?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
nfs: Enable building with user namespaces enabled.
[linux-2.6.git] / fs / binfmt_elf_fdpic.c
blobdc84732e554f679dee6de0d95e7eab2b5bcda5b9
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 *);
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 register_binfmt(&elf_fdpic_format);
95 return 0;
96 }
97
98 static void __exit exit_elf_fdpic_binfmt(void)
99 {
100 unregister_binfmt(&elf_fdpic_format);
101 }
102
103 core_initcall(init_elf_fdpic_binfmt);
104 module_exit(exit_elf_fdpic_binfmt);
105
106 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
107 {
108 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
109 return 0;
110 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
111 return 0;
112 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
113 return 0;
114 if (!file->f_op || !file->f_op->mmap)
115 return 0;
116 return 1;
117 }
118
119 /*****************************************************************************/
120 /*
121 * read the program headers table into memory
122 */
123 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
124 struct file *file)
125 {
126 struct elf32_phdr *phdr;
127 unsigned long size;
128 int retval, loop;
129
130 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
131 return -ENOMEM;
132 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
133 return -ENOMEM;
134
135 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
136 params->phdrs = kmalloc(size, GFP_KERNEL);
137 if (!params->phdrs)
138 return -ENOMEM;
139
140 retval = kernel_read(file, params->hdr.e_phoff,
141 (char *) params->phdrs, size);
142 if (unlikely(retval != size))
143 return retval < 0 ? retval : -ENOEXEC;
144
145 /* determine stack size for this binary */
146 phdr = params->phdrs;
147 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
148 if (phdr->p_type != PT_GNU_STACK)
149 continue;
150
151 if (phdr->p_flags & PF_X)
152 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
153 else
154 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
155
156 params->stack_size = phdr->p_memsz;
157 break;
158 }
159
160 return 0;
161 }
162
163 /*****************************************************************************/
164 /*
165 * load an fdpic binary into various bits of memory
166 */
167 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
168 {
169 struct elf_fdpic_params exec_params, interp_params;
170 struct pt_regs *regs = current_pt_regs();
171 struct elf_phdr *phdr;
172 unsigned long stack_size, entryaddr;
173 #ifdef ELF_FDPIC_PLAT_INIT
174 unsigned long dynaddr;
175 #endif
176 #ifndef CONFIG_MMU
177 unsigned long stack_prot;
178 #endif
179 struct file *interpreter = NULL; /* to shut gcc up */
180 char *interpreter_name = NULL;
181 int executable_stack;
182 int retval, i;
183
184 kdebug("____ LOAD %d ____", current->pid);
185
186 memset(&exec_params, 0, sizeof(exec_params));
187 memset(&interp_params, 0, sizeof(interp_params));
188
189 exec_params.hdr = *(struct elfhdr *) bprm->buf;
190 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
191
192 /* check that this is a binary we know how to deal with */
193 retval = -ENOEXEC;
194 if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
195 goto error;
196
197 /* read the program header table */
198 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
199 if (retval < 0)
200 goto error;
201
202 /* scan for a program header that specifies an interpreter */
203 phdr = exec_params.phdrs;
204
205 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
206 switch (phdr->p_type) {
207 case PT_INTERP:
208 retval = -ENOMEM;
209 if (phdr->p_filesz > PATH_MAX)
210 goto error;
211 retval = -ENOENT;
212 if (phdr->p_filesz < 2)
213 goto error;
214
215 /* read the name of the interpreter into memory */
216 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
217 if (!interpreter_name)
218 goto error;
219
220 retval = kernel_read(bprm->file,
221 phdr->p_offset,
222 interpreter_name,
223 phdr->p_filesz);
224 if (unlikely(retval != phdr->p_filesz)) {
225 if (retval >= 0)
226 retval = -ENOEXEC;
227 goto error;
228 }
229
230 retval = -ENOENT;
231 if (interpreter_name[phdr->p_filesz - 1] != '\0')
232 goto error;
233
234 kdebug("Using ELF interpreter %s", interpreter_name);
235
236 /* replace the program with the interpreter */
237 interpreter = open_exec(interpreter_name);
238 retval = PTR_ERR(interpreter);
239 if (IS_ERR(interpreter)) {
240 interpreter = NULL;
241 goto error;
242 }
243
244 /*
245 * If the binary is not readable then enforce
246 * mm->dumpable = 0 regardless of the interpreter's
247 * permissions.
248 */
249 would_dump(bprm, interpreter);
250
251 retval = kernel_read(interpreter, 0, bprm->buf,
252 BINPRM_BUF_SIZE);
253 if (unlikely(retval != BINPRM_BUF_SIZE)) {
254 if (retval >= 0)
255 retval = -ENOEXEC;
256 goto error;
257 }
258
259 interp_params.hdr = *((struct elfhdr *) bprm->buf);
260 break;
261
262 case PT_LOAD:
263 #ifdef CONFIG_MMU
264 if (exec_params.load_addr == 0)
265 exec_params.load_addr = phdr->p_vaddr;
266 #endif
267 break;
268 }
269
270 }
271
272 if (elf_check_const_displacement(&exec_params.hdr))
273 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
274
275 /* perform insanity checks on the interpreter */
276 if (interpreter_name) {
277 retval = -ELIBBAD;
278 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
279 goto error;
280
281 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
282
283 /* read the interpreter's program header table */
284 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
285 if (retval < 0)
286 goto error;
287 }
288
289 stack_size = exec_params.stack_size;
290 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
291 executable_stack = EXSTACK_ENABLE_X;
292 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
293 executable_stack = EXSTACK_DISABLE_X;
294 else
295 executable_stack = EXSTACK_DEFAULT;
296
297 if (stack_size == 0) {
298 stack_size = interp_params.stack_size;
299 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
300 executable_stack = EXSTACK_ENABLE_X;
301 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
302 executable_stack = EXSTACK_DISABLE_X;
303 else
304 executable_stack = EXSTACK_DEFAULT;
305 }
306
307 retval = -ENOEXEC;
308 if (stack_size == 0)
309 goto error;
310
311 if (elf_check_const_displacement(&interp_params.hdr))
312 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
313
314 /* flush all traces of the currently running executable */
315 retval = flush_old_exec(bprm);
