1/*
2 * linux/fs/binfmt_elf.c
3 *
4 * These are the functions used to load ELF format executables as used
5 * on SVr4 machines. Information on the format may be found in the book
6 * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
7 * Tools".
8 *
9 * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
10 */
11
12#include <linux/module.h>
13#include <linux/kernel.h>
14#include <linux/fs.h>
15#include <linux/mm.h>
16#include <linux/mman.h>
17#include <linux/errno.h>
18#include <linux/signal.h>
19#include <linux/binfmts.h>
20#include <linux/string.h>
21#include <linux/file.h>
22#include <linux/slab.h>
23#include <linux/personality.h>
24#include <linux/elfcore.h>
25#include <linux/init.h>
26#include <linux/highuid.h>
27#include <linux/compiler.h>
28#include <linux/highmem.h>
29#include <linux/pagemap.h>
30#include <linux/vmalloc.h>
31#include <linux/security.h>
32#include <linux/random.h>
33#include <linux/elf.h>
34#include <linux/elf-randomize.h>
35#include <linux/utsname.h>
36#include <linux/coredump.h>
37#include <linux/sched.h>
38#include <linux/sched/coredump.h>
39#include <linux/sched/task_stack.h>
40#include <linux/sched/cputime.h>
41#include <linux/cred.h>
42#include <linux/dax.h>
43#include <linux/uaccess.h>
44#include <asm/param.h>
45#include <asm/page.h>
46
47#ifndef user_long_t
48#define user_long_t long
49#endif
50#ifndef user_siginfo_t
51#define user_siginfo_t siginfo_t
52#endif
53
54/* That's for binfmt_elf_fdpic to deal with */
55#ifndef elf_check_fdpic
56#define elf_check_fdpic(ex) false
57#endif
58
59static int load_elf_binary(struct linux_binprm *bprm);
60static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
61 int, int, unsigned long);
62
63#ifdef CONFIG_USELIB
64static int load_elf_library(struct file *);
65#else
66#define load_elf_library NULL
67#endif
68
69/*
70 * If we don't support core dumping, then supply a NULL so we
71 * don't even try.
72 */
73#ifdef CONFIG_ELF_CORE
74static int elf_core_dump(struct coredump_params *cprm);
75#else
76#define elf_core_dump NULL
77#endif
78
79#if ELF_EXEC_PAGESIZE > PAGE_SIZE
80#define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
81#else
82#define ELF_MIN_ALIGN PAGE_SIZE
83#endif
84
85#ifndef ELF_CORE_EFLAGS
86#define ELF_CORE_EFLAGS 0
87#endif
88
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))
92
93static 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,
99};
100
101#define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
102
103static int set_brk(unsigned long start, unsigned long end, int prot)
104{
105 start = ELF_PAGEALIGN(start);
106 end = ELF_PAGEALIGN(end);
107 if (end > start) {
108 /*
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).
112 */
113 int error = vm_brk_flags(start, end - start,
114 prot & PROT_EXEC ? VM_EXEC : 0);
115 if (error)
116 return error;
117 }
118 current->mm->start_brk = current->mm->brk = end;
119 return 0;
120}
121
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
125 be in memory
126 */
127static int padzero(unsigned long elf_bss)
128{
129 unsigned long nbyte;
130
131 nbyte = ELF_PAGEOFFSET(elf_bss);
132 if (nbyte) {
133 nbyte = ELF_MIN_ALIGN - nbyte;
134 if (clear_user((void __user *) elf_bss, nbyte))
135 return -EFAULT;
136 }
137 return 0;
138}
139
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; \
147 old_sp; })
148#else
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; })
153#endif
154
155#ifndef ELF_BASE_PLATFORM
156/*
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.
160 */
161#define ELF_BASE_PLATFORM NULL
162#endif
163
164static int
165create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
166 unsigned long load_addr, unsigned long interp_load_addr)
167{
168 unsigned long p = bprm->p;
169 int argc = bprm->argc;
170 int envc = bprm->envc;
171 elf_addr_t __user *sp;
172 elf_addr_t __user *u_platform;
173 elf_addr_t __user *u_base_platform;
174 elf_addr_t __user *u_rand_bytes;
175 const char *k_platform = ELF_PLATFORM;
176 const char *k_base_platform = ELF_BASE_PLATFORM;
177 unsigned char k_rand_bytes[16];
178 int items;
179 elf_addr_t *elf_info;
180 int ei_index = 0;
181 const struct cred *cred = current_cred();
182 struct vm_area_struct *vma;
183
184 /*
185 * In some cases (e.g. Hyper-Threading), we want to avoid L1
186 * evictions by the processes running on the same package. One
187 * thing we can do is to shuffle the initial stack for them.
188 */
189
190 p = arch_align_stack(p);
191
192 /*
193 * If this architecture has a platform capability string, copy it
194 * to userspace. In some cases (Sparc), this info is impossible
195 * for userspace to get any other way, in others (i386) it is
196 * merely difficult.
197 */
198 u_platform = NULL;
199 if (k_platform) {
200 size_t len = strlen(k_platform) + 1;
201
202 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
203 if (__copy_to_user(u_platform, k_platform, len))
204 return -EFAULT;
205 }
206
207 /*
208 * If this architecture has a "base" platform capability
209 * string, copy it to userspace.
210 */
211 u_base_platform = NULL;
212 if (k_base_platform) {
213 size_t len = strlen(k_base_platform) + 1;
214
215 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
216 if (__copy_to_user(u_base_platform, k_base_platform, len))
217 return -EFAULT;
218 }
219
220 /*
221 * Generate 16 random bytes for userspace PRNG seeding.
222 */
223 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
224 u_rand_bytes = (elf_addr_t __user *)
225 STACK_ALLOC(p, sizeof(k_rand_bytes));
226 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
227 return -EFAULT;
228
229 /* Create the ELF interpreter info */
230 elf_info = (elf_addr_t *)current->mm->saved_auxv;
231 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
232#define NEW_AUX_ENT(id, val) \
233 do { \
234 elf_info[ei_index++] = id; \
235 elf_info[ei_index++] = val; \
236 } while (0)
237
238#ifdef ARCH_DLINFO
239 /*
240 * ARCH_DLINFO must come first so PPC can do its special alignment of
241 * AUXV.
242 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
243 * ARCH_DLINFO changes
244 */
245 ARCH_DLINFO;
246#endif
247 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
248 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
249 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
250 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
251 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
252 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
253 NEW_AUX_ENT(AT_BASE, interp_load_addr);
254 NEW_AUX_ENT(AT_FLAGS, 0);
255 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
256 NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
257 NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
258 NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
259 NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
260 NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
261 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
262#ifdef ELF_HWCAP2
263 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
264#endif
265 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
266 if (k_platform) {
267 NEW_AUX_ENT(AT_PLATFORM,
268 (elf_addr_t)(unsigned long)u_platform);
269 }
270 if (k_base_platform) {
271 NEW_AUX_ENT(AT_BASE_PLATFORM,
272 (elf_addr_t)(unsigned long)u_base_platform);
273 }
274 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
275 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
276 }
277#undef NEW_AUX_ENT
278 /* AT_NULL is zero; clear the rest too */
279 memset(&elf_info[ei_index], 0,
280 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
281
282 /* And advance past the AT_NULL entry. */
283 ei_index += 2;
284
285 sp = STACK_ADD(p, ei_index);
286
287 items = (argc + 1) + (envc + 1) + 1;
288 bprm->p = STACK_ROUND(sp, items);
289
290 /* Point sp at the lowest address on the stack */
291#ifdef CONFIG_STACK_GROWSUP
292 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
293 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
294#else
295 sp = (elf_addr_t __user *)bprm->p;
296#endif
297
298
299 /*
300 * Grow the stack manually; some architectures have a limit on how
301 * far ahead a user-space access may be in order to grow the stack.
302 */
303 vma = find_extend_vma(current->mm, bprm->p);
304 if (!vma)
305 return -EFAULT;
306
307 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
308 if (__put_user(argc, sp++))
309 return -EFAULT;
310
311 /* Populate list of argv pointers back to argv strings. */
312 p = current->mm->arg_end = current->mm->arg_start;
313 while (argc-- > 0) {
314 size_t len;
315 if (__put_user((elf_addr_t)p, sp++))
316 return -EFAULT;
317 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
318 if (!len || len > MAX_ARG_STRLEN)
319 return -EINVAL;
320 p += len;
321 }
322 if (__put_user(0, sp++))
323 return -EFAULT;
324 current->mm->arg_end = p;
325
326 /* Populate list of envp pointers back to envp strings. */
327 current->mm->env_end = current->mm->env_start = p;
328 while (envc-- > 0) {
329 size_t len;
330 if (__put_user((elf_addr_t)p, sp++))
331 return -EFAULT;
332 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
333 if (!len || len > MAX_ARG_STRLEN)
334 return -EINVAL;
335 p += len;
336 }
337 if (__put_user(0, sp++))
338 return -EFAULT;
339 current->mm->env_end = p;
340
341 /* Put the elf_info on the stack in the right place. */
342 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
343 return -EFAULT;
344 return 0;
345}
346
347#ifndef elf_map
348
349static unsigned long elf_map(struct file *filep, unsigned long addr,
350 struct elf_phdr *eppnt, int prot, int type,
351 unsigned long total_size)
352{
353 unsigned long map_addr;
354 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
355 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
356 addr = ELF_PAGESTART(addr);
357 size = ELF_PAGEALIGN(size);
358
359 /* mmap() will return -EINVAL if given a zero size, but a
360 * segment with zero filesize is perfectly valid */
361 if (!size)
362 return addr;
363
364 /*
365 * total_size is the size of the ELF (interpreter) image.
