?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: x86 emulator: Use opcode::execute for 0F 01 opcode
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / binfmt_elf_fdpic.c
blob63039ed9576f75c396ed0dae71bec1e4abf53058
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 if (file_permission(interpreter, MAY_READ) < 0)
249 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
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 current->flags &= ~PF_FORKNOEXEC;
339
340 #ifdef CONFIG_MMU
341 elf_fdpic_arch_lay_out_mm(&exec_params,
342 &interp_params,
343 &current->mm->start_stack,
344 &current->mm->start_brk);
345
346 retval = setup_arg_pages(bprm, current->mm->start_stack,
347 executable_stack);
348 if (retval < 0) {
349 send_sig(SIGKILL, current, 0);
350 goto error_kill;
351 }
352 #endif
353
354 /* load the executable and interpreter into memory */
355 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
356 "executable");
357 if (retval < 0)
358 goto error_kill;
359
360 if (interpreter_name) {
361 retval = elf_fdpic_map_file(&interp_params, interpreter,
362 current->mm, "interpreter");
363 if (retval < 0) {
364 printk(KERN_ERR "Unable to load interpreter\n");
365 goto error_kill;
366 }
367
368 allow_write_access(interpreter);
369 fput(interpreter);
370 interpreter = NULL;
371 }
372
373 #ifdef CONFIG_MMU
374 if (!current->mm->start_brk)
375 current->mm->start_brk = current->mm->end_data;
376
377 current->mm->brk = current->mm->start_brk =
378 PAGE_ALIGN(current->mm->start_brk);
379
380 #else
381 /* create a stack and brk area big enough for everyone
382 * - the brk heap starts at the bottom and works up
383 * - the stack starts at the top and works down
384 */
385 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
386 if (stack_size < PAGE_SIZE * 2)
387 stack_size = PAGE_SIZE * 2;
388
389 stack_prot = PROT_READ | PROT_WRITE;
390 if (executable_stack == EXSTACK_ENABLE_X ||
391 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
392 stack_prot |= PROT_EXEC;
393
394 down_write(&current->mm->mmap_sem);
395 current->mm->start_brk = do_mmap(NULL, 0, stack_size, stack_prot,
396 MAP_PRIVATE | MAP_ANONYMOUS |
397 MAP_UNINITIALIZED | MAP_GROWSDOWN,
398 0);
399
400 if (IS_ERR_VALUE(current->mm->start_brk)) {
401 up_write(&current->mm->mmap_sem);
402 retval = current->mm->start_brk;
403 current->mm->start_brk = 0;
404 goto error_kill;
405 }
406
407 up_write(&current->mm->mmap_sem);
408
409 current->mm->brk = current->mm->start_brk;
410 current->mm->context.end_brk = current->mm->start_brk;
411 current->mm->context.end_brk +=
412 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
413 current->mm->start_stack = current->mm->start_brk + stack_size;
414 #endif
415
416 install_exec_creds(bprm);
417 current->flags &= ~PF_FORKNOEXEC;
418 if (create_elf_fdpic_tables(bprm, current->mm,
419 &exec_params, &interp_params) < 0)
420 goto error_kill;
421
422 kdebug("- start_code %lx", current->mm->start_code);
423 kdebug("- end_code %lx", current->mm->end_code);
424 kdebug("- start_data %lx", current->mm->start_data);
425 kdebug("- end_data %lx", current->mm->end_data);
426 kdebug("- start_brk %lx", current->mm->start_brk);
427 kdebug("- brk %lx", current->mm->brk);
428 kdebug("- start_stack %lx", current->mm->start_stack);
429
430 #ifdef ELF_FDPIC_PLAT_INIT
431 /*
432 * The ABI may specify that certain registers be set up in special
433 * ways (on i386 %edx is the address of a DT_FINI function, for
434 * example. This macro performs whatever initialization to
435 * the regs structure is required.
436 */
437 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
438 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
439 dynaddr);
440 #endif
441
442 /* everything is now ready... get the userspace context ready to roll */
443 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
444 start_thread(regs, entryaddr, current->mm->start_stack);
445
446 retval = 0;
447
448 error:
449 if (interpreter) {
450 allow_write_access(interpreter);
451 fput(interpreter);
452 }
453 kfree(interpreter_name);
454 kfree(exec_params.phdrs);
455 kfree(exec_params.loadmap);
456 kfree(interp_params.phdrs);
457 kfree(interp_params.loadmap);
458 return retval;
459
460 /* unrecoverable error - kill the process */
461 error_kill:
462 send_sig(SIGSEGV, current, 0);
463 goto error;
464
465 }
466
467 /*****************************************************************************/
468
469 #ifndef ELF_BASE_PLATFORM
470 /*
471 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
472 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
473 * will be copied to the user stack in the same manner as AT_PLATFORM.
474 */
475 #define ELF_BASE_PLATFORM NULL
476 #endif
477
478 /*
479 * present useful information to the program by shovelling it onto the new
480 * process's stack
481 */
482 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
483 struct mm_struct *mm,
484 struct elf_fdpic_params *exec_params,
485 struct elf_fdpic_params *interp_params)
486 {
487 const struct cred *cred = current_cred();
488 unsigned long sp, csp, nitems;
489 elf_caddr_t __user *argv, *envp;
490 size_t platform_len = 0, len;
491 char *k_platform, *k_base_platform;
492 char __user *u_platform, *u_base_platform, *p;
493 long hwcap;
494 int loop;
495 int nr; /* reset for each csp adjustment */
496
497 #ifdef CONFIG_MMU
498 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
499 * by the processes running on the same package. One thing we can do is
500 * to shuffle the initial stack for them, so we give the architecture
501 * an opportunity to do so here.
