]> git.ipfire.org Git - people/arne_f/kernel.git/blob - fs/binfmt_elf_fdpic.c
projects / people / arne_f / kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
cls_matchall: use tcf_exts_get_net() before call_rcu()
[people/arne_f/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 loff_t pos = params->hdr.e_phoff;
149
150 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
151 return -ENOMEM;
152 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
153 return -ENOMEM;
154
155 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
156 params->phdrs = kmalloc(size, GFP_KERNEL);
157 if (!params->phdrs)
158 return -ENOMEM;
159
160 retval = kernel_read(file, params->phdrs, size, &pos);
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 loff_t pos;
203
204 kdebug("____ LOAD %d ____", current->pid);
205
206 memset(&exec_params, 0, sizeof(exec_params));
207 memset(&interp_params, 0, sizeof(interp_params));
208
209 exec_params.hdr = *(struct elfhdr *) bprm->buf;
210 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
211
212 /* check that this is a binary we know how to deal with */
213 retval = -ENOEXEC;
214 if (!is_elf(&exec_params.hdr, bprm->file))
215 goto error;
216 if (!elf_check_fdpic(&exec_params.hdr)) {
217 #ifdef CONFIG_MMU
218 /* binfmt_elf handles non-fdpic elf except on nommu */
219 goto error;
220 #else
221 /* nommu can only load ET_DYN (PIE) ELF */
222 if (exec_params.hdr.e_type != ET_DYN)
223 goto error;
224 #endif
225 }
226
227 /* read the program header table */
228 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
229 if (retval < 0)
230 goto error;
231
232 /* scan for a program header that specifies an interpreter */
233 phdr = exec_params.phdrs;
234
235 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
236 switch (phdr->p_type) {
237 case PT_INTERP:
238 retval = -ENOMEM;
239 if (phdr->p_filesz > PATH_MAX)
240 goto error;
241 retval = -ENOENT;
242 if (phdr->p_filesz < 2)
243 goto error;
244
245 /* read the name of the interpreter into memory */
246 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
247 if (!interpreter_name)
248 goto error;
249
250 pos = phdr->p_offset;
251 retval = kernel_read(bprm->file, interpreter_name,
252 phdr->p_filesz, &pos);
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 pos = 0;
281 retval = kernel_read(interpreter, bprm->buf,
282 BINPRM_BUF_SIZE, &pos);
283 if (unlikely(retval != BINPRM_BUF_SIZE)) {
284 if (retval >= 0)
285 retval = -ENOEXEC;
286 goto error;
287 }
288
289 interp_params.hdr = *((struct elfhdr *) bprm->buf);
290 break;
291
292 case PT_LOAD:
293 #ifdef CONFIG_MMU
294 if (exec_params.load_addr == 0)
295 exec_params.load_addr = phdr->p_vaddr;
296 #endif
297 break;
298 }
299
300 }
301
302 if (is_constdisp(&exec_params.hdr))
303 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
304
305 /* perform insanity checks on the interpreter */
306 if (interpreter_name) {
307 retval = -ELIBBAD;
308 if (!is_elf(&interp_params.hdr, interpreter))
309 goto error;
310
311 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
312
313 /* read the interpreter's program header table */
314 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
315 if (retval < 0)
316 goto error;
317 }
318
319 stack_size = exec_params.stack_size;
320 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
321 executable_stack = EXSTACK_ENABLE_X;
322 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
323 executable_stack = EXSTACK_DISABLE_X;
324 else
325 executable_stack = EXSTACK_DEFAULT;
326
327 if (stack_size == 0) {
328 stack_size = interp_params.stack_size;
329 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
330 executable_stack = EXSTACK_ENABLE_X;
331 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
332 executable_stack = EXSTACK_DISABLE_X;
333 else
334 executable_stack = EXSTACK_DEFAULT;
335 }
336
337 retval = -ENOEXEC;
338 if (stack_size == 0)
339 stack_size = 131072UL; /* same as exec.c's default commit */
340
341 if (is_constdisp(&interp_params.hdr))
342 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
343
344 /* flush all traces of the currently running executable */
345 retval = flush_old_exec(bprm);
346 if (retval)
347 goto error;
348
349 /* there's now no turning back... the old userspace image is dead,
350 * defunct, deceased, etc.
351 */
352 if (elf_check_fdpic(&exec_params.hdr))
353 set_personality(PER_LINUX_FDPIC);
354 else
355 set_personality(PER_LINUX);
356 if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
357 current->personality |= READ_IMPLIES_EXEC;
358
359 setup_new_exec(bprm);
360
361 set_binfmt(&elf_fdpic_format);
362
363 current->mm->start_code = 0;
364 current->mm->end_code = 0;
365 current->mm->start_stack = 0;
366 current->mm->start_data = 0;
367 current->mm->end_data = 0;
368 current->mm->context.exec_fdpic_loadmap = 0;
369 current->mm->context.interp_fdpic_loadmap = 0;
370
371 #ifdef CONFIG_MMU
372 elf_fdpic_arch_lay_out_mm(&exec_params,
373 &interp_params,
374 &current->mm->start_stack,
375 &current->mm->start_brk);
376
377 retval = setup_arg_pages(bprm, current->mm->start_stack,
378 executable_stack);
379 if (retval < 0)
380 goto error;
381 #endif
382
383 /* load the executable and interpreter into memory */
384 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
385 "executable");
386 if (retval < 0)
387 goto error;
388
389 if (interpreter_name) {
390 retval = elf_fdpic_map_file(&interp_params, interpreter,
391 current->mm, "interpreter");
392 if (retval < 0) {
393 printk(KERN_ERR "Unable to load interpreter\n");
394 goto error;
395 }
396
397 allow_write_access(interpreter);
398 fput(interpreter);
399 interpreter = NULL;
400 }
401
402 #ifdef CONFIG_MMU
403 if (!current->mm->start_brk)
404 current->mm->start_brk = current->mm->end_data;
405
406 current->mm->brk = current->mm->start_brk =
407 PAGE_ALIGN(current->mm->start_brk);
408
409 #else
410 /* create a stack area and zero-size brk area */
411 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
412 if (stack_size < PAGE_SIZE * 2)
413 stack_size = PAGE_SIZE * 2;
414
415 stack_prot = PROT_READ | PROT_WRITE;
416 if (executable_stack == EXSTACK_ENABLE_X ||
417 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
418 stack_prot |= PROT_EXEC;
419
420 current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
421 MAP_PRIVATE | MAP_ANONYMOUS |
422 MAP_UNINITIALIZED | MAP_GROWSDOWN,
423 0);
424
425 if (IS_ERR_VALUE(current->mm->start_brk)) {
426 retval = current->mm->start_brk;
427 current->mm->start_brk = 0;
428 goto error;
429 }
430
431 current->mm->brk = current->mm->start_brk;
432 current->mm->context.end_brk = current->mm->start_brk;
433 current->mm->start_stack = current->mm->start_brk + stack_size;
434 #endif
435
436 install_exec_creds(bprm);
437 if (create_elf_fdpic_tables(bprm, current->mm,
438 &exec_params, &interp_params) < 0)
439 goto error;
440
441 kdebug("- start_code %lx", current->mm->start_code);
442 kdebug("- end_code %lx", current->mm->end_code);
443 kdebug("- start_data %lx", current->mm->start_data);
444 kdebug("- end_data %lx", current->mm->end_data);
445 kdebug("- start_brk %lx", current->mm->start_brk);
446 kdebug("- brk %lx", current->mm->brk);
447 kdebug("- start_stack %lx", current->mm->start_stack);
448
449 #ifdef ELF_FDPIC_PLAT_INIT
450 /*
451 * The ABI may specify that certain registers be set up in special
452 * ways (on i386 %edx is the address of a DT_FINI function, for
453 * example. This macro performs whatever initialization to
454 * the regs structure is required.
