
1971 struct elf_note_info {
1972 struct memelfnote *notes;
1973 struct memelfnote *notes_files;
1974 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1975 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1976 struct list_head thread_list;
1977 elf_fpregset_t *fpu;
1978 #ifdef ELF_CORE_COPY_XFPREGS
1979 elf_fpxregset_t *xfpu;
1981 user_siginfo_t csigdata;
1982 int thread_status_size;
1986 static int elf_note_info_init(struct elf_note_info *info)
1988 memset(info, 0, sizeof(*info));
1989 INIT_LIST_HEAD(&info->thread_list);
1991 /* Allocate space for ELF notes */
1992 info->notes = kmalloc_array(8, sizeof(struct memelfnote), GFP_KERNEL);
1995 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1998 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1999 if (!info->prstatus)
2001 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2004 #ifdef ELF_CORE_COPY_XFPREGS
2005 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
2012 static int fill_note_info(struct elfhdr *elf, int phdrs,
2013 struct elf_note_info *info,
2014 const kernel_siginfo_t *siginfo, struct pt_regs *regs)
2016 struct core_thread *ct;
2017 struct elf_thread_status *ets;
2019 if (!elf_note_info_init(info))
2022 for (ct = current->mm->core_state->dumper.next;
2023 ct; ct = ct->next) {
2024 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2028 ets->thread = ct->task;
2029 list_add(&ets->list, &info->thread_list);
2032 list_for_each_entry(ets, &info->thread_list, list) {
2035 sz = elf_dump_thread_status(siginfo->si_signo, ets);
2036 info->thread_status_size += sz;
2038 /* now collect the dump for the current */
2039 memset(info->prstatus, 0, sizeof(*info->prstatus));
2040 fill_prstatus(info->prstatus, current, siginfo->si_signo);
2041 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2044 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2047 * Set up the notes in similar form to SVR4 core dumps made
2048 * with info from their /proc.
2051 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2052 sizeof(*info->prstatus), info->prstatus);
2053 fill_psinfo(info->psinfo, current->group_leader, current->mm);
2054 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2055 sizeof(*info->psinfo), info->psinfo);
2057 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2058 fill_auxv_note(info->notes + 3, current->mm);
2061 if (fill_files_note(info->notes + info->numnote) == 0) {
2062 info->notes_files = info->notes + info->numnote;
2066 /* Try to dump the FPU. */
2067 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2069 if (info->prstatus->pr_fpvalid)
2070 fill_note(info->notes + info->numnote++,
2071 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2072 #ifdef ELF_CORE_COPY_XFPREGS
2073 if (elf_core_copy_task_xfpregs(current, info->xfpu))
2074 fill_note(info->notes + info->numnote++,
2075 "LINUX", ELF_CORE_XFPREG_TYPE,
2076 sizeof(*info->xfpu), info->xfpu);
2082 static size_t get_note_info_size(struct elf_note_info *info)
2087 for (i = 0; i < info->numnote; i++)
2088 sz += notesize(info->notes + i);
2090 sz += info->thread_status_size;
2095 static int write_note_info(struct elf_note_info *info,
2096 struct coredump_params *cprm)
2098 struct elf_thread_status *ets;
2101 for (i = 0; i < info->numnote; i++)
2102 if (!writenote(info->notes + i, cprm))
2105 /* write out the thread status notes section */
2106 list_for_each_entry(ets, &info->thread_list, list) {
2107 for (i = 0; i < ets->num_notes; i++)
2108 if (!writenote(&ets->notes[i], cprm))
2115 static void free_note_info(struct elf_note_info *info)
2117 while (!list_empty(&info->thread_list)) {
2118 struct list_head *tmp = info->thread_list.next;
2120 kfree(list_entry(tmp, struct elf_thread_status, list));
2123 /* Free data possibly allocated by fill_files_note(): */
2124 if (info->notes_files)
2125 kvfree(info->notes_files->data);
2127 kfree(info->prstatus);
2128 kfree(info->psinfo);
2131 #ifdef ELF_CORE_COPY_XFPREGS
2138 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2139 struct vm_area_struct *gate_vma)
2141 struct vm_area_struct *ret = tsk->mm->mmap;
2148 * Helper function for iterating across a vma list. It ensures that the caller
2149 * will visit `gate_vma' prior to terminating the search.