316 if (retval)
317 goto error;
318
319 /* there's now no turning back... the old userspace image is dead,
320 * defunct, deceased, etc. after this point we have to exit via
321 * error_kill */
322 set_personality(PER_LINUX_FDPIC);
323 if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
324 current->personality |= READ_IMPLIES_EXEC;
325
326 setup_new_exec(bprm);
327
328 set_binfmt(&elf_fdpic_format);
329
330 current->mm->start_code = 0;
331 current->mm->end_code = 0;
332 current->mm->start_stack = 0;
333 current->mm->start_data = 0;
334 current->mm->end_data = 0;
335 current->mm->context.exec_fdpic_loadmap = 0;
336 current->mm->context.interp_fdpic_loadmap = 0;
337
338 #ifdef CONFIG_MMU
339 elf_fdpic_arch_lay_out_mm(&exec_params,
340 &interp_params,
341 &current->mm->start_stack,
342 &current->mm->start_brk);
343
344 retval = setup_arg_pages(bprm, current->mm->start_stack,
345 executable_stack);
346 if (retval < 0) {
347 send_sig(SIGKILL, current, 0);
348 goto error_kill;
349 }
350 #endif
351
352 /* load the executable and interpreter into memory */
353 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
354 "executable");
355 if (retval < 0)
356 goto error_kill;
357
358 if (interpreter_name) {
359 retval = elf_fdpic_map_file(&interp_params, interpreter,
360 current->mm, "interpreter");
361 if (retval < 0) {
362 printk(KERN_ERR "Unable to load interpreter\n");
363 goto error_kill;
364 }
365
366 allow_write_access(interpreter);
367 fput(interpreter);
368 interpreter = NULL;
369 }
370
371 #ifdef CONFIG_MMU
372 if (!current->mm->start_brk)
373 current->mm->start_brk = current->mm->end_data;
374
375 current->mm->brk = current->mm->start_brk =
376 PAGE_ALIGN(current->mm->start_brk);
377
378 #else
379 /* create a stack and brk area big enough for everyone
380 * - the brk heap starts at the bottom and works up
381 * - the stack starts at the top and works down
382 */
383 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
384 if (stack_size < PAGE_SIZE * 2)
385 stack_size = PAGE_SIZE * 2;
386
387 stack_prot = PROT_READ | PROT_WRITE;
388 if (executable_stack == EXSTACK_ENABLE_X ||
389 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
390 stack_prot |= PROT_EXEC;
391
392 current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
393 MAP_PRIVATE | MAP_ANONYMOUS |
394 MAP_UNINITIALIZED | MAP_GROWSDOWN,
395 0);
396
397 if (IS_ERR_VALUE(current->mm->start_brk)) {
398 retval = current->mm->start_brk;
399 current->mm->start_brk = 0;
400 goto error_kill;
401 }
402
403 current->mm->brk = current->mm->start_brk;
404 current->mm->context.end_brk = current->mm->start_brk;
405 current->mm->context.end_brk +=
406 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
407 current->mm->start_stack = current->mm->start_brk + stack_size;
408 #endif
409
410 install_exec_creds(bprm);
411 if (create_elf_fdpic_tables(bprm, current->mm,
412 &exec_params, &interp_params) < 0)
413 goto error_kill;
414
415 kdebug("- start_code %lx", current->mm->start_code);
416 kdebug("- end_code %lx", current->mm->end_code);
417 kdebug("- start_data %lx", current->mm->start_data);
418 kdebug("- end_data %lx", current->mm->end_data);
419 kdebug("- start_brk %lx", current->mm->start_brk);
420 kdebug("- brk %lx", current->mm->brk);
421 kdebug("- start_stack %lx", current->mm->start_stack);
422
423 #ifdef ELF_FDPIC_PLAT_INIT
424 /*
425 * The ABI may specify that certain registers be set up in special
426 * ways (on i386 %edx is the address of a DT_FINI function, for
427 * example. This macro performs whatever initialization to
428 * the regs structure is required.
429 */
430 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
431 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
432 dynaddr);
433 #endif
434
435 /* everything is now ready... get the userspace context ready to roll */
436 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
437 start_thread(regs, entryaddr, current->mm->start_stack);
438
439 retval = 0;
440
441 error:
442 if (interpreter) {
443 allow_write_access(interpreter);
444 fput(interpreter);
445 }
446 kfree(interpreter_name);
447 kfree(exec_params.phdrs);
448 kfree(exec_params.loadmap);
449 kfree(interp_params.phdrs);
450 kfree(interp_params.loadmap);
451 return retval;
452
453 /* unrecoverable error - kill the process */
454 error_kill:
455 send_sig(SIGSEGV, current, 0);
456 goto error;
457
458 }
459
460 /*****************************************************************************/
461
462 #ifndef ELF_BASE_PLATFORM
463 /*
464 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
465 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
466 * will be copied to the user stack in the same manner as AT_PLATFORM.
467 */
468 #define ELF_BASE_PLATFORM NULL
469 #endif
470
471 /*
472 * present useful information to the program by shovelling it onto the new
473 * process's stack
474 */
475 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
476 struct mm_struct *mm,
477 struct elf_fdpic_params *exec_params,
478 struct elf_fdpic_params *interp_params)
479 {
480 const struct cred *cred = current_cred();
481 unsigned long sp, csp, nitems;
482 elf_caddr_t __user *argv, *envp;
483 size_t platform_len = 0, len;
484 char *k_platform, *k_base_platform;
485 char __user *u_platform, *u_base_platform, *p;
486 long hwcap;
487 int loop;
488 int nr; /* reset for each csp adjustment */
489
490 #ifdef CONFIG_MMU
491 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
492 * by the processes running on the same package. One thing we can do is
493 * to shuffle the initial stack for them, so we give the architecture
494 * an opportunity to do so here.
495 */
496 sp = arch_align_stack(bprm->p);
497 #else
498 sp = mm->start_stack;
499
500 /* stack the program arguments and environment */
501 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
502 return -EFAULT;
503 #endif
504
505 hwcap = ELF_HWCAP;
506
507 /*
508 * If this architecture has a platform capability string, copy it
509 * to userspace. In some cases (Sparc), this info is impossible
510 * for userspace to get any other way, in others (i386) it is
511 * merely difficult.
512 */
513 k_platform = ELF_PLATFORM;
514 u_platform = NULL;
515
516 if (k_platform) {
517 platform_len = strlen(k_platform) + 1;
518 sp -= platform_len;
519 u_platform = (char __user *) sp;
520 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
521 return -EFAULT;
522 }
523
524 /*
525 * If this architecture has a "base" platform capability
526 * string, copy it to userspace.
527 */
528 k_base_platform = ELF_BASE_PLATFORM;
529 u_base_platform = NULL;
530
531 if (k_base_platform) {
532 platform_len = strlen(k_base_platform) + 1;
533 sp -= platform_len;
534 u_base_platform = (char __user *) sp;
535 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
536 return -EFAULT;
537 }
538
539 sp &= ~7UL;
540
541 /* stack the load map(s) */
542 len = sizeof(struct elf32_fdpic_loadmap);
543 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
544 sp = (sp - len) & ~7UL;
545 exec_params->map_addr = sp;
546
547 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
548 return -EFAULT;
549
550 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
551
552 if (interp_params->loadmap) {
553 len = sizeof(struct elf32_fdpic_loadmap);
554 len += sizeof(struct elf32_fdpic_loadseg) *
555 interp_params->loadmap->nsegs;
556 sp = (sp - len) & ~7UL;
557 interp_params->map_addr = sp;
558
559 if (copy_to_user((void __user *) sp, interp_params->loadmap,
560 len) != 0)
561 return -EFAULT;
562
563 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
564 }
565
566 /* force 16 byte _final_ alignment here for generality */
567 #define DLINFO_ITEMS 15
568
569 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
570 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
571
572 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
573 nitems++;
574
575 csp = sp;
576 sp -= nitems * 2 * sizeof(unsigned long);
577 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
578 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
579 sp -= 1 * sizeof(unsigned long); /* argc */
580
581 csp -= sp & 15UL;
582 sp -= sp & 15UL;
583
584 /* put the ELF interpreter info on the stack */
585 #define NEW_AUX_ENT(id, val) \
586 do { \
587 struct { unsigned long _id, _val; } __user *ent; \
588 \
589 ent = (void __user *) csp; \
590 __put_user((id), &ent[nr]._id); \
591 __put_user((val), &ent[nr]._val); \
592 nr++; \
593 } while (0)
594
595 nr = 0;
596 csp -= 2 * sizeof(unsigned long);
597 NEW_AUX_ENT(AT_NULL, 0);
598 if (k_platform) {
599 nr = 0;
600 csp -= 2 * sizeof(unsigned long);
601 NEW_AUX_ENT(AT_PLATFORM,
602 (elf_addr_t) (unsigned long) u_platform);
603 }
604
605 if (k_base_platform) {
606 nr = 0;
607 csp -= 2 * sizeof(unsigned long);
608 NEW_AUX_ENT(AT_BASE_PLATFORM,
609 (elf_addr_t) (unsigned long) u_base_platform);
610 }
611
612 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
613 nr = 0;
614 csp -= 2 * sizeof(unsigned long);
615 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
616 }
617
618 nr = 0;
619 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
620 NEW_AUX_ENT(AT_HWCAP, hwcap);
621 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
622 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
623 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
624 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
625 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
626 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
627 NEW_AUX_ENT(AT_FLAGS, 0);
628 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
629 NEW_AUX_ENT(AT_UID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
630 NEW_AUX_ENT(AT_EUID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
631 NEW_AUX_ENT(AT_GID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
632 NEW_AUX_ENT(AT_EGID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
633 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
634 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
635
636 #ifdef ARCH_DLINFO
637 nr = 0;
638 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
639
640 /* ARCH_DLINFO must come last so platform specific code can enforce
641 * special alignment requirements on the AUXV if necessary (eg. PPC).
642 */
643 ARCH_DLINFO;
644 #endif
645 #undef NEW_AUX_ENT
646
647 /* allocate room for argv[] and envv[] */
648 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
649 envp = (elf_caddr_t __user *) csp;
650 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
651 argv = (elf_caddr_t __user *) csp;
652
653 /* stack argc */
654 csp -= sizeof(unsigned long);
655 __put_user(bprm->argc, (unsigned long __user *) csp);
656
657 BUG_ON(csp != sp);
658
659 /* fill in the argv[] array */
660 #ifdef CONFIG_MMU
661 current->mm->arg_start = bprm->p;
662 #else
663 current->mm->arg_start = current->mm->start_stack -
664 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
665 #endif
666
667 p = (char __user *) current->mm->arg_start;
668 for (loop = bprm->argc; loop > 0; loop--) {
669 __put_user((elf_caddr_t) p, argv++);
670 len = strnlen_user(p, MAX_ARG_STRLEN);
671 if (!len || len > MAX_ARG_STRLEN)
672 return -EINVAL;
673 p += len;
674 }
675 __put_user(NULL, argv);
676 current->mm->arg_end = (unsigned long) p;
677
678 /* fill in the envv[] array */
679 current->mm->env_start = (unsigned long) p;
680 for (loop = bprm->envc; loop > 0; loop--) {
681 __put_user((elf_caddr_t)(unsigned long) p, envp++);
682 len = strnlen_user(p, MAX_ARG_STRLEN);
683 if (!len || len > MAX_ARG_STRLEN)
684 return -EINVAL;
685 p += len;
686 }
687 __put_user(NULL, envp);
688 current->mm->env_end = (unsigned long) p;
689
690 mm->start_stack = (unsigned long) sp;
691 return 0;
692 }
693
694 /*****************************************************************************/
695 /*
696 * transfer the program arguments and environment from the holding pages onto
697 * the stack
698 */
699 #ifndef CONFIG_MMU
700 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
701 unsigned long *_sp)
702 {
703 unsigned long index, stop, sp;
704 char *src;
705 int ret = 0;
706
707 stop = bprm->p >> PAGE_SHIFT;
708 sp = *_sp;
709
710 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
711 src = kmap(bprm->page[index]);
712 sp -= PAGE_SIZE;
713 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
714 ret = -EFAULT;
715 kunmap(bprm->page[index]);
716 if (ret < 0)
717 goto out;
718 }
719
720 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
721
722 out:
723 return ret;
724 }
725 #endif
726
727 /*****************************************************************************/
728 /*
729 * load the appropriate binary image (executable or interpreter) into memory
730 * - we assume no MMU is available
731 * - if no other PIC bits are set in params->hdr->e_flags
732 * - we assume that the LOADable segments in the binary are independently relocatable
733 * - we assume R/O executable segments are shareable
734 * - else
735 * - we assume the loadable parts of the image to require fixed displacement
736 * - the image is not shareable
737 */
738 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
739 struct file *file,
740 struct mm_struct *mm,
741 const char *what)
742 {
743 struct elf32_fdpic_loadmap *loadmap;
744 #ifdef CONFIG_MMU
745 struct elf32_fdpic_loadseg *mseg;
746 #endif
747 struct elf32_fdpic_loadseg *seg;
748 struct elf32_phdr *phdr;
749 unsigned long load_addr, stop;
750 unsigned nloads, tmp;
751 size_t size;
752 int loop, ret;
753
754 /* allocate a load map table */
755 nloads = 0;
756 for (loop = 0; loop < params->hdr.e_phnum; loop++)
757 if (params->phdrs[loop].p_type == PT_LOAD)
758 nloads++;
759
760 if (nloads == 0)
761 return -ELIBBAD;
762
763 size = sizeof(*loadmap) + nloads * sizeof(*seg);
764 loadmap = kzalloc(size, GFP_KERNEL);
765 if (!loadmap)
766 return -ENOMEM;
767
768 params->loadmap = loadmap;
769
770 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
771 loadmap->nsegs = nloads;
772
773 load_addr = params->load_addr;
774 seg = loadmap->segs;
775
776 /* map the requested LOADs into the memory space */
777 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
778 case ELF_FDPIC_FLAG_CONSTDISP:
779 case ELF_FDPIC_FLAG_CONTIGUOUS:
780 #ifndef CONFIG_MMU
781 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
782 if (ret < 0)
783 return ret;
784 break;
785 #endif
786 default:
787 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
788 if (ret < 0)
789 return ret;
790 break;
791 }
792
793 /* map the entry point */
794 if (params->hdr.e_entry) {
795 seg = loadmap->segs;
796 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
797 if (params->hdr.e_entry >= seg->p_vaddr &&
798 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
799 params->entry_addr =
800 (params->hdr.e_entry - seg->p_vaddr) +
801 seg->addr;
802 break;
803 }
804 }
805 }
806
807 /* determine where the program header table has wound up if mapped */
808 stop = params->hdr.e_phoff;
809 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
810 phdr = params->phdrs;
811
812 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
813 if (phdr->p_type != PT_LOAD)
814 continue;
815
816 if (phdr->p_offset > params->hdr.e_phoff ||
817 phdr->p_offset + phdr->p_filesz < stop)
818 continue;
819
820 seg = loadmap->segs;
821 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
822 if (phdr->p_vaddr >= seg->p_vaddr &&
823 phdr->p_vaddr + phdr->p_filesz <=
824 seg->p_vaddr + seg->p_memsz) {
825 params->ph_addr =
826 (phdr->p_vaddr - seg->p_vaddr) +
827 seg->addr +
828 params->hdr.e_phoff - phdr->p_offset;
829 break;
830 }
831 }
832 break;
833 }
834
835 /* determine where the dynamic section has wound up if there is one */
836 phdr = params->phdrs;
837 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
838 if (phdr->p_type != PT_DYNAMIC)
839 continue;
840
841 seg = loadmap->segs;
842 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
843 if (phdr->p_vaddr >= seg->p_vaddr &&
844 phdr->p_vaddr + phdr->p_memsz <=
845 seg->p_vaddr + seg->p_memsz) {
846 params->dynamic_addr =
847 (phdr->p_vaddr - seg->p_vaddr) +
848 seg->addr;
849
850 /* check the dynamic section contains at least
851 * one item, and that the last item is a NULL
852 * entry */
853 if (phdr->p_memsz == 0 ||
854 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
855 goto dynamic_error;
856
857 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
858 if (((Elf32_Dyn *)
859 params->dynamic_addr)[tmp - 1].d_tag != 0)
860 goto dynamic_error;
861 break;
862 }
863 }
864 break;
865 }
866
867 /* now elide adjacent segments in the load map on MMU linux
868 * - on uClinux the holes between may actually be filled with system
869 * stuff or stuff from other processes
870 */
871 #ifdef CONFIG_MMU
872 nloads = loadmap->nsegs;
873 mseg = loadmap->segs;
874 seg = mseg + 1;
875 for (loop = 1; loop < nloads; loop++) {
876 /* see if we have a candidate for merging */
877 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
878 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
879 if (load_addr == (seg->addr & PAGE_MASK)) {
880 mseg->p_memsz +=
881 load_addr -
882 (mseg->addr + mseg->p_memsz);
883 mseg->p_memsz += seg->addr & ~PAGE_MASK;
884 mseg->p_memsz += seg->p_memsz;
885 loadmap->nsegs--;
886 continue;
887 }
888 }
889
890 mseg++;
891 if (mseg != seg)
892 *mseg = *seg;
893 }
894 #endif
895
896 kdebug("Mapped Object [%s]:", what);
897 kdebug("- elfhdr : %lx", params->elfhdr_addr);
898 kdebug("- entry : %lx", params->entry_addr);
899 kdebug("- PHDR[] : %lx", params->ph_addr);
900 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
901 seg = loadmap->segs;
902 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
903 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
904 loop,
905 seg->addr, seg->addr + seg->p_memsz - 1,
906 seg->p_vaddr, seg->p_memsz);
907
908 return 0;
909
910 dynamic_error:
911 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
912 what, file->f_path.dentry->d_inode->i_ino);
913 return -ELIBBAD;
914 }
915
916 /*****************************************************************************/
917 /*
918 * map a file with constant displacement under uClinux
919 */
920 #ifndef CONFIG_MMU
921 static int elf_fdpic_map_file_constdisp_on_uclinux(
922 struct elf_fdpic_params *params,
923 struct file *file,
924 struct mm_struct *mm)
925 {
926 struct elf32_fdpic_loadseg *seg;
927 struct elf32_phdr *phdr;
928 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
929 loff_t fpos;
930 int loop, ret;
931
932 load_addr = params->load_addr;
933 seg = params->loadmap->segs;
934
935 /* determine the bounds of the contiguous overall allocation we must
936 * make */
937 phdr = params->phdrs;
938 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
939 if (params->phdrs[loop].p_type != PT_LOAD)
940 continue;
941
942 if (base > phdr->p_vaddr)
943 base = phdr->p_vaddr;
944 if (top < phdr->p_vaddr + phdr->p_memsz)
945 top = phdr->p_vaddr + phdr->p_memsz;
946 }
947
948 /* allocate one big anon block for everything */
949 mflags = MAP_PRIVATE;
950 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
951 mflags |= MAP_EXECUTABLE;
952
953 maddr = vm_mmap(NULL, load_addr, top - base,
954 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
