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/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/param.h>
40 #include <asm/pgalloc.h>
41
42 typedef char *elf_caddr_t;
43
44 #if 0
45 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
46 #else
47 #define kdebug(fmt, ...) do {} while(0)
48 #endif
49
50 #if 0
51 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
52 #else
53 #define kdcore(fmt, ...) do {} while(0)
54 #endif
55
56 MODULE_LICENSE("GPL");
57
58 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
59 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
60 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
61 struct mm_struct *, const char *);
62
63 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
64 struct elf_fdpic_params *,
65 struct elf_fdpic_params *);
66
67 #ifndef CONFIG_MMU
68 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
69 unsigned long *);
70 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
71 struct file *,
72 struct mm_struct *);
73 #endif
74
75 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
76 struct file *, struct mm_struct *);
77
78 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
79 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *, unsigned long limit);
80 #endif
81
82 static struct linux_binfmt elf_fdpic_format = {
83 .module = THIS_MODULE,
84 .load_binary = load_elf_fdpic_binary,
85 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
86 .core_dump = elf_fdpic_core_dump,
87 #endif
88 .min_coredump = ELF_EXEC_PAGESIZE,
89 };
90
init_elf_fdpic_binfmt(void)91 static int __init init_elf_fdpic_binfmt(void)
92 {
93 return register_binfmt(&elf_fdpic_format);
94 }
95
exit_elf_fdpic_binfmt(void)96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 unregister_binfmt(&elf_fdpic_format);
99 }
100
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103
is_elf_fdpic(struct elfhdr * hdr,struct file * file)104 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
105 {
106 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 return 0;
108 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 return 0;
110 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
111 return 0;
112 if (!file->f_op || !file->f_op->mmap)
113 return 0;
114 return 1;
115 }
116
117 /*****************************************************************************/
118 /*
119 * read the program headers table into memory
120 */
elf_fdpic_fetch_phdrs(struct elf_fdpic_params * params,struct file * file)121 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
122 struct file *file)
123 {
124 struct elf32_phdr *phdr;
125 unsigned long size;
126 int retval, loop;
127
128 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
129 return -ENOMEM;
130 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
131 return -ENOMEM;
132
133 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
134 params->phdrs = kmalloc(size, GFP_KERNEL);
135 if (!params->phdrs)
136 return -ENOMEM;
137
138 retval = kernel_read(file, params->hdr.e_phoff,
139 (char *) params->phdrs, size);
140 if (unlikely(retval != size))
141 return retval < 0 ? retval : -ENOEXEC;
142
143 /* determine stack size for this binary */
144 phdr = params->phdrs;
145 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
146 if (phdr->p_type != PT_GNU_STACK)
147 continue;
148
149 if (phdr->p_flags & PF_X)
150 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
151 else
152 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
153
154 params->stack_size = phdr->p_memsz;
155 break;
156 }
157
158 return 0;
159 }
160
161 /*****************************************************************************/
162 /*
163 * load an fdpic binary into various bits of memory
164 */
load_elf_fdpic_binary(struct linux_binprm * bprm,struct pt_regs * regs)165 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
166 struct pt_regs *regs)
167 {
168 struct elf_fdpic_params exec_params, interp_params;
169 struct elf_phdr *phdr;
170 unsigned long stack_size, entryaddr;
171 #ifdef ELF_FDPIC_PLAT_INIT
172 unsigned long dynaddr;
173 #endif
174 struct file *interpreter = NULL; /* to shut gcc up */
175 char *interpreter_name = NULL;
176 int executable_stack;
177 int retval, i;
178
179 kdebug("____ LOAD %d ____", current->pid);
180
181 memset(&exec_params, 0, sizeof(exec_params));
182 memset(&interp_params, 0, sizeof(interp_params));
183
184 exec_params.hdr = *(struct elfhdr *) bprm->buf;
185 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
186
187 /* check that this is a binary we know how to deal with */
188 retval = -ENOEXEC;
189 if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
190 goto error;
191
192 /* read the program header table */
193 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
194 if (retval < 0)
195 goto error;
196
197 /* scan for a program header that specifies an interpreter */
198 phdr = exec_params.phdrs;
199
200 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
201 switch (phdr->p_type) {
202 case PT_INTERP:
203 retval = -ENOMEM;
204 if (phdr->p_filesz > PATH_MAX)
205 goto error;
206 retval = -ENOENT;
207 if (phdr->p_filesz < 2)
208 goto error;
209
210 /* read the name of the interpreter into memory */
211 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
212 if (!interpreter_name)
213 goto error;
214
215 retval = kernel_read(bprm->file,
216 phdr->p_offset,
217 interpreter_name,
218 phdr->p_filesz);
219 if (unlikely(retval != phdr->p_filesz)) {
220 if (retval >= 0)
221 retval = -ENOEXEC;
222 goto error;
223 }
224
225 retval = -ENOENT;
226 if (interpreter_name[phdr->p_filesz - 1] != '\0')
227 goto error;
228
229 kdebug("Using ELF interpreter %s", interpreter_name);
230
231 /* replace the program with the interpreter */
232 interpreter = open_exec(interpreter_name);
233 retval = PTR_ERR(interpreter);
234 if (IS_ERR(interpreter)) {
235 interpreter = NULL;
236 goto error;
237 }
238
239 /*
240 * If the binary is not readable then enforce
241 * mm->dumpable = 0 regardless of the interpreter's
242 * permissions.