502 */
503 sp = arch_align_stack(bprm->p);
504 #else
505 sp = mm->start_stack;
506
507 /* stack the program arguments and environment */
508 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
509 return -EFAULT;
510 #endif
511
512 hwcap = ELF_HWCAP;
513
514 /*
515 * If this architecture has a platform capability string, copy it
516 * to userspace. In some cases (Sparc), this info is impossible
517 * for userspace to get any other way, in others (i386) it is
518 * merely difficult.
519 */
520 k_platform = ELF_PLATFORM;
521 u_platform = NULL;
522
523 if (k_platform) {
524 platform_len = strlen(k_platform) + 1;
525 sp -= platform_len;
526 u_platform = (char __user *) sp;
527 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
528 return -EFAULT;
529 }
530
531 /*
532 * If this architecture has a "base" platform capability
533 * string, copy it to userspace.
534 */
535 k_base_platform = ELF_BASE_PLATFORM;
536 u_base_platform = NULL;
537
538 if (k_base_platform) {
539 platform_len = strlen(k_base_platform) + 1;
540 sp -= platform_len;
541 u_base_platform = (char __user *) sp;
542 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
543 return -EFAULT;
544 }
545
546 sp &= ~7UL;
547
548 /* stack the load map(s) */
549 len = sizeof(struct elf32_fdpic_loadmap);
550 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
551 sp = (sp - len) & ~7UL;
552 exec_params->map_addr = sp;
553
554 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
555 return -EFAULT;
556
557 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
558
559 if (interp_params->loadmap) {
560 len = sizeof(struct elf32_fdpic_loadmap);
561 len += sizeof(struct elf32_fdpic_loadseg) *
562 interp_params->loadmap->nsegs;
563 sp = (sp - len) & ~7UL;
564 interp_params->map_addr = sp;
565
566 if (copy_to_user((void __user *) sp, interp_params->loadmap,
567 len) != 0)
568 return -EFAULT;
569
570 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
571 }
572
573 /* force 16 byte _final_ alignment here for generality */
574 #define DLINFO_ITEMS 15
575
576 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
577 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
578
579 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
580 nitems++;
581
582 csp = sp;
583 sp -= nitems * 2 * sizeof(unsigned long);
584 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
585 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
586 sp -= 1 * sizeof(unsigned long); /* argc */
587
588 csp -= sp & 15UL;
589 sp -= sp & 15UL;
590
591 /* put the ELF interpreter info on the stack */
592 #define NEW_AUX_ENT(id, val) \
593 do { \
594 struct { unsigned long _id, _val; } __user *ent; \
595 \
596 ent = (void __user *) csp; \
597 __put_user((id), &ent[nr]._id); \
598 __put_user((val), &ent[nr]._val); \
599 nr++; \
600 } while (0)
601
602 nr = 0;
603 csp -= 2 * sizeof(unsigned long);
604 NEW_AUX_ENT(AT_NULL, 0);
605 if (k_platform) {
606 nr = 0;
607 csp -= 2 * sizeof(unsigned long);
608 NEW_AUX_ENT(AT_PLATFORM,
609 (elf_addr_t) (unsigned long) u_platform);
610 }
611
612 if (k_base_platform) {
613 nr = 0;
614 csp -= 2 * sizeof(unsigned long);
615 NEW_AUX_ENT(AT_BASE_PLATFORM,
616 (elf_addr_t) (unsigned long) u_base_platform);
617 }
618
619 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
620 nr = 0;
621 csp -= 2 * sizeof(unsigned long);
622 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
623 }
624
625 nr = 0;
626 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
627 NEW_AUX_ENT(AT_HWCAP, hwcap);
628 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
629 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
630 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
631 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
632 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
633 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
634 NEW_AUX_ENT(AT_FLAGS, 0);
635 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
636 NEW_AUX_ENT(AT_UID, (elf_addr_t) cred->uid);
637 NEW_AUX_ENT(AT_EUID, (elf_addr_t) cred->euid);
638 NEW_AUX_ENT(AT_GID, (elf_addr_t) cred->gid);
639 NEW_AUX_ENT(AT_EGID, (elf_addr_t) cred->egid);
640 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
641 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
642
643 #ifdef ARCH_DLINFO
644 nr = 0;
645 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
646
647 /* ARCH_DLINFO must come last so platform specific code can enforce
648 * special alignment requirements on the AUXV if necessary (eg. PPC).
