]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/binfmt_elf_fdpic.c
UBUNTU: Ubuntu-4.13.0-45.50
[mirror_ubuntu-artful-kernel.git] / fs / binfmt_elf_fdpic.c
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2 *
3 * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 * Derived from binfmt_elf.c
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13 #include <linux/module.h>
14
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/sched/coredump.h>
19 #include <linux/sched/task_stack.h>
20 #include <linux/sched/cputime.h>
21 #include <linux/mm.h>
22 #include <linux/mman.h>
23 #include <linux/errno.h>
24 #include <linux/signal.h>
25 #include <linux/binfmts.h>
26 #include <linux/string.h>
27 #include <linux/file.h>
28 #include <linux/fcntl.h>
29 #include <linux/slab.h>
30 #include <linux/pagemap.h>
31 #include <linux/security.h>
32 #include <linux/highmem.h>
33 #include <linux/highuid.h>
34 #include <linux/personality.h>
35 #include <linux/ptrace.h>
36 #include <linux/init.h>
37 #include <linux/elf.h>
38 #include <linux/elf-fdpic.h>
39 #include <linux/elfcore.h>
40 #include <linux/coredump.h>
41 #include <linux/dax.h>
42
43 #include <linux/uaccess.h>
44 #include <asm/param.h>
45 #include <asm/pgalloc.h>
46
47 typedef char *elf_caddr_t;
48
49 #if 0
50 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
51 #else
52 #define kdebug(fmt, ...) do {} while(0)
53 #endif
54
55 #if 0
56 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
57 #else
58 #define kdcore(fmt, ...) do {} while(0)
59 #endif
60
61 MODULE_LICENSE("GPL");
62
63 static int load_elf_fdpic_binary(struct linux_binprm *);
64 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
65 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
66 struct mm_struct *, const char *);
67
68 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
69 struct elf_fdpic_params *,
70 struct elf_fdpic_params *);
71
72 #ifndef CONFIG_MMU
73 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
74 struct file *,
75 struct mm_struct *);
76 #endif
77
78 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
79 struct file *, struct mm_struct *);
80
81 #ifdef CONFIG_ELF_CORE
82 static int elf_fdpic_core_dump(struct coredump_params *cprm);
83 #endif
84
85 static struct linux_binfmt elf_fdpic_format = {
86 .module = THIS_MODULE,
87 .load_binary = load_elf_fdpic_binary,
88 #ifdef CONFIG_ELF_CORE
89 .core_dump = elf_fdpic_core_dump,
90 #endif
91 .min_coredump = ELF_EXEC_PAGESIZE,
92 };
93
94 static int __init init_elf_fdpic_binfmt(void)
95 {
96 register_binfmt(&elf_fdpic_format);
97 return 0;
98 }
99
100 static void __exit exit_elf_fdpic_binfmt(void)
101 {
102 unregister_binfmt(&elf_fdpic_format);
103 }
104
105 core_initcall(init_elf_fdpic_binfmt);
106 module_exit(exit_elf_fdpic_binfmt);
107
108 static int is_elf(struct elfhdr *hdr, struct file *file)
109 {
110 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
111 return 0;
112 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
113 return 0;
114 if (!elf_check_arch(hdr))
115 return 0;
116 if (!file->f_op->mmap)
117 return 0;
118 return 1;
119 }
120
121 #ifndef elf_check_fdpic
122 #define elf_check_fdpic(x) 0
123 #endif
124
125 #ifndef elf_check_const_displacement
126 #define elf_check_const_displacement(x) 0
127 #endif
128
129 static int is_constdisp(struct elfhdr *hdr)
130 {
131 if (!elf_check_fdpic(hdr))
132 return 1;
133 if (elf_check_const_displacement(hdr))
134 return 1;
135 return 0;
136 }
137
138 /*****************************************************************************/
139 /*
140 * read the program headers table into memory
141 */
142 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
143 struct file *file)
144 {
145 struct elf32_phdr *phdr;
146 unsigned long size;
147 int retval, loop;
148
149 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
150 return -ENOMEM;
151 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
152 return -ENOMEM;
153
154 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
155 params->phdrs = kmalloc(size, GFP_KERNEL);
156 if (!params->phdrs)
157 return -ENOMEM;
158
159 retval = kernel_read(file, params->hdr.e_phoff,
160 (char *) params->phdrs, size);
161 if (unlikely(retval != size))
162 return retval < 0 ? retval : -ENOEXEC;
163
164 /* determine stack size for this binary */
165 phdr = params->phdrs;
166 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
167 if (phdr->p_type != PT_GNU_STACK)
168 continue;
169
170 if (phdr->p_flags & PF_X)
171 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
172 else
173 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
174
175 params->stack_size = phdr->p_memsz;
176 break;
177 }
178
179 return 0;
180 }
181
182 /*****************************************************************************/
183 /*
184 * load an fdpic binary into various bits of memory
185 */
186 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
187 {
188 struct elf_fdpic_params exec_params, interp_params;
189 struct pt_regs *regs = current_pt_regs();
190 struct elf_phdr *phdr;
191 unsigned long stack_size, entryaddr;
192 #ifdef ELF_FDPIC_PLAT_INIT
193 unsigned long dynaddr;
194 #endif
195 #ifndef CONFIG_MMU
196 unsigned long stack_prot;
197 #endif
198 struct file *interpreter = NULL; /* to shut gcc up */
199 char *interpreter_name = NULL;
200 int executable_stack;
201 int retval, i;
202
203 kdebug("____ LOAD %d ____", current->pid);
204
205 memset(&exec_params, 0, sizeof(exec_params));
206 memset(&interp_params, 0, sizeof(interp_params));
207
208 exec_params.hdr = *(struct elfhdr *) bprm->buf;
209 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
210
211 /* check that this is a binary we know how to deal with */
212 retval = -ENOEXEC;
213 if (!is_elf(&exec_params.hdr, bprm->file))
214 goto error;
215 if (!elf_check_fdpic(&exec_params.hdr)) {
216 #ifdef CONFIG_MMU
217 /* binfmt_elf handles non-fdpic elf except on nommu */
218 goto error;
219 #else
220 /* nommu can only load ET_DYN (PIE) ELF */
221 if (exec_params.hdr.e_type != ET_DYN)
222 goto error;
223 #endif
224 }
225
226 /* read the program header table */
227 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
228 if (retval < 0)
229 goto error;
230
231 /* scan for a program header that specifies an interpreter */
232 phdr = exec_params.phdrs;
233
234 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
235 switch (phdr->p_type) {
236 case PT_INTERP:
237 retval = -ENOMEM;
238 if (phdr->p_filesz > PATH_MAX)
239 goto error;
240 retval = -ENOENT;
241 if (phdr->p_filesz < 2)
242 goto error;
243
244 /* read the name of the interpreter into memory */
245 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
246 if (!interpreter_name)
247 goto error;
248
249 retval = kernel_read(bprm->file,
250 phdr->p_offset,
251 interpreter_name,
252 phdr->p_filesz);
253 if (unlikely(retval != phdr->p_filesz)) {
254 if (retval >= 0)
255 retval = -ENOEXEC;
256 goto error;
257 }
258
259 retval = -ENOENT;
260 if (interpreter_name[phdr->p_filesz - 1] != '\0')
261 goto error;
262
263 kdebug("Using ELF interpreter %s", interpreter_name);
264
265 /* replace the program with the interpreter */
266 interpreter = open_exec(interpreter_name);
267 retval = PTR_ERR(interpreter);
268 if (IS_ERR(interpreter)) {
269 interpreter = NULL;
270 goto error;
271 }
272
273 /*
274 * If the binary is not readable then enforce
275 * mm->dumpable = 0 regardless of the interpreter's
276 * permissions.