955 if (IS_ERR_VALUE(maddr))
956 return (int) maddr;
957
958 if (load_addr != 0)
959 load_addr += PAGE_ALIGN(top - base);
960
961 /* and then load the file segments into it */
962 phdr = params->phdrs;
963 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
964 if (params->phdrs[loop].p_type != PT_LOAD)
965 continue;
966
967 fpos = phdr->p_offset;
968
969 seg->addr = maddr + (phdr->p_vaddr - base);
970 seg->p_vaddr = phdr->p_vaddr;
971 seg->p_memsz = phdr->p_memsz;
972
973 ret = file->f_op->read(file, (void *) seg->addr,
974 phdr->p_filesz, &fpos);
975 if (ret < 0)
976 return ret;
977
978 /* map the ELF header address if in this segment */
979 if (phdr->p_offset == 0)
980 params->elfhdr_addr = seg->addr;
981
982 /* clear any space allocated but not loaded */
983 if (phdr->p_filesz < phdr->p_memsz) {
984 if (clear_user((void *) (seg->addr + phdr->p_filesz),
985 phdr->p_memsz - phdr->p_filesz))
986 return -EFAULT;
987 }
988
989 if (mm) {
990 if (phdr->p_flags & PF_X) {
991 if (!mm->start_code) {
992 mm->start_code = seg->addr;
993 mm->end_code = seg->addr +
994 phdr->p_memsz;
995 }
996 } else if (!mm->start_data) {
997 mm->start_data = seg->addr;
998 mm->end_data = seg->addr + phdr->p_memsz;
999 }
1000 }
1001
1002 seg++;
1003 }
1004
1005 return 0;
1006 }
1007 #endif
1008
1009 /*****************************************************************************/
1010 /*
1011 * map a binary by direct mmap() of the individual PT_LOAD segments
1012 */
1013 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1014 struct file *file,
1015 struct mm_struct *mm)
1016 {
1017 struct elf32_fdpic_loadseg *seg;
1018 struct elf32_phdr *phdr;
1019 unsigned long load_addr, delta_vaddr;
1020 int loop, dvset;
1021
1022 load_addr = params->load_addr;
1023 delta_vaddr = 0;
1024 dvset = 0;
1025
1026 seg = params->loadmap->segs;
1027
1028 /* deal with each load segment separately */
1029 phdr = params->phdrs;
1030 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1031 unsigned long maddr, disp, excess, excess1;
1032 int prot = 0, flags;
1033
1034 if (phdr->p_type != PT_LOAD)
1035 continue;
1036
1037 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1038 (unsigned long) phdr->p_vaddr,
1039 (unsigned long) phdr->p_offset,
1040 (unsigned long) phdr->p_filesz,
1041 (unsigned long) phdr->p_memsz);
1042
1043 /* determine the mapping parameters */
1044 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1045 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1046 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1047
1048 flags = MAP_PRIVATE | MAP_DENYWRITE;
1049 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1050 flags |= MAP_EXECUTABLE;
1051
1052 maddr = 0;
1053
1054 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1055 case ELF_FDPIC_FLAG_INDEPENDENT:
1056 /* PT_LOADs are independently locatable */
1057 break;
1058
1059 case ELF_FDPIC_FLAG_HONOURVADDR:
1060 /* the specified virtual address must be honoured */
1061 maddr = phdr->p_vaddr;
1062 flags |= MAP_FIXED;
1063 break;
1064
1065 case ELF_FDPIC_FLAG_CONSTDISP:
1066 /* constant displacement
1067 * - can be mapped anywhere, but must be mapped as a
1068 * unit
1069 */
1070 if (!dvset) {
1071 maddr = load_addr;
1072 delta_vaddr = phdr->p_vaddr;
1073 dvset = 1;
1074 } else {
1075 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1076 flags |= MAP_FIXED;
1077 }
1078 break;
1079
1080 case ELF_FDPIC_FLAG_CONTIGUOUS:
1081 /* contiguity handled later */
1082 break;
1083
1084 default:
1085 BUG();
1086 }
1087
1088 maddr &= PAGE_MASK;
1089
1090 /* create the mapping */
1091 disp = phdr->p_vaddr & ~PAGE_MASK;
1092 maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1093 phdr->p_offset - disp);
1094
1095 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1096 loop, phdr->p_memsz + disp, prot, flags,
1097 phdr->p_offset - disp, maddr);
1098
1099 if (IS_ERR_VALUE(maddr))
1100 return (int) maddr;
1101
1102 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1103 ELF_FDPIC_FLAG_CONTIGUOUS)
1104 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1105
1106 seg->addr = maddr + disp;
1107 seg->p_vaddr = phdr->p_vaddr;
1108 seg->p_memsz = phdr->p_memsz;
1109
1110 /* map the ELF header address if in this segment */
1111 if (phdr->p_offset == 0)
1112 params->elfhdr_addr = seg->addr;
1113
1114 /* clear the bit between beginning of mapping and beginning of
1115 * PT_LOAD */
1116 if (prot & PROT_WRITE && disp > 0) {
1117 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1118 if (clear_user((void __user *) maddr, disp))
1119 return -EFAULT;
1120 maddr += disp;
1121 }
1122
1123 /* clear any space allocated but not loaded
1124 * - on uClinux we can just clear the lot
1125 * - on MMU linux we'll get a SIGBUS beyond the last page
1126 * extant in the file
1127 */
1128 excess = phdr->p_memsz - phdr->p_filesz;
1129 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1130
1131 #ifdef CONFIG_MMU
1132 if (excess > excess1) {
1133 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1134 unsigned long xmaddr;
1135
1136 flags |= MAP_FIXED | MAP_ANONYMOUS;
1137 xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1138 prot, flags, 0);
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 if (clear_user((void __user *) maddr + phdr->p_filesz,
1153 excess1))
1154 return -EFAULT;
1155 }
1156
1157 #else
1158 if (excess > 0) {
1159 kdebug("clear[%d] ad=%lx sz=%lx",
1160 loop, maddr + phdr->p_filesz, excess);
1161 if (clear_user((void *) maddr + phdr->p_filesz, excess))
1162 return -EFAULT;
1163 }
1164 #endif
1165
1166 if (mm) {
1167 if (phdr->p_flags & PF_X) {
1168 if (!mm->start_code) {
1169 mm->start_code = maddr;
1170 mm->end_code = maddr + phdr->p_memsz;
1171 }
1172 } else if (!mm->start_data) {
1173 mm->start_data = maddr;
1174 mm->end_data = maddr + phdr->p_memsz;
1175 }
1176 }
1177
1178 seg++;
1179 }
1180
1181 return 0;
1182 }
1183
1184 /*****************************************************************************/
1185 /*
1186 * ELF-FDPIC core dumper
1187 *
1188 * Modelled on fs/exec.c:aout_core_dump()
1189 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1190 *
1191 * Modelled on fs/binfmt_elf.c core dumper
1192 */
1193 #ifdef CONFIG_ELF_CORE
1194
1195 /*
1196 * Decide whether a segment is worth dumping; default is yes to be
1197 * sure (missing info is worse than too much; etc).
1198 * Personally I'd include everything, and use the coredump limit...
1199 *
1200 * I think we should skip something. But I am not sure how. H.J.
1201 */
1202 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1203 {
1204 int dump_ok;
1205
1206 /* Do not dump I/O mapped devices or special mappings */
1207 if (vma->vm_flags & VM_IO) {
1208 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1209 return 0;
1210 }
1211
1212 /* If we may not read the contents, don't allow us to dump
1213 * them either. "dump_write()" can't handle it anyway.
1214 */
1215 if (!(vma->vm_flags & VM_READ)) {
1216 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1217 return 0;
1218 }
1219
1220 /* By default, dump shared memory if mapped from an anonymous file. */
1221 if (vma->vm_flags & VM_SHARED) {
1222 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1223 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1224 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1225 vma->vm_flags, dump_ok ? "yes" : "no");
1226 return dump_ok;
1227 }
1228
1229 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1230 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1231 vma->vm_flags, dump_ok ? "yes" : "no");
1232 return dump_ok;
1233 }
1234
1235 #ifdef CONFIG_MMU
1236 /* By default, if it hasn't been written to, don't write it out */
1237 if (!vma->anon_vma) {
1238 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1239 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1240 vma->vm_flags, dump_ok ? "yes" : "no");
1241 return dump_ok;
1242 }
1243 #endif
1244
1245 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1246 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1247 dump_ok ? "yes" : "no");
1248 return dump_ok;
1249 }
1250
1251 /* An ELF note in memory */
1252 struct memelfnote
1253 {
1254 const char *name;
1255 int type;
1256 unsigned int datasz;
1257 void *data;
1258 };
1259
1260 static int notesize(struct memelfnote *en)
1261 {
1262 int sz;
1263
1264 sz = sizeof(struct elf_note);
1265 sz += roundup(strlen(en->name) + 1, 4);
1266 sz += roundup(en->datasz, 4);
1267
1268 return sz;
1269 }
1270
1271 /* #define DEBUG */
1272
1273 #define DUMP_WRITE(addr, nr, foffset) \
1274 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1275
1276 static int alignfile(struct file *file, loff_t *foffset)
1277 {
1278 static const char buf[4] = { 0, };
1279 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1280 return 1;
1281 }
1282
1283 static int writenote(struct memelfnote *men, struct file *file,
1284 loff_t *foffset)
1285 {
1286 struct elf_note en;
1287 en.n_namesz = strlen(men->name) + 1;
1288 en.n_descsz = men->datasz;
1289 en.n_type = men->type;
1290
1291 DUMP_WRITE(&en, sizeof(en), foffset);
1292 DUMP_WRITE(men->name, en.n_namesz, foffset);
1293 if (!alignfile(file, foffset))
1294 return 0;
1295 DUMP_WRITE(men->data, men->datasz, foffset);
1296 if (!alignfile(file, foffset))
1297 return 0;
1298
1299 return 1;
1300 }
1301 #undef DUMP_WRITE
1302
1303 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1304 {
1305 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1306 elf->e_ident[EI_CLASS] = ELF_CLASS;
1307 elf->e_ident[EI_DATA] = ELF_DATA;
1308 elf->e_ident[EI_VERSION] = EV_CURRENT;
1309 elf->e_ident[EI_OSABI] = ELF_OSABI;
1310 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1311
1312 elf->e_type = ET_CORE;
1313 elf->e_machine = ELF_ARCH;
1314 elf->e_version = EV_CURRENT;
1315 elf->e_entry = 0;
1316 elf->e_phoff = sizeof(struct elfhdr);
1317 elf->e_shoff = 0;
1318 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1319 elf->e_ehsize = sizeof(struct elfhdr);
1320 elf->e_phentsize = sizeof(struct elf_phdr);
1321 elf->e_phnum = segs;
1322 elf->e_shentsize = 0;
1323 elf->e_shnum = 0;
1324 elf->e_shstrndx = 0;
1325 return;
1326 }
1327
1328 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1329 {
1330 phdr->p_type = PT_NOTE;
1331 phdr->p_offset = offset;
1332 phdr->p_vaddr = 0;
1333 phdr->p_paddr = 0;
1334 phdr->p_filesz = sz;
1335 phdr->p_memsz = 0;
1336 phdr->p_flags = 0;
1337 phdr->p_align = 0;
1338 return;
1339 }
1340
1341 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1342 unsigned int sz, void *data)
1343 {
1344 note->name = name;
1345 note->type = type;
1346 note->datasz = sz;
1347 note->data = data;
1348 return;
1349 }
1350
1351 /*
1352 * fill up all the fields in prstatus from the given task struct, except
1353 * registers which need to be filled up separately.
1354 */
1355 static void fill_prstatus(struct elf_prstatus *prstatus,
1356 struct task_struct *p, long signr)
1357 {
1358 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1359 prstatus->pr_sigpend = p->pending.signal.sig[0];
1360 prstatus->pr_sighold = p->blocked.sig[0];
1361 rcu_read_lock();
1362 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1363 rcu_read_unlock();
1364 prstatus->pr_pid = task_pid_vnr(p);
1365 prstatus->pr_pgrp = task_pgrp_vnr(p);
1366 prstatus->pr_sid = task_session_vnr(p);
1367 if (thread_group_leader(p)) {
1368 struct task_cputime cputime;
1369
1370 /*
1371 * This is the record for the group leader. It shows the
1372 * group-wide total, not its individual thread total.
1373 */
1374 thread_group_cputime(p, &cputime);
1375 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1376 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1377 } else {
1378 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1379 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1380 }
1381 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1382 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1383
1384 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1385 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1386 }
1387
1388 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1389 struct mm_struct *mm)
1390 {
1391 const struct cred *cred;
1392 unsigned int i, len;
1393
1394 /* first copy the parameters from user space */
1395 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1396
1397 len = mm->arg_end - mm->arg_start;
1398 if (len >= ELF_PRARGSZ)
1399 len = ELF_PRARGSZ - 1;
1400 if (copy_from_user(&psinfo->pr_psargs,
1401 (const char __user *) mm->arg_start, len))
1402 return -EFAULT;
1403 for (i = 0; i < len; i++)
1404 if (psinfo->pr_psargs[i] == 0)
1405 psinfo->pr_psargs[i] = ' ';
1406 psinfo->pr_psargs[len] = 0;
1407
1408 rcu_read_lock();
1409 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1410 rcu_read_unlock();
1411 psinfo->pr_pid = task_pid_vnr(p);
1412 psinfo->pr_pgrp = task_pgrp_vnr(p);
1413 psinfo->pr_sid = task_session_vnr(p);
1414
1415 i = p->state ? ffz(~p->state) + 1 : 0;
1416 psinfo->pr_state = i;
1417 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1418 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1419 psinfo->pr_nice = task_nice(p);
1420 psinfo->pr_flag = p->flags;
1421 rcu_read_lock();
1422 cred = __task_cred(p);
1423 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1424 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1425 rcu_read_unlock();
1426 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1427
1428 return 0;
1429 }
1430
1431 /* Here is the structure in which status of each thread is captured. */
1432 struct elf_thread_status
1433 {
1434 struct list_head list;
1435 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1436 elf_fpregset_t fpu; /* NT_PRFPREG */
1437 struct task_struct *thread;
1438 #ifdef ELF_CORE_COPY_XFPREGS
1439 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1440 #endif
1441 struct memelfnote notes[3];
1442 int num_notes;
1443 };
1444
1445 /*
1446 * In order to add the specific thread information for the elf file format,
1447 * we need to keep a linked list of every thread's pr_status and then create
1448 * a single section for them in the final core file.
1449 */
1450 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1451 {
1452 struct task_struct *p = t->thread;
1453 int sz = 0;
1454
1455 t->num_notes = 0;
1456
1457 fill_prstatus(&t->prstatus, p, signr);
1458 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1459
1460 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1461 &t->prstatus);
1462 t->num_notes++;
1463 sz += notesize(&t->notes[0]);
1464
1465 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1466 if (t->prstatus.pr_fpvalid) {
1467 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1468 &t->fpu);
1469 t->num_notes++;
1470 sz += notesize(&t->notes[1]);
1471 }
1472
1473 #ifdef ELF_CORE_COPY_XFPREGS
1474 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1475 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1476 sizeof(t->xfpu), &t->xfpu);
1477 t->num_notes++;
1478 sz += notesize(&t->notes[2]);
1479 }
1480 #endif
1481 return sz;
1482 }
1483
1484 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1485 elf_addr_t e_shoff, int segs)
1486 {
1487 elf->e_shoff = e_shoff;
1488 elf->e_shentsize = sizeof(*shdr4extnum);
1489 elf->e_shnum = 1;
1490 elf->e_shstrndx = SHN_UNDEF;
1491
1492 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1493
1494 shdr4extnum->sh_type = SHT_NULL;
1495 shdr4extnum->sh_size = elf->e_shnum;
1496 shdr4extnum->sh_link = elf->e_shstrndx;
1497 shdr4extnum->sh_info = segs;
1498 }
1499
1500 /*
1501 * dump the segments for an MMU process
1502 */
1503 #ifdef CONFIG_MMU
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 int err = 0;
1509
1510 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1511 unsigned long addr;
1512
1513 if (!maydump(vma, mm_flags))
1514 continue;
1515
1516 for (addr = vma->vm_start; addr < vma->vm_end;
1517 addr += PAGE_SIZE) {
1518 struct page *page = get_dump_page(addr);
1519 if (page) {
1520 void *kaddr = kmap(page);
1521 *size += PAGE_SIZE;
1522 if (*size > *limit)
1523 err = -EFBIG;
1524 else if (!dump_write(file, kaddr, PAGE_SIZE))
1525 err = -EIO;
1526 kunmap(page);
1527 page_cache_release(page);
1528 } else if (!dump_seek(file, PAGE_SIZE))
1529 err = -EFBIG;
1530 if (err)
1531 goto out;
1532 }
1533 }
1534 out:
1535 return err;
1536 }
1537 #endif
1538
1539 /*
1540 * dump the segments for a NOMMU process
1541 */
1542 #ifndef CONFIG_MMU
1543 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1544 unsigned long *limit, unsigned long mm_flags)
1545 {
1546 struct vm_area_struct *vma;
1547
1548 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1549 if (!maydump(vma, mm_flags))
1550 continue;
1551
1552 if ((*size += PAGE_SIZE) > *limit)
1553 return -EFBIG;
1554
1555 if (!dump_write(file, (void *) vma->vm_start,
1556 vma->vm_end - vma->vm_start))
1557 return -EIO;
1558 }
1559
1560 return 0;
1561 }
1562 #endif
1563
1564 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1565 {
1566 struct vm_area_struct *vma;
1567 size_t size = 0;
1568
1569 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1570 if (maydump(vma, mm_flags))
1571 size += vma->vm_end - vma->vm_start;
1572 return size;
1573 }
1574
1575 /*
1576 * Actual dumper
1577 *
1578 * This is a two-pass process; first we find the offsets of the bits,
1579 * and then they are actually written out. If we run out of core limit
1580 * we just truncate.
1581 */
1582 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1583 {
1584 #define NUM_NOTES 6
1585 int has_dumped = 0;
1586 mm_segment_t fs;
1587 int segs;
1588 size_t size = 0;
1589 int i;
1590 struct vm_area_struct *vma;
1591 struct elfhdr *elf = NULL;
1592 loff_t offset = 0, dataoff, foffset;
1593 int numnote;
1594 struct memelfnote *notes = NULL;
1595 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1596 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1597 LIST_HEAD(thread_list);
1598 struct list_head *t;
1599 elf_fpregset_t *fpu = NULL;
1600 #ifdef ELF_CORE_COPY_XFPREGS
1601 elf_fpxregset_t *xfpu = NULL;
1602 #endif
1603 int thread_status_size = 0;
1604 elf_addr_t *auxv;
1605 struct elf_phdr *phdr4note = NULL;
1606 struct elf_shdr *shdr4extnum = NULL;
1607 Elf_Half e_phnum;
1608 elf_addr_t e_shoff;
1609
1610 /*
1611 * We no longer stop all VM operations.
1612 *
1613 * This is because those proceses that could possibly change map_count
1614 * or the mmap / vma pages are now blocked in do_exit on current
1615 * finishing this core dump.
1616 *
1617 * Only ptrace can touch these memory addresses, but it doesn't change
1618 * the map_count or the pages allocated. So no possibility of crashing
1619 * exists while dumping the mm->vm_next areas to the core file.
1620 */
1621
1622 /* alloc memory for large data structures: too large to be on stack */
1623 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1624 if (!elf)
1625 goto cleanup;
1626 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1627 if (!prstatus)
1628 goto cleanup;
1629 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1630 if (!psinfo)
1631 goto cleanup;
1632 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1633 if (!notes)
1634 goto cleanup;
1635 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1636 if (!fpu)
1637 goto cleanup;
1638 #ifdef ELF_CORE_COPY_XFPREGS
1639 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1640 if (!xfpu)
1641 goto cleanup;
1642 #endif
1643
1644 if (cprm->siginfo->si_signo) {
1645 struct core_thread *ct;
1646 struct elf_thread_status *tmp;
1647
1648 for (ct = current->mm->core_state->dumper.next;
1649 ct; ct = ct->next) {
1650 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1651 if (!tmp)
1652 goto cleanup;
1653
1654 tmp->thread = ct->task;
1655 list_add(&tmp->list, &thread_list);
1656 }
1657
1658 list_for_each(t, &thread_list) {
1659 struct elf_thread_status *tmp;
1660 int sz;
1661
1662 tmp = list_entry(t, struct elf_thread_status, list);
1663 sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1664 thread_status_size += sz;
1665 }
1666 }
1667
1668 /* now collect the dump for the current */
1669 fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1670 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1671
1672 segs = current->mm->map_count;
1673 segs += elf_core_extra_phdrs();
1674
1675 /* for notes section */
1676 segs++;
1677
1678 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1679 * this, kernel supports extended numbering. Have a look at
1680 * include/linux/elf.h for further information. */
1681 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1682
1683 /* Set up header */
1684 fill_elf_fdpic_header(elf, e_phnum);
1685
1686 has_dumped = 1;
1687 current->flags |= PF_DUMPCORE;
1688
1689 /*
1690 * Set up the notes in similar form to SVR4 core dumps made
1691 * with info from their /proc.
1692 */
1693
1694 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1695 fill_psinfo(psinfo, current->group_leader, current->mm);
1696 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1697
1698 numnote = 2;
1699
1700 auxv = (elf_addr_t *) current->mm->saved_auxv;
1701
1702 i = 0;
1703 do
1704 i += 2;
1705 while (auxv[i - 2] != AT_NULL);
1706 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1707 i * sizeof(elf_addr_t), auxv);
1708
1709 /* Try to dump the FPU. */
1710 if ((prstatus->pr_fpvalid =
1711 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1712 fill_note(notes + numnote++,
1713 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1714 #ifdef ELF_CORE_COPY_XFPREGS
1715 if (elf_core_copy_task_xfpregs(current, xfpu))
1716 fill_note(notes + numnote++,
1717 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1718 #endif
1719
1720 fs = get_fs();
1721 set_fs(KERNEL_DS);
1722
1723 offset += sizeof(*elf); /* Elf header */
1724 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1725 foffset = offset;
1726
1727 /* Write notes phdr entry */
1728 {
1729 int sz = 0;
1730
1731 for (i = 0; i < numnote; i++)
1732 sz += notesize(notes + i);
1733
1734 sz += thread_status_size;
1735
1736 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1737 if (!phdr4note)
1738 goto end_coredump;
1739
1740 fill_elf_note_phdr(phdr4note, sz, offset);
1741 offset += sz;
1742 }
1743
1744 /* Page-align dumped data */
1745 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1746
1747 offset += elf_core_vma_data_size(cprm->mm_flags);
1748 offset += elf_core_extra_data_size();
1749 e_shoff = offset;
1750
1751 if (e_phnum == PN_XNUM) {
1752 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1753 if (!shdr4extnum)
1754 goto end_coredump;
1755 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1756 }
1757
1758 offset = dataoff;
1759
1760 size += sizeof(*elf);
1761 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1762 goto end_coredump;
1763
1764 size += sizeof(*phdr4note);
1765 if (size > cprm->limit
1766 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1767 goto end_coredump;
1768
1769 /* write program headers for segments dump */
1770 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1771 struct elf_phdr phdr;
1772 size_t sz;
1773
1774 sz = vma->vm_end - vma->vm_start;
1775
1776 phdr.p_type = PT_LOAD;
1777 phdr.p_offset = offset;
1778 phdr.p_vaddr = vma->vm_start;
1779 phdr.p_paddr = 0;
1780 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1781 phdr.p_memsz = sz;
1782 offset += phdr.p_filesz;
1783 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1784 if (vma->vm_flags & VM_WRITE)
1785 phdr.p_flags |= PF_W;
1786 if (vma->vm_flags & VM_EXEC)
1787 phdr.p_flags |= PF_X;
1788 phdr.p_align = ELF_EXEC_PAGESIZE;
1789
1790 size += sizeof(phdr);
1791 if (size > cprm->limit
1792 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
1793 goto end_coredump;
1794 }
1795
1796 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
1797 goto end_coredump;
1798
1799 /* write out the notes section */
1800 for (i = 0; i < numnote; i++)
1801 if (!writenote(notes + i, cprm->file, &foffset))
1802 goto end_coredump;
1803
1804 /* write out the thread status notes section */
1805 list_for_each(t, &thread_list) {
1806 struct elf_thread_status *tmp =
1807 list_entry(t, struct elf_thread_status, list);
1808
1809 for (i = 0; i < tmp->num_notes; i++)
1810 if (!writenote(&tmp->notes[i], cprm->file, &foffset))
1811 goto end_coredump;
1812 }
1813
1814 if (!dump_seek(cprm->file, dataoff - foffset))
1815 goto end_coredump;
1816
1817 if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit,
1818 cprm->mm_flags) < 0)
1819 goto end_coredump;
1820
1821 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
1822 goto end_coredump;
1823
1824 if (e_phnum == PN_XNUM) {
1825 size += sizeof(*shdr4extnum);
1826 if (size > cprm->limit
1827 || !dump_write(cprm->file, shdr4extnum,
1828 sizeof(*shdr4extnum)))
1829 goto end_coredump;
1830 }
1831
1832 if (cprm->file->f_pos != offset) {
1833 /* Sanity check */
1834 printk(KERN_WARNING
1835 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1836 cprm->file->f_pos, offset);
1837 }
1838
1839 end_coredump:
1840 set_fs(fs);
1841
1842 cleanup:
1843 while (!list_empty(&thread_list)) {
1844 struct list_head *tmp = thread_list.next;
1845 list_del(tmp);
1846 kfree(list_entry(tmp, struct elf_thread_status, list));
1847 }
1848 kfree(phdr4note);
1849 kfree(elf);
1850 kfree(prstatus);
1851 kfree(psinfo);
1852 kfree(notes);
1853 kfree(fpu);
1854 kfree(shdr4extnum);
1855 #ifdef ELF_CORE_COPY_XFPREGS
1856 kfree(xfpu);
1857 #endif
1858 return has_dumped;
1859 #undef NUM_NOTES
1860 }
1861
1862 #endif /* CONFIG_ELF_CORE */
Linus' kernel tree