projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'v3.10.71' 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 retval = kernel_read(interpreter, 0, bprm->buf,
686 BINPRM_BUF_SIZE);
687 if (retval != BINPRM_BUF_SIZE) {
688 if (retval >= 0)
689 retval = -EIO;
690 goto out_free_dentry;
691 }
692
693 /* Get the exec headers */
694 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
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
760 if (elf_ppnt->p_type != PT_LOAD)
761 continue;
762
763 if (unlikely (elf_brk > elf_bss)) {
764 unsigned long nbyte;
765
766 /* There was a PT_LOAD segment with p_memsz > p_filesz
767 before this one. Map anonymous pages, if needed,
768 and clear the area. */
769 retval = set_brk(elf_bss + load_bias,
770 elf_brk + load_bias);
771 if (retval) {
772 send_sig(SIGKILL, current, 0);
773 goto out_free_dentry;
774 }
775 nbyte = ELF_PAGEOFFSET(elf_bss);
776 if (nbyte) {
777 nbyte = ELF_MIN_ALIGN - nbyte;
778 if (nbyte > elf_brk - elf_bss)
779 nbyte = elf_brk - elf_bss;
780 if (clear_user((void __user *)elf_bss +
781 load_bias, nbyte)) {
782 /*
783 * This bss-zeroing can fail if the ELF
784 * file specifies odd protections. So
785 * we don't check the return value
786 */
787 }
788 }
789 }
790
791 if (elf_ppnt->p_flags & PF_R)
792 elf_prot |= PROT_READ;
793 if (elf_ppnt->p_flags & PF_W)
794 elf_prot |= PROT_WRITE;
795 if (elf_ppnt->p_flags & PF_X)
796 elf_prot |= PROT_EXEC;
797
798 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
799
800 vaddr = elf_ppnt->p_vaddr;
801 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
802 elf_flags |= MAP_FIXED;
803 } else if (loc->elf_ex.e_type == ET_DYN) {
804 /* Try and get dynamic programs out of the way of the
805 * default mmap base, as well as whatever program they
806 * might try to exec. This is because the brk will
807 * follow the loader, and is not movable. */
808 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
809 /* Memory randomization might have been switched off
810 * in runtime via sysctl or explicit setting of
811 * personality flags.
812 * If that is the case, retain the original non-zero
813 * load_bias value in order to establish proper
814 * non-randomized mappings.
815 */
816 if (current->flags & PF_RANDOMIZE)
817 load_bias = 0;
818 else
819 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
820 #else
821 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
822 #endif
823 }
824
825 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
826 elf_prot, elf_flags, 0);
827 if (BAD_ADDR(error)) {
828 send_sig(SIGKILL, current, 0);
829 retval = IS_ERR((void *)error) ?
830 PTR_ERR((void*)error) : -EINVAL;
831 goto out_free_dentry;
832 }
833
834 if (!load_addr_set) {
835 load_addr_set = 1;
836 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
837 if (loc->elf_ex.e_type == ET_DYN) {
838 load_bias += error -
839 ELF_PAGESTART(load_bias + vaddr);
840 load_addr += load_bias;
841 reloc_func_desc = load_bias;
842 }
843 }
844 k = elf_ppnt->p_vaddr;
845 if (k < start_code)
846 start_code = k;
847 if (start_data < k)
848 start_data = k;
849
850 /*
851 * Check to see if the section's size will overflow the
852 * allowed task size. Note that p_filesz must always be
853 * <= p_memsz so it is only necessary to check p_memsz.
854 */
855 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
856 elf_ppnt->p_memsz > TASK_SIZE ||
857 TASK_SIZE - elf_ppnt->p_memsz < k) {
858 /* set_brk can never work. Avoid overflows. */
859 send_sig(SIGKILL, current, 0);
860 retval = -EINVAL;
861 goto out_free_dentry;
862 }
863
864 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
865
866 if (k > elf_bss)
867 elf_bss = k;
868 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
869 end_code = k;
870 if (end_data < k)
871 end_data = k;
872 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
873 if (k > elf_brk)
874 elf_brk = k;
875 }
876
877 loc->elf_ex.e_entry += load_bias;
878 elf_bss += load_bias;
879 elf_brk += load_bias;
880 start_code += load_bias;
881 end_code += load_bias;
882 start_data += load_bias;
883 end_data += load_bias;
884
885 /* Calling set_brk effectively mmaps the pages that we need
886 * for the bss and break sections. We must do this before
887 * mapping in the interpreter, to make sure it doesn't wind
888 * up getting placed where the bss needs to go.
889 */
890 retval = set_brk(elf_bss, elf_brk);
891 if (retval) {
892 send_sig(SIGKILL, current, 0);
893 goto out_free_dentry;
894 }
895 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
896 send_sig(SIGSEGV, current, 0);
897 retval = -EFAULT; /* Nobody gets to see this, but.. */
898 goto out_free_dentry;
899 }
900
901 if (elf_interpreter) {
902 unsigned long interp_map_addr = 0;
903
904 elf_entry = load_elf_interp(&loc->interp_elf_ex,
905 interpreter,
906 &interp_map_addr,
907 load_bias);
908 if (!IS_ERR((void *)elf_entry)) {
909 /*
910 * load_elf_interp() returns relocation
911 * adjustment
912 */
913 interp_load_addr = elf_entry;
914 elf_entry += loc->interp_elf_ex.e_entry;
915 }
916 if (BAD_ADDR(elf_entry)) {
917 force_sig(SIGSEGV, current);
918 retval = IS_ERR((void *)elf_entry) ?
919 (int)elf_entry : -EINVAL;
920 goto out_free_dentry;
921 }
922 reloc_func_desc = interp_load_addr;
923
924 allow_write_access(interpreter);
925 fput(interpreter);
926 kfree(elf_interpreter);
927 } else {
928 elf_entry = loc->elf_ex.e_entry;
929 if (BAD_ADDR(elf_entry)) {
930 force_sig(SIGSEGV, current);
931 retval = -EINVAL;
932 goto out_free_dentry;
933 }
934 }
935
936 kfree(elf_phdata);
937
938 set_binfmt(&elf_format);
939
940 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
941 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
942 if (retval < 0) {
943 send_sig(SIGKILL, current, 0);
944 goto out;
945 }
946 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
947
948 install_exec_creds(bprm);
949 retval = create_elf_tables(bprm, &loc->elf_ex,
950 load_addr, interp_load_addr);
951 if (retval < 0) {
952 send_sig(SIGKILL, current, 0);
953 goto out;
954 }
955 /* N.B. passed_fileno might not be initialized? */
956 current->mm->end_code = end_code;
957 current->mm->start_code = start_code;
958 current->mm->start_data = start_data;
959 current->mm->end_data = end_data;
960 current->mm->start_stack = bprm->p;
961
962 #ifdef arch_randomize_brk
963 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
964 current->mm->brk = current->mm->start_brk =
965 arch_randomize_brk(current->mm);
966 #ifdef CONFIG_COMPAT_BRK
967 current->brk_randomized = 1;
968 #endif
969 }
970 #endif
971
972 if (current->personality & MMAP_PAGE_ZERO) {
973 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
974 and some applications "depend" upon this behavior.
975 Since we do not have the power to recompile these, we
976 emulate the SVr4 behavior. Sigh. */
977 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
978 MAP_FIXED | MAP_PRIVATE, 0);
979 }
980
981 #ifdef ELF_PLAT_INIT
982 /*
983 * The ABI may specify that certain registers be set up in special
984 * ways (on i386 %edx is the address of a DT_FINI function, for
985 * example. In addition, it may also specify (eg, PowerPC64 ELF)
986 * that the e_entry field is the address of the function descriptor
987 * for the startup routine, rather than the address of the startup
988 * routine itself. This macro performs whatever initialization to
989 * the regs structure is required as well as any relocations to the
990 * function descriptor entries when executing dynamically links apps.
991 */
992 ELF_PLAT_INIT(regs, reloc_func_desc);
993 #endif
994
995 start_thread(regs, elf_entry, bprm->p);
996 retval = 0;
997 out:
998 kfree(loc);
999 out_ret:
1000 return retval;
1001
1002 /* error cleanup */
1003 out_free_dentry:
1004 allow_write_access(interpreter);
1005 if (interpreter)
1006 fput(interpreter);
1007 out_free_interp:
1008 kfree(elf_interpreter);
1009 out_free_ph:
1010 kfree(elf_phdata);
1011 goto out;
1012 }
1013
1014 /* This is really simpleminded and specialized - we are loading an
1015 a.out library that is given an ELF header. */
1016 static int load_elf_library(struct file *file)
1017 {
1018 struct elf_phdr *elf_phdata;
1019 struct elf_phdr *eppnt;
1020 unsigned long elf_bss, bss, len;
1021 int retval, error, i, j;
1022 struct elfhdr elf_ex;
1023
1024 error = -ENOEXEC;
1025 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1026 if (retval != sizeof(elf_ex))
1027 goto out;
1028
1029 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1030 goto out;
1031
1032 /* First of all, some simple consistency checks */
1033 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1034 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1035 goto out;
1036
1037 /* Now read in all of the header information */
1038
1039 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1040 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1041
1042 error = -ENOMEM;
1043 elf_phdata = kmalloc(j, GFP_KERNEL);
1044 if (!elf_phdata)
1045 goto out;
1046
1047 eppnt = elf_phdata;
1048 error = -ENOEXEC;
1049 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1050 if (retval != j)
1051 goto out_free_ph;
1052
1053 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1054 if ((eppnt + i)->p_type == PT_LOAD)
1055 j++;
1056 if (j != 1)
1057 goto out_free_ph;
1058
1059 while (eppnt->p_type != PT_LOAD)
1060 eppnt++;
1061
1062 /* Now use mmap to map the library into memory. */
1063 error = vm_mmap(file,
1064 ELF_PAGESTART(eppnt->p_vaddr),
1065 (eppnt->p_filesz +
1066 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1067 PROT_READ | PROT_WRITE | PROT_EXEC,
1068 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1069 (eppnt->p_offset -
1070 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1071 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1072 goto out_free_ph;
1073
1074 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1075 if (padzero(elf_bss)) {
1076 error = -EFAULT;
1077 goto out_free_ph;
1078 }
1079
1080 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1081 ELF_MIN_ALIGN - 1);
1082 bss = eppnt->p_memsz + eppnt->p_vaddr;
1083 if (bss > len)
1084 vm_brk(len, bss - len);
1085 error = 0;
1086
1087 out_free_ph:
1088 kfree(elf_phdata);
1089 out:
1090 return error;
1091 }
1092
1093 #ifdef CONFIG_ELF_CORE
1094 /*
1095 * ELF core dumper
1096 *
1097 * Modelled on fs/exec.c:aout_core_dump()
1098 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1099 */
1100
1101 #ifdef CONFIG_MTK_EXTMEM
1102 extern bool extmem_in_mspace(struct vm_area_struct *vma);
1103 extern unsigned long get_virt_from_mspace(unsigned long pa);
1104 #endif
1105
1106 /*
1107 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1108 * that are useful for post-mortem analysis are included in every core dump.
1109 * In that way we ensure that the core dump is fully interpretable later
1110 * without matching up the same kernel and hardware config to see what PC values
1111 * meant. These special mappings include - vDSO, vsyscall, and other
1112 * architecture specific mappings
1113 */
1114 static bool always_dump_vma(struct vm_area_struct *vma)
1115 {
1116 /* Any vsyscall mappings? */
1117 if (vma == get_gate_vma(vma->vm_mm))
1118 return true;
1119 /*
1120 * arch_vma_name() returns non-NULL for special architecture mappings,
1121 * such as vDSO sections.
1122 */
1123 if (arch_vma_name(vma))
1124 return true;
1125
1126 #ifdef CONFIG_MTK_EXTMEM
1127 if (extmem_in_mspace(vma)) {
1128 return true;
1129 }
1130 #endif
1131 return false;
1132 }
1133
1134 /*
1135 * Decide what to dump of a segment, part, all or none.
1136 */
1137 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1138 unsigned long mm_flags)
1139 {
1140 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1141
1142 /* always dump the vdso and vsyscall sections */
1143 if (always_dump_vma(vma))
1144 goto whole;
1145
1146 if (vma->vm_flags & VM_DONTDUMP)
1147 return 0;
1148
1149 /* Hugetlb memory check */
1150 if (vma->vm_flags & VM_HUGETLB) {
1151 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1152 goto whole;
1153 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1154 goto whole;
1155 return 0;
1156 }
1157
1158 /* Do not dump I/O mapped devices or special mappings */
1159 if (vma->vm_flags & VM_IO)
1160 return 0;
1161
1162 /* By default, dump shared memory if mapped from an anonymous file. */
1163 if (vma->vm_flags & VM_SHARED) {
1164 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1165 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1166 goto whole;
1167 return 0;
1168 }
1169
1170 /* Dump segments that have been written to. */
1171 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1172 goto whole;
1173 if (vma->vm_file == NULL)
1174 return 0;
1175
1176 if (FILTER(MAPPED_PRIVATE))
1177 goto whole;
1178
1179 /*
1180 * If this looks like the beginning of a DSO or executable mapping,
1181 * check for an ELF header. If we find one, dump the first page to
1182 * aid in determining what was mapped here.
1183 */
1184 if (FILTER(ELF_HEADERS) &&
1185 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1186 u32 __user *header = (u32 __user *) vma->vm_start;
1187 u32 word;
1188 mm_segment_t fs = get_fs();
1189 /*
1190 * Doing it this way gets the constant folded by GCC.
1191 */
1192 union {
1193 u32 cmp;
1194 char elfmag[SELFMAG];
1195 } magic;
1196 BUILD_BUG_ON(SELFMAG != sizeof word);
1197 magic.elfmag[EI_MAG0] = ELFMAG0;
1198 magic.elfmag[EI_MAG1] = ELFMAG1;
1199 magic.elfmag[EI_MAG2] = ELFMAG2;
1200 magic.elfmag[EI_MAG3] = ELFMAG3;
1201 /*
1202 * Switch to the user "segment" for get_user(),
1203 * then put back what elf_core_dump() had in place.
1204 */
1205 set_fs(USER_DS);
1206 if (unlikely(get_user(word, header)))
1207 word = 0;
1208 set_fs(fs);
1209 if (word == magic.cmp)
1210 return PAGE_SIZE;
1211 }
1212
1213 #undef FILTER
1214
1215 return 0;
1216
1217 whole:
1218 return vma->vm_end - vma->vm_start;
1219 }
1220
1221 /* An ELF note in memory */
1222 struct memelfnote
1223 {
1224 const char *name;
1225 int type;
1226 unsigned int datasz;
1227 void *data;
1228 };
1229
1230 static int notesize(struct memelfnote *en)
1231 {
1232 int sz;
1233
1234 sz = sizeof(struct elf_note);
1235 sz += roundup(strlen(en->name) + 1, 4);
1236 sz += roundup(en->datasz, 4);
1237
1238 return sz;
1239 }
1240
1241 #define DUMP_WRITE(addr, nr, foffset) \
1242 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1243
1244 static int alignfile(struct file *file, loff_t *foffset)
1245 {
1246 static const char buf[4] = { 0, };
1247 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1248 return 1;
1249 }
1250
1251 static int writenote(struct memelfnote *men, struct file *file,
1252 loff_t *foffset)
1253 {
1254 struct elf_note en;
1255 en.n_namesz = strlen(men->name) + 1;
1256 en.n_descsz = men->datasz;
1257 en.n_type = men->type;
1258
1259 DUMP_WRITE(&en, sizeof(en), foffset);
1260 DUMP_WRITE(men->name, en.n_namesz, foffset);
1261 if (!alignfile(file, foffset))
1262 return 0;
1263 DUMP_WRITE(men->data, men->datasz, foffset);
1264 if (!alignfile(file, foffset))
1265 return 0;
1266
1267 return 1;
1268 }
1269 #undef DUMP_WRITE
1270
1271 static void fill_elf_header(struct elfhdr *elf, int segs,
1272 u16 machine, u32 flags)
1273 {
1274 memset(elf, 0, sizeof(*elf));
1275
1276 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1277 elf->e_ident[EI_CLASS] = ELF_CLASS;
1278 elf->e_ident[EI_DATA] = ELF_DATA;
1279 elf->e_ident[EI_VERSION] = EV_CURRENT;
1280 elf->e_ident[EI_OSABI] = ELF_OSABI;
1281
1282 elf->e_type = ET_CORE;
1283 elf->e_machine = machine;
1284 elf->e_version = EV_CURRENT;
1285 elf->e_phoff = sizeof(struct elfhdr);
1286 elf->e_flags = flags;
1287 elf->e_ehsize = sizeof(struct elfhdr);
1288 elf->e_phentsize = sizeof(struct elf_phdr);
1289 elf->e_phnum = segs;
1290
1291 return;
1292 }
1293
1294 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1295 {
1296 phdr->p_type = PT_NOTE;
1297 phdr->p_offset = offset;
1298 phdr->p_vaddr = 0;
1299 phdr->p_paddr = 0;
1300 phdr->p_filesz = sz;
1301 phdr->p_memsz = 0;
1302 phdr->p_flags = 0;
1303 phdr->p_align = 0;
1304 return;
1305 }
1306
1307 static void fill_note(struct memelfnote *note, const char *name, int type,
1308 unsigned int sz, void *data)
1309 {
1310 note->name = name;
1311 note->type = type;
1312 note->datasz = sz;
1313 note->data = data;
1314 return;
1315 }
1316
1317 /*
1318 * fill up all the fields in prstatus from the given task struct, except
1319 * registers which need to be filled up separately.
1320 */
1321 static void fill_prstatus(struct elf_prstatus *prstatus,
1322 struct task_struct *p, long signr)
1323 {
1324 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1325 prstatus->pr_sigpend = p->pending.signal.sig[0];
1326 prstatus->pr_sighold = p->blocked.sig[0];
1327 rcu_read_lock();
1328 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1329 rcu_read_unlock();
1330 prstatus->pr_pid = task_pid_vnr(p);
1331 prstatus->pr_pgrp = task_pgrp_vnr(p);
1332 prstatus->pr_sid = task_session_vnr(p);
1333 if (thread_group_leader(p)) {
1334 struct task_cputime cputime;
1335
1336 /*
1337 * This is the record for the group leader. It shows the
1338 * group-wide total, not its individual thread total.
1339 */
1340 thread_group_cputime(p, &cputime);
1341 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1342 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1343 } else {
1344 cputime_t utime, stime;
1345
1346 task_cputime(p, &utime, &stime);
1347 cputime_to_timeval(utime, &prstatus->pr_utime);
1348 cputime_to_timeval(stime, &prstatus->pr_stime);
1349 }
1350 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1351 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1352 }
1353
1354 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1355 struct mm_struct *mm)
1356 {
1357 const struct cred *cred;
1358 unsigned int i, len;
1359
1360 /* first copy the parameters from user space */
1361 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1362
1363 len = mm->arg_end - mm->arg_start;
1364 if (len >= ELF_PRARGSZ)
1365 len = ELF_PRARGSZ-1;
1366 if (copy_from_user(&psinfo->pr_psargs,
1367 (const char __user *)mm->arg_start, len))
1368 return -EFAULT;
1369 for(i = 0; i < len; i++)
1370 if (psinfo->pr_psargs[i] == 0)
1371 psinfo->pr_psargs[i] = ' ';
1372 psinfo->pr_psargs[len] = 0;
1373
1374 rcu_read_lock();
1375 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1376 rcu_read_unlock();
1377 psinfo->pr_pid = task_pid_vnr(p);
1378 psinfo->pr_pgrp = task_pgrp_vnr(p);
1379 psinfo->pr_sid = task_session_vnr(p);
1380
1381 i = p->state ? ffz(~p->state) + 1 : 0;
1382 psinfo->pr_state = i;
1383 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1384 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1385 psinfo->pr_nice = task_nice(p);
1386 psinfo->pr_flag = p->flags;
1387 rcu_read_lock();
1388 cred = __task_cred(p);
1389 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1390 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1391 rcu_read_unlock();
1392 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1393
1394 return 0;
1395 }
1396
1397 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1398 {
1399 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1400 int i = 0;
1401 do
1402 i += 2;
1403 while (auxv[i - 2] != AT_NULL);
1404 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1405 }
1406
1407 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1408 siginfo_t *siginfo)
1409 {
1410 mm_segment_t old_fs = get_fs();
1411 set_fs(KERNEL_DS);
1412 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1413 set_fs(old_fs);
1414 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1415 }
1416
1417 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1418 /*
1419 * Format of NT_FILE note:
1420 *
1421 * long count -- how many files are mapped
1422 * long page_size -- units for file_ofs
1423 * array of [COUNT] elements of
1424 * long start
1425 * long end
1426 * long file_ofs
1427 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1428 */
1429 static int fill_files_note(struct memelfnote *note)
1430 {
1431 struct vm_area_struct *vma;
1432 unsigned count, size, names_ofs, remaining, n;
1433 user_long_t *data;
1434 user_long_t *start_end_ofs;
1435 char *name_base, *name_curpos;
1436
1437 /* *Estimated* file count and total data size needed */
1438 count = current->mm->map_count;
1439 size = count * 64;
1440
1441 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1442 alloc:
1443 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1444 return -EINVAL;
1445 size = round_up(size, PAGE_SIZE);
1446 data = vmalloc(size);
1447 if (!data)
1448 return -ENOMEM;
1449
1450 start_end_ofs = data + 2;
1451 name_base = name_curpos = ((char *)data) + names_ofs;
1452 remaining = size - names_ofs;
1453 count = 0;
1454 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1455 struct file *file;
1456 const char *filename;
1457
1458 file = vma->vm_file;
1459 if (!file)
1460 continue;
1461 filename = d_path(&file->f_path, name_curpos, remaining);
1462 if (IS_ERR(filename)) {
1463 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1464 vfree(data);
1465 size = size * 5 / 4;
1466 goto alloc;
1467 }
1468 continue;
1469 }
1470
1471 /* d_path() fills at the end, move name down */
1472 /* n = strlen(filename) + 1: */
1473 n = (name_curpos + remaining) - filename;
1474 remaining = filename - name_curpos;
1475 memmove(name_curpos, filename, n);
1476 name_curpos += n;
1477
1478 *start_end_ofs++ = vma->vm_start;
1479 *start_end_ofs++ = vma->vm_end;
1480 *start_end_ofs++ = vma->vm_pgoff;
1481 count++;
1482 }
1483
1484 /* Now we know exact count of files, can store it */
1485 data[0] = count;
1486 data[1] = PAGE_SIZE;
1487 /*
1488 * Count usually is less than current->mm->map_count,
1489 * we need to move filenames down.
1490 */
1491 n = current->mm->map_count - count;
1492 if (n != 0) {
1493 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1494 memmove(name_base - shift_bytes, name_base,
1495 name_curpos - name_base);
1496 name_curpos -= shift_bytes;
1497 }
1498
1499 size = name_curpos - (char *)data;
1500 fill_note(note, "CORE", NT_FILE, size, data);
1501 return 0;
1502 }
1503
1504 #ifdef CORE_DUMP_USE_REGSET
1505 #include <linux/regset.h>
1506
1507 struct elf_thread_core_info {
1508 struct elf_thread_core_info *next;
1509 struct task_struct *task;
1510 struct elf_prstatus prstatus;
1511 struct memelfnote notes[0];
1512 };
1513
1514 struct elf_note_info {
1515 struct elf_thread_core_info *thread;
1516 struct memelfnote psinfo;
1517 struct memelfnote signote;
1518 struct memelfnote auxv;
1519 struct memelfnote files;
1520 user_siginfo_t csigdata;
1521 size_t size;
1522 int thread_notes;
1523 };
1524
1525 /*
1526 * When a regset has a writeback hook, we call it on each thread before
1527 * dumping user memory. On register window machines, this makes sure the
1528 * user memory backing the register data is up to date before we read it.
1529 */
1530 static void do_thread_regset_writeback(struct task_struct *task,
1531 const struct user_regset *regset)
1532 {
1533 if (regset->writeback)
1534 regset->writeback(task, regset, 1);
1535 }
1536
1537 #ifndef PR_REG_SIZE
1538 #define PR_REG_SIZE(S) sizeof(S)
1539 #endif
1540
1541 #ifndef PRSTATUS_SIZE
1542 #define PRSTATUS_SIZE(S) sizeof(S)
1543 #endif
1544
1545 #ifndef PR_REG_PTR
1546 #define PR_REG_PTR(S) (&((S)->pr_reg))
1547 #endif
1548
1549 #ifndef SET_PR_FPVALID
1550 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1551 #endif
1552
1553 static int fill_thread_core_info(struct elf_thread_core_info *t,
1554 const struct user_regset_view *view,
1555 long signr, size_t *total)
1556 {
1557 unsigned int i;
1558
1559 /*
1560 * NT_PRSTATUS is the one special case, because the regset data
1561 * goes into the pr_reg field inside the note contents, rather
1562 * than being the whole note contents. We fill the reset in here.
1563 * We assume that regset 0 is NT_PRSTATUS.
1564 */
1565 fill_prstatus(&t->prstatus, t->task, signr);
1566 (void) view->regsets[0].get(t->task, &view->regsets[0],
1567 0, PR_REG_SIZE(t->prstatus.pr_reg),
1568 PR_REG_PTR(&t->prstatus), NULL);
1569
1570 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1571 PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1572 *total += notesize(&t->notes[0]);
1573
1574 do_thread_regset_writeback(t->task, &view->regsets[0]);
1575
1576 /*
1577 * Each other regset might generate a note too. For each regset
1578 * that has no core_note_type or is inactive, we leave t->notes[i]
1579 * all zero and we'll know to skip writing it later.
1580 */
1581 for (i = 1; i < view->n; ++i) {
1582 const struct user_regset *regset = &view->regsets[i];
1583 do_thread_regset_writeback(t->task, regset);
1584 if (regset->core_note_type && regset->get &&
1585 (!regset->active || regset->active(t->task, regset))) {
1586 int ret;
1587 size_t size = regset->n * regset->size;
1588 void *data = kmalloc(size, GFP_KERNEL);
1589 if (unlikely(!data))
1590 return 0;
1591 ret = regset->get(t->task, regset,
1592 0, size, data, NULL);
1593 if (unlikely(ret))
1594 kfree(data);
1595 else {
1596 if (regset->core_note_type != NT_PRFPREG)
1597 fill_note(&t->notes[i], "LINUX",
1598 regset->core_note_type,
1599 size, data);
1600 else {
1601 SET_PR_FPVALID(&t->prstatus, 1);
1602 fill_note(&t->notes[i], "CORE",
1603 NT_PRFPREG, size, data);
1604 }
1605 *total += notesize(&t->notes[i]);
1606 }
1607 }
1608 }
1609
1610 return 1;
1611 }
1612
1613 static int fill_note_info(struct elfhdr *elf, int phdrs,
1614 struct elf_note_info *info,
1615 siginfo_t *siginfo, struct pt_regs *regs)
1616 {
1617 struct task_struct *dump_task = current;
1618 const struct user_regset_view *view = task_user_regset_view(dump_task);
1619 struct elf_thread_core_info *t;
1620 struct elf_prpsinfo *psinfo;
1621 struct core_thread *ct;
1622 unsigned int i;
1623
1624 info->size = 0;
1625 info->thread = NULL;
1626
1627 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1628 if (psinfo == NULL) {
1629 info->psinfo.data = NULL; /* So we don't free this wrongly */
1630 return 0;
1631 }
1632
1633 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1634
1635 /*
1636 * Figure out how many notes we're going to need for each thread.
1637 */
1638 info->thread_notes = 0;
1639 for (i = 0; i < view->n; ++i)
1640 if (view->regsets[i].core_note_type != 0)
1641 ++info->thread_notes;
1642
1643 /*
1644 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1645 * since it is our one special case.
1646 */
1647 if (unlikely(info->thread_notes == 0) ||
1648 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1649 WARN_ON(1);
1650 return 0;
1651 }
1652
1653 /*
1654 * Initialize the ELF file header.
1655 */
1656 fill_elf_header(elf, phdrs,
1657 view->e_machine, view->e_flags);
1658
1659 /*
1660 * Allocate a structure for each thread.
1661 */
1662 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1663 t = kzalloc(offsetof(struct elf_thread_core_info,
1664 notes[info->thread_notes]),
1665 GFP_KERNEL);
1666 if (unlikely(!t))
1667 return 0;
1668
1669 t->task = ct->task;
1670 if (ct->task == dump_task || !info->thread) {
1671 t->next = info->thread;
1672 info->thread = t;
1673 } else {
1674 /*
1675 * Make sure to keep the original task at
1676 * the head of the list.
1677 */
1678 t->next = info->thread->next;
1679 info->thread->next = t;
1680 }
1681 }
1682
1683 /*
1684 * Now fill in each thread's information.
1685 */
1686 for (t = info->thread; t != NULL; t = t->next)
1687 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1688 return 0;
1689
1690 /*
1691 * Fill in the two process-wide notes.
1692 */
1693 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1694 info->size += notesize(&info->psinfo);
1695
1696 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1697 info->size += notesize(&info->signote);
1698
1699 fill_auxv_note(&info->auxv, current->mm);
1700 info->size += notesize(&info->auxv);
1701
1702 if (fill_files_note(&info->files) == 0)
1703 info->size += notesize(&info->files);
1704
1705 return 1;
1706 }
1707
1708 static size_t get_note_info_size(struct elf_note_info *info)
1709 {
1710 return info->size;
1711 }
1712
1713 /*
1714 * Write all the notes for each thread. When writing the first thread, the
1715 * process-wide notes are interleaved after the first thread-specific note.
1716 */
1717 static int write_note_info(struct elf_note_info *info,
1718 struct file *file, loff_t *foffset)
1719 {
1720 bool first = 1;
1721 struct elf_thread_core_info *t = info->thread;
1722
1723 do {
1724 int i;
1725
1726 if (!writenote(&t->notes[0], file, foffset))
1727 return 0;
1728
1729 if (first && !writenote(&info->psinfo, file, foffset))
1730 return 0;
1731 if (first && !writenote(&info->signote, file, foffset))
1732 return 0;
1733 if (first && !writenote(&info->auxv, file, foffset))
1734 return 0;
1735 if (first && info->files.data &&
1736 !writenote(&info->files, file, foffset))
1737 return 0;
1738
1739 for (i = 1; i < info->thread_notes; ++i)
1740 if (t->notes[i].data &&
1741 !writenote(&t->notes[i], file, foffset))
1742 return 0;
1743
1744 first = 0;
1745 t = t->next;
1746 } while (t);
1747
1748 return 1;
1749 }
1750
1751 static void free_note_info(struct elf_note_info *info)
1752 {
1753 struct elf_thread_core_info *threads = info->thread;
1754 while (threads) {
1755 unsigned int i;
1756 struct elf_thread_core_info *t = threads;
1757 threads = t->next;
1758 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1759 for (i = 1; i < info->thread_notes; ++i)
1760 kfree(t->notes[i].data);
1761 kfree(t);
1762 }
1763 kfree(info->psinfo.data);
1764 vfree(info->files.data);
1765 }
1766
1767 #else
1768
1769 /* Here is the structure in which status of each thread is captured. */
1770 struct elf_thread_status
1771 {
1772 struct list_head list;
1773 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1774 elf_fpregset_t fpu; /* NT_PRFPREG */
1775 struct task_struct *thread;
1776 #ifdef ELF_CORE_COPY_XFPREGS
1777 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1778 #endif
1779 struct memelfnote notes[3];
1780 int num_notes;
1781 };
1782
1783 /*
1784 * In order to add the specific thread information for the elf file format,
1785 * we need to keep a linked list of every threads pr_status and then create
1786 * a single section for them in the final core file.
1787 */
1788 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1789 {
1790 int sz = 0;
1791 struct task_struct *p = t->thread;
1792 t->num_notes = 0;
1793
1794 fill_prstatus(&t->prstatus, p, signr);
1795 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1796
1797 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1798 &(t->prstatus));
1799 t->num_notes++;
1800 sz += notesize(&t->notes[0]);
1801
1802 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1803 &t->fpu))) {
1804 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1805 &(t->fpu));
1806 t->num_notes++;
1807 sz += notesize(&t->notes[1]);
1808 }
1809
1810 #ifdef ELF_CORE_COPY_XFPREGS
1811 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1812 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1813 sizeof(t->xfpu), &t->xfpu);
1814 t->num_notes++;
1815 sz += notesize(&t->notes[2]);
1816 }
1817 #endif
1818 return sz;
1819 }
1820
1821 struct elf_note_info {
1822 struct memelfnote *notes;
1823 struct memelfnote *notes_files;
1824 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1825 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1826 struct list_head thread_list;
1827 elf_fpregset_t *fpu;
1828 #ifdef ELF_CORE_COPY_XFPREGS
1829 elf_fpxregset_t *xfpu;
1830 #endif
1831 user_siginfo_t csigdata;
1832 int thread_status_size;
1833 int numnote;
1834 };
1835
1836 static int elf_note_info_init(struct elf_note_info *info)
1837 {
1838 memset(info, 0, sizeof(*info));
1839 INIT_LIST_HEAD(&info->thread_list);
1840
1841 /* Allocate space for ELF notes */
1842 info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1843 if (!info->notes)
1844 return 0;
1845 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1846 if (!info->psinfo)
1847 return 0;
1848 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1849 if (!info->prstatus)
1850 return 0;
1851 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1852 if (!info->fpu)
1853 return 0;
1854 #ifdef ELF_CORE_COPY_XFPREGS
1855 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1856 if (!info->xfpu)
1857 return 0;
1858 #endif
1859 return 1;
1860 }
1861
1862 static int fill_note_info(struct elfhdr *elf, int phdrs,
1863 struct elf_note_info *info,
1864 siginfo_t *siginfo, struct pt_regs *regs)
1865 {
1866 struct list_head *t;
1867
1868 if (!elf_note_info_init(info))
1869 return 0;
1870
1871 if (siginfo->si_signo) {
1872 struct core_thread *ct;
1873 struct elf_thread_status *ets;
1874
1875 for (ct = current->mm->core_state->dumper.next;
1876 ct; ct = ct->next) {
1877 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1878 if (!ets)
1879 return 0;
1880
1881 ets->thread = ct->task;
1882 list_add(&ets->list, &info->thread_list);
1883 }
1884
1885 list_for_each(t, &info->thread_list) {
1886 int sz;
1887
1888 ets = list_entry(t, struct elf_thread_status, list);
1889 sz = elf_dump_thread_status(siginfo->si_signo, ets);
1890 info->thread_status_size += sz;
1891 }
1892 }
1893 /* now collect the dump for the current */
1894 memset(info->prstatus, 0, sizeof(*info->prstatus));
1895 fill_prstatus(info->prstatus, current, siginfo->si_signo);
1896 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1897
1898 /* Set up header */
1899 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1900
1901 /*
1902 * Set up the notes in similar form to SVR4 core dumps made
1903 * with info from their /proc.
1904 */
1905
1906 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1907 sizeof(*info->prstatus), info->prstatus);
1908 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1909 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1910 sizeof(*info->psinfo), info->psinfo);
1911
1912 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1913 fill_auxv_note(info->notes + 3, current->mm);
1914 info->numnote = 4;
1915
1916 if (fill_files_note(info->notes + info->numnote) == 0) {
1917 info->notes_files = info->notes + info->numnote;
1918 info->numnote++;
1919 }
1920
1921 /* Try to dump the FPU. */
1922 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1923 info->fpu);
1924 if (info->prstatus->pr_fpvalid)
1925 fill_note(info->notes + info->numnote++,
1926 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1927 #ifdef ELF_CORE_COPY_XFPREGS
1928 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1929 fill_note(info->notes + info->numnote++,
1930 "LINUX", ELF_CORE_XFPREG_TYPE,
1931 sizeof(*info->xfpu), info->xfpu);
1932 #endif
1933
1934 return 1;
1935 }
1936
1937 static size_t get_note_info_size(struct elf_note_info *info)
1938 {
1939 int sz = 0;
1940 int i;
1941
1942 for (i = 0; i < info->numnote; i++)
1943 sz += notesize(info->notes + i);
1944
1945 sz += info->thread_status_size;
1946
1947 return sz;
1948 }
1949
1950 static int write_note_info(struct elf_note_info *info,
1951 struct file *file, loff_t *foffset)
1952 {
1953 int i;
1954 struct list_head *t;
1955
1956 for (i = 0; i < info->numnote; i++)
1957 if (!writenote(info->notes + i, file, foffset))
1958 return 0;
1959
1960 /* write out the thread status notes section */
1961 list_for_each(t, &info->thread_list) {
1962 struct elf_thread_status *tmp =
1963 list_entry(t, struct elf_thread_status, list);
1964
1965 for (i = 0; i < tmp->num_notes; i++)
1966 if (!writenote(&tmp->notes[i], file, foffset))
1967 return 0;
1968 }
1969
1970 return 1;
1971 }
1972
1973 static void free_note_info(struct elf_note_info *info)
1974 {
1975 while (!list_empty(&info->thread_list)) {
1976 struct list_head *tmp = info->thread_list.next;
1977 list_del(tmp);
1978 kfree(list_entry(tmp, struct elf_thread_status, list));
1979 }
1980
1981 /* Free data possibly allocated by fill_files_note(): */
1982 if (info->notes_files)
1983 vfree(info->notes_files->data);
1984
1985 kfree(info->prstatus);
1986 kfree(info->psinfo);
1987 kfree(info->notes);
1988 kfree(info->fpu);
1989 #ifdef ELF_CORE_COPY_XFPREGS
1990 kfree(info->xfpu);
1991 #endif
1992 }
1993
1994 #endif
1995
1996 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1997 struct vm_area_struct *gate_vma)
1998 {
1999 struct vm_area_struct *ret = tsk->mm->mmap;
2000
2001 if (ret)
2002 return ret;
2003 return gate_vma;
2004 }
2005 /*
2006 * Helper function for iterating across a vma list. It ensures that the caller
2007 * will visit `gate_vma' prior to terminating the search.
2008 */
2009 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2010 struct vm_area_struct *gate_vma)
2011 {
2012 struct vm_area_struct *ret;
2013
2014 ret = this_vma->vm_next;
2015 if (ret)
2016 return ret;
2017 if (this_vma == gate_vma)
2018 return NULL;
2019 return gate_vma;
2020 }
2021
2022 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2023 elf_addr_t e_shoff, int segs)
2024 {
2025 elf->e_shoff = e_shoff;
2026 elf->e_shentsize = sizeof(*shdr4extnum);
2027 elf->e_shnum = 1;
2028 elf->e_shstrndx = SHN_UNDEF;
2029
2030 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2031
2032 shdr4extnum->sh_type = SHT_NULL;
2033 shdr4extnum->sh_size = elf->e_shnum;
2034 shdr4extnum->sh_link = elf->e_shstrndx;
2035 shdr4extnum->sh_info = segs;
2036 }
2037
2038 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2039 unsigned long mm_flags)
2040 {
2041 struct vm_area_struct *vma;
2042 size_t size = 0;
2043
2044 for (vma = first_vma(current, gate_vma); vma != NULL;
2045 vma = next_vma(vma, gate_vma))
2046 size += vma_dump_size(vma, mm_flags);
2047 return size;
2048 }
2049
2050 /*
2051 * Actual dumper
2052 *
2053 * This is a two-pass process; first we find the offsets of the bits,
2054 * and then they are actually written out. If we run out of core limit
2055 * we just truncate.
2056 */
2057 static int elf_core_dump(struct coredump_params *cprm)
2058 {
2059 int has_dumped = 0;
2060 mm_segment_t fs;
2061 int segs;
2062 size_t size = 0;
2063 struct vm_area_struct *vma, *gate_vma;
2064 struct elfhdr *elf = NULL;
2065 loff_t offset = 0, dataoff, foffset;
2066 struct elf_note_info info = { };
2067 struct elf_phdr *phdr4note = NULL;
2068 struct elf_shdr *shdr4extnum = NULL;
2069 Elf_Half e_phnum;
2070 elf_addr_t e_shoff;
2071
2072 printk(KERN_WARNING "coredump(%d): start\n", current->pid);
2073
2074 /*
2075 * We no longer stop all VM operations.
2076 *
2077 * This is because those proceses that could possibly change map_count
2078 * or the mmap / vma pages are now blocked in do_exit on current
2079 * finishing this core dump.
2080 *
2081 * Only ptrace can touch these memory addresses, but it doesn't change
2082 * the map_count or the pages allocated. So no possibility of crashing
2083 * exists while dumping the mm->vm_next areas to the core file.
2084 */
2085
2086 /* alloc memory for large data structures: too large to be on stack */
2087 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2088 if (!elf)
2089 goto out;
2090 /*
2091 * The number of segs are recored into ELF header as 16bit value.
2092 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2093 */
2094 segs = current->mm->map_count;
2095 segs += elf_core_extra_phdrs();
2096
2097 gate_vma = get_gate_vma(current->mm);
2098 if (gate_vma != NULL)
2099 segs++;
2100
2101 /* for notes section */
2102 segs++;
2103
2104 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2105 * this, kernel supports extended numbering. Have a look at
2106 * include/linux/elf.h for further information. */
2107 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2108
2109 /*
2110 * Collect all the non-memory information about the process for the
2111 * notes. This also sets up the file header.
2112 */
2113 if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2114 goto cleanup;
2115
2116 has_dumped = 1;
2117
2118 fs = get_fs();
2119 set_fs(KERNEL_DS);
2120
2121 offset += sizeof(*elf); /* Elf header */
2122 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2123 foffset = offset;
2124
2125 /* Write notes phdr entry */
2126 {
2127 size_t sz = get_note_info_size(&info);
2128
2129 sz += elf_coredump_extra_notes_size();
2130
2131 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2132 if (!phdr4note)
2133 goto end_coredump;
2134
2135 fill_elf_note_phdr(phdr4note, sz, offset);
2136 offset += sz;
2137 }
2138
2139 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2140
2141 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2142 offset += elf_core_extra_data_size();
2143 e_shoff = offset;
2144
2145 if (e_phnum == PN_XNUM) {
2146 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2147 if (!shdr4extnum)
2148 goto end_coredump;
2149 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2150 }
2151
2152 offset = dataoff;
2153
2154 size += sizeof(*elf);
2155 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2156 goto end_coredump;
2157
2158 size += sizeof(*phdr4note);
2159 if (size > cprm->limit
2160 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2161 goto end_coredump;
2162
2163 /* Write program headers for segments dump */
2164 for (vma = first_vma(current, gate_vma); vma != NULL;
2165 vma = next_vma(vma, gate_vma)) {
2166 struct elf_phdr phdr;
2167
2168 phdr.p_type = PT_LOAD;
2169 phdr.p_offset = offset;
2170 phdr.p_vaddr = vma->vm_start;
2171 phdr.p_paddr = 0;
2172 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2173 phdr.p_memsz = vma->vm_end - vma->vm_start;
2174 offset += phdr.p_filesz;
2175 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2176 if (vma->vm_flags & VM_WRITE)
2177 phdr.p_flags |= PF_W;
2178 if (vma->vm_flags & VM_EXEC)
2179 phdr.p_flags |= PF_X;
2180 phdr.p_align = ELF_EXEC_PAGESIZE;
2181
2182 size += sizeof(phdr);
2183 if (size > cprm->limit
2184 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2185 goto end_coredump;
2186 }
2187
2188 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2189 goto end_coredump;
2190
2191 /* write out the notes section */
2192 if (!write_note_info(&info, cprm->file, &foffset))
2193 goto end_coredump;
2194
2195 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2196 goto end_coredump;
2197
2198 /* Align to page */
2199 if (!dump_seek(cprm->file, dataoff - foffset))
2200 goto end_coredump;
2201
2202 printk(KERN_WARNING "coredump(%d): write output program header and notes\n", current->pid);
2203
2204 for (vma = first_vma(current, gate_vma); vma != NULL;
2205 vma = next_vma(vma, gate_vma)) {
2206 unsigned long addr;
2207 unsigned long end;
2208
2209 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2210
2211 #ifdef CONFIG_MTK_EXTMEM
2212 if (extmem_in_mspace(vma)) {
2213 void *extmem_va = (void *)get_virt_from_mspace(vma->vm_pgoff << PAGE_SHIFT);
2214 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE, extmem_va += PAGE_SIZE) {
2215 int stop;
2216 int dump_write_ret = dump_write(cprm->file, extmem_va, PAGE_SIZE);
2217 stop = ((size += PAGE_SIZE) > cprm->limit) || (!dump_write_ret);
2218 if (stop) {
2219 printk(KERN_WARNING "[EXT_MEM]stop addr:0x%lx, size:%zx, limit:0x%lx, dump_write_ret:%d\n",
2220 addr, size, cprm->limit, dump_write_ret);
2221 goto end_coredump;
2222 }
2223 }
2224 continue;
2225 }
2226 #endif
2227
2228 //printk(KERN_WARNING "coredump(%d): write out load vm start:%08lx, end:%08lx\n", current->pid, vma->vm_start, end);
2229 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2230 struct page *page;
2231 int stop;
2232
2233 page = get_dump_page(addr);
2234 if (page) {
2235 void *kaddr = kmap(page);
2236 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2237 !dump_write(cprm->file, kaddr,
2238 PAGE_SIZE);
2239 kunmap(page);
2240 page_cache_release(page);
2241 if (stop) {
2242 printk(KERN_WARNING "coredump(%d): failed to write core dump\n", current->pid);
2243 }
2244 } else {
2245 stop = !dump_seek(cprm->file, PAGE_SIZE);
2246 if (stop) {
2247 printk(KERN_WARNING "coredump(%d): failed to seek core dump\n", current->pid);
2248 }
2249 }
2250 if (stop)
2251 goto end_coredump;
2252 }
2253 }
2254
2255 printk(KERN_WARNING "coredump(%d): write loads\n", current->pid);
2256
2257 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2258 goto end_coredump;
2259
2260 if (e_phnum == PN_XNUM) {
2261 size += sizeof(*shdr4extnum);
2262 if (size > cprm->limit
2263 || !dump_write(cprm->file, shdr4extnum,
2264 sizeof(*shdr4extnum)))
2265 goto end_coredump;
2266 }
2267
2268 printk(KERN_WARNING "coredump(%d): write out completed %lld\n", current->pid, offset);
2269
2270 end_coredump:
2271 set_fs(fs);
2272
2273 cleanup:
2274 free_note_info(&info);
2275 kfree(shdr4extnum);
2276 kfree(phdr4note);
2277 kfree(elf);
2278 out:
2279 return has_dumped;
2280 }
2281
2282 #endif /* CONFIG_ELF_CORE */
2283
2284 static int __init init_elf_binfmt(void)
2285 {
2286 register_binfmt(&elf_format);
2287 return 0;
2288 }
2289
2290 static void __exit exit_elf_binfmt(void)
2291 {
2292 /* Remove the COFF and ELF loaders. */
2293 unregister_binfmt(&elf_format);
2294 }
2295
2296 core_initcall(init_elf_binfmt);
2297 module_exit(exit_elf_binfmt);
2298 MODULE_LICENSE("GPL");
kernel source for telekom puls (alcatel/mediatek)