455 */
456 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
457 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
458 dynaddr);
459 #endif
460
461 /* everything is now ready... get the userspace context ready to roll */
462 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
463 start_thread(regs, entryaddr, current->mm->start_stack);
464
465 retval = 0;
466
467 error:
468 if (interpreter) {
469 allow_write_access(interpreter);
470 fput(interpreter);
471 }
472 kfree(interpreter_name);
473 kfree(exec_params.phdrs);
474 kfree(exec_params.loadmap);
475 kfree(interp_params.phdrs);
476 kfree(interp_params.loadmap);
477 return retval;
478 }
479
480 /*****************************************************************************/
481
482 #ifndef ELF_BASE_PLATFORM
483 /*
484 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
485 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
486 * will be copied to the user stack in the same manner as AT_PLATFORM.
487 */
488 #define ELF_BASE_PLATFORM NULL
489 #endif
490
491 /*
492 * present useful information to the program by shovelling it onto the new
493 * process's stack
494 */
495 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
496 struct mm_struct *mm,
497 struct elf_fdpic_params *exec_params,
498 struct elf_fdpic_params *interp_params)
499 {
500 const struct cred *cred = current_cred();
501 unsigned long sp, csp, nitems;
502 elf_caddr_t __user *argv, *envp;
503 size_t platform_len = 0, len;
504 char *k_platform, *k_base_platform;
505 char __user *u_platform, *u_base_platform, *p;
506 int loop;
507 int nr; /* reset for each csp adjustment */
508
509 #ifdef CONFIG_MMU
510 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
511 * by the processes running on the same package. One thing we can do is
512 * to shuffle the initial stack for them, so we give the architecture
513 * an opportunity to do so here.
514 */
515 sp = arch_align_stack(bprm->p);
516 #else
517 sp = mm->start_stack;
518
519 /* stack the program arguments and environment */
520 if (transfer_args_to_stack(bprm, &sp) < 0)
521 return -EFAULT;
522 sp &= ~15;
523 #endif
524
525 /*
526 * If this architecture has a platform capability string, copy it
527 * to userspace. In some cases (Sparc), this info is impossible
528 * for userspace to get any other way, in others (i386) it is
529 * merely difficult.
530 */
531 k_platform = ELF_PLATFORM;
532 u_platform = NULL;
533
534 if (k_platform) {
535 platform_len = strlen(k_platform) + 1;
536 sp -= platform_len;
537 u_platform = (char __user *) sp;
538 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
539 return -EFAULT;
540 }
541
542 /*
543 * If this architecture has a "base" platform capability
544 * string, copy it to userspace.
545 */
546 k_base_platform = ELF_BASE_PLATFORM;
547 u_base_platform = NULL;
548
549 if (k_base_platform) {
550 platform_len = strlen(k_base_platform) + 1;
551 sp -= platform_len;
552 u_base_platform = (char __user *) sp;
553 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
554 return -EFAULT;
555 }
556
557 sp &= ~7UL;
558
559 /* stack the load map(s) */
560 len = sizeof(struct elf32_fdpic_loadmap);
561 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
562 sp = (sp - len) & ~7UL;
563 exec_params->map_addr = sp;
564
565 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
566 return -EFAULT;
567
568 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
569
570 if (interp_params->loadmap) {
571 len = sizeof(struct elf32_fdpic_loadmap);
572 len += sizeof(struct elf32_fdpic_loadseg) *
573 interp_params->loadmap->nsegs;
574 sp = (sp - len) & ~7UL;
575 interp_params->map_addr = sp;
576
577 if (copy_to_user((void __user *) sp, interp_params->loadmap,
578 len) != 0)
579 return -EFAULT;
580
581 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
582 }
583
584 /* force 16 byte _final_ alignment here for generality */
585 #define DLINFO_ITEMS 15
586
587 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
588 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
589
590 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
591 nitems++;
592
593 csp = sp;
594 sp -= nitems * 2 * sizeof(unsigned long);
595 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
596 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
597 sp -= 1 * sizeof(unsigned long); /* argc */
598
599 csp -= sp & 15UL;
600 sp -= sp & 15UL;
601
602 /* put the ELF interpreter info on the stack */
603 #define NEW_AUX_ENT(id, val) \
604 do { \
605 struct { unsigned long _id, _val; } __user *ent; \
606 \
607 ent = (void __user *) csp; \
608 __put_user((id), &ent[nr]._id); \
609 __put_user((val), &ent[nr]._val); \
610 nr++; \
611 } while (0)
612
613 nr = 0;
614 csp -= 2 * sizeof(unsigned long);
615 NEW_AUX_ENT(AT_NULL, 0);
616 if (k_platform) {
617 nr = 0;
618 csp -= 2 * sizeof(unsigned long);
619 NEW_AUX_ENT(AT_PLATFORM,
620 (elf_addr_t) (unsigned long) u_platform);
621 }
622
623 if (k_base_platform) {
624 nr = 0;
625 csp -= 2 * sizeof(unsigned long);
626 NEW_AUX_ENT(AT_BASE_PLATFORM,
627 (elf_addr_t) (unsigned long) u_base_platform);
628 }
629
630 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
631 nr = 0;
632 csp -= 2 * sizeof(unsigned long);
633 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
634 }
635
636 nr = 0;
637 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
638 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
639 #ifdef ELF_HWCAP2
640 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
641 #endif
642 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
643 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
644 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
645 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
646 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
647 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
648 NEW_AUX_ENT(AT_FLAGS, 0);
649 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
650 NEW_AUX_ENT(AT_UID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
651 NEW_AUX_ENT(AT_EUID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
652 NEW_AUX_ENT(AT_GID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
653 NEW_AUX_ENT(AT_EGID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
654 NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
655 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
656
657 #ifdef ARCH_DLINFO
658 nr = 0;
659 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
660
661 /* ARCH_DLINFO must come last so platform specific code can enforce
662 * special alignment requirements on the AUXV if necessary (eg. PPC).
