?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
get rid of useless dget_parent() in btrfs rename() and link()
[linux-2.6.git] / fs / binfmt_elf_fdpic.c
blob30745f459fafb97175206b396ab23cf3c5e3246a
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 #include <linux/coredump.h>
38
39 #include <asm/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/pgalloc.h>
42
43 typedef char *elf_caddr_t;
44
45 #if 0
46 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
47 #else
48 #define kdebug(fmt, ...) do {} while(0)
49 #endif
50
51 #if 0
52 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
53 #else
54 #define kdcore(fmt, ...) do {} while(0)
55 #endif
56
57 MODULE_LICENSE("GPL");
58
59 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
60 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
61 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
62 struct mm_struct *, const char *);
63
64 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
65 struct elf_fdpic_params *,
66 struct elf_fdpic_params *);
67
68 #ifndef CONFIG_MMU
69 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
70 unsigned long *);
71 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
72 struct file *,
73 struct mm_struct *);
74 #endif
75
76 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
77 struct file *, struct mm_struct *);
78
79 #ifdef CONFIG_ELF_CORE
80 static int elf_fdpic_core_dump(struct coredump_params *cprm);
81 #endif
82
83 static struct linux_binfmt elf_fdpic_format = {
84 .module = THIS_MODULE,
85 .load_binary = load_elf_fdpic_binary,
86 #ifdef CONFIG_ELF_CORE
87 .core_dump = elf_fdpic_core_dump,
88 #endif
89 .min_coredump = ELF_EXEC_PAGESIZE,
90 };
91
92 static int __init init_elf_fdpic_binfmt(void)
93 {
94 return register_binfmt(&elf_fdpic_format);
95 }
96
97 static void __exit exit_elf_fdpic_binfmt(void)
98 {
99 unregister_binfmt(&elf_fdpic_format);
100 }
101
102 core_initcall(init_elf_fdpic_binfmt);
103 module_exit(exit_elf_fdpic_binfmt);
104
105 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
106 {
107 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
108 return 0;
109 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
110 return 0;
111 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
112 return 0;
113 if (!file->f_op || !file->f_op->mmap)
114 return 0;
115 return 1;
116 }
117
118 /*****************************************************************************/
119 /*
120 * read the program headers table into memory
121 */
122 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
123 struct file *file)
124 {
125 struct elf32_phdr *phdr;
126 unsigned long size;
127 int retval, loop;
128
129 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
130 return -ENOMEM;
131 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
132 return -ENOMEM;
133
134 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
135 params->phdrs = kmalloc(size, GFP_KERNEL);
136 if (!params->phdrs)
137 return -ENOMEM;
138
139 retval = kernel_read(file, params->hdr.e_phoff,
140 (char *) params->phdrs, size);
141 if (unlikely(retval != size))
142 return retval < 0 ? retval : -ENOEXEC;
143
144 /* determine stack size for this binary */
145 phdr = params->phdrs;
146 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
147 if (phdr->p_type != PT_GNU_STACK)
148 continue;
149
150 if (phdr->p_flags & PF_X)
151 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
152 else
153 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
154
155 params->stack_size = phdr->p_memsz;
156 break;
157 }
158
159 return 0;
160 }
161
162 /*****************************************************************************/
163 /*
164 * load an fdpic binary into various bits of memory
165 */
166 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
167 struct pt_regs *regs)
168 {
169 struct elf_fdpic_params exec_params, interp_params;
170 struct elf_phdr *phdr;
171 unsigned long stack_size, entryaddr;
172 #ifdef ELF_FDPIC_PLAT_INIT
173 unsigned long dynaddr;
174 #endif
175 #ifndef CONFIG_MMU
176 unsigned long stack_prot;
177 #endif
178 struct file *interpreter = NULL; /* to shut gcc up */
179 char *interpreter_name = NULL;
180 int executable_stack;
181 int retval, i;
182
183 kdebug("____ LOAD %d ____", current->pid);
184
185 memset(&exec_params, 0, sizeof(exec_params));
186 memset(&interp_params, 0, sizeof(interp_params));
187
188 exec_params.hdr = *(struct elfhdr *) bprm->buf;
189 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
190
191 /* check that this is a binary we know how to deal with */
192 retval = -ENOEXEC;
193 if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
194 goto error;
195
196 /* read the program header table */
197 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
198 if (retval < 0)
199 goto error;
200
201 /* scan for a program header that specifies an interpreter */
202 phdr = exec_params.phdrs;
203
204 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
205 switch (phdr->p_type) {
206 case PT_INTERP:
207 retval = -ENOMEM;
208 if (phdr->p_filesz > PATH_MAX)
209 goto error;
210 retval = -ENOENT;
211 if (phdr->p_filesz < 2)
212 goto error;
213
214 /* read the name of the interpreter into memory */
215 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
216 if (!interpreter_name)
217 goto error;
218
219 retval = kernel_read(bprm->file,
220 phdr->p_offset,
221 interpreter_name,
222 phdr->p_filesz);
223 if (unlikely(retval != phdr->p_filesz)) {
224 if (retval >= 0)
225 retval = -ENOEXEC;
226 goto error;
227 }
228
229 retval = -ENOENT;
230 if (interpreter_name[phdr->p_filesz - 1] != '\0')
231 goto error;
232
233 kdebug("Using ELF interpreter %s", interpreter_name);
234
235 /* replace the program with the interpreter */
236 interpreter = open_exec(interpreter_name);
237 retval = PTR_ERR(interpreter);
238 if (IS_ERR(interpreter)) {
239 interpreter = NULL;
240 goto error;
241 }
242
243 /*
244 * If the binary is not readable then enforce
245 * mm->dumpable = 0 regardless of the interpreter's
246 * permissions.
