?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
deal with races in /proc/*/{syscall,stack,personality}
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / binfmt_elf_fdpic.c
blobe7a0bb47252c75dc50c437c3bdd5ca69ed73cdb2
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2 *
3 * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 * Derived from binfmt_elf.c
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13 #include <linux/module.h>
14
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/param.h>
40 #include <asm/pgalloc.h>
41
42 typedef char *elf_caddr_t;
43
44 #if 0
45 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
46 #else
47 #define kdebug(fmt, ...) do {} while(0)
48 #endif
49
50 #if 0
51 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
52 #else
53 #define kdcore(fmt, ...) do {} while(0)
54 #endif
55
56 MODULE_LICENSE("GPL");
57
58 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
59 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
60 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
61 struct mm_struct *, const char *);
62
63 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
64 struct elf_fdpic_params *,
65 struct elf_fdpic_params *);
66
67 #ifndef CONFIG_MMU
68 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
69 unsigned long *);
70 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
71 struct file *,
72 struct mm_struct *);
73 #endif
74
75 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
76 struct file *, struct mm_struct *);
77
78 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
79 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *, unsigned long limit);
80 #endif
81
82 static struct linux_binfmt elf_fdpic_format = {
83 .module = THIS_MODULE,
84 .load_binary = load_elf_fdpic_binary,
85 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
86 .core_dump = elf_fdpic_core_dump,
87 #endif
88 .min_coredump = ELF_EXEC_PAGESIZE,
89 };
90
91 static int __init init_elf_fdpic_binfmt(void)
92 {
93 return register_binfmt(&elf_fdpic_format);
94 }
95
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 unregister_binfmt(&elf_fdpic_format);
99 }
100
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103
104 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
105 {
106 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 return 0;
108 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 return 0;
110 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
111 return 0;
112 if (!file->f_op || !file->f_op->mmap)
113 return 0;
114 return 1;
115 }
116
117 /*****************************************************************************/
118 /*
119 * read the program headers table into memory
120 */
121 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
122 struct file *file)
123 {
124 struct elf32_phdr *phdr;
125 unsigned long size;
126 int retval, loop;
127
128 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
129 return -ENOMEM;
130 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
131 return -ENOMEM;
132
133 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
134 params->phdrs = kmalloc(size, GFP_KERNEL);
135 if (!params->phdrs)
136 return -ENOMEM;
137
138 retval = kernel_read(file, params->hdr.e_phoff,
139 (char *) params->phdrs, size);
140 if (unlikely(retval != size))
141 return retval < 0 ? retval : -ENOEXEC;
142
143 /* determine stack size for this binary */
144 phdr = params->phdrs;
145 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
146 if (phdr->p_type != PT_GNU_STACK)
147 continue;
148
149 if (phdr->p_flags & PF_X)
150 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
151 else
152 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
153
154 params->stack_size = phdr->p_memsz;
155 break;
156 }
157
158 return 0;
159 }
160
161 /*****************************************************************************/
162 /*
163 * load an fdpic binary into various bits of memory
164 */
165 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
166 struct pt_regs *regs)
167 {
168 struct elf_fdpic_params exec_params, interp_params;
169 struct elf_phdr *phdr;
170 unsigned long stack_size, entryaddr;
171 #ifdef ELF_FDPIC_PLAT_INIT
172 unsigned long dynaddr;
173 #endif
174 #ifndef CONFIG_MMU
175 unsigned long stack_prot;
176 #endif
177 struct file *interpreter = NULL; /* to shut gcc up */
178 char *interpreter_name = NULL;
179 int executable_stack;
180 int retval, i;
181
182 kdebug("____ LOAD %d ____", current->pid);
183
184 memset(&exec_params, 0, sizeof(exec_params));
185 memset(&interp_params, 0, sizeof(interp_params));
186
187 exec_params.hdr = *(struct elfhdr *) bprm->buf;
188 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
189
190 /* check that this is a binary we know how to deal with */
191 retval = -ENOEXEC;
192 if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
193 goto error;
194
195 /* read the program header table */
196 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
197 if (retval < 0)
198 goto error;
199
200 /* scan for a program header that specifies an interpreter */
201 phdr = exec_params.phdrs;
202
203 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
204 switch (phdr->p_type) {
205 case PT_INTERP:
206 retval = -ENOMEM;
207 if (phdr->p_filesz > PATH_MAX)
208 goto error;
209 retval = -ENOENT;
210 if (phdr->p_filesz < 2)
211 goto error;
212
213 /* read the name of the interpreter into memory */
214 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
215 if (!interpreter_name)
216 goto error;
217
218 retval = kernel_read(bprm->file,
219 phdr->p_offset,
220 interpreter_name,
221 phdr->p_filesz);
222 if (unlikely(retval != phdr->p_filesz)) {
223 if (retval >= 0)
224 retval = -ENOEXEC;
225 goto error;
226 }
227
228 retval = -ENOENT;
229 if (interpreter_name[phdr->p_filesz - 1] != '\0')
230 goto error;
231
232 kdebug("Using ELF interpreter %s", interpreter_name);
233
234 /* replace the program with the interpreter */
235 interpreter = open_exec(interpreter_name);
236 retval = PTR_ERR(interpreter);
237 if (IS_ERR(interpreter)) {
238 interpreter = NULL;
239 goto error;
240 }
241
242 /*
243 * If the binary is not readable then enforce
244 * mm->dumpable = 0 regardless of the interpreter's
245 * permissions.
246 */
247 if (file_permission(interpreter, MAY_READ) < 0)
248 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
249
250 retval = kernel_read(interpreter, 0, bprm->buf,
251 BINPRM_BUF_SIZE);
252 if (unlikely(retval != BINPRM_BUF_SIZE)) {
253 if (retval >= 0)
254 retval = -ENOEXEC;
255 goto error;
256 }
257
258 interp_params.hdr = *((struct elfhdr *) bprm->buf);
259 break;
260
261 case PT_LOAD:
262 #ifdef CONFIG_MMU
263 if (exec_params.load_addr == 0)
264 exec_params.load_addr = phdr->p_vaddr;
265 #endif
266 break;
267 }
268
269 }
270
271 if (elf_check_const_displacement(&exec_params.hdr))
272 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
273
274 /* perform insanity checks on the interpreter */
275 if (interpreter_name) {
276 retval = -ELIBBAD;
277 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
278 goto error;
279
280 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
281
282 /* read the interpreter's program header table */
283 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
284 if (retval < 0)
285 goto error;
286 }
287
288 stack_size = exec_params.stack_size;
289 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
290 executable_stack = EXSTACK_ENABLE_X;
291 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
292 executable_stack = EXSTACK_DISABLE_X;
293 else
294 executable_stack = EXSTACK_DEFAULT;
295
296 if (stack_size == 0) {
297 stack_size = interp_params.stack_size;
298 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
299 executable_stack = EXSTACK_ENABLE_X;
300 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
301 executable_stack = EXSTACK_DISABLE_X;
302 else
303 executable_stack = EXSTACK_DEFAULT;
304 }
305
306 retval = -ENOEXEC;
307 if (stack_size == 0)
308 goto error;
309
310 if (elf_check_const_displacement(&interp_params.hdr))
311 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
312
313 /* flush all traces of the currently running executable */
314 retval = flush_old_exec(bprm);
315 if (retval)
316 goto error;
317
318 /* there's now no turning back... the old userspace image is dead,
319 * defunct, deceased, etc. after this point we have to exit via
320 * error_kill */
321 set_personality(PER_LINUX_FDPIC);
322 if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
323 current->personality |= READ_IMPLIES_EXEC;
324
325 setup_new_exec(bprm);
326
327 set_binfmt(&elf_fdpic_format);
328
329 current->mm->start_code = 0;
330 current->mm->end_code = 0;
331 current->mm->start_stack = 0;
332 current->mm->start_data = 0;
333 current->mm->end_data = 0;
334 current->mm->context.exec_fdpic_loadmap = 0;
335 current->mm->context.interp_fdpic_loadmap = 0;
336
337 current->flags &= ~PF_FORKNOEXEC;
338
339 #ifdef CONFIG_MMU
340 elf_fdpic_arch_lay_out_mm(&exec_params,
341 &interp_params,
342 &current->mm->start_stack,
343 &current->mm->start_brk);
344
345 retval = setup_arg_pages(bprm, current->mm->start_stack,
346 executable_stack);
347 if (retval < 0) {
348 send_sig(SIGKILL, current, 0);
349 goto error_kill;
350 }
351 #endif
352
353 /* load the executable and interpreter into memory */
354 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
355 "executable");
356 if (retval < 0)
357 goto error_kill;
358
359 if (interpreter_name) {
360 retval = elf_fdpic_map_file(&interp_params, interpreter,
361 current->mm, "interpreter");
362 if (retval < 0) {
363 printk(KERN_ERR "Unable to load interpreter\n");
364 goto error_kill;
365 }
366
367 allow_write_access(interpreter);
368 fput(interpreter);
369 interpreter = NULL;
370 }
371
372 #ifdef CONFIG_MMU
373 if (!current->mm->start_brk)
374 current->mm->start_brk = current->mm->end_data;
375
376 current->mm->brk = current->mm->start_brk =
377 PAGE_ALIGN(current->mm->start_brk);
378
379 #else
380 /* create a stack and brk area big enough for everyone
381 * - the brk heap starts at the bottom and works up
382 * - the stack starts at the top and works down
383 */
384 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
385 if (stack_size < PAGE_SIZE * 2)
386 stack_size = PAGE_SIZE * 2;
387
388 stack_prot = PROT_READ | PROT_WRITE;
389 if (executable_stack == EXSTACK_ENABLE_X ||
390 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
391 stack_prot |= PROT_EXEC;
392
393 down_write(&current->mm->mmap_sem);
394 current->mm->start_brk = do_mmap(NULL, 0, stack_size, stack_prot,
395 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
396 0);
397
398 if (IS_ERR_VALUE(current->mm->start_brk)) {
399 up_write(&current->mm->mmap_sem);
400 retval = current->mm->start_brk;
401 current->mm->start_brk = 0;
402 goto error_kill;
403 }
404
405 up_write(&current->mm->mmap_sem);
406
407 current->mm->brk = current->mm->start_brk;
408 current->mm->context.end_brk = current->mm->start_brk;
409 current->mm->context.end_brk +=
410 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
411 current->mm->start_stack = current->mm->start_brk + stack_size;
412 #endif
413
414 install_exec_creds(bprm);
415 current->flags &= ~PF_FORKNOEXEC;
416 if (create_elf_fdpic_tables(bprm, current->mm,
417 &exec_params, &interp_params) < 0)
418 goto error_kill;
419
420 kdebug("- start_code %lx", current->mm->start_code);
421 kdebug("- end_code %lx", current->mm->end_code);
422 kdebug("- start_data %lx", current->mm->start_data);
423 kdebug("- end_data %lx", current->mm->end_data);
424 kdebug("- start_brk %lx", current->mm->start_brk);
425 kdebug("- brk %lx", current->mm->brk);
426 kdebug("- start_stack %lx", current->mm->start_stack);
427
428 #ifdef ELF_FDPIC_PLAT_INIT
429 /*
430 * The ABI may specify that certain registers be set up in special
431 * ways (on i386 %edx is the address of a DT_FINI function, for
432 * example. This macro performs whatever initialization to
433 * the regs structure is required.
434 */
435 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
436 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
437 dynaddr);
438 #endif
439
440 /* everything is now ready... get the userspace context ready to roll */
441 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
442 start_thread(regs, entryaddr, current->mm->start_stack);
443
444 retval = 0;
445
446 error:
447 if (interpreter) {
448 allow_write_access(interpreter);
449 fput(interpreter);
450 }
451 kfree(interpreter_name);
452 kfree(exec_params.phdrs);
453 kfree(exec_params.loadmap);
454 kfree(interp_params.phdrs);
455 kfree(interp_params.loadmap);
456 return retval;
457
458 /* unrecoverable error - kill the process */
459 error_kill:
460 send_sig(SIGSEGV, current, 0);
461 goto error;
462
463 }
464
465 /*****************************************************************************/
466
467 #ifndef ELF_BASE_PLATFORM
468 /*
469 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
470 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
471 * will be copied to the user stack in the same manner as AT_PLATFORM.
472 */
473 #define ELF_BASE_PLATFORM NULL
474 #endif
475
476 /*
477 * present useful information to the program by shovelling it onto the new
478 * process's stack
479 */
480 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
481 struct mm_struct *mm,
482 struct elf_fdpic_params *exec_params,
483 struct elf_fdpic_params *interp_params)
484 {
485 const struct cred *cred = current_cred();
486 unsigned long sp, csp, nitems;
487 elf_caddr_t __user *argv, *envp;
488 size_t platform_len = 0, len;
489 char *k_platform, *k_base_platform;
490 char __user *u_platform, *u_base_platform, *p;
491 long hwcap;
492 int loop;
493 int nr; /* reset for each csp adjustment */
494
495 #ifdef CONFIG_MMU
496 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
497 * by the processes running on the same package. One thing we can do is
498 * to shuffle the initial stack for them, so we give the architecture
499 * an opportunity to do so here.
500 */
501 sp = arch_align_stack(bprm->p);
502 #else
503 sp = mm->start_stack;
504
505 /* stack the program arguments and environment */
506 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
507 return -EFAULT;
508 #endif
509
510 hwcap = ELF_HWCAP;
511
512 /*
513 * If this architecture has a platform capability string, copy it
514 * to userspace. In some cases (Sparc), this info is impossible
515 * for userspace to get any other way, in others (i386) it is
516 * merely difficult.
517 */
518 k_platform = ELF_PLATFORM;
519 u_platform = NULL;
520
521 if (k_platform) {
522 platform_len = strlen(k_platform) + 1;
523 sp -= platform_len;
524 u_platform = (char __user *) sp;
525 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
526 return -EFAULT;
527 }
528
529 /*
530 * If this architecture has a "base" platform capability
531 * string, copy it to userspace.
532 */
533 k_base_platform = ELF_BASE_PLATFORM;
534 u_base_platform = NULL;
535
536 if (k_base_platform) {
537 platform_len = strlen(k_base_platform) + 1;
538 sp -= platform_len;
539 u_base_platform = (char __user *) sp;
540 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
541 return -EFAULT;
542 }
543
544 sp &= ~7UL;
545
546 /* stack the load map(s) */
547 len = sizeof(struct elf32_fdpic_loadmap);
548 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
549 sp = (sp - len) & ~7UL;
550 exec_params->map_addr = sp;
551
552 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
553 return -EFAULT;
554
555 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
556
557 if (interp_params->loadmap) {
558 len = sizeof(struct elf32_fdpic_loadmap);
559 len += sizeof(struct elf32_fdpic_loadseg) *
560 interp_params->loadmap->nsegs;
561 sp = (sp - len) & ~7UL;
562 interp_params->map_addr = sp;
563
564 if (copy_to_user((void __user *) sp, interp_params->loadmap,
565 len) != 0)
566 return -EFAULT;
567
568 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
569 }
570
571 /* force 16 byte _final_ alignment here for generality */
572 #define DLINFO_ITEMS 15
573
574 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
575 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
576
577 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
578 nitems++;
579
580 csp = sp;
581 sp -= nitems * 2 * sizeof(unsigned long);
582 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
583 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
584 sp -= 1 * sizeof(unsigned long); /* argc */
585
586 csp -= sp & 15UL;
587 sp -= sp & 15UL;
588
589 /* put the ELF interpreter info on the stack */
590 #define NEW_AUX_ENT(id, val) \
591 do { \
592 struct { unsigned long _id, _val; } __user *ent; \
593 \
594 ent = (void __user *) csp; \
595 __put_user((id), &ent[nr]._id); \
596 __put_user((val), &ent[nr]._val); \
597 nr++; \
598 } while (0)
599
600 nr = 0;
601 csp -= 2 * sizeof(unsigned long);
602 NEW_AUX_ENT(AT_NULL, 0);
603 if (k_platform) {
604 nr = 0;
605 csp -= 2 * sizeof(unsigned long);
606 NEW_AUX_ENT(AT_PLATFORM,
607 (elf_addr_t) (unsigned long) u_platform);
608 }
609
610 if (k_base_platform) {
611 nr = 0;
612 csp -= 2 * sizeof(unsigned long);
613 NEW_AUX_ENT(AT_BASE_PLATFORM,
614 (elf_addr_t) (unsigned long) u_base_platform);
615 }
616
617 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
618 nr = 0;
619 csp -= 2 * sizeof(unsigned long);
620 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
621 }
622
623 nr = 0;
624 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
625 NEW_AUX_ENT(AT_HWCAP, hwcap);
626 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
627 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
628 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
629 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
630 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
631 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
632 NEW_AUX_ENT(AT_FLAGS, 0);
633 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
634 NEW_AUX_ENT(AT_UID, (elf_addr_t) cred->uid);
635 NEW_AUX_ENT(AT_EUID, (elf_addr_t) cred->euid);
636 NEW_AUX_ENT(AT_GID, (elf_addr_t) cred->gid);
637 NEW_AUX_ENT(AT_EGID, (elf_addr_t) cred->egid);
638 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
639 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
640
641 #ifdef ARCH_DLINFO
642 nr = 0;
643 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
644
645 /* ARCH_DLINFO must come last so platform specific code can enforce
646 * special alignment requirements on the AUXV if necessary (eg. PPC).
647 */
648 ARCH_DLINFO;
649 #endif
650 #undef NEW_AUX_ENT
651
652 /* allocate room for argv[] and envv[] */
653 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
654 envp = (elf_caddr_t __user *) csp;
655 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
656 argv = (elf_caddr_t __user *) csp;
657
658 /* stack argc */
659 csp -= sizeof(unsigned long);
660 __put_user(bprm->argc, (unsigned long __user *) csp);
661
662 BUG_ON(csp != sp);
663
664 /* fill in the argv[] array */
665 #ifdef CONFIG_MMU
666 current->mm->arg_start = bprm->p;
667 #else
668 current->mm->arg_start = current->mm->start_stack -
669 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
670 #endif
671
672 p = (char __user *) current->mm->arg_start;
673 for (loop = bprm->argc; loop > 0; loop--) {
674 __put_user((elf_caddr_t) p, argv++);
675 len = strnlen_user(p, MAX_ARG_STRLEN);
676 if (!len || len > MAX_ARG_STRLEN)
677 return -EINVAL;
678 p += len;
679 }
680 __put_user(NULL, argv);
681 current->mm->arg_end = (unsigned long) p;
682
683 /* fill in the envv[] array */
684 current->mm->env_start = (unsigned long) p;
685 for (loop = bprm->envc; loop > 0; loop--) {
686 __put_user((elf_caddr_t)(unsigned long) p, envp++);
687 len = strnlen_user(p, MAX_ARG_STRLEN);
688 if (!len || len > MAX_ARG_STRLEN)
689 return -EINVAL;
690 p += len;
691 }
692 __put_user(NULL, envp);
693 current->mm->env_end = (unsigned long) p;
694
695 mm->start_stack = (unsigned long) sp;
696 return 0;
697 }
698
699 /*****************************************************************************/
700 /*
701 * transfer the program arguments and environment from the holding pages onto
702 * the stack
703 */
704 #ifndef CONFIG_MMU
705 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
706 unsigned long *_sp)
707 {
708 unsigned long index, stop, sp;
709 char *src;
710 int ret = 0;
711
712 stop = bprm->p >> PAGE_SHIFT;
713 sp = *_sp;
714
715 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
716 src = kmap(bprm->page[index]);
717 sp -= PAGE_SIZE;
718 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
719 ret = -EFAULT;
720 kunmap(bprm->page[index]);
721 if (ret < 0)
722 goto out;
723 }
724
725 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
726
727 out:
728 return ret;
729 }
730 #endif
731
732 /*****************************************************************************/
733 /*
734 * load the appropriate binary image (executable or interpreter) into memory
735 * - we assume no MMU is available
736 * - if no other PIC bits are set in params->hdr->e_flags
737 * - we assume that the LOADable segments in the binary are independently relocatable
738 * - we assume R/O executable segments are shareable
739 * - else
740 * - we assume the loadable parts of the image to require fixed displacement
741 * - the image is not shareable
742 */
743 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
744 struct file *file,
745 struct mm_struct *mm,
746 const char *what)
747 {
748 struct elf32_fdpic_loadmap *loadmap;
749 #ifdef CONFIG_MMU
750 struct elf32_fdpic_loadseg *mseg;
751 #endif
752 struct elf32_fdpic_loadseg *seg;
753 struct elf32_phdr *phdr;
754 unsigned long load_addr, stop;
755 unsigned nloads, tmp;
756 size_t size;
757 int loop, ret;
758
759 /* allocate a load map table */
760 nloads = 0;
761 for (loop = 0; loop < params->hdr.e_phnum; loop++)
762 if (params->phdrs[loop].p_type == PT_LOAD)
763 nloads++;
764
765 if (nloads == 0)
766 return -ELIBBAD;
767
768 size = sizeof(*loadmap) + nloads * sizeof(*seg);
769 loadmap = kzalloc(size, GFP_KERNEL);
770 if (!loadmap)
771 return -ENOMEM;
772
773 params->loadmap = loadmap;
774
775 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
776 loadmap->nsegs = nloads;
777
778 load_addr = params->load_addr;
779 seg = loadmap->segs;
780
781 /* map the requested LOADs into the memory space */
782 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
783 case ELF_FDPIC_FLAG_CONSTDISP:
784 case ELF_FDPIC_FLAG_CONTIGUOUS:
785 #ifndef CONFIG_MMU
786 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
787 if (ret < 0)
788 return ret;
789 break;
790 #endif
791 default:
792 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
793 if (ret < 0)
794 return ret;
795 break;
796 }
797
798 /* map the entry point */
799 if (params->hdr.e_entry) {
800 seg = loadmap->segs;
801 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
802 if (params->hdr.e_entry >= seg->p_vaddr &&
803 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
804 params->entry_addr =
805 (params->hdr.e_entry - seg->p_vaddr) +
806 seg->addr;
807 break;
808 }
809 }
810 }
811
812 /* determine where the program header table has wound up if mapped */
813 stop = params->hdr.e_phoff;
814 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
815 phdr = params->phdrs;
816
817 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
818 if (phdr->p_type != PT_LOAD)
819 continue;
820
821 if (phdr->p_offset > params->hdr.e_phoff ||
822 phdr->p_offset + phdr->p_filesz < stop)
823 continue;
824
825 seg = loadmap->segs;
826 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
827 if (phdr->p_vaddr >= seg->p_vaddr &&
828 phdr->p_vaddr + phdr->p_filesz <=
829 seg->p_vaddr + seg->p_memsz) {
830 params->ph_addr =
831 (phdr->p_vaddr - seg->p_vaddr) +
832 seg->addr +
833 params->hdr.e_phoff - phdr->p_offset;
834 break;
835 }
836 }
837 break;
838 }
839
840 /* determine where the dynamic section has wound up if there is one */
841 phdr = params->phdrs;
842 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
843 if (phdr->p_type != PT_DYNAMIC)
844 continue;
845
846 seg = loadmap->segs;
847 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
848 if (phdr->p_vaddr >= seg->p_vaddr &&
849 phdr->p_vaddr + phdr->p_memsz <=
850 seg->p_vaddr + seg->p_memsz) {
851 params->dynamic_addr =
852 (phdr->p_vaddr - seg->p_vaddr) +
853 seg->addr;
854
855 /* check the dynamic section contains at least
856 * one item, and that the last item is a NULL
857 * entry */
858 if (phdr->p_memsz == 0 ||
859 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
860 goto dynamic_error;
861
862 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
863 if (((Elf32_Dyn *)
864 params->dynamic_addr)[tmp - 1].d_tag != 0)
865 goto dynamic_error;
866 break;
867 }
868 }
869 break;
870 }
871
872 /* now elide adjacent segments in the load map on MMU linux
873 * - on uClinux the holes between may actually be filled with system
874 * stuff or stuff from other processes
875 */
876 #ifdef CONFIG_MMU
877 nloads = loadmap->nsegs;
878 mseg = loadmap->segs;
879 seg = mseg + 1;
880 for (loop = 1; loop < nloads; loop++) {
881 /* see if we have a candidate for merging */
882 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
883 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
884 if (load_addr == (seg->addr & PAGE_MASK)) {
885 mseg->p_memsz +=
886 load_addr -
887 (mseg->addr + mseg->p_memsz);
888 mseg->p_memsz += seg->addr & ~PAGE_MASK;
889 mseg->p_memsz += seg->p_memsz;
890 loadmap->nsegs--;
891 continue;
892 }
893 }
894
895 mseg++;
896 if (mseg != seg)
897 *mseg = *seg;
898 }
899 #endif
900
901 kdebug("Mapped Object [%s]:", what);
902 kdebug("- elfhdr : %lx", params->elfhdr_addr);
903 kdebug("- entry : %lx", params->entry_addr);
904 kdebug("- PHDR[] : %lx", params->ph_addr);
905 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
906 seg = loadmap->segs;
907 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
908 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
909 loop,
910 seg->addr, seg->addr + seg->p_memsz - 1,
911 seg->p_vaddr, seg->p_memsz);
912
913 return 0;
914
915 dynamic_error:
916 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
917 what, file->f_path.dentry->d_inode->i_ino);
918 return -ELIBBAD;
919 }
920
921 /*****************************************************************************/
922 /*
923 * map a file with constant displacement under uClinux
924 */
925 #ifndef CONFIG_MMU
926 static int elf_fdpic_map_file_constdisp_on_uclinux(
927 struct elf_fdpic_params *params,
928 struct file *file,
929 struct mm_struct *mm)
930 {
931 struct elf32_fdpic_loadseg *seg;
932 struct elf32_phdr *phdr;
933 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
934 loff_t fpos;
935 int loop, ret;
936
937 load_addr = params->load_addr;
938 seg = params->loadmap->segs;
939
940 /* determine the bounds of the contiguous overall allocation we must
941 * make */
942 phdr = params->phdrs;
943 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
944 if (params->phdrs[loop].p_type != PT_LOAD)
945 continue;
946
947 if (base > phdr->p_vaddr)
948 base = phdr->p_vaddr;
949 if (top < phdr->p_vaddr + phdr->p_memsz)
950 top = phdr->p_vaddr + phdr->p_memsz;
951 }
952
953 /* allocate one big anon block for everything */
954 mflags = MAP_PRIVATE;
955 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
956 mflags |= MAP_EXECUTABLE;
957
958 down_write(&mm->mmap_sem);
959 maddr = do_mmap(NULL, load_addr, top - base,
960 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
961 up_write(&mm->mmap_sem);
962 if (IS_ERR_VALUE(maddr))
963 return (int) maddr;
964
965 if (load_addr != 0)
966 load_addr += PAGE_ALIGN(top - base);
967
968 /* and then load the file segments into it */
969 phdr = params->phdrs;
970 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
971 if (params->phdrs[loop].p_type != PT_LOAD)
972 continue;
973
974 fpos = phdr->p_offset;
975
976 seg->addr = maddr + (phdr->p_vaddr - base);
977 seg->p_vaddr = phdr->p_vaddr;
978 seg->p_memsz = phdr->p_memsz;
979
980 ret = file->f_op->read(file, (void *) seg->addr,
981 phdr->p_filesz, &fpos);
982 if (ret < 0)
983 return ret;
984
985 /* map the ELF header address if in this segment */
986 if (phdr->p_offset == 0)
987 params->elfhdr_addr = seg->addr;
988
989 /* clear any space allocated but not loaded */
990 if (phdr->p_filesz < phdr->p_memsz) {
991 ret = clear_user((void *) (seg->addr + phdr->p_filesz),
992 phdr->p_memsz - phdr->p_filesz);
993 if (ret)
994 return ret;
995 }
996
997 if (mm) {
998 if (phdr->p_flags & PF_X) {
999 if (!mm->start_code) {
1000 mm->start_code = seg->addr;
1001 mm->end_code = seg->addr +
1002 phdr->p_memsz;
1003 }
1004 } else if (!mm->start_data) {
1005 mm->start_data = seg->addr;
1006 #ifndef CONFIG_MMU
1007 mm->end_data = seg->addr + phdr->p_memsz;
1008 #endif
1009 }
1010
1011 #ifdef CONFIG_MMU
1012 if (seg->addr + phdr->p_memsz > mm->end_data)
1013 mm->end_data = seg->addr + phdr->p_memsz;
1014 #endif
1015 }
1016
1017 seg++;
1018 }
1019
1020 return 0;
1021 }
1022 #endif
1023
1024 /*****************************************************************************/
1025 /*
1026 * map a binary by direct mmap() of the individual PT_LOAD segments
1027 */
1028 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1029 struct file *file,
1030 struct mm_struct *mm)
1031 {
1032 struct elf32_fdpic_loadseg *seg;
1033 struct elf32_phdr *phdr;
1034 unsigned long load_addr, delta_vaddr;
1035 int loop, dvset, ret;
1036
1037 load_addr = params->load_addr;
1038 delta_vaddr = 0;
1039 dvset = 0;
1040
1041 seg = params->loadmap->segs;
1042
1043 /* deal with each load segment separately */
1044 phdr = params->phdrs;
1045 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1046 unsigned long maddr, disp, excess, excess1;
1047 int prot = 0, flags;
1048
1049 if (phdr->p_type != PT_LOAD)
1050 continue;
1051
1052 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1053 (unsigned long) phdr->p_vaddr,
1054 (unsigned long) phdr->p_offset,
1055 (unsigned long) phdr->p_filesz,
1056 (unsigned long) phdr->p_memsz);
1057
1058 /* determine the mapping parameters */
1059 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1060 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1061 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1062
1063 flags = MAP_PRIVATE | MAP_DENYWRITE;
1064 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1065 flags |= MAP_EXECUTABLE;
1066
1067 maddr = 0;
1068
1069 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1070 case ELF_FDPIC_FLAG_INDEPENDENT:
1071 /* PT_LOADs are independently locatable */
1072 break;
1073
1074 case ELF_FDPIC_FLAG_HONOURVADDR:
1075 /* the specified virtual address must be honoured */
1076 maddr = phdr->p_vaddr;
1077 flags |= MAP_FIXED;
1078 break;
1079
1080 case ELF_FDPIC_FLAG_CONSTDISP:
1081 /* constant displacement
1082 * - can be mapped anywhere, but must be mapped as a
1083 * unit
1084 */
1085 if (!dvset) {
1086 maddr = load_addr;
1087 delta_vaddr = phdr->p_vaddr;
1088 dvset = 1;
1089 } else {
1090 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1091 flags |= MAP_FIXED;
1092 }
1093 break;
1094
1095 case ELF_FDPIC_FLAG_CONTIGUOUS:
1096 /* contiguity handled later */
1097 break;
1098
1099 default:
1100 BUG();
1101 }
1102
1103 maddr &= PAGE_MASK;
1104
1105 /* create the mapping */
1106 disp = phdr->p_vaddr & ~PAGE_MASK;
1107 down_write(&mm->mmap_sem);
1108 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1109 phdr->p_offset - disp);
1110 up_write(&mm->mmap_sem);
1111
1112 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1113 loop, phdr->p_memsz + disp, prot, flags,
1114 phdr->p_offset - disp, maddr);
1115
1116 if (IS_ERR_VALUE(maddr))
1117 return (int) maddr;
1118
1119 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1120 ELF_FDPIC_FLAG_CONTIGUOUS)
1121 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1122
1123 seg->addr = maddr + disp;
1124 seg->p_vaddr = phdr->p_vaddr;
1125 seg->p_memsz = phdr->p_memsz;
1126
1127 /* map the ELF header address if in this segment */
1128 if (phdr->p_offset == 0)
1129 params->elfhdr_addr = seg->addr;
1130
1131 /* clear the bit between beginning of mapping and beginning of
1132 * PT_LOAD */
1133 if (prot & PROT_WRITE && disp > 0) {
1134 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1135 ret = clear_user((void __user *) maddr, disp);
1136 if (ret)
1137 return ret;
1138 maddr += disp;
1139 }
1140
1141 /* clear any space allocated but not loaded
1142 * - on uClinux we can just clear the lot
1143 * - on MMU linux we'll get a SIGBUS beyond the last page
1144 * extant in the file
1145 */
1146 excess = phdr->p_memsz - phdr->p_filesz;
1147 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1148
1149 #ifdef CONFIG_MMU
1150 if (excess > excess1) {
1151 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1152 unsigned long xmaddr;
1153
1154 flags |= MAP_FIXED | MAP_ANONYMOUS;
1155 down_write(&mm->mmap_sem);
1156 xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1157 prot, flags, 0);
1158 up_write(&mm->mmap_sem);
1159
1160 kdebug("mmap[%d] <anon>"
1161 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1162 loop, xaddr, excess - excess1, prot, flags,
1163 xmaddr);
1164
1165 if (xmaddr != xaddr)
1166 return -ENOMEM;
1167 }
1168
1169 if (prot & PROT_WRITE && excess1 > 0) {
1170 kdebug("clear[%d] ad=%lx sz=%lx",
1171 loop, maddr + phdr->p_filesz, excess1);
1172 ret = clear_user((void __user *) maddr + phdr->p_filesz,
1173 excess1);
1174 if (ret)
1175 return ret;
1176 }
1177
1178 #else
1179 if (excess > 0) {
1180 kdebug("clear[%d] ad=%lx sz=%lx",
1181 loop, maddr + phdr->p_filesz, excess);
1182 ret = clear_user((void *) maddr + phdr->p_filesz, excess);
1183 if (ret)
1184 return ret;
1185 }
1186 #endif
1187
1188 if (mm) {
1189 if (phdr->p_flags & PF_X) {
1190 if (!mm->start_code) {
1191 mm->start_code = maddr;
1192 mm->end_code = maddr + phdr->p_memsz;
1193 }
1194 } else if (!mm->start_data) {
1195 mm->start_data = maddr;
1196 mm->end_data = maddr + phdr->p_memsz;
1197 }
1198 }
1199
1200 seg++;
1201 }
1202
1203 return 0;
1204 }
1205
1206 /*****************************************************************************/
1207 /*
1208 * ELF-FDPIC core dumper
1209 *
1210 * Modelled on fs/exec.c:aout_core_dump()
1211 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1212 *
1213 * Modelled on fs/binfmt_elf.c core dumper
1214 */
1215 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1216
1217 /*
1218 * These are the only things you should do on a core-file: use only these
1219 * functions to write out all the necessary info.
1220 */
1221 static int dump_write(struct file *file, const void *addr, int nr)
1222 {
1223 return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1224 }
1225
1226 static int dump_seek(struct file *file, loff_t off)
1227 {
1228 if (file->f_op->llseek) {
1229 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1230 return 0;
1231 } else {
1232 file->f_pos = off;
1233 }
1234 return 1;
1235 }
1236
1237 /*
1238 * Decide whether a segment is worth dumping; default is yes to be
1239 * sure (missing info is worse than too much; etc).
1240 * Personally I'd include everything, and use the coredump limit...
1241 *
1242 * I think we should skip something. But I am not sure how. H.J.
1243 */
1244 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1245 {
1246 int dump_ok;
1247
1248 /* Do not dump I/O mapped devices or special mappings */
1249 if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1250 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1251 return 0;
1252 }
1253
1254 /* If we may not read the contents, don't allow us to dump
1255 * them either. "dump_write()" can't handle it anyway.
1256 */
1257 if (!(vma->vm_flags & VM_READ)) {
1258 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1259 return 0;
1260 }
1261
1262 /* By default, dump shared memory if mapped from an anonymous file. */
1263 if (vma->vm_flags & VM_SHARED) {
1264 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1265 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1266 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1267 vma->vm_flags, dump_ok ? "yes" : "no");
1268 return dump_ok;
1269 }
1270
1271 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1272 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1273 vma->vm_flags, dump_ok ? "yes" : "no");
1274 return dump_ok;
1275 }
1276
1277 #ifdef CONFIG_MMU
1278 /* By default, if it hasn't been written to, don't write it out */
1279 if (!vma->anon_vma) {
1280 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1281 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1282 vma->vm_flags, dump_ok ? "yes" : "no");
1283 return dump_ok;
1284 }
1285 #endif
1286
1287 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1288 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1289 dump_ok ? "yes" : "no");
1290 return dump_ok;
1291 }
1292
1293 /* An ELF note in memory */
1294 struct memelfnote
1295 {
1296 const char *name;
1297 int type;
1298 unsigned int datasz;
1299 void *data;
1300 };
1301
1302 static int notesize(struct memelfnote *en)
1303 {
1304 int sz;
1305
1306 sz = sizeof(struct elf_note);
1307 sz += roundup(strlen(en->name) + 1, 4);
1308 sz += roundup(en->datasz, 4);
1309
1310 return sz;
1311 }
1312
1313 /* #define DEBUG */
1314
1315 #define DUMP_WRITE(addr, nr) \
1316 do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1317 #define DUMP_SEEK(off) \
1318 do { if (!dump_seek(file, (off))) return 0; } while(0)
1319
1320 static int writenote(struct memelfnote *men, struct file *file)
1321 {
1322 struct elf_note en;
1323
1324 en.n_namesz = strlen(men->name) + 1;
1325 en.n_descsz = men->datasz;
1326 en.n_type = men->type;
1327
1328 DUMP_WRITE(&en, sizeof(en));
1329 DUMP_WRITE(men->name, en.n_namesz);
1330 /* XXX - cast from long long to long to avoid need for libgcc.a */
1331 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1332 DUMP_WRITE(men->data, men->datasz);
1333 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1334
1335 return 1;
1336 }
1337 #undef DUMP_WRITE
1338 #undef DUMP_SEEK
1339
1340 #define DUMP_WRITE(addr, nr) \
1341 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1342 goto end_coredump;
1343
1344 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1345 {
1346 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1347 elf->e_ident[EI_CLASS] = ELF_CLASS;
1348 elf->e_ident[EI_DATA] = ELF_DATA;
1349 elf->e_ident[EI_VERSION] = EV_CURRENT;
1350 elf->e_ident[EI_OSABI] = ELF_OSABI;
1351 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1352
1353 elf->e_type = ET_CORE;
1354 elf->e_machine = ELF_ARCH;
1355 elf->e_version = EV_CURRENT;
1356 elf->e_entry = 0;
1357 elf->e_phoff = sizeof(struct elfhdr);
1358 elf->e_shoff = 0;
1359 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1360 elf->e_ehsize = sizeof(struct elfhdr);
1361 elf->e_phentsize = sizeof(struct elf_phdr);
1362 elf->e_phnum = segs;
1363 elf->e_shentsize = 0;
1364 elf->e_shnum = 0;
1365 elf->e_shstrndx = 0;
1366 return;
1367 }
1368
1369 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1370 {
1371 phdr->p_type = PT_NOTE;
1372 phdr->p_offset = offset;
1373 phdr->p_vaddr = 0;
1374 phdr->p_paddr = 0;
1375 phdr->p_filesz = sz;
1376 phdr->p_memsz = 0;
1377 phdr->p_flags = 0;
1378 phdr->p_align = 0;
1379 return;
1380 }
1381
1382 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1383 unsigned int sz, void *data)
1384 {
1385 note->name = name;
1386 note->type = type;
1387 note->datasz = sz;
1388 note->data = data;
1389 return;
1390 }
1391
1392 /*
1393 * fill up all the fields in prstatus from the given task struct, except
1394 * registers which need to be filled up seperately.
1395 */
1396 static void fill_prstatus(struct elf_prstatus *prstatus,
1397 struct task_struct *p, long signr)
1398 {
1399 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1400 prstatus->pr_sigpend = p->pending.signal.sig[0];
1401 prstatus->pr_sighold = p->blocked.sig[0];
1402 rcu_read_lock();
1403 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1404 rcu_read_unlock();
1405 prstatus->pr_pid = task_pid_vnr(p);
1406 prstatus->pr_pgrp = task_pgrp_vnr(p);
1407 prstatus->pr_sid = task_session_vnr(p);
1408 if (thread_group_leader(p)) {
1409 struct task_cputime cputime;
1410
1411 /*
1412 * This is the record for the group leader. It shows the
1413 * group-wide total, not its individual thread total.
1414 */
1415 thread_group_cputime(p, &cputime);
1416 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1417 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1418 } else {
1419 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1420 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1421 }
1422 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1423 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1424
1425 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1426 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1427 }
1428
1429 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1430 struct mm_struct *mm)
1431 {
1432 const struct cred *cred;
1433 unsigned int i, len;
1434
1435 /* first copy the parameters from user space */
1436 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1437
1438 len = mm->arg_end - mm->arg_start;
1439 if (len >= ELF_PRARGSZ)
1440 len = ELF_PRARGSZ - 1;
1441 if (copy_from_user(&psinfo->pr_psargs,
1442 (const char __user *) mm->arg_start, len))
1443 return -EFAULT;
1444 for (i = 0; i < len; i++)
1445 if (psinfo->pr_psargs[i] == 0)
1446 psinfo->pr_psargs[i] = ' ';
1447 psinfo->pr_psargs[len] = 0;
1448
1449 rcu_read_lock();
1450 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1451 rcu_read_unlock();
1452 psinfo->pr_pid = task_pid_vnr(p);
1453 psinfo->pr_pgrp = task_pgrp_vnr(p);
1454 psinfo->pr_sid = task_session_vnr(p);
1455
1456 i = p->state ? ffz(~p->state) + 1 : 0;
1457 psinfo->pr_state = i;
1458 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1459 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1460 psinfo->pr_nice = task_nice(p);
1461 psinfo->pr_flag = p->flags;
1462 rcu_read_lock();
1463 cred = __task_cred(p);
1464 SET_UID(psinfo->pr_uid, cred->uid);
1465 SET_GID(psinfo->pr_gid, cred->gid);
1466 rcu_read_unlock();
1467 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1468
1469 return 0;
1470 }
1471
1472 /* Here is the structure in which status of each thread is captured. */
1473 struct elf_thread_status
1474 {
1475 struct list_head list;
1476 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1477 elf_fpregset_t fpu; /* NT_PRFPREG */
1478 struct task_struct *thread;
1479 #ifdef ELF_CORE_COPY_XFPREGS
1480 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1481 #endif
1482 struct memelfnote notes[3];
1483 int num_notes;
1484 };
1485
1486 /*
1487 * In order to add the specific thread information for the elf file format,
1488 * we need to keep a linked list of every thread's pr_status and then create
1489 * a single section for them in the final core file.
1490 */
1491 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1492 {
1493 struct task_struct *p = t->thread;
1494 int sz = 0;
1495
1496 t->num_notes = 0;
1497
1498 fill_prstatus(&t->prstatus, p, signr);
1499 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1500
1501 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1502 &t->prstatus);
1503 t->num_notes++;
1504 sz += notesize(&t->notes[0]);
1505
1506 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1507 if (t->prstatus.pr_fpvalid) {
1508 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1509 &t->fpu);
1510 t->num_notes++;
1511 sz += notesize(&t->notes[1]);
1512 }
1513
1514 #ifdef ELF_CORE_COPY_XFPREGS
1515 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1516 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1517 sizeof(t->xfpu), &t->xfpu);
1518 t->num_notes++;
1519 sz += notesize(&t->notes[2]);
1520 }
1521 #endif
1522 return sz;
1523 }
1524
1525 /*
1526 * dump the segments for an MMU process
1527 */
1528 #ifdef CONFIG_MMU
1529 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1530 unsigned long *limit, unsigned long mm_flags)
1531 {
1532 struct vm_area_struct *vma;
1533 int err = 0;
1534
1535 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1536 unsigned long addr;
1537
1538 if (!maydump(vma, mm_flags))
1539 continue;
1540
1541 for (addr = vma->vm_start; addr < vma->vm_end;
1542 addr += PAGE_SIZE) {
1543 struct page *page = get_dump_page(addr);
1544 if (page) {
1545 void *kaddr = kmap(page);
1546 *size += PAGE_SIZE;
1547 if (*size > *limit)
1548 err = -EFBIG;
1549 else if (!dump_write(file, kaddr, PAGE_SIZE))
1550 err = -EIO;
1551 kunmap(page);
1552 page_cache_release(page);
1553 } else if (!dump_seek(file, file->f_pos + PAGE_SIZE))
1554 err = -EFBIG;
1555 if (err)
1556 goto out;
1557 }
1558 }
1559 out:
1560 return err;
1561 }
1562 #endif
1563
1564 /*
1565 * dump the segments for a NOMMU process
1566 */
1567 #ifndef CONFIG_MMU
1568 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1569 unsigned long *limit, unsigned long mm_flags)
1570 {
1571 struct vm_area_struct *vma;
1572
1573 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1574 if (!maydump(vma, mm_flags))
1575 continue;
1576
1577 if ((*size += PAGE_SIZE) > *limit)
1578 return -EFBIG;
1579
1580 if (!dump_write(file, (void *) vma->vm_start,
1581 vma->vm_end - vma->vm_start))
1582 return -EIO;
1583 }
1584
1585 return 0;
1586 }
1587 #endif
1588
1589 /*
1590 * Actual dumper
1591 *
1592 * This is a two-pass process; first we find the offsets of the bits,
1593 * and then they are actually written out. If we run out of core limit
1594 * we just truncate.
1595 */
1596 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1597 struct file *file, unsigned long limit)
1598 {
1599 #define NUM_NOTES 6
1600 int has_dumped = 0;
1601 mm_segment_t fs;
1602 int segs;
1603 size_t size = 0;
1604 int i;
1605 struct vm_area_struct *vma;
1606 struct elfhdr *elf = NULL;
1607 loff_t offset = 0, dataoff;
1608 int numnote;
1609 struct memelfnote *notes = NULL;
1610 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1611 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1612 LIST_HEAD(thread_list);
1613 struct list_head *t;
1614 elf_fpregset_t *fpu = NULL;
1615 #ifdef ELF_CORE_COPY_XFPREGS
1616 elf_fpxregset_t *xfpu = NULL;
1617 #endif
1618 int thread_status_size = 0;
1619 elf_addr_t *auxv;
1620 unsigned long mm_flags;
1621
1622 /*
1623 * We no longer stop all VM operations.
1624 *
1625 * This is because those proceses that could possibly change map_count
1626 * or the mmap / vma pages are now blocked in do_exit on current
1627 * finishing this core dump.
1628 *
1629 * Only ptrace can touch these memory addresses, but it doesn't change
1630 * the map_count or the pages allocated. So no possibility of crashing
1631 * exists while dumping the mm->vm_next areas to the core file.
1632 */
1633
1634 /* alloc memory for large data structures: too large to be on stack */
1635 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1636 if (!elf)
1637 goto cleanup;
1638 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1639 if (!prstatus)
1640 goto cleanup;
1641 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1642 if (!psinfo)
1643 goto cleanup;
1644 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1645 if (!notes)
1646 goto cleanup;
1647 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1648 if (!fpu)
1649 goto cleanup;
1650 #ifdef ELF_CORE_COPY_XFPREGS
1651 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1652 if (!xfpu)
1653 goto cleanup;
1654 #endif
1655
1656 if (signr) {
1657 struct core_thread *ct;
1658 struct elf_thread_status *tmp;
1659
1660 for (ct = current->mm->core_state->dumper.next;
1661 ct; ct = ct->next) {
1662 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1663 if (!tmp)
1664 goto cleanup;
1665
1666 tmp->thread = ct->task;
1667 list_add(&tmp->list, &thread_list);
1668 }
1669
1670 list_for_each(t, &thread_list) {
1671 struct elf_thread_status *tmp;
1672 int sz;
1673
1674 tmp = list_entry(t, struct elf_thread_status, list);
1675 sz = elf_dump_thread_status(signr, tmp);
1676 thread_status_size += sz;
1677 }
1678 }
1679
1680 /* now collect the dump for the current */
1681 fill_prstatus(prstatus, current, signr);
1682 elf_core_copy_regs(&prstatus->pr_reg, regs);
1683
1684 segs = current->mm->map_count;
1685 #ifdef ELF_CORE_EXTRA_PHDRS
1686 segs += ELF_CORE_EXTRA_PHDRS;
1687 #endif
1688
1689 /* Set up header */
1690 fill_elf_fdpic_header(elf, segs + 1); /* including notes section */
1691
1692 has_dumped = 1;
1693 current->flags |= PF_DUMPCORE;
1694
1695 /*
1696 * Set up the notes in similar form to SVR4 core dumps made
1697 * with info from their /proc.
1698 */
1699
1700 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1701 fill_psinfo(psinfo, current->group_leader, current->mm);
1702 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1703
1704 numnote = 2;
1705
1706 auxv = (elf_addr_t *) current->mm->saved_auxv;
1707
1708 i = 0;
1709 do
1710 i += 2;
1711 while (auxv[i - 2] != AT_NULL);
1712 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1713 i * sizeof(elf_addr_t), auxv);
1714
1715 /* Try to dump the FPU. */
1716 if ((prstatus->pr_fpvalid =
1717 elf_core_copy_task_fpregs(current, regs, fpu)))
1718 fill_note(notes + numnote++,
1719 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1720 #ifdef ELF_CORE_COPY_XFPREGS
1721 if (elf_core_copy_task_xfpregs(current, xfpu))
1722 fill_note(notes + numnote++,
1723 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1724 #endif
1725
1726 fs = get_fs();
1727 set_fs(KERNEL_DS);
1728
1729 DUMP_WRITE(elf, sizeof(*elf));
1730 offset += sizeof(*elf); /* Elf header */
1731 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers */
1732
1733 /* Write notes phdr entry */
1734 {
1735 struct elf_phdr phdr;
1736 int sz = 0;
1737
1738 for (i = 0; i < numnote; i++)
1739 sz += notesize(notes + i);
1740
1741 sz += thread_status_size;
1742
1743 fill_elf_note_phdr(&phdr, sz, offset);
1744 offset += sz;
1745 DUMP_WRITE(&phdr, sizeof(phdr));
1746 }
1747
1748 /* Page-align dumped data */
1749 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1750
1751 /*
1752 * We must use the same mm->flags while dumping core to avoid
1753 * inconsistency between the program headers and bodies, otherwise an
1754 * unusable core file can be generated.
1755 */
1756 mm_flags = current->mm->flags;
1757
1758 /* write program headers for segments dump */
1759 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1760 struct elf_phdr phdr;
1761 size_t sz;
1762
1763 sz = vma->vm_end - vma->vm_start;
1764
1765 phdr.p_type = PT_LOAD;
1766 phdr.p_offset = offset;
1767 phdr.p_vaddr = vma->vm_start;
1768 phdr.p_paddr = 0;
1769 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1770 phdr.p_memsz = sz;
1771 offset += phdr.p_filesz;
1772 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1773 if (vma->vm_flags & VM_WRITE)
1774 phdr.p_flags |= PF_W;
1775 if (vma->vm_flags & VM_EXEC)
1776 phdr.p_flags |= PF_X;
1777 phdr.p_align = ELF_EXEC_PAGESIZE;
1778
1779 DUMP_WRITE(&phdr, sizeof(phdr));
1780 }
1781
1782 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1783 ELF_CORE_WRITE_EXTRA_PHDRS;
1784 #endif
1785
1786 /* write out the notes section */
1787 for (i = 0; i < numnote; i++)
1788 if (!writenote(notes + i, file))
1789 goto end_coredump;
1790
1791 /* write out the thread status notes section */
1792 list_for_each(t, &thread_list) {
1793 struct elf_thread_status *tmp =
1794 list_entry(t, struct elf_thread_status, list);
1795
1796 for (i = 0; i < tmp->num_notes; i++)
1797 if (!writenote(&tmp->notes[i], file))
1798 goto end_coredump;
1799 }
1800
1801 if (!dump_seek(file, dataoff))
1802 goto end_coredump;
1803
1804 if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1805 goto end_coredump;
1806
1807 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1808 ELF_CORE_WRITE_EXTRA_DATA;
1809 #endif
1810
1811 if (file->f_pos != offset) {
1812 /* Sanity check */
1813 printk(KERN_WARNING
1814 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1815 file->f_pos, offset);
1816 }
1817
1818 end_coredump:
1819 set_fs(fs);
1820
1821 cleanup:
1822 while (!list_empty(&thread_list)) {
1823 struct list_head *tmp = thread_list.next;
1824 list_del(tmp);
1825 kfree(list_entry(tmp, struct elf_thread_status, list));
1826 }
1827
1828 kfree(elf);
1829 kfree(prstatus);
1830 kfree(psinfo);
1831 kfree(notes);
1832 kfree(fpu);
1833 #ifdef ELF_CORE_COPY_XFPREGS
1834 kfree(xfpu);
1835 #endif
1836 return has_dumped;
1837 #undef NUM_NOTES
1838 }
1839
1840 #endif /* USE_ELF_CORE_DUMP */
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree