Merge tag 'v3.10.97' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / binfmt_elf.c
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/utsname.h>
35 #include <linux/coredump.h>
36 #include <linux/sched.h>
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/page.h>
40
41 #ifndef user_long_t
42 #define user_long_t long
43 #endif
44 #ifndef user_siginfo_t
45 #define user_siginfo_t siginfo_t
46 #endif
47
48 static int load_elf_binary(struct linux_binprm *bprm);
49 static int load_elf_library(struct file *);
50 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
51 int, int, unsigned long);
52
53 /*
54 * If we don't support core dumping, then supply a NULL so we
55 * don't even try.
56 */
57 #ifdef CONFIG_ELF_CORE
58 static int elf_core_dump(struct coredump_params *cprm);
59 #else
60 #define elf_core_dump NULL
61 #endif
62
63 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
64 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
65 #else
66 #define ELF_MIN_ALIGN PAGE_SIZE
67 #endif
68
69 #ifndef ELF_CORE_EFLAGS
70 #define ELF_CORE_EFLAGS 0
71 #endif
72
73 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
74 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
75 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
76
77 static struct linux_binfmt elf_format = {
78 .module = THIS_MODULE,
79 .load_binary = load_elf_binary,
80 .load_shlib = load_elf_library,
81 .core_dump = elf_core_dump,
82 .min_coredump = ELF_EXEC_PAGESIZE,
83 };
84
85 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
86
87 static int set_brk(unsigned long start, unsigned long end)
88 {
89 start = ELF_PAGEALIGN(start);
90 end = ELF_PAGEALIGN(end);
91 if (end > start) {
92 unsigned long addr;
93 addr = vm_brk(start, end - start);
94 if (BAD_ADDR(addr))
95 return addr;
96 }
97 current->mm->start_brk = current->mm->brk = end;
98 return 0;
99 }
100
101 /* We need to explicitly zero any fractional pages
102 after the data section (i.e. bss). This would
103 contain the junk from the file that should not
104 be in memory
105 */
106 static int padzero(unsigned long elf_bss)
107 {
108 unsigned long nbyte;
109
110 nbyte = ELF_PAGEOFFSET(elf_bss);
111 if (nbyte) {
112 nbyte = ELF_MIN_ALIGN - nbyte;
113 if (clear_user((void __user *) elf_bss, nbyte))
114 return -EFAULT;
115 }
116 return 0;
117 }
118
119 /* Let's use some macros to make this stack manipulation a little clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
126 old_sp; })
127 #else
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 (((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
132 #endif
133
134 #ifndef ELF_BASE_PLATFORM
135 /*
136 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
137 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
138 * will be copied to the user stack in the same manner as AT_PLATFORM.
139 */
140 #define ELF_BASE_PLATFORM NULL
141 #endif
142
143 static int
144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
145 unsigned long load_addr, unsigned long interp_load_addr)
146 {
147 unsigned long p = bprm->p;
148 int argc = bprm->argc;
149 int envc = bprm->envc;
150 elf_addr_t __user *argv;
151 elf_addr_t __user *envp;
152 elf_addr_t __user *sp;
153 elf_addr_t __user *u_platform;
154 elf_addr_t __user *u_base_platform;
155 elf_addr_t __user *u_rand_bytes;
156 const char *k_platform = ELF_PLATFORM;
157 const char *k_base_platform = ELF_BASE_PLATFORM;
158 unsigned char k_rand_bytes[16];
159 int items;
160 elf_addr_t *elf_info;
161 int ei_index = 0;
162 const struct cred *cred = current_cred();
163 struct vm_area_struct *vma;
164
165 /*
166 * In some cases (e.g. Hyper-Threading), we want to avoid L1
167 * evictions by the processes running on the same package. One
168 * thing we can do is to shuffle the initial stack for them.
169 */
170
171 p = arch_align_stack(p);
172
173 /*
174 * If this architecture has a platform capability string, copy it
175 * to userspace. In some cases (Sparc), this info is impossible
176 * for userspace to get any other way, in others (i386) it is
177 * merely difficult.
178 */
179 u_platform = NULL;
180 if (k_platform) {
181 size_t len = strlen(k_platform) + 1;
182
183 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
184 if (__copy_to_user(u_platform, k_platform, len))
185 return -EFAULT;
186 }
187
188 /*
189 * If this architecture has a "base" platform capability
190 * string, copy it to userspace.
191 */
192 u_base_platform = NULL;
193 if (k_base_platform) {
194 size_t len = strlen(k_base_platform) + 1;
195
196 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 if (__copy_to_user(u_base_platform, k_base_platform, len))
198 return -EFAULT;
199 }
200
201 /*
202 * Generate 16 random bytes for userspace PRNG seeding.
203 */
204 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
205 u_rand_bytes = (elf_addr_t __user *)
206 STACK_ALLOC(p, sizeof(k_rand_bytes));
207 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
208 return -EFAULT;
209
210 /* Create the ELF interpreter info */
211 elf_info = (elf_addr_t *)current->mm->saved_auxv;
212 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
213 #define NEW_AUX_ENT(id, val) \
214 do { \
215 elf_info[ei_index++] = id; \
216 elf_info[ei_index++] = val; \
217 } while (0)
218
219 #ifdef ARCH_DLINFO
220 /*
221 * ARCH_DLINFO must come first so PPC can do its special alignment of
222 * AUXV.
223 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
224 * ARCH_DLINFO changes
225 */
226 ARCH_DLINFO;
227 #endif
228 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
229 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
230 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
231 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
232 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
233 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
234 NEW_AUX_ENT(AT_BASE, interp_load_addr);
235 NEW_AUX_ENT(AT_FLAGS, 0);
236 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
237 NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
238 NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
239 NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
240 NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
241 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
242 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
243 #ifdef ELF_HWCAP2
244 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
245 #endif
246 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
247 if (k_platform) {
248 NEW_AUX_ENT(AT_PLATFORM,
249 (elf_addr_t)(unsigned long)u_platform);
250 }
251 if (k_base_platform) {
252 NEW_AUX_ENT(AT_BASE_PLATFORM,
253 (elf_addr_t)(unsigned long)u_base_platform);
254 }
255 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
256 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
257 }
258 #undef NEW_AUX_ENT
259 /* AT_NULL is zero; clear the rest too */
260 memset(&elf_info[ei_index], 0,
261 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
262
263 /* And advance past the AT_NULL entry. */
264 ei_index += 2;
265
266 sp = STACK_ADD(p, ei_index);
267
268 items = (argc + 1) + (envc + 1) + 1;
269 bprm->p = STACK_ROUND(sp, items);
270
271 /* Point sp at the lowest address on the stack */
272 #ifdef CONFIG_STACK_GROWSUP
273 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
274 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
275 #else
276 sp = (elf_addr_t __user *)bprm->p;
277 #endif
278
279
280 /*
281 * Grow the stack manually; some architectures have a limit on how
282 * far ahead a user-space access may be in order to grow the stack.
283 */
284 vma = find_extend_vma(current->mm, bprm->p);
285 if (!vma)
286 return -EFAULT;
287
288 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
289 if (__put_user(argc, sp++))
290 return -EFAULT;
291 argv = sp;
292 envp = argv + argc + 1;
293
294 /* Populate argv and envp */
295 p = current->mm->arg_end = current->mm->arg_start;
296 while (argc-- > 0) {
297 size_t len;
298 if (__put_user((elf_addr_t)p, argv++))
299 return -EFAULT;
300 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
301 if (!len || len > MAX_ARG_STRLEN)
302 return -EINVAL;
303 p += len;
304 }
305 if (__put_user(0, argv))
306 return -EFAULT;
307 current->mm->arg_end = current->mm->env_start = p;
308 while (envc-- > 0) {
309 size_t len;
310 if (__put_user((elf_addr_t)p, envp++))
311 return -EFAULT;
312 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
313 if (!len || len > MAX_ARG_STRLEN)
314 return -EINVAL;
315 p += len;
316 }
317 if (__put_user(0, envp))
318 return -EFAULT;
319 current->mm->env_end = p;
320
321 /* Put the elf_info on the stack in the right place. */
322 sp = (elf_addr_t __user *)envp + 1;
323 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
324 return -EFAULT;
325 return 0;
326 }
327
328 #ifndef elf_map
329
330 static unsigned long elf_map(struct file *filep, unsigned long addr,
331 struct elf_phdr *eppnt, int prot, int type,
332 unsigned long total_size)
333 {
334 unsigned long map_addr;
335 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
336 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
337 addr = ELF_PAGESTART(addr);
338 size = ELF_PAGEALIGN(size);
339
340 /* mmap() will return -EINVAL if given a zero size, but a
341 * segment with zero filesize is perfectly valid */
342 if (!size)
343 return addr;
344
345 /*
346 * total_size is the size of the ELF (interpreter) image.