277 */
278 would_dump(bprm, interpreter);
279
280 retval = kernel_read(interpreter, 0, bprm->buf,
281 BINPRM_BUF_SIZE);
282 if (unlikely(retval != BINPRM_BUF_SIZE)) {
283 if (retval >= 0)
284 retval = -ENOEXEC;
285 goto error;
286 }
287
288 interp_params.hdr = *((struct elfhdr *) bprm->buf);
289 break;
290
291 case PT_LOAD:
292 #ifdef CONFIG_MMU
293 if (exec_params.load_addr == 0)
294 exec_params.load_addr = phdr->p_vaddr;
295 #endif
296 break;
297 }
298
299 }
300
301 if (is_constdisp(&exec_params.hdr))
302 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
303
304 /* perform insanity checks on the interpreter */
305 if (interpreter_name) {
306 retval = -ELIBBAD;
307 if (!is_elf(&interp_params.hdr, interpreter))
308 goto error;
309
310 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
311
312 /* read the interpreter's program header table */
313 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
314 if (retval < 0)
315 goto error;
316 }
317
318 stack_size = exec_params.stack_size;
319 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
320 executable_stack = EXSTACK_ENABLE_X;
321 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
322 executable_stack = EXSTACK_DISABLE_X;
323 else
324 executable_stack = EXSTACK_DEFAULT;
325
326 if (stack_size == 0) {
327 stack_size = interp_params.stack_size;
328 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
329 executable_stack = EXSTACK_ENABLE_X;
330 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
331 executable_stack = EXSTACK_DISABLE_X;
332 else
333 executable_stack = EXSTACK_DEFAULT;
334 }
335
336 retval = -ENOEXEC;
337 if (stack_size == 0)
338 stack_size = 131072UL; /* same as exec.c's default commit */
339
340 if (is_constdisp(&interp_params.hdr))
341 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
342
343 /* flush all traces of the currently running executable */
344 retval = flush_old_exec(bprm);
345 if (retval)
346 goto error;
347
348 /* there's now no turning back... the old userspace image is dead,
349 * defunct, deceased, etc.
350 */
351 if (elf_check_fdpic(&exec_params.hdr))
352 set_personality(PER_LINUX_FDPIC);
353 else
354 set_personality(PER_LINUX);
355 if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
356 current->personality |= READ_IMPLIES_EXEC;
357
358 setup_new_exec(bprm);
359
360 set_binfmt(&elf_fdpic_format);
361
362 current->mm->start_code = 0;
363 current->mm->end_code = 0;
364 current->mm->start_stack = 0;
365 current->mm->start_data = 0;
366 current->mm->end_data = 0;
367 current->mm->context.exec_fdpic_loadmap = 0;
368 current->mm->context.interp_fdpic_loadmap = 0;
369
370 #ifdef CONFIG_MMU
371 elf_fdpic_arch_lay_out_mm(&exec_params,
372 &interp_params,
373 &current->mm->start_stack,
374 &current->mm->start_brk);
375
376 retval = setup_arg_pages(bprm, current->mm->start_stack,
377 executable_stack);
378 if (retval < 0)
379 goto error;
380 #endif
381
382 /* load the executable and interpreter into memory */
383 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
384 "executable");
385 if (retval < 0)
386 goto error;
387
388 if (interpreter_name) {
389 retval = elf_fdpic_map_file(&interp_params, interpreter,
390 current->mm, "interpreter");
391 if (retval < 0) {
392 printk(KERN_ERR "Unable to load interpreter\n");
393 goto error;
394 }
395
396 allow_write_access(interpreter);
397 fput(interpreter);
398 interpreter = NULL;
399 }
400
401 #ifdef CONFIG_MMU
402 if (!current->mm->start_brk)
403 current->mm->start_brk = current->mm->end_data;
404
405 current->mm->brk = current->mm->start_brk =
406 PAGE_ALIGN(current->mm->start_brk);
407
408 #else
409 /* create a stack area and zero-size brk area */
410 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
411 if (stack_size < PAGE_SIZE * 2)
412 stack_size = PAGE_SIZE * 2;
413
414 stack_prot = PROT_READ | PROT_WRITE;
415 if (executable_stack == EXSTACK_ENABLE_X ||
416 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
417 stack_prot |= PROT_EXEC;
418
419 current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
420 MAP_PRIVATE | MAP_ANONYMOUS |
421 MAP_UNINITIALIZED | MAP_GROWSDOWN,
422 0);
423
424 if (IS_ERR_VALUE(current->mm->start_brk)) {
425 retval = current->mm->start_brk;
426 current->mm->start_brk = 0;
427 goto error;
428 }
429
430 current->mm->brk = current->mm->start_brk;
431 current->mm->context.end_brk = current->mm->start_brk;
432 current->mm->start_stack = current->mm->start_brk + stack_size;
433 #endif
434
435 install_exec_creds(bprm);
436 if (create_elf_fdpic_tables(bprm, current->mm,
437 &exec_params, &interp_params) < 0)
438 goto error;
439
440 kdebug("- start_code %lx", current->mm->start_code);
441 kdebug("- end_code %lx", current->mm->end_code);
442 kdebug("- start_data %lx", current->mm->start_data);
443 kdebug("- end_data %lx", current->mm->end_data);
444 kdebug("- start_brk %lx", current->mm->start_brk);
445 kdebug("- brk %lx", current->mm->brk);
446 kdebug("- start_stack %lx", current->mm->start_stack);
447
448 #ifdef ELF_FDPIC_PLAT_INIT
449 /*
450 * The ABI may specify that certain registers be set up in special
451 * ways (on i386 %edx is the address of a DT_FINI function, for
452 * example. This macro performs whatever initialization to
453 * the regs structure is required.
454 */
455 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
456 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
457 dynaddr);
458 #endif
459
460 /* everything is now ready... get the userspace context ready to roll */
461 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
462 start_thread(regs, entryaddr, current->mm->start_stack);
463
464 retval = 0;
465
466 error:
467 if (interpreter) {
468 allow_write_access(interpreter);
469 fput(interpreter);
470 }
471 kfree(interpreter_name);
472 kfree(exec_params.phdrs);
473 kfree(exec_params.loadmap);
474 kfree(interp_params.phdrs);
475 kfree(interp_params.loadmap);
476 return retval;
477 }
478
479 /*****************************************************************************/
480
481 #ifndef ELF_BASE_PLATFORM
482 /*
483 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
484 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
485 * will be copied to the user stack in the same manner as AT_PLATFORM.
486 */
487 #define ELF_BASE_PLATFORM NULL
488 #endif
489
490 /*
491 * present useful information to the program by shovelling it onto the new
492 * process's stack
493 */
494 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
495 struct mm_struct *mm,
496 struct elf_fdpic_params *exec_params,
497 struct elf_fdpic_params *interp_params)
498 {
499 const struct cred *cred = current_cred();
500 unsigned long sp, csp, nitems;
501 elf_caddr_t __user *argv, *envp;
502 size_t platform_len = 0, len;
503 char *k_platform, *k_base_platform;
504 char __user *u_platform, *u_base_platform, *p;
505 int loop;
506 int nr; /* reset for each csp adjustment */
507
508 #ifdef CONFIG_MMU
509 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
510 * by the processes running on the same package. One thing we can do is
511 * to shuffle the initial stack for them, so we give the architecture
512 * an opportunity to do so here.
513 */
514 sp = arch_align_stack(bprm->p);
515 #else
516 sp = mm->start_stack;
517
518 /* stack the program arguments and environment */
519 if (transfer_args_to_stack(bprm, &sp) < 0)
520 return -EFAULT;
521 sp &= ~15;
522 #endif
523
524 /*
525 * If this architecture has a platform capability string, copy it
526 * to userspace. In some cases (Sparc), this info is impossible
527 * for userspace to get any other way, in others (i386) it is
528 * merely difficult.
529 */
530 k_platform = ELF_PLATFORM;
531 u_platform = NULL;
532
533 if (k_platform) {
534 platform_len = strlen(k_platform) + 1;
535 sp -= platform_len;
536 u_platform = (char __user *) sp;
537 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
538 return -EFAULT;
539 }
540
541 /*
542 * If this architecture has a "base" platform capability
543 * string, copy it to userspace.
544 */
545 k_base_platform = ELF_BASE_PLATFORM;
546 u_base_platform = NULL;
547
548 if (k_base_platform) {
549 platform_len = strlen(k_base_platform) + 1;
550 sp -= platform_len;
551 u_base_platform = (char __user *) sp;
552 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
553 return -EFAULT;
554 }
555
556 sp &= ~7UL;
557
558 /* stack the load map(s) */
559 len = sizeof(struct elf32_fdpic_loadmap);
560 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
561 sp = (sp - len) & ~7UL;
562 exec_params->map_addr = sp;
563
564 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
565 return -EFAULT;
566
567 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
568
569 if (interp_params->loadmap) {
570 len = sizeof(struct elf32_fdpic_loadmap);
571 len += sizeof(struct elf32_fdpic_loadseg) *
572 interp_params->loadmap->nsegs;
573 sp = (sp - len) & ~7UL;
574 interp_params->map_addr = sp;
575
576 if (copy_to_user((void __user *) sp, interp_params->loadmap,
577 len) != 0)
578 return -EFAULT;
579
580 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
581 }
582
583 /* force 16 byte _final_ alignment here for generality */
584 #define DLINFO_ITEMS 15
585
586 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
587 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
588
589 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
590 nitems++;
591
592 csp = sp;
593 sp -= nitems * 2 * sizeof(unsigned long);
594 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
595 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
596 sp -= 1 * sizeof(unsigned long); /* argc */
597
598 csp -= sp & 15UL;
599 sp -= sp & 15UL;
600
601 /* put the ELF interpreter info on the stack */
602 #define NEW_AUX_ENT(id, val) \
603 do { \
604 struct { unsigned long _id, _val; } __user *ent; \
605 \
606 ent = (void __user *) csp; \
607 __put_user((id), &ent[nr]._id); \
608 __put_user((val), &ent[nr]._val); \
609 nr++; \
610 } while (0)
611
612 nr = 0;
613 csp -= 2 * sizeof(unsigned long);
614 NEW_AUX_ENT(AT_NULL, 0);
615 if (k_platform) {
616 nr = 0;
617 csp -= 2 * sizeof(unsigned long);
618 NEW_AUX_ENT(AT_PLATFORM,
619 (elf_addr_t) (unsigned long) u_platform);
620 }
621
622 if (k_base_platform) {
623 nr = 0;
624 csp -= 2 * sizeof(unsigned long);
625 NEW_AUX_ENT(AT_BASE_PLATFORM,
626 (elf_addr_t) (unsigned long) u_base_platform);
627 }
628
629 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
630 nr = 0;
631 csp -= 2 * sizeof(unsigned long);
632 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
633 }
634
635 nr = 0;
636 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
637 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
638 #ifdef ELF_HWCAP2
639 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
640 #endif
641 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
642 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
643 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
644 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
645 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
646 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
647 NEW_AUX_ENT(AT_FLAGS, 0);
648 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
649 NEW_AUX_ENT(AT_UID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
650 NEW_AUX_ENT(AT_EUID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
651 NEW_AUX_ENT(AT_GID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
652 NEW_AUX_ENT(AT_EGID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
653 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
654 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
655
656 #ifdef ARCH_DLINFO
657 nr = 0;
658 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
659
660 /* ARCH_DLINFO must come last so platform specific code can enforce
661 * special alignment requirements on the AUXV if necessary (eg. PPC).