663 */
664 ARCH_DLINFO;
665 #endif
666 #undef NEW_AUX_ENT
667
668 /* allocate room for argv[] and envv[] */
669 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
670 envp = (elf_caddr_t __user *) csp;
671 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
672 argv = (elf_caddr_t __user *) csp;
673
674 /* stack argc */
675 csp -= sizeof(unsigned long);
676 __put_user(bprm->argc, (unsigned long __user *) csp);
677
678 BUG_ON(csp != sp);
679
680 /* fill in the argv[] array */
681 #ifdef CONFIG_MMU
682 current->mm->arg_start = bprm->p;
683 #else
684 current->mm->arg_start = current->mm->start_stack -
685 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
686 #endif
687
688 p = (char __user *) current->mm->arg_start;
689 for (loop = bprm->argc; loop > 0; loop--) {
690 __put_user((elf_caddr_t) p, argv++);
691 len = strnlen_user(p, MAX_ARG_STRLEN);
692 if (!len || len > MAX_ARG_STRLEN)
693 return -EINVAL;
694 p += len;
695 }
696 __put_user(NULL, argv);
697 current->mm->arg_end = (unsigned long) p;
698
699 /* fill in the envv[] array */
700 current->mm->env_start = (unsigned long) p;
701 for (loop = bprm->envc; loop > 0; loop--) {
702 __put_user((elf_caddr_t)(unsigned long) p, envp++);
703 len = strnlen_user(p, MAX_ARG_STRLEN);
704 if (!len || len > MAX_ARG_STRLEN)
705 return -EINVAL;
706 p += len;
707 }
708 __put_user(NULL, envp);
709 current->mm->env_end = (unsigned long) p;
710
711 mm->start_stack = (unsigned long) sp;
712 return 0;
713 }
714
715 /*****************************************************************************/
716 /*
717 * load the appropriate binary image (executable or interpreter) into memory
718 * - we assume no MMU is available
719 * - if no other PIC bits are set in params->hdr->e_flags
720 * - we assume that the LOADable segments in the binary are independently relocatable
721 * - we assume R/O executable segments are shareable
722 * - else
723 * - we assume the loadable parts of the image to require fixed displacement
724 * - the image is not shareable
725 */
726 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
727 struct file *file,
728 struct mm_struct *mm,
729 const char *what)
730 {
731 struct elf32_fdpic_loadmap *loadmap;
732 #ifdef CONFIG_MMU
733 struct elf32_fdpic_loadseg *mseg;
734 #endif
735 struct elf32_fdpic_loadseg *seg;
736 struct elf32_phdr *phdr;
737 unsigned long load_addr, stop;
738 unsigned nloads, tmp;
739 size_t size;
740 int loop, ret;
741
742 /* allocate a load map table */
743 nloads = 0;
744 for (loop = 0; loop < params->hdr.e_phnum; loop++)
745 if (params->phdrs[loop].p_type == PT_LOAD)
746 nloads++;
747
748 if (nloads == 0)
749 return -ELIBBAD;
750
751 size = sizeof(*loadmap) + nloads * sizeof(*seg);
752 loadmap = kzalloc(size, GFP_KERNEL);
753 if (!loadmap)
754 return -ENOMEM;
755
756 params->loadmap = loadmap;
757
758 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
759 loadmap->nsegs = nloads;
760
761 load_addr = params->load_addr;
762 seg = loadmap->segs;
763
764 /* map the requested LOADs into the memory space */
765 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
766 case ELF_FDPIC_FLAG_CONSTDISP:
767 case ELF_FDPIC_FLAG_CONTIGUOUS:
768 #ifndef CONFIG_MMU
769 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
770 if (ret < 0)
771 return ret;
772 break;
773 #endif
774 default:
775 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
776 if (ret < 0)
777 return ret;
778 break;
779 }
780
781 /* map the entry point */
782 if (params->hdr.e_entry) {
783 seg = loadmap->segs;
784 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
785 if (params->hdr.e_entry >= seg->p_vaddr &&
786 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
787 params->entry_addr =
788 (params->hdr.e_entry - seg->p_vaddr) +
789 seg->addr;
790 break;
791 }
792 }
793 }
794
795 /* determine where the program header table has wound up if mapped */
796 stop = params->hdr.e_phoff;
797 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
798 phdr = params->phdrs;
799
800 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
801 if (phdr->p_type != PT_LOAD)
802 continue;
803
804 if (phdr->p_offset > params->hdr.e_phoff ||
805 phdr->p_offset + phdr->p_filesz < stop)
806 continue;
807
808 seg = loadmap->segs;
809 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
810 if (phdr->p_vaddr >= seg->p_vaddr &&
811 phdr->p_vaddr + phdr->p_filesz <=
812 seg->p_vaddr + seg->p_memsz) {
813 params->ph_addr =
814 (phdr->p_vaddr - seg->p_vaddr) +
815 seg->addr +
816 params->hdr.e_phoff - phdr->p_offset;
817 break;
818 }
819 }
820 break;
821 }
822
823 /* determine where the dynamic section has wound up if there is one */
824 phdr = params->phdrs;
825 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
826 if (phdr->p_type != PT_DYNAMIC)
827 continue;
828
829 seg = loadmap->segs;
830 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
831 if (phdr->p_vaddr >= seg->p_vaddr &&
832 phdr->p_vaddr + phdr->p_memsz <=
833 seg->p_vaddr + seg->p_memsz) {
834 params->dynamic_addr =
835 (phdr->p_vaddr - seg->p_vaddr) +
836 seg->addr;
837
838 /* check the dynamic section contains at least