347 * The _first_ mmap needs to know the full size, otherwise
348 * randomization might put this image into an overlapping
349 * position with the ELF binary image. (since size < total_size)
350 * So we first map the 'big' image - and unmap the remainder at
351 * the end. (which unmap is needed for ELF images with holes.)
352 */
353 if (total_size) {
354 total_size = ELF_PAGEALIGN(total_size);
355 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
356 if (!BAD_ADDR(map_addr))
357 vm_munmap(map_addr+size, total_size-size);
358 } else
359 map_addr = vm_mmap(filep, addr, size, prot, type, off);
360
361 return(map_addr);
362 }
363
364 #endif /* !elf_map */
365
366 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
367 {
368 int i, first_idx = -1, last_idx = -1;
369
370 for (i = 0; i < nr; i++) {
371 if (cmds[i].p_type == PT_LOAD) {
372 last_idx = i;
373 if (first_idx == -1)
374 first_idx = i;
375 }
376 }
377 if (first_idx == -1)
378 return 0;
379
380 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
381 ELF_PAGESTART(cmds[first_idx].p_vaddr);
382 }
383
384
385 /* This is much more generalized than the library routine read function,
386 so we keep this separate. Technically the library read function
387 is only provided so that we can read a.out libraries that have
388 an ELF header */
389
390 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
391 struct file *interpreter, unsigned long *interp_map_addr,
392 unsigned long no_base)
393 {
394 struct elf_phdr *elf_phdata;
395 struct elf_phdr *eppnt;
396 unsigned long load_addr = 0;
397 int load_addr_set = 0;
398 unsigned long last_bss = 0, elf_bss = 0;
399 unsigned long error = ~0UL;
400 unsigned long total_size;
401 int retval, i, size;
402
403 /* First of all, some simple consistency checks */
404 if (interp_elf_ex->e_type != ET_EXEC &&
405 interp_elf_ex->e_type != ET_DYN)
406 goto out;
407 if (!elf_check_arch(interp_elf_ex))
408 goto out;
409 if (!interpreter->f_op || !interpreter->f_op->mmap)
410 goto out;
411
412 /*
413 * If the size of this structure has changed, then punt, since
414 * we will be doing the wrong thing.
415 */
416 if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
417 goto out;
418 if (interp_elf_ex->e_phnum < 1 ||
419 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
420 goto out;
421
422 /* Now read in all of the header information */
423 size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
424 if (size > ELF_MIN_ALIGN)
425 goto out;
426 elf_phdata = kmalloc(size, GFP_KERNEL);
427 if (!elf_phdata)
428 goto out;
429
430 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
431 (char *)elf_phdata, size);
432 error = -EIO;
433 if (retval != size) {
434 if (retval < 0)
435 error = retval;
436 goto out_close;
437 }
438
439 total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
440 if (!total_size) {
441 error = -EINVAL;
442 goto out_close;
443 }
444
445 eppnt = elf_phdata;
446 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
447 if (eppnt->p_type == PT_LOAD) {
448 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
449 int elf_prot = 0;
450 unsigned long vaddr = 0;
451 unsigned long k, map_addr;
452
453 if (eppnt->p_flags & PF_R)
454 elf_prot = PROT_READ;
455 if (eppnt->p_flags & PF_W)
456 elf_prot |= PROT_WRITE;
457 if (eppnt->p_flags & PF_X)
458 elf_prot |= PROT_EXEC;
459 vaddr = eppnt->p_vaddr;
460 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
461 elf_type |= MAP_FIXED;
462 else if (no_base && interp_elf_ex->e_type == ET_DYN)
463 load_addr = -vaddr;
464
465 map_addr = elf_map(interpreter, load_addr + vaddr,
466 eppnt, elf_prot, elf_type, total_size);
467 total_size = 0;
468 if (!*interp_map_addr)
469 *interp_map_addr = map_addr;
470 error = map_addr;
471 if (BAD_ADDR(map_addr))
472 goto out_close;
473
474 if (!load_addr_set &&
475 interp_elf_ex->e_type == ET_DYN) {
476 load_addr = map_addr - ELF_PAGESTART(vaddr);
477 load_addr_set = 1;
478 }
479
480 /*
481 * Check to see if the section's size will overflow the
482 * allowed task size. Note that p_filesz must always be
483 * <= p_memsize so it's only necessary to check p_memsz.
484 */
485 k = load_addr + eppnt->p_vaddr;
486 if (BAD_ADDR(k) ||
487 eppnt->p_filesz > eppnt->p_memsz ||
488 eppnt->p_memsz > TASK_SIZE ||
489 TASK_SIZE - eppnt->p_memsz < k) {
490 error = -ENOMEM;
491 goto out_close;
492 }
493
494 /*
495 * Find the end of the file mapping for this phdr, and
496 * keep track of the largest address we see for this.
497 */
498 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
499 if (k > elf_bss)
500 elf_bss = k;
501
502 /*
503 * Do the same thing for the memory mapping - between
504 * elf_bss and last_bss is the bss section.
505 */
506 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
507 if (k > last_bss)
508 last_bss = k;
509 }
510 }
511
512 if (last_bss > elf_bss) {
513 /*
514 * Now fill out the bss section. First pad the last page up
515 * to the page boundary, and then perform a mmap to make sure
516 * that there are zero-mapped pages up to and including the
517 * last bss page.
518 */
519 if (padzero(elf_bss)) {
520 error = -EFAULT;
521 goto out_close;
522 }
523
524 /* What we have mapped so far */
525 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
526
527 /* Map the last of the bss segment */
528 error = vm_brk(elf_bss, last_bss - elf_bss);
529 if (BAD_ADDR(error))
530 goto out_close;
531 }
532
533 error = load_addr;
534
535 out_close:
536 kfree(elf_phdata);
537 out:
538 return error;
539 }
540
541 /*
542 * These are the functions used to load ELF style executables and shared
543 * libraries. There is no binary dependent code anywhere else.
544 */
545
546 #define INTERPRETER_NONE 0
547 #define INTERPRETER_ELF 2
548
549 #ifndef STACK_RND_MASK
550 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
551 #endif
552
553 static unsigned long randomize_stack_top(unsigned long stack_top)
554 {
555 unsigned long random_variable = 0;
556
557 if ((current->flags & PF_RANDOMIZE) &&
558 !(current->personality & ADDR_NO_RANDOMIZE)) {
559 random_variable = (unsigned long) get_random_int();
560 random_variable &= STACK_RND_MASK;
561 random_variable <<= PAGE_SHIFT;
562 }
563 #ifdef CONFIG_STACK_GROWSUP
564 return PAGE_ALIGN(stack_top) + random_variable;
565 #else
566 return PAGE_ALIGN(stack_top) - random_variable;
567 #endif
568 }
569
570 static int load_elf_binary(struct linux_binprm *bprm)
571 {
572 struct file *interpreter = NULL; /* to shut gcc up */
573 unsigned long load_addr = 0, load_bias = 0;
574 int load_addr_set = 0;
575 char * elf_interpreter = NULL;
576 unsigned long error;
577 struct elf_phdr *elf_ppnt, *elf_phdata;
578 unsigned long elf_bss, elf_brk;
579 int retval, i;
580 unsigned int size;
581 unsigned long elf_entry;
582 unsigned long interp_load_addr = 0;
583 unsigned long start_code, end_code, start_data, end_data;
584 unsigned long reloc_func_desc __maybe_unused = 0;
585 int executable_stack = EXSTACK_DEFAULT;
586 unsigned long def_flags = 0;
587 struct pt_regs *regs = current_pt_regs();
588 struct {
589 struct elfhdr elf_ex;
590 struct elfhdr interp_elf_ex;
591 } *loc;
592
593 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
594 if (!loc) {
595 retval = -ENOMEM;
596 goto out_ret;
597 }
598
599 /* Get the exec-header */
600 loc->elf_ex = *((struct elfhdr *)bprm->buf);
601
602 retval = -ENOEXEC;
603 /* First of all, some simple consistency checks */
604 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
605 goto out;
606
607 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
608 goto out;
609 if (!elf_check_arch(&loc->elf_ex))
610 goto out;
611 if (!bprm->file->f_op || !bprm->file->f_op->mmap)
612 goto out;
613
614 /* Now read in all of the header information */
615 if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
616 goto out;
617 if (loc->elf_ex.e_phnum < 1 ||
618 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
619 goto out;
620 size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
621 retval = -ENOMEM;
622 elf_phdata = kmalloc(size, GFP_KERNEL);
623 if (!elf_phdata)
624 goto out;
625
626 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
627 (char *)elf_phdata, size);
628 if (retval != size) {
629 if (retval >= 0)
630 retval = -EIO;
631 goto out_free_ph;
632 }
633
634 elf_ppnt = elf_phdata;
635 elf_bss = 0;
636 elf_brk = 0;
637
638 start_code = ~0UL;
639 end_code = 0;
640 start_data = 0;
641 end_data = 0;
642
643 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
644 if (elf_ppnt->p_type == PT_INTERP) {
645 /* This is the program interpreter used for
646 * shared libraries - for now assume that this
647 * is an a.out format binary
648 */
649 retval = -ENOEXEC;
650 if (elf_ppnt->p_filesz > PATH_MAX ||
651 elf_ppnt->p_filesz < 2)
652 goto out_free_ph;
653
654 retval = -ENOMEM;
655 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
656 GFP_KERNEL);
657 if (!elf_interpreter)
658 goto out_free_ph;
659
660 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
661 elf_interpreter,
662 elf_ppnt->p_filesz);
663 if (retval != elf_ppnt->p_filesz) {
664 if (retval >= 0)
665 retval = -EIO;
666 goto out_free_interp;
667 }
668 /* make sure path is NULL terminated */
669 retval = -ENOEXEC;
670 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
671 goto out_free_interp;
672
673 interpreter = open_exec(elf_interpreter);
674 retval = PTR_ERR(interpreter);
675 if (IS_ERR(interpreter))
676 goto out_free_interp;
677
678 /*
679 * If the binary is not readable then enforce
680 * mm->dumpable = 0 regardless of the interpreter's
681 * permissions.