243 */
244 if (file_permission(interpreter, MAY_READ) < 0)
245 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
246
247 retval = kernel_read(interpreter, 0, bprm->buf,
248 BINPRM_BUF_SIZE);
249 if (unlikely(retval != BINPRM_BUF_SIZE)) {
250 if (retval >= 0)
251 retval = -ENOEXEC;
252 goto error;
253 }
254
255 interp_params.hdr = *((struct elfhdr *) bprm->buf);
256 break;
257
258 case PT_LOAD:
259 #ifdef CONFIG_MMU
260 if (exec_params.load_addr == 0)
261 exec_params.load_addr = phdr->p_vaddr;
262 #endif
263 break;
264 }
265
266 }
267
268 if (elf_check_const_displacement(&exec_params.hdr))
269 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
270
271 /* perform insanity checks on the interpreter */
272 if (interpreter_name) {
273 retval = -ELIBBAD;
274 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
275 goto error;
276
277 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
278
279 /* read the interpreter's program header table */
280 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
281 if (retval < 0)
282 goto error;
283 }
284
285 stack_size = exec_params.stack_size;
286 if (stack_size < interp_params.stack_size)
287 stack_size = interp_params.stack_size;
288
289 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
290 executable_stack = EXSTACK_ENABLE_X;
291 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
292 executable_stack = EXSTACK_DISABLE_X;
293 else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
294 executable_stack = EXSTACK_ENABLE_X;
295 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
296 executable_stack = EXSTACK_DISABLE_X;
297 else
298 executable_stack = EXSTACK_DEFAULT;
299
300 retval = -ENOEXEC;
301 if (stack_size == 0)
302 goto error;
303
304 if (elf_check_const_displacement(&interp_params.hdr))
305 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
306
307 /* flush all traces of the currently running executable */
308 retval = flush_old_exec(bprm);
309 if (retval)
310 goto error;
311
312 /* there's now no turning back... the old userspace image is dead,
313 * defunct, deceased, etc. after this point we have to exit via
314 * error_kill */
315 set_personality(PER_LINUX_FDPIC);
316 set_binfmt(&elf_fdpic_format);
317
318 current->mm->start_code = 0;
319 current->mm->end_code = 0;
320 current->mm->start_stack = 0;
321 current->mm->start_data = 0;
322 current->mm->end_data = 0;
323 current->mm->context.exec_fdpic_loadmap = 0;
324 current->mm->context.interp_fdpic_loadmap = 0;
325
326 current->flags &= ~PF_FORKNOEXEC;
327
328 #ifdef CONFIG_MMU
329 elf_fdpic_arch_lay_out_mm(&exec_params,
330 &interp_params,
331 ¤t->mm->start_stack,
332 ¤t->mm->start_brk);
333
334 retval = setup_arg_pages(bprm, current->mm->start_stack,
335 executable_stack);
336 if (retval < 0) {
337 send_sig(SIGKILL, current, 0);
338 goto error_kill;
339 }
340 #endif
341
342 /* load the executable and interpreter into memory */
343 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
344 "executable");
345 if (retval < 0)
346 goto error_kill;
347
348 if (interpreter_name) {
349 retval = elf_fdpic_map_file(&interp_params, interpreter,
350 current->mm, "interpreter");
351 if (retval < 0) {
352 printk(KERN_ERR "Unable to load interpreter\n");
353 goto error_kill;
354 }
355
356 allow_write_access(interpreter);
357 fput(interpreter);
358 interpreter = NULL;
359 }
360
361 #ifdef CONFIG_MMU
362 if (!current->mm->start_brk)
363 current->mm->start_brk = current->mm->end_data;
364
365 current->mm->brk = current->mm->start_brk =
366 PAGE_ALIGN(current->mm->start_brk);
367
368 #else
369 /* create a stack and brk area big enough for everyone
370 * - the brk heap starts at the bottom and works up
371 * - the stack starts at the top and works down
372 */
373 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
374 if (stack_size < PAGE_SIZE * 2)
375 stack_size = PAGE_SIZE * 2;
376
377 down_write(¤t->mm->mmap_sem);
378 current->mm->start_brk = do_mmap(NULL, 0, stack_size,
379 PROT_READ | PROT_WRITE | PROT_EXEC,
380 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
381 0);
382
383 if (IS_ERR_VALUE(current->mm->start_brk)) {
384 up_write(¤t->mm->mmap_sem);
385 retval = current->mm->start_brk;
386 current->mm->start_brk = 0;
387 goto error_kill;
388 }
389
390 up_write(¤t->mm->mmap_sem);
391
392 current->mm->brk = current->mm->start_brk;
393 current->mm->context.end_brk = current->mm->start_brk;
394 current->mm->context.end_brk +=
395 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
396 current->mm->start_stack = current->mm->start_brk + stack_size;
397 #endif
398
399 install_exec_creds(bprm);
400 current->flags &= ~PF_FORKNOEXEC;
401 if (create_elf_fdpic_tables(bprm, current->mm,
402 &exec_params, &interp_params) < 0)
403 goto error_kill;
404
405 kdebug("- start_code %lx", current->mm->start_code);
406 kdebug("- end_code %lx", current->mm->end_code);
407 kdebug("- start_data %lx", current->mm->start_data);
408 kdebug("- end_data %lx", current->mm->end_data);
409 kdebug("- start_brk %lx", current->mm->start_brk);
410 kdebug("- brk %lx", current->mm->brk);
411 kdebug("- start_stack %lx", current->mm->start_stack);
412
413 #ifdef ELF_FDPIC_PLAT_INIT
414 /*
415 * The ABI may specify that certain registers be set up in special
416 * ways (on i386 %edx is the address of a DT_FINI function, for
417 * example. This macro performs whatever initialization to
418 * the regs structure is required.
419 */
420 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
421 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
422 dynaddr);
423 #endif
424
425 /* everything is now ready... get the userspace context ready to roll */
426 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
427 start_thread(regs, entryaddr, current->mm->start_stack);
428
429 retval = 0;
430
431 error:
432 if (interpreter) {
433 allow_write_access(interpreter);
434 fput(interpreter);
435 }
436 kfree(interpreter_name);
437 kfree(exec_params.phdrs);
438 kfree(exec_params.loadmap);
439 kfree(interp_params.phdrs);
440 kfree(interp_params.loadmap);
441 return retval;
442
443 /* unrecoverable error - kill the process */
444 error_kill:
445 send_sig(SIGSEGV, current, 0);
446 goto error;
447
448 }
449
450 /*****************************************************************************/
451
452 #ifndef ELF_BASE_PLATFORM
453 /*
454 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
455 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
456 * will be copied to the user stack in the same manner as AT_PLATFORM.
457 */
458 #define ELF_BASE_PLATFORM NULL
459 #endif
460
461 /*
462 * present useful information to the program by shovelling it onto the new
463 * process's stack
464 */
create_elf_fdpic_tables(struct linux_binprm * bprm,struct mm_struct * mm,struct elf_fdpic_params * exec_params,struct elf_fdpic_params * interp_params)465 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
466 struct mm_struct *mm,
467 struct elf_fdpic_params *exec_params,
468 struct elf_fdpic_params *interp_params)
469 {
470 const struct cred *cred = current_cred();
471 unsigned long sp, csp, nitems;
472 elf_caddr_t __user *argv, *envp;
473 size_t platform_len = 0, len;
474 char *k_platform, *k_base_platform;
475 char __user *u_platform, *u_base_platform, *p;
476 long hwcap;
477 int loop;
478 int nr; /* reset for each csp adjustment */
479
480 #ifdef CONFIG_MMU
481 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
482 * by the processes running on the same package. One thing we can do is
483 * to shuffle the initial stack for them, so we give the architecture
484 * an opportunity to do so here.
485 */
486 sp = arch_align_stack(bprm->p);
487 #else
488 sp = mm->start_stack;
489
490 /* stack the program arguments and environment */
491 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
492 return -EFAULT;
493 #endif
494
495 hwcap = ELF_HWCAP;
496
497 /*
498 * If this architecture has a platform capability string, copy it
499 * to userspace. In some cases (Sparc), this info is impossible
500 * for userspace to get any other way, in others (i386) it is
501 * merely difficult.
502 */
503 k_platform = ELF_PLATFORM;
504 u_platform = NULL;
505
506 if (k_platform) {
507 platform_len = strlen(k_platform) + 1;
508 sp -= platform_len;
509 u_platform = (char __user *) sp;
510 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
511 return -EFAULT;
512 }
513
514 /*
515 * If this architecture has a "base" platform capability
516 * string, copy it to userspace.