247 */
248 would_dump(bprm, interpreter);
249
250 retval = kernel_read(interpreter, 0, bprm->buf,
251 BINPRM_BUF_SIZE);
252 if (unlikely(retval != BINPRM_BUF_SIZE)) {
253 if (retval >= 0)
254 retval = -ENOEXEC;
255 goto error;
256 }
257
258 interp_params.hdr = *((struct elfhdr *) bprm->buf);
259 break;
260
261 case PT_LOAD:
262 #ifdef CONFIG_MMU
263 if (exec_params.load_addr == 0)
264 exec_params.load_addr = phdr->p_vaddr;
265 #endif
266 break;
267 }
268
269 }
270
271 if (elf_check_const_displacement(&exec_params.hdr))
272 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
273
274 /* perform insanity checks on the interpreter */
275 if (interpreter_name) {
276 retval = -ELIBBAD;
277 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
278 goto error;
279
280 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
281
282 /* read the interpreter's program header table */
283 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
284 if (retval < 0)
285 goto error;
286 }
287
288 stack_size = exec_params.stack_size;
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
294 executable_stack = EXSTACK_DEFAULT;
295
296 if (stack_size == 0) {
297 stack_size = interp_params.stack_size;
298 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
299 executable_stack = EXSTACK_ENABLE_X;
300 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
301 executable_stack = EXSTACK_DISABLE_X;
302 else
303 executable_stack = EXSTACK_DEFAULT;
304 }
305
306 retval = -ENOEXEC;
307 if (stack_size == 0)
308 goto error;
309
310 if (elf_check_const_displacement(&interp_params.hdr))
311 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
312
313 /* flush all traces of the currently running executable */
314 retval = flush_old_exec(bprm);
315 if (retval)
316 goto error;
317
318 /* there's now no turning back... the old userspace image is dead,
319 * defunct, deceased, etc. after this point we have to exit via
320 * error_kill */
321 set_personality(PER_LINUX_FDPIC);
322 if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
323 current->personality |= READ_IMPLIES_EXEC;
324
325 setup_new_exec(bprm);
326
327 set_binfmt(&elf_fdpic_format);
328
329 current->mm->start_code = 0;
330 current->mm->end_code = 0;
331 current->mm->start_stack = 0;
332 current->mm->start_data = 0;
333 current->mm->end_data = 0;
334 current->mm->context.exec_fdpic_loadmap = 0;
335 current->mm->context.interp_fdpic_loadmap = 0;
336
337 current->flags &= ~PF_FORKNOEXEC;
338
339 #ifdef CONFIG_MMU
340 elf_fdpic_arch_lay_out_mm(&exec_params,
341 &interp_params,
342 &current->mm->start_stack,
343 &current->mm->start_brk);
344
345 retval = setup_arg_pages(bprm, current->mm->start_stack,
346 executable_stack);
347 if (retval < 0) {
348 send_sig(SIGKILL, current, 0);
349 goto error_kill;
350 }
351 #endif
352
353 /* load the executable and interpreter into memory */
354 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
355 "executable");
356 if (retval < 0)
357 goto error_kill;
358
359 if (interpreter_name) {
360 retval = elf_fdpic_map_file(&interp_params, interpreter,
361 current->mm, "interpreter");
362 if (retval < 0) {
363 printk(KERN_ERR "Unable to load interpreter\n");
364 goto error_kill;
365 }
366
367 allow_write_access(interpreter);
368 fput(interpreter);
369 interpreter = NULL;
370 }
371
372 #ifdef CONFIG_MMU
373 if (!current->mm->start_brk)
374 current->mm->start_brk = current->mm->end_data;
375
376 current->mm->brk = current->mm->start_brk =
377 PAGE_ALIGN(current->mm->start_brk);
378
379 #else
380 /* create a stack and brk area big enough for everyone
381 * - the brk heap starts at the bottom and works up
382 * - the stack starts at the top and works down
383 */
384 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
385 if (stack_size < PAGE_SIZE * 2)
386 stack_size = PAGE_SIZE * 2;
387
388 stack_prot = PROT_READ | PROT_WRITE;
389 if (executable_stack == EXSTACK_ENABLE_X ||
390 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
391 stack_prot |= PROT_EXEC;
392
393 down_write(&current->mm->mmap_sem);
394 current->mm->start_brk = do_mmap(NULL, 0, stack_size, stack_prot,
395 MAP_PRIVATE | MAP_ANONYMOUS |
396 MAP_UNINITIALIZED | MAP_GROWSDOWN,
397 0);
398
399 if (IS_ERR_VALUE(current->mm->start_brk)) {
400 up_write(&current->mm->mmap_sem);
401 retval = current->mm->start_brk;
402 current->mm->start_brk = 0;
403 goto error_kill;
404 }
405
406 up_write(&current->mm->mmap_sem);
407
408 current->mm->brk = current->mm->start_brk;
409 current->mm->context.end_brk = current->mm->start_brk;
410 current->mm->context.end_brk +=
411 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
412 current->mm->start_stack = current->mm->start_brk + stack_size;
413 #endif
414
415 install_exec_creds(bprm);
416 current->flags &= ~PF_FORKNOEXEC;
417 if (create_elf_fdpic_tables(bprm, current->mm,
418 &exec_params, &interp_params) < 0)
419 goto error_kill;
420
421 kdebug("- start_code %lx", current->mm->start_code);
422 kdebug("- end_code %lx", current->mm->end_code);
423 kdebug("- start_data %lx", current->mm->start_data);
424 kdebug("- end_data %lx", current->mm->end_data);
425 kdebug("- start_brk %lx", current->mm->start_brk);
426 kdebug("- brk %lx", current->mm->brk);
427 kdebug("- start_stack %lx", current->mm->start_stack);
428
429 #ifdef ELF_FDPIC_PLAT_INIT
430 /*
431 * The ABI may specify that certain registers be set up in special
432 * ways (on i386 %edx is the address of a DT_FINI function, for
433 * example. This macro performs whatever initialization to
434 * the regs structure is required.
435 */
436 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
437 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
438 dynaddr);
439 #endif
440
441 /* everything is now ready... get the userspace context ready to roll */
442 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
443 start_thread(regs, entryaddr, current->mm->start_stack);
444
445 retval = 0;
446
447 error:
448 if (interpreter) {
449 allow_write_access(interpreter);
450 fput(interpreter);
451 }
452 kfree(interpreter_name);
453 kfree(exec_params.phdrs);
454 kfree(exec_params.loadmap);
455 kfree(interp_params.phdrs);
456 kfree(interp_params.loadmap);
457 return retval;
458
459 /* unrecoverable error - kill the process */
460 error_kill:
461 send_sig(SIGSEGV, current, 0);
462 goto error;
463
464 }
465
466 /*****************************************************************************/
467
468 #ifndef ELF_BASE_PLATFORM
469 /*
470 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
471 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
472 * will be copied to the user stack in the same manner as AT_PLATFORM.
473 */
474 #define ELF_BASE_PLATFORM NULL
475 #endif
476
477 /*
478 * present useful information to the program by shovelling it onto the new
479 * process's stack
480 */
481 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
482 struct mm_struct *mm,
483 struct elf_fdpic_params *exec_params,
484 struct elf_fdpic_params *interp_params)
485 {
486 const struct cred *cred = current_cred();
487 unsigned long sp, csp, nitems;
488 elf_caddr_t __user *argv, *envp;
489 size_t platform_len = 0, len;
490 char *k_platform, *k_base_platform;
491 char __user *u_platform, *u_base_platform, *p;
492 long hwcap;
493 int loop;
494 int nr; /* reset for each csp adjustment */
495
496 #ifdef CONFIG_MMU
497 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
498 * by the processes running on the same package. One thing we can do is
499 * to shuffle the initial stack for them, so we give the architecture
500 * an opportunity to do so here.
501 */
502 sp = arch_align_stack(bprm->p);
503 #else
504 sp = mm->start_stack;
505
506 /* stack the program arguments and environment */
507 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
508 return -EFAULT;
509 #endif
510
511 hwcap = ELF_HWCAP;
512
513 /*
514 * If this architecture has a platform capability string, copy it
515 * to userspace. In some cases (Sparc), this info is impossible
516 * for userspace to get any other way, in others (i386) it is
517 * merely difficult.