366 * The _first_ mmap needs to know the full size, otherwise
367 * randomization might put this image into an overlapping
368 * position with the ELF binary image. (since size < total_size)
369 * So we first map the 'big' image - and unmap the remainder at
370 * the end. (which unmap is needed for ELF images with holes.)
371 */
372 if (total_size) {
373 total_size = ELF_PAGEALIGN(total_size);
374 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
375 if (!BAD_ADDR(map_addr))
376 vm_munmap(map_addr+size, total_size-size);
377 } else
378 map_addr = vm_mmap(filep, addr, size, prot, type, off);
379
380 if ((type & MAP_FIXED_NOREPLACE) &&
381 PTR_ERR((void *)map_addr) == -EEXIST)
382 pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
383 task_pid_nr(current), current->comm, (void *)addr);
384
385 return(map_addr);
386}
387
388#endif /* !elf_map */
389
390static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
391{
392 int i, first_idx = -1, last_idx = -1;
393
394 for (i = 0; i < nr; i++) {
395 if (cmds[i].p_type == PT_LOAD) {
396 last_idx = i;
397 if (first_idx == -1)
398 first_idx = i;
399 }
400 }
401 if (first_idx == -1)
402 return 0;
403
404 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
405 ELF_PAGESTART(cmds[first_idx].p_vaddr);
406}
407
408/**
409 * load_elf_phdrs() - load ELF program headers
410 * @elf_ex: ELF header of the binary whose program headers should be loaded
411 * @elf_file: the opened ELF binary file
412 *
413 * Loads ELF program headers from the binary file elf_file, which has the ELF
414 * header pointed to by elf_ex, into a newly allocated array. The caller is
415 * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
416 */
417static struct elf_phdr *load_elf_phdrs(struct elfhdr *elf_ex,
418 struct file *elf_file)
419{
420 struct elf_phdr *elf_phdata = NULL;
421 int retval, size, err = -1;
422 loff_t pos = elf_ex->e_phoff;
423
424 /*
425 * If the size of this structure has changed, then punt, since
426 * we will be doing the wrong thing.
427 */
428 if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
429 goto out;
430
431 /* Sanity check the number of program headers... */
432 if (elf_ex->e_phnum < 1 ||
433 elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
434 goto out;
435
436 /* ...and their total size. */
437 size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
438 if (size > ELF_MIN_ALIGN)
439 goto out;
440
441 elf_phdata = kmalloc(size, GFP_KERNEL);
442 if (!elf_phdata)
443 goto out;
444
445 /* Read in the program headers */
446 retval = kernel_read(elf_file, elf_phdata, size, &pos);
447 if (retval != size) {
448 err = (retval < 0) ? retval : -EIO;
449 goto out;
450 }
451
452 /* Success! */
453 err = 0;
454out:
455 if (err) {
456 kfree(elf_phdata);
457 elf_phdata = NULL;
458 }
459 return elf_phdata;
460}
461
462#ifndef CONFIG_ARCH_BINFMT_ELF_STATE
463
464/**
465 * struct arch_elf_state - arch-specific ELF loading state
466 *
467 * This structure is used to preserve architecture specific data during
468 * the loading of an ELF file, throughout the checking of architecture
469 * specific ELF headers & through to the point where the ELF load is
470 * known to be proceeding (ie. SET_PERSONALITY).
471 *
472 * This implementation is a dummy for architectures which require no
473 * specific state.
474 */
475struct arch_elf_state {
476};
477
478#define INIT_ARCH_ELF_STATE {}
479
480/**
481 * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
482 * @ehdr: The main ELF header
483 * @phdr: The program header to check
484 * @elf: The open ELF file
485 * @is_interp: True if the phdr is from the interpreter of the ELF being
486 * loaded, else false.
487 * @state: Architecture-specific state preserved throughout the process
488 * of loading the ELF.
489 *
490 * Inspects the program header phdr to validate its correctness and/or
491 * suitability for the system. Called once per ELF program header in the
492 * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
493 * interpreter.
494 *
495 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
496 * with that return code.
497 */
498static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
499 struct elf_phdr *phdr,
500 struct file *elf, bool is_interp,
501 struct arch_elf_state *state)
502{
503 /* Dummy implementation, always proceed */
504 return 0;
505}
506
507/**
508 * arch_check_elf() - check an ELF executable
509 * @ehdr: The main ELF header
510 * @has_interp: True if the ELF has an interpreter, else false.
511 * @interp_ehdr: The interpreter's ELF header
512 * @state: Architecture-specific state preserved throughout the process
513 * of loading the ELF.
514 *
515 * Provides a final opportunity for architecture code to reject the loading
516 * of the ELF & cause an exec syscall to return an error. This is called after
517 * all program headers to be checked by arch_elf_pt_proc have been.
518 *
519 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
520 * with that return code.
521 */
522static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
523 struct elfhdr *interp_ehdr,
524 struct arch_elf_state *state)
525{
526 /* Dummy implementation, always proceed */
527 return 0;
528}
529
530#endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
531
532/* This is much more generalized than the library routine read function,
533 so we keep this separate. Technically the library read function
534 is only provided so that we can read a.out libraries that have
535 an ELF header */
536
537static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
538 struct file *interpreter, unsigned long *interp_map_addr,
539 unsigned long no_base, struct elf_phdr *interp_elf_phdata)
540{
541 struct elf_phdr *eppnt;
542 unsigned long load_addr = 0;
543 int load_addr_set = 0;
544 unsigned long last_bss = 0, elf_bss = 0;
545 int bss_prot = 0;
546 unsigned long error = ~0UL;
547 unsigned long total_size;
548 int i;
549
550 /* First of all, some simple consistency checks */
551 if (interp_elf_ex->e_type != ET_EXEC &&
552 interp_elf_ex->e_type != ET_DYN)
553 goto out;
554 if (!elf_check_arch(interp_elf_ex) ||
555 elf_check_fdpic(interp_elf_ex))
556 goto out;
557 if (!interpreter->f_op->mmap)
558 goto out;
559
560 total_size = total_mapping_size(interp_elf_phdata,
561 interp_elf_ex->e_phnum);
562 if (!total_size) {
563 error = -EINVAL;
564 goto out;
565 }
566
567 eppnt = interp_elf_phdata;
568 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
569 if (eppnt->p_type == PT_LOAD) {
570 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
571 int elf_prot = 0;
572 unsigned long vaddr = 0;
573 unsigned long k, map_addr;
574
575 if (eppnt->p_flags & PF_R)
576 elf_prot = PROT_READ;
577 if (eppnt->p_flags & PF_W)
578 elf_prot |= PROT_WRITE;
579 if (eppnt->p_flags & PF_X)
580 elf_prot |= PROT_EXEC;
581 vaddr = eppnt->p_vaddr;
582 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
583 elf_type |= MAP_FIXED_NOREPLACE;
584 else if (no_base && interp_elf_ex->e_type == ET_DYN)
585 load_addr = -vaddr;
586
587 map_addr = elf_map(interpreter, load_addr + vaddr,
588 eppnt, elf_prot, elf_type, total_size);
589 total_size = 0;
590 if (!*interp_map_addr)
591 *interp_map_addr = map_addr;
592 error = map_addr;
593 if (BAD_ADDR(map_addr))
594 goto out;
595
596 if (!load_addr_set &&
597 interp_elf_ex->e_type == ET_DYN) {
598 load_addr = map_addr - ELF_PAGESTART(vaddr);
599 load_addr_set = 1;
600 }
601
602 /*
603 * Check to see if the section's size will overflow the
604 * allowed task size. Note that p_filesz must always be
605 * <= p_memsize so it's only necessary to check p_memsz.
606 */
607 k = load_addr + eppnt->p_vaddr;
608 if (BAD_ADDR(k) ||
609 eppnt->p_filesz > eppnt->p_memsz ||
610 eppnt->p_memsz > TASK_SIZE ||
611 TASK_SIZE - eppnt->p_memsz < k) {
612 error = -ENOMEM;
613 goto out;
614 }
615
616 /*
617 * Find the end of the file mapping for this phdr, and
618 * keep track of the largest address we see for this.
619 */
620 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
621 if (k > elf_bss)
622 elf_bss = k;
623
624 /*
625 * Do the same thing for the memory mapping - between
626 * elf_bss and last_bss is the bss section.
627 */
628 k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
629 if (k > last_bss) {
630 last_bss = k;
631 bss_prot = elf_prot;
632 }
633 }
634 }
635
636 /*
637 * Now fill out the bss section: first pad the last page from
638 * the file up to the page boundary, and zero it from elf_bss
639 * up to the end of the page.
640 */
641 if (padzero(elf_bss)) {
642 error = -EFAULT;
643 goto out;
644 }
645 /*
646 * Next, align both the file and mem bss up to the page size,
647 * since this is where elf_bss was just zeroed up to, and where
648 * last_bss will end after the vm_brk_flags() below.
649 */
650 elf_bss = ELF_PAGEALIGN(elf_bss);
651 last_bss = ELF_PAGEALIGN(last_bss);
652 /* Finally, if there is still more bss to allocate, do it. */
653 if (last_bss > elf_bss) {
654 error = vm_brk_flags(elf_bss, last_bss - elf_bss,
655 bss_prot & PROT_EXEC ? VM_EXEC : 0);
656 if (error)
657 goto out;
658 }
659
660 error = load_addr;
661out:
662 return error;
663}
664
665/*
666 * These are the functions used to load ELF style executables and shared
667 * libraries. There is no binary dependent code anywhere else.
668 */
669
670#ifndef STACK_RND_MASK
671#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
672#endif
673
674static unsigned long randomize_stack_top(unsigned long stack_top)
675{
676 unsigned long random_variable = 0;
677
678 if (current->flags & PF_RANDOMIZE) {
679 random_variable = get_random_long();
680 random_variable &= STACK_RND_MASK;
681 random_variable <<= PAGE_SHIFT;
682 }
683#ifdef CONFIG_STACK_GROWSUP
684 return PAGE_ALIGN(stack_top) + random_variable;
685#else
686 return PAGE_ALIGN(stack_top) - random_variable;
687#endif
688}
689
690static int load_elf_binary(struct linux_binprm *bprm)
691{
692 struct file *interpreter = NULL; /* to shut gcc up */
693 unsigned long load_addr = 0, load_bias = 0;
694 int load_addr_set = 0;
695 char * elf_interpreter = NULL;
696 unsigned long error;
697 struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
698 unsigned long elf_bss, elf_brk;
699 int bss_prot = 0;
700 int retval, i;
701 unsigned long elf_entry;
702 unsigned long interp_load_addr = 0;
703 unsigned long start_code, end_code, start_data, end_data;
704 unsigned long reloc_func_desc __maybe_unused = 0;
705 int executable_stack = EXSTACK_DEFAULT;
706 struct pt_regs *regs = current_pt_regs();
707 struct {
708 struct elfhdr elf_ex;
709 struct elfhdr interp_elf_ex;
710 } *loc;
711 struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
712 loff_t pos;
713
714 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
715 if (!loc) {
716 retval = -ENOMEM;
717 goto out_ret;
718 }
719
720 /* Get the exec-header */
721 loc->elf_ex = *((struct elfhdr *)bprm->buf);
722
723 retval = -ENOEXEC;
724 /* First of all, some simple consistency checks */
725 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
726 goto out;
727
728 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
729 goto out;
730 if (!elf_check_arch(&loc->elf_ex))
731 goto out;
732 if (elf_check_fdpic(&loc->elf_ex))
733 goto out;
734 if (!bprm->file->f_op->mmap)
735 goto out;
736
737 elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
738 if (!elf_phdata)
739 goto out;
740
741 elf_ppnt = elf_phdata;
742 elf_bss = 0;
743 elf_brk = 0;
744
745 start_code = ~0UL;
746 end_code = 0;
747 start_data = 0;
748 end_data = 0;
749
750 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
751 if (elf_ppnt->p_type == PT_INTERP) {
752 /* This is the program interpreter used for
753 * shared libraries - for now assume that this
754 * is an a.out format binary
755 */
756 retval = -ENOEXEC;
757 if (elf_ppnt->p_filesz > PATH_MAX ||
758 elf_ppnt->p_filesz < 2)
759 goto out_free_ph;
760
761 retval = -ENOMEM;
762 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
763 GFP_KERNEL);
764 if (!elf_interpreter)
765 goto out_free_ph;
766
767 pos = elf_ppnt->p_offset;
768 retval = kernel_read(bprm->file, elf_interpreter,
769 elf_ppnt->p_filesz, &pos);
770 if (retval != elf_ppnt->p_filesz) {
771 if (retval >= 0)
772 retval = -EIO;
773 goto out_free_interp;
774 }
775 /* make sure path is NULL terminated */
776 retval = -ENOEXEC;
777 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
778 goto out_free_interp;
779
780 interpreter = open_exec(elf_interpreter);
781 retval = PTR_ERR(interpreter);
782 if (IS_ERR(interpreter))
783 goto out_free_interp;
784
785 /*
786 * If the binary is not readable then enforce
787 * mm->dumpable = 0 regardless of the interpreter's
788 * permissions.
789 */
790 would_dump(bprm, interpreter);
791
792 /* Get the exec headers */
793 pos = 0;
794 retval = kernel_read(interpreter, &loc->interp_elf_ex,
795 sizeof(loc->interp_elf_ex), &pos);
796 if (retval != sizeof(loc->interp_elf_ex)) {
797 if (retval >= 0)
798 retval = -EIO;
799 goto out_free_dentry;
800 }
801
802 break;
803 }
804 elf_ppnt++;
805 }
806
807 elf_ppnt = elf_phdata;
808 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
809 switch (elf_ppnt->p_type) {
810 case PT_GNU_STACK:
811 if (elf_ppnt->p_flags & PF_X)
812 executable_stack = EXSTACK_ENABLE_X;
813 else
814 executable_stack = EXSTACK_DISABLE_X;
815 break;
816
817 case PT_LOPROC ... PT_HIPROC:
818 retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
819 bprm->file, false,
820 &arch_state);
821 if (retval)
822 goto out_free_dentry;
823 break;
824 }
825
826 /* Some simple consistency checks for the interpreter */
827 if (elf_interpreter) {
828 retval = -ELIBBAD;
829 /* Not an ELF interpreter */
830 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
831 goto out_free_dentry;
832 /* Verify the interpreter has a valid arch */
833 if (!elf_check_arch(&loc->interp_elf_ex) ||
834 elf_check_fdpic(&loc->interp_elf_ex))
835 goto out_free_dentry;
836
837 /* Load the interpreter program headers */
838 interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
839 interpreter);
840 if (!interp_elf_phdata)
841 goto out_free_dentry;
842
843 /* Pass PT_LOPROC..PT_HIPROC headers to arch code */
844 elf_ppnt = interp_elf_phdata;
845 for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
846 switch (elf_ppnt->p_type) {
847 case PT_LOPROC ... PT_HIPROC:
848 retval = arch_elf_pt_proc(&loc->interp_elf_ex,
849 elf_ppnt, interpreter,
850 true, &arch_state);
851 if (retval)
852 goto out_free_dentry;
853 break;
854 }
855 }
856
857 /*
858 * Allow arch code to reject the ELF at this point, whilst it's
859 * still possible to return an error to the code that invoked
860 * the exec syscall.
861 */
862 retval = arch_check_elf(&loc->elf_ex,
863 !!interpreter, &loc->interp_elf_ex,
864 &arch_state);
865 if (retval)
866 goto out_free_dentry;
867
868 /* Flush all traces of the currently running executable */
869 retval = flush_old_exec(bprm);
870 if (retval)
871 goto out_free_dentry;
872
873 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
874 may depend on the personality. */
875 SET_PERSONALITY2(loc->elf_ex, &arch_state);
876 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
877 current->personality |= READ_IMPLIES_EXEC;
878
879 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
880 current->flags |= PF_RANDOMIZE;
881
882 setup_new_exec(bprm);
883 install_exec_creds(bprm);
884
885 /* Do this so that we can load the interpreter, if need be. We will
886 change some of these later */
887 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
888 executable_stack);
889 if (retval < 0)
890 goto out_free_dentry;
891
892 current->mm->start_stack = bprm->p;
893
894 /* Now we do a little grungy work by mmapping the ELF image into
895 the correct location in memory. */
896 for(i = 0, elf_ppnt = elf_phdata;
897 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
898 int elf_prot = 0, elf_flags, elf_fixed = MAP_FIXED_NOREPLACE;
899 unsigned long k, vaddr;
900 unsigned long total_size = 0;
901
902 if (elf_ppnt->p_type != PT_LOAD)
903 continue;
904
905 if (unlikely (elf_brk > elf_bss)) {
906 unsigned long nbyte;
907
908 /* There was a PT_LOAD segment with p_memsz > p_filesz
909 before this one. Map anonymous pages, if needed,
910 and clear the area. */
911 retval = set_brk(elf_bss + load_bias,
912 elf_brk + load_bias,
913 bss_prot);
914 if (retval)
915 goto out_free_dentry;
916 nbyte = ELF_PAGEOFFSET(elf_bss);
917 if (nbyte) {
918 nbyte = ELF_MIN_ALIGN - nbyte;
919 if (nbyte > elf_brk - elf_bss)
920 nbyte = elf_brk - elf_bss;
921 if (clear_user((void __user *)elf_bss +
922 load_bias, nbyte)) {
923 /*
924 * This bss-zeroing can fail if the ELF
925 * file specifies odd protections. So
926 * we don't check the return value
927 */
928 }
929 }
930
931 /*
932 * Some binaries have overlapping elf segments and then
933 * we have to forcefully map over an existing mapping
934 * e.g. over this newly established brk mapping.