2151 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2152 struct vm_area_struct *gate_vma)
2154 struct vm_area_struct *ret;
2156 ret = this_vma->vm_next;
2159 if (this_vma == gate_vma)
2164 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2165 elf_addr_t e_shoff, int segs)
2167 elf->e_shoff = e_shoff;
2168 elf->e_shentsize = sizeof(*shdr4extnum);
2170 elf->e_shstrndx = SHN_UNDEF;
2172 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2174 shdr4extnum->sh_type = SHT_NULL;
2175 shdr4extnum->sh_size = elf->e_shnum;
2176 shdr4extnum->sh_link = elf->e_shstrndx;
2177 shdr4extnum->sh_info = segs;
2183 * This is a two-pass process; first we find the offsets of the bits,
2184 * and then they are actually written out. If we run out of core limit
2187 static int elf_core_dump(struct coredump_params *cprm)
2192 size_t vma_data_size = 0;
2193 struct vm_area_struct *vma, *gate_vma;
2195 loff_t offset = 0, dataoff;
2196 struct elf_note_info info = { };
2197 struct elf_phdr *phdr4note = NULL;
2198 struct elf_shdr *shdr4extnum = NULL;
2201 elf_addr_t *vma_filesz = NULL;
2204 * We no longer stop all VM operations.
2206 * This is because those proceses that could possibly change map_count
2207 * or the mmap / vma pages are now blocked in do_exit on current
2208 * finishing this core dump.
2210 * Only ptrace can touch these memory addresses, but it doesn't change
2211 * the map_count or the pages allocated. So no possibility of crashing
2212 * exists while dumping the mm->vm_next areas to the core file.
2216 * The number of segs are recored into ELF header as 16bit value.
2217 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2219 segs = current->mm->map_count;
2220 segs += elf_core_extra_phdrs();
2222 gate_vma = get_gate_vma(current->mm);
2223 if (gate_vma != NULL)
2226 /* for notes section */
2229 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2230 * this, kernel supports extended numbering. Have a look at
2231 * include/linux/elf.h for further information. */
2232 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2235 * Collect all the non-memory information about the process for the
2236 * notes. This also sets up the file header.
2238 if (!fill_note_info(&elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2246 offset += sizeof(elf); /* Elf header */
2247 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2249 /* Write notes phdr entry */
2251 size_t sz = get_note_info_size(&info);
2253 sz += elf_coredump_extra_notes_size();
2255 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2259 fill_elf_note_phdr(phdr4note, sz, offset);
2263 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2266 * Zero vma process will get ZERO_SIZE_PTR here.
2267 * Let coredump continue for register state at least.
2269 vma_filesz = kvmalloc(array_size(sizeof(*vma_filesz), (segs - 1)),
2274 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2275 vma = next_vma(vma, gate_vma)) {
2276 unsigned long dump_size;
2278 dump_size = vma_dump_size(vma, cprm->mm_flags);
2279 vma_filesz[i++] = dump_size;
2280 vma_data_size += dump_size;
2283 offset += vma_data_size;
2284 offset += elf_core_extra_data_size();
2287 if (e_phnum == PN_XNUM) {
2288 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2291 fill_extnum_info(&elf, shdr4extnum, e_shoff, segs);
2296 if (!dump_emit(cprm, &elf, sizeof(elf)))
2299 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2302 /* Write program headers for segments dump */
2303 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2304 vma = next_vma(vma, gate_vma)) {
2305 struct elf_phdr phdr;
2307 phdr.p_type = PT_LOAD;
2308 phdr.p_offset = offset;
2309 phdr.p_vaddr = vma->vm_start;
2311 phdr.p_filesz = vma_filesz[i++];
2312 phdr.p_memsz = vma->vm_end - vma->vm_start;
2313 offset += phdr.p_filesz;
2314 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2315 if (vma->vm_flags & VM_WRITE)
2316 phdr.p_flags |= PF_W;
2317 if (vma->vm_flags & VM_EXEC)
2318 phdr.p_flags |= PF_X;
2319 phdr.p_align = ELF_EXEC_PAGESIZE;
2321 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2325 if (!elf_core_write_extra_phdrs(cprm, offset))
2328 /* write out the notes section */
2329 if (!write_note_info(&info, cprm))
2332 if (elf_coredump_extra_notes_write(cprm))
2336 if (!dump_skip(cprm, dataoff - cprm->pos))
2339 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2340 vma = next_vma(vma, gate_vma)) {
2344 end = vma->vm_start + vma_filesz[i++];
2346 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2350 page = get_dump_page(addr);
2352 void *kaddr = kmap(page);
2353 stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2357 stop = !dump_skip(cprm, PAGE_SIZE);
2362 dump_truncate(cprm);
2364 if (!elf_core_write_extra_data(cprm))
2367 if (e_phnum == PN_XNUM) {
2368 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2376 free_note_info(&info);
2383 #endif /* CONFIG_ELF_CORE */
2385 static int __init init_elf_binfmt(void)
2387 register_binfmt(&elf_format);
2391 static void __exit exit_elf_binfmt(void)
2393 /* Remove the COFF and ELF loaders. */
2394 unregister_binfmt(&elf_format);
2397 core_initcall(init_elf_binfmt);
2398 module_exit(exit_elf_binfmt);
2399 MODULE_LICENSE("GPL");