517 */
518 k_base_platform = ELF_BASE_PLATFORM;
519 u_base_platform = NULL;
520
521 if (k_base_platform) {
522 platform_len = strlen(k_base_platform) + 1;
523 sp -= platform_len;
524 u_base_platform = (char __user *) sp;
525 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
526 return -EFAULT;
527 }
528
529 sp &= ~7UL;
530
531 /* stack the load map(s) */
532 len = sizeof(struct elf32_fdpic_loadmap);
533 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
534 sp = (sp - len) & ~7UL;
535 exec_params->map_addr = sp;
536
537 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
538 return -EFAULT;
539
540 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
541
542 if (interp_params->loadmap) {
543 len = sizeof(struct elf32_fdpic_loadmap);
544 len += sizeof(struct elf32_fdpic_loadseg) *
545 interp_params->loadmap->nsegs;
546 sp = (sp - len) & ~7UL;
547 interp_params->map_addr = sp;
548
549 if (copy_to_user((void __user *) sp, interp_params->loadmap,
550 len) != 0)
551 return -EFAULT;
552
553 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
554 }
555
556 /* force 16 byte _final_ alignment here for generality */
557 #define DLINFO_ITEMS 15
558
559 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
560 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
561
562 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
563 nitems++;
564
565 csp = sp;
566 sp -= nitems * 2 * sizeof(unsigned long);
567 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
568 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
569 sp -= 1 * sizeof(unsigned long); /* argc */
570
571 csp -= sp & 15UL;
572 sp -= sp & 15UL;
573
574 /* put the ELF interpreter info on the stack */
575 #define NEW_AUX_ENT(id, val) \
576 do { \
577 struct { unsigned long _id, _val; } __user *ent; \
578 \
579 ent = (void __user *) csp; \
580 __put_user((id), &ent[nr]._id); \
581 __put_user((val), &ent[nr]._val); \
582 nr++; \
583 } while (0)
584
585 nr = 0;
586 csp -= 2 * sizeof(unsigned long);
587 NEW_AUX_ENT(AT_NULL, 0);
588 if (k_platform) {
589 nr = 0;
590 csp -= 2 * sizeof(unsigned long);
591 NEW_AUX_ENT(AT_PLATFORM,
592 (elf_addr_t) (unsigned long) u_platform);
593 }
594
595 if (k_base_platform) {
596 nr = 0;
597 csp -= 2 * sizeof(unsigned long);
598 NEW_AUX_ENT(AT_BASE_PLATFORM,
599 (elf_addr_t) (unsigned long) u_base_platform);
600 }
601
602 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
603 nr = 0;
604 csp -= 2 * sizeof(unsigned long);
605 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
606 }
607
608 nr = 0;
609 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
610 NEW_AUX_ENT(AT_HWCAP, hwcap);
611 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
612 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
613 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
614 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
615 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
616 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
617 NEW_AUX_ENT(AT_FLAGS, 0);
618 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
619 NEW_AUX_ENT(AT_UID, (elf_addr_t) cred->uid);
620 NEW_AUX_ENT(AT_EUID, (elf_addr_t) cred->euid);
621 NEW_AUX_ENT(AT_GID, (elf_addr_t) cred->gid);
622 NEW_AUX_ENT(AT_EGID, (elf_addr_t) cred->egid);
623 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
624 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
625
626 #ifdef ARCH_DLINFO
627 nr = 0;
628 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
629
630 /* ARCH_DLINFO must come last so platform specific code can enforce
631 * special alignment requirements on the AUXV if necessary (eg. PPC).
632 */
633 ARCH_DLINFO;
634 #endif
635 #undef NEW_AUX_ENT
636
637 /* allocate room for argv[] and envv[] */
638 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
639 envp = (elf_caddr_t __user *) csp;
640 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
641 argv = (elf_caddr_t __user *) csp;
642
643 /* stack argc */
644 csp -= sizeof(unsigned long);
645 __put_user(bprm->argc, (unsigned long __user *) csp);
646
647 BUG_ON(csp != sp);
648
649 /* fill in the argv[] array */
650 #ifdef CONFIG_MMU
651 current->mm->arg_start = bprm->p;
652 #else
653 current->mm->arg_start = current->mm->start_stack -
654 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
655 #endif
656
657 p = (char __user *) current->mm->arg_start;
658 for (loop = bprm->argc; loop > 0; loop--) {
659 __put_user((elf_caddr_t) p, argv++);
660 len = strnlen_user(p, MAX_ARG_STRLEN);
661 if (!len || len > MAX_ARG_STRLEN)
662 return -EINVAL;
663 p += len;
664 }
665 __put_user(NULL, argv);
666 current->mm->arg_end = (unsigned long) p;
667
668 /* fill in the envv[] array */
669 current->mm->env_start = (unsigned long) p;
670 for (loop = bprm->envc; loop > 0; loop--) {
671 __put_user((elf_caddr_t)(unsigned long) p, envp++);
672 len = strnlen_user(p, MAX_ARG_STRLEN);
673 if (!len || len > MAX_ARG_STRLEN)
674 return -EINVAL;
675 p += len;
676 }
677 __put_user(NULL, envp);
678 current->mm->env_end = (unsigned long) p;
679
680 mm->start_stack = (unsigned long) sp;
681 return 0;
682 }
683
684 /*****************************************************************************/
685 /*
686 * transfer the program arguments and environment from the holding pages onto
687 * the stack
688 */
689 #ifndef CONFIG_MMU
elf_fdpic_transfer_args_to_stack(struct linux_binprm * bprm,unsigned long * _sp)690 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
691 unsigned long *_sp)
692 {
693 unsigned long index, stop, sp;
694 char *src;
695 int ret = 0;
696
697 stop = bprm->p >> PAGE_SHIFT;
698 sp = *_sp;
699
700 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
701 src = kmap(bprm->page[index]);
702 sp -= PAGE_SIZE;
703 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
704 ret = -EFAULT;
705 kunmap(bprm->page[index]);
706 if (ret < 0)
707 goto out;
708 }
709
710 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
711
712 out:
713 return ret;
714 }
715 #endif
716
717 /*****************************************************************************/
718 /*
719 * load the appropriate binary image (executable or interpreter) into memory
720 * - we assume no MMU is available
721 * - if no other PIC bits are set in params->hdr->e_flags
722 * - we assume that the LOADable segments in the binary are independently relocatable
723 * - we assume R/O executable segments are shareable
724 * - else
725 * - we assume the loadable parts of the image to require fixed displacement
726 * - the image is not shareable
727 */
elf_fdpic_map_file(struct elf_fdpic_params * params,struct file * file,struct mm_struct * mm,const char * what)728 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
729 struct file *file,
730 struct mm_struct *mm,
731 const char *what)
732 {
733 struct elf32_fdpic_loadmap *loadmap;
734 #ifdef CONFIG_MMU
735 struct elf32_fdpic_loadseg *mseg;
736 #endif
737 struct elf32_fdpic_loadseg *seg;
738 struct elf32_phdr *phdr;
739 unsigned long load_addr, stop;
740 unsigned nloads, tmp;
741 size_t size;
742 int loop, ret;
743
744 /* allocate a load map table */
745 nloads = 0;
746 for (loop = 0; loop < params->hdr.e_phnum; loop++)
747 if (params->phdrs[loop].p_type == PT_LOAD)
748 nloads++;
749
750 if (nloads == 0)
751 return -ELIBBAD;
752
753 size = sizeof(*loadmap) + nloads * sizeof(*seg);
754 loadmap = kzalloc(size, GFP_KERNEL);
755 if (!loadmap)
756 return -ENOMEM;
757
758 params->loadmap = loadmap;
759
760 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
761 loadmap->nsegs = nloads;
762
763 load_addr = params->load_addr;
764 seg = loadmap->segs;
765
766 /* map the requested LOADs into the memory space */
767 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
768 case ELF_FDPIC_FLAG_CONSTDISP:
769 case ELF_FDPIC_FLAG_CONTIGUOUS:
770 #ifndef CONFIG_MMU
771 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
772 if (ret < 0)
773 return ret;
774 break;
775 #endif
776 default:
777 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
778 if (ret < 0)
779 return ret;
780 break;
781 }
782
783 /* map the entry point */
784 if (params->hdr.e_entry) {
785 seg = loadmap->segs;
786 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
787 if (params->hdr.e_entry >= seg->p_vaddr &&
788 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
789 params->entry_addr =
790 (params->hdr.e_entry - seg->p_vaddr) +
791 seg->addr;
792 break;
793 }
794 }
795 }
796
797 /* determine where the program header table has wound up if mapped */
798 stop = params->hdr.e_phoff;
799 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
800 phdr = params->phdrs;
801
802 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
803 if (phdr->p_type != PT_LOAD)
804 continue;
805
806 if (phdr->p_offset > params->hdr.e_phoff ||
807 phdr->p_offset + phdr->p_filesz < stop)
808 continue;
809
810 seg = loadmap->segs;
811 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
812 if (phdr->p_vaddr >= seg->p_vaddr &&
813 phdr->p_vaddr + phdr->p_filesz <=
814 seg->p_vaddr + seg->p_memsz) {
815 params->ph_addr =
816 (phdr->p_vaddr - seg->p_vaddr) +
817 seg->addr +
818 params->hdr.e_phoff - phdr->p_offset;
819 break;
820 }
821 }
822 break;
823 }
824
825 /* determine where the dynamic section has wound up if there is one */
826 phdr = params->phdrs;
827 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
828 if (phdr->p_type != PT_DYNAMIC)
829 continue;
830
831 seg = loadmap->segs;
832 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
833 if (phdr->p_vaddr >= seg->p_vaddr &&
834 phdr->p_vaddr + phdr->p_memsz <=
835 seg->p_vaddr + seg->p_memsz) {
836 params->dynamic_addr =
837 (phdr->p_vaddr - seg->p_vaddr) +
838 seg->addr;
839
840 /* check the dynamic section contains at least
841 * one item, and that the last item is a NULL
842 * entry */
843 if (phdr->p_memsz == 0 ||
844 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
845 goto dynamic_error;
846
847 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
848 if (((Elf32_Dyn *)
849 params->dynamic_addr)[tmp - 1].d_tag != 0)
850 goto dynamic_error;
851 break;
852 }
853 }
854 break;
855 }
856
857 /* now elide adjacent segments in the load map on MMU linux
858 * - on uClinux the holes between may actually be filled with system
859 * stuff or stuff from other processes
860 */
861 #ifdef CONFIG_MMU
862 nloads = loadmap->nsegs;
863 mseg = loadmap->segs;
864 seg = mseg + 1;
865 for (loop = 1; loop < nloads; loop++) {
866 /* see if we have a candidate for merging */
867 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
868 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
869 if (load_addr == (seg->addr & PAGE_MASK)) {
870 mseg->p_memsz +=
871 load_addr -
872 (mseg->addr + mseg->p_memsz);
873 mseg->p_memsz += seg->addr & ~PAGE_MASK;
874 mseg->p_memsz += seg->p_memsz;
875 loadmap->nsegs--;
876 continue;
877 }
878 }
879
880 mseg++;
881 if (mseg != seg)
882 *mseg = *seg;
883 }
884 #endif
885
886 kdebug("Mapped Object [%s]:", what);
887 kdebug("- elfhdr : %lx", params->elfhdr_addr);
888 kdebug("- entry : %lx", params->entry_addr);
889 kdebug("- PHDR[] : %lx", params->ph_addr);
890 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
891 seg = loadmap->segs;
892 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
893 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
894 loop,
895 seg->addr, seg->addr + seg->p_memsz - 1,
896 seg->p_vaddr, seg->p_memsz);
897
898 return 0;
899
900 dynamic_error:
901 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
902 what, file->f_path.dentry->d_inode->i_ino);
903 return -ELIBBAD;
904 }
905
906 /*****************************************************************************/
907 /*
908 * map a file with constant displacement under uClinux
909 */
910 #ifndef CONFIG_MMU
elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params * params,struct file * file,struct mm_struct * mm)911 static int elf_fdpic_map_file_constdisp_on_uclinux(
912 struct elf_fdpic_params *params,
913 struct file *file,
914 struct mm_struct *mm)
915 {
916 struct elf32_fdpic_loadseg *seg;
917 struct elf32_phdr *phdr;
918 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
919 loff_t fpos;
920 int loop, ret;
921
922 load_addr = params->load_addr;
923 seg = params->loadmap->segs;
924
925 /* determine the bounds of the contiguous overall allocation we must
926 * make */
927 phdr = params->phdrs;
928 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
929 if (params->phdrs[loop].p_type != PT_LOAD)
930 continue;
931
932 if (base > phdr->p_vaddr)
933 base = phdr->p_vaddr;
934 if (top < phdr->p_vaddr + phdr->p_memsz)
935 top = phdr->p_vaddr + phdr->p_memsz;
936 }
937
938 /* allocate one big anon block for everything */
939 mflags = MAP_PRIVATE;
940 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
941 mflags |= MAP_EXECUTABLE;
942
943 down_write(&mm->mmap_sem);
944 maddr = do_mmap(NULL, load_addr, top - base,
945 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
946 up_write(&mm->mmap_sem);
947 if (IS_ERR_VALUE(maddr))
948 return (int) maddr;
949
950 if (load_addr != 0)
951 load_addr += PAGE_ALIGN(top - base);
952
953 /* and then load the file segments into it */
954 phdr = params->phdrs;
955 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
956 if (params->phdrs[loop].p_type != PT_LOAD)