935 */
936 elf_fixed = MAP_FIXED;
937 }
938
939 if (elf_ppnt->p_flags & PF_R)
940 elf_prot |= PROT_READ;
941 if (elf_ppnt->p_flags & PF_W)
942 elf_prot |= PROT_WRITE;
943 if (elf_ppnt->p_flags & PF_X)
944 elf_prot |= PROT_EXEC;
945
946 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
947
948 vaddr = elf_ppnt->p_vaddr;
949 /*
950 * If we are loading ET_EXEC or we have already performed
951 * the ET_DYN load_addr calculations, proceed normally.
952 */
953 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
954 elf_flags |= elf_fixed;
955 } else if (loc->elf_ex.e_type == ET_DYN) {
956 /*
957 * This logic is run once for the first LOAD Program
958 * Header for ET_DYN binaries to calculate the
959 * randomization (load_bias) for all the LOAD
960 * Program Headers, and to calculate the entire
961 * size of the ELF mapping (total_size). (Note that
962 * load_addr_set is set to true later once the
963 * initial mapping is performed.)
964 *
965 * There are effectively two types of ET_DYN
966 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
967 * and loaders (ET_DYN without INTERP, since they
968 * _are_ the ELF interpreter). The loaders must
969 * be loaded away from programs since the program
970 * may otherwise collide with the loader (especially
971 * for ET_EXEC which does not have a randomized
972 * position). For example to handle invocations of
973 * "./ld.so someprog" to test out a new version of
974 * the loader, the subsequent program that the
975 * loader loads must avoid the loader itself, so
976 * they cannot share the same load range. Sufficient
977 * room for the brk must be allocated with the
978 * loader as well, since brk must be available with
979 * the loader.
980 *
981 * Therefore, programs are loaded offset from
982 * ELF_ET_DYN_BASE and loaders are loaded into the
983 * independently randomized mmap region (0 load_bias
984 * without MAP_FIXED).
985 */
986 if (elf_interpreter) {
987 load_bias = ELF_ET_DYN_BASE;
988 if (current->flags & PF_RANDOMIZE)
989 load_bias += arch_mmap_rnd();
990 elf_flags |= elf_fixed;
991 } else
992 load_bias = 0;
993
994 /*
995 * Since load_bias is used for all subsequent loading
996 * calculations, we must lower it by the first vaddr
997 * so that the remaining calculations based on the
998 * ELF vaddrs will be correctly offset. The result
999 * is then page aligned.
1000 */
1001 load_bias = ELF_PAGESTART(load_bias - vaddr);
1002
1003 total_size = total_mapping_size(elf_phdata,
1004 loc->elf_ex.e_phnum);
1005 if (!total_size) {
1006 retval = -EINVAL;
1007 goto out_free_dentry;
1008 }
1009 }
1010
1011 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1012 elf_prot, elf_flags, total_size);
1013 if (BAD_ADDR(error)) {
1014 retval = IS_ERR((void *)error) ?
1015 PTR_ERR((void*)error) : -EINVAL;
1016 goto out_free_dentry;
1017 }
1018
1019 if (!load_addr_set) {
1020 load_addr_set = 1;
1021 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1022 if (loc->elf_ex.e_type == ET_DYN) {
1023 load_bias += error -
1024 ELF_PAGESTART(load_bias + vaddr);
1025 load_addr += load_bias;
1026 reloc_func_desc = load_bias;
1027 }
1028 }
1029 k = elf_ppnt->p_vaddr;
1030 if (k < start_code)
1031 start_code = k;
1032 if (start_data < k)
1033 start_data = k;
1034
1035 /*
1036 * Check to see if the section's size will overflow the
1037 * allowed task size. Note that p_filesz must always be
1038 * <= p_memsz so it is only necessary to check p_memsz.
1039 */
1040 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1041 elf_ppnt->p_memsz > TASK_SIZE ||
1042 TASK_SIZE - elf_ppnt->p_memsz < k) {
1043 /* set_brk can never work. Avoid overflows. */
1044 retval = -EINVAL;
1045 goto out_free_dentry;
1046 }
1047
1048 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1049
1050 if (k > elf_bss)
1051 elf_bss = k;
1052 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1053 end_code = k;
1054 if (end_data < k)
1055 end_data = k;
1056 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1057 if (k > elf_brk) {
1058 bss_prot = elf_prot;
1059 elf_brk = k;
1060 }
1061 }
1062
1063 loc->elf_ex.e_entry += load_bias;
1064 elf_bss += load_bias;
1065 elf_brk += load_bias;
1066 start_code += load_bias;
1067 end_code += load_bias;
1068 start_data += load_bias;
1069 end_data += load_bias;
1070
1071 /* Calling set_brk effectively mmaps the pages that we need
1072 * for the bss and break sections. We must do this before
1073 * mapping in the interpreter, to make sure it doesn't wind
1074 * up getting placed where the bss needs to go.
1075 */
1076 retval = set_brk(elf_bss, elf_brk, bss_prot);
1077 if (retval)
1078 goto out_free_dentry;
1079 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1080 retval = -EFAULT; /* Nobody gets to see this, but.. */
1081 goto out_free_dentry;
1082 }
1083
1084 if (elf_interpreter) {
1085 unsigned long interp_map_addr = 0;
1086
1087 elf_entry = load_elf_interp(&loc->interp_elf_ex,
1088 interpreter,
1089 &interp_map_addr,
1090 load_bias, interp_elf_phdata);
1091 if (!IS_ERR((void *)elf_entry)) {
1092 /*
1093 * load_elf_interp() returns relocation
1094 * adjustment
1095 */
1096 interp_load_addr = elf_entry;
1097 elf_entry += loc->interp_elf_ex.e_entry;
1098 }
1099 if (BAD_ADDR(elf_entry)) {
1100 retval = IS_ERR((void *)elf_entry) ?
1101 (int)elf_entry : -EINVAL;
1102 goto out_free_dentry;
1103 }
1104 reloc_func_desc = interp_load_addr;
1105
1106 allow_write_access(interpreter);
1107 fput(interpreter);
1108 kfree(elf_interpreter);
1109 } else {
1110 elf_entry = loc->elf_ex.e_entry;
1111 if (BAD_ADDR(elf_entry)) {
1112 retval = -EINVAL;
1113 goto out_free_dentry;
1114 }
1115 }
1116
1117 kfree(interp_elf_phdata);
1118 kfree(elf_phdata);
1119
1120 set_binfmt(&elf_format);
1121
1122#ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1123 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1124 if (retval < 0)
1125 goto out;
1126#endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1127
1128 retval = create_elf_tables(bprm, &loc->elf_ex,
1129 load_addr, interp_load_addr);
1130 if (retval < 0)
1131 goto out;
1132 /* N.B. passed_fileno might not be initialized? */
1133 current->mm->end_code = end_code;
1134 current->mm->start_code = start_code;
1135 current->mm->start_data = start_data;
1136 current->mm->end_data = end_data;
1137 current->mm->start_stack = bprm->p;
1138
1139 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1140 current->mm->brk = current->mm->start_brk =
1141 arch_randomize_brk(current->mm);
1142#ifdef compat_brk_randomized
1143 current->brk_randomized = 1;
1144#endif
1145 }
1146
1147 if (current->personality & MMAP_PAGE_ZERO) {
1148 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1149 and some applications "depend" upon this behavior.
1150 Since we do not have the power to recompile these, we
1151 emulate the SVr4 behavior. Sigh. */
1152 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1153 MAP_FIXED | MAP_PRIVATE, 0);
1154 }
1155
1156#ifdef ELF_PLAT_INIT
1157 /*
1158 * The ABI may specify that certain registers be set up in special
1159 * ways (on i386 %edx is the address of a DT_FINI function, for
1160 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1161 * that the e_entry field is the address of the function descriptor
1162 * for the startup routine, rather than the address of the startup
1163 * routine itself. This macro performs whatever initialization to
1164 * the regs structure is required as well as any relocations to the
1165 * function descriptor entries when executing dynamically links apps.
1166 */
1167 ELF_PLAT_INIT(regs, reloc_func_desc);
1168#endif
1169
1170 finalize_exec(bprm);
1171 start_thread(regs, elf_entry, bprm->p);
1172 retval = 0;
1173out:
1174 kfree(loc);
1175out_ret:
1176 return retval;
1177
1178 /* error cleanup */
1179out_free_dentry:
1180 kfree(interp_elf_phdata);
1181 allow_write_access(interpreter);
1182 if (interpreter)
1183 fput(interpreter);
1184out_free_interp:
1185 kfree(elf_interpreter);
1186out_free_ph:
1187 kfree(elf_phdata);
1188 goto out;
1189}
1190
1191#ifdef CONFIG_USELIB
1192/* This is really simpleminded and specialized - we are loading an
1193 a.out library that is given an ELF header. */
1194static int load_elf_library(struct file *file)
1195{
1196 struct elf_phdr *elf_phdata;
1197 struct elf_phdr *eppnt;
1198 unsigned long elf_bss, bss, len;
1199 int retval, error, i, j;
1200 struct elfhdr elf_ex;
1201 loff_t pos = 0;
1202
1203 error = -ENOEXEC;
1204 retval = kernel_read(file, &elf_ex, sizeof(elf_ex), &pos);
1205 if (retval != sizeof(elf_ex))
1206 goto out;
1207
1208 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1209 goto out;
1210
1211 /* First of all, some simple consistency checks */
1212 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1213 !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1214 goto out;
1215 if (elf_check_fdpic(&elf_ex))
1216 goto out;
1217
1218 /* Now read in all of the header information */
1219
1220 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1221 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1222
1223 error = -ENOMEM;
1224 elf_phdata = kmalloc(j, GFP_KERNEL);
1225 if (!elf_phdata)
1226 goto out;
1227
1228 eppnt = elf_phdata;
1229 error = -ENOEXEC;
1230 pos = elf_ex.e_phoff;
1231 retval = kernel_read(file, eppnt, j, &pos);
1232 if (retval != j)
1233 goto out_free_ph;
1234
1235 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1236 if ((eppnt + i)->p_type == PT_LOAD)
1237 j++;
1238 if (j != 1)
1239 goto out_free_ph;
1240
1241 while (eppnt->p_type != PT_LOAD)
1242 eppnt++;
1243
1244 /* Now use mmap to map the library into memory. */
1245 error = vm_mmap(file,
1246 ELF_PAGESTART(eppnt->p_vaddr),
1247 (eppnt->p_filesz +
1248 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1249 PROT_READ | PROT_WRITE | PROT_EXEC,
1250 MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_DENYWRITE,
1251 (eppnt->p_offset -
1252 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1253 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1254 goto out_free_ph;
1255
1256 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1257 if (padzero(elf_bss)) {
1258 error = -EFAULT;
1259 goto out_free_ph;
1260 }
1261
1262 len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1263 bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1264 if (bss > len) {
1265 error = vm_brk(len, bss - len);
1266 if (error)
1267 goto out_free_ph;
1268 }
1269 error = 0;
1270
1271out_free_ph:
1272 kfree(elf_phdata);
1273out:
1274 return error;
1275}
1276#endif /* #ifdef CONFIG_USELIB */
1277
1278#ifdef CONFIG_ELF_CORE
1279/*
1280 * ELF core dumper
1281 *
1282 * Modelled on fs/exec.c:aout_core_dump()
1283 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1284 */
1285
1286/*
1287 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1288 * that are useful for post-mortem analysis are included in every core dump.
1289 * In that way we ensure that the core dump is fully interpretable later
1290 * without matching up the same kernel and hardware config to see what PC values
1291 * meant. These special mappings include - vDSO, vsyscall, and other
1292 * architecture specific mappings
1293 */
1294static bool always_dump_vma(struct vm_area_struct *vma)
1295{
1296 /* Any vsyscall mappings? */
1297 if (vma == get_gate_vma(vma->vm_mm))
1298 return true;
1299
1300 /*
1301 * Assume that all vmas with a .name op should always be dumped.
1302 * If this changes, a new vm_ops field can easily be added.
1303 */
1304 if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1305 return true;
1306
1307 /*
1308 * arch_vma_name() returns non-NULL for special architecture mappings,
1309 * such as vDSO sections.
1310 */
1311 if (arch_vma_name(vma))
1312 return true;
1313
1314 return false;
1315}
1316
1317/*
1318 * Decide what to dump of a segment, part, all or none.
1319 */
1320static unsigned long vma_dump_size(struct vm_area_struct *vma,
1321 unsigned long mm_flags)
1322{
1323#define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1324
1325 /* always dump the vdso and vsyscall sections */
1326 if (always_dump_vma(vma))
1327 goto whole;
1328
1329 if (vma->vm_flags & VM_DONTDUMP)
1330 return 0;
1331
1332 /* support for DAX */
1333 if (vma_is_dax(vma)) {
1334 if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1335 goto whole;
1336 if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1337 goto whole;
1338 return 0;
1339 }
1340
1341 /* Hugetlb memory check */
1342 if (vma->vm_flags & VM_HUGETLB) {
1343 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1344 goto whole;
1345 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1346 goto whole;
1347 return 0;
1348 }
1349
1350 /* Do not dump I/O mapped devices or special mappings */
1351 if (vma->vm_flags & VM_IO)
1352 return 0;
1353
1354 /* By default, dump shared memory if mapped from an anonymous file. */
1355 if (vma->vm_flags & VM_SHARED) {
1356 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1357 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1358 goto whole;
1359 return 0;
1360 }
1361
1362 /* Dump segments that have been written to. */
1363 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1364 goto whole;
1365 if (vma->vm_file == NULL)
1366 return 0;
1367
1368 if (FILTER(MAPPED_PRIVATE))
1369 goto whole;
1370
1371 /*
1372 * If this looks like the beginning of a DSO or executable mapping,
1373 * check for an ELF header. If we find one, dump the first page to
1374 * aid in determining what was mapped here.
1375 */
1376 if (FILTER(ELF_HEADERS) &&
1377 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1378 u32 __user *header = (u32 __user *) vma->vm_start;
1379 u32 word;
1380 mm_segment_t fs = get_fs();
1381 /*
1382 * Doing it this way gets the constant folded by GCC.
1383 */
1384 union {
1385 u32 cmp;
1386 char elfmag[SELFMAG];
1387 } magic;
1388 BUILD_BUG_ON(SELFMAG != sizeof word);
1389 magic.elfmag[EI_MAG0] = ELFMAG0;
1390 magic.elfmag[EI_MAG1] = ELFMAG1;
1391 magic.elfmag[EI_MAG2] = ELFMAG2;
1392 magic.elfmag[EI_MAG3] = ELFMAG3;
1393 /*
1394 * Switch to the user "segment" for get_user(),
1395 * then put back what elf_core_dump() had in place.
1396 */
1397 set_fs(USER_DS);
1398 if (unlikely(get_user(word, header)))
1399 word = 0;
1400 set_fs(fs);
1401 if (word == magic.cmp)
1402 return PAGE_SIZE;
1403 }
1404
1405#undef FILTER
1406
1407 return 0;
1408
1409whole:
1410 return vma->vm_end - vma->vm_start;
1411}
1412
1413/* An ELF note in memory */
1414struct memelfnote
1415{
1416 const char *name;
1417 int type;
1418 unsigned int datasz;
1419 void *data;
1420};
1421
1422static int notesize(struct memelfnote *en)
1423{
1424 int sz;
1425
1426 sz = sizeof(struct elf_note);
1427 sz += roundup(strlen(en->name) + 1, 4);
1428 sz += roundup(en->datasz, 4);
1429
1430 return sz;
1431}
1432
1433static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1434{
1435 struct elf_note en;
1436 en.n_namesz = strlen(men->name) + 1;
1437 en.n_descsz = men->datasz;
1438 en.n_type = men->type;
1439
1440 return dump_emit(cprm, &en, sizeof(en)) &&
1441 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1442 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1443}
1444
1445static void fill_elf_header(struct elfhdr *elf, int segs,
1446 u16 machine, u32 flags)
1447{
1448 memset(elf, 0, sizeof(*elf));
1449
1450 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1451 elf->e_ident[EI_CLASS] = ELF_CLASS;
1452 elf->e_ident[EI_DATA] = ELF_DATA;
1453 elf->e_ident[EI_VERSION] = EV_CURRENT;
1454 elf->e_ident[EI_OSABI] = ELF_OSABI;
1455
1456 elf->e_type = ET_CORE;
1457 elf->e_machine = machine;
1458 elf->e_version = EV_CURRENT;
1459 elf->e_phoff = sizeof(struct elfhdr);
1460 elf->e_flags = flags;
1461 elf->e_ehsize = sizeof(struct elfhdr);
1462 elf->e_phentsize = sizeof(struct elf_phdr);
1463 elf->e_phnum = segs;
1464
1465 return;
1466}
1467
1468static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1469{
1470 phdr->p_type = PT_NOTE;
1471 phdr->p_offset = offset;
1472 phdr->p_vaddr = 0;
1473 phdr->p_paddr = 0;
1474 phdr->p_filesz = sz;
1475 phdr->p_memsz = 0;
1476 phdr->p_flags = 0;
1477 phdr->p_align = 0;
1478 return;
1479}
1480
1481static void fill_note(struct memelfnote *note, const char *name, int type,
1482 unsigned int sz, void *data)
1483{
1484 note->name = name;
1485 note->type = type;
1486 note->datasz = sz;
1487 note->data = data;
1488 return;
1489}
1490
1491/*
1492 * fill up all the fields in prstatus from the given task struct, except
1493 * registers which need to be filled up separately.
1494 */
1495static void fill_prstatus(struct elf_prstatus *prstatus,
1496 struct task_struct *p, long signr)
1497{
1498 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1499 prstatus->pr_sigpend = p->pending.signal.sig[0];
1500 prstatus->pr_sighold = p->blocked.sig[0];
1501 rcu_read_lock();
1502 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1503 rcu_read_unlock();
1504 prstatus->pr_pid = task_pid_vnr(p);
1505 prstatus->pr_pgrp = task_pgrp_vnr(p);
1506 prstatus->pr_sid = task_session_vnr(p);
1507 if (thread_group_leader(p)) {
1508 struct task_cputime cputime;
1509
1510 /*
1511 * This is the record for the group leader. It shows the
1512 * group-wide total, not its individual thread total.
1513 */
1514 thread_group_cputime(p, &cputime);
1515 prstatus->pr_utime = ns_to_timeval(cputime.utime);
1516 prstatus->pr_stime = ns_to_timeval(cputime.stime);
1517 } else {
1518 u64 utime, stime;
1519
1520 task_cputime(p, &utime, &stime);
1521 prstatus->pr_utime = ns_to_timeval(utime);
1522 prstatus->pr_stime = ns_to_timeval(stime);
1523 }
1524
1525 prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1526 prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1527}
1528
1529static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1530 struct mm_struct *mm)
1531{
1532 const struct cred *cred;
1533 unsigned int i, len;
1534
1535 /* first copy the parameters from user space */
1536 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1537
1538 len = mm->arg_end - mm->arg_start;
1539 if (len >= ELF_PRARGSZ)
1540 len = ELF_PRARGSZ-1;
1541 if (copy_from_user(&psinfo->pr_psargs,
1542 (const char __user *)mm->arg_start, len))
1543 return -EFAULT;
1544 for(i = 0; i < len; i++)
1545 if (psinfo->pr_psargs[i] == 0)
1546 psinfo->pr_psargs[i] = ' ';
1547 psinfo->pr_psargs[len] = 0;
1548
1549 rcu_read_lock();
1550 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1551 rcu_read_unlock();
1552 psinfo->pr_pid = task_pid_vnr(p);
1553 psinfo->pr_pgrp = task_pgrp_vnr(p);
1554 psinfo->pr_sid = task_session_vnr(p);
1555
1556 i = p->state ? ffz(~p->state) + 1 : 0;
1557 psinfo->pr_state = i;
1558 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1559 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1560 psinfo->pr_nice = task_nice(p);
1561 psinfo->pr_flag = p->flags;
1562 rcu_read_lock();
1563 cred = __task_cred(p);
1564 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1565 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1566 rcu_read_unlock();
1567 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1568
1569 return 0;
1570}
1571
1572static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1573{
1574 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1575 int i = 0;
1576 do
1577 i += 2;
1578 while (auxv[i - 2] != AT_NULL);
1579 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1580}
1581
1582static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1583 const siginfo_t *siginfo)
1584{
1585 mm_segment_t old_fs = get_fs();
1586 set_fs(KERNEL_DS);
1587 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1588 set_fs(old_fs);
1589 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1590}
1591
1592#define MAX_FILE_NOTE_SIZE (4*1024*1024)
1593/*
1594 * Format of NT_FILE note:
1595 *
1596 * long count -- how many files are mapped
1597 * long page_size -- units for file_ofs
1598 * array of [COUNT] elements of
1599 * long start
1600 * long end
1601 * long file_ofs
1602 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1603 */
1604static int fill_files_note(struct memelfnote *note)
1605{
1606 struct vm_area_struct *vma;
1607 unsigned count, size, names_ofs, remaining, n;
1608 user_long_t *data;
1609 user_long_t *start_end_ofs;
1610 char *name_base, *name_curpos;
1611
1612 /* *Estimated* file count and total data size needed */
1613 count = current->mm->map_count;
1614 if (count > UINT_MAX / 64)
1615 return -EINVAL;
1616 size = count * 64;
1617
1618 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1619 alloc:
1620 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1621 return -EINVAL;
1622 size = round_up(size, PAGE_SIZE);
1623 data = kvmalloc(size, GFP_KERNEL);
1624 if (ZERO_OR_NULL_PTR(data))
1625 return -ENOMEM;
1626
1627 start_end_ofs = data + 2;
1628 name_base = name_curpos = ((char *)data) + names_ofs;
1629 remaining = size - names_ofs;
1630 count = 0;
1631 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1632 struct file *file;
1633 const char *filename;
1634
1635 file = vma->vm_file;
1636 if (!file)
1637 continue;
1638 filename = file_path(file, name_curpos, remaining);
1639 if (IS_ERR(filename)) {
1640 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1641 kvfree(data);
1642 size = size * 5 / 4;
1643 goto alloc;
1644 }
1645 continue;
1646 }
1647
1648 /* file_path() fills at the end, move name down */
1649 /* n = strlen(filename) + 1: */
1650 n = (name_curpos + remaining) - filename;
1651 remaining = filename - name_curpos;
1652 memmove(name_curpos, filename, n);
1653 name_curpos += n;
1654
1655 *start_end_ofs++ = vma->vm_start;
1656 *start_end_ofs++ = vma->vm_end;
1657 *start_end_ofs++ = vma->vm_pgoff;
1658 count++;
1659 }
1660
1661 /* Now we know exact count of files, can store it */
1662 data[0] = count;
1663 data[1] = PAGE_SIZE;
1664 /*
1665 * Count usually is less than current->mm->map_count,
1666 * we need to move filenames down.
1667 */
1668 n = current->mm->map_count - count;
1669 if (n != 0) {
1670 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1671 memmove(name_base - shift_bytes, name_base,
1672 name_curpos - name_base);
1673 name_curpos -= shift_bytes;
1674 }
1675
1676 size = name_curpos - (char *)data;
1677 fill_note(note, "CORE", NT_FILE, size, data);
1678 return 0;
1679}
1680
1681#ifdef CORE_DUMP_USE_REGSET
1682#include <linux/regset.h>
1683
1684struct elf_thread_core_info {
1685 struct elf_thread_core_info *next;
1686 struct task_struct *task;
1687 struct elf_prstatus prstatus;
1688 struct memelfnote notes[0];
1689};
1690
1691struct elf_note_info {
1692 struct elf_thread_core_info *thread;
1693 struct memelfnote psinfo;
1694 struct memelfnote signote;
1695 struct memelfnote auxv;
1696 struct memelfnote files;
1697 user_siginfo_t csigdata;
1698 size_t size;
1699 int thread_notes;
1700};
1701
1702/*
1703 * When a regset has a writeback hook, we call it on each thread before
1704 * dumping user memory. On register window machines, this makes sure the
1705 * user memory backing the register data is up to date before we read it.
1706 */
1707static void do_thread_regset_writeback(struct task_struct *task,
1708 const struct user_regset *regset)
1709{
1710 if (regset->writeback)
1711 regset->writeback(task, regset, 1);
1712}
1713
1714#ifndef PRSTATUS_SIZE
1715#define PRSTATUS_SIZE(S, R) sizeof(S)
1716#endif
1717
1718#ifndef SET_PR_FPVALID
1719#define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1720#endif
1721
1722static int fill_thread_core_info(struct elf_thread_core_info *t,
1723 const struct user_regset_view *view,
1724 long signr, size_t *total)
1725{
1726 unsigned int i;
1727 unsigned int regset0_size = regset_size(t->task, &view->regsets[0]);
1728
1729 /*
1730 * NT_PRSTATUS is the one special case, because the regset data
1731 * goes into the pr_reg field inside the note contents, rather
1732 * than being the whole note contents. We fill the reset in here.
1733 * We assume that regset 0 is NT_PRSTATUS.
1734 */
1735 fill_prstatus(&t->prstatus, t->task, signr);
1736 (void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset0_size,
1737 &t->prstatus.pr_reg, NULL);
1738
1739 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1740 PRSTATUS_SIZE(t->prstatus, regset0_size), &t->prstatus);
1741 *total += notesize(&t->notes[0]);
1742
1743 do_thread_regset_writeback(t->task, &view->regsets[0]);
1744
1745 /*
1746 * Each other regset might generate a note too. For each regset
1747 * that has no core_note_type or is inactive, we leave t->notes[i]
1748 * all zero and we'll know to skip writing it later.
1749 */
1750 for (i = 1; i < view->n; ++i) {
1751 const struct user_regset *regset = &view->regsets[i];
1752 do_thread_regset_writeback(t->task, regset);
1753 if (regset->core_note_type && regset->get &&
1754 (!regset->active || regset->active(t->task, regset) > 0)) {
1755 int ret;
1756 size_t size = regset_size(t->task, regset);
1757 void *data = kmalloc(size, GFP_KERNEL);
1758 if (unlikely(!data))
1759 return 0;
1760 ret = regset->get(t->task, regset,
1761 0, size, data, NULL);
1762 if (unlikely(ret))
1763 kfree(data);
1764 else {
1765 if (regset->core_note_type != NT_PRFPREG)
1766 fill_note(&t->notes[i], "LINUX",
1767 regset->core_note_type,
1768 size, data);
1769 else {
1770 SET_PR_FPVALID(&t->prstatus,
1771 1, regset0_size);
1772 fill_note(&t->notes[i], "CORE",
1773 NT_PRFPREG, size, data);
1774 }
1775 *total += notesize(&t->notes[i]);
1776 }
1777 }
1778 }
1779
1780 return 1;
1781}
1782
1783static int fill_note_info(struct elfhdr *elf, int phdrs,
1784 struct elf_note_info *info,
1785 const siginfo_t *siginfo, struct pt_regs *regs)
1786{
1787 struct task_struct *dump_task = current;
1788 const struct user_regset_view *view = task_user_regset_view(dump_task);
1789 struct elf_thread_core_info *t;
1790 struct elf_prpsinfo *psinfo;
1791 struct core_thread *ct;
1792 unsigned int i;
1793
1794 info->size = 0;
1795 info->thread = NULL;
1796
1797 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1798 if (psinfo == NULL) {
1799 info->psinfo.data = NULL; /* So we don't free this wrongly */
1800 return 0;
1801 }
1802
1803 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1804
1805 /*
1806 * Figure out how many notes we're going to need for each thread.
1807 */
1808 info->thread_notes = 0;
1809 for (i = 0; i < view->n; ++i)
1810 if (view->regsets[i].core_note_type != 0)
1811 ++info->thread_notes;
1812
1813 /*
1814 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1815 * since it is our one special case.
1816 */
1817 if (unlikely(info->thread_notes == 0) ||
1818 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1819 WARN_ON(1);
1820 return 0;
1821 }
1822
1823 /*
1824 * Initialize the ELF file header.
1825 */
1826 fill_elf_header(elf, phdrs,
1827 view->e_machine, view->e_flags);
1828
1829 /*
1830 * Allocate a structure for each thread.
1831 */
1832 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1833 t = kzalloc(offsetof(struct elf_thread_core_info,
1834 notes[info->thread_notes]),
1835 GFP_KERNEL);
1836 if (unlikely(!t))
1837 return 0;
1838
1839 t->task = ct->task;
1840 if (ct->task == dump_task || !info->thread) {
1841 t->next = info->thread;
1842 info->thread = t;
1843 } else {
1844 /*
1845 * Make sure to keep the original task at
1846 * the head of the list.
1847 */
1848 t->next = info->thread->next;
1849 info->thread->next = t;
1850 }
1851 }
1852
1853 /*
1854 * Now fill in each thread's information.
1855 */
1856 for (t = info->thread; t != NULL; t = t->next)
1857 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1858 return 0;
1859
1860 /*
1861 * Fill in the two process-wide notes.
1862 */
1863 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1864 info->size += notesize(&info->psinfo);
1865
1866 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1867 info->size += notesize(&info->signote);
1868
1869 fill_auxv_note(&info->auxv, current->mm);
1870 info->size += notesize(&info->auxv);
1871
1872 if (fill_files_note(&info->files) == 0)
1873 info->size += notesize(&info->files);
1874
1875 return 1;
1876}
1877
1878static size_t get_note_info_size(struct elf_note_info *info)
1879{
1880 return info->size;
1881}
1882
1883/*
1884 * Write all the notes for each thread. When writing the first thread, the
1885 * process-wide notes are interleaved after the first thread-specific note.
1886 */
1887static int write_note_info(struct elf_note_info *info,
1888 struct coredump_params *cprm)
1889{
1890 bool first = true;
1891 struct elf_thread_core_info *t = info->thread;
1892
1893 do {
1894 int i;
1895
1896 if (!writenote(&t->notes[0], cprm))
1897 return 0;
1898
1899 if (first && !writenote(&info->psinfo, cprm))
1900 return 0;
1901 if (first && !writenote(&info->signote, cprm))
1902 return 0;
1903 if (first && !writenote(&info->auxv, cprm))
1904 return 0;
1905 if (first && info->files.data &&
1906 !writenote(&info->files, cprm))
1907 return 0;
1908
1909 for (i = 1; i < info->thread_notes; ++i)
1910 if (t->notes[i].data &&
1911 !writenote(&t->notes[i], cprm))
1912 return 0;
1913
1914 first = false;
1915 t = t->next;
1916 } while (t);
1917
1918 return 1;
1919}
1920
1921static void free_note_info(struct elf_note_info *info)
1922{
1923 struct elf_thread_core_info *threads = info->thread;
1924 while (threads) {
1925 unsigned int i;
1926 struct elf_thread_core_info *t = threads;
1927 threads = t->next;
1928 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1929 for (i = 1; i < info->thread_notes; ++i)
1930 kfree(t->notes[i].data);
1931 kfree(t);
1932 }
1933 kfree(info->psinfo.data);
1934 kvfree(info->files.data);
1935}
1936
1937#else
1938
1939/* Here is the structure in which status of each thread is captured. */
1940struct elf_thread_status
1941{
1942 struct list_head list;
1943 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1944 elf_fpregset_t fpu; /* NT_PRFPREG */
1945 struct task_struct *thread;
1946#ifdef ELF_CORE_COPY_XFPREGS
1947 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1948#endif
1949 struct memelfnote notes[3];
1950 int num_notes;
1951};
1952
1953/*
1954 * In order to add the specific thread information for the elf file format,
1955 * we need to keep a linked list of every threads pr_status and then create
1956 * a single section for them in the final core file.
1957 */
1958static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1959{
1960 int sz = 0;
1961 struct task_struct *p = t->thread;
1962 t->num_notes = 0;
1963
1964 fill_prstatus(&t->prstatus, p, signr);
1965 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1966
1967 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1968 &(t->prstatus));
1969 t->num_notes++;
1970 sz += notesize(&t->notes[0]);
1971
1972 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1973 &t->fpu))) {
1974 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1975 &(t->fpu));
1976 t->num_notes++;
1977 sz += notesize(&t->notes[1]);
1978 }
1979
1980#ifdef ELF_CORE_COPY_XFPREGS
1981 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1982 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1983 sizeof(t->xfpu), &t->xfpu);
1984 t->num_notes++;
1985 sz += notesize(&t->notes[2]);
1986 }
1987#endif
1988 return sz;
1989}
1990
1991struct elf_note_info {
1992 struct memelfnote *notes;
1993 struct memelfnote *notes_files;
1994 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1995 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1996 struct list_head thread_list;
1997 elf_fpregset_t *fpu;
1998#ifdef ELF_CORE_COPY_XFPREGS
1999 elf_fpxregset_t *xfpu;
2000#endif
2001 user_siginfo_t csigdata;
2002 int thread_status_size;
2003 int numnote;
2004};
2005
2006static int elf_note_info_init(struct elf_note_info *info)
2007{
2008 memset(info, 0, sizeof(*info));
2009 INIT_LIST_HEAD(&info->thread_list);
2010
2011 /* Allocate space for ELF notes */
2012 info->notes = kmalloc_array(8, sizeof(struct memelfnote), GFP_KERNEL);
2013 if (!info->notes)
2014 return 0;
2015 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2016 if (!info->psinfo)
2017 return 0;
2018 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2019 if (!info->prstatus)
2020 return 0;
2021 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2022 if (!info->fpu)
2023 return 0;
2024#ifdef ELF_CORE_COPY_XFPREGS
2025 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
2026 if (!info->xfpu)
2027 return 0;
2028#endif
2029 return 1;
2030}
2031
2032static int fill_note_info(struct elfhdr *elf, int phdrs,
2033 struct elf_note_info *info,
2034 const siginfo_t *siginfo, struct pt_regs *regs)
2035{
2036 struct list_head *t;
2037 struct core_thread *ct;
2038 struct elf_thread_status *ets;
2039
2040 if (!elf_note_info_init(info))
2041 return 0;
2042
2043 for (ct = current->mm->core_state->dumper.next;
2044 ct; ct = ct->next) {
2045 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2046 if (!ets)
2047 return 0;
2048
2049 ets->thread = ct->task;
2050 list_add(&ets->list, &info->thread_list);
2051 }
2052
2053 list_for_each(t, &info->thread_list) {
2054 int sz;
2055
2056 ets = list_entry(t, struct elf_thread_status, list);
2057 sz = elf_dump_thread_status(siginfo->si_signo, ets);
2058 info->thread_status_size += sz;
2059 }
2060 /* now collect the dump for the current */
2061 memset(info->prstatus, 0, sizeof(*info->prstatus));
2062 fill_prstatus(info->prstatus, current, siginfo->si_signo);
2063 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2064
2065 /* Set up header */
2066 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2067
2068 /*
2069 * Set up the notes in similar form to SVR4 core dumps made
2070 * with info from their /proc.
2071 */
2072
2073 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2074 sizeof(*info->prstatus), info->prstatus);
2075 fill_psinfo(info->psinfo, current->group_leader, current->mm);
2076 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2077 sizeof(*info->psinfo), info->psinfo);
2078
2079 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2080 fill_auxv_note(info->notes + 3, current->mm);
2081 info->numnote = 4;
2082
2083 if (fill_files_note(info->notes + info->numnote) == 0) {
2084 info->notes_files = info->notes + info->numnote;
2085 info->numnote++;
2086 }
2087
2088 /* Try to dump the FPU. */
2089 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2090 info->fpu);
2091 if (info->prstatus->pr_fpvalid)
2092 fill_note(info->notes + info->numnote++,
2093 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2094#ifdef ELF_CORE_COPY_XFPREGS
2095 if (elf_core_copy_task_xfpregs(current, info->xfpu))
2096 fill_note(info->notes + info->numnote++,
2097 "LINUX", ELF_CORE_XFPREG_TYPE,
2098 sizeof(*info->xfpu), info->xfpu);
2099#endif
2100
2101 return 1;
2102}
2103
2104static size_t get_note_info_size(struct elf_note_info *info)
2105{
2106 int sz = 0;
2107 int i;
2108
2109 for (i = 0; i < info->numnote; i++)
2110 sz += notesize(info->notes + i);
2111
2112 sz += info->thread_status_size;
2113
2114 return sz;
2115}
2116
2117static int write_note_info(struct elf_note_info *info,
2118 struct coredump_params *cprm)
2119{
2120 int i;
2121 struct list_head *t;
2122
2123 for (i = 0; i < info->numnote; i++)
2124 if (!writenote(info->notes + i, cprm))
2125 return 0;
2126
2127 /* write out the thread status notes section */
2128 list_for_each(t, &info->thread_list) {
2129 struct elf_thread_status *tmp =
2130 list_entry(t, struct elf_thread_status, list);
2131
2132 for (i = 0; i < tmp->num_notes; i++)
2133 if (!writenote(&tmp->notes[i], cprm))
2134 return 0;
2135 }
2136
2137 return 1;
2138}
2139
2140static void free_note_info(struct elf_note_info *info)
2141{
2142 while (!list_empty(&info->thread_list)) {
2143 struct list_head *tmp = info->thread_list.next;
2144 list_del(tmp);
2145 kfree(list_entry(tmp, struct elf_thread_status, list));
2146 }
2147
2148 /* Free data possibly allocated by fill_files_note(): */
2149 if (info->notes_files)
2150 kvfree(info->notes_files->data);
2151
2152 kfree(info->prstatus);
2153 kfree(info->psinfo);
2154 kfree(info->notes);
2155 kfree(info->fpu);
2156#ifdef ELF_CORE_COPY_XFPREGS
2157 kfree(info->xfpu);
2158#endif
2159}
2160
2161#endif
2162
2163static struct vm_area_struct *first_vma(struct task_struct *tsk,
2164 struct vm_area_struct *gate_vma)
2165{
2166 struct vm_area_struct *ret = tsk->mm->mmap;
2167
2168 if (ret)
2169 return ret;
2170 return gate_vma;
2171}
2172/*
2173 * Helper function for iterating across a vma list. It ensures that the caller
2174 * will visit `gate_vma' prior to terminating the search.
2175 */
2176static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2177 struct vm_area_struct *gate_vma)
2178{
2179 struct vm_area_struct *ret;
2180
2181 ret = this_vma->vm_next;
2182 if (ret)
2183 return ret;
2184 if (this_vma == gate_vma)
2185 return NULL;
2186 return gate_vma;
2187}
2188
2189static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2190 elf_addr_t e_shoff, int segs)
2191{
2192 elf->e_shoff = e_shoff;
2193 elf->e_shentsize = sizeof(*shdr4extnum);
2194 elf->e_shnum = 1;
2195 elf->e_shstrndx = SHN_UNDEF;
2196
2197 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2198
2199 shdr4extnum->sh_type = SHT_NULL;
2200 shdr4extnum->sh_size = elf->e_shnum;
2201 shdr4extnum->sh_link = elf->e_shstrndx;
2202 shdr4extnum->sh_info = segs;
2203}
2204
2205/*
2206 * Actual dumper
2207 *
2208 * This is a two-pass process; first we find the offsets of the bits,
2209 * and then they are actually written out. If we run out of core limit
2210 * we just truncate.
2211 */
2212static int elf_core_dump(struct coredump_params *cprm)
2213{
2214 int has_dumped = 0;
2215 mm_segment_t fs;
2216 int segs, i;
2217 size_t vma_data_size = 0;
2218 struct vm_area_struct *vma, *gate_vma;
2219 struct elfhdr *elf = NULL;
2220 loff_t offset = 0, dataoff;
2221 struct elf_note_info info = { };
2222 struct elf_phdr *phdr4note = NULL;
2223 struct elf_shdr *shdr4extnum = NULL;
2224 Elf_Half e_phnum;
2225 elf_addr_t e_shoff;
2226 elf_addr_t *vma_filesz = NULL;
2227
2228 /*
2229 * We no longer stop all VM operations.
2230 *
2231 * This is because those proceses that could possibly change map_count
2232 * or the mmap / vma pages are now blocked in do_exit on current
2233 * finishing this core dump.
2234 *
2235 * Only ptrace can touch these memory addresses, but it doesn't change
2236 * the map_count or the pages allocated. So no possibility of crashing
2237 * exists while dumping the mm->vm_next areas to the core file.
2238 */
2239
2240 /* alloc memory for large data structures: too large to be on stack */
2241 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2242 if (!elf)
2243 goto out;
2244 /*
2245 * The number of segs are recored into ELF header as 16bit value.
2246 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2247 */
2248 segs = current->mm->map_count;
2249 segs += elf_core_extra_phdrs();
2250
2251 gate_vma = get_gate_vma(current->mm);
2252 if (gate_vma != NULL)
2253 segs++;
2254
2255 /* for notes section */
2256 segs++;
2257
2258 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2259 * this, kernel supports extended numbering. Have a look at
2260 * include/linux/elf.h for further information. */
2261 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2262
2263 /*
2264 * Collect all the non-memory information about the process for the
2265 * notes. This also sets up the file header.
2266 */
2267 if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2268 goto cleanup;
2269
2270 has_dumped = 1;
2271
2272 fs = get_fs();
2273 set_fs(KERNEL_DS);
2274
2275 offset += sizeof(*elf); /* Elf header */
2276 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2277
2278 /* Write notes phdr entry */
2279 {
2280 size_t sz = get_note_info_size(&info);
2281
2282 sz += elf_coredump_extra_notes_size();
2283
2284 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2285 if (!phdr4note)
2286 goto end_coredump;
2287
2288 fill_elf_note_phdr(phdr4note, sz, offset);
2289 offset += sz;
2290 }
2291
2292 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2293
2294 if (segs - 1 > ULONG_MAX / sizeof(*vma_filesz))
2295 goto end_coredump;
2296 vma_filesz = kvmalloc(array_size(sizeof(*vma_filesz), (segs - 1)),
2297 GFP_KERNEL);
2298 if (ZERO_OR_NULL_PTR(vma_filesz))
2299 goto end_coredump;
2300
2301 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2302 vma = next_vma(vma, gate_vma)) {
2303 unsigned long dump_size;
2304
2305 dump_size = vma_dump_size(vma, cprm->mm_flags);
2306 vma_filesz[i++] = dump_size;
2307 vma_data_size += dump_size;
2308 }
2309
2310 offset += vma_data_size;
2311 offset += elf_core_extra_data_size();
2312 e_shoff = offset;
2313
2314 if (e_phnum == PN_XNUM) {
2315 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2316 if (!shdr4extnum)
2317 goto end_coredump;
2318 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2319 }
2320
2321 offset = dataoff;
2322
2323 if (!dump_emit(cprm, elf, sizeof(*elf)))
2324 goto end_coredump;
2325
2326 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2327 goto end_coredump;
2328
2329 /* Write program headers for segments dump */
2330 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2331 vma = next_vma(vma, gate_vma)) {
2332 struct elf_phdr phdr;
2333
2334 phdr.p_type = PT_LOAD;
2335 phdr.p_offset = offset;
2336 phdr.p_vaddr = vma->vm_start;
2337 phdr.p_paddr = 0;
2338 phdr.p_filesz = vma_filesz[i++];
2339 phdr.p_memsz = vma->vm_end - vma->vm_start;
2340 offset += phdr.p_filesz;
2341 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2342 if (vma->vm_flags & VM_WRITE)
2343 phdr.p_flags |= PF_W;
2344 if (vma->vm_flags & VM_EXEC)
2345 phdr.p_flags |= PF_X;
2346 phdr.p_align = ELF_EXEC_PAGESIZE;
2347
2348 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2349 goto end_coredump;
2350 }
2351
2352 if (!elf_core_write_extra_phdrs(cprm, offset))
2353 goto end_coredump;
2354
2355 /* write out the notes section */
2356 if (!write_note_info(&info, cprm))
2357 goto end_coredump;
2358
2359 if (elf_coredump_extra_notes_write(cprm))
2360 goto end_coredump;
2361
2362 /* Align to page */
2363 if (!dump_skip(cprm, dataoff - cprm->pos))
2364 goto end_coredump;
2365
2366 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2367 vma = next_vma(vma, gate_vma)) {
2368 unsigned long addr;
2369 unsigned long end;
2370
2371 end = vma->vm_start + vma_filesz[i++];
2372
2373 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2374 struct page *page;
2375 int stop;
2376
2377 page = get_dump_page(addr);
2378 if (page) {
2379 void *kaddr = kmap(page);
2380 stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2381 kunmap(page);
2382 put_page(page);
2383 } else
2384 stop = !dump_skip(cprm, PAGE_SIZE);
2385 if (stop)
2386 goto end_coredump;
2387 }
2388 }
2389 dump_truncate(cprm);
2390
2391 if (!elf_core_write_extra_data(cprm))
2392 goto end_coredump;
2393
2394 if (e_phnum == PN_XNUM) {
2395 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2396 goto end_coredump;
2397 }
2398
2399end_coredump:
2400 set_fs(fs);
2401
2402cleanup:
2403 free_note_info(&info);
2404 kfree(shdr4extnum);
2405 kvfree(vma_filesz);
2406 kfree(phdr4note);
2407 kfree(elf);
2408out:
2409 return has_dumped;
2410}
2411
2412#endif /* CONFIG_ELF_CORE */
2413
2414static int __init init_elf_binfmt(void)
2415{
2416 register_binfmt(&elf_format);
2417 return 0;
2418}
2419
2420static void __exit exit_elf_binfmt(void)
2421{
2422 /* Remove the COFF and ELF loaders. */
2423 unregister_binfmt(&elf_format);
2424}
2425
2426core_initcall(init_elf_binfmt);
2427module_exit(exit_elf_binfmt);
2428MODULE_LICENSE("GPL");
2429