649 */
650 ARCH_DLINFO;
651 #endif
652 #undef NEW_AUX_ENT
653
654 /* allocate room for argv[] and envv[] */
655 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
656 envp = (elf_caddr_t __user *) csp;
657 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
658 argv = (elf_caddr_t __user *) csp;
659
660 /* stack argc */
661 csp -= sizeof(unsigned long);
662 __put_user(bprm->argc, (unsigned long __user *) csp);
663
664 BUG_ON(csp != sp);
665
666 /* fill in the argv[] array */
667 #ifdef CONFIG_MMU
668 current->mm->arg_start = bprm->p;
669 #else
670 current->mm->arg_start = current->mm->start_stack -
671 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
672 #endif
673
674 p = (char __user *) current->mm->arg_start;
675 for (loop = bprm->argc; loop > 0; loop--) {
676 __put_user((elf_caddr_t) p, argv++);
677 len = strnlen_user(p, MAX_ARG_STRLEN);
678 if (!len || len > MAX_ARG_STRLEN)
679 return -EINVAL;
680 p += len;
681 }
682 __put_user(NULL, argv);
683 current->mm->arg_end = (unsigned long) p;
684
685 /* fill in the envv[] array */
686 current->mm->env_start = (unsigned long) p;
687 for (loop = bprm->envc; loop > 0; loop--) {
688 __put_user((elf_caddr_t)(unsigned long) p, envp++);
689 len = strnlen_user(p, MAX_ARG_STRLEN);
690 if (!len || len > MAX_ARG_STRLEN)
691 return -EINVAL;
692 p += len;
693 }
694 __put_user(NULL, envp);
695 current->mm->env_end = (unsigned long) p;
696
697 mm->start_stack = (unsigned long) sp;
698 return 0;
699 }
700
701 /*****************************************************************************/
702 /*
703 * transfer the program arguments and environment from the holding pages onto
704 * the stack
705 */
706 #ifndef CONFIG_MMU
707 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
708 unsigned long *_sp)
709 {
710 unsigned long index, stop, sp;
711 char *src;
712 int ret = 0;
713
714 stop = bprm->p >> PAGE_SHIFT;
715 sp = *_sp;
716
717 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
718 src = kmap(bprm->page[index]);
719 sp -= PAGE_SIZE;
720 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
721 ret = -EFAULT;
722 kunmap(bprm->page[index]);
723 if (ret < 0)
724 goto out;
725 }
726
727 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
728
729 out:
730 return ret;
731 }
732 #endif
733
734 /*****************************************************************************/
735 /*
736 * load the appropriate binary image (executable or interpreter) into memory
737 * - we assume no MMU is available
738 * - if no other PIC bits are set in params->hdr->e_flags
739 * - we assume that the LOADable segments in the binary are independently relocatable
740 * - we assume R/O executable segments are shareable
741 * - else
742 * - we assume the loadable parts of the image to require fixed displacement
743 * - the image is not shareable
744 */
745 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
746 struct file *file,
747 struct mm_struct *mm,
748 const char *what)
749 {
750 struct elf32_fdpic_loadmap *loadmap;
751 #ifdef CONFIG_MMU
752 struct elf32_fdpic_loadseg *mseg;
753 #endif
754 struct elf32_fdpic_loadseg *seg;
755 struct elf32_phdr *phdr;
756 unsigned long load_addr, stop;
757 unsigned nloads, tmp;
758 size_t size;
759 int loop, ret;
760
761 /* allocate a load map table */
762 nloads = 0;
763 for (loop = 0; loop < params->hdr.e_phnum; loop++)
764 if (params->phdrs[loop].p_type == PT_LOAD)
765 nloads++;
766
767 if (nloads == 0)
768 return -ELIBBAD;
769
770 size = sizeof(*loadmap) + nloads * sizeof(*seg);
771 loadmap = kzalloc(size, GFP_KERNEL);
772 if (!loadmap)
773 return -ENOMEM;
774
775 params->loadmap = loadmap;
776
777 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
778 loadmap->nsegs = nloads;
779
780 load_addr = params->load_addr;
781 seg = loadmap->segs;
782
783 /* map the requested LOADs into the memory space */
784 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
785 case ELF_FDPIC_FLAG_CONSTDISP:
786 case ELF_FDPIC_FLAG_CONTIGUOUS:
787 #ifndef CONFIG_MMU
788 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
789 if (ret < 0)
790 return ret;
791 break;
792 #endif
793 default:
794 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
795 if (ret < 0)
796 return ret;
797 break;
798 }
799
800 /* map the entry point */
801 if (params->hdr.e_entry) {
802 seg = loadmap->segs;
803 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
804 if (params->hdr.e_entry >= seg->p_vaddr &&
805 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
806 params->entry_addr =
807 (params->hdr.e_entry - seg->p_vaddr) +
808 seg->addr;
809 break;
810 }
811 }
812 }
813
814 /* determine where the program header table has wound up if mapped */
815 stop = params->hdr.e_phoff;
816 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
817 phdr = params->phdrs;
818
819 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
820 if (phdr->p_type != PT_LOAD)
821 continue;
822
823 if (phdr->p_offset > params->hdr.e_phoff ||
824 phdr->p_offset + phdr->p_filesz < stop)
825 continue;
826
827 seg = loadmap->segs;
828 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
829 if (phdr->p_vaddr >= seg->p_vaddr &&
830 phdr->p_vaddr + phdr->p_filesz <=
831 seg->p_vaddr + seg->p_memsz) {
832 params->ph_addr =
833 (phdr->p_vaddr - seg->p_vaddr) +
834 seg->addr +
835 params->hdr.e_phoff - phdr->p_offset;
836 break;
837 }
838 }
839 break;
840 }
841
842 /* determine where the dynamic section has wound up if there is one */
843 phdr = params->phdrs;
844 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
845 if (phdr->p_type != PT_DYNAMIC)
846 continue;
847
848 seg = loadmap->segs;
849 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
850 if (phdr->p_vaddr >= seg->p_vaddr &&
851 phdr->p_vaddr + phdr->p_memsz <=
852 seg->p_vaddr + seg->p_memsz) {
853 params->dynamic_addr =
854 (phdr->p_vaddr - seg->p_vaddr) +
855 seg->addr;
856
857 /* check the dynamic section contains at least
858 * one item, and that the last item is a NULL
859 * entry */
860 if (phdr->p_memsz == 0 ||
861 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
862 goto dynamic_error;
863
864 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
865 if (((Elf32_Dyn *)
866 params->dynamic_addr)[tmp - 1].d_tag != 0)
867 goto dynamic_error;
868 break;
869 }
870 }
871 break;
872 }
873
874 /* now elide adjacent segments in the load map on MMU linux
875 * - on uClinux the holes between may actually be filled with system
876 * stuff or stuff from other processes
877 */
878 #ifdef CONFIG_MMU
879 nloads = loadmap->nsegs;
880 mseg = loadmap->segs;
881 seg = mseg + 1;
882 for (loop = 1; loop < nloads; loop++) {
883 /* see if we have a candidate for merging */
884 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
885 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
886 if (load_addr == (seg->addr & PAGE_MASK)) {
887 mseg->p_memsz +=
888 load_addr -
889 (mseg->addr + mseg->p_memsz);
890 mseg->p_memsz += seg->addr & ~PAGE_MASK;
891 mseg->p_memsz += seg->p_memsz;
892 loadmap->nsegs--;
893 continue;
894 }
895 }
896
897 mseg++;
898 if (mseg != seg)
899 *mseg = *seg;
900 }
901 #endif
902
903 kdebug("Mapped Object [%s]:", what);
904 kdebug("- elfhdr : %lx", params->elfhdr_addr);
905 kdebug("- entry : %lx", params->entry_addr);
906 kdebug("- PHDR[] : %lx", params->ph_addr);
907 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
908 seg = loadmap->segs;
909 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
910 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
911 loop,
912 seg->addr, seg->addr + seg->p_memsz - 1,
913 seg->p_vaddr, seg->p_memsz);
914
915 return 0;
916
917 dynamic_error:
918 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
919 what, file->f_path.dentry->d_inode->i_ino);
920 return -ELIBBAD;
921 }
922
923 /*****************************************************************************/
924 /*
925 * map a file with constant displacement under uClinux
926 */
927 #ifndef CONFIG_MMU
928 static int elf_fdpic_map_file_constdisp_on_uclinux(
929 struct elf_fdpic_params *params,
930 struct file *file,
931 struct mm_struct *mm)
932 {
933 struct elf32_fdpic_loadseg *seg;
934 struct elf32_phdr *phdr;
935 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
936 loff_t fpos;
937 int loop, ret;
938
939 load_addr = params->load_addr;
940 seg = params->loadmap->segs;
941
942 /* determine the bounds of the contiguous overall allocation we must
943 * make */
944 phdr = params->phdrs;
945 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
946 if (params->phdrs[loop].p_type != PT_LOAD)
947 continue;
948
949 if (base > phdr->p_vaddr)
950 base = phdr->p_vaddr;
951 if (top < phdr->p_vaddr + phdr->p_memsz)
952 top = phdr->p_vaddr + phdr->p_memsz;
953 }
954
955 /* allocate one big anon block for everything */
956 mflags = MAP_PRIVATE;
957 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
958 mflags |= MAP_EXECUTABLE;
959
960 down_write(&mm->mmap_sem);
961 maddr = do_mmap(NULL, load_addr, top - base,
962 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
963 up_write(&mm->mmap_sem);
964 if (IS_ERR_VALUE(maddr))
965 return (int) maddr;
966
967 if (load_addr != 0)
968 load_addr += PAGE_ALIGN(top - base);
969
970 /* and then load the file segments into it */
971 phdr = params->phdrs;
972 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
973 if (params->phdrs[loop].p_type != PT_LOAD)
974 continue;
975
976 fpos = phdr->p_offset;
977
978 seg->addr = maddr + (phdr->p_vaddr - base);
979 seg->p_vaddr = phdr->p_vaddr;
980 seg->p_memsz = phdr->p_memsz;
981
982 ret = file->f_op->read(file, (void *) seg->addr,
983 phdr->p_filesz, &fpos);
984 if (ret < 0)
985 return ret;
986
987 /* map the ELF header address if in this segment */
988 if (phdr->p_offset == 0)
989 params->elfhdr_addr = seg->addr;
990
991 /* clear any space allocated but not loaded */
992 if (phdr->p_filesz < phdr->p_memsz) {
993 if (clear_user((void *) (seg->addr + phdr->p_filesz),
994 phdr->p_memsz - phdr->p_filesz))
995 return -EFAULT;
996 }
997
998 if (mm) {
999 if (phdr->p_flags & PF_X) {
1000 if (!mm->start_code) {
1001 mm->start_code = seg->addr;
1002 mm->end_code = seg->addr +
1003 phdr->p_memsz;
1004 }
1005 } else if (!mm->start_data) {
1006 mm->start_data = seg->addr;
1007 mm->end_data = seg->addr + phdr->p_memsz;
1008 }
1009 }
1010
1011 seg++;
1012 }
1013
1014 return 0;
1015 }
1016 #endif
1017
1018 /*****************************************************************************/
1019 /*
1020 * map a binary by direct mmap() of the individual PT_LOAD segments
1021 */
1022 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1023 struct file *file,
1024 struct mm_struct *mm)
1025 {
1026 struct elf32_fdpic_loadseg *seg;
1027 struct elf32_phdr *phdr;
1028 unsigned long load_addr, delta_vaddr;
1029 int loop, dvset;
1030
1031 load_addr = params->load_addr;
1032 delta_vaddr = 0;
1033 dvset = 0;
1034
1035 seg = params->loadmap->segs;
1036
1037 /* deal with each load segment separately */
1038 phdr = params->phdrs;
1039 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1040 unsigned long maddr, disp, excess, excess1;
1041 int prot = 0, flags;
1042
1043 if (phdr->p_type != PT_LOAD)
1044 continue;
1045
1046 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1047 (unsigned long) phdr->p_vaddr,
1048 (unsigned long) phdr->p_offset,
1049 (unsigned long) phdr->p_filesz,
1050 (unsigned long) phdr->p_memsz);
1051
1052 /* determine the mapping parameters */
1053 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1054 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1055 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1056
1057 flags = MAP_PRIVATE | MAP_DENYWRITE;
1058 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1059 flags |= MAP_EXECUTABLE;
1060
1061 maddr = 0;
1062
1063 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1064 case ELF_FDPIC_FLAG_INDEPENDENT:
1065 /* PT_LOADs are independently locatable */
1066 break;
1067
1068 case ELF_FDPIC_FLAG_HONOURVADDR:
1069 /* the specified virtual address must be honoured */
1070 maddr = phdr->p_vaddr;
1071 flags |= MAP_FIXED;
1072 break;
1073
1074 case ELF_FDPIC_FLAG_CONSTDISP:
1075 /* constant displacement
1076 * - can be mapped anywhere, but must be mapped as a
1077 * unit
1078 */
1079 if (!dvset) {
1080 maddr = load_addr;
1081 delta_vaddr = phdr->p_vaddr;
1082 dvset = 1;
1083 } else {
1084 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1085 flags |= MAP_FIXED;
1086 }
1087 break;
1088
1089 case ELF_FDPIC_FLAG_CONTIGUOUS:
1090 /* contiguity handled later */
1091 break;
1092
1093 default:
1094 BUG();
1095 }
1096
1097 maddr &= PAGE_MASK;
1098
1099 /* create the mapping */
1100 disp = phdr->p_vaddr & ~PAGE_MASK;
1101 down_write(&mm->mmap_sem);
1102 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1103 phdr->p_offset - disp);
1104 up_write(&mm->mmap_sem);
1105
1106 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1107 loop, phdr->p_memsz + disp, prot, flags,
1108 phdr->p_offset - disp, maddr);
1109
1110 if (IS_ERR_VALUE(maddr))
1111 return (int) maddr;
1112
1113 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1114 ELF_FDPIC_FLAG_CONTIGUOUS)
1115 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1116
1117 seg->addr = maddr + disp;
1118 seg->p_vaddr = phdr->p_vaddr;
1119 seg->p_memsz = phdr->p_memsz;
1120
1121 /* map the ELF header address if in this segment */
1122 if (phdr->p_offset == 0)
1123 params->elfhdr_addr = seg->addr;
1124
1125 /* clear the bit between beginning of mapping and beginning of
1126 * PT_LOAD */
1127 if (prot & PROT_WRITE && disp > 0) {
1128 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1129 if (clear_user((void __user *) maddr, disp))
1130 return -EFAULT;
1131 maddr += disp;
1132 }
1133
1134 /* clear any space allocated but not loaded
1135 * - on uClinux we can just clear the lot
1136 * - on MMU linux we'll get a SIGBUS beyond the last page
1137 * extant in the file
1138 */
1139 excess = phdr->p_memsz - phdr->p_filesz;
1140 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1141
1142 #ifdef CONFIG_MMU
1143 if (excess > excess1) {
1144 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1145 unsigned long xmaddr;
1146
1147 flags |= MAP_FIXED | MAP_ANONYMOUS;
1148 down_write(&mm->mmap_sem);
1149 xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1150 prot, flags, 0);
1151 up_write(&mm->mmap_sem);
1152
1153 kdebug("mmap[%d] <anon>"
1154 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1155 loop, xaddr, excess - excess1, prot, flags,
1156 xmaddr);
1157
1158 if (xmaddr != xaddr)
1159 return -ENOMEM;
1160 }
1161
1162 if (prot & PROT_WRITE && excess1 > 0) {
1163 kdebug("clear[%d] ad=%lx sz=%lx",
1164 loop, maddr + phdr->p_filesz, excess1);
1165 if (clear_user((void __user *) maddr + phdr->p_filesz,
1166 excess1))
1167 return -EFAULT;
1168 }
1169
1170 #else
1171 if (excess > 0) {
1172 kdebug("clear[%d] ad=%lx sz=%lx",
1173 loop, maddr + phdr->p_filesz, excess);
1174 if (clear_user((void *) maddr + phdr->p_filesz, excess))
1175 return -EFAULT;
1176 }
1177 #endif
1178
1179 if (mm) {
1180 if (phdr->p_flags & PF_X) {
1181 if (!mm->start_code) {
1182 mm->start_code = maddr;
1183 mm->end_code = maddr + phdr->p_memsz;
1184 }
1185 } else if (!mm->start_data) {
1186 mm->start_data = maddr;
1187 mm->end_data = maddr + phdr->p_memsz;
1188 }
1189 }
1190
1191 seg++;
1192 }
1193
1194 return 0;
1195 }
1196
1197 /*****************************************************************************/
1198 /*
1199 * ELF-FDPIC core dumper
1200 *
1201 * Modelled on fs/exec.c:aout_core_dump()
1202 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1203 *
1204 * Modelled on fs/binfmt_elf.c core dumper
1205 */
1206 #ifdef CONFIG_ELF_CORE
1207
1208 /*
1209 * Decide whether a segment is worth dumping; default is yes to be
1210 * sure (missing info is worse than too much; etc).
1211 * Personally I'd include everything, and use the coredump limit...
1212 *
1213 * I think we should skip something. But I am not sure how. H.J.
1214 */
1215 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1216 {
1217 int dump_ok;
1218
1219 /* Do not dump I/O mapped devices or special mappings */
1220 if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1221 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1222 return 0;
1223 }
1224
1225 /* If we may not read the contents, don't allow us to dump
1226 * them either. "dump_write()" can't handle it anyway.