957 continue;
958
959 fpos = phdr->p_offset;
960
961 seg->addr = maddr + (phdr->p_vaddr - base);
962 seg->p_vaddr = phdr->p_vaddr;
963 seg->p_memsz = phdr->p_memsz;
964
965 ret = file->f_op->read(file, (void *) seg->addr,
966 phdr->p_filesz, &fpos);
967 if (ret < 0)
968 return ret;
969
970 /* map the ELF header address if in this segment */
971 if (phdr->p_offset == 0)
972 params->elfhdr_addr = seg->addr;
973
974 /* clear any space allocated but not loaded */
975 if (phdr->p_filesz < phdr->p_memsz)
976 clear_user((void *) (seg->addr + phdr->p_filesz),
977 phdr->p_memsz - phdr->p_filesz);
978
979 if (mm) {
980 if (phdr->p_flags & PF_X) {
981 if (!mm->start_code) {
982 mm->start_code = seg->addr;
983 mm->end_code = seg->addr +
984 phdr->p_memsz;
985 }
986 } else if (!mm->start_data) {
987 mm->start_data = seg->addr;
988 #ifndef CONFIG_MMU
989 mm->end_data = seg->addr + phdr->p_memsz;
990 #endif
991 }
992
993 #ifdef CONFIG_MMU
994 if (seg->addr + phdr->p_memsz > mm->end_data)
995 mm->end_data = seg->addr + phdr->p_memsz;
996 #endif
997 }
998
999 seg++;
1000 }
1001
1002 return 0;
1003 }
1004 #endif
1005
1006 /*****************************************************************************/
1007 /*
1008 * map a binary by direct mmap() of the individual PT_LOAD segments
1009 */
elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params * params,struct file * file,struct mm_struct * mm)1010 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1011 struct file *file,
1012 struct mm_struct *mm)
1013 {
1014 struct elf32_fdpic_loadseg *seg;
1015 struct elf32_phdr *phdr;
1016 unsigned long load_addr, delta_vaddr;
1017 int loop, dvset;
1018
1019 load_addr = params->load_addr;
1020 delta_vaddr = 0;
1021 dvset = 0;
1022
1023 seg = params->loadmap->segs;
1024
1025 /* deal with each load segment separately */
1026 phdr = params->phdrs;
1027 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1028 unsigned long maddr, disp, excess, excess1;
1029 int prot = 0, flags;
1030
1031 if (phdr->p_type != PT_LOAD)
1032 continue;
1033
1034 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1035 (unsigned long) phdr->p_vaddr,
1036 (unsigned long) phdr->p_offset,
1037 (unsigned long) phdr->p_filesz,
1038 (unsigned long) phdr->p_memsz);
1039
1040 /* determine the mapping parameters */
1041 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1042 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1043 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1044
1045 flags = MAP_PRIVATE | MAP_DENYWRITE;
1046 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1047 flags |= MAP_EXECUTABLE;
1048
1049 maddr = 0;
1050
1051 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1052 case ELF_FDPIC_FLAG_INDEPENDENT:
1053 /* PT_LOADs are independently locatable */
1054 break;
1055
1056 case ELF_FDPIC_FLAG_HONOURVADDR:
1057 /* the specified virtual address must be honoured */
1058 maddr = phdr->p_vaddr;
1059 flags |= MAP_FIXED;
1060 break;
1061
1062 case ELF_FDPIC_FLAG_CONSTDISP:
1063 /* constant displacement
1064 * - can be mapped anywhere, but must be mapped as a
1065 * unit
1066 */
1067 if (!dvset) {
1068 maddr = load_addr;
1069 delta_vaddr = phdr->p_vaddr;
1070 dvset = 1;
1071 } else {
1072 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1073 flags |= MAP_FIXED;
1074 }
1075 break;
1076
1077 case ELF_FDPIC_FLAG_CONTIGUOUS:
1078 /* contiguity handled later */
1079 break;
1080
1081 default:
1082 BUG();
1083 }
1084
1085 maddr &= PAGE_MASK;
1086
1087 /* create the mapping */
1088 disp = phdr->p_vaddr & ~PAGE_MASK;
1089 down_write(&mm->mmap_sem);
1090 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1091 phdr->p_offset - disp);
1092 up_write(&mm->mmap_sem);
1093
1094 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1095 loop, phdr->p_memsz + disp, prot, flags,
1096 phdr->p_offset - disp, maddr);
1097
1098 if (IS_ERR_VALUE(maddr))
1099 return (int) maddr;
1100
1101 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1102 ELF_FDPIC_FLAG_CONTIGUOUS)
1103 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1104
1105 seg->addr = maddr + disp;
1106 seg->p_vaddr = phdr->p_vaddr;
1107 seg->p_memsz = phdr->p_memsz;
1108
1109 /* map the ELF header address if in this segment */
1110 if (phdr->p_offset == 0)
1111 params->elfhdr_addr = seg->addr;
1112
1113 /* clear the bit between beginning of mapping and beginning of
1114 * PT_LOAD */
1115 if (prot & PROT_WRITE && disp > 0) {
1116 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1117 clear_user((void __user *) maddr, disp);
1118 maddr += disp;
1119 }
1120
1121 /* clear any space allocated but not loaded
1122 * - on uClinux we can just clear the lot
1123 * - on MMU linux we'll get a SIGBUS beyond the last page
1124 * extant in the file
1125 */
1126 excess = phdr->p_memsz - phdr->p_filesz;
1127 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1128
1129 #ifdef CONFIG_MMU
1130 if (excess > excess1) {
1131 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1132 unsigned long xmaddr;
1133
1134 flags |= MAP_FIXED | MAP_ANONYMOUS;
1135 down_write(&mm->mmap_sem);
1136 xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1137 prot, flags, 0);
1138 up_write(&mm->mmap_sem);
1139
1140 kdebug("mmap[%d] <anon>"
1141 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1142 loop, xaddr, excess - excess1, prot, flags,
1143 xmaddr);
1144
1145 if (xmaddr != xaddr)
1146 return -ENOMEM;
1147 }
1148
1149 if (prot & PROT_WRITE && excess1 > 0) {
1150 kdebug("clear[%d] ad=%lx sz=%lx",
1151 loop, maddr + phdr->p_filesz, excess1);
1152 clear_user((void __user *) maddr + phdr->p_filesz,
1153 excess1);
1154 }
1155
1156 #else
1157 if (excess > 0) {
1158 kdebug("clear[%d] ad=%lx sz=%lx",
1159 loop, maddr + phdr->p_filesz, excess);
1160 clear_user((void *) maddr + phdr->p_filesz, excess);
1161 }
1162 #endif
1163
1164 if (mm) {
1165 if (phdr->p_flags & PF_X) {
1166 if (!mm->start_code) {
1167 mm->start_code = maddr;
1168 mm->end_code = maddr + phdr->p_memsz;
1169 }
1170 } else if (!mm->start_data) {
1171 mm->start_data = maddr;
1172 mm->end_data = maddr + phdr->p_memsz;
1173 }
1174 }
1175
1176 seg++;
1177 }
1178
1179 return 0;
1180 }
1181
1182 /*****************************************************************************/
1183 /*
1184 * ELF-FDPIC core dumper
1185 *
1186 * Modelled on fs/exec.c:aout_core_dump()
1187 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1188 *
1189 * Modelled on fs/binfmt_elf.c core dumper
1190 */
1191 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1192
1193 /*
1194 * These are the only things you should do on a core-file: use only these
1195 * functions to write out all the necessary info.
1196 */
dump_write(struct file * file,const void * addr,int nr)1197 static int dump_write(struct file *file, const void *addr, int nr)
1198 {
1199 return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1200 }
1201
dump_seek(struct file * file,loff_t off)1202 static int dump_seek(struct file *file, loff_t off)
1203 {
1204 if (file->f_op->llseek) {
1205 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1206 return 0;
1207 } else {
1208 file->f_pos = off;
1209 }
1210 return 1;
1211 }
1212
1213 /*
1214 * Decide whether a segment is worth dumping; default is yes to be
1215 * sure (missing info is worse than too much; etc).
1216 * Personally I'd include everything, and use the coredump limit...
1217 *
1218 * I think we should skip something. But I am not sure how. H.J.
1219 */
maydump(struct vm_area_struct * vma,unsigned long mm_flags)1220 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1221 {
1222 int dump_ok;
1223
1224 /* Do not dump I/O mapped devices or special mappings */
1225 if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1226 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1227 return 0;
1228 }
1229
1230 /* If we may not read the contents, don't allow us to dump
1231 * them either. "dump_write()" can't handle it anyway.
1232 */
1233 if (!(vma->vm_flags & VM_READ)) {
1234 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1235 return 0;
1236 }
1237
1238 /* By default, dump shared memory if mapped from an anonymous file. */
1239 if (vma->vm_flags & VM_SHARED) {
1240 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1241 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1242 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1243 vma->vm_flags, dump_ok ? "yes" : "no");
1244 return dump_ok;
1245 }
1246
1247 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1248 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1249 vma->vm_flags, dump_ok ? "yes" : "no");
1250 return dump_ok;
1251 }
1252
1253 #ifdef CONFIG_MMU
1254 /* By default, if it hasn't been written to, don't write it out */
1255 if (!vma->anon_vma) {
1256 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1257 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1258 vma->vm_flags, dump_ok ? "yes" : "no");
1259 return dump_ok;
1260 }
1261 #endif
1262
1263 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1264 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1265 dump_ok ? "yes" : "no");
1266 return dump_ok;
1267 }
1268
1269 /* An ELF note in memory */
1270 struct memelfnote
1271 {
1272 const char *name;
1273 int type;
1274 unsigned int datasz;
1275 void *data;
1276 };
1277
notesize(struct memelfnote * en)1278 static int notesize(struct memelfnote *en)
1279 {
1280 int sz;
1281
1282 sz = sizeof(struct elf_note);
1283 sz += roundup(strlen(en->name) + 1, 4);
1284 sz += roundup(en->datasz, 4);
1285
1286 return sz;
1287 }
1288
1289 /* #define DEBUG */
1290
1291 #define DUMP_WRITE(addr, nr) \
1292 do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1293 #define DUMP_SEEK(off) \
1294 do { if (!dump_seek(file, (off))) return 0; } while(0)
1295
writenote(struct memelfnote * men,struct file * file)1296 static int writenote(struct memelfnote *men, struct file *file)
1297 {
1298 struct elf_note en;
1299
1300 en.n_namesz = strlen(men->name) + 1;
1301 en.n_descsz = men->datasz;
1302 en.n_type = men->type;
1303
1304 DUMP_WRITE(&en, sizeof(en));
1305 DUMP_WRITE(men->name, en.n_namesz);
1306 /* XXX - cast from long long to long to avoid need for libgcc.a */
1307 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1308 DUMP_WRITE(men->data, men->datasz);
1309 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1310
1311 return 1;
1312 }
1313 #undef DUMP_WRITE
1314 #undef DUMP_SEEK
1315
1316 #define DUMP_WRITE(addr, nr) \
1317 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1318 goto end_coredump;
1319 #define DUMP_SEEK(off) \
1320 if (!dump_seek(file, (off))) \
1321 goto end_coredump;
1322
fill_elf_fdpic_header(struct elfhdr * elf,int segs)1323 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1324 {
1325 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1326 elf->e_ident[EI_CLASS] = ELF_CLASS;
1327 elf->e_ident[EI_DATA] = ELF_DATA;
1328 elf->e_ident[EI_VERSION] = EV_CURRENT;
1329 elf->e_ident[EI_OSABI] = ELF_OSABI;
1330 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1331
1332 elf->e_type = ET_CORE;
1333 elf->e_machine = ELF_ARCH;
1334 elf->e_version = EV_CURRENT;
1335 elf->e_entry = 0;
1336 elf->e_phoff = sizeof(struct elfhdr);
1337 elf->e_shoff = 0;
1338 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1339 elf->e_ehsize = sizeof(struct elfhdr);
1340 elf->e_phentsize = sizeof(struct elf_phdr);
1341 elf->e_phnum = segs;
1342 elf->e_shentsize = 0;
1343 elf->e_shnum = 0;
1344 elf->e_shstrndx = 0;
1345 return;
1346 }
1347
fill_elf_note_phdr(struct elf_phdr * phdr,int sz,loff_t offset)1348 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1349 {
1350 phdr->p_type = PT_NOTE;
1351 phdr->p_offset = offset;
1352 phdr->p_vaddr = 0;
1353 phdr->p_paddr = 0;
1354 phdr->p_filesz = sz;
1355 phdr->p_memsz = 0;
1356 phdr->p_flags = 0;
1357 phdr->p_align = 0;
1358 return;
1359 }
1360
fill_note(struct memelfnote * note,const char * name,int type,unsigned int sz,void * data)1361 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1362 unsigned int sz, void *data)
1363 {
1364 note->name = name;
1365 note->type = type;
1366 note->datasz = sz;
1367 note->data = data;
1368 return;
1369 }
1370
1371 /*
1372 * fill up all the fields in prstatus from the given task struct, except
1373 * registers which need to be filled up seperately.
1374 */
fill_prstatus(struct elf_prstatus * prstatus,struct task_struct * p,long signr)1375 static void fill_prstatus(struct elf_prstatus *prstatus,
1376 struct task_struct *p, long signr)
1377 {
1378 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1379 prstatus->pr_sigpend = p->pending.signal.sig[0];
1380 prstatus->pr_sighold = p->blocked.sig[0];
1381 prstatus->pr_pid = task_pid_vnr(p);
1382 prstatus->pr_ppid = task_pid_vnr(p->parent);
1383 prstatus->pr_pgrp = task_pgrp_vnr(p);
1384 prstatus->pr_sid = task_session_vnr(p);
1385 if (thread_group_leader(p)) {
1386 struct task_cputime cputime;
1387
1388 /*
1389 * This is the record for the group leader. It shows the
1390 * group-wide total, not its individual thread total.
1391 */
1392 thread_group_cputime(p, &cputime);
1393 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1394 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1395 } else {
1396 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1397 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1398 }
1399 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1400 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1401
1402 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1403 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1404 }
1405
fill_psinfo(struct elf_prpsinfo * psinfo,struct task_struct * p,struct mm_struct * mm)1406 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1407 struct mm_struct *mm)
1408 {
1409 const struct cred *cred;
1410 unsigned int i, len;
1411
1412 /* first copy the parameters from user space */
1413 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1414
1415 len = mm->arg_end - mm->arg_start;
1416 if (len >= ELF_PRARGSZ)
1417 len = ELF_PRARGSZ - 1;
1418 if (copy_from_user(&psinfo->pr_psargs,
1419 (const char __user *) mm->arg_start, len))
1420 return -EFAULT;
1421 for (i = 0; i < len; i++)
1422 if (psinfo->pr_psargs[i] == 0)
1423 psinfo->pr_psargs[i] = ' ';
1424 psinfo->pr_psargs[len] = 0;
1425
1426 psinfo->pr_pid = task_pid_vnr(p);
1427 psinfo->pr_ppid = task_pid_vnr(p->parent);
1428 psinfo->pr_pgrp = task_pgrp_vnr(p);
1429 psinfo->pr_sid = task_session_vnr(p);
1430
1431 i = p->state ? ffz(~p->state) + 1 : 0;
1432 psinfo->pr_state = i;
1433 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1434 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1435 psinfo->pr_nice = task_nice(p);
1436 psinfo->pr_flag = p->flags;
1437 rcu_read_lock();
1438 cred = __task_cred(p);
1439 SET_UID(psinfo->pr_uid, cred->uid);
1440 SET_GID(psinfo->pr_gid, cred->gid);
1441 rcu_read_unlock();
1442 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1443
1444 return 0;
1445 }
1446
1447 /* Here is the structure in which status of each thread is captured. */
1448 struct elf_thread_status
1449 {
1450 struct list_head list;
1451 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1452 elf_fpregset_t fpu; /* NT_PRFPREG */
1453 struct task_struct *thread;
1454 #ifdef ELF_CORE_COPY_XFPREGS
1455 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1456 #endif
1457 struct memelfnote notes[3];
1458 int num_notes;
1459 };
1460
1461 /*
1462 * In order to add the specific thread information for the elf file format,
1463 * we need to keep a linked list of every thread's pr_status and then create
1464 * a single section for them in the final core file.
1465 */
elf_dump_thread_status(long signr,struct elf_thread_status * t)1466 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1467 {
1468 struct task_struct *p = t->thread;
1469 int sz = 0;
1470
1471 t->num_notes = 0;
1472
1473 fill_prstatus(&t->prstatus, p, signr);
1474 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1475
1476 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1477 &t->prstatus);
1478 t->num_notes++;
1479 sz += notesize(&t->notes[0]);
1480
1481 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1482 if (t->prstatus.pr_fpvalid) {
1483 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1484 &t->fpu);
1485 t->num_notes++;
1486 sz += notesize(&t->notes[1]);
1487 }
1488
1489 #ifdef ELF_CORE_COPY_XFPREGS
1490 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1491 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1492 sizeof(t->xfpu), &t->xfpu);
1493 t->num_notes++;
1494 sz += notesize(&t->notes[2]);
1495 }
1496 #endif
1497 return sz;
1498 }
1499
1500 /*
1501 * dump the segments for an MMU process
1502 */
1503 #ifdef CONFIG_MMU
elf_fdpic_dump_segments(struct file * file,size_t * size,unsigned long * limit,unsigned long mm_flags)1504 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1505 unsigned long *limit, unsigned long mm_flags)
1506 {
1507 struct vm_area_struct *vma;
1508
1509 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1510 unsigned long addr;
1511
1512 if (!maydump(vma, mm_flags))
1513 continue;
1514
1515 for (addr = vma->vm_start;
1516 addr < vma->vm_end;
1517 addr += PAGE_SIZE
1518 ) {
1519 struct vm_area_struct *vma;
1520 struct page *page;
1521
1522 if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1523 &page, &vma) <= 0) {
1524 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1525 }
1526 else if (page == ZERO_PAGE(0)) {
1527 page_cache_release(page);
1528 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1529 }
1530 else {
1531 void *kaddr;
1532
1533 flush_cache_page(vma, addr, page_to_pfn(page));
1534 kaddr = kmap(page);
1535 if ((*size += PAGE_SIZE) > *limit ||
1536 !dump_write(file, kaddr, PAGE_SIZE)
1537 ) {
1538 kunmap(page);
1539 page_cache_release(page);
1540 return -EIO;
1541 }
1542 kunmap(page);
1543 page_cache_release(page);
1544 }
1545 }
1546 }
1547
1548 return 0;
1549
1550 end_coredump:
1551 return -EFBIG;
1552 }
1553 #endif
1554
1555 /*
1556 * dump the segments for a NOMMU process
1557 */
1558 #ifndef CONFIG_MMU
elf_fdpic_dump_segments(struct file * file,size_t * size,unsigned long * limit,unsigned long mm_flags)1559 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1560 unsigned long *limit, unsigned long mm_flags)
1561 {
1562 struct vm_area_struct *vma;
1563
1564 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1565 if (!maydump(vma, mm_flags))
1566 continue;
1567
1568 if ((*size += PAGE_SIZE) > *limit)
1569 return -EFBIG;
1570
1571 if (!dump_write(file, (void *) vma->vm_start,
1572 vma->vm_end - vma->vm_start))
1573 return -EIO;
1574 }
1575
1576 return 0;
1577 }
1578 #endif
1579
1580 /*
1581 * Actual dumper
1582 *
1583 * This is a two-pass process; first we find the offsets of the bits,
1584 * and then they are actually written out. If we run out of core limit
1585 * we just truncate.
1586 */
elf_fdpic_core_dump(long signr,struct pt_regs * regs,struct file * file,unsigned long limit)1587 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1588 struct file *file, unsigned long limit)
1589 {
1590 #define NUM_NOTES 6
1591 int has_dumped = 0;
1592 mm_segment_t fs;
1593 int segs;
1594 size_t size = 0;
1595 int i;
1596 struct vm_area_struct *vma;
1597 struct elfhdr *elf = NULL;
1598 loff_t offset = 0, dataoff;
1599 int numnote;
1600 struct memelfnote *notes = NULL;
1601 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1602 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1603 LIST_HEAD(thread_list);
1604 struct list_head *t;
1605 elf_fpregset_t *fpu = NULL;
1606 #ifdef ELF_CORE_COPY_XFPREGS
1607 elf_fpxregset_t *xfpu = NULL;
1608 #endif
1609 int thread_status_size = 0;
1610 elf_addr_t *auxv;
1611 unsigned long mm_flags;
1612
1613 /*
1614 * We no longer stop all VM operations.