839 * one item, and that the last item is a NULL
840 * entry */
841 if (phdr->p_memsz == 0 ||
842 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
843 goto dynamic_error;
844
845 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
846 if (((Elf32_Dyn *)
847 params->dynamic_addr)[tmp - 1].d_tag != 0)
848 goto dynamic_error;
849 break;
850 }
851 }
852 break;
853 }
854
855 /* now elide adjacent segments in the load map on MMU linux
856 * - on uClinux the holes between may actually be filled with system
857 * stuff or stuff from other processes
858 */
859 #ifdef CONFIG_MMU
860 nloads = loadmap->nsegs;
861 mseg = loadmap->segs;
862 seg = mseg + 1;
863 for (loop = 1; loop < nloads; loop++) {
864 /* see if we have a candidate for merging */
865 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
866 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
867 if (load_addr == (seg->addr & PAGE_MASK)) {
868 mseg->p_memsz +=
869 load_addr -
870 (mseg->addr + mseg->p_memsz);
871 mseg->p_memsz += seg->addr & ~PAGE_MASK;
872 mseg->p_memsz += seg->p_memsz;
873 loadmap->nsegs--;
874 continue;
875 }
876 }
877
878 mseg++;
879 if (mseg != seg)
880 *mseg = *seg;
881 }
882 #endif
883
884 kdebug("Mapped Object [%s]:", what);
885 kdebug("- elfhdr : %lx", params->elfhdr_addr);
886 kdebug("- entry : %lx", params->entry_addr);
887 kdebug("- PHDR[] : %lx", params->ph_addr);
888 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
889 seg = loadmap->segs;
890 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
891 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
892 loop,
893 seg->addr, seg->addr + seg->p_memsz - 1,
894 seg->p_vaddr, seg->p_memsz);
895
896 return 0;
897
898 dynamic_error:
899 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
900 what, file_inode(file)->i_ino);
901 return -ELIBBAD;
902 }
903
904 /*****************************************************************************/
905 /*
906 * map a file with constant displacement under uClinux
907 */
908 #ifndef CONFIG_MMU
909 static int elf_fdpic_map_file_constdisp_on_uclinux(
910 struct elf_fdpic_params *params,
911 struct file *file,
912 struct mm_struct *mm)
913 {
914 struct elf32_fdpic_loadseg *seg;
915 struct elf32_phdr *phdr;
916 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
917 int loop, ret;
918
919 load_addr = params->load_addr;
920 seg = params->loadmap->segs;
921
922 /* determine the bounds of the contiguous overall allocation we must
923 * make */
924 phdr = params->phdrs;
925 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
926 if (params->phdrs[loop].p_type != PT_LOAD)
927 continue;
928
929 if (base > phdr->p_vaddr)
930 base = phdr->p_vaddr;
931 if (top < phdr->p_vaddr + phdr->p_memsz)
932 top = phdr->p_vaddr + phdr->p_memsz;
933 }
934
935 /* allocate one big anon block for everything */
936 mflags = MAP_PRIVATE;
937 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
938 mflags |= MAP_EXECUTABLE;
939
940 maddr = vm_mmap(NULL, load_addr, top - base,
941 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
942 if (IS_ERR_VALUE(maddr))
943 return (int) maddr;
944
945 if (load_addr != 0)
946 load_addr += PAGE_ALIGN(top - base);
947
948 /* and then load the file segments into it */
949 phdr = params->phdrs;
950 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
951 if (params->phdrs[loop].p_type != PT_LOAD)
952 continue;
953
954 seg->addr = maddr + (phdr->p_vaddr - base);
955 seg->p_vaddr = phdr->p_vaddr;
956 seg->p_memsz = phdr->p_memsz;
957
958 ret = read_code(file, seg->addr, phdr->p_offset,
959 phdr->p_filesz);
960 if (ret < 0)
961 return ret;
962
963 /* map the ELF header address if in this segment */
964 if (phdr->p_offset == 0)
965 params->elfhdr_addr = seg->addr;
966
967 /* clear any space allocated but not loaded */
968 if (phdr->p_filesz < phdr->p_memsz) {
969 if (clear_user((void *) (seg->addr + phdr->p_filesz),
970 phdr->p_memsz - phdr->p_filesz))
971 return -EFAULT;
972 }
973
974 if (mm) {
975 if (phdr->p_flags & PF_X) {
976 if (!mm->start_code) {
977 mm->start_code = seg->addr;
978 mm->end_code = seg->addr +
979 phdr->p_memsz;
980 }
981 } else if (!mm->start_data) {
982 mm->start_data = seg->addr;
983 mm->end_data = seg->addr + phdr->p_memsz;
984 }
985 }
986
987 seg++;
988 }
989
990 return 0;
991 }
992 #endif
993
994 /*****************************************************************************/
995 /*
996 * map a binary by direct mmap() of the individual PT_LOAD segments
997 */
998 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
999 struct file *file,
1000 struct mm_struct *mm)
1001 {
1002 struct elf32_fdpic_loadseg *seg;
1003 struct elf32_phdr *phdr;
1004 unsigned long load_addr, delta_vaddr;
1005 int loop, dvset;
1006
1007 load_addr = params->load_addr;
1008 delta_vaddr = 0;
1009 dvset = 0;
1010
1011 seg = params->loadmap->segs;
1012
1013 /* deal with each load segment separately */
1014 phdr = params->phdrs;
1015 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1016 unsigned long maddr, disp, excess, excess1;
1017 int prot = 0, flags;
1018
1019 if (phdr->p_type != PT_LOAD)
1020 continue;
1021
1022 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1023 (unsigned long) phdr->p_vaddr,
1024 (unsigned long) phdr->p_offset,
1025 (unsigned long) phdr->p_filesz,
1026 (unsigned long) phdr->p_memsz);
1027
1028 /* determine the mapping parameters */
1029 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1030 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1031 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1032
1033 flags = MAP_PRIVATE | MAP_DENYWRITE;
1034 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1035 flags |= MAP_EXECUTABLE;
1036
1037 maddr = 0;
1038
1039 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1040 case ELF_FDPIC_FLAG_INDEPENDENT:
1041 /* PT_LOADs are independently locatable */
1042 break;
1043
1044 case ELF_FDPIC_FLAG_HONOURVADDR:
1045 /* the specified virtual address must be honoured */
1046 maddr = phdr->p_vaddr;
1047 flags |= MAP_FIXED;
1048 break;
1049
1050 case ELF_FDPIC_FLAG_CONSTDISP:
1051 /* constant displacement
1052 * - can be mapped anywhere, but must be mapped as a
1053 * unit
1054 */
1055 if (!dvset) {
1056 maddr = load_addr;
1057 delta_vaddr = phdr->p_vaddr;
1058 dvset = 1;
1059 } else {
1060 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1061 flags |= MAP_FIXED;
1062 }
1063 break;
1064
1065 case ELF_FDPIC_FLAG_CONTIGUOUS:
1066 /* contiguity handled later */
1067 break;
1068
1069 default:
1070 BUG();
1071 }
1072
1073 maddr &= PAGE_MASK;
1074
1075 /* create the mapping */
1076 disp = phdr->p_vaddr & ~PAGE_MASK;
1077 maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1078 phdr->p_offset - disp);
1079
1080 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1081 loop, phdr->p_memsz + disp, prot, flags,
1082 phdr->p_offset - disp, maddr);
1083
1084 if (IS_ERR_VALUE(maddr))
1085 return (int) maddr;
1086
1087 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1088 ELF_FDPIC_FLAG_CONTIGUOUS)
1089 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1090
1091 seg->addr = maddr + disp;
1092 seg->p_vaddr = phdr->p_vaddr;
1093 seg->p_memsz = phdr->p_memsz;
1094
1095 /* map the ELF header address if in this segment */
1096 if (phdr->p_offset == 0)
1097 params->elfhdr_addr = seg->addr;
1098
1099 /* clear the bit between beginning of mapping and beginning of
1100 * PT_LOAD */
1101 if (prot & PROT_WRITE && disp > 0) {
1102 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1103 if (clear_user((void __user *) maddr, disp))
1104 return -EFAULT;
1105 maddr += disp;
1106 }
1107
1108 /* clear any space allocated but not loaded
1109 * - on uClinux we can just clear the lot
1110 * - on MMU linux we'll get a SIGBUS beyond the last page
1111 * extant in the file
1112 */
1113 excess = phdr->p_memsz - phdr->p_filesz;
1114 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1115
1116 #ifdef CONFIG_MMU
1117 if (excess > excess1) {
1118 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1119 unsigned long xmaddr;
1120
1121 flags |= MAP_FIXED | MAP_ANONYMOUS;
1122 xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1123 prot, flags, 0);
1124
1125 kdebug("mmap[%d] <anon>"
1126 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1127 loop, xaddr, excess - excess1, prot, flags,
1128 xmaddr);
1129
1130 if (xmaddr != xaddr)
1131 return -ENOMEM;
1132 }
1133
1134 if (prot & PROT_WRITE && excess1 > 0) {
1135 kdebug("clear[%d] ad=%lx sz=%lx",
1136 loop, maddr + phdr->p_filesz, excess1);
1137 if (clear_user((void __user *) maddr + phdr->p_filesz,
1138 excess1))
1139 return -EFAULT;
1140 }
1141
1142 #else
1143 if (excess > 0) {
1144 kdebug("clear[%d] ad=%lx sz=%lx",
1145 loop, maddr + phdr->p_filesz, excess);
1146 if (clear_user((void *) maddr + phdr->p_filesz, excess))
1147 return -EFAULT;
1148 }
1149 #endif
1150
1151 if (mm) {
1152 if (phdr->p_flags & PF_X) {
1153 if (!mm->start_code) {
1154 mm->start_code = maddr;
1155 mm->end_code = maddr + phdr->p_memsz;
1156 }
1157 } else if (!mm->start_data) {
1158 mm->start_data = maddr;
1159 mm->end_data = maddr + phdr->p_memsz;
1160 }
1161 }
1162
1163 seg++;
1164 }
1165
1166 return 0;
1167 }
1168
1169 /*****************************************************************************/
1170 /*
1171 * ELF-FDPIC core dumper
1172 *
1173 * Modelled on fs/exec.c:aout_core_dump()
1174 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1175 *
1176 * Modelled on fs/binfmt_elf.c core dumper
1177 */
1178 #ifdef CONFIG_ELF_CORE
1179
1180 /*
1181 * Decide whether a segment is worth dumping; default is yes to be
1182 * sure (missing info is worse than too much; etc).
1183 * Personally I'd include everything, and use the coredump limit...
1184 *
1185 * I think we should skip something. But I am not sure how. H.J.
1186 */
1187 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1188 {
1189 int dump_ok;
1190
1191 /* Do not dump I/O mapped devices or special mappings */
1192 if (vma->vm_flags & VM_IO) {
1193 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1194 return 0;
1195 }
1196
1197 /* If we may not read the contents, don't allow us to dump
1198 * them either. "dump_write()" can't handle it anyway.
1199 */
1200 if (!(vma->vm_flags & VM_READ)) {
1201 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1202 return 0;
1203 }
1204
1205 /* support for DAX */
1206 if (vma_is_dax(vma)) {
1207 if (vma->vm_flags & VM_SHARED) {
1208 dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1209 kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1210 vma->vm_flags, dump_ok ? "yes" : "no");
1211 } else {
1212 dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1213 kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1214 vma->vm_flags, dump_ok ? "yes" : "no");
1215 }
1216 return dump_ok;
1217 }
1218
1219 /* By default, dump shared memory if mapped from an anonymous file. */
1220 if (vma->vm_flags & VM_SHARED) {
1221 if (file_inode(vma->vm_file)->i_nlink == 0) {
1222 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1223 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1224 vma->vm_flags, dump_ok ? "yes" : "no");
1225 return dump_ok;
1226 }
1227
1228 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1229 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1230 vma->vm_flags, dump_ok ? "yes" : "no");
1231 return dump_ok;
1232 }
1233
1234 #ifdef CONFIG_MMU
1235 /* By default, if it hasn't been written to, don't write it out */
1236 if (!vma->anon_vma) {
1237 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1238 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1239 vma->vm_flags, dump_ok ? "yes" : "no");
1240 return dump_ok;
1241 }
1242 #endif
1243
1244 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1245 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1246 dump_ok ? "yes" : "no");
1247 return dump_ok;
1248 }
1249
1250 /* An ELF note in memory */
1251 struct memelfnote
1252 {
1253 const char *name;
1254 int type;
1255 unsigned int datasz;
1256 void *data;
1257 };
1258
1259 static int notesize(struct memelfnote *en)
1260 {
1261 int sz;
1262
1263 sz = sizeof(struct elf_note);
1264 sz += roundup(strlen(en->name) + 1, 4);
1265 sz += roundup(en->datasz, 4);
1266
1267 return sz;
1268 }
1269
1270 /* #define DEBUG */
1271
1272 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1273 {
1274 struct elf_note en;
1275 en.n_namesz = strlen(men->name) + 1;
1276 en.n_descsz = men->datasz;
1277 en.n_type = men->type;
1278
1279 return dump_emit(cprm, &en, sizeof(en)) &&
1280 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1281 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1282 }
1283
1284 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1285 {
1286 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1287 elf->e_ident[EI_CLASS] = ELF_CLASS;
1288 elf->e_ident[EI_DATA] = ELF_DATA;
1289 elf->e_ident[EI_VERSION] = EV_CURRENT;
1290 elf->e_ident[EI_OSABI] = ELF_OSABI;
1291 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1292
1293 elf->e_type = ET_CORE;
1294 elf->e_machine = ELF_ARCH;
1295 elf->e_version = EV_CURRENT;
1296 elf->e_entry = 0;
1297 elf->e_phoff = sizeof(struct elfhdr);
1298 elf->e_shoff = 0;
1299 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1300 elf->e_ehsize = sizeof(struct elfhdr);
1301 elf->e_phentsize = sizeof(struct elf_phdr);
1302 elf->e_phnum = segs;
1303 elf->e_shentsize = 0;
1304 elf->e_shnum = 0;
1305 elf->e_shstrndx = 0;
1306 return;
1307 }
1308
1309 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1310 {
1311 phdr->p_type = PT_NOTE;
1312 phdr->p_offset = offset;
1313 phdr->p_vaddr = 0;
1314 phdr->p_paddr = 0;
1315 phdr->p_filesz = sz;
1316 phdr->p_memsz = 0;
1317 phdr->p_flags = 0;
1318 phdr->p_align = 0;
1319 return;
1320 }
1321
1322 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1323 unsigned int sz, void *data)
1324 {
1325 note->name = name;
1326 note->type = type;
1327 note->datasz = sz;
1328 note->data = data;
1329 return;
1330 }
1331
1332 /*
1333 * fill up all the fields in prstatus from the given task struct, except
1334 * registers which need to be filled up separately.
1335 */
1336 static void fill_prstatus(struct elf_prstatus *prstatus,
1337 struct task_struct *p, long signr)
1338 {
1339 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1340 prstatus->pr_sigpend = p->pending.signal.sig[0];
1341 prstatus->pr_sighold = p->blocked.sig[0];
1342 rcu_read_lock();
1343 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1344 rcu_read_unlock();
1345 prstatus->pr_pid = task_pid_vnr(p);
1346 prstatus->pr_pgrp = task_pgrp_vnr(p);
1347 prstatus->pr_sid = task_session_vnr(p);
1348 if (thread_group_leader(p)) {
1349 struct task_cputime cputime;
1350
1351 /*
1352 * This is the record for the group leader. It shows the
1353 * group-wide total, not its individual thread total.
1354 */
1355 thread_group_cputime(p, &cputime);
1356 prstatus->pr_utime = ns_to_timeval(cputime.utime);
1357 prstatus->pr_stime = ns_to_timeval(cputime.stime);
1358 } else {
1359 u64 utime, stime;
1360
1361 task_cputime(p, &utime, &stime);
1362 prstatus->pr_utime = ns_to_timeval(utime);
1363 prstatus->pr_stime = ns_to_timeval(stime);
1364 }
1365 prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1366 prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1367
1368 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1369 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1370 }
1371
1372 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1373 struct mm_struct *mm)
1374 {
1375 const struct cred *cred;
1376 unsigned int i, len;
1377
1378 /* first copy the parameters from user space */
1379 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1380
1381 len = mm->arg_end - mm->arg_start;
1382 if (len >= ELF_PRARGSZ)
1383 len = ELF_PRARGSZ - 1;
1384 if (copy_from_user(&psinfo->pr_psargs,
1385 (const char __user *) mm->arg_start, len))
1386 return -EFAULT;
1387 for (i = 0; i < len; i++)
1388 if (psinfo->pr_psargs[i] == 0)
1389 psinfo->pr_psargs[i] = ' ';
1390 psinfo->pr_psargs[len] = 0;
1391
1392 rcu_read_lock();
1393 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1394 rcu_read_unlock();
1395 psinfo->pr_pid = task_pid_vnr(p);
1396 psinfo->pr_pgrp = task_pgrp_vnr(p);
1397 psinfo->pr_sid = task_session_vnr(p);
1398
1399 i = p->state ? ffz(~p->state) + 1 : 0;
1400 psinfo->pr_state = i;
1401 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1402 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1403 psinfo->pr_nice = task_nice(p);
1404 psinfo->pr_flag = p->flags;
1405 rcu_read_lock();
1406 cred = __task_cred(p);
1407 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1408 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1409 rcu_read_unlock();
1410 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1411
1412 return 0;
1413 }
1414
1415 /* Here is the structure in which status of each thread is captured. */
1416 struct elf_thread_status
1417 {
1418 struct list_head list;
1419 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1420 elf_fpregset_t fpu; /* NT_PRFPREG */
1421 struct task_struct *thread;
1422 #ifdef ELF_CORE_COPY_XFPREGS
1423 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1424 #endif
1425 struct memelfnote notes[3];
1426 int num_notes;
1427 };
1428
1429 /*
1430 * In order to add the specific thread information for the elf file format,
1431 * we need to keep a linked list of every thread's pr_status and then create
1432 * a single section for them in the final core file.
1433 */
1434 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1435 {
1436 struct task_struct *p = t->thread;
1437 int sz = 0;
1438
1439 t->num_notes = 0;
1440
1441 fill_prstatus(&t->prstatus, p, signr);
1442 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1443
1444 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1445 &t->prstatus);
1446 t->num_notes++;
1447 sz += notesize(&t->notes[0]);
1448
1449 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1450 if (t->prstatus.pr_fpvalid) {
1451 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1452 &t->fpu);
1453 t->num_notes++;
1454 sz += notesize(&t->notes[1]);
1455 }
1456
1457 #ifdef ELF_CORE_COPY_XFPREGS
1458 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1459 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1460 sizeof(t->xfpu), &t->xfpu);
1461 t->num_notes++;
1462 sz += notesize(&t->notes[2]);
1463 }
1464 #endif
1465 return sz;
1466 }
1467
1468 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1469 elf_addr_t e_shoff, int segs)
1470 {
1471 elf->e_shoff = e_shoff;
1472 elf->e_shentsize = sizeof(*shdr4extnum);
1473 elf->e_shnum = 1;
1474 elf->e_shstrndx = SHN_UNDEF;
1475
1476 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1477
1478 shdr4extnum->sh_type = SHT_NULL;
1479 shdr4extnum->sh_size = elf->e_shnum;
1480 shdr4extnum->sh_link = elf->e_shstrndx;
1481 shdr4extnum->sh_info = segs;
1482 }
1483
1484 /*
1485 * dump the segments for an MMU process
1486 */
1487 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1488 {
1489 struct vm_area_struct *vma;
1490
1491 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1492 unsigned long addr;
1493
1494 if (!maydump(vma, cprm->mm_flags))
1495 continue;
1496
1497 #ifdef CONFIG_MMU
1498 for (addr = vma->vm_start; addr < vma->vm_end;
1499 addr += PAGE_SIZE) {
1500 bool res;
1501 struct page *page = get_dump_page(addr);
1502 if (page) {
1503 void *kaddr = kmap(page);
1504 res = dump_emit(cprm, kaddr, PAGE_SIZE);
1505 kunmap(page);
1506 put_page(page);
1507 } else {
1508 res = dump_skip(cprm, PAGE_SIZE);
1509 }
1510 if (!res)
1511 return false;
1512 }
1513 #else
1514 if (!dump_emit(cprm, (void *) vma->vm_start,
1515 vma->vm_end - vma->vm_start))
1516 return false;
1517 #endif
1518 }
1519 return true;
1520 }
1521
1522 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1523 {
1524 struct vm_area_struct *vma;
1525 size_t size = 0;
1526
1527 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1528 if (maydump(vma, mm_flags))
1529 size += vma->vm_end - vma->vm_start;
1530 return size;
1531 }
1532
1533 /*
1534 * Actual dumper
1535 *
1536 * This is a two-pass process; first we find the offsets of the bits,
1537 * and then they are actually written out. If we run out of core limit
1538 * we just truncate.
1539 */
1540 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1541 {
1542 #define NUM_NOTES 6
1543 int has_dumped = 0;
1544 mm_segment_t fs;
1545 int segs;
1546 int i;
1547 struct vm_area_struct *vma;
1548 struct elfhdr *elf = NULL;
1549 loff_t offset = 0, dataoff;
1550 int numnote;
1551 struct memelfnote *notes = NULL;
1552 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1553 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1554 LIST_HEAD(thread_list);
1555 struct list_head *t;
1556 elf_fpregset_t *fpu = NULL;
1557 #ifdef ELF_CORE_COPY_XFPREGS
1558 elf_fpxregset_t *xfpu = NULL;
1559 #endif
1560 int thread_status_size = 0;
1561 elf_addr_t *auxv;
1562 struct elf_phdr *phdr4note = NULL;
1563 struct elf_shdr *shdr4extnum = NULL;
1564 Elf_Half e_phnum;
1565 elf_addr_t e_shoff;
1566 struct core_thread *ct;
1567 struct elf_thread_status *tmp;
1568
1569 /*
1570 * We no longer stop all VM operations.
1571 *
1572 * This is because those proceses that could possibly change map_count
1573 * or the mmap / vma pages are now blocked in do_exit on current
1574 * finishing this core dump.
1575 *
1576 * Only ptrace can touch these memory addresses, but it doesn't change
1577 * the map_count or the pages allocated. So no possibility of crashing
1578 * exists while dumping the mm->vm_next areas to the core file.
1579 */
1580
1581 /* alloc memory for large data structures: too large to be on stack */
1582 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1583 if (!elf)
1584 goto cleanup;
1585 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1586 if (!prstatus)
1587 goto cleanup;
1588 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1589 if (!psinfo)
1590 goto cleanup;
1591 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1592 if (!notes)
1593 goto cleanup;
1594 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1595 if (!fpu)
1596 goto cleanup;
1597 #ifdef ELF_CORE_COPY_XFPREGS
1598 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1599 if (!xfpu)
1600 goto cleanup;
1601 #endif
1602
1603 for (ct = current->mm->core_state->dumper.next;
1604 ct; ct = ct->next) {
1605 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1606 if (!tmp)
1607 goto cleanup;
1608
1609 tmp->thread = ct->task;
1610 list_add(&tmp->list, &thread_list);
1611 }
1612
1613 list_for_each(t, &thread_list) {
1614 struct elf_thread_status *tmp;
1615 int sz;
1616
1617 tmp = list_entry(t, struct elf_thread_status, list);
1618 sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1619 thread_status_size += sz;
1620 }
1621
1622 /* now collect the dump for the current */
1623 fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1624 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1625
1626 segs = current->mm->map_count;
1627 segs += elf_core_extra_phdrs();
1628
1629 /* for notes section */
1630 segs++;
1631
1632 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1633 * this, kernel supports extended numbering. Have a look at
1634 * include/linux/elf.h for further information. */
1635 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1636
1637 /* Set up header */
1638 fill_elf_fdpic_header(elf, e_phnum);
1639
1640 has_dumped = 1;
1641 /*
1642 * Set up the notes in similar form to SVR4 core dumps made
1643 * with info from their /proc.
1644 */
1645
1646 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1647 fill_psinfo(psinfo, current->group_leader, current->mm);
1648 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1649
1650 numnote = 2;
1651
1652 auxv = (elf_addr_t *) current->mm->saved_auxv;
1653
1654 i = 0;
1655 do
1656 i += 2;
1657 while (auxv[i - 2] != AT_NULL);
1658 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1659 i * sizeof(elf_addr_t), auxv);
1660
1661 /* Try to dump the FPU. */
1662 if ((prstatus->pr_fpvalid =
1663 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1664 fill_note(notes + numnote++,
1665 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1666 #ifdef ELF_CORE_COPY_XFPREGS
1667 if (elf_core_copy_task_xfpregs(current, xfpu))
1668 fill_note(notes + numnote++,
1669 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1670 #endif
1671
1672 fs = get_fs();
1673 set_fs(KERNEL_DS);
1674
1675 offset += sizeof(*elf); /* Elf header */
1676 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1677
1678 /* Write notes phdr entry */
1679 {
1680 int sz = 0;
1681
1682 for (i = 0; i < numnote; i++)
1683 sz += notesize(notes + i);
1684
1685 sz += thread_status_size;
1686
1687 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1688 if (!phdr4note)
1689 goto end_coredump;
1690
1691 fill_elf_note_phdr(phdr4note, sz, offset);
1692 offset += sz;
1693 }
1694
1695 /* Page-align dumped data */
1696 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1697
1698 offset += elf_core_vma_data_size(cprm->mm_flags);
1699 offset += elf_core_extra_data_size();
1700 e_shoff = offset;
1701
1702 if (e_phnum == PN_XNUM) {
1703 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1704 if (!shdr4extnum)
1705 goto end_coredump;
1706 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1707 }
1708
1709 offset = dataoff;
1710
1711 if (!dump_emit(cprm, elf, sizeof(*elf)))
1712 goto end_coredump;
1713
1714 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1715 goto end_coredump;
1716
1717 /* write program headers for segments dump */
1718 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1719 struct elf_phdr phdr;
1720 size_t sz;
1721
1722 sz = vma->vm_end - vma->vm_start;
1723
1724 phdr.p_type = PT_LOAD;
1725 phdr.p_offset = offset;
1726 phdr.p_vaddr = vma->vm_start;
1727 phdr.p_paddr = 0;
1728 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1729 phdr.p_memsz = sz;
1730 offset += phdr.p_filesz;
1731 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1732 if (vma->vm_flags & VM_WRITE)
1733 phdr.p_flags |= PF_W;
1734 if (vma->vm_flags & VM_EXEC)
1735 phdr.p_flags |= PF_X;
1736 phdr.p_align = ELF_EXEC_PAGESIZE;
1737
1738 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1739 goto end_coredump;
1740 }
1741
1742 if (!elf_core_write_extra_phdrs(cprm, offset))
1743 goto end_coredump;
1744
1745 /* write out the notes section */
1746 for (i = 0; i < numnote; i++)
1747 if (!writenote(notes + i, cprm))
1748 goto end_coredump;
1749
1750 /* write out the thread status notes section */
1751 list_for_each(t, &thread_list) {
1752 struct elf_thread_status *tmp =
1753 list_entry(t, struct elf_thread_status, list);
1754
1755 for (i = 0; i < tmp->num_notes; i++)
1756 if (!writenote(&tmp->notes[i], cprm))
1757 goto end_coredump;
1758 }
1759
1760 if (!dump_skip(cprm, dataoff - cprm->pos))
1761 goto end_coredump;
1762
1763 if (!elf_fdpic_dump_segments(cprm))
1764 goto end_coredump;
1765
1766 if (!elf_core_write_extra_data(cprm))
1767 goto end_coredump;
1768
1769 if (e_phnum == PN_XNUM) {
1770 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1771 goto end_coredump;
1772 }
1773
1774 if (cprm->file->f_pos != offset) {
1775 /* Sanity check */
1776 printk(KERN_WARNING
1777 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1778 cprm->file->f_pos, offset);
1779 }
1780
1781 end_coredump:
1782 set_fs(fs);
1783
1784 cleanup:
1785 while (!list_empty(&thread_list)) {
1786 struct list_head *tmp = thread_list.next;
1787 list_del(tmp);
1788 kfree(list_entry(tmp, struct elf_thread_status, list));
1789 }
1790 kfree(phdr4note);
1791 kfree(elf);
1792 kfree(prstatus);
1793 kfree(psinfo);
1794 kfree(notes);
1795 kfree(fpu);
1796 kfree(shdr4extnum);
1797 #ifdef ELF_CORE_COPY_XFPREGS
1798 kfree(xfpu);
1799 #endif
1800 return has_dumped;
1801 #undef NUM_NOTES
1802 }
1803
1804 #endif /* CONFIG_ELF_CORE */
Arne's tree of linux kernel.