682 */
683 would_dump(bprm, interpreter);
684
685 /* Get the exec headers */
686 retval = kernel_read(interpreter, 0,
687 (void *)&loc->interp_elf_ex,
688 sizeof(loc->interp_elf_ex));
689 if (retval != sizeof(loc->interp_elf_ex)) {
690 if (retval >= 0)
691 retval = -EIO;
692 goto out_free_dentry;
693 }
694
695 break;
696 }
697 elf_ppnt++;
698 }
699
700 elf_ppnt = elf_phdata;
701 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
702 if (elf_ppnt->p_type == PT_GNU_STACK) {
703 if (elf_ppnt->p_flags & PF_X)
704 executable_stack = EXSTACK_ENABLE_X;
705 else
706 executable_stack = EXSTACK_DISABLE_X;
707 break;
708 }
709
710 /* Some simple consistency checks for the interpreter */
711 if (elf_interpreter) {
712 retval = -ELIBBAD;
713 /* Not an ELF interpreter */
714 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
715 goto out_free_dentry;
716 /* Verify the interpreter has a valid arch */
717 if (!elf_check_arch(&loc->interp_elf_ex))
718 goto out_free_dentry;
719 }
720
721 /* Flush all traces of the currently running executable */
722 retval = flush_old_exec(bprm);
723 if (retval)
724 goto out_free_dentry;
725
726 /* OK, This is the point of no return */
727 current->mm->def_flags = def_flags;
728
729 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
730 may depend on the personality. */
731 SET_PERSONALITY(loc->elf_ex);
732 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
733 current->personality |= READ_IMPLIES_EXEC;
734
735 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
736 current->flags |= PF_RANDOMIZE;
737
738 setup_new_exec(bprm);
739
740 /* Do this so that we can load the interpreter, if need be. We will
741 change some of these later */
742 current->mm->free_area_cache = current->mm->mmap_base;
743 current->mm->cached_hole_size = 0;
744 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
745 executable_stack);
746 if (retval < 0) {
747 send_sig(SIGKILL, current, 0);
748 goto out_free_dentry;
749 }
750
751 current->mm->start_stack = bprm->p;
752
753 /* Now we do a little grungy work by mmapping the ELF image into
754 the correct location in memory. */
755 for(i = 0, elf_ppnt = elf_phdata;
756 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
757 int elf_prot = 0, elf_flags;
758 unsigned long k, vaddr;
759 unsigned long total_size = 0;
760
761 if (elf_ppnt->p_type != PT_LOAD)
762 continue;
763
764 if (unlikely (elf_brk > elf_bss)) {
765 unsigned long nbyte;
766
767 /* There was a PT_LOAD segment with p_memsz > p_filesz
768 before this one. Map anonymous pages, if needed,
769 and clear the area. */
770 retval = set_brk(elf_bss + load_bias,
771 elf_brk + load_bias);
772 if (retval) {
773 send_sig(SIGKILL, current, 0);
774 goto out_free_dentry;
775 }
776 nbyte = ELF_PAGEOFFSET(elf_bss);
777 if (nbyte) {
778 nbyte = ELF_MIN_ALIGN - nbyte;
779 if (nbyte > elf_brk - elf_bss)
780 nbyte = elf_brk - elf_bss;
781 if (clear_user((void __user *)elf_bss +
782 load_bias, nbyte)) {
783 /*
784 * This bss-zeroing can fail if the ELF
785 * file specifies odd protections. So
786 * we don't check the return value
787 */
788 }
789 }
790 }
791
792 if (elf_ppnt->p_flags & PF_R)
793 elf_prot |= PROT_READ;
794 if (elf_ppnt->p_flags & PF_W)
795 elf_prot |= PROT_WRITE;
796 if (elf_ppnt->p_flags & PF_X)
797 elf_prot |= PROT_EXEC;
798
799 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
800
801 vaddr = elf_ppnt->p_vaddr;
802 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
803 elf_flags |= MAP_FIXED;
804 } else if (loc->elf_ex.e_type == ET_DYN) {
805 /* Try and get dynamic programs out of the way of the
806 * default mmap base, as well as whatever program they
807 * might try to exec. This is because the brk will
808 * follow the loader, and is not movable. */
809 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
810 /* Memory randomization might have been switched off
811 * in runtime via sysctl or explicit setting of
812 * personality flags.
813 * If that is the case, retain the original non-zero
814 * load_bias value in order to establish proper
815 * non-randomized mappings.
816 */
817 if (current->flags & PF_RANDOMIZE)
818 load_bias = 0;
819 else
820 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
821 #else
822 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
823 #endif
824 total_size = total_mapping_size(elf_phdata,
825 loc->elf_ex.e_phnum);
826 if (!total_size) {
827 retval = -EINVAL;
828 goto out_free_dentry;
829 }
830 }
831
832 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
833 elf_prot, elf_flags, total_size);
834 if (BAD_ADDR(error)) {
835 send_sig(SIGKILL, current, 0);
836 retval = IS_ERR((void *)error) ?
837 PTR_ERR((void*)error) : -EINVAL;
838 goto out_free_dentry;
839 }
840
841 if (!load_addr_set) {
842 load_addr_set = 1;
843 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
844 if (loc->elf_ex.e_type == ET_DYN) {
845 load_bias += error -
846 ELF_PAGESTART(load_bias + vaddr);
847 load_addr += load_bias;
848 reloc_func_desc = load_bias;
849 }
850 }
851 k = elf_ppnt->p_vaddr;
852 if (k < start_code)
853 start_code = k;
854 if (start_data < k)
855 start_data = k;
856
857 /*
858 * Check to see if the section's size will overflow the
859 * allowed task size. Note that p_filesz must always be
860 * <= p_memsz so it is only necessary to check p_memsz.
861 */
862 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
863 elf_ppnt->p_memsz > TASK_SIZE ||
864 TASK_SIZE - elf_ppnt->p_memsz < k) {
865 /* set_brk can never work. Avoid overflows. */
866 send_sig(SIGKILL, current, 0);
867 retval = -EINVAL;
868 goto out_free_dentry;
869 }
870
871 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
872
873 if (k > elf_bss)
874 elf_bss = k;
875 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
876 end_code = k;
877 if (end_data < k)
878 end_data = k;
879 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
880 if (k > elf_brk)
881 elf_brk = k;
882 }
883
884 loc->elf_ex.e_entry += load_bias;
885 elf_bss += load_bias;
886 elf_brk += load_bias;
887 start_code += load_bias;
888 end_code += load_bias;
889 start_data += load_bias;
890 end_data += load_bias;
891
892 /* Calling set_brk effectively mmaps the pages that we need
893 * for the bss and break sections. We must do this before
894 * mapping in the interpreter, to make sure it doesn't wind
895 * up getting placed where the bss needs to go.
896 */
897 retval = set_brk(elf_bss, elf_brk);
898 if (retval) {
899 send_sig(SIGKILL, current, 0);
900 goto out_free_dentry;
901 }
902 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
903 send_sig(SIGSEGV, current, 0);
904 retval = -EFAULT; /* Nobody gets to see this, but.. */
905 goto out_free_dentry;
906 }
907
908 if (elf_interpreter) {
909 unsigned long interp_map_addr = 0;
910
911 elf_entry = load_elf_interp(&loc->interp_elf_ex,
912 interpreter,
913 &interp_map_addr,
914 load_bias);
915 if (!IS_ERR((void *)elf_entry)) {
916 /*
917 * load_elf_interp() returns relocation
918 * adjustment
919 */
920 interp_load_addr = elf_entry;
921 elf_entry += loc->interp_elf_ex.e_entry;
922 }
923 if (BAD_ADDR(elf_entry)) {
924 force_sig(SIGSEGV, current);
925 retval = IS_ERR((void *)elf_entry) ?
926 (int)elf_entry : -EINVAL;
927 goto out_free_dentry;
928 }
929 reloc_func_desc = interp_load_addr;
930
931 allow_write_access(interpreter);
932 fput(interpreter);
933 kfree(elf_interpreter);
934 } else {
935 elf_entry = loc->elf_ex.e_entry;
936 if (BAD_ADDR(elf_entry)) {
937 force_sig(SIGSEGV, current);
938 retval = -EINVAL;
939 goto out_free_dentry;
940 }
941 }
942
943 kfree(elf_phdata);
944
945 set_binfmt(&elf_format);
946
947 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
948 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
949 if (retval < 0) {
950 send_sig(SIGKILL, current, 0);
951 goto out;
952 }
953 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
954
955 install_exec_creds(bprm);
956 retval = create_elf_tables(bprm, &loc->elf_ex,
957 load_addr, interp_load_addr);
958 if (retval < 0) {
959 send_sig(SIGKILL, current, 0);
960 goto out;
961 }
962 /* N.B. passed_fileno might not be initialized? */
963 current->mm->end_code = end_code;
964 current->mm->start_code = start_code;
965 current->mm->start_data = start_data;
966 current->mm->end_data = end_data;
967 current->mm->start_stack = bprm->p;
968
969 #ifdef arch_randomize_brk
970 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
971 current->mm->brk = current->mm->start_brk =
972 arch_randomize_brk(current->mm);
973 #ifdef CONFIG_COMPAT_BRK
974 current->brk_randomized = 1;
975 #endif
976 }
977 #endif
978
979 if (current->personality & MMAP_PAGE_ZERO) {
980 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
981 and some applications "depend" upon this behavior.
982 Since we do not have the power to recompile these, we
983 emulate the SVr4 behavior. Sigh. */
984 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
985 MAP_FIXED | MAP_PRIVATE, 0);
986 }
987
988 #ifdef ELF_PLAT_INIT
989 /*
990 * The ABI may specify that certain registers be set up in special
991 * ways (on i386 %edx is the address of a DT_FINI function, for
992 * example. In addition, it may also specify (eg, PowerPC64 ELF)
993 * that the e_entry field is the address of the function descriptor
994 * for the startup routine, rather than the address of the startup
995 * routine itself. This macro performs whatever initialization to
996 * the regs structure is required as well as any relocations to the
997 * function descriptor entries when executing dynamically links apps.
998 */
999 ELF_PLAT_INIT(regs, reloc_func_desc);
1000 #endif
1001
1002 start_thread(regs, elf_entry, bprm->p);
1003 retval = 0;
1004 out:
1005 kfree(loc);
1006 out_ret:
1007 return retval;
1008
1009 /* error cleanup */
1010 out_free_dentry:
1011 allow_write_access(interpreter);
1012 if (interpreter)
1013 fput(interpreter);
1014 out_free_interp:
1015 kfree(elf_interpreter);
1016 out_free_ph:
1017 kfree(elf_phdata);
1018 goto out;
1019 }
1020
1021 /* This is really simpleminded and specialized - we are loading an
1022 a.out library that is given an ELF header. */
1023 static int load_elf_library(struct file *file)
1024 {
1025 struct elf_phdr *elf_phdata;
1026 struct elf_phdr *eppnt;
1027 unsigned long elf_bss, bss, len;
1028 int retval, error, i, j;
1029 struct elfhdr elf_ex;
1030
1031 error = -ENOEXEC;
1032 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1033 if (retval != sizeof(elf_ex))
1034 goto out;
1035
1036 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1037 goto out;
1038
1039 /* First of all, some simple consistency checks */
1040 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1041 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1042 goto out;
1043
1044 /* Now read in all of the header information */
1045
1046 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1047 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1048
1049 error = -ENOMEM;
1050 elf_phdata = kmalloc(j, GFP_KERNEL);
1051 if (!elf_phdata)
1052 goto out;
1053
1054 eppnt = elf_phdata;
1055 error = -ENOEXEC;
1056 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1057 if (retval != j)
1058 goto out_free_ph;
1059
1060 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1061 if ((eppnt + i)->p_type == PT_LOAD)
1062 j++;
1063 if (j != 1)
1064 goto out_free_ph;
1065
1066 while (eppnt->p_type != PT_LOAD)
1067 eppnt++;
1068
1069 /* Now use mmap to map the library into memory. */
1070 error = vm_mmap(file,
1071 ELF_PAGESTART(eppnt->p_vaddr),
1072 (eppnt->p_filesz +
1073 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1074 PROT_READ | PROT_WRITE | PROT_EXEC,
1075 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1076 (eppnt->p_offset -
1077 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1078 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1079 goto out_free_ph;
1080
1081 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1082 if (padzero(elf_bss)) {
1083 error = -EFAULT;
1084 goto out_free_ph;
1085 }
1086
1087 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1088 ELF_MIN_ALIGN - 1);
1089 bss = eppnt->p_memsz + eppnt->p_vaddr;
1090 if (bss > len)
1091 vm_brk(len, bss - len);
1092 error = 0;
1093
1094 out_free_ph:
1095 kfree(elf_phdata);
1096 out:
1097 return error;
1098 }
1099
1100 #ifdef CONFIG_ELF_CORE
1101 /*
1102 * ELF core dumper
1103 *
1104 * Modelled on fs/exec.c:aout_core_dump()
1105 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1106 */
1107
1108 #ifdef CONFIG_MTK_EXTMEM
1109 extern bool extmem_in_mspace(struct vm_area_struct *vma);
1110 extern unsigned long get_virt_from_mspace(unsigned long pa);
1111 #endif
1112
1113 /*
1114 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1115 * that are useful for post-mortem analysis are included in every core dump.
1116 * In that way we ensure that the core dump is fully interpretable later
1117 * without matching up the same kernel and hardware config to see what PC values
1118 * meant. These special mappings include - vDSO, vsyscall, and other
1119 * architecture specific mappings
1120 */
1121 static bool always_dump_vma(struct vm_area_struct *vma)
1122 {
1123 /* Any vsyscall mappings? */
1124 if (vma == get_gate_vma(vma->vm_mm))
1125 return true;
1126 /*
1127 * arch_vma_name() returns non-NULL for special architecture mappings,
1128 * such as vDSO sections.
1129 */
1130 if (arch_vma_name(vma))
1131 return true;
1132
1133 #ifdef CONFIG_MTK_EXTMEM
1134 if (extmem_in_mspace(vma)) {
1135 return true;
1136 }
1137 #endif
1138 return false;
1139 }
1140
1141 /*
1142 * Decide what to dump of a segment, part, all or none.
1143 */
1144 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1145 unsigned long mm_flags)
1146 {
1147 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1148
1149 /* always dump the vdso and vsyscall sections */
1150 if (always_dump_vma(vma))
1151 goto whole;
1152
1153 if (vma->vm_flags & VM_DONTDUMP)
1154 return 0;
1155
1156 /* Hugetlb memory check */
1157 if (vma->vm_flags & VM_HUGETLB) {
1158 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1159 goto whole;
1160 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1161 goto whole;
1162 return 0;
1163 }
1164
1165 /* Do not dump I/O mapped devices or special mappings */
1166 if (vma->vm_flags & VM_IO)
1167 return 0;
1168
1169 /* By default, dump shared memory if mapped from an anonymous file. */
1170 if (vma->vm_flags & VM_SHARED) {
1171 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1172 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1173 goto whole;
1174 return 0;
1175 }
1176
1177 /* Dump segments that have been written to. */
1178 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1179 goto whole;
1180 if (vma->vm_file == NULL)
1181 return 0;
1182
1183 if (FILTER(MAPPED_PRIVATE))
1184 goto whole;
1185
1186 /*
1187 * If this looks like the beginning of a DSO or executable mapping,
1188 * check for an ELF header. If we find one, dump the first page to
1189 * aid in determining what was mapped here.
1190 */
1191 if (FILTER(ELF_HEADERS) &&
1192 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1193 u32 __user *header = (u32 __user *) vma->vm_start;
1194 u32 word;
1195 mm_segment_t fs = get_fs();
1196 /*
1197 * Doing it this way gets the constant folded by GCC.
1198 */
1199 union {
1200 u32 cmp;
1201 char elfmag[SELFMAG];
1202 } magic;
1203 BUILD_BUG_ON(SELFMAG != sizeof word);
1204 magic.elfmag[EI_MAG0] = ELFMAG0;
1205 magic.elfmag[EI_MAG1] = ELFMAG1;
1206 magic.elfmag[EI_MAG2] = ELFMAG2;
1207 magic.elfmag[EI_MAG3] = ELFMAG3;
1208 /*
1209 * Switch to the user "segment" for get_user(),
1210 * then put back what elf_core_dump() had in place.
1211 */
1212 set_fs(USER_DS);
1213 if (unlikely(get_user(word, header)))
1214 word = 0;
1215 set_fs(fs);
1216 if (word == magic.cmp)
1217 return PAGE_SIZE;
1218 }
1219
1220 #undef FILTER
1221
1222 return 0;
1223
1224 whole:
1225 return vma->vm_end - vma->vm_start;
1226 }
1227
1228 /* An ELF note in memory */
1229 struct memelfnote
1230 {
1231 const char *name;
1232 int type;
1233 unsigned int datasz;
1234 void *data;
1235 };
1236
1237 static int notesize(struct memelfnote *en)
1238 {
1239 int sz;
1240
1241 sz = sizeof(struct elf_note);
1242 sz += roundup(strlen(en->name) + 1, 4);
1243 sz += roundup(en->datasz, 4);
1244
1245 return sz;
1246 }
1247
1248 #define DUMP_WRITE(addr, nr, foffset) \
1249 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1250
1251 static int alignfile(struct file *file, loff_t *foffset)
1252 {
1253 static const char buf[4] = { 0, };
1254 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1255 return 1;
1256 }
1257
1258 static int writenote(struct memelfnote *men, struct file *file,
1259 loff_t *foffset)
1260 {
1261 struct elf_note en;
1262 en.n_namesz = strlen(men->name) + 1;
1263 en.n_descsz = men->datasz;
1264 en.n_type = men->type;
1265
1266 DUMP_WRITE(&en, sizeof(en), foffset);
1267 DUMP_WRITE(men->name, en.n_namesz, foffset);
1268 if (!alignfile(file, foffset))
1269 return 0;
1270 DUMP_WRITE(men->data, men->datasz, foffset);
1271 if (!alignfile(file, foffset))
1272 return 0;
1273
1274 return 1;
1275 }
1276 #undef DUMP_WRITE
1277
1278 static void fill_elf_header(struct elfhdr *elf, int segs,
1279 u16 machine, u32 flags)
1280 {
1281 memset(elf, 0, sizeof(*elf));
1282
1283 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1284 elf->e_ident[EI_CLASS] = ELF_CLASS;
1285 elf->e_ident[EI_DATA] = ELF_DATA;
1286 elf->e_ident[EI_VERSION] = EV_CURRENT;
1287 elf->e_ident[EI_OSABI] = ELF_OSABI;
1288
1289 elf->e_type = ET_CORE;
1290 elf->e_machine = machine;
1291 elf->e_version = EV_CURRENT;
1292 elf->e_phoff = sizeof(struct elfhdr);
1293 elf->e_flags = flags;
1294 elf->e_ehsize = sizeof(struct elfhdr);
1295 elf->e_phentsize = sizeof(struct elf_phdr);
1296 elf->e_phnum = segs;
1297
1298 return;
1299 }
1300
1301 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1302 {
1303 phdr->p_type = PT_NOTE;
1304 phdr->p_offset = offset;
1305 phdr->p_vaddr = 0;
1306 phdr->p_paddr = 0;
1307 phdr->p_filesz = sz;
1308 phdr->p_memsz = 0;
1309 phdr->p_flags = 0;
1310 phdr->p_align = 0;
1311 return;
1312 }
1313
1314 static void fill_note(struct memelfnote *note, const char *name, int type,
1315 unsigned int sz, void *data)
1316 {
1317 note->name = name;
1318 note->type = type;
1319 note->datasz = sz;
1320 note->data = data;
1321 return;
1322 }
1323
1324 /*
1325 * fill up all the fields in prstatus from the given task struct, except
1326 * registers which need to be filled up separately.
1327 */
1328 static void fill_prstatus(struct elf_prstatus *prstatus,
1329 struct task_struct *p, long signr)
1330 {
1331 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1332 prstatus->pr_sigpend = p->pending.signal.sig[0];
1333 prstatus->pr_sighold = p->blocked.sig[0];
1334 rcu_read_lock();
1335 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1336 rcu_read_unlock();
1337 prstatus->pr_pid = task_pid_vnr(p);
1338 prstatus->pr_pgrp = task_pgrp_vnr(p);
1339 prstatus->pr_sid = task_session_vnr(p);
1340 if (thread_group_leader(p)) {
1341 struct task_cputime cputime;
1342
1343 /*
1344 * This is the record for the group leader. It shows the
1345 * group-wide total, not its individual thread total.
1346 */
1347 thread_group_cputime(p, &cputime);
1348 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1349 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1350 } else {
1351 cputime_t utime, stime;
1352
1353 task_cputime(p, &utime, &stime);
1354 cputime_to_timeval(utime, &prstatus->pr_utime);
1355 cputime_to_timeval(stime, &prstatus->pr_stime);
1356 }
1357 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1358 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1359 }
1360
1361 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1362 struct mm_struct *mm)
1363 {
1364 const struct cred *cred;
1365 unsigned int i, len;
1366
1367 /* first copy the parameters from user space */
1368 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1369
1370 len = mm->arg_end - mm->arg_start;
1371 if (len >= ELF_PRARGSZ)
1372 len = ELF_PRARGSZ-1;
1373 if (copy_from_user(&psinfo->pr_psargs,
1374 (const char __user *)mm->arg_start, len))
1375 return -EFAULT;
1376 for(i = 0; i < len; i++)
1377 if (psinfo->pr_psargs[i] == 0)
1378 psinfo->pr_psargs[i] = ' ';
1379 psinfo->pr_psargs[len] = 0;
1380
1381 rcu_read_lock();
1382 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1383 rcu_read_unlock();
1384 psinfo->pr_pid = task_pid_vnr(p);
1385 psinfo->pr_pgrp = task_pgrp_vnr(p);
1386 psinfo->pr_sid = task_session_vnr(p);
1387
1388 i = p->state ? ffz(~p->state) + 1 : 0;
1389 psinfo->pr_state = i;
1390 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1391 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1392 psinfo->pr_nice = task_nice(p);
1393 psinfo->pr_flag = p->flags;
1394 rcu_read_lock();
1395 cred = __task_cred(p);
1396 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1397 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1398 rcu_read_unlock();
1399 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1400
1401 return 0;
1402 }
1403
1404 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1405 {
1406 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1407 int i = 0;
1408 do
1409 i += 2;
1410 while (auxv[i - 2] != AT_NULL);
1411 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1412 }
1413
1414 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1415 siginfo_t *siginfo)
1416 {
1417 mm_segment_t old_fs = get_fs();
1418 set_fs(KERNEL_DS);
1419 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1420 set_fs(old_fs);
1421 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1422 }
1423
1424 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1425 /*
1426 * Format of NT_FILE note:
1427 *
1428 * long count -- how many files are mapped
1429 * long page_size -- units for file_ofs
1430 * array of [COUNT] elements of
1431 * long start
1432 * long end
1433 * long file_ofs
1434 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1435 */
1436 static int fill_files_note(struct memelfnote *note)
1437 {
1438 struct vm_area_struct *vma;
1439 unsigned count, size, names_ofs, remaining, n;
1440 user_long_t *data;
1441 user_long_t *start_end_ofs;
1442 char *name_base, *name_curpos;
1443
1444 /* *Estimated* file count and total data size needed */
1445 count = current->mm->map_count;
1446 size = count * 64;
1447
1448 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1449 alloc:
1450 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1451 return -EINVAL;
1452 size = round_up(size, PAGE_SIZE);
1453 data = vmalloc(size);
1454 if (!data)
1455 return -ENOMEM;
1456
1457 start_end_ofs = data + 2;
1458 name_base = name_curpos = ((char *)data) + names_ofs;
1459 remaining = size - names_ofs;
1460 count = 0;
1461 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1462 struct file *file;
1463 const char *filename;
1464
1465 file = vma->vm_file;
1466 if (!file)
1467 continue;
1468 filename = d_path(&file->f_path, name_curpos, remaining);
1469 if (IS_ERR(filename)) {
1470 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1471 vfree(data);
1472 size = size * 5 / 4;
1473 goto alloc;
1474 }
1475 continue;
1476 }
1477
1478 /* d_path() fills at the end, move name down */
1479 /* n = strlen(filename) + 1: */
1480 n = (name_curpos + remaining) - filename;
1481 remaining = filename - name_curpos;
1482 memmove(name_curpos, filename, n);
1483 name_curpos += n;
1484
1485 *start_end_ofs++ = vma->vm_start;
1486 *start_end_ofs++ = vma->vm_end;
1487 *start_end_ofs++ = vma->vm_pgoff;
1488 count++;
1489 }
1490
1491 /* Now we know exact count of files, can store it */
1492 data[0] = count;
1493 data[1] = PAGE_SIZE;
1494 /*
1495 * Count usually is less than current->mm->map_count,
1496 * we need to move filenames down.
1497 */
1498 n = current->mm->map_count - count;
1499 if (n != 0) {
1500 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1501 memmove(name_base - shift_bytes, name_base,
1502 name_curpos - name_base);
1503 name_curpos -= shift_bytes;
1504 }
1505
1506 size = name_curpos - (char *)data;
1507 fill_note(note, "CORE", NT_FILE, size, data);
1508 return 0;
1509 }
1510
1511 #ifdef CORE_DUMP_USE_REGSET
1512 #include <linux/regset.h>
1513
1514 struct elf_thread_core_info {
1515 struct elf_thread_core_info *next;
1516 struct task_struct *task;
1517 struct elf_prstatus prstatus;
1518 struct memelfnote notes[0];
1519 };
1520
1521 struct elf_note_info {
1522 struct elf_thread_core_info *thread;
1523 struct memelfnote psinfo;
1524 struct memelfnote signote;
1525 struct memelfnote auxv;
1526 struct memelfnote files;
1527 user_siginfo_t csigdata;
1528 size_t size;
1529 int thread_notes;
1530 };
1531
1532 /*
1533 * When a regset has a writeback hook, we call it on each thread before
1534 * dumping user memory. On register window machines, this makes sure the
1535 * user memory backing the register data is up to date before we read it.
1536 */
1537 static void do_thread_regset_writeback(struct task_struct *task,
1538 const struct user_regset *regset)
1539 {
1540 if (regset->writeback)
1541 regset->writeback(task, regset, 1);
1542 }
1543
1544 #ifndef PR_REG_SIZE
1545 #define PR_REG_SIZE(S) sizeof(S)
1546 #endif
1547
1548 #ifndef PRSTATUS_SIZE
1549 #define PRSTATUS_SIZE(S) sizeof(S)
1550 #endif
1551
1552 #ifndef PR_REG_PTR
1553 #define PR_REG_PTR(S) (&((S)->pr_reg))
1554 #endif
1555
1556 #ifndef SET_PR_FPVALID
1557 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1558 #endif
1559
1560 static int fill_thread_core_info(struct elf_thread_core_info *t,
1561 const struct user_regset_view *view,
1562 long signr, size_t *total)
1563 {
1564 unsigned int i;
1565
1566 /*
1567 * NT_PRSTATUS is the one special case, because the regset data
1568 * goes into the pr_reg field inside the note contents, rather
1569 * than being the whole note contents. We fill the reset in here.
1570 * We assume that regset 0 is NT_PRSTATUS.
1571 */
1572 fill_prstatus(&t->prstatus, t->task, signr);
1573 (void) view->regsets[0].get(t->task, &view->regsets[0],
1574 0, PR_REG_SIZE(t->prstatus.pr_reg),
1575 PR_REG_PTR(&t->prstatus), NULL);
1576
1577 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1578 PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1579 *total += notesize(&t->notes[0]);
1580
1581 do_thread_regset_writeback(t->task, &view->regsets[0]);
1582
1583 /*
1584 * Each other regset might generate a note too. For each regset
1585 * that has no core_note_type or is inactive, we leave t->notes[i]
1586 * all zero and we'll know to skip writing it later.
1587 */
1588 for (i = 1; i < view->n; ++i) {
1589 const struct user_regset *regset = &view->regsets[i];
1590 do_thread_regset_writeback(t->task, regset);
1591 if (regset->core_note_type && regset->get &&
1592 (!regset->active || regset->active(t->task, regset))) {
1593 int ret;
1594 size_t size = regset->n * regset->size;
1595 void *data = kmalloc(size, GFP_KERNEL);
1596 if (unlikely(!data))
1597 return 0;
1598 ret = regset->get(t->task, regset,
1599 0, size, data, NULL);
1600 if (unlikely(ret))
1601 kfree(data);
1602 else {
1603 if (regset->core_note_type != NT_PRFPREG)
1604 fill_note(&t->notes[i], "LINUX",
1605 regset->core_note_type,
1606 size, data);
1607 else {
1608 SET_PR_FPVALID(&t->prstatus, 1);
1609 fill_note(&t->notes[i], "CORE",
1610 NT_PRFPREG, size, data);
1611 }
1612 *total += notesize(&t->notes[i]);
1613 }
1614 }
1615 }
1616
1617 return 1;
1618 }
1619
1620 static int fill_note_info(struct elfhdr *elf, int phdrs,
1621 struct elf_note_info *info,
1622 siginfo_t *siginfo, struct pt_regs *regs)
1623 {
1624 struct task_struct *dump_task = current;
1625 const struct user_regset_view *view = task_user_regset_view(dump_task);
1626 struct elf_thread_core_info *t;
1627 struct elf_prpsinfo *psinfo;
1628 struct core_thread *ct;
1629 unsigned int i;
1630
1631 info->size = 0;
1632 info->thread = NULL;
1633
1634 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1635 if (psinfo == NULL) {
1636 info->psinfo.data = NULL; /* So we don't free this wrongly */
1637 return 0;
1638 }
1639
1640 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1641
1642 /*
1643 * Figure out how many notes we're going to need for each thread.
1644 */
1645 info->thread_notes = 0;
1646 for (i = 0; i < view->n; ++i)
1647 if (view->regsets[i].core_note_type != 0)
1648 ++info->thread_notes;
1649
1650 /*
1651 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1652 * since it is our one special case.
1653 */
1654 if (unlikely(info->thread_notes == 0) ||
1655 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1656 WARN_ON(1);
1657 return 0;
1658 }
1659
1660 /*
1661 * Initialize the ELF file header.
1662 */
1663 fill_elf_header(elf, phdrs,
1664 view->e_machine, view->e_flags);
1665
1666 /*
1667 * Allocate a structure for each thread.
1668 */
1669 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1670 t = kzalloc(offsetof(struct elf_thread_core_info,
1671 notes[info->thread_notes]),
1672 GFP_KERNEL);
1673 if (unlikely(!t))
1674 return 0;
1675
1676 t->task = ct->task;
1677 if (ct->task == dump_task || !info->thread) {
1678 t->next = info->thread;
1679 info->thread = t;
1680 } else {
1681 /*
1682 * Make sure to keep the original task at
1683 * the head of the list.
1684 */
1685 t->next = info->thread->next;
1686 info->thread->next = t;
1687 }
1688 }
1689
1690 /*
1691 * Now fill in each thread's information.
1692 */
1693 for (t = info->thread; t != NULL; t = t->next)
1694 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1695 return 0;
1696
1697 /*
1698 * Fill in the two process-wide notes.
1699 */
1700 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1701 info->size += notesize(&info->psinfo);
1702
1703 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1704 info->size += notesize(&info->signote);
1705
1706 fill_auxv_note(&info->auxv, current->mm);
1707 info->size += notesize(&info->auxv);
1708
1709 if (fill_files_note(&info->files) == 0)
1710 info->size += notesize(&info->files);
1711
1712 return 1;
1713 }
1714
1715 static size_t get_note_info_size(struct elf_note_info *info)
1716 {
1717 return info->size;
1718 }
1719
1720 /*
1721 * Write all the notes for each thread. When writing the first thread, the
1722 * process-wide notes are interleaved after the first thread-specific note.
1723 */
1724 static int write_note_info(struct elf_note_info *info,
1725 struct file *file, loff_t *foffset)
1726 {
1727 bool first = 1;
1728 struct elf_thread_core_info *t = info->thread;
1729
1730 do {
1731 int i;
1732
1733 if (!writenote(&t->notes[0], file, foffset))
1734 return 0;
1735
1736 if (first && !writenote(&info->psinfo, file, foffset))
1737 return 0;
1738 if (first && !writenote(&info->signote, file, foffset))
1739 return 0;
1740 if (first && !writenote(&info->auxv, file, foffset))
1741 return 0;
1742 if (first && info->files.data &&
1743 !writenote(&info->files, file, foffset))
1744 return 0;
1745
1746 for (i = 1; i < info->thread_notes; ++i)
1747 if (t->notes[i].data &&
1748 !writenote(&t->notes[i], file, foffset))
1749 return 0;
1750
1751 first = 0;
1752 t = t->next;
1753 } while (t);
1754
1755 return 1;
1756 }
1757
1758 static void free_note_info(struct elf_note_info *info)
1759 {
1760 struct elf_thread_core_info *threads = info->thread;
1761 while (threads) {
1762 unsigned int i;
1763 struct elf_thread_core_info *t = threads;
1764 threads = t->next;
1765 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1766 for (i = 1; i < info->thread_notes; ++i)
1767 kfree(t->notes[i].data);
1768 kfree(t);
1769 }
1770 kfree(info->psinfo.data);
1771 vfree(info->files.data);
1772 }
1773
1774 #else
1775
1776 /* Here is the structure in which status of each thread is captured. */
1777 struct elf_thread_status
1778 {
1779 struct list_head list;
1780 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1781 elf_fpregset_t fpu; /* NT_PRFPREG */
1782 struct task_struct *thread;
1783 #ifdef ELF_CORE_COPY_XFPREGS
1784 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1785 #endif
1786 struct memelfnote notes[3];
1787 int num_notes;
1788 };
1789
1790 /*
1791 * In order to add the specific thread information for the elf file format,
1792 * we need to keep a linked list of every threads pr_status and then create
1793 * a single section for them in the final core file.
1794 */
1795 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1796 {
1797 int sz = 0;
1798 struct task_struct *p = t->thread;
1799 t->num_notes = 0;
1800
1801 fill_prstatus(&t->prstatus, p, signr);
1802 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1803
1804 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1805 &(t->prstatus));
1806 t->num_notes++;
1807 sz += notesize(&t->notes[0]);
1808
1809 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1810 &t->fpu))) {
1811 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1812 &(t->fpu));
1813 t->num_notes++;
1814 sz += notesize(&t->notes[1]);
1815 }
1816
1817 #ifdef ELF_CORE_COPY_XFPREGS
1818 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1819 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1820 sizeof(t->xfpu), &t->xfpu);
1821 t->num_notes++;
1822 sz += notesize(&t->notes[2]);
1823 }
1824 #endif
1825 return sz;
1826 }
1827
1828 struct elf_note_info {
1829 struct memelfnote *notes;
1830 struct memelfnote *notes_files;
1831 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1832 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1833 struct list_head thread_list;
1834 elf_fpregset_t *fpu;
1835 #ifdef ELF_CORE_COPY_XFPREGS
1836 elf_fpxregset_t *xfpu;
1837 #endif
1838 user_siginfo_t csigdata;
1839 int thread_status_size;
1840 int numnote;
1841 };
1842
1843 static int elf_note_info_init(struct elf_note_info *info)
1844 {
1845 memset(info, 0, sizeof(*info));
1846 INIT_LIST_HEAD(&info->thread_list);
1847
1848 /* Allocate space for ELF notes */
1849 info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1850 if (!info->notes)
1851 return 0;
1852 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1853 if (!info->psinfo)
1854 return 0;
1855 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1856 if (!info->prstatus)
1857 return 0;
1858 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1859 if (!info->fpu)
1860 return 0;
1861 #ifdef ELF_CORE_COPY_XFPREGS
1862 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1863 if (!info->xfpu)
1864 return 0;
1865 #endif
1866 return 1;
1867 }
1868
1869 static int fill_note_info(struct elfhdr *elf, int phdrs,
1870 struct elf_note_info *info,
1871 siginfo_t *siginfo, struct pt_regs *regs)
1872 {
1873 struct list_head *t;
1874
1875 if (!elf_note_info_init(info))
1876 return 0;
1877
1878 if (siginfo->si_signo) {
1879 struct core_thread *ct;
1880 struct elf_thread_status *ets;
1881
1882 for (ct = current->mm->core_state->dumper.next;
1883 ct; ct = ct->next) {
1884 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1885 if (!ets)
1886 return 0;
1887
1888 ets->thread = ct->task;
1889 list_add(&ets->list, &info->thread_list);
1890 }
1891
1892 list_for_each(t, &info->thread_list) {
1893 int sz;
1894
1895 ets = list_entry(t, struct elf_thread_status, list);
1896 sz = elf_dump_thread_status(siginfo->si_signo, ets);
1897 info->thread_status_size += sz;
1898 }
1899 }
1900 /* now collect the dump for the current */
1901 memset(info->prstatus, 0, sizeof(*info->prstatus));
1902 fill_prstatus(info->prstatus, current, siginfo->si_signo);
1903 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1904
1905 /* Set up header */
1906 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1907
1908 /*
1909 * Set up the notes in similar form to SVR4 core dumps made
1910 * with info from their /proc.
1911 */
1912
1913 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1914 sizeof(*info->prstatus), info->prstatus);
1915 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1916 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1917 sizeof(*info->psinfo), info->psinfo);
1918
1919 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1920 fill_auxv_note(info->notes + 3, current->mm);
1921 info->numnote = 4;
1922
1923 if (fill_files_note(info->notes + info->numnote) == 0) {
1924 info->notes_files = info->notes + info->numnote;
1925 info->numnote++;
1926 }
1927
1928 /* Try to dump the FPU. */
1929 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1930 info->fpu);
1931 if (info->prstatus->pr_fpvalid)
1932 fill_note(info->notes + info->numnote++,
1933 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1934 #ifdef ELF_CORE_COPY_XFPREGS
1935 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1936 fill_note(info->notes + info->numnote++,
1937 "LINUX", ELF_CORE_XFPREG_TYPE,
1938 sizeof(*info->xfpu), info->xfpu);
1939 #endif
1940
1941 return 1;
1942 }
1943
1944 static size_t get_note_info_size(struct elf_note_info *info)
1945 {
1946 int sz = 0;
1947 int i;
1948
1949 for (i = 0; i < info->numnote; i++)
1950 sz += notesize(info->notes + i);
1951
1952 sz += info->thread_status_size;
1953
1954 return sz;
1955 }
1956
1957 static int write_note_info(struct elf_note_info *info,
1958 struct file *file, loff_t *foffset)
1959 {
1960 int i;
1961 struct list_head *t;
1962
1963 for (i = 0; i < info->numnote; i++)
1964 if (!writenote(info->notes + i, file, foffset))
1965 return 0;
1966
1967 /* write out the thread status notes section */
1968 list_for_each(t, &info->thread_list) {
1969 struct elf_thread_status *tmp =
1970 list_entry(t, struct elf_thread_status, list);
1971
1972 for (i = 0; i < tmp->num_notes; i++)
1973 if (!writenote(&tmp->notes[i], file, foffset))
1974 return 0;
1975 }
1976
1977 return 1;
1978 }
1979
1980 static void free_note_info(struct elf_note_info *info)
1981 {
1982 while (!list_empty(&info->thread_list)) {
1983 struct list_head *tmp = info->thread_list.next;
1984 list_del(tmp);
1985 kfree(list_entry(tmp, struct elf_thread_status, list));
1986 }
1987
1988 /* Free data possibly allocated by fill_files_note(): */
1989 if (info->notes_files)
1990 vfree(info->notes_files->data);
1991
1992 kfree(info->prstatus);
1993 kfree(info->psinfo);
1994 kfree(info->notes);
1995 kfree(info->fpu);
1996 #ifdef ELF_CORE_COPY_XFPREGS
1997 kfree(info->xfpu);
1998 #endif
1999 }
2000
2001 #endif
2002
2003 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2004 struct vm_area_struct *gate_vma)
2005 {
2006 struct vm_area_struct *ret = tsk->mm->mmap;
2007
2008 if (ret)
2009 return ret;
2010 return gate_vma;
2011 }
2012 /*
2013 * Helper function for iterating across a vma list. It ensures that the caller
2014 * will visit `gate_vma' prior to terminating the search.
2015 */
2016 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2017 struct vm_area_struct *gate_vma)
2018 {
2019 struct vm_area_struct *ret;
2020
2021 ret = this_vma->vm_next;
2022 if (ret)
2023 return ret;
2024 if (this_vma == gate_vma)
2025 return NULL;
2026 return gate_vma;
2027 }
2028
2029 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2030 elf_addr_t e_shoff, int segs)
2031 {
2032 elf->e_shoff = e_shoff;
2033 elf->e_shentsize = sizeof(*shdr4extnum);
2034 elf->e_shnum = 1;
2035 elf->e_shstrndx = SHN_UNDEF;
2036
2037 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2038
2039 shdr4extnum->sh_type = SHT_NULL;
2040 shdr4extnum->sh_size = elf->e_shnum;
2041 shdr4extnum->sh_link = elf->e_shstrndx;
2042 shdr4extnum->sh_info = segs;
2043 }
2044
2045 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2046 unsigned long mm_flags)
2047 {
2048 struct vm_area_struct *vma;
2049 size_t size = 0;
2050
2051 for (vma = first_vma(current, gate_vma); vma != NULL;
2052 vma = next_vma(vma, gate_vma))
2053 size += vma_dump_size(vma, mm_flags);
2054 return size;
2055 }
2056
2057 /*
2058 * Actual dumper
2059 *
2060 * This is a two-pass process; first we find the offsets of the bits,
2061 * and then they are actually written out. If we run out of core limit
2062 * we just truncate.
2063 */
2064 static int elf_core_dump(struct coredump_params *cprm)
2065 {
2066 int has_dumped = 0;
2067 mm_segment_t fs;
2068 int segs;
2069 size_t size = 0;
2070 struct vm_area_struct *vma, *gate_vma;
2071 struct elfhdr *elf = NULL;
2072 loff_t offset = 0, dataoff, foffset;
2073 struct elf_note_info info = { };
2074 struct elf_phdr *phdr4note = NULL;
2075 struct elf_shdr *shdr4extnum = NULL;
2076 Elf_Half e_phnum;
2077 elf_addr_t e_shoff;
2078
2079 printk(KERN_WARNING "coredump(%d): start\n", current->pid);
2080
2081 /*
2082 * We no longer stop all VM operations.
2083 *
2084 * This is because those proceses that could possibly change map_count
2085 * or the mmap / vma pages are now blocked in do_exit on current
2086 * finishing this core dump.
2087 *
2088 * Only ptrace can touch these memory addresses, but it doesn't change
2089 * the map_count or the pages allocated. So no possibility of crashing
2090 * exists while dumping the mm->vm_next areas to the core file.
2091 */
2092
2093 /* alloc memory for large data structures: too large to be on stack */
2094 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2095 if (!elf)
2096 goto out;
2097 /*
2098 * The number of segs are recored into ELF header as 16bit value.
2099 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2100 */
2101 segs = current->mm->map_count;
2102 segs += elf_core_extra_phdrs();
2103
2104 gate_vma = get_gate_vma(current->mm);
2105 if (gate_vma != NULL)
2106 segs++;
2107
2108 /* for notes section */
2109 segs++;
2110
2111 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2112 * this, kernel supports extended numbering. Have a look at
2113 * include/linux/elf.h for further information. */
2114 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2115
2116 /*
2117 * Collect all the non-memory information about the process for the
2118 * notes. This also sets up the file header.
2119 */
2120 if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2121 goto cleanup;
2122
2123 has_dumped = 1;
2124
2125 fs = get_fs();
2126 set_fs(KERNEL_DS);
2127
2128 offset += sizeof(*elf); /* Elf header */
2129 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2130 foffset = offset;
2131
2132 /* Write notes phdr entry */
2133 {
2134 size_t sz = get_note_info_size(&info);
2135
2136 sz += elf_coredump_extra_notes_size();
2137
2138 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2139 if (!phdr4note)
2140 goto end_coredump;
2141
2142 fill_elf_note_phdr(phdr4note, sz, offset);
2143 offset += sz;
2144 }
2145
2146 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2147
2148 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2149 offset += elf_core_extra_data_size();
2150 e_shoff = offset;
2151
2152 if (e_phnum == PN_XNUM) {
2153 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2154 if (!shdr4extnum)
2155 goto end_coredump;
2156 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2157 }
2158
2159 offset = dataoff;
2160
2161 size += sizeof(*elf);
2162 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2163 goto end_coredump;
2164
2165 size += sizeof(*phdr4note);
2166 if (size > cprm->limit
2167 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2168 goto end_coredump;
2169
2170 /* Write program headers for segments dump */
2171 for (vma = first_vma(current, gate_vma); vma != NULL;
2172 vma = next_vma(vma, gate_vma)) {
2173 struct elf_phdr phdr;
2174
2175 phdr.p_type = PT_LOAD;
2176 phdr.p_offset = offset;
2177 phdr.p_vaddr = vma->vm_start;
2178 phdr.p_paddr = 0;
2179 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2180 phdr.p_memsz = vma->vm_end - vma->vm_start;
2181 offset += phdr.p_filesz;
2182 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2183 if (vma->vm_flags & VM_WRITE)
2184 phdr.p_flags |= PF_W;
2185 if (vma->vm_flags & VM_EXEC)
2186 phdr.p_flags |= PF_X;
2187 phdr.p_align = ELF_EXEC_PAGESIZE;
2188
2189 size += sizeof(phdr);
2190 if (size > cprm->limit
2191 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2192 goto end_coredump;
2193 }
2194
2195 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2196 goto end_coredump;
2197
2198 /* write out the notes section */
2199 if (!write_note_info(&info, cprm->file, &foffset))
2200 goto end_coredump;
2201
2202 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2203 goto end_coredump;
2204
2205 /* Align to page */
2206 if (!dump_seek(cprm->file, dataoff - foffset))
2207 goto end_coredump;
2208
2209 printk(KERN_WARNING "coredump(%d): write output program header and notes\n", current->pid);
2210
2211 for (vma = first_vma(current, gate_vma); vma != NULL;
2212 vma = next_vma(vma, gate_vma)) {
2213 unsigned long addr;
2214 unsigned long end;
2215
2216 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2217
2218 #ifdef CONFIG_MTK_EXTMEM
2219 if (extmem_in_mspace(vma)) {
2220 void *extmem_va = (void *)get_virt_from_mspace(vma->vm_pgoff << PAGE_SHIFT);
2221 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE, extmem_va += PAGE_SIZE) {
2222 int stop;
2223 int dump_write_ret = dump_write(cprm->file, extmem_va, PAGE_SIZE);
2224 stop = ((size += PAGE_SIZE) > cprm->limit) || (!dump_write_ret);
2225 if (stop) {
2226 printk(KERN_WARNING "[EXT_MEM]stop addr:0x%lx, size:%zx, limit:0x%lx, dump_write_ret:%d\n",
2227 addr, size, cprm->limit, dump_write_ret);
2228 goto end_coredump;
2229 }
2230 }
2231 continue;
2232 }
2233 #endif
2234
2235 //printk(KERN_WARNING "coredump(%d): write out load vm start:%08lx, end:%08lx\n", current->pid, vma->vm_start, end);
2236 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2237 struct page *page;
2238 int stop;
2239
2240 page = get_dump_page(addr);
2241 if (page) {
2242 void *kaddr = kmap(page);
2243 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2244 !dump_write(cprm->file, kaddr,
2245 PAGE_SIZE);
2246 kunmap(page);
2247 page_cache_release(page);
2248 if (stop) {
2249 printk(KERN_WARNING "coredump(%d): failed to write core dump\n", current->pid);
2250 }
2251 } else {
2252 stop = !dump_seek(cprm->file, PAGE_SIZE);
2253 if (stop) {
2254 printk(KERN_WARNING "coredump(%d): failed to seek core dump\n", current->pid);
2255 }
2256 }
2257 if (stop)
2258 goto end_coredump;
2259 }
2260 }
2261
2262 printk(KERN_WARNING "coredump(%d): write loads\n", current->pid);
2263
2264 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2265 goto end_coredump;
2266
2267 if (e_phnum == PN_XNUM) {
2268 size += sizeof(*shdr4extnum);
2269 if (size > cprm->limit
2270 || !dump_write(cprm->file, shdr4extnum,
2271 sizeof(*shdr4extnum)))
2272 goto end_coredump;
2273 }
2274
2275 printk(KERN_WARNING "coredump(%d): write out completed %lld\n", current->pid, offset);
2276
2277 end_coredump:
2278 set_fs(fs);
2279
2280 cleanup:
2281 free_note_info(&info);
2282 kfree(shdr4extnum);
2283 kfree(phdr4note);
2284 kfree(elf);
2285 out:
2286 return has_dumped;
2287 }
2288
2289 #endif /* CONFIG_ELF_CORE */
2290
2291 static int __init init_elf_binfmt(void)
2292 {
2293 register_binfmt(&elf_format);
2294 return 0;
2295 }
2296
2297 static void __exit exit_elf_binfmt(void)
2298 {
2299 /* Remove the COFF and ELF loaders. */
2300 unregister_binfmt(&elf_format);
2301 }
2302
2303 core_initcall(init_elf_binfmt);
2304 module_exit(exit_elf_binfmt);
2305 MODULE_LICENSE("GPL");