1615 *
1616 * This is because those proceses that could possibly change map_count
1617 * or the mmap / vma pages are now blocked in do_exit on current
1618 * finishing this core dump.
1619 *
1620 * Only ptrace can touch these memory addresses, but it doesn't change
1621 * the map_count or the pages allocated. So no possibility of crashing
1622 * exists while dumping the mm->vm_next areas to the core file.
1623 */
1624
1625 /* alloc memory for large data structures: too large to be on stack */
1626 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1627 if (!elf)
1628 goto cleanup;
1629 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1630 if (!prstatus)
1631 goto cleanup;
1632 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1633 if (!psinfo)
1634 goto cleanup;
1635 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1636 if (!notes)
1637 goto cleanup;
1638 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1639 if (!fpu)
1640 goto cleanup;
1641 #ifdef ELF_CORE_COPY_XFPREGS
1642 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1643 if (!xfpu)
1644 goto cleanup;
1645 #endif
1646
1647 if (signr) {
1648 struct core_thread *ct;
1649 struct elf_thread_status *tmp;
1650
1651 for (ct = current->mm->core_state->dumper.next;
1652 ct; ct = ct->next) {
1653 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1654 if (!tmp)
1655 goto cleanup;
1656
1657 tmp->thread = ct->task;
1658 list_add(&tmp->list, &thread_list);
1659 }
1660
1661 list_for_each(t, &thread_list) {
1662 struct elf_thread_status *tmp;
1663 int sz;
1664
1665 tmp = list_entry(t, struct elf_thread_status, list);
1666 sz = elf_dump_thread_status(signr, tmp);
1667 thread_status_size += sz;
1668 }
1669 }
1670
1671 /* now collect the dump for the current */
1672 fill_prstatus(prstatus, current, signr);
1673 elf_core_copy_regs(&prstatus->pr_reg, regs);
1674
1675 segs = current->mm->map_count;
1676 #ifdef ELF_CORE_EXTRA_PHDRS
1677 segs += ELF_CORE_EXTRA_PHDRS;
1678 #endif
1679
1680 /* Set up header */
1681 fill_elf_fdpic_header(elf, segs + 1); /* including notes section */
1682
1683 has_dumped = 1;
1684 current->flags |= PF_DUMPCORE;
1685
1686 /*
1687 * Set up the notes in similar form to SVR4 core dumps made
1688 * with info from their /proc.
1689 */
1690
1691 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1692 fill_psinfo(psinfo, current->group_leader, current->mm);
1693 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1694
1695 numnote = 2;
1696
1697 auxv = (elf_addr_t *) current->mm->saved_auxv;
1698
1699 i = 0;
1700 do
1701 i += 2;
1702 while (auxv[i - 2] != AT_NULL);
1703 fill_note(¬es[numnote++], "CORE", NT_AUXV,
1704 i * sizeof(elf_addr_t), auxv);
1705
1706 /* Try to dump the FPU. */
1707 if ((prstatus->pr_fpvalid =
1708 elf_core_copy_task_fpregs(current, regs, fpu)))
1709 fill_note(notes + numnote++,
1710 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1711 #ifdef ELF_CORE_COPY_XFPREGS
1712 if (elf_core_copy_task_xfpregs(current, xfpu))
1713 fill_note(notes + numnote++,
1714 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1715 #endif
1716
1717 fs = get_fs();
1718 set_fs(KERNEL_DS);
1719
1720 DUMP_WRITE(elf, sizeof(*elf));
1721 offset += sizeof(*elf); /* Elf header */
1722 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers */
1723
1724 /* Write notes phdr entry */
1725 {
1726 struct elf_phdr phdr;
1727 int sz = 0;
1728
1729 for (i = 0; i < numnote; i++)
1730 sz += notesize(notes + i);
1731
1732 sz += thread_status_size;
1733
1734 fill_elf_note_phdr(&phdr, sz, offset);
1735 offset += sz;
1736 DUMP_WRITE(&phdr, sizeof(phdr));
1737 }
1738
1739 /* Page-align dumped data */
1740 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1741
1742 /*
1743 * We must use the same mm->flags while dumping core to avoid
1744 * inconsistency between the program headers and bodies, otherwise an
1745 * unusable core file can be generated.
1746 */
1747 mm_flags = current->mm->flags;
1748
1749 /* write program headers for segments dump */
1750 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1751 struct elf_phdr phdr;
1752 size_t sz;
1753
1754 sz = vma->vm_end - vma->vm_start;
1755
1756 phdr.p_type = PT_LOAD;
1757 phdr.p_offset = offset;
1758 phdr.p_vaddr = vma->vm_start;
1759 phdr.p_paddr = 0;
1760 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1761 phdr.p_memsz = sz;
1762 offset += phdr.p_filesz;
1763 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1764 if (vma->vm_flags & VM_WRITE)
1765 phdr.p_flags |= PF_W;
1766 if (vma->vm_flags & VM_EXEC)
1767 phdr.p_flags |= PF_X;
1768 phdr.p_align = ELF_EXEC_PAGESIZE;
1769
1770 DUMP_WRITE(&phdr, sizeof(phdr));
1771 }
1772
1773 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1774 ELF_CORE_WRITE_EXTRA_PHDRS;
1775 #endif
1776
1777 /* write out the notes section */
1778 for (i = 0; i < numnote; i++)
1779 if (!writenote(notes + i, file))
1780 goto end_coredump;
1781
1782 /* write out the thread status notes section */
1783 list_for_each(t, &thread_list) {
1784 struct elf_thread_status *tmp =
1785 list_entry(t, struct elf_thread_status, list);
1786
1787 for (i = 0; i < tmp->num_notes; i++)
1788 if (!writenote(&tmp->notes[i], file))
1789 goto end_coredump;
1790 }
1791
1792 DUMP_SEEK(dataoff);
1793
1794 if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1795 goto end_coredump;
1796
1797 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1798 ELF_CORE_WRITE_EXTRA_DATA;
1799 #endif
1800
1801 if (file->f_pos != offset) {
1802 /* Sanity check */
1803 printk(KERN_WARNING
1804 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1805 file->f_pos, offset);
1806 }
1807
1808 end_coredump:
1809 set_fs(fs);
1810
1811 cleanup:
1812 while (!list_empty(&thread_list)) {
1813 struct list_head *tmp = thread_list.next;
1814 list_del(tmp);
1815 kfree(list_entry(tmp, struct elf_thread_status, list));
1816 }
1817
1818 kfree(elf);
1819 kfree(prstatus);
1820 kfree(psinfo);
1821 kfree(notes);
1822 kfree(fpu);
1823 #ifdef ELF_CORE_COPY_XFPREGS
1824 kfree(xfpu);
1825 #endif
1826 return has_dumped;
1827 #undef NUM_NOTES
1828 }
1829
1830 #endif /* USE_ELF_CORE_DUMP */
1831