1227 */
1228 if (!(vma->vm_flags & VM_READ)) {
1229 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1230 return 0;
1231 }
1232
1233 /* By default, dump shared memory if mapped from an anonymous file. */
1234 if (vma->vm_flags & VM_SHARED) {
1235 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1236 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1237 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1238 vma->vm_flags, dump_ok ? "yes" : "no");
1239 return dump_ok;
1240 }
1241
1242 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1243 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1244 vma->vm_flags, dump_ok ? "yes" : "no");
1245 return dump_ok;
1246 }
1247
1248 #ifdef CONFIG_MMU
1249 /* By default, if it hasn't been written to, don't write it out */
1250 if (!vma->anon_vma) {
1251 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1252 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1253 vma->vm_flags, dump_ok ? "yes" : "no");
1254 return dump_ok;
1255 }
1256 #endif
1257
1258 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1259 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1260 dump_ok ? "yes" : "no");
1261 return dump_ok;
1262 }
1263
1264 /* An ELF note in memory */
1265 struct memelfnote
1266 {
1267 const char *name;
1268 int type;
1269 unsigned int datasz;
1270 void *data;
1271 };
1272
1273 static int notesize(struct memelfnote *en)
1274 {
1275 int sz;
1276
1277 sz = sizeof(struct elf_note);
1278 sz += roundup(strlen(en->name) + 1, 4);
1279 sz += roundup(en->datasz, 4);
1280
1281 return sz;
1282 }
1283
1284 /* #define DEBUG */
1285
1286 #define DUMP_WRITE(addr, nr, foffset) \
1287 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1288
1289 static int alignfile(struct file *file, loff_t *foffset)
1290 {
1291 static const char buf[4] = { 0, };
1292 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1293 return 1;
1294 }
1295
1296 static int writenote(struct memelfnote *men, struct file *file,
1297 loff_t *foffset)
1298 {
1299 struct elf_note en;
1300 en.n_namesz = strlen(men->name) + 1;
1301 en.n_descsz = men->datasz;
1302 en.n_type = men->type;
1303
1304 DUMP_WRITE(&en, sizeof(en), foffset);
1305 DUMP_WRITE(men->name, en.n_namesz, foffset);
1306 if (!alignfile(file, foffset))
1307 return 0;
1308 DUMP_WRITE(men->data, men->datasz, foffset);
1309 if (!alignfile(file, foffset))
1310 return 0;
1311
1312 return 1;
1313 }
1314 #undef DUMP_WRITE
1315
1316 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1317 {
1318 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1319 elf->e_ident[EI_CLASS] = ELF_CLASS;
1320 elf->e_ident[EI_DATA] = ELF_DATA;
1321 elf->e_ident[EI_VERSION] = EV_CURRENT;
1322 elf->e_ident[EI_OSABI] = ELF_OSABI;
1323 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1324
1325 elf->e_type = ET_CORE;
1326 elf->e_machine = ELF_ARCH;
1327 elf->e_version = EV_CURRENT;
1328 elf->e_entry = 0;
1329 elf->e_phoff = sizeof(struct elfhdr);
1330 elf->e_shoff = 0;
1331 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1332 elf->e_ehsize = sizeof(struct elfhdr);
1333 elf->e_phentsize = sizeof(struct elf_phdr);
1334 elf->e_phnum = segs;
1335 elf->e_shentsize = 0;
1336 elf->e_shnum = 0;
1337 elf->e_shstrndx = 0;
1338 return;
1339 }
1340
1341 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1342 {
1343 phdr->p_type = PT_NOTE;
1344 phdr->p_offset = offset;
1345 phdr->p_vaddr = 0;
1346 phdr->p_paddr = 0;
1347 phdr->p_filesz = sz;
1348 phdr->p_memsz = 0;
1349 phdr->p_flags = 0;
1350 phdr->p_align = 0;
1351 return;
1352 }
1353
1354 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1355 unsigned int sz, void *data)
1356 {
1357 note->name = name;
1358 note->type = type;
1359 note->datasz = sz;
1360 note->data = data;
1361 return;
1362 }
1363
1364 /*
1365 * fill up all the fields in prstatus from the given task struct, except
1366 * registers which need to be filled up separately.
1367 */
1368 static void fill_prstatus(struct elf_prstatus *prstatus,
1369 struct task_struct *p, long signr)
1370 {
1371 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1372 prstatus->pr_sigpend = p->pending.signal.sig[0];
1373 prstatus->pr_sighold = p->blocked.sig[0];
1374 rcu_read_lock();
1375 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1376 rcu_read_unlock();
1377 prstatus->pr_pid = task_pid_vnr(p);
1378 prstatus->pr_pgrp = task_pgrp_vnr(p);
1379 prstatus->pr_sid = task_session_vnr(p);
1380 if (thread_group_leader(p)) {
1381 struct task_cputime cputime;
1382
1383 /*
1384 * This is the record for the group leader. It shows the
1385 * group-wide total, not its individual thread total.
1386 */
1387 thread_group_cputime(p, &cputime);
1388 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1389 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1390 } else {
1391 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1392 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1393 }
1394 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1395 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1396
1397 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1398 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1399 }
1400
1401 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1402 struct mm_struct *mm)
1403 {
1404 const struct cred *cred;
1405 unsigned int i, len;
1406
1407 /* first copy the parameters from user space */
1408 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1409
1410 len = mm->arg_end - mm->arg_start;
1411 if (len >= ELF_PRARGSZ)
1412 len = ELF_PRARGSZ - 1;
1413 if (copy_from_user(&psinfo->pr_psargs,
1414 (const char __user *) mm->arg_start, len))
1415 return -EFAULT;
1416 for (i = 0; i < len; i++)
1417 if (psinfo->pr_psargs[i] == 0)
1418 psinfo->pr_psargs[i] = ' ';
1419 psinfo->pr_psargs[len] = 0;
1420
1421 rcu_read_lock();
1422 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1423 rcu_read_unlock();
1424 psinfo->pr_pid = task_pid_vnr(p);
1425 psinfo->pr_pgrp = task_pgrp_vnr(p);
1426 psinfo->pr_sid = task_session_vnr(p);
1427
1428 i = p->state ? ffz(~p->state) + 1 : 0;
1429 psinfo->pr_state = i;
1430 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1431 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1432 psinfo->pr_nice = task_nice(p);
1433 psinfo->pr_flag = p->flags;
1434 rcu_read_lock();
1435 cred = __task_cred(p);
1436 SET_UID(psinfo->pr_uid, cred->uid);
1437 SET_GID(psinfo->pr_gid, cred->gid);
1438 rcu_read_unlock();
1439 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1440
1441 return 0;
1442 }
1443
1444 /* Here is the structure in which status of each thread is captured. */
1445 struct elf_thread_status
1446 {
1447 struct list_head list;
1448 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1449 elf_fpregset_t fpu; /* NT_PRFPREG */
1450 struct task_struct *thread;
1451 #ifdef ELF_CORE_COPY_XFPREGS
1452 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1453 #endif
1454 struct memelfnote notes[3];
1455 int num_notes;
1456 };
1457
1458 /*
1459 * In order to add the specific thread information for the elf file format,
1460 * we need to keep a linked list of every thread's pr_status and then create
1461 * a single section for them in the final core file.