662 */
663 ARCH_DLINFO;
664 #endif
665 #undef NEW_AUX_ENT
666
667 /* allocate room for argv[] and envv[] */
668 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
669 envp = (elf_caddr_t __user *) csp;
670 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
671 argv = (elf_caddr_t __user *) csp;
672
673 /* stack argc */
674 csp -= sizeof(unsigned long);
675 __put_user(bprm->argc, (unsigned long __user *) csp);
676
677 BUG_ON(csp != sp);
678
679 /* fill in the argv[] array */
680 #ifdef CONFIG_MMU
681 current->mm->arg_start = bprm->p;
682 #else
683 current->mm->arg_start = current->mm->start_stack -
684 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
685 #endif
686
687 p = (char __user *) current->mm->arg_start;
688 for (loop = bprm->argc; loop > 0; loop--) {
689 __put_user((elf_caddr_t) p, argv++);
690 len = strnlen_user(p, MAX_ARG_STRLEN);
691 if (!len || len > MAX_ARG_STRLEN)
692 return -EINVAL;
693 p += len;
694 }
695 __put_user(NULL, argv);
696 current->mm->arg_end = (unsigned long) p;
697
698 /* fill in the envv[] array */
699 current->mm->env_start = (unsigned long) p;
700 for (loop = bprm->envc; loop > 0; loop--) {
701 __put_user((elf_caddr_t)(unsigned long) p, envp++);
702 len = strnlen_user(p, MAX_ARG_STRLEN);
703 if (!len || len > MAX_ARG_STRLEN)
704 return -EINVAL;
705 p += len;
706 }
707 __put_user(NULL, envp);
708 current->mm->env_end = (unsigned long) p;
709
710 mm->start_stack = (unsigned long) sp;
711 return 0;
712 }
713
714 /*****************************************************************************/
715 /*
716 * load the appropriate binary image (executable or interpreter) into memory
717 * - we assume no MMU is available
718 * - if no other PIC bits are set in params->hdr->e_flags
719 * - we assume that the LOADable segments in the binary are independently relocatable
720 * - we assume R/O executable segments are shareable
721 * - else
722 * - we assume the loadable parts of the image to require fixed displacement
723 * - the image is not shareable
724 */
725 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
726 struct file *file,
727 struct mm_struct *mm,
728 const char *what)
729 {
730 struct elf32_fdpic_loadmap *loadmap;
731 #ifdef CONFIG_MMU
732 struct elf32_fdpic_loadseg *mseg;
733 #endif
734 struct elf32_fdpic_loadseg *seg;
735 struct elf32_phdr *phdr;
736 unsigned long load_addr, stop;
737 unsigned nloads, tmp;
738 size_t size;
739 int loop, ret;
740
741 /* allocate a load map table */
742 nloads = 0;
743 for (loop = 0; loop < params->hdr.e_phnum; loop++)
744 if (params->phdrs[loop].p_type == PT_LOAD)
745 nloads++;
746
747 if (nloads == 0)
748 return -ELIBBAD;
749
750 size = sizeof(*loadmap) + nloads * sizeof(*seg);
751 loadmap = kzalloc(size, GFP_KERNEL);
752 if (!loadmap)
753 return -ENOMEM;
754
755 params->loadmap = loadmap;
756
757 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
758 loadmap->nsegs = nloads;
759
760 load_addr = params->load_addr;
761 seg = loadmap->segs;
762
763 /* map the requested LOADs into the memory space */
764 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
765 case ELF_FDPIC_FLAG_CONSTDISP:
766 case ELF_FDPIC_FLAG_CONTIGUOUS:
767 #ifndef CONFIG_MMU
768 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
769 if (ret < 0)
770 return ret;
771 break;
772 #endif
773 default:
774 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
775 if (ret < 0)
776 return ret;
777 break;
778 }
779
780 /* map the entry point */
781 if (params->hdr.e_entry) {
782 seg = loadmap->segs;
783 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
784 if (params->hdr.e_entry >= seg->p_vaddr &&
785 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
786 params->entry_addr =
787 (params->hdr.e_entry - seg->p_vaddr) +
788 seg->addr;
789 break;
790 }
791 }
792 }
793
794 /* determine where the program header table has wound up if mapped */
795 stop = params->hdr.e_phoff;
796 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
797 phdr = params->phdrs;
798
799 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
800 if (phdr->p_type != PT_LOAD)
801 continue;
802
803 if (phdr->p_offset > params->hdr.e_phoff ||
804 phdr->p_offset + phdr->p_filesz < stop)
805 continue;
806
807 seg = loadmap->segs;
808 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
809 if (phdr->p_vaddr >= seg->p_vaddr &&
810 phdr->p_vaddr + phdr->p_filesz <=
811 seg->p_vaddr + seg->p_memsz) {
812 params->ph_addr =
813 (phdr->p_vaddr - seg->p_vaddr) +
814 seg->addr +
815 params->hdr.e_phoff - phdr->p_offset;
816 break;
817 }
818 }
819 break;
820 }
821
822 /* determine where the dynamic section has wound up if there is one */
823 phdr = params->phdrs;
824 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
825 if (phdr->p_type != PT_DYNAMIC)
826 continue;
827
828 seg = loadmap->segs;
829 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
830 if (phdr->p_vaddr >= seg->p_vaddr &&
831 phdr->p_vaddr + phdr->p_memsz <=
832 seg->p_vaddr + seg->p_memsz) {
833 params->dynamic_addr =
834 (phdr->p_vaddr - seg->p_vaddr) +
835 seg->addr;
836
837 /* check the dynamic section contains at least
838 * one item, and that the last item is a NULL
839 * entry */
840 if (phdr->p_memsz == 0 ||
841 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
842 goto dynamic_error;
843
844 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
845 if (((Elf32_Dyn *)
846 params->dynamic_addr)[tmp - 1].d_tag != 0)
847 goto dynamic_error;
848 break;
849 }
850 }
851 break;
852 }
853
854 /* now elide adjacent segments in the load map on MMU linux
855 * - on uClinux the holes between may actually be filled with system
856 * stuff or stuff from other processes
857 */
858 #ifdef CONFIG_MMU
859 nloads = loadmap->nsegs;
860 mseg = loadmap->segs;
861 seg = mseg + 1;
862 for (loop = 1; loop < nloads; loop++) {
863 /* see if we have a candidate for merging */
864 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
865 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
866 if (load_addr == (seg->addr & PAGE_MASK)) {
867 mseg->p_memsz +=
868 load_addr -
869 (mseg->addr + mseg->p_memsz);
870 mseg->p_memsz += seg->addr & ~PAGE_MASK;
871 mseg->p_memsz += seg->p_memsz;
872 loadmap->nsegs--;
873 continue;
874 }
875 }
876
877 mseg++;
878 if (mseg != seg)
879 *mseg = *seg;
880 }
881 #endif
882
883 kdebug("Mapped Object [%s]:", what);
884 kdebug("- elfhdr : %lx", params->elfhdr_addr);
885 kdebug("- entry : %lx", params->entry_addr);
886 kdebug("- PHDR[] : %lx", params->ph_addr);
887 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
888 seg = loadmap->segs;
889 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
890 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
891 loop,
892 seg->addr, seg->addr + seg->p_memsz - 1,
893 seg->p_vaddr, seg->p_memsz);
894
895 return 0;
896
897 dynamic_error:
898 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
899 what, file_inode(file)->i_ino);
900 return -ELIBBAD;
901 }
902
903 /*****************************************************************************/
904 /*
905 * map a file with constant displacement under uClinux
906 */
907 #ifndef CONFIG_MMU
908 static int elf_fdpic_map_file_constdisp_on_uclinux(
909 struct elf_fdpic_params *params,
910 struct file *file,
911 struct mm_struct *mm)
912 {
913 struct elf32_fdpic_loadseg *seg;
914 struct elf32_phdr *phdr;
915 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
916 int loop, ret;
917
918 load_addr = params->load_addr;
919 seg = params->loadmap->segs;
920
921 /* determine the bounds of the contiguous overall allocation we must
922 * make */
923 phdr = params->phdrs;
924 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
925 if (params->phdrs[loop].p_type != PT_LOAD)
926 continue;
927
928 if (base > phdr->p_vaddr)
929 base = phdr->p_vaddr;
930 if (top < phdr->p_vaddr + phdr->p_memsz)
931 top = phdr->p_vaddr + phdr->p_memsz;
932 }
933
934 /* allocate one big anon block for everything */
935 mflags = MAP_PRIVATE;
936 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
937 mflags |= MAP_EXECUTABLE;
938
939 maddr = vm_mmap(NULL, load_addr, top - base,
940 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
941 if (IS_ERR_VALUE(maddr))
942 return (int) maddr;
943
944 if (load_addr != 0)
945 load_addr += PAGE_ALIGN(top - base);
946
947 /* and then load the file segments into it */
948 phdr = params->phdrs;
949 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
950 if (params->phdrs[loop].p_type != PT_LOAD)
951 continue;
952
953 seg->addr = maddr + (phdr->p_vaddr - base);
954 seg->p_vaddr = phdr->p_vaddr;
955 seg->p_memsz = phdr->p_memsz;
956
957 ret = read_code(file, seg->addr, phdr->p_offset,
958 phdr->p_filesz);
959 if (ret < 0)
960 return ret;
961
962 /* map the ELF header address if in this segment */
963 if (phdr->p_offset == 0)
964 params->elfhdr_addr = seg->addr;
965
966 /* clear any space allocated but not loaded */
967 if (phdr->p_filesz < phdr->p_memsz) {
968 if (clear_user((void *) (seg->addr + phdr->p_filesz),
969 phdr->p_memsz - phdr->p_filesz))
970 return -EFAULT;
971 }
972
973 if (mm) {
974 if (phdr->p_flags & PF_X) {
975 if (!mm->start_code) {
976 mm->start_code = seg->addr;
977 mm->end_code = seg->addr +
978 phdr->p_memsz;
979 }
980 } else if (!mm->start_data) {
981 mm->start_data = seg->addr;
982 mm->end_data = seg->addr + phdr->p_memsz;
983 }
984 }
985
986 seg++;
987 }
988
989 return 0;
990 }
991 #endif
992
993 /*****************************************************************************/
994 /*
995 * map a binary by direct mmap() of the individual PT_LOAD segments
996 */
997 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
998 struct file *file,
999 struct mm_struct *mm)
1000 {
1001 struct elf32_fdpic_loadseg *seg;
1002 struct elf32_phdr *phdr;
1003 unsigned long load_addr, delta_vaddr;
1004 int loop, dvset;
1005
1006 load_addr = params->load_addr;
1007 delta_vaddr = 0;
1008 dvset = 0;
1009
1010 seg = params->loadmap->segs;
1011
1012 /* deal with each load segment separately */
1013 phdr = params->phdrs;
1014 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1015 unsigned long maddr, disp, excess, excess1;
1016 int prot = 0, flags;
1017
1018 if (phdr->p_type != PT_LOAD)
1019 continue;
1020
1021 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1022 (unsigned long) phdr->p_vaddr,
1023 (unsigned long) phdr->p_offset,
1024 (unsigned long) phdr->p_filesz,
1025 (unsigned long) phdr->p_memsz);
1026
1027 /* determine the mapping parameters */
1028 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1029 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1030 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1031
1032 flags = MAP_PRIVATE | MAP_DENYWRITE;
1033 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1034 flags |= MAP_EXECUTABLE;
1035
1036 maddr = 0;
1037
1038 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1039 case ELF_FDPIC_FLAG_INDEPENDENT:
1040 /* PT_LOADs are independently locatable */
1041 break;
1042
1043 case ELF_FDPIC_FLAG_HONOURVADDR:
1044 /* the specified virtual address must be honoured */
1045 maddr = phdr->p_vaddr;
1046 flags |= MAP_FIXED;
1047 break;
1048
1049 case ELF_FDPIC_FLAG_CONSTDISP:
1050 /* constant displacement
1051 * - can be mapped anywhere, but must be mapped as a
1052 * unit
1053 */
1054 if (!dvset) {
1055 maddr = load_addr;
1056 delta_vaddr = phdr->p_vaddr;
1057 dvset = 1;
1058 } else {
1059 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1060 flags |= MAP_FIXED;
1061 }
1062 break;
1063
1064 case ELF_FDPIC_FLAG_CONTIGUOUS:
1065 /* contiguity handled later */
1066 break;
1067
1068 default:
1069 BUG();
1070 }
1071
1072 maddr &= PAGE_MASK;
1073
1074 /* create the mapping */
1075 disp = phdr->p_vaddr & ~PAGE_MASK;
1076 maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1077 phdr->p_offset - disp);
1078
1079 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1080 loop, phdr->p_memsz + disp, prot, flags,
1081 phdr->p_offset - disp, maddr);
1082
1083 if (IS_ERR_VALUE(maddr))
1084 return (int) maddr;
1085
1086 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1087 ELF_FDPIC_FLAG_CONTIGUOUS)
1088 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1089
1090 seg->addr = maddr + disp;
1091 seg->p_vaddr = phdr->p_vaddr;
1092 seg->p_memsz = phdr->p_memsz;
1093
1094 /* map the ELF header address if in this segment */
1095 if (phdr->p_offset == 0)
1096 params->elfhdr_addr = seg->addr;
1097
1098 /* clear the bit between beginning of mapping and beginning of
1099 * PT_LOAD */
1100 if (prot & PROT_WRITE && disp > 0) {
1101 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1102 if (clear_user((void __user *) maddr, disp))
1103 return -EFAULT;
1104 maddr += disp;
1105 }
1106
1107 /* clear any space allocated but not loaded
1108 * - on uClinux we can just clear the lot
1109 * - on MMU linux we'll get a SIGBUS beyond the last page
1110 * extant in the file
1111 */
1112 excess = phdr->p_memsz - phdr->p_filesz;
1113 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1114
1115 #ifdef CONFIG_MMU
1116 if (excess > excess1) {
1117 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1118 unsigned long xmaddr;
1119
1120 flags |= MAP_FIXED | MAP_ANONYMOUS;
1121 xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1122 prot, flags, 0);
1123
1124 kdebug("mmap[%d] <anon>"
1125 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1126 loop, xaddr, excess - excess1, prot, flags,
1127 xmaddr);
1128
1129 if (xmaddr != xaddr)
1130 return -ENOMEM;
1131 }
1132
1133 if (prot & PROT_WRITE && excess1 > 0) {
1134 kdebug("clear[%d] ad=%lx sz=%lx",
1135 loop, maddr + phdr->p_filesz, excess1);
1136 if (clear_user((void __user *) maddr + phdr->p_filesz,
1137 excess1))
1138 return -EFAULT;
1139 }
1140
1141 #else
1142 if (excess > 0) {
1143 kdebug("clear[%d] ad=%lx sz=%lx",
1144 loop, maddr + phdr->p_filesz, excess);
1145 if (clear_user((void *) maddr + phdr->p_filesz, excess))
1146 return -EFAULT;
1147 }
1148 #endif
1149
1150 if (mm) {
1151 if (phdr->p_flags & PF_X) {
1152 if (!mm->start_code) {
1153 mm->start_code = maddr;
1154 mm->end_code = maddr + phdr->p_memsz;
1155 }
1156 } else if (!mm->start_data) {
1157 mm->start_data = maddr;
1158 mm->end_data = maddr + phdr->p_memsz;
1159 }
1160 }
1161
1162 seg++;
1163 }
1164
1165 return 0;
1166 }
1167
1168 /*****************************************************************************/
1169 /*
1170 * ELF-FDPIC core dumper
1171 *
1172 * Modelled on fs/exec.c:aout_core_dump()
1173 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1174 *
1175 * Modelled on fs/binfmt_elf.c core dumper
1176 */
1177 #ifdef CONFIG_ELF_CORE
1178
1179 /*
1180 * Decide whether a segment is worth dumping; default is yes to be
1181 * sure (missing info is worse than too much; etc).
1182 * Personally I'd include everything, and use the coredump limit...
1183 *
1184 * I think we should skip something. But I am not sure how. H.J.
1185 */
1186 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1187 {
1188 int dump_ok;
1189
1190 /* Do not dump I/O mapped devices or special mappings */
1191 if (vma->vm_flags & VM_IO) {
1192 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1193 return 0;
1194 }
1195
1196 /* If we may not read the contents, don't allow us to dump
1197 * them either. "dump_write()" can't handle it anyway.
1198 */
1199 if (!(vma->vm_flags & VM_READ)) {
1200 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1201 return 0;
1202 }
1203
1204 /* support for DAX */
1205 if (vma_is_dax(vma)) {
1206 if (vma->vm_flags & VM_SHARED) {
1207 dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1208 kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1209 vma->vm_flags, dump_ok ? "yes" : "no");
1210 } else {
1211 dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1212 kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1213 vma->vm_flags, dump_ok ? "yes" : "no");
1214 }
1215 return dump_ok;
1216 }
1217
1218 /* By default, dump shared memory if mapped from an anonymous file. */
1219 if (vma->vm_flags & VM_SHARED) {
1220 if (file_inode(vma->vm_file)->i_nlink == 0) {
1221 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1222 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1223 vma->vm_flags, dump_ok ? "yes" : "no");
1224 return dump_ok;
1225 }
1226
1227 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1228 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1229 vma->vm_flags, dump_ok ? "yes" : "no");
1230 return dump_ok;
1231 }
1232
1233 #ifdef CONFIG_MMU
1234 /* By default, if it hasn't been written to, don't write it out */
1235 if (!vma->anon_vma) {
1236 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1237 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1238 vma->vm_flags, dump_ok ? "yes" : "no");
1239 return dump_ok;
1240 }
1241 #endif
1242
1243 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1244 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1245 dump_ok ? "yes" : "no");
1246 return dump_ok;
1247 }
1248
1249 /* An ELF note in memory */
1250 struct memelfnote
1251 {
1252 const char *name;
1253 int type;
1254 unsigned int datasz;
1255 void *data;
1256 };
1257
1258 static int notesize(struct memelfnote *en)
1259 {
1260 int sz;
1261
1262 sz = sizeof(struct elf_note);
1263 sz += roundup(strlen(en->name) + 1, 4);
1264 sz += roundup(en->datasz, 4);
1265
1266 return sz;
1267 }
1268
1269 /* #define DEBUG */
1270
1271 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1272 {
1273 struct elf_note en;
1274 en.n_namesz = strlen(men->name) + 1;
1275 en.n_descsz = men->datasz;
1276 en.n_type = men->type;
1277
1278 return dump_emit(cprm, &en, sizeof(en)) &&
1279 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1280 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1281 }
1282
1283 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1284 {
1285 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1286 elf->e_ident[EI_CLASS] = ELF_CLASS;
1287 elf->e_ident[EI_DATA] = ELF_DATA;
1288 elf->e_ident[EI_VERSION] = EV_CURRENT;
1289 elf->e_ident[EI_OSABI] = ELF_OSABI;
1290 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1291
1292 elf->e_type = ET_CORE;
1293 elf->e_machine = ELF_ARCH;
1294 elf->e_version = EV_CURRENT;
1295 elf->e_entry = 0;
1296 elf->e_phoff = sizeof(struct elfhdr);
1297 elf->e_shoff = 0;
1298 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1299 elf->e_ehsize = sizeof(struct elfhdr);
1300 elf->e_phentsize = sizeof(struct elf_phdr);
1301 elf->e_phnum = segs;
1302 elf->e_shentsize = 0;
1303 elf->e_shnum = 0;
1304 elf->e_shstrndx = 0;
1305 return;
1306 }
1307
1308 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1309 {
1310 phdr->p_type = PT_NOTE;
1311 phdr->p_offset = offset;
1312 phdr->p_vaddr = 0;
1313 phdr->p_paddr = 0;
1314 phdr->p_filesz = sz;
1315 phdr->p_memsz = 0;
1316 phdr->p_flags = 0;
1317 phdr->p_align = 0;
1318 return;
1319 }
1320
1321 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1322 unsigned int sz, void *data)
1323 {
1324 note->name = name;
1325 note->type = type;
1326 note->datasz = sz;
1327 note->data = data;
1328 return;
1329 }
1330
1331 /*
1332 * fill up all the fields in prstatus from the given task struct, except
1333 * registers which need to be filled up separately.
1334 */
1335 static void fill_prstatus(struct elf_prstatus *prstatus,
1336 struct task_struct *p, long signr)
1337 {
1338 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1339 prstatus->pr_sigpend = p->pending.signal.sig[0];
1340 prstatus->pr_sighold = p->blocked.sig[0];
1341 rcu_read_lock();
1342 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1343 rcu_read_unlock();
1344 prstatus->pr_pid = task_pid_vnr(p);
1345 prstatus->pr_pgrp = task_pgrp_vnr(p);
1346 prstatus->pr_sid = task_session_vnr(p);
1347 if (thread_group_leader(p)) {
1348 struct task_cputime cputime;
1349
1350 /*
1351 * This is the record for the group leader. It shows the
1352 * group-wide total, not its individual thread total.
1353 */
1354 thread_group_cputime(p, &cputime);
1355 prstatus->pr_utime = ns_to_timeval(cputime.utime);
1356 prstatus->pr_stime = ns_to_timeval(cputime.stime);
1357 } else {
1358 u64 utime, stime;
1359
1360 task_cputime(p, &utime, &stime);
1361 prstatus->pr_utime = ns_to_timeval(utime);
1362 prstatus->pr_stime = ns_to_timeval(stime);
1363 }
1364 prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1365 prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1366
1367 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1368 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1369 }
1370
1371 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1372 struct mm_struct *mm)
1373 {
1374 const struct cred *cred;
1375 unsigned int i, len;
1376
1377 /* first copy the parameters from user space */
1378 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1379
1380 len = mm->arg_end - mm->arg_start;
1381 if (len >= ELF_PRARGSZ)
1382 len = ELF_PRARGSZ - 1;
1383 if (copy_from_user(&psinfo->pr_psargs,
1384 (const char __user *) mm->arg_start, len))
1385 return -EFAULT;
1386 for (i = 0; i < len; i++)
1387 if (psinfo->pr_psargs[i] == 0)
1388 psinfo->pr_psargs[i] = ' ';
1389 psinfo->pr_psargs[len] = 0;
1390
1391 rcu_read_lock();
1392 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1393 rcu_read_unlock();
1394 psinfo->pr_pid = task_pid_vnr(p);
1395 psinfo->pr_pgrp = task_pgrp_vnr(p);
1396 psinfo->pr_sid = task_session_vnr(p);
1397
1398 i = p->state ? ffz(~p->state) + 1 : 0;
1399 psinfo->pr_state = i;
1400 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1401 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1402 psinfo->pr_nice = task_nice(p);
1403 psinfo->pr_flag = p->flags;
1404 rcu_read_lock();
1405 cred = __task_cred(p);
1406 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1407 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1408 rcu_read_unlock();
1409 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1410
1411 return 0;
1412 }
1413
1414 /* Here is the structure in which status of each thread is captured. */
1415 struct elf_thread_status
1416 {
1417 struct list_head list;
1418 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1419 elf_fpregset_t fpu; /* NT_PRFPREG */
1420 struct task_struct *thread;
1421 #ifdef ELF_CORE_COPY_XFPREGS
1422 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1423 #endif
1424 struct memelfnote notes[3];
1425 int num_notes;
1426 };
1427
1428 /*
1429 * In order to add the specific thread information for the elf file format,
1430 * we need to keep a linked list of every thread's pr_status and then create
1431 * a single section for them in the final core file.
1432 */
1433 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1434 {
1435 struct task_struct *p = t->thread;
1436 int sz = 0;
1437
1438 t->num_notes = 0;
1439
1440 fill_prstatus(&t->prstatus, p, signr);
1441 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1442
1443 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1444 &t->prstatus);
1445 t->num_notes++;
1446 sz += notesize(&t->notes[0]);
1447
1448 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1449 if (t->prstatus.pr_fpvalid) {
1450 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1451 &t->fpu);
1452 t->num_notes++;
1453 sz += notesize(&t->notes[1]);
1454 }
1455
1456 #ifdef ELF_CORE_COPY_XFPREGS
1457 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1458 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1459 sizeof(t->xfpu), &t->xfpu);
1460 t->num_notes++;
1461 sz += notesize(&t->notes[2]);
1462 }
1463 #endif
1464 return sz;
1465 }
1466
1467 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1468 elf_addr_t e_shoff, int segs)
1469 {
1470 elf->e_shoff = e_shoff;
1471 elf->e_shentsize = sizeof(*shdr4extnum);
1472 elf->e_shnum = 1;
1473 elf->e_shstrndx = SHN_UNDEF;
1474
1475 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1476
1477 shdr4extnum->sh_type = SHT_NULL;
1478 shdr4extnum->sh_size = elf->e_shnum;
1479 shdr4extnum->sh_link = elf->e_shstrndx;
1480 shdr4extnum->sh_info = segs;
1481 }
1482
1483 /*
1484 * dump the segments for an MMU process
1485 */
1486 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1487 {
1488 struct vm_area_struct *vma;
1489
1490 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1491 unsigned long addr;
1492
1493 if (!maydump(vma, cprm->mm_flags))
1494 continue;
1495
1496 #ifdef CONFIG_MMU
1497 for (addr = vma->vm_start; addr < vma->vm_end;
1498 addr += PAGE_SIZE) {
1499 bool res;
1500 struct page *page = get_dump_page(addr);
1501 if (page) {
1502 void *kaddr = kmap(page);
1503 res = dump_emit(cprm, kaddr, PAGE_SIZE);
1504 kunmap(page);
1505 put_page(page);
1506 } else {
1507 res = dump_skip(cprm, PAGE_SIZE);
1508 }
1509 if (!res)
1510 return false;
1511 }
1512 #else
1513 if (!dump_emit(cprm, (void *) vma->vm_start,
1514 vma->vm_end - vma->vm_start))
1515 return false;
1516 #endif
1517 }
1518 return true;
1519 }
1520
1521 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1522 {
1523 struct vm_area_struct *vma;
1524 size_t size = 0;
1525
1526 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1527 if (maydump(vma, mm_flags))
1528 size += vma->vm_end - vma->vm_start;
1529 return size;
1530 }
1531
1532 /*
1533 * Actual dumper
1534 *
1535 * This is a two-pass process; first we find the offsets of the bits,
1536 * and then they are actually written out. If we run out of core limit
1537 * we just truncate.
1538 */
1539 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1540 {
1541 #define NUM_NOTES 6
1542 int has_dumped = 0;
1543 mm_segment_t fs;
1544 int segs;
1545 int i;
1546 struct vm_area_struct *vma;
1547 struct elfhdr *elf = NULL;
1548 loff_t offset = 0, dataoff;
1549 int numnote;
1550 struct memelfnote *notes = NULL;
1551 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1552 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1553 LIST_HEAD(thread_list);
1554 struct list_head *t;
1555 elf_fpregset_t *fpu = NULL;
1556 #ifdef ELF_CORE_COPY_XFPREGS
1557 elf_fpxregset_t *xfpu = NULL;
1558 #endif
1559 int thread_status_size = 0;
1560 elf_addr_t *auxv;
1561 struct elf_phdr *phdr4note = NULL;
1562 struct elf_shdr *shdr4extnum = NULL;
1563 Elf_Half e_phnum;
1564 elf_addr_t e_shoff;
1565 struct core_thread *ct;
1566 struct elf_thread_status *tmp;
1567
1568 /*
1569 * We no longer stop all VM operations.
1570 *
1571 * This is because those proceses that could possibly change map_count
1572 * or the mmap / vma pages are now blocked in do_exit on current
1573 * finishing this core dump.
1574 *
1575 * Only ptrace can touch these memory addresses, but it doesn't change
1576 * the map_count or the pages allocated. So no possibility of crashing
1577 * exists while dumping the mm->vm_next areas to the core file.
1578 */
1579
1580 /* alloc memory for large data structures: too large to be on stack */
1581 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1582 if (!elf)
1583 goto cleanup;
1584 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1585 if (!prstatus)
1586 goto cleanup;
1587 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1588 if (!psinfo)
1589 goto cleanup;
1590 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1591 if (!notes)
1592 goto cleanup;
1593 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1594 if (!fpu)
1595 goto cleanup;
1596 #ifdef ELF_CORE_COPY_XFPREGS
1597 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1598 if (!xfpu)
1599 goto cleanup;
1600 #endif
1601
1602 for (ct = current->mm->core_state->dumper.next;
1603 ct; ct = ct->next) {
1604 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1605 if (!tmp)
1606 goto cleanup;
1607
1608 tmp->thread = ct->task;
1609 list_add(&tmp->list, &thread_list);
1610 }
1611
1612 list_for_each(t, &thread_list) {
1613 struct elf_thread_status *tmp;
1614 int sz;
1615
1616 tmp = list_entry(t, struct elf_thread_status, list);
1617 sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1618 thread_status_size += sz;
1619 }
1620
1621 /* now collect the dump for the current */
1622 fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1623 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1624
1625 segs = current->mm->map_count;
1626 segs += elf_core_extra_phdrs();
1627
1628 /* for notes section */
1629 segs++;
1630
1631 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1632 * this, kernel supports extended numbering. Have a look at
1633 * include/linux/elf.h for further information. */
1634 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1635
1636 /* Set up header */
1637 fill_elf_fdpic_header(elf, e_phnum);
1638
1639 has_dumped = 1;
1640 /*
1641 * Set up the notes in similar form to SVR4 core dumps made
1642 * with info from their /proc.
1643 */
1644
1645 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1646 fill_psinfo(psinfo, current->group_leader, current->mm);
1647 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1648
1649 numnote = 2;
1650
1651 auxv = (elf_addr_t *) current->mm->saved_auxv;
1652
1653 i = 0;
1654 do
1655 i += 2;
1656 while (auxv[i - 2] != AT_NULL);
1657 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1658 i * sizeof(elf_addr_t), auxv);
1659
1660 /* Try to dump the FPU. */
1661 if ((prstatus->pr_fpvalid =
1662 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1663 fill_note(notes + numnote++,
1664 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1665 #ifdef ELF_CORE_COPY_XFPREGS
1666 if (elf_core_copy_task_xfpregs(current, xfpu))
1667 fill_note(notes + numnote++,
1668 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1669 #endif
1670
1671 fs = get_fs();
1672 set_fs(KERNEL_DS);
1673
1674 offset += sizeof(*elf); /* Elf header */
1675 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1676
1677 /* Write notes phdr entry */
1678 {
1679 int sz = 0;
1680
1681 for (i = 0; i < numnote; i++)
1682 sz += notesize(notes + i);
1683
1684 sz += thread_status_size;
1685
1686 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1687 if (!phdr4note)
1688 goto end_coredump;
1689
1690 fill_elf_note_phdr(phdr4note, sz, offset);
1691 offset += sz;
1692 }
1693
1694 /* Page-align dumped data */
1695 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1696
1697 offset += elf_core_vma_data_size(cprm->mm_flags);
1698 offset += elf_core_extra_data_size();
1699 e_shoff = offset;
1700
1701 if (e_phnum == PN_XNUM) {
1702 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1703 if (!shdr4extnum)
1704 goto end_coredump;
1705 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1706 }
1707
1708 offset = dataoff;
1709
1710 if (!dump_emit(cprm, elf, sizeof(*elf)))
1711 goto end_coredump;
1712
1713 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1714 goto end_coredump;
1715
1716 /* write program headers for segments dump */
1717 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1718 struct elf_phdr phdr;
1719 size_t sz;
1720
1721 sz = vma->vm_end - vma->vm_start;
1722
1723 phdr.p_type = PT_LOAD;
1724 phdr.p_offset = offset;
1725 phdr.p_vaddr = vma->vm_start;
1726 phdr.p_paddr = 0;
1727 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1728 phdr.p_memsz = sz;
1729 offset += phdr.p_filesz;
1730 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1731 if (vma->vm_flags & VM_WRITE)
1732 phdr.p_flags |= PF_W;
1733 if (vma->vm_flags & VM_EXEC)
1734 phdr.p_flags |= PF_X;
1735 phdr.p_align = ELF_EXEC_PAGESIZE;
1736
1737 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1738 goto end_coredump;
1739 }
1740
1741 if (!elf_core_write_extra_phdrs(cprm, offset))
1742 goto end_coredump;
1743
1744 /* write out the notes section */
1745 for (i = 0; i < numnote; i++)
1746 if (!writenote(notes + i, cprm))
1747 goto end_coredump;
1748
1749 /* write out the thread status notes section */
1750 list_for_each(t, &thread_list) {
1751 struct elf_thread_status *tmp =
1752 list_entry(t, struct elf_thread_status, list);
1753
1754 for (i = 0; i < tmp->num_notes; i++)
1755 if (!writenote(&tmp->notes[i], cprm))
1756 goto end_coredump;
1757 }
1758
1759 if (!dump_skip(cprm, dataoff - cprm->pos))
1760 goto end_coredump;
1761
1762 if (!elf_fdpic_dump_segments(cprm))
1763 goto end_coredump;
1764
1765 if (!elf_core_write_extra_data(cprm))
1766 goto end_coredump;
1767
1768 if (e_phnum == PN_XNUM) {
1769 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1770 goto end_coredump;
1771 }
1772
1773 if (cprm->file->f_pos != offset) {
1774 /* Sanity check */
1775 printk(KERN_WARNING
1776 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1777 cprm->file->f_pos, offset);
1778 }
1779
1780 end_coredump:
1781 set_fs(fs);
1782
1783 cleanup:
1784 while (!list_empty(&thread_list)) {
1785 struct list_head *tmp = thread_list.next;
1786 list_del(tmp);
1787 kfree(list_entry(tmp, struct elf_thread_status, list));
1788 }
1789 kfree(phdr4note);
1790 kfree(elf);
1791 kfree(prstatus);
1792 kfree(psinfo);
1793 kfree(notes);
1794 kfree(fpu);
1795 kfree(shdr4extnum);
1796 #ifdef ELF_CORE_COPY_XFPREGS
1797 kfree(xfpu);
1798 #endif
1799 return has_dumped;
1800 #undef NUM_NOTES
1801 }
1802
1803 #endif /* CONFIG_ELF_CORE */