518 */
519 k_platform = ELF_PLATFORM;
520 u_platform = NULL;
521
522 if (k_platform) {
523 platform_len = strlen(k_platform) + 1;
524 sp -= platform_len;
525 u_platform = (char __user *) sp;
526 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
527 return -EFAULT;
528 }
529
530 /*
531 * If this architecture has a "base" platform capability
532 * string, copy it to userspace.
533 */
534 k_base_platform = ELF_BASE_PLATFORM;
535 u_base_platform = NULL;
536
537 if (k_base_platform) {
538 platform_len = strlen(k_base_platform) + 1;
539 sp -= platform_len;
540 u_base_platform = (char __user *) sp;
541 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
542 return -EFAULT;
543 }
544
545 sp &= ~7UL;
546
547 /* stack the load map(s) */
548 len = sizeof(struct elf32_fdpic_loadmap);
549 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
550 sp = (sp - len) & ~7UL;
551 exec_params->map_addr = sp;
552
553 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
554 return -EFAULT;
555
556 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
557
558 if (interp_params->loadmap) {
559 len = sizeof(struct elf32_fdpic_loadmap);
560 len += sizeof(struct elf32_fdpic_loadseg) *
561 interp_params->loadmap->nsegs;
562 sp = (sp - len) & ~7UL;
563 interp_params->map_addr = sp;
564
565 if (copy_to_user((void __user *) sp, interp_params->loadmap,
566 len) != 0)
567 return -EFAULT;
568
569 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
570 }
571
572 /* force 16 byte _final_ alignment here for generality */
573 #define DLINFO_ITEMS 15
574
575 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
576 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
577
578 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
579 nitems++;
580
581 csp = sp;
582 sp -= nitems * 2 * sizeof(unsigned long);
583 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
584 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
585 sp -= 1 * sizeof(unsigned long); /* argc */
586
587 csp -= sp & 15UL;
588 sp -= sp & 15UL;
589
590 /* put the ELF interpreter info on the stack */
591 #define NEW_AUX_ENT(id, val) \
592 do { \
593 struct { unsigned long _id, _val; } __user *ent; \
594 \
595 ent = (void __user *) csp; \
596 __put_user((id), &ent[nr]._id); \
597 __put_user((val), &ent[nr]._val); \
598 nr++; \
599 } while (0)
600
601 nr = 0;
602 csp -= 2 * sizeof(unsigned long);
603 NEW_AUX_ENT(AT_NULL, 0);
604 if (k_platform) {
605 nr = 0;
606 csp -= 2 * sizeof(unsigned long);
607 NEW_AUX_ENT(AT_PLATFORM,
608 (elf_addr_t) (unsigned long) u_platform);
609 }
610
611 if (k_base_platform) {
612 nr = 0;
613 csp -= 2 * sizeof(unsigned long);
614 NEW_AUX_ENT(AT_BASE_PLATFORM,
615 (elf_addr_t) (unsigned long) u_base_platform);
616 }
617
618 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
619 nr = 0;
620 csp -= 2 * sizeof(unsigned long);
621 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
622 }
623
624 nr = 0;
625 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
626 NEW_AUX_ENT(AT_HWCAP, hwcap);
627 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
628 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
629 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
630 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
631 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
632 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
633 NEW_AUX_ENT(AT_FLAGS, 0);
634 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
635 NEW_AUX_ENT(AT_UID, (elf_addr_t) cred->uid);
636 NEW_AUX_ENT(AT_EUID, (elf_addr_t) cred->euid);
637 NEW_AUX_ENT(AT_GID, (elf_addr_t) cred->gid);
638 NEW_AUX_ENT(AT_EGID, (elf_addr_t) cred->egid);
639 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
640 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
641
642 #ifdef ARCH_DLINFO
643 nr = 0;
644 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
645
646 /* ARCH_DLINFO must come last so platform specific code can enforce
647 * special alignment requirements on the AUXV if necessary (eg. PPC).
648 */
649 ARCH_DLINFO;
650 #endif
651 #undef NEW_AUX_ENT
652
653 /* allocate room for argv[] and envv[] */
654 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
655 envp = (elf_caddr_t __user *) csp;
656 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
657 argv = (elf_caddr_t __user *) csp;
658
659 /* stack argc */
660 csp -= sizeof(unsigned long);
661 __put_user(bprm->argc, (unsigned long __user *) csp);
662
663 BUG_ON(csp != sp);
664
665 /* fill in the argv[] array */
666 #ifdef CONFIG_MMU
667 current->mm->arg_start = bprm->p;
668 #else
669 current->mm->arg_start = current->mm->start_stack -
670 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
671 #endif
672
673 p = (char __user *) current->mm->arg_start;
674 for (loop = bprm->argc; loop > 0; loop--) {
675 __put_user((elf_caddr_t) p, argv++);
676 len = strnlen_user(p, MAX_ARG_STRLEN);
677 if (!len || len > MAX_ARG_STRLEN)
678 return -EINVAL;
679 p += len;
680 }
681 __put_user(NULL, argv);
682 current->mm->arg_end = (unsigned long) p;
683
684 /* fill in the envv[] array */
685 current->mm->env_start = (unsigned long) p;
686 for (loop = bprm->envc; loop > 0; loop--) {
687 __put_user((elf_caddr_t)(unsigned long) p, envp++);
688 len = strnlen_user(p, MAX_ARG_STRLEN);
689 if (!len || len > MAX_ARG_STRLEN)
690 return -EINVAL;
691 p += len;
692 }
693 __put_user(NULL, envp);
694 current->mm->env_end = (unsigned long) p;
695
696 mm->start_stack = (unsigned long) sp;
697 return 0;
698 }
699
700 /*****************************************************************************/
701 /*
702 * transfer the program arguments and environment from the holding pages onto
703 * the stack
704 */
705 #ifndef CONFIG_MMU
706 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
707 unsigned long *_sp)
708 {
709 unsigned long index, stop, sp;
710 char *src;
711 int ret = 0;
712
713 stop = bprm->p >> PAGE_SHIFT;
714 sp = *_sp;
715
716 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
717 src = kmap(bprm->page[index]);
718 sp -= PAGE_SIZE;
719 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
720 ret = -EFAULT;
721 kunmap(bprm->page[index]);
722 if (ret < 0)
723 goto out;
724 }
725
726 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
727
728 out:
729 return ret;
730 }
731 #endif
732
733 /*****************************************************************************/
734 /*
735 * load the appropriate binary image (executable or interpreter) into memory
736 * - we assume no MMU is available
737 * - if no other PIC bits are set in params->hdr->e_flags
738 * - we assume that the LOADable segments in the binary are independently relocatable
739 * - we assume R/O executable segments are shareable
740 * - else
741 * - we assume the loadable parts of the image to require fixed displacement
742 * - the image is not shareable
743 */
744 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
745 struct file *file,
746 struct mm_struct *mm,
747 const char *what)
748 {
749 struct elf32_fdpic_loadmap *loadmap;
750 #ifdef CONFIG_MMU
751 struct elf32_fdpic_loadseg *mseg;
752 #endif
753 struct elf32_fdpic_loadseg *seg;
754 struct elf32_phdr *phdr;
755 unsigned long load_addr, stop;
756 unsigned nloads, tmp;
757 size_t size;
758 int loop, ret;
759
760 /* allocate a load map table */
761 nloads = 0;
762 for (loop = 0; loop < params->hdr.e_phnum; loop++)
763 if (params->phdrs[loop].p_type == PT_LOAD)
764 nloads++;
765
766 if (nloads == 0)
767 return -ELIBBAD;
768
769 size = sizeof(*loadmap) + nloads * sizeof(*seg);
770 loadmap = kzalloc(size, GFP_KERNEL);
771 if (!loadmap)
772 return -ENOMEM;
773
774 params->loadmap = loadmap;
775
776 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
777 loadmap->nsegs = nloads;
778
779 load_addr = params->load_addr;
780 seg = loadmap->segs;
781
782 /* map the requested LOADs into the memory space */
783 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
784 case ELF_FDPIC_FLAG_CONSTDISP:
785 case ELF_FDPIC_FLAG_CONTIGUOUS:
786 #ifndef CONFIG_MMU
787 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
788 if (ret < 0)
789 return ret;
790 break;
791 #endif
792 default:
793 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
794 if (ret < 0)
795 return ret;
796 break;
797 }
798
799 /* map the entry point */
800 if (params->hdr.e_entry) {
801 seg = loadmap->segs;
802 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
803 if (params->hdr.e_entry >= seg->p_vaddr &&
804 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
805 params->entry_addr =
806 (params->hdr.e_entry - seg->p_vaddr) +
807 seg->addr;
808 break;
809 }
810 }
811 }
812
813 /* determine where the program header table has wound up if mapped */
814 stop = params->hdr.e_phoff;
815 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
816 phdr = params->phdrs;
817
818 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
819 if (phdr->p_type != PT_LOAD)
820 continue;
821
822 if (phdr->p_offset > params->hdr.e_phoff ||
823 phdr->p_offset + phdr->p_filesz < stop)
824 continue;
825
826 seg = loadmap->segs;
827 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
828 if (phdr->p_vaddr >= seg->p_vaddr &&
829 phdr->p_vaddr + phdr->p_filesz <=
830 seg->p_vaddr + seg->p_memsz) {
831 params->ph_addr =
832 (phdr->p_vaddr - seg->p_vaddr) +
833 seg->addr +
834 params->hdr.e_phoff - phdr->p_offset;
835 break;
836 }
837 }
838 break;
839 }
840
841 /* determine where the dynamic section has wound up if there is one */
842 phdr = params->phdrs;
843 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
844 if (phdr->p_type != PT_DYNAMIC)
845 continue;
846
847 seg = loadmap->segs;
848 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
849 if (phdr->p_vaddr >= seg->p_vaddr &&
850 phdr->p_vaddr + phdr->p_memsz <=
851 seg->p_vaddr + seg->p_memsz) {
852 params->dynamic_addr =
853 (phdr->p_vaddr - seg->p_vaddr) +
854 seg->addr;
855
856 /* check the dynamic section contains at least
857 * one item, and that the last item is a NULL
858 * entry */
859 if (phdr->p_memsz == 0 ||
860 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
861 goto dynamic_error;
862
863 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
864 if (((Elf32_Dyn *)
865 params->dynamic_addr)[tmp - 1].d_tag != 0)
866 goto dynamic_error;
867 break;
868 }
869 }
870 break;
871 }
872
873 /* now elide adjacent segments in the load map on MMU linux
874 * - on uClinux the holes between may actually be filled with system
875 * stuff or stuff from other processes
876 */
877 #ifdef CONFIG_MMU
878 nloads = loadmap->nsegs;
879 mseg = loadmap->segs;
880 seg = mseg + 1;
881 for (loop = 1; loop < nloads; loop++) {
882 /* see if we have a candidate for merging */
883 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
884 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
885 if (load_addr == (seg->addr & PAGE_MASK)) {
886 mseg->p_memsz +=
887 load_addr -
888 (mseg->addr + mseg->p_memsz);
889 mseg->p_memsz += seg->addr & ~PAGE_MASK;
890 mseg->p_memsz += seg->p_memsz;
891 loadmap->nsegs--;
892 continue;
893 }
894 }
895
896 mseg++;
897 if (mseg != seg)
898 *mseg = *seg;
899 }
900 #endif
901
902 kdebug("Mapped Object [%s]:", what);
903 kdebug("- elfhdr : %lx", params->elfhdr_addr);
904 kdebug("- entry : %lx", params->entry_addr);
905 kdebug("- PHDR[] : %lx", params->ph_addr);
906 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
907 seg = loadmap->segs;
908 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
909 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
910 loop,
911 seg->addr, seg->addr + seg->p_memsz - 1,
912 seg->p_vaddr, seg->p_memsz);
913
914 return 0;
915
916 dynamic_error:
917 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
918 what, file->f_path.dentry->d_inode->i_ino);
919 return -ELIBBAD;
920 }
921
922 /*****************************************************************************/
923 /*
924 * map a file with constant displacement under uClinux
925 */
926 #ifndef CONFIG_MMU
927 static int elf_fdpic_map_file_constdisp_on_uclinux(
928 struct elf_fdpic_params *params,
929 struct file *file,
930 struct mm_struct *mm)
931 {
932 struct elf32_fdpic_loadseg *seg;
933 struct elf32_phdr *phdr;
934 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
935 loff_t fpos;
936 int loop, ret;
937
938 load_addr = params->load_addr;
939 seg = params->loadmap->segs;
940
941 /* determine the bounds of the contiguous overall allocation we must
942 * make */
943 phdr = params->phdrs;
944 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
945 if (params->phdrs[loop].p_type != PT_LOAD)
946 continue;
947
948 if (base > phdr->p_vaddr)
949 base = phdr->p_vaddr;
950 if (top < phdr->p_vaddr + phdr->p_memsz)
951 top = phdr->p_vaddr + phdr->p_memsz;
952 }
953
954 /* allocate one big anon block for everything */
955 mflags = MAP_PRIVATE;
956 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
957 mflags |= MAP_EXECUTABLE;
958
959 down_write(&mm->mmap_sem);
960 maddr = do_mmap(NULL, load_addr, top - base,
961 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
962 up_write(&mm->mmap_sem);
963 if (IS_ERR_VALUE(maddr))
964 return (int) maddr;
965
966 if (load_addr != 0)
967 load_addr += PAGE_ALIGN(top - base);
968
969 /* and then load the file segments into it */
970 phdr = params->phdrs;
971 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
972 if (params->phdrs[loop].p_type != PT_LOAD)
973 continue;
974
975 fpos = phdr->p_offset;
976
977 seg->addr = maddr + (phdr->p_vaddr - base);
978 seg->p_vaddr = phdr->p_vaddr;
979 seg->p_memsz = phdr->p_memsz;
980
981 ret = file->f_op->read(file, (void *) seg->addr,
982 phdr->p_filesz, &fpos);
983 if (ret < 0)
984 return ret;
985
986 /* map the ELF header address if in this segment */
987 if (phdr->p_offset == 0)
988 params->elfhdr_addr = seg->addr;
989
990 /* clear any space allocated but not loaded */
991 if (phdr->p_filesz < phdr->p_memsz) {
992 if (clear_user((void *) (seg->addr + phdr->p_filesz),
993 phdr->p_memsz - phdr->p_filesz))
994 return -EFAULT;
995 }
996
997 if (mm) {
998 if (phdr->p_flags & PF_X) {
999 if (!mm->start_code) {
1000 mm->start_code = seg->addr;
1001 mm->end_code = seg->addr +
1002 phdr->p_memsz;
1003 }
1004 } else if (!mm->start_data) {
1005 mm->start_data = seg->addr;
1006 mm->end_data = seg->addr + phdr->p_memsz;
1007 }
1008 }
1009
1010 seg++;
1011 }
1012
1013 return 0;
1014 }
1015 #endif
1016
1017 /*****************************************************************************/
1018 /*
1019 * map a binary by direct mmap() of the individual PT_LOAD segments
1020 */
1021 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1022 struct file *file,
1023 struct mm_struct *mm)
1024 {
1025 struct elf32_fdpic_loadseg *seg;
1026 struct elf32_phdr *phdr;
1027 unsigned long load_addr, delta_vaddr;
1028 int loop, dvset;
1029
1030 load_addr = params->load_addr;
1031 delta_vaddr = 0;
1032 dvset = 0;
1033
1034 seg = params->loadmap->segs;
1035
1036 /* deal with each load segment separately */
1037 phdr = params->phdrs;
1038 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1039 unsigned long maddr, disp, excess, excess1;
1040 int prot = 0, flags;
1041
1042 if (phdr->p_type != PT_LOAD)
1043 continue;
1044
1045 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1046 (unsigned long) phdr->p_vaddr,
1047 (unsigned long) phdr->p_offset,
1048 (unsigned long) phdr->p_filesz,
1049 (unsigned long) phdr->p_memsz);
1050
1051 /* determine the mapping parameters */
1052 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1053 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1054 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1055
1056 flags = MAP_PRIVATE | MAP_DENYWRITE;
1057 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1058 flags |= MAP_EXECUTABLE;
1059
1060 maddr = 0;
1061
1062 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1063 case ELF_FDPIC_FLAG_INDEPENDENT:
1064 /* PT_LOADs are independently locatable */
1065 break;
1066
1067 case ELF_FDPIC_FLAG_HONOURVADDR:
1068 /* the specified virtual address must be honoured */
1069 maddr = phdr->p_vaddr;
1070 flags |= MAP_FIXED;
1071 break;
1072
1073 case ELF_FDPIC_FLAG_CONSTDISP:
1074 /* constant displacement
1075 * - can be mapped anywhere, but must be mapped as a
1076 * unit
1077 */
1078 if (!dvset) {
1079 maddr = load_addr;
1080 delta_vaddr = phdr->p_vaddr;
1081 dvset = 1;
1082 } else {
1083 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1084 flags |= MAP_FIXED;
1085 }
1086 break;
1087
1088 case ELF_FDPIC_FLAG_CONTIGUOUS:
1089 /* contiguity handled later */
1090 break;
1091
1092 default:
1093 BUG();
1094 }
1095
1096 maddr &= PAGE_MASK;
1097
1098 /* create the mapping */
1099 disp = phdr->p_vaddr & ~PAGE_MASK;
1100 down_write(&mm->mmap_sem);
1101 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1102 phdr->p_offset - disp);
1103 up_write(&mm->mmap_sem);
1104
1105 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1106 loop, phdr->p_memsz + disp, prot, flags,
1107 phdr->p_offset - disp, maddr);
1108
1109 if (IS_ERR_VALUE(maddr))
1110 return (int) maddr;
1111
1112 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1113 ELF_FDPIC_FLAG_CONTIGUOUS)
1114 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1115
1116 seg->addr = maddr + disp;
1117 seg->p_vaddr = phdr->p_vaddr;
1118 seg->p_memsz = phdr->p_memsz;
1119
1120 /* map the ELF header address if in this segment */
1121 if (phdr->p_offset == 0)
1122 params->elfhdr_addr = seg->addr;
1123
1124 /* clear the bit between beginning of mapping and beginning of
1125 * PT_LOAD */
1126 if (prot & PROT_WRITE && disp > 0) {
1127 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1128 if (clear_user((void __user *) maddr, disp))
1129 return -EFAULT;
1130 maddr += disp;
1131 }
1132
1133 /* clear any space allocated but not loaded
1134 * - on uClinux we can just clear the lot
1135 * - on MMU linux we'll get a SIGBUS beyond the last page
1136 * extant in the file
1137 */
1138 excess = phdr->p_memsz - phdr->p_filesz;
1139 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1140
1141 #ifdef CONFIG_MMU
1142 if (excess > excess1) {
1143 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1144 unsigned long xmaddr;
1145
1146 flags |= MAP_FIXED | MAP_ANONYMOUS;
1147 down_write(&mm->mmap_sem);
1148 xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1149 prot, flags, 0);
1150 up_write(&mm->mmap_sem);
1151
1152 kdebug("mmap[%d] <anon>"
1153 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1154 loop, xaddr, excess - excess1, prot, flags,
1155 xmaddr);
1156
1157 if (xmaddr != xaddr)
1158 return -ENOMEM;
1159 }
1160
1161 if (prot & PROT_WRITE && excess1 > 0) {
1162 kdebug("clear[%d] ad=%lx sz=%lx",
1163 loop, maddr + phdr->p_filesz, excess1);
1164 if (clear_user((void __user *) maddr + phdr->p_filesz,
1165 excess1))
1166 return -EFAULT;
1167 }
1168
1169 #else
1170 if (excess > 0) {
1171 kdebug("clear[%d] ad=%lx sz=%lx",
1172 loop, maddr + phdr->p_filesz, excess);
1173 if (clear_user((void *) maddr + phdr->p_filesz, excess))
1174 return -EFAULT;
1175 }
1176 #endif
1177
1178 if (mm) {
1179 if (phdr->p_flags & PF_X) {
1180 if (!mm->start_code) {
1181 mm->start_code = maddr;
1182 mm->end_code = maddr + phdr->p_memsz;
1183 }
1184 } else if (!mm->start_data) {
1185 mm->start_data = maddr;
1186 mm->end_data = maddr + phdr->p_memsz;
1187 }
1188 }
1189
1190 seg++;
1191 }
1192
1193 return 0;
1194 }
1195
1196 /*****************************************************************************/
1197 /*
1198 * ELF-FDPIC core dumper
1199 *
1200 * Modelled on fs/exec.c:aout_core_dump()
1201 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1202 *
1203 * Modelled on fs/binfmt_elf.c core dumper
1204 */
1205 #ifdef CONFIG_ELF_CORE
1206
1207 /*
1208 * Decide whether a segment is worth dumping; default is yes to be
1209 * sure (missing info is worse than too much; etc).
1210 * Personally I'd include everything, and use the coredump limit...
1211 *
1212 * I think we should skip something. But I am not sure how. H.J.
1213 */
1214 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1215 {
1216 int dump_ok;
1217
1218 /* Do not dump I/O mapped devices or special mappings */
1219 if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1220 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1221 return 0;
1222 }
1223
1224 /* If we may not read the contents, don't allow us to dump
1225 * them either. "dump_write()" can't handle it anyway.
1226 */
1227 if (!(vma->vm_flags & VM_READ)) {
1228 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1229 return 0;
1230 }
1231
1232 /* By default, dump shared memory if mapped from an anonymous file. */
1233 if (vma->vm_flags & VM_SHARED) {
1234 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1235 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1236 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1237 vma->vm_flags, dump_ok ? "yes" : "no");
1238 return dump_ok;
1239 }
1240
1241 dump_ok = test_bit(MMF_DUMP_MAPPED_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 #ifdef CONFIG_MMU
1248 /* By default, if it hasn't been written to, don't write it out */
1249 if (!vma->anon_vma) {
1250 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1251 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1252 vma->vm_flags, dump_ok ? "yes" : "no");
1253 return dump_ok;
1254 }
1255 #endif
1256
1257 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1258 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1259 dump_ok ? "yes" : "no");
1260 return dump_ok;
1261 }
1262
1263 /* An ELF note in memory */
1264 struct memelfnote
1265 {
1266 const char *name;
1267 int type;
1268 unsigned int datasz;
1269 void *data;
1270 };
1271
1272 static int notesize(struct memelfnote *en)
1273 {
1274 int sz;
1275
1276 sz = sizeof(struct elf_note);
1277 sz += roundup(strlen(en->name) + 1, 4);
1278 sz += roundup(en->datasz, 4);
1279
1280 return sz;
1281 }
1282
1283 /* #define DEBUG */
1284
1285 #define DUMP_WRITE(addr, nr, foffset) \
1286 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1287
1288 static int alignfile(struct file *file, loff_t *foffset)
1289 {
1290 static const char buf[4] = { 0, };
1291 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1292 return 1;
1293 }
1294
1295 static int writenote(struct memelfnote *men, struct file *file,
1296 loff_t *foffset)
1297 {
1298 struct elf_note en;
1299 en.n_namesz = strlen(men->name) + 1;
1300 en.n_descsz = men->datasz;
1301 en.n_type = men->type;
1302
1303 DUMP_WRITE(&en, sizeof(en), foffset);
1304 DUMP_WRITE(men->name, en.n_namesz, foffset);
1305 if (!alignfile(file, foffset))
1306 return 0;
1307 DUMP_WRITE(men->data, men->datasz, foffset);
1308 if (!alignfile(file, foffset))
1309 return 0;
1310
1311 return 1;
1312 }
1313 #undef DUMP_WRITE
1314
1315 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1316 {
1317 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1318 elf->e_ident[EI_CLASS] = ELF_CLASS;
1319 elf->e_ident[EI_DATA] = ELF_DATA;
1320 elf->e_ident[EI_VERSION] = EV_CURRENT;
1321 elf->e_ident[EI_OSABI] = ELF_OSABI;
1322 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1323
1324 elf->e_type = ET_CORE;
1325 elf->e_machine = ELF_ARCH;
1326 elf->e_version = EV_CURRENT;
1327 elf->e_entry = 0;
1328 elf->e_phoff = sizeof(struct elfhdr);
1329 elf->e_shoff = 0;
1330 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1331 elf->e_ehsize = sizeof(struct elfhdr);
1332 elf->e_phentsize = sizeof(struct elf_phdr);
1333 elf->e_phnum = segs;
1334 elf->e_shentsize = 0;
1335 elf->e_shnum = 0;
1336 elf->e_shstrndx = 0;
1337 return;
1338 }
1339
1340 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1341 {
1342 phdr->p_type = PT_NOTE;
1343 phdr->p_offset = offset;
1344 phdr->p_vaddr = 0;
1345 phdr->p_paddr = 0;
1346 phdr->p_filesz = sz;
1347 phdr->p_memsz = 0;
1348 phdr->p_flags = 0;
1349 phdr->p_align = 0;
1350 return;
1351 }
1352
1353 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1354 unsigned int sz, void *data)
1355 {
1356 note->name = name;
1357 note->type = type;
1358 note->datasz = sz;
1359 note->data = data;
1360 return;
1361 }
1362
1363 /*
1364 * fill up all the fields in prstatus from the given task struct, except
1365 * registers which need to be filled up separately.
1366 */
1367 static void fill_prstatus(struct elf_prstatus *prstatus,
1368 struct task_struct *p, long signr)
1369 {
1370 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1371 prstatus->pr_sigpend = p->pending.signal.sig[0];
1372 prstatus->pr_sighold = p->blocked.sig[0];
1373 rcu_read_lock();
1374 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1375 rcu_read_unlock();
1376 prstatus->pr_pid = task_pid_vnr(p);
1377 prstatus->pr_pgrp = task_pgrp_vnr(p);
1378 prstatus->pr_sid = task_session_vnr(p);
1379 if (thread_group_leader(p)) {
1380 struct task_cputime cputime;
1381
1382 /*
1383 * This is the record for the group leader. It shows the
1384 * group-wide total, not its individual thread total.
1385 */
1386 thread_group_cputime(p, &cputime);
1387 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1388 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1389 } else {
1390 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1391 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1392 }
1393 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1394 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1395
1396 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1397 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1398 }
1399
1400 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1401 struct mm_struct *mm)
1402 {
1403 const struct cred *cred;
1404 unsigned int i, len;
1405
1406 /* first copy the parameters from user space */
1407 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1408
1409 len = mm->arg_end - mm->arg_start;
1410 if (len >= ELF_PRARGSZ)
1411 len = ELF_PRARGSZ - 1;
1412 if (copy_from_user(&psinfo->pr_psargs,
1413 (const char __user *) mm->arg_start, len))
1414 return -EFAULT;
1415 for (i = 0; i < len; i++)
1416 if (psinfo->pr_psargs[i] == 0)
1417 psinfo->pr_psargs[i] = ' ';
1418 psinfo->pr_psargs[len] = 0;
1419
1420 rcu_read_lock();
1421 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1422 rcu_read_unlock();
1423 psinfo->pr_pid = task_pid_vnr(p);
1424 psinfo->pr_pgrp = task_pgrp_vnr(p);
1425 psinfo->pr_sid = task_session_vnr(p);
1426
1427 i = p->state ? ffz(~p->state) + 1 : 0;
1428 psinfo->pr_state = i;
1429 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1430 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1431 psinfo->pr_nice = task_nice(p);
1432 psinfo->pr_flag = p->flags;
1433 rcu_read_lock();
1434 cred = __task_cred(p);
1435 SET_UID(psinfo->pr_uid, cred->uid);
1436 SET_GID(psinfo->pr_gid, cred->gid);
1437 rcu_read_unlock();
1438 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1439
1440 return 0;
1441 }
1442
1443 /* Here is the structure in which status of each thread is captured. */
1444 struct elf_thread_status
1445 {
1446 struct list_head list;
1447 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1448 elf_fpregset_t fpu; /* NT_PRFPREG */
1449 struct task_struct *thread;
1450 #ifdef ELF_CORE_COPY_XFPREGS
1451 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1452 #endif
1453 struct memelfnote notes[3];
1454 int num_notes;
1455 };
1456
1457 /*
1458 * In order to add the specific thread information for the elf file format,
1459 * we need to keep a linked list of every thread's pr_status and then create
1460 * a single section for them in the final core file.
1461 */
1462 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1463 {
1464 struct task_struct *p = t->thread;
1465 int sz = 0;
1466
1467 t->num_notes = 0;
1468
1469 fill_prstatus(&t->prstatus, p, signr);
1470 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1471
1472 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1473 &t->prstatus);
1474 t->num_notes++;
1475 sz += notesize(&t->notes[0]);
1476
1477 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1478 if (t->prstatus.pr_fpvalid) {
1479 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1480 &t->fpu);
1481 t->num_notes++;
1482 sz += notesize(&t->notes[1]);
1483 }
1484
1485 #ifdef ELF_CORE_COPY_XFPREGS
1486 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1487 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1488 sizeof(t->xfpu), &t->xfpu);
1489 t->num_notes++;
1490 sz += notesize(&t->notes[2]);
1491 }
1492 #endif
1493 return sz;
1494 }
1495
1496 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1497 elf_addr_t e_shoff, int segs)
1498 {
1499 elf->e_shoff = e_shoff;
1500 elf->e_shentsize = sizeof(*shdr4extnum);
1501 elf->e_shnum = 1;
1502 elf->e_shstrndx = SHN_UNDEF;
1503
1504 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1505
1506 shdr4extnum->sh_type = SHT_NULL;
1507 shdr4extnum->sh_size = elf->e_shnum;
1508 shdr4extnum->sh_link = elf->e_shstrndx;
1509 shdr4extnum->sh_info = segs;
1510 }
1511
1512 /*
1513 * dump the segments for an MMU process
1514 */
1515 #ifdef CONFIG_MMU
1516 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1517 unsigned long *limit, unsigned long mm_flags)
1518 {
1519 struct vm_area_struct *vma;
1520 int err = 0;
1521
1522 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1523 unsigned long addr;
1524
1525 if (!maydump(vma, mm_flags))
1526 continue;
1527
1528 for (addr = vma->vm_start; addr < vma->vm_end;
1529 addr += PAGE_SIZE) {
1530 struct page *page = get_dump_page(addr);
1531 if (page) {
1532 void *kaddr = kmap(page);
1533 *size += PAGE_SIZE;
1534 if (*size > *limit)
1535 err = -EFBIG;
1536 else if (!dump_write(file, kaddr, PAGE_SIZE))
1537 err = -EIO;
1538 kunmap(page);
1539 page_cache_release(page);
1540 } else if (!dump_seek(file, PAGE_SIZE))
1541 err = -EFBIG;
1542 if (err)
1543 goto out;
1544 }
1545 }
1546 out:
1547 return err;
1548 }
1549 #endif
1550
1551 /*
1552 * dump the segments for a NOMMU process
1553 */
1554 #ifndef CONFIG_MMU
1555 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1556 unsigned long *limit, unsigned long mm_flags)
1557 {
1558 struct vm_area_struct *vma;
1559
1560 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1561 if (!maydump(vma, mm_flags))
1562 continue;
1563
1564 if ((*size += PAGE_SIZE) > *limit)
1565 return -EFBIG;
1566
1567 if (!dump_write(file, (void *) vma->vm_start,
1568 vma->vm_end - vma->vm_start))
1569 return -EIO;
1570 }
1571
1572 return 0;
1573 }
1574 #endif
1575
1576 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1577 {
1578 struct vm_area_struct *vma;
1579 size_t size = 0;
1580
1581 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1582 if (maydump(vma, mm_flags))
1583 size += vma->vm_end - vma->vm_start;
1584 return size;
1585 }
1586
1587 /*
1588 * Actual dumper
1589 *
1590 * This is a two-pass process; first we find the offsets of the bits,
1591 * and then they are actually written out. If we run out of core limit
1592 * we just truncate.
1593 */
1594 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1595 {
1596 #define NUM_NOTES 6
1597 int has_dumped = 0;
1598 mm_segment_t fs;
1599 int segs;
1600 size_t size = 0;
1601 int i;
1602 struct vm_area_struct *vma;
1603 struct elfhdr *elf = NULL;
1604 loff_t offset = 0, dataoff, foffset;
1605 int numnote;
1606 struct memelfnote *notes = NULL;
1607 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1608 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1609 LIST_HEAD(thread_list);
1610 struct list_head *t;
1611 elf_fpregset_t *fpu = NULL;
1612 #ifdef ELF_CORE_COPY_XFPREGS
1613 elf_fpxregset_t *xfpu = NULL;
1614 #endif
1615 int thread_status_size = 0;
1616 elf_addr_t *auxv;
1617 struct elf_phdr *phdr4note = NULL;
1618 struct elf_shdr *shdr4extnum = NULL;
1619 Elf_Half e_phnum;
1620 elf_addr_t e_shoff;
1621
1622 /*
1623 * We no longer stop all VM operations.
1624 *
1625 * This is because those proceses that could possibly change map_count
1626 * or the mmap / vma pages are now blocked in do_exit on current
1627 * finishing this core dump.
1628 *
1629 * Only ptrace can touch these memory addresses, but it doesn't change
1630 * the map_count or the pages allocated. So no possibility of crashing
1631 * exists while dumping the mm->vm_next areas to the core file.
1632 */
1633
1634 /* alloc memory for large data structures: too large to be on stack */
1635 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1636 if (!elf)
1637 goto cleanup;
1638 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1639 if (!prstatus)
1640 goto cleanup;
1641 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1642 if (!psinfo)
1643 goto cleanup;
1644 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1645 if (!notes)
1646 goto cleanup;
1647 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1648 if (!fpu)
1649 goto cleanup;
1650 #ifdef ELF_CORE_COPY_XFPREGS
1651 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1652 if (!xfpu)
1653 goto cleanup;
1654 #endif
1655
1656 if (cprm->signr) {
1657 struct core_thread *ct;
1658 struct elf_thread_status *tmp;
1659
1660 for (ct = current->mm->core_state->dumper.next;
1661 ct; ct = ct->next) {
1662 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1663 if (!tmp)
1664 goto cleanup;
1665
1666 tmp->thread = ct->task;
1667 list_add(&tmp->list, &thread_list);
1668 }
1669
1670 list_for_each(t, &thread_list) {
1671 struct elf_thread_status *tmp;
1672 int sz;
1673
1674 tmp = list_entry(t, struct elf_thread_status, list);
1675 sz = elf_dump_thread_status(cprm->signr, tmp);
1676 thread_status_size += sz;
1677 }
1678 }
1679
1680 /* now collect the dump for the current */
1681 fill_prstatus(prstatus, current, cprm->signr);
1682 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1683
1684 segs = current->mm->map_count;
1685 segs += elf_core_extra_phdrs();
1686
1687 /* for notes section */
1688 segs++;
1689
1690 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1691 * this, kernel supports extended numbering. Have a look at
1692 * include/linux/elf.h for further information. */
1693 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1694
1695 /* Set up header */
1696 fill_elf_fdpic_header(elf, e_phnum);
1697
1698 has_dumped = 1;
1699 current->flags |= PF_DUMPCORE;
1700
1701 /*
1702 * Set up the notes in similar form to SVR4 core dumps made
1703 * with info from their /proc.
1704 */
1705
1706 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1707 fill_psinfo(psinfo, current->group_leader, current->mm);
1708 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1709
1710 numnote = 2;
1711
1712 auxv = (elf_addr_t *) current->mm->saved_auxv;
1713
1714 i = 0;
1715 do
1716 i += 2;
1717 while (auxv[i - 2] != AT_NULL);
1718 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1719 i * sizeof(elf_addr_t), auxv);
1720
1721 /* Try to dump the FPU. */
1722 if ((prstatus->pr_fpvalid =
1723 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1724 fill_note(notes + numnote++,
1725 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1726 #ifdef ELF_CORE_COPY_XFPREGS
1727 if (elf_core_copy_task_xfpregs(current, xfpu))
1728 fill_note(notes + numnote++,
1729 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1730 #endif
1731
1732 fs = get_fs();
1733 set_fs(KERNEL_DS);
1734
1735 offset += sizeof(*elf); /* Elf header */
1736 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1737 foffset = offset;
1738
1739 /* Write notes phdr entry */
1740 {
1741 int sz = 0;
1742
1743 for (i = 0; i < numnote; i++)
1744 sz += notesize(notes + i);
1745
1746 sz += thread_status_size;
1747
1748 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1749 if (!phdr4note)
1750 goto end_coredump;
1751
1752 fill_elf_note_phdr(phdr4note, sz, offset);
1753 offset += sz;
1754 }
1755
1756 /* Page-align dumped data */
1757 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1758
1759 offset += elf_core_vma_data_size(cprm->mm_flags);
1760 offset += elf_core_extra_data_size();
1761 e_shoff = offset;
1762
1763 if (e_phnum == PN_XNUM) {
1764 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1765 if (!shdr4extnum)
1766 goto end_coredump;
1767 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1768 }
1769
1770 offset = dataoff;
1771
1772 size += sizeof(*elf);
1773 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1774 goto end_coredump;
1775
1776 size += sizeof(*phdr4note);
1777 if (size > cprm->limit
1778 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1779 goto end_coredump;
1780
1781 /* write program headers for segments dump */
1782 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1783 struct elf_phdr phdr;
1784 size_t sz;
1785
1786 sz = vma->vm_end - vma->vm_start;
1787
1788 phdr.p_type = PT_LOAD;
1789 phdr.p_offset = offset;
1790 phdr.p_vaddr = vma->vm_start;
1791 phdr.p_paddr = 0;
1792 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1793 phdr.p_memsz = sz;
1794 offset += phdr.p_filesz;
1795 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1796 if (vma->vm_flags & VM_WRITE)
1797 phdr.p_flags |= PF_W;
1798 if (vma->vm_flags & VM_EXEC)
1799 phdr.p_flags |= PF_X;
1800 phdr.p_align = ELF_EXEC_PAGESIZE;
1801
1802 size += sizeof(phdr);
1803 if (size > cprm->limit
1804 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
1805 goto end_coredump;
1806 }
1807
1808 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
1809 goto end_coredump;
1810
1811 /* write out the notes section */
1812 for (i = 0; i < numnote; i++)
1813 if (!writenote(notes + i, cprm->file, &foffset))
1814 goto end_coredump;
1815
1816 /* write out the thread status notes section */
1817 list_for_each(t, &thread_list) {
1818 struct elf_thread_status *tmp =
1819 list_entry(t, struct elf_thread_status, list);
1820
1821 for (i = 0; i < tmp->num_notes; i++)
1822 if (!writenote(&tmp->notes[i], cprm->file, &foffset))
1823 goto end_coredump;
1824 }
1825
1826 if (!dump_seek(cprm->file, dataoff - foffset))
1827 goto end_coredump;
1828
1829 if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit,
1830 cprm->mm_flags) < 0)
1831 goto end_coredump;
1832
1833 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
1834 goto end_coredump;
1835
1836 if (e_phnum == PN_XNUM) {
1837 size += sizeof(*shdr4extnum);
1838 if (size > cprm->limit
1839 || !dump_write(cprm->file, shdr4extnum,
1840 sizeof(*shdr4extnum)))
1841 goto end_coredump;
1842 }
1843
1844 if (cprm->file->f_pos != offset) {
1845 /* Sanity check */
1846 printk(KERN_WARNING
1847 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1848 cprm->file->f_pos, offset);
1849 }
1850
1851 end_coredump:
1852 set_fs(fs);
1853
1854 cleanup:
1855 while (!list_empty(&thread_list)) {
1856 struct list_head *tmp = thread_list.next;
1857 list_del(tmp);
1858 kfree(list_entry(tmp, struct elf_thread_status, list));
1859 }
1860 kfree(phdr4note);
1861 kfree(elf);
1862 kfree(prstatus);
1863 kfree(psinfo);
1864 kfree(notes);
1865 kfree(fpu);
1866 kfree(shdr4extnum);
1867 #ifdef ELF_CORE_COPY_XFPREGS
1868 kfree(xfpu);
1869 #endif
1870 return has_dumped;
1871 #undef NUM_NOTES
1872 }
1873
1874 #endif /* CONFIG_ELF_CORE */
Linus' kernel tree