1462 */
1463 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1464 {
1465 struct task_struct *p = t->thread;
1466 int sz = 0;
1467
1468 t->num_notes = 0;
1469
1470 fill_prstatus(&t->prstatus, p, signr);
1471 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1472
1473 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1474 &t->prstatus);
1475 t->num_notes++;
1476 sz += notesize(&t->notes[0]);
1477
1478 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1479 if (t->prstatus.pr_fpvalid) {
1480 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1481 &t->fpu);
1482 t->num_notes++;
1483 sz += notesize(&t->notes[1]);
1484 }
1485
1486 #ifdef ELF_CORE_COPY_XFPREGS
1487 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1488 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1489 sizeof(t->xfpu), &t->xfpu);
1490 t->num_notes++;
1491 sz += notesize(&t->notes[2]);
1492 }
1493 #endif
1494 return sz;
1495 }
1496
1497 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1498 elf_addr_t e_shoff, int segs)
1499 {
1500 elf->e_shoff = e_shoff;
1501 elf->e_shentsize = sizeof(*shdr4extnum);
1502 elf->e_shnum = 1;
1503 elf->e_shstrndx = SHN_UNDEF;
1504
1505 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1506
1507 shdr4extnum->sh_type = SHT_NULL;
1508 shdr4extnum->sh_size = elf->e_shnum;
1509 shdr4extnum->sh_link = elf->e_shstrndx;
1510 shdr4extnum->sh_info = segs;
1511 }
1512
1513 /*
1514 * dump the segments for an MMU process
1515 */
1516 #ifdef CONFIG_MMU
1517 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1518 unsigned long *limit, unsigned long mm_flags)
1519 {
1520 struct vm_area_struct *vma;
1521 int err = 0;
1522
1523 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1524 unsigned long addr;
1525
1526 if (!maydump(vma, mm_flags))
1527 continue;
1528
1529 for (addr = vma->vm_start; addr < vma->vm_end;
1530 addr += PAGE_SIZE) {
1531 struct page *page = get_dump_page(addr);
1532 if (page) {
1533 void *kaddr = kmap(page);
1534 *size += PAGE_SIZE;
1535 if (*size > *limit)
1536 err = -EFBIG;
1537 else if (!dump_write(file, kaddr, PAGE_SIZE))
1538 err = -EIO;
1539 kunmap(page);
1540 page_cache_release(page);
1541 } else if (!dump_seek(file, PAGE_SIZE))
1542 err = -EFBIG;
1543 if (err)
1544 goto out;
1545 }
1546 }
1547 out:
1548 return err;
1549 }
1550 #endif
1551
1552 /*
1553 * dump the segments for a NOMMU process
1554 */
1555 #ifndef CONFIG_MMU
1556 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1557 unsigned long *limit, unsigned long mm_flags)
1558 {
1559 struct vm_area_struct *vma;
1560
1561 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1562 if (!maydump(vma, mm_flags))
1563 continue;
1564
1565 if ((*size += PAGE_SIZE) > *limit)
1566 return -EFBIG;
1567
1568 if (!dump_write(file, (void *) vma->vm_start,
1569 vma->vm_end - vma->vm_start))
1570 return -EIO;
1571 }
1572
1573 return 0;
1574 }
1575 #endif
1576
1577 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1578 {
1579 struct vm_area_struct *vma;
1580 size_t size = 0;
1581
1582 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1583 if (maydump(vma, mm_flags))
1584 size += vma->vm_end - vma->vm_start;
1585 return size;
1586 }
1587
1588 /*
1589 * Actual dumper
1590 *
1591 * This is a two-pass process; first we find the offsets of the bits,
1592 * and then they are actually written out. If we run out of core limit
1593 * we just truncate.
1594 */
1595 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1596 {
1597 #define NUM_NOTES 6
1598 int has_dumped = 0;
1599 mm_segment_t fs;
1600 int segs;
1601 size_t size = 0;
1602 int i;
1603 struct vm_area_struct *vma;
1604 struct elfhdr *elf = NULL;
1605 loff_t offset = 0, dataoff, foffset;
1606 int numnote;
1607 struct memelfnote *notes = NULL;
1608 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1609 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1610 LIST_HEAD(thread_list);
1611 struct list_head *t;
1612 elf_fpregset_t *fpu = NULL;
1613 #ifdef ELF_CORE_COPY_XFPREGS
1614 elf_fpxregset_t *xfpu = NULL;
1615 #endif
1616 int thread_status_size = 0;
1617 elf_addr_t *auxv;
1618 struct elf_phdr *phdr4note = NULL;
1619 struct elf_shdr *shdr4extnum = NULL;
1620 Elf_Half e_phnum;
1621 elf_addr_t e_shoff;
1622
1623 /*
1624 * We no longer stop all VM operations.
1625 *
1626 * This is because those proceses that could possibly change map_count
1627 * or the mmap / vma pages are now blocked in do_exit on current
1628 * finishing this core dump.
1629 *
1630 * Only ptrace can touch these memory addresses, but it doesn't change
1631 * the map_count or the pages allocated. So no possibility of crashing
1632 * exists while dumping the mm->vm_next areas to the core file.
1633 */
1634
1635 /* alloc memory for large data structures: too large to be on stack */
1636 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1637 if (!elf)
1638 goto cleanup;
1639 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1640 if (!prstatus)
1641 goto cleanup;
1642 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1643 if (!psinfo)
1644 goto cleanup;
1645 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1646 if (!notes)
1647 goto cleanup;
1648 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1649 if (!fpu)
1650 goto cleanup;
1651 #ifdef ELF_CORE_COPY_XFPREGS
1652 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1653 if (!xfpu)
1654 goto cleanup;
1655 #endif
1656
1657 if (cprm->signr) {
1658 struct core_thread *ct;
1659 struct elf_thread_status *tmp;
1660
1661 for (ct = current->mm->core_state->dumper.next;
1662 ct; ct = ct->next) {
1663 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1664 if (!tmp)
1665 goto cleanup;
1666
1667 tmp->thread = ct->task;
1668 list_add(&tmp->list, &thread_list);
1669 }
1670
1671 list_for_each(t, &thread_list) {
1672 struct elf_thread_status *tmp;
1673 int sz;
1674
1675 tmp = list_entry(t, struct elf_thread_status, list);
1676 sz = elf_dump_thread_status(cprm->signr, tmp);
1677 thread_status_size += sz;
1678 }
1679 }
1680
1681 /* now collect the dump for the current */
1682 fill_prstatus(prstatus, current, cprm->signr);
1683 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1684
1685 segs = current->mm->map_count;
1686 segs += elf_core_extra_phdrs();
1687
1688 /* for notes section */
1689 segs++;
1690
1691 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1692 * this, kernel supports extended numbering. Have a look at
1693 * include/linux/elf.h for further information. */
1694 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1695
1696 /* Set up header */
1697 fill_elf_fdpic_header(elf, e_phnum);
1698
1699 has_dumped = 1;
1700 current->flags |= PF_DUMPCORE;
1701
1702 /*
1703 * Set up the notes in similar form to SVR4 core dumps made
1704 * with info from their /proc.
1705 */
1706
1707 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1708 fill_psinfo(psinfo, current->group_leader, current->mm);
1709 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1710
1711 numnote = 2;
1712
1713 auxv = (elf_addr_t *) current->mm->saved_auxv;
1714
1715 i = 0;
1716 do
1717 i += 2;
1718 while (auxv[i - 2] != AT_NULL);
1719 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1720 i * sizeof(elf_addr_t), auxv);
1721
1722 /* Try to dump the FPU. */
1723 if ((prstatus->pr_fpvalid =
1724 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1725 fill_note(notes + numnote++,
1726 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1727 #ifdef ELF_CORE_COPY_XFPREGS
1728 if (elf_core_copy_task_xfpregs(current, xfpu))
1729 fill_note(notes + numnote++,
1730 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1731 #endif
1732
1733 fs = get_fs();
1734 set_fs(KERNEL_DS);
1735
1736 offset += sizeof(*elf); /* Elf header */
1737 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1738 foffset = offset;
1739
1740 /* Write notes phdr entry */
1741 {
1742 int sz = 0;
1743
1744 for (i = 0; i < numnote; i++)
1745 sz += notesize(notes + i);
1746
1747 sz += thread_status_size;
1748
1749 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1750 if (!phdr4note)
1751 goto end_coredump;
1752
1753 fill_elf_note_phdr(phdr4note, sz, offset);
1754 offset += sz;
1755 }
1756
1757 /* Page-align dumped data */
1758 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1759
1760 offset += elf_core_vma_data_size(cprm->mm_flags);
1761 offset += elf_core_extra_data_size();
1762 e_shoff = offset;
1763
1764 if (e_phnum == PN_XNUM) {
1765 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1766 if (!shdr4extnum)
1767 goto end_coredump;
1768 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1769 }
1770
1771 offset = dataoff;
1772
1773 size += sizeof(*elf);
1774 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1775 goto end_coredump;
1776
1777 size += sizeof(*phdr4note);
1778 if (size > cprm->limit
1779 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1780 goto end_coredump;
1781
1782 /* write program headers for segments dump */
1783 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1784 struct elf_phdr phdr;
1785 size_t sz;
1786
1787 sz = vma->vm_end - vma->vm_start;
1788
1789 phdr.p_type = PT_LOAD;
1790 phdr.p_offset = offset;
1791 phdr.p_vaddr = vma->vm_start;
1792 phdr.p_paddr = 0;
1793 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1794 phdr.p_memsz = sz;
1795 offset += phdr.p_filesz;
1796 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1797 if (vma->vm_flags & VM_WRITE)
1798 phdr.p_flags |= PF_W;
1799 if (vma->vm_flags & VM_EXEC)
1800 phdr.p_flags |= PF_X;
1801 phdr.p_align = ELF_EXEC_PAGESIZE;
1802
1803 size += sizeof(phdr);
1804 if (size > cprm->limit
1805 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
1806 goto end_coredump;
1807 }
1808
1809 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
1810 goto end_coredump;
1811
1812 /* write out the notes section */
1813 for (i = 0; i < numnote; i++)
1814 if (!writenote(notes + i, cprm->file, &foffset))
1815 goto end_coredump;
1816
1817 /* write out the thread status notes section */
1818 list_for_each(t, &thread_list) {
1819 struct elf_thread_status *tmp =
1820 list_entry(t, struct elf_thread_status, list);
1821
1822 for (i = 0; i < tmp->num_notes; i++)
1823 if (!writenote(&tmp->notes[i], cprm->file, &foffset))
1824 goto end_coredump;
1825 }
1826
1827 if (!dump_seek(cprm->file, dataoff - foffset))
1828 goto end_coredump;
1829
1830 if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit,
1831 cprm->mm_flags) < 0)
1832 goto end_coredump;
1833
1834 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
1835 goto end_coredump;
1836
1837 if (e_phnum == PN_XNUM) {
1838 size += sizeof(*shdr4extnum);
1839 if (size > cprm->limit
1840 || !dump_write(cprm->file, shdr4extnum,
1841 sizeof(*shdr4extnum)))
1842 goto end_coredump;
1843 }
1844
1845 if (cprm->file->f_pos != offset) {
1846 /* Sanity check */
1847 printk(KERN_WARNING
1848 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1849 cprm->file->f_pos, offset);
1850 }
1851
1852 end_coredump:
1853 set_fs(fs);
1854
1855 cleanup:
1856 while (!list_empty(&thread_list)) {
1857 struct list_head *tmp = thread_list.next;
1858 list_del(tmp);
1859 kfree(list_entry(tmp, struct elf_thread_status, list));
1860 }
1861 kfree(phdr4note);
1862 kfree(elf);
1863 kfree(prstatus);
1864 kfree(psinfo);
1865 kfree(notes);
1866 kfree(fpu);
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 */
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree