Merge branch 'parisc-4.7-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
[deliverable/linux.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/elf-randomize.h>
35 #include <linux/utsname.h>
36 #include <linux/coredump.h>
37 #include <linux/sched.h>
38 #include <linux/dax.h>
39 #include <asm/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/page.h>
42
43 #ifndef user_long_t
44 #define user_long_t long
45 #endif
46 #ifndef user_siginfo_t
47 #define user_siginfo_t siginfo_t
48 #endif
49
50 static int load_elf_binary(struct linux_binprm *bprm);
51 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
52 int, int, unsigned long);
53
54 #ifdef CONFIG_USELIB
55 static int load_elf_library(struct file *);
56 #else
57 #define load_elf_library NULL
58 #endif
59
60 /*
61 * If we don't support core dumping, then supply a NULL so we
62 * don't even try.
63 */
64 #ifdef CONFIG_ELF_CORE
65 static int elf_core_dump(struct coredump_params *cprm);
66 #else
67 #define elf_core_dump NULL
68 #endif
69
70 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
71 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
72 #else
73 #define ELF_MIN_ALIGN PAGE_SIZE
74 #endif
75
76 #ifndef ELF_CORE_EFLAGS
77 #define ELF_CORE_EFLAGS 0
78 #endif
79
80 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
81 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
82 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
83
84 static struct linux_binfmt elf_format = {
85 .module = THIS_MODULE,
86 .load_binary = load_elf_binary,
87 .load_shlib = load_elf_library,
88 .core_dump = elf_core_dump,
89 .min_coredump = ELF_EXEC_PAGESIZE,
90 };
91
92 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
93
94 static int set_brk(unsigned long start, unsigned long end)
95 {
96 start = ELF_PAGEALIGN(start);
97 end = ELF_PAGEALIGN(end);
98 if (end > start) {
99 unsigned long addr;
100 addr = vm_brk(start, end - start);
101 if (BAD_ADDR(addr))
102 return addr;
103 }
104 current->mm->start_brk = current->mm->brk = end;
105 return 0;
106 }
107
108 /* We need to explicitly zero any fractional pages
109 after the data section (i.e. bss). This would
110 contain the junk from the file that should not
111 be in memory
112 */
113 static int padzero(unsigned long elf_bss)
114 {
115 unsigned long nbyte;
116
117 nbyte = ELF_PAGEOFFSET(elf_bss);
118 if (nbyte) {
119 nbyte = ELF_MIN_ALIGN - nbyte;
120 if (clear_user((void __user *) elf_bss, nbyte))
121 return -EFAULT;
122 }
123 return 0;
124 }
125
126 /* Let's use some macros to make this stack manipulation a little clearer */
127 #ifdef CONFIG_STACK_GROWSUP
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
129 #define STACK_ROUND(sp, items) \
130 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ \
132 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
133 old_sp; })
134 #else
135 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
136 #define STACK_ROUND(sp, items) \
137 (((unsigned long) (sp - items)) &~ 15UL)
138 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
139 #endif
140
141 #ifndef ELF_BASE_PLATFORM
142 /*
143 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
144 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
145 * will be copied to the user stack in the same manner as AT_PLATFORM.
146 */
147 #define ELF_BASE_PLATFORM NULL
148 #endif
149
150 static int
151 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
152 unsigned long load_addr, unsigned long interp_load_addr)
153 {
154 unsigned long p = bprm->p;
155 int argc = bprm->argc;
156 int envc = bprm->envc;
157 elf_addr_t __user *argv;
158 elf_addr_t __user *envp;
159 elf_addr_t __user *sp;
160 elf_addr_t __user *u_platform;
161 elf_addr_t __user *u_base_platform;
162 elf_addr_t __user *u_rand_bytes;
163 const char *k_platform = ELF_PLATFORM;
164 const char *k_base_platform = ELF_BASE_PLATFORM;
165 unsigned char k_rand_bytes[16];
166 int items;
167 elf_addr_t *elf_info;
168 int ei_index = 0;
169 const struct cred *cred = current_cred();
170 struct vm_area_struct *vma;
171
172 /*
173 * In some cases (e.g. Hyper-Threading), we want to avoid L1
174 * evictions by the processes running on the same package. One
175 * thing we can do is to shuffle the initial stack for them.
176 */
177
178 p = arch_align_stack(p);
179
180 /*
181 * If this architecture has a platform capability string, copy it
182 * to userspace. In some cases (Sparc), this info is impossible
183 * for userspace to get any other way, in others (i386) it is
184 * merely difficult.
185 */
186 u_platform = NULL;
187 if (k_platform) {
188 size_t len = strlen(k_platform) + 1;
189
190 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
191 if (__copy_to_user(u_platform, k_platform, len))
192 return -EFAULT;
193 }
194
195 /*
196 * If this architecture has a "base" platform capability
197 * string, copy it to userspace.
198 */
199 u_base_platform = NULL;
200 if (k_base_platform) {
201 size_t len = strlen(k_base_platform) + 1;
202
203 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
204 if (__copy_to_user(u_base_platform, k_base_platform, len))
205 return -EFAULT;
206 }
207
208 /*
209 * Generate 16 random bytes for userspace PRNG seeding.
210 */
211 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
212 u_rand_bytes = (elf_addr_t __user *)
213 STACK_ALLOC(p, sizeof(k_rand_bytes));
214 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
215 return -EFAULT;
216
217 /* Create the ELF interpreter info */
218 elf_info = (elf_addr_t *)current->mm->saved_auxv;
219 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
220 #define NEW_AUX_ENT(id, val) \
221 do { \
222 elf_info[ei_index++] = id; \
223 elf_info[ei_index++] = val; \
224 } while (0)
225
226 #ifdef ARCH_DLINFO
227 /*
228 * ARCH_DLINFO must come first so PPC can do its special alignment of
229 * AUXV.
230 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
231 * ARCH_DLINFO changes
232 */
233 ARCH_DLINFO;
234 #endif
235 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
236 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
237 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
238 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
239 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
240 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
241 NEW_AUX_ENT(AT_BASE, interp_load_addr);
242 NEW_AUX_ENT(AT_FLAGS, 0);
243 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
244 NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
245 NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
246 NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
247 NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
248 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
249 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
250 #ifdef ELF_HWCAP2
251 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
252 #endif
253 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
254 if (k_platform) {
255 NEW_AUX_ENT(AT_PLATFORM,
256 (elf_addr_t)(unsigned long)u_platform);
257 }
258 if (k_base_platform) {
259 NEW_AUX_ENT(AT_BASE_PLATFORM,
260 (elf_addr_t)(unsigned long)u_base_platform);
261 }
262 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
263 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
264 }
265 #undef NEW_AUX_ENT
266 /* AT_NULL is zero; clear the rest too */
267 memset(&elf_info[ei_index], 0,
268 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
269
270 /* And advance past the AT_NULL entry. */
271 ei_index += 2;
272
273 sp = STACK_ADD(p, ei_index);
274
275 items = (argc + 1) + (envc + 1) + 1;
276 bprm->p = STACK_ROUND(sp, items);
277
278 /* Point sp at the lowest address on the stack */
279 #ifdef CONFIG_STACK_GROWSUP
280 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
281 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
282 #else
283 sp = (elf_addr_t __user *)bprm->p;
284 #endif
285
286
287 /*
288 * Grow the stack manually; some architectures have a limit on how
289 * far ahead a user-space access may be in order to grow the stack.
290 */
291 vma = find_extend_vma(current->mm, bprm->p);
292 if (!vma)
293 return -EFAULT;
294
295 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
296 if (__put_user(argc, sp++))
297 return -EFAULT;
298 argv = sp;
299 envp = argv + argc + 1;
300
301 /* Populate argv and envp */
302 p = current->mm->arg_end = current->mm->arg_start;
303 while (argc-- > 0) {
304 size_t len;
305 if (__put_user((elf_addr_t)p, argv++))
306 return -EFAULT;
307 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
308 if (!len || len > MAX_ARG_STRLEN)
309 return -EINVAL;
310 p += len;
311 }
312 if (__put_user(0, argv))
313 return -EFAULT;
314 current->mm->arg_end = current->mm->env_start = p;
315 while (envc-- > 0) {
316 size_t len;
317 if (__put_user((elf_addr_t)p, envp++))
318 return -EFAULT;
319 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
320 if (!len || len > MAX_ARG_STRLEN)
321 return -EINVAL;
322 p += len;
323 }
324 if (__put_user(0, envp))
325 return -EFAULT;
326 current->mm->env_end = p;
327
328 /* Put the elf_info on the stack in the right place. */
329 sp = (elf_addr_t __user *)envp + 1;
330 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
331 return -EFAULT;
332 return 0;
333 }
334
335 #ifndef elf_map
336
337 static unsigned long elf_map(struct file *filep, unsigned long addr,
338 struct elf_phdr *eppnt, int prot, int type,
339 unsigned long total_size)
340 {
341 unsigned long map_addr;
342 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
343 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
344 addr = ELF_PAGESTART(addr);
345 size = ELF_PAGEALIGN(size);
346
347 /* mmap() will return -EINVAL if given a zero size, but a
348 * segment with zero filesize is perfectly valid */
349 if (!size)
350 return addr;
351
352 /*
353 * total_size is the size of the ELF (interpreter) image.
354 * The _first_ mmap needs to know the full size, otherwise
355 * randomization might put this image into an overlapping
356 * position with the ELF binary image. (since size < total_size)
357 * So we first map the 'big' image - and unmap the remainder at
358 * the end. (which unmap is needed for ELF images with holes.)
359 */
360 if (total_size) {
361 total_size = ELF_PAGEALIGN(total_size);
362 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
363 if (!BAD_ADDR(map_addr))
364 vm_munmap(map_addr+size, total_size-size);
365 } else
366 map_addr = vm_mmap(filep, addr, size, prot, type, off);
367
368 return(map_addr);
369 }
370
371 #endif /* !elf_map */
372
373 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
374 {
375 int i, first_idx = -1, last_idx = -1;
376
377 for (i = 0; i < nr; i++) {
378 if (cmds[i].p_type == PT_LOAD) {
379 last_idx = i;
380 if (first_idx == -1)
381 first_idx = i;
382 }
383 }
384 if (first_idx == -1)
385 return 0;
386
387 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
388 ELF_PAGESTART(cmds[first_idx].p_vaddr);
389 }
390
391 /**
392 * load_elf_phdrs() - load ELF program headers
393 * @elf_ex: ELF header of the binary whose program headers should be loaded
394 * @elf_file: the opened ELF binary file
395 *
396 * Loads ELF program headers from the binary file elf_file, which has the ELF
397 * header pointed to by elf_ex, into a newly allocated array. The caller is
398 * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
399 */
400 static struct elf_phdr *load_elf_phdrs(struct elfhdr *elf_ex,
401 struct file *elf_file)
402 {
403 struct elf_phdr *elf_phdata = NULL;
404 int retval, size, err = -1;
405
406 /*
407 * If the size of this structure has changed, then punt, since
408 * we will be doing the wrong thing.
409 */
410 if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
411 goto out;
412
413 /* Sanity check the number of program headers... */
414 if (elf_ex->e_phnum < 1 ||
415 elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
416 goto out;
417
418 /* ...and their total size. */
419 size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
420 if (size > ELF_MIN_ALIGN)
421 goto out;
422
423 elf_phdata = kmalloc(size, GFP_KERNEL);
424 if (!elf_phdata)
425 goto out;
426
427 /* Read in the program headers */
428 retval = kernel_read(elf_file, elf_ex->e_phoff,
429 (char *)elf_phdata, size);
430 if (retval != size) {
431 err = (retval < 0) ? retval : -EIO;
432 goto out;
433 }
434
435 /* Success! */
436 err = 0;
437 out:
438 if (err) {
439 kfree(elf_phdata);
440 elf_phdata = NULL;
441 }
442 return elf_phdata;
443 }
444
445 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
446
447 /**
448 * struct arch_elf_state - arch-specific ELF loading state
449 *
450 * This structure is used to preserve architecture specific data during
451 * the loading of an ELF file, throughout the checking of architecture
452 * specific ELF headers & through to the point where the ELF load is
453 * known to be proceeding (ie. SET_PERSONALITY).
454 *
455 * This implementation is a dummy for architectures which require no
456 * specific state.
457 */
458 struct arch_elf_state {
459 };
460
461 #define INIT_ARCH_ELF_STATE {}
462
463 /**
464 * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
465 * @ehdr: The main ELF header
466 * @phdr: The program header to check
467 * @elf: The open ELF file
468 * @is_interp: True if the phdr is from the interpreter of the ELF being
469 * loaded, else false.
470 * @state: Architecture-specific state preserved throughout the process
471 * of loading the ELF.
472 *
473 * Inspects the program header phdr to validate its correctness and/or
474 * suitability for the system. Called once per ELF program header in the
475 * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
476 * interpreter.
477 *
478 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
479 * with that return code.
480 */
481 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
482 struct elf_phdr *phdr,
483 struct file *elf, bool is_interp,
484 struct arch_elf_state *state)
485 {
486 /* Dummy implementation, always proceed */
487 return 0;
488 }
489
490 /**
491 * arch_check_elf() - check an ELF executable
492 * @ehdr: The main ELF header
493 * @has_interp: True if the ELF has an interpreter, else false.
494 * @interp_ehdr: The interpreter's ELF header
495 * @state: Architecture-specific state preserved throughout the process
496 * of loading the ELF.
497 *
498 * Provides a final opportunity for architecture code to reject the loading
499 * of the ELF & cause an exec syscall to return an error. This is called after
500 * all program headers to be checked by arch_elf_pt_proc have been.
501 *
502 * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
503 * with that return code.
504 */
505 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
506 struct elfhdr *interp_ehdr,
507 struct arch_elf_state *state)
508 {
509 /* Dummy implementation, always proceed */
510 return 0;
511 }
512
513 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
514
515 /* This is much more generalized than the library routine read function,
516 so we keep this separate. Technically the library read function
517 is only provided so that we can read a.out libraries that have
518 an ELF header */
519
520 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
521 struct file *interpreter, unsigned long *interp_map_addr,
522 unsigned long no_base, struct elf_phdr *interp_elf_phdata)
523 {
524 struct elf_phdr *eppnt;
525 unsigned long load_addr = 0;
526 int load_addr_set = 0;
527 unsigned long last_bss = 0, elf_bss = 0;
528 unsigned long error = ~0UL;
529 unsigned long total_size;
530 int i;
531
532 /* First of all, some simple consistency checks */
533 if (interp_elf_ex->e_type != ET_EXEC &&
534 interp_elf_ex->e_type != ET_DYN)
535 goto out;
536 if (!elf_check_arch(interp_elf_ex))
537 goto out;
538 if (!interpreter->f_op->mmap)
539 goto out;
540
541 total_size = total_mapping_size(interp_elf_phdata,
542 interp_elf_ex->e_phnum);
543 if (!total_size) {
544 error = -EINVAL;
545 goto out;
546 }
547
548 eppnt = interp_elf_phdata;
549 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
550 if (eppnt->p_type == PT_LOAD) {
551 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
552 int elf_prot = 0;
553 unsigned long vaddr = 0;
554 unsigned long k, map_addr;
555
556 if (eppnt->p_flags & PF_R)
557 elf_prot = PROT_READ;
558 if (eppnt->p_flags & PF_W)
559 elf_prot |= PROT_WRITE;
560 if (eppnt->p_flags & PF_X)
561 elf_prot |= PROT_EXEC;
562 vaddr = eppnt->p_vaddr;
563 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
564 elf_type |= MAP_FIXED;
565 else if (no_base && interp_elf_ex->e_type == ET_DYN)
566 load_addr = -vaddr;
567
568 map_addr = elf_map(interpreter, load_addr + vaddr,
569 eppnt, elf_prot, elf_type, total_size);
570 total_size = 0;
571 if (!*interp_map_addr)
572 *interp_map_addr = map_addr;
573 error = map_addr;
574 if (BAD_ADDR(map_addr))
575 goto out;
576
577 if (!load_addr_set &&
578 interp_elf_ex->e_type == ET_DYN) {
579 load_addr = map_addr - ELF_PAGESTART(vaddr);
580 load_addr_set = 1;
581 }
582
583 /*
584 * Check to see if the section's size will overflow the
585 * allowed task size. Note that p_filesz must always be
586 * <= p_memsize so it's only necessary to check p_memsz.
587 */
588 k = load_addr + eppnt->p_vaddr;
589 if (BAD_ADDR(k) ||
590 eppnt->p_filesz > eppnt->p_memsz ||
591 eppnt->p_memsz > TASK_SIZE ||
592 TASK_SIZE - eppnt->p_memsz < k) {
593 error = -ENOMEM;
594 goto out;
595 }
596
597 /*
598 * Find the end of the file mapping for this phdr, and
599 * keep track of the largest address we see for this.
600 */
601 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
602 if (k > elf_bss)
603 elf_bss = k;
604
605 /*
606 * Do the same thing for the memory mapping - between
607 * elf_bss and last_bss is the bss section.
608 */
609 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
610 if (k > last_bss)
611 last_bss = k;
612 }
613 }
614
615 if (last_bss > elf_bss) {
616 /*
617 * Now fill out the bss section. First pad the last page up
618 * to the page boundary, and then perform a mmap to make sure
619 * that there are zero-mapped pages up to and including the
620 * last bss page.
621 */
622 if (padzero(elf_bss)) {
623 error = -EFAULT;
624 goto out;
625 }
626
627 /* What we have mapped so far */
628 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
629
630 /* Map the last of the bss segment */
631 error = vm_brk(elf_bss, last_bss - elf_bss);
632 if (BAD_ADDR(error))
633 goto out;
634 }
635
636 error = load_addr;
637 out:
638 return error;
639 }
640
641 /*
642 * These are the functions used to load ELF style executables and shared
643 * libraries. There is no binary dependent code anywhere else.
644 */
645
646 #ifndef STACK_RND_MASK
647 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
648 #endif
649
650 static unsigned long randomize_stack_top(unsigned long stack_top)
651 {
652 unsigned long random_variable = 0;
653
654 if ((current->flags & PF_RANDOMIZE) &&
655 !(current->personality & ADDR_NO_RANDOMIZE)) {
656 random_variable = get_random_long();
657 random_variable &= STACK_RND_MASK;
658 random_variable <<= PAGE_SHIFT;
659 }
660 #ifdef CONFIG_STACK_GROWSUP
661 return PAGE_ALIGN(stack_top) + random_variable;
662 #else
663 return PAGE_ALIGN(stack_top) - random_variable;
664 #endif
665 }
666
667 static int load_elf_binary(struct linux_binprm *bprm)
668 {
669 struct file *interpreter = NULL; /* to shut gcc up */
670 unsigned long load_addr = 0, load_bias = 0;
671 int load_addr_set = 0;
672 char * elf_interpreter = NULL;
673 unsigned long error;
674 struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
675 unsigned long elf_bss, elf_brk;
676 int retval, i;
677 unsigned long elf_entry;
678 unsigned long interp_load_addr = 0;
679 unsigned long start_code, end_code, start_data, end_data;
680 unsigned long reloc_func_desc __maybe_unused = 0;
681 int executable_stack = EXSTACK_DEFAULT;
682 struct pt_regs *regs = current_pt_regs();
683 struct {
684 struct elfhdr elf_ex;
685 struct elfhdr interp_elf_ex;
686 } *loc;
687 struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
688
689 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
690 if (!loc) {
691 retval = -ENOMEM;
692 goto out_ret;
693 }
694
695 /* Get the exec-header */
696 loc->elf_ex = *((struct elfhdr *)bprm->buf);
697
698 retval = -ENOEXEC;
699 /* First of all, some simple consistency checks */
700 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
701 goto out;
702
703 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
704 goto out;
705 if (!elf_check_arch(&loc->elf_ex))
706 goto out;
707 if (!bprm->file->f_op->mmap)
708 goto out;
709
710 elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
711 if (!elf_phdata)
712 goto out;
713
714 elf_ppnt = elf_phdata;
715 elf_bss = 0;
716 elf_brk = 0;
717
718 start_code = ~0UL;
719 end_code = 0;
720 start_data = 0;
721 end_data = 0;
722
723 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
724 if (elf_ppnt->p_type == PT_INTERP) {
725 /* This is the program interpreter used for
726 * shared libraries - for now assume that this
727 * is an a.out format binary
728 */
729 retval = -ENOEXEC;
730 if (elf_ppnt->p_filesz > PATH_MAX ||
731 elf_ppnt->p_filesz < 2)
732 goto out_free_ph;
733
734 retval = -ENOMEM;
735 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
736 GFP_KERNEL);
737 if (!elf_interpreter)
738 goto out_free_ph;
739
740 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
741 elf_interpreter,
742 elf_ppnt->p_filesz);
743 if (retval != elf_ppnt->p_filesz) {
744 if (retval >= 0)
745 retval = -EIO;
746 goto out_free_interp;
747 }
748 /* make sure path is NULL terminated */
749 retval = -ENOEXEC;
750 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
751 goto out_free_interp;
752
753 interpreter = open_exec(elf_interpreter);
754 retval = PTR_ERR(interpreter);
755 if (IS_ERR(interpreter))
756 goto out_free_interp;
757
758 /*
759 * If the binary is not readable then enforce
760 * mm->dumpable = 0 regardless of the interpreter's
761 * permissions.
762 */
763 would_dump(bprm, interpreter);
764
765 /* Get the exec headers */
766 retval = kernel_read(interpreter, 0,
767 (void *)&loc->interp_elf_ex,
768 sizeof(loc->interp_elf_ex));
769 if (retval != sizeof(loc->interp_elf_ex)) {
770 if (retval >= 0)
771 retval = -EIO;
772 goto out_free_dentry;
773 }
774
775 break;
776 }
777 elf_ppnt++;
778 }
779
780 elf_ppnt = elf_phdata;
781 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
782 switch (elf_ppnt->p_type) {
783 case PT_GNU_STACK:
784 if (elf_ppnt->p_flags & PF_X)
785 executable_stack = EXSTACK_ENABLE_X;
786 else
787 executable_stack = EXSTACK_DISABLE_X;
788 break;
789
790 case PT_LOPROC ... PT_HIPROC:
791 retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
792 bprm->file, false,
793 &arch_state);
794 if (retval)
795 goto out_free_dentry;
796 break;
797 }
798
799 /* Some simple consistency checks for the interpreter */
800 if (elf_interpreter) {
801 retval = -ELIBBAD;
802 /* Not an ELF interpreter */
803 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
804 goto out_free_dentry;
805 /* Verify the interpreter has a valid arch */
806 if (!elf_check_arch(&loc->interp_elf_ex))
807 goto out_free_dentry;
808
809 /* Load the interpreter program headers */
810 interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
811 interpreter);
812 if (!interp_elf_phdata)
813 goto out_free_dentry;
814
815 /* Pass PT_LOPROC..PT_HIPROC headers to arch code */
816 elf_ppnt = interp_elf_phdata;
817 for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
818 switch (elf_ppnt->p_type) {
819 case PT_LOPROC ... PT_HIPROC:
820 retval = arch_elf_pt_proc(&loc->interp_elf_ex,
821 elf_ppnt, interpreter,
822 true, &arch_state);
823 if (retval)
824 goto out_free_dentry;
825 break;
826 }
827 }
828
829 /*
830 * Allow arch code to reject the ELF at this point, whilst it's
831 * still possible to return an error to the code that invoked
832 * the exec syscall.
833 */
834 retval = arch_check_elf(&loc->elf_ex,
835 !!interpreter, &loc->interp_elf_ex,
836 &arch_state);
837 if (retval)
838 goto out_free_dentry;
839
840 /* Flush all traces of the currently running executable */
841 retval = flush_old_exec(bprm);
842 if (retval)
843 goto out_free_dentry;
844
845 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
846 may depend on the personality. */
847 SET_PERSONALITY2(loc->elf_ex, &arch_state);
848 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
849 current->personality |= READ_IMPLIES_EXEC;
850
851 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
852 current->flags |= PF_RANDOMIZE;
853
854 setup_new_exec(bprm);
855
856 /* Do this so that we can load the interpreter, if need be. We will
857 change some of these later */
858 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
859 executable_stack);
860 if (retval < 0)
861 goto out_free_dentry;
862
863 current->mm->start_stack = bprm->p;
864
865 /* Now we do a little grungy work by mmapping the ELF image into
866 the correct location in memory. */
867 for(i = 0, elf_ppnt = elf_phdata;
868 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
869 int elf_prot = 0, elf_flags;
870 unsigned long k, vaddr;
871 unsigned long total_size = 0;
872
873 if (elf_ppnt->p_type != PT_LOAD)
874 continue;
875
876 if (unlikely (elf_brk > elf_bss)) {
877 unsigned long nbyte;
878
879 /* There was a PT_LOAD segment with p_memsz > p_filesz
880 before this one. Map anonymous pages, if needed,
881 and clear the area. */
882 retval = set_brk(elf_bss + load_bias,
883 elf_brk + load_bias);
884 if (retval)
885 goto out_free_dentry;
886 nbyte = ELF_PAGEOFFSET(elf_bss);
887 if (nbyte) {
888 nbyte = ELF_MIN_ALIGN - nbyte;
889 if (nbyte > elf_brk - elf_bss)
890 nbyte = elf_brk - elf_bss;
891 if (clear_user((void __user *)elf_bss +
892 load_bias, nbyte)) {
893 /*
894 * This bss-zeroing can fail if the ELF
895 * file specifies odd protections. So
896 * we don't check the return value
897 */
898 }
899 }
900 }
901
902 if (elf_ppnt->p_flags & PF_R)
903 elf_prot |= PROT_READ;
904 if (elf_ppnt->p_flags & PF_W)
905 elf_prot |= PROT_WRITE;
906 if (elf_ppnt->p_flags & PF_X)
907 elf_prot |= PROT_EXEC;
908
909 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
910
911 vaddr = elf_ppnt->p_vaddr;
912 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
913 elf_flags |= MAP_FIXED;
914 } else if (loc->elf_ex.e_type == ET_DYN) {
915 /* Try and get dynamic programs out of the way of the
916 * default mmap base, as well as whatever program they
917 * might try to exec. This is because the brk will
918 * follow the loader, and is not movable. */
919 load_bias = ELF_ET_DYN_BASE - vaddr;
920 if (current->flags & PF_RANDOMIZE)
921 load_bias += arch_mmap_rnd();
922 load_bias = ELF_PAGESTART(load_bias);
923 total_size = total_mapping_size(elf_phdata,
924 loc->elf_ex.e_phnum);
925 if (!total_size) {
926 retval = -EINVAL;
927 goto out_free_dentry;
928 }
929 }
930
931 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
932 elf_prot, elf_flags, total_size);
933 if (BAD_ADDR(error)) {
934 retval = IS_ERR((void *)error) ?
935 PTR_ERR((void*)error) : -EINVAL;
936 goto out_free_dentry;
937 }
938
939 if (!load_addr_set) {
940 load_addr_set = 1;
941 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
942 if (loc->elf_ex.e_type == ET_DYN) {
943 load_bias += error -
944 ELF_PAGESTART(load_bias + vaddr);
945 load_addr += load_bias;
946 reloc_func_desc = load_bias;
947 }
948 }
949 k = elf_ppnt->p_vaddr;
950 if (k < start_code)
951 start_code = k;
952 if (start_data < k)
953 start_data = k;
954
955 /*
956 * Check to see if the section's size will overflow the
957 * allowed task size. Note that p_filesz must always be
958 * <= p_memsz so it is only necessary to check p_memsz.
959 */
960 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
961 elf_ppnt->p_memsz > TASK_SIZE ||
962 TASK_SIZE - elf_ppnt->p_memsz < k) {
963 /* set_brk can never work. Avoid overflows. */
964 retval = -EINVAL;
965 goto out_free_dentry;
966 }
967
968 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
969
970 if (k > elf_bss)
971 elf_bss = k;
972 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
973 end_code = k;
974 if (end_data < k)
975 end_data = k;
976 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
977 if (k > elf_brk)
978 elf_brk = k;
979 }
980
981 loc->elf_ex.e_entry += load_bias;
982 elf_bss += load_bias;
983 elf_brk += load_bias;
984 start_code += load_bias;
985 end_code += load_bias;
986 start_data += load_bias;
987 end_data += load_bias;
988
989 /* Calling set_brk effectively mmaps the pages that we need
990 * for the bss and break sections. We must do this before
991 * mapping in the interpreter, to make sure it doesn't wind
992 * up getting placed where the bss needs to go.
993 */
994 retval = set_brk(elf_bss, elf_brk);
995 if (retval)
996 goto out_free_dentry;
997 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
998 retval = -EFAULT; /* Nobody gets to see this, but.. */
999 goto out_free_dentry;
1000 }
1001
1002 if (elf_interpreter) {
1003 unsigned long interp_map_addr = 0;
1004
1005 elf_entry = load_elf_interp(&loc->interp_elf_ex,
1006 interpreter,
1007 &interp_map_addr,
1008 load_bias, interp_elf_phdata);
1009 if (!IS_ERR((void *)elf_entry)) {
1010 /*
1011 * load_elf_interp() returns relocation
1012 * adjustment
1013 */
1014 interp_load_addr = elf_entry;
1015 elf_entry += loc->interp_elf_ex.e_entry;
1016 }
1017 if (BAD_ADDR(elf_entry)) {
1018 retval = IS_ERR((void *)elf_entry) ?
1019 (int)elf_entry : -EINVAL;
1020 goto out_free_dentry;
1021 }
1022 reloc_func_desc = interp_load_addr;
1023
1024 allow_write_access(interpreter);
1025 fput(interpreter);
1026 kfree(elf_interpreter);
1027 } else {
1028 elf_entry = loc->elf_ex.e_entry;
1029 if (BAD_ADDR(elf_entry)) {
1030 retval = -EINVAL;
1031 goto out_free_dentry;
1032 }
1033 }
1034
1035 kfree(interp_elf_phdata);
1036 kfree(elf_phdata);
1037
1038 set_binfmt(&elf_format);
1039
1040 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1041 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1042 if (retval < 0)
1043 goto out;
1044 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1045
1046 install_exec_creds(bprm);
1047 retval = create_elf_tables(bprm, &loc->elf_ex,
1048 load_addr, interp_load_addr);
1049 if (retval < 0)
1050 goto out;
1051 /* N.B. passed_fileno might not be initialized? */
1052 current->mm->end_code = end_code;
1053 current->mm->start_code = start_code;
1054 current->mm->start_data = start_data;
1055 current->mm->end_data = end_data;
1056 current->mm->start_stack = bprm->p;
1057
1058 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1059 current->mm->brk = current->mm->start_brk =
1060 arch_randomize_brk(current->mm);
1061 #ifdef compat_brk_randomized
1062 current->brk_randomized = 1;
1063 #endif
1064 }
1065
1066 if (current->personality & MMAP_PAGE_ZERO) {
1067 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1068 and some applications "depend" upon this behavior.
1069 Since we do not have the power to recompile these, we
1070 emulate the SVr4 behavior. Sigh. */
1071 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1072 MAP_FIXED | MAP_PRIVATE, 0);
1073 }
1074
1075 #ifdef ELF_PLAT_INIT
1076 /*
1077 * The ABI may specify that certain registers be set up in special
1078 * ways (on i386 %edx is the address of a DT_FINI function, for
1079 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1080 * that the e_entry field is the address of the function descriptor
1081 * for the startup routine, rather than the address of the startup
1082 * routine itself. This macro performs whatever initialization to
1083 * the regs structure is required as well as any relocations to the
1084 * function descriptor entries when executing dynamically links apps.
1085 */
1086 ELF_PLAT_INIT(regs, reloc_func_desc);
1087 #endif
1088
1089 start_thread(regs, elf_entry, bprm->p);
1090 retval = 0;
1091 out:
1092 kfree(loc);
1093 out_ret:
1094 return retval;
1095
1096 /* error cleanup */
1097 out_free_dentry:
1098 kfree(interp_elf_phdata);
1099 allow_write_access(interpreter);
1100 if (interpreter)
1101 fput(interpreter);
1102 out_free_interp:
1103 kfree(elf_interpreter);
1104 out_free_ph:
1105 kfree(elf_phdata);
1106 goto out;
1107 }
1108
1109 #ifdef CONFIG_USELIB
1110 /* This is really simpleminded and specialized - we are loading an
1111 a.out library that is given an ELF header. */
1112 static int load_elf_library(struct file *file)
1113 {
1114 struct elf_phdr *elf_phdata;
1115 struct elf_phdr *eppnt;
1116 unsigned long elf_bss, bss, len;
1117 int retval, error, i, j;
1118 struct elfhdr elf_ex;
1119
1120 error = -ENOEXEC;
1121 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1122 if (retval != sizeof(elf_ex))
1123 goto out;
1124
1125 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1126 goto out;
1127
1128 /* First of all, some simple consistency checks */
1129 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1130 !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1131 goto out;
1132
1133 /* Now read in all of the header information */
1134
1135 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1136 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1137
1138 error = -ENOMEM;
1139 elf_phdata = kmalloc(j, GFP_KERNEL);
1140 if (!elf_phdata)
1141 goto out;
1142
1143 eppnt = elf_phdata;
1144 error = -ENOEXEC;
1145 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1146 if (retval != j)
1147 goto out_free_ph;
1148
1149 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1150 if ((eppnt + i)->p_type == PT_LOAD)
1151 j++;
1152 if (j != 1)
1153 goto out_free_ph;
1154
1155 while (eppnt->p_type != PT_LOAD)
1156 eppnt++;
1157
1158 /* Now use mmap to map the library into memory. */
1159 error = vm_mmap(file,
1160 ELF_PAGESTART(eppnt->p_vaddr),
1161 (eppnt->p_filesz +
1162 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1163 PROT_READ | PROT_WRITE | PROT_EXEC,
1164 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1165 (eppnt->p_offset -
1166 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1167 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1168 goto out_free_ph;
1169
1170 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1171 if (padzero(elf_bss)) {
1172 error = -EFAULT;
1173 goto out_free_ph;
1174 }
1175
1176 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1177 ELF_MIN_ALIGN - 1);
1178 bss = eppnt->p_memsz + eppnt->p_vaddr;
1179 if (bss > len) {
1180 error = vm_brk(len, bss - len);
1181 if (BAD_ADDR(error))
1182 goto out_free_ph;
1183 }
1184 error = 0;
1185
1186 out_free_ph:
1187 kfree(elf_phdata);
1188 out:
1189 return error;
1190 }
1191 #endif /* #ifdef CONFIG_USELIB */
1192
1193 #ifdef CONFIG_ELF_CORE
1194 /*
1195 * ELF core dumper
1196 *
1197 * Modelled on fs/exec.c:aout_core_dump()
1198 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1199 */
1200
1201 /*
1202 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1203 * that are useful for post-mortem analysis are included in every core dump.
1204 * In that way we ensure that the core dump is fully interpretable later
1205 * without matching up the same kernel and hardware config to see what PC values
1206 * meant. These special mappings include - vDSO, vsyscall, and other
1207 * architecture specific mappings
1208 */
1209 static bool always_dump_vma(struct vm_area_struct *vma)
1210 {
1211 /* Any vsyscall mappings? */
1212 if (vma == get_gate_vma(vma->vm_mm))
1213 return true;
1214
1215 /*
1216 * Assume that all vmas with a .name op should always be dumped.
1217 * If this changes, a new vm_ops field can easily be added.
1218 */
1219 if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1220 return true;
1221
1222 /*
1223 * arch_vma_name() returns non-NULL for special architecture mappings,
1224 * such as vDSO sections.
1225 */
1226 if (arch_vma_name(vma))
1227 return true;
1228
1229 return false;
1230 }
1231
1232 /*
1233 * Decide what to dump of a segment, part, all or none.
1234 */
1235 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1236 unsigned long mm_flags)
1237 {
1238 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1239
1240 /* always dump the vdso and vsyscall sections */
1241 if (always_dump_vma(vma))
1242 goto whole;
1243
1244 if (vma->vm_flags & VM_DONTDUMP)
1245 return 0;
1246
1247 /* support for DAX */
1248 if (vma_is_dax(vma)) {
1249 if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1250 goto whole;
1251 if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1252 goto whole;
1253 return 0;
1254 }
1255
1256 /* Hugetlb memory check */
1257 if (vma->vm_flags & VM_HUGETLB) {
1258 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1259 goto whole;
1260 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1261 goto whole;
1262 return 0;
1263 }
1264
1265 /* Do not dump I/O mapped devices or special mappings */
1266 if (vma->vm_flags & VM_IO)
1267 return 0;
1268
1269 /* By default, dump shared memory if mapped from an anonymous file. */
1270 if (vma->vm_flags & VM_SHARED) {
1271 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1272 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1273 goto whole;
1274 return 0;
1275 }
1276
1277 /* Dump segments that have been written to. */
1278 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1279 goto whole;
1280 if (vma->vm_file == NULL)
1281 return 0;
1282
1283 if (FILTER(MAPPED_PRIVATE))
1284 goto whole;
1285
1286 /*
1287 * If this looks like the beginning of a DSO or executable mapping,
1288 * check for an ELF header. If we find one, dump the first page to
1289 * aid in determining what was mapped here.
1290 */
1291 if (FILTER(ELF_HEADERS) &&
1292 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1293 u32 __user *header = (u32 __user *) vma->vm_start;
1294 u32 word;
1295 mm_segment_t fs = get_fs();
1296 /*
1297 * Doing it this way gets the constant folded by GCC.
1298 */
1299 union {
1300 u32 cmp;
1301 char elfmag[SELFMAG];
1302 } magic;
1303 BUILD_BUG_ON(SELFMAG != sizeof word);
1304 magic.elfmag[EI_MAG0] = ELFMAG0;
1305 magic.elfmag[EI_MAG1] = ELFMAG1;
1306 magic.elfmag[EI_MAG2] = ELFMAG2;
1307 magic.elfmag[EI_MAG3] = ELFMAG3;
1308 /*
1309 * Switch to the user "segment" for get_user(),
1310 * then put back what elf_core_dump() had in place.
1311 */
1312 set_fs(USER_DS);
1313 if (unlikely(get_user(word, header)))
1314 word = 0;
1315 set_fs(fs);
1316 if (word == magic.cmp)
1317 return PAGE_SIZE;
1318 }
1319
1320 #undef FILTER
1321
1322 return 0;
1323
1324 whole:
1325 return vma->vm_end - vma->vm_start;
1326 }
1327
1328 /* An ELF note in memory */
1329 struct memelfnote
1330 {
1331 const char *name;
1332 int type;
1333 unsigned int datasz;
1334 void *data;
1335 };
1336
1337 static int notesize(struct memelfnote *en)
1338 {
1339 int sz;
1340
1341 sz = sizeof(struct elf_note);
1342 sz += roundup(strlen(en->name) + 1, 4);
1343 sz += roundup(en->datasz, 4);
1344
1345 return sz;
1346 }
1347
1348 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1349 {
1350 struct elf_note en;
1351 en.n_namesz = strlen(men->name) + 1;
1352 en.n_descsz = men->datasz;
1353 en.n_type = men->type;
1354
1355 return dump_emit(cprm, &en, sizeof(en)) &&
1356 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1357 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1358 }
1359
1360 static void fill_elf_header(struct elfhdr *elf, int segs,
1361 u16 machine, u32 flags)
1362 {
1363 memset(elf, 0, sizeof(*elf));
1364
1365 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1366 elf->e_ident[EI_CLASS] = ELF_CLASS;
1367 elf->e_ident[EI_DATA] = ELF_DATA;
1368 elf->e_ident[EI_VERSION] = EV_CURRENT;
1369 elf->e_ident[EI_OSABI] = ELF_OSABI;
1370
1371 elf->e_type = ET_CORE;
1372 elf->e_machine = machine;
1373 elf->e_version = EV_CURRENT;
1374 elf->e_phoff = sizeof(struct elfhdr);
1375 elf->e_flags = flags;
1376 elf->e_ehsize = sizeof(struct elfhdr);
1377 elf->e_phentsize = sizeof(struct elf_phdr);
1378 elf->e_phnum = segs;
1379
1380 return;
1381 }
1382
1383 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1384 {
1385 phdr->p_type = PT_NOTE;
1386 phdr->p_offset = offset;
1387 phdr->p_vaddr = 0;
1388 phdr->p_paddr = 0;
1389 phdr->p_filesz = sz;
1390 phdr->p_memsz = 0;
1391 phdr->p_flags = 0;
1392 phdr->p_align = 0;
1393 return;
1394 }
1395
1396 static void fill_note(struct memelfnote *note, const char *name, int type,
1397 unsigned int sz, void *data)
1398 {
1399 note->name = name;
1400 note->type = type;
1401 note->datasz = sz;
1402 note->data = data;
1403 return;
1404 }
1405
1406 /*
1407 * fill up all the fields in prstatus from the given task struct, except
1408 * registers which need to be filled up separately.
1409 */
1410 static void fill_prstatus(struct elf_prstatus *prstatus,
1411 struct task_struct *p, long signr)
1412 {
1413 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1414 prstatus->pr_sigpend = p->pending.signal.sig[0];
1415 prstatus->pr_sighold = p->blocked.sig[0];
1416 rcu_read_lock();
1417 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1418 rcu_read_unlock();
1419 prstatus->pr_pid = task_pid_vnr(p);
1420 prstatus->pr_pgrp = task_pgrp_vnr(p);
1421 prstatus->pr_sid = task_session_vnr(p);
1422 if (thread_group_leader(p)) {
1423 struct task_cputime cputime;
1424
1425 /*
1426 * This is the record for the group leader. It shows the
1427 * group-wide total, not its individual thread total.
1428 */
1429 thread_group_cputime(p, &cputime);
1430 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1431 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1432 } else {
1433 cputime_t utime, stime;
1434
1435 task_cputime(p, &utime, &stime);
1436 cputime_to_timeval(utime, &prstatus->pr_utime);
1437 cputime_to_timeval(stime, &prstatus->pr_stime);
1438 }
1439 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1440 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1441 }
1442
1443 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1444 struct mm_struct *mm)
1445 {
1446 const struct cred *cred;
1447 unsigned int i, len;
1448
1449 /* first copy the parameters from user space */
1450 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1451
1452 len = mm->arg_end - mm->arg_start;
1453 if (len >= ELF_PRARGSZ)
1454 len = ELF_PRARGSZ-1;
1455 if (copy_from_user(&psinfo->pr_psargs,
1456 (const char __user *)mm->arg_start, len))
1457 return -EFAULT;
1458 for(i = 0; i < len; i++)
1459 if (psinfo->pr_psargs[i] == 0)
1460 psinfo->pr_psargs[i] = ' ';
1461 psinfo->pr_psargs[len] = 0;
1462
1463 rcu_read_lock();
1464 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1465 rcu_read_unlock();
1466 psinfo->pr_pid = task_pid_vnr(p);
1467 psinfo->pr_pgrp = task_pgrp_vnr(p);
1468 psinfo->pr_sid = task_session_vnr(p);
1469
1470 i = p->state ? ffz(~p->state) + 1 : 0;
1471 psinfo->pr_state = i;
1472 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1473 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1474 psinfo->pr_nice = task_nice(p);
1475 psinfo->pr_flag = p->flags;
1476 rcu_read_lock();
1477 cred = __task_cred(p);
1478 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1479 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1480 rcu_read_unlock();
1481 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1482
1483 return 0;
1484 }
1485
1486 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1487 {
1488 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1489 int i = 0;
1490 do
1491 i += 2;
1492 while (auxv[i - 2] != AT_NULL);
1493 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1494 }
1495
1496 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1497 const siginfo_t *siginfo)
1498 {
1499 mm_segment_t old_fs = get_fs();
1500 set_fs(KERNEL_DS);
1501 copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1502 set_fs(old_fs);
1503 fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1504 }
1505
1506 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1507 /*
1508 * Format of NT_FILE note:
1509 *
1510 * long count -- how many files are mapped
1511 * long page_size -- units for file_ofs
1512 * array of [COUNT] elements of
1513 * long start
1514 * long end
1515 * long file_ofs
1516 * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1517 */
1518 static int fill_files_note(struct memelfnote *note)
1519 {
1520 struct vm_area_struct *vma;
1521 unsigned count, size, names_ofs, remaining, n;
1522 user_long_t *data;
1523 user_long_t *start_end_ofs;
1524 char *name_base, *name_curpos;
1525
1526 /* *Estimated* file count and total data size needed */
1527 count = current->mm->map_count;
1528 size = count * 64;
1529
1530 names_ofs = (2 + 3 * count) * sizeof(data[0]);
1531 alloc:
1532 if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1533 return -EINVAL;
1534 size = round_up(size, PAGE_SIZE);
1535 data = vmalloc(size);
1536 if (!data)
1537 return -ENOMEM;
1538
1539 start_end_ofs = data + 2;
1540 name_base = name_curpos = ((char *)data) + names_ofs;
1541 remaining = size - names_ofs;
1542 count = 0;
1543 for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1544 struct file *file;
1545 const char *filename;
1546
1547 file = vma->vm_file;
1548 if (!file)
1549 continue;
1550 filename = file_path(file, name_curpos, remaining);
1551 if (IS_ERR(filename)) {
1552 if (PTR_ERR(filename) == -ENAMETOOLONG) {
1553 vfree(data);
1554 size = size * 5 / 4;
1555 goto alloc;
1556 }
1557 continue;
1558 }
1559
1560 /* file_path() fills at the end, move name down */
1561 /* n = strlen(filename) + 1: */
1562 n = (name_curpos + remaining) - filename;
1563 remaining = filename - name_curpos;
1564 memmove(name_curpos, filename, n);
1565 name_curpos += n;
1566
1567 *start_end_ofs++ = vma->vm_start;
1568 *start_end_ofs++ = vma->vm_end;
1569 *start_end_ofs++ = vma->vm_pgoff;
1570 count++;
1571 }
1572
1573 /* Now we know exact count of files, can store it */
1574 data[0] = count;
1575 data[1] = PAGE_SIZE;
1576 /*
1577 * Count usually is less than current->mm->map_count,
1578 * we need to move filenames down.
1579 */
1580 n = current->mm->map_count - count;
1581 if (n != 0) {
1582 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1583 memmove(name_base - shift_bytes, name_base,
1584 name_curpos - name_base);
1585 name_curpos -= shift_bytes;
1586 }
1587
1588 size = name_curpos - (char *)data;
1589 fill_note(note, "CORE", NT_FILE, size, data);
1590 return 0;
1591 }
1592
1593 #ifdef CORE_DUMP_USE_REGSET
1594 #include <linux/regset.h>
1595
1596 struct elf_thread_core_info {
1597 struct elf_thread_core_info *next;
1598 struct task_struct *task;
1599 struct elf_prstatus prstatus;
1600 struct memelfnote notes[0];
1601 };
1602
1603 struct elf_note_info {
1604 struct elf_thread_core_info *thread;
1605 struct memelfnote psinfo;
1606 struct memelfnote signote;
1607 struct memelfnote auxv;
1608 struct memelfnote files;
1609 user_siginfo_t csigdata;
1610 size_t size;
1611 int thread_notes;
1612 };
1613
1614 /*
1615 * When a regset has a writeback hook, we call it on each thread before
1616 * dumping user memory. On register window machines, this makes sure the
1617 * user memory backing the register data is up to date before we read it.
1618 */
1619 static void do_thread_regset_writeback(struct task_struct *task,
1620 const struct user_regset *regset)
1621 {
1622 if (regset->writeback)
1623 regset->writeback(task, regset, 1);
1624 }
1625
1626 #ifndef PR_REG_SIZE
1627 #define PR_REG_SIZE(S) sizeof(S)
1628 #endif
1629
1630 #ifndef PRSTATUS_SIZE
1631 #define PRSTATUS_SIZE(S) sizeof(S)
1632 #endif
1633
1634 #ifndef PR_REG_PTR
1635 #define PR_REG_PTR(S) (&((S)->pr_reg))
1636 #endif
1637
1638 #ifndef SET_PR_FPVALID
1639 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1640 #endif
1641
1642 static int fill_thread_core_info(struct elf_thread_core_info *t,
1643 const struct user_regset_view *view,
1644 long signr, size_t *total)
1645 {
1646 unsigned int i;
1647
1648 /*
1649 * NT_PRSTATUS is the one special case, because the regset data
1650 * goes into the pr_reg field inside the note contents, rather
1651 * than being the whole note contents. We fill the reset in here.
1652 * We assume that regset 0 is NT_PRSTATUS.
1653 */
1654 fill_prstatus(&t->prstatus, t->task, signr);
1655 (void) view->regsets[0].get(t->task, &view->regsets[0],
1656 0, PR_REG_SIZE(t->prstatus.pr_reg),
1657 PR_REG_PTR(&t->prstatus), NULL);
1658
1659 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1660 PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1661 *total += notesize(&t->notes[0]);
1662
1663 do_thread_regset_writeback(t->task, &view->regsets[0]);
1664
1665 /*
1666 * Each other regset might generate a note too. For each regset
1667 * that has no core_note_type or is inactive, we leave t->notes[i]
1668 * all zero and we'll know to skip writing it later.
1669 */
1670 for (i = 1; i < view->n; ++i) {
1671 const struct user_regset *regset = &view->regsets[i];
1672 do_thread_regset_writeback(t->task, regset);
1673 if (regset->core_note_type && regset->get &&
1674 (!regset->active || regset->active(t->task, regset))) {
1675 int ret;
1676 size_t size = regset->n * regset->size;
1677 void *data = kmalloc(size, GFP_KERNEL);
1678 if (unlikely(!data))
1679 return 0;
1680 ret = regset->get(t->task, regset,
1681 0, size, data, NULL);
1682 if (unlikely(ret))
1683 kfree(data);
1684 else {
1685 if (regset->core_note_type != NT_PRFPREG)
1686 fill_note(&t->notes[i], "LINUX",
1687 regset->core_note_type,
1688 size, data);
1689 else {
1690 SET_PR_FPVALID(&t->prstatus, 1);
1691 fill_note(&t->notes[i], "CORE",
1692 NT_PRFPREG, size, data);
1693 }
1694 *total += notesize(&t->notes[i]);
1695 }
1696 }
1697 }
1698
1699 return 1;
1700 }
1701
1702 static int fill_note_info(struct elfhdr *elf, int phdrs,
1703 struct elf_note_info *info,
1704 const siginfo_t *siginfo, struct pt_regs *regs)
1705 {
1706 struct task_struct *dump_task = current;
1707 const struct user_regset_view *view = task_user_regset_view(dump_task);
1708 struct elf_thread_core_info *t;
1709 struct elf_prpsinfo *psinfo;
1710 struct core_thread *ct;
1711 unsigned int i;
1712
1713 info->size = 0;
1714 info->thread = NULL;
1715
1716 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1717 if (psinfo == NULL) {
1718 info->psinfo.data = NULL; /* So we don't free this wrongly */
1719 return 0;
1720 }
1721
1722 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1723
1724 /*
1725 * Figure out how many notes we're going to need for each thread.
1726 */
1727 info->thread_notes = 0;
1728 for (i = 0; i < view->n; ++i)
1729 if (view->regsets[i].core_note_type != 0)
1730 ++info->thread_notes;
1731
1732 /*
1733 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1734 * since it is our one special case.
1735 */
1736 if (unlikely(info->thread_notes == 0) ||
1737 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1738 WARN_ON(1);
1739 return 0;
1740 }
1741
1742 /*
1743 * Initialize the ELF file header.
1744 */
1745 fill_elf_header(elf, phdrs,
1746 view->e_machine, view->e_flags);
1747
1748 /*
1749 * Allocate a structure for each thread.
1750 */
1751 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1752 t = kzalloc(offsetof(struct elf_thread_core_info,
1753 notes[info->thread_notes]),
1754 GFP_KERNEL);
1755 if (unlikely(!t))
1756 return 0;
1757
1758 t->task = ct->task;
1759 if (ct->task == dump_task || !info->thread) {
1760 t->next = info->thread;
1761 info->thread = t;
1762 } else {
1763 /*
1764 * Make sure to keep the original task at
1765 * the head of the list.
1766 */
1767 t->next = info->thread->next;
1768 info->thread->next = t;
1769 }
1770 }
1771
1772 /*
1773 * Now fill in each thread's information.
1774 */
1775 for (t = info->thread; t != NULL; t = t->next)
1776 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1777 return 0;
1778
1779 /*
1780 * Fill in the two process-wide notes.
1781 */
1782 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1783 info->size += notesize(&info->psinfo);
1784
1785 fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1786 info->size += notesize(&info->signote);
1787
1788 fill_auxv_note(&info->auxv, current->mm);
1789 info->size += notesize(&info->auxv);
1790
1791 if (fill_files_note(&info->files) == 0)
1792 info->size += notesize(&info->files);
1793
1794 return 1;
1795 }
1796
1797 static size_t get_note_info_size(struct elf_note_info *info)
1798 {
1799 return info->size;
1800 }
1801
1802 /*
1803 * Write all the notes for each thread. When writing the first thread, the
1804 * process-wide notes are interleaved after the first thread-specific note.
1805 */
1806 static int write_note_info(struct elf_note_info *info,
1807 struct coredump_params *cprm)
1808 {
1809 bool first = true;
1810 struct elf_thread_core_info *t = info->thread;
1811
1812 do {
1813 int i;
1814
1815 if (!writenote(&t->notes[0], cprm))
1816 return 0;
1817
1818 if (first && !writenote(&info->psinfo, cprm))
1819 return 0;
1820 if (first && !writenote(&info->signote, cprm))
1821 return 0;
1822 if (first && !writenote(&info->auxv, cprm))
1823 return 0;
1824 if (first && info->files.data &&
1825 !writenote(&info->files, cprm))
1826 return 0;
1827
1828 for (i = 1; i < info->thread_notes; ++i)
1829 if (t->notes[i].data &&
1830 !writenote(&t->notes[i], cprm))
1831 return 0;
1832
1833 first = false;
1834 t = t->next;
1835 } while (t);
1836
1837 return 1;
1838 }
1839
1840 static void free_note_info(struct elf_note_info *info)
1841 {
1842 struct elf_thread_core_info *threads = info->thread;
1843 while (threads) {
1844 unsigned int i;
1845 struct elf_thread_core_info *t = threads;
1846 threads = t->next;
1847 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1848 for (i = 1; i < info->thread_notes; ++i)
1849 kfree(t->notes[i].data);
1850 kfree(t);
1851 }
1852 kfree(info->psinfo.data);
1853 vfree(info->files.data);
1854 }
1855
1856 #else
1857
1858 /* Here is the structure in which status of each thread is captured. */
1859 struct elf_thread_status
1860 {
1861 struct list_head list;
1862 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1863 elf_fpregset_t fpu; /* NT_PRFPREG */
1864 struct task_struct *thread;
1865 #ifdef ELF_CORE_COPY_XFPREGS
1866 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1867 #endif
1868 struct memelfnote notes[3];
1869 int num_notes;
1870 };
1871
1872 /*
1873 * In order to add the specific thread information for the elf file format,
1874 * we need to keep a linked list of every threads pr_status and then create
1875 * a single section for them in the final core file.
1876 */
1877 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1878 {
1879 int sz = 0;
1880 struct task_struct *p = t->thread;
1881 t->num_notes = 0;
1882
1883 fill_prstatus(&t->prstatus, p, signr);
1884 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1885
1886 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1887 &(t->prstatus));
1888 t->num_notes++;
1889 sz += notesize(&t->notes[0]);
1890
1891 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1892 &t->fpu))) {
1893 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1894 &(t->fpu));
1895 t->num_notes++;
1896 sz += notesize(&t->notes[1]);
1897 }
1898
1899 #ifdef ELF_CORE_COPY_XFPREGS
1900 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1901 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1902 sizeof(t->xfpu), &t->xfpu);
1903 t->num_notes++;
1904 sz += notesize(&t->notes[2]);
1905 }
1906 #endif
1907 return sz;
1908 }
1909
1910 struct elf_note_info {
1911 struct memelfnote *notes;
1912 struct memelfnote *notes_files;
1913 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1914 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1915 struct list_head thread_list;
1916 elf_fpregset_t *fpu;
1917 #ifdef ELF_CORE_COPY_XFPREGS
1918 elf_fpxregset_t *xfpu;
1919 #endif
1920 user_siginfo_t csigdata;
1921 int thread_status_size;
1922 int numnote;
1923 };
1924
1925 static int elf_note_info_init(struct elf_note_info *info)
1926 {
1927 memset(info, 0, sizeof(*info));
1928 INIT_LIST_HEAD(&info->thread_list);
1929
1930 /* Allocate space for ELF notes */
1931 info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1932 if (!info->notes)
1933 return 0;
1934 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1935 if (!info->psinfo)
1936 return 0;
1937 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1938 if (!info->prstatus)
1939 return 0;
1940 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1941 if (!info->fpu)
1942 return 0;
1943 #ifdef ELF_CORE_COPY_XFPREGS
1944 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1945 if (!info->xfpu)
1946 return 0;
1947 #endif
1948 return 1;
1949 }
1950
1951 static int fill_note_info(struct elfhdr *elf, int phdrs,
1952 struct elf_note_info *info,
1953 const siginfo_t *siginfo, struct pt_regs *regs)
1954 {
1955 struct list_head *t;
1956 struct core_thread *ct;
1957 struct elf_thread_status *ets;
1958
1959 if (!elf_note_info_init(info))
1960 return 0;
1961
1962 for (ct = current->mm->core_state->dumper.next;
1963 ct; ct = ct->next) {
1964 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1965 if (!ets)
1966 return 0;
1967
1968 ets->thread = ct->task;
1969 list_add(&ets->list, &info->thread_list);
1970 }
1971
1972 list_for_each(t, &info->thread_list) {
1973 int sz;
1974
1975 ets = list_entry(t, struct elf_thread_status, list);
1976 sz = elf_dump_thread_status(siginfo->si_signo, ets);
1977 info->thread_status_size += sz;
1978 }
1979 /* now collect the dump for the current */
1980 memset(info->prstatus, 0, sizeof(*info->prstatus));
1981 fill_prstatus(info->prstatus, current, siginfo->si_signo);
1982 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1983
1984 /* Set up header */
1985 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1986
1987 /*
1988 * Set up the notes in similar form to SVR4 core dumps made
1989 * with info from their /proc.
1990 */
1991
1992 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1993 sizeof(*info->prstatus), info->prstatus);
1994 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1995 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1996 sizeof(*info->psinfo), info->psinfo);
1997
1998 fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1999 fill_auxv_note(info->notes + 3, current->mm);
2000 info->numnote = 4;
2001
2002 if (fill_files_note(info->notes + info->numnote) == 0) {
2003 info->notes_files = info->notes + info->numnote;
2004 info->numnote++;
2005 }
2006
2007 /* Try to dump the FPU. */
2008 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2009 info->fpu);
2010 if (info->prstatus->pr_fpvalid)
2011 fill_note(info->notes + info->numnote++,
2012 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2013 #ifdef ELF_CORE_COPY_XFPREGS
2014 if (elf_core_copy_task_xfpregs(current, info->xfpu))
2015 fill_note(info->notes + info->numnote++,
2016 "LINUX", ELF_CORE_XFPREG_TYPE,
2017 sizeof(*info->xfpu), info->xfpu);
2018 #endif
2019
2020 return 1;
2021 }
2022
2023 static size_t get_note_info_size(struct elf_note_info *info)
2024 {
2025 int sz = 0;
2026 int i;
2027
2028 for (i = 0; i < info->numnote; i++)
2029 sz += notesize(info->notes + i);
2030
2031 sz += info->thread_status_size;
2032
2033 return sz;
2034 }
2035
2036 static int write_note_info(struct elf_note_info *info,
2037 struct coredump_params *cprm)
2038 {
2039 int i;
2040 struct list_head *t;
2041
2042 for (i = 0; i < info->numnote; i++)
2043 if (!writenote(info->notes + i, cprm))
2044 return 0;
2045
2046 /* write out the thread status notes section */
2047 list_for_each(t, &info->thread_list) {
2048 struct elf_thread_status *tmp =
2049 list_entry(t, struct elf_thread_status, list);
2050
2051 for (i = 0; i < tmp->num_notes; i++)
2052 if (!writenote(&tmp->notes[i], cprm))
2053 return 0;
2054 }
2055
2056 return 1;
2057 }
2058
2059 static void free_note_info(struct elf_note_info *info)
2060 {
2061 while (!list_empty(&info->thread_list)) {
2062 struct list_head *tmp = info->thread_list.next;
2063 list_del(tmp);
2064 kfree(list_entry(tmp, struct elf_thread_status, list));
2065 }
2066
2067 /* Free data possibly allocated by fill_files_note(): */
2068 if (info->notes_files)
2069 vfree(info->notes_files->data);
2070
2071 kfree(info->prstatus);
2072 kfree(info->psinfo);
2073 kfree(info->notes);
2074 kfree(info->fpu);
2075 #ifdef ELF_CORE_COPY_XFPREGS
2076 kfree(info->xfpu);
2077 #endif
2078 }
2079
2080 #endif
2081
2082 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2083 struct vm_area_struct *gate_vma)
2084 {
2085 struct vm_area_struct *ret = tsk->mm->mmap;
2086
2087 if (ret)
2088 return ret;
2089 return gate_vma;
2090 }
2091 /*
2092 * Helper function for iterating across a vma list. It ensures that the caller
2093 * will visit `gate_vma' prior to terminating the search.
2094 */
2095 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2096 struct vm_area_struct *gate_vma)
2097 {
2098 struct vm_area_struct *ret;
2099
2100 ret = this_vma->vm_next;
2101 if (ret)
2102 return ret;
2103 if (this_vma == gate_vma)
2104 return NULL;
2105 return gate_vma;
2106 }
2107
2108 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2109 elf_addr_t e_shoff, int segs)
2110 {
2111 elf->e_shoff = e_shoff;
2112 elf->e_shentsize = sizeof(*shdr4extnum);
2113 elf->e_shnum = 1;
2114 elf->e_shstrndx = SHN_UNDEF;
2115
2116 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2117
2118 shdr4extnum->sh_type = SHT_NULL;
2119 shdr4extnum->sh_size = elf->e_shnum;
2120 shdr4extnum->sh_link = elf->e_shstrndx;
2121 shdr4extnum->sh_info = segs;
2122 }
2123
2124 /*
2125 * Actual dumper
2126 *
2127 * This is a two-pass process; first we find the offsets of the bits,
2128 * and then they are actually written out. If we run out of core limit
2129 * we just truncate.
2130 */
2131 static int elf_core_dump(struct coredump_params *cprm)
2132 {
2133 int has_dumped = 0;
2134 mm_segment_t fs;
2135 int segs, i;
2136 size_t vma_data_size = 0;
2137 struct vm_area_struct *vma, *gate_vma;
2138 struct elfhdr *elf = NULL;
2139 loff_t offset = 0, dataoff;
2140 struct elf_note_info info = { };
2141 struct elf_phdr *phdr4note = NULL;
2142 struct elf_shdr *shdr4extnum = NULL;
2143 Elf_Half e_phnum;
2144 elf_addr_t e_shoff;
2145 elf_addr_t *vma_filesz = NULL;
2146
2147 /*
2148 * We no longer stop all VM operations.
2149 *
2150 * This is because those proceses that could possibly change map_count
2151 * or the mmap / vma pages are now blocked in do_exit on current
2152 * finishing this core dump.
2153 *
2154 * Only ptrace can touch these memory addresses, but it doesn't change
2155 * the map_count or the pages allocated. So no possibility of crashing
2156 * exists while dumping the mm->vm_next areas to the core file.
2157 */
2158
2159 /* alloc memory for large data structures: too large to be on stack */
2160 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2161 if (!elf)
2162 goto out;
2163 /*
2164 * The number of segs are recored into ELF header as 16bit value.
2165 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2166 */
2167 segs = current->mm->map_count;
2168 segs += elf_core_extra_phdrs();
2169
2170 gate_vma = get_gate_vma(current->mm);
2171 if (gate_vma != NULL)
2172 segs++;
2173
2174 /* for notes section */
2175 segs++;
2176
2177 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2178 * this, kernel supports extended numbering. Have a look at
2179 * include/linux/elf.h for further information. */
2180 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2181
2182 /*
2183 * Collect all the non-memory information about the process for the
2184 * notes. This also sets up the file header.
2185 */
2186 if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2187 goto cleanup;
2188
2189 has_dumped = 1;
2190
2191 fs = get_fs();
2192 set_fs(KERNEL_DS);
2193
2194 offset += sizeof(*elf); /* Elf header */
2195 offset += segs * sizeof(struct elf_phdr); /* Program headers */
2196
2197 /* Write notes phdr entry */
2198 {
2199 size_t sz = get_note_info_size(&info);
2200
2201 sz += elf_coredump_extra_notes_size();
2202
2203 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2204 if (!phdr4note)
2205 goto end_coredump;
2206
2207 fill_elf_note_phdr(phdr4note, sz, offset);
2208 offset += sz;
2209 }
2210
2211 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2212
2213 vma_filesz = kmalloc_array(segs - 1, sizeof(*vma_filesz), GFP_KERNEL);
2214 if (!vma_filesz)
2215 goto end_coredump;
2216
2217 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2218 vma = next_vma(vma, gate_vma)) {
2219 unsigned long dump_size;
2220
2221 dump_size = vma_dump_size(vma, cprm->mm_flags);
2222 vma_filesz[i++] = dump_size;
2223 vma_data_size += dump_size;
2224 }
2225
2226 offset += vma_data_size;
2227 offset += elf_core_extra_data_size();
2228 e_shoff = offset;
2229
2230 if (e_phnum == PN_XNUM) {
2231 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2232 if (!shdr4extnum)
2233 goto end_coredump;
2234 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2235 }
2236
2237 offset = dataoff;
2238
2239 if (!dump_emit(cprm, elf, sizeof(*elf)))
2240 goto end_coredump;
2241
2242 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2243 goto end_coredump;
2244
2245 /* Write program headers for segments dump */
2246 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2247 vma = next_vma(vma, gate_vma)) {
2248 struct elf_phdr phdr;
2249
2250 phdr.p_type = PT_LOAD;
2251 phdr.p_offset = offset;
2252 phdr.p_vaddr = vma->vm_start;
2253 phdr.p_paddr = 0;
2254 phdr.p_filesz = vma_filesz[i++];
2255 phdr.p_memsz = vma->vm_end - vma->vm_start;
2256 offset += phdr.p_filesz;
2257 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2258 if (vma->vm_flags & VM_WRITE)
2259 phdr.p_flags |= PF_W;
2260 if (vma->vm_flags & VM_EXEC)
2261 phdr.p_flags |= PF_X;
2262 phdr.p_align = ELF_EXEC_PAGESIZE;
2263
2264 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2265 goto end_coredump;
2266 }
2267
2268 if (!elf_core_write_extra_phdrs(cprm, offset))
2269 goto end_coredump;
2270
2271 /* write out the notes section */
2272 if (!write_note_info(&info, cprm))
2273 goto end_coredump;
2274
2275 if (elf_coredump_extra_notes_write(cprm))
2276 goto end_coredump;
2277
2278 /* Align to page */
2279 if (!dump_skip(cprm, dataoff - cprm->file->f_pos))
2280 goto end_coredump;
2281
2282 for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2283 vma = next_vma(vma, gate_vma)) {
2284 unsigned long addr;
2285 unsigned long end;
2286
2287 end = vma->vm_start + vma_filesz[i++];
2288
2289 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2290 struct page *page;
2291 int stop;
2292
2293 page = get_dump_page(addr);
2294 if (page) {
2295 void *kaddr = kmap(page);
2296 stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2297 kunmap(page);
2298 put_page(page);
2299 } else
2300 stop = !dump_skip(cprm, PAGE_SIZE);
2301 if (stop)
2302 goto end_coredump;
2303 }
2304 }
2305
2306 if (!elf_core_write_extra_data(cprm))
2307 goto end_coredump;
2308
2309 if (e_phnum == PN_XNUM) {
2310 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2311 goto end_coredump;
2312 }
2313
2314 end_coredump:
2315 set_fs(fs);
2316
2317 cleanup:
2318 free_note_info(&info);
2319 kfree(shdr4extnum);
2320 kfree(vma_filesz);
2321 kfree(phdr4note);
2322 kfree(elf);
2323 out:
2324 return has_dumped;
2325 }
2326
2327 #endif /* CONFIG_ELF_CORE */
2328
2329 static int __init init_elf_binfmt(void)
2330 {
2331 register_binfmt(&elf_format);
2332 return 0;
2333 }
2334
2335 static void __exit exit_elf_binfmt(void)
2336 {
2337 /* Remove the COFF and ELF loaders. */
2338 unregister_binfmt(&elf_format);
2339 }
2340
2341 core_initcall(init_elf_binfmt);
2342 module_exit(exit_elf_binfmt);
2343 MODULE_LICENSE("GPL");
This page took 0.075508 seconds and 6 git commands to generate.