?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
lockd: release reference to nsm_handle in nlm_host_rebooted
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / binfmt_elf_fdpic.c
blobc57d9ce5ff7ef85605786fc26200b3fe76df93df
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 #ifdef CONFIG_ELF_CORE
79 static int elf_fdpic_core_dump(struct coredump_params *cprm);
80 #endif
81
82 static struct linux_binfmt elf_fdpic_format = {
83 .module = THIS_MODULE,
84 .load_binary = load_elf_fdpic_binary,
85 #ifdef 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 set_binfmt(&elf_fdpic_format);
325
326 current->mm->start_code = 0;
327 current->mm->end_code = 0;
328 current->mm->start_stack = 0;
329 current->mm->start_data = 0;
330 current->mm->end_data = 0;
331 current->mm->context.exec_fdpic_loadmap = 0;
332 current->mm->context.interp_fdpic_loadmap = 0;
333
334 current->flags &= ~PF_FORKNOEXEC;
335
336 #ifdef CONFIG_MMU
337 elf_fdpic_arch_lay_out_mm(&exec_params,
338 &interp_params,
339 &current->mm->start_stack,
340 &current->mm->start_brk);
341
342 retval = setup_arg_pages(bprm, current->mm->start_stack,
343 executable_stack);
344 if (retval < 0) {
345 send_sig(SIGKILL, current, 0);
346 goto error_kill;
347 }
348 #endif
349
350 /* load the executable and interpreter into memory */
351 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
352 "executable");
353 if (retval < 0)
354 goto error_kill;
355
356 if (interpreter_name) {
357 retval = elf_fdpic_map_file(&interp_params, interpreter,
358 current->mm, "interpreter");
359 if (retval < 0) {
360 printk(KERN_ERR "Unable to load interpreter\n");
361 goto error_kill;
362 }
363
364 allow_write_access(interpreter);
365 fput(interpreter);
366 interpreter = NULL;
367 }
368
369 #ifdef CONFIG_MMU
370 if (!current->mm->start_brk)
371 current->mm->start_brk = current->mm->end_data;
372
373 current->mm->brk = current->mm->start_brk =
374 PAGE_ALIGN(current->mm->start_brk);
375
376 #else
377 /* create a stack and brk area big enough for everyone
378 * - the brk heap starts at the bottom and works up
379 * - the stack starts at the top and works down
380 */
381 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
382 if (stack_size < PAGE_SIZE * 2)
383 stack_size = PAGE_SIZE * 2;
384
385 stack_prot = PROT_READ | PROT_WRITE;
386 if (executable_stack == EXSTACK_ENABLE_X ||
387 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
388 stack_prot |= PROT_EXEC;
389
390 down_write(&current->mm->mmap_sem);
391 current->mm->start_brk = do_mmap(NULL, 0, stack_size, stack_prot,
392 MAP_PRIVATE | MAP_ANONYMOUS |
393 MAP_UNINITIALIZED | MAP_GROWSDOWN,
394 0);
395
396 if (IS_ERR_VALUE(current->mm->start_brk)) {
397 up_write(&current->mm->mmap_sem);
398 retval = current->mm->start_brk;
399 current->mm->start_brk = 0;
400 goto error_kill;
401 }
402
403 up_write(&current->mm->mmap_sem);
404
405 current->mm->brk = current->mm->start_brk;
406 current->mm->context.end_brk = current->mm->start_brk;
407 current->mm->context.end_brk +=
408 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
409 current->mm->start_stack = current->mm->start_brk + stack_size;
410 #endif
411
412 install_exec_creds(bprm);
413 current->flags &= ~PF_FORKNOEXEC;
414 if (create_elf_fdpic_tables(bprm, current->mm,
415 &exec_params, &interp_params) < 0)
416 goto error_kill;
417
418 kdebug("- start_code %lx", current->mm->start_code);
419 kdebug("- end_code %lx", current->mm->end_code);
420 kdebug("- start_data %lx", current->mm->start_data);
421 kdebug("- end_data %lx", current->mm->end_data);
422 kdebug("- start_brk %lx", current->mm->start_brk);
423 kdebug("- brk %lx", current->mm->brk);
424 kdebug("- start_stack %lx", current->mm->start_stack);
425
426 #ifdef ELF_FDPIC_PLAT_INIT
427 /*
428 * The ABI may specify that certain registers be set up in special
429 * ways (on i386 %edx is the address of a DT_FINI function, for
430 * example. This macro performs whatever initialization to
431 * the regs structure is required.
432 */
433 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
434 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
435 dynaddr);
436 #endif
437
438 /* everything is now ready... get the userspace context ready to roll */
439 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
440 start_thread(regs, entryaddr, current->mm->start_stack);
441
442 retval = 0;
443
444 error:
445 if (interpreter) {
446 allow_write_access(interpreter);
447 fput(interpreter);
448 }
449 kfree(interpreter_name);
450 kfree(exec_params.phdrs);
451 kfree(exec_params.loadmap);
452 kfree(interp_params.phdrs);
453 kfree(interp_params.loadmap);
454 return retval;
455
456 /* unrecoverable error - kill the process */
457 error_kill:
458 send_sig(SIGSEGV, current, 0);
459 goto error;
460
461 }
462
463 /*****************************************************************************/
464
465 #ifndef ELF_BASE_PLATFORM
466 /*
467 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
468 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
469 * will be copied to the user stack in the same manner as AT_PLATFORM.
470 */
471 #define ELF_BASE_PLATFORM NULL
472 #endif
473
474 /*
475 * present useful information to the program by shovelling it onto the new
476 * process's stack
477 */
478 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
479 struct mm_struct *mm,
480 struct elf_fdpic_params *exec_params,
481 struct elf_fdpic_params *interp_params)
482 {
483 const struct cred *cred = current_cred();
484 unsigned long sp, csp, nitems;
485 elf_caddr_t __user *argv, *envp;
486 size_t platform_len = 0, len;
487 char *k_platform, *k_base_platform;
488 char __user *u_platform, *u_base_platform, *p;
489 long hwcap;
490 int loop;
491 int nr; /* reset for each csp adjustment */
492
493 #ifdef CONFIG_MMU
494 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
495 * by the processes running on the same package. One thing we can do is
496 * to shuffle the initial stack for them, so we give the architecture
497 * an opportunity to do so here.
498 */
499 sp = arch_align_stack(bprm->p);
500 #else
501 sp = mm->start_stack;
502
503 /* stack the program arguments and environment */
504 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
505 return -EFAULT;
506 #endif
507
508 hwcap = ELF_HWCAP;
509
510 /*
511 * If this architecture has a platform capability string, copy it
512 * to userspace. In some cases (Sparc), this info is impossible
513 * for userspace to get any other way, in others (i386) it is
514 * merely difficult.
515 */
516 k_platform = ELF_PLATFORM;
517 u_platform = NULL;
518
519 if (k_platform) {
520 platform_len = strlen(k_platform) + 1;
521 sp -= platform_len;
522 u_platform = (char __user *) sp;
523 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
524 return -EFAULT;
525 }
526
527 /*
528 * If this architecture has a "base" platform capability
529 * string, copy it to userspace.
530 */
531 k_base_platform = ELF_BASE_PLATFORM;
532 u_base_platform = NULL;
533
534 if (k_base_platform) {
535 platform_len = strlen(k_base_platform) + 1;
536 sp -= platform_len;
537 u_base_platform = (char __user *) sp;
538 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
539 return -EFAULT;
540 }
541
542 sp &= ~7UL;
543
544 /* stack the load map(s) */
545 len = sizeof(struct elf32_fdpic_loadmap);
546 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
547 sp = (sp - len) & ~7UL;
548 exec_params->map_addr = sp;
549
550 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
551 return -EFAULT;
552
553 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
554
555 if (interp_params->loadmap) {
556 len = sizeof(struct elf32_fdpic_loadmap);
557 len += sizeof(struct elf32_fdpic_loadseg) *
558 interp_params->loadmap->nsegs;
559 sp = (sp - len) & ~7UL;
560 interp_params->map_addr = sp;
561
562 if (copy_to_user((void __user *) sp, interp_params->loadmap,
563 len) != 0)
564 return -EFAULT;
565
566 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
567 }
568
569 /* force 16 byte _final_ alignment here for generality */
570 #define DLINFO_ITEMS 15
571
572 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
573 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
574
575 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
576 nitems++;
577
578 csp = sp;
579 sp -= nitems * 2 * sizeof(unsigned long);
580 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
581 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
582 sp -= 1 * sizeof(unsigned long); /* argc */
583
584 csp -= sp & 15UL;
585 sp -= sp & 15UL;
586
587 /* put the ELF interpreter info on the stack */
588 #define NEW_AUX_ENT(id, val) \
589 do { \
590 struct { unsigned long _id, _val; } __user *ent; \
591 \
592 ent = (void __user *) csp; \
593 __put_user((id), &ent[nr]._id); \
594 __put_user((val), &ent[nr]._val); \
595 nr++; \
596 } while (0)
597
598 nr = 0;
599 csp -= 2 * sizeof(unsigned long);
600 NEW_AUX_ENT(AT_NULL, 0);
601 if (k_platform) {
602 nr = 0;
603 csp -= 2 * sizeof(unsigned long);
604 NEW_AUX_ENT(AT_PLATFORM,
605 (elf_addr_t) (unsigned long) u_platform);
606 }
607
608 if (k_base_platform) {
609 nr = 0;
610 csp -= 2 * sizeof(unsigned long);
611 NEW_AUX_ENT(AT_BASE_PLATFORM,
612 (elf_addr_t) (unsigned long) u_base_platform);
613 }
614
615 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
616 nr = 0;
617 csp -= 2 * sizeof(unsigned long);
618 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
619 }
620
621 nr = 0;
622 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
623 NEW_AUX_ENT(AT_HWCAP, hwcap);
624 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
625 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
626 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
627 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
628 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
629 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
630 NEW_AUX_ENT(AT_FLAGS, 0);
631 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
632 NEW_AUX_ENT(AT_UID, (elf_addr_t) cred->uid);
633 NEW_AUX_ENT(AT_EUID, (elf_addr_t) cred->euid);
634 NEW_AUX_ENT(AT_GID, (elf_addr_t) cred->gid);
635 NEW_AUX_ENT(AT_EGID, (elf_addr_t) cred->egid);
636 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
637 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
638
639 #ifdef ARCH_DLINFO
640 nr = 0;
641 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
642
643 /* ARCH_DLINFO must come last so platform specific code can enforce
644 * special alignment requirements on the AUXV if necessary (eg. PPC).
645 */
646 ARCH_DLINFO;
647 #endif
648 #undef NEW_AUX_ENT
649
650 /* allocate room for argv[] and envv[] */
651 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
652 envp = (elf_caddr_t __user *) csp;
653 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
654 argv = (elf_caddr_t __user *) csp;
655
656 /* stack argc */
657 csp -= sizeof(unsigned long);
658 __put_user(bprm->argc, (unsigned long __user *) csp);
659
660 BUG_ON(csp != sp);
661
662 /* fill in the argv[] array */
663 #ifdef CONFIG_MMU
664 current->mm->arg_start = bprm->p;
665 #else
666 current->mm->arg_start = current->mm->start_stack -
667 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
668 #endif
669
670 p = (char __user *) current->mm->arg_start;
671 for (loop = bprm->argc; loop > 0; loop--) {
672 __put_user((elf_caddr_t) p, argv++);
673 len = strnlen_user(p, MAX_ARG_STRLEN);
674 if (!len || len > MAX_ARG_STRLEN)
675 return -EINVAL;
676 p += len;
677 }
678 __put_user(NULL, argv);
679 current->mm->arg_end = (unsigned long) p;
680
681 /* fill in the envv[] array */
682 current->mm->env_start = (unsigned long) p;
683 for (loop = bprm->envc; loop > 0; loop--) {
684 __put_user((elf_caddr_t)(unsigned long) p, envp++);
685 len = strnlen_user(p, MAX_ARG_STRLEN);
686 if (!len || len > MAX_ARG_STRLEN)
687 return -EINVAL;
688 p += len;
689 }
690 __put_user(NULL, envp);
691 current->mm->env_end = (unsigned long) p;
692
693 mm->start_stack = (unsigned long) sp;
694 return 0;
695 }
696
697 /*****************************************************************************/
698 /*
699 * transfer the program arguments and environment from the holding pages onto
700 * the stack
701 */
702 #ifndef CONFIG_MMU
703 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
704 unsigned long *_sp)
705 {
706 unsigned long index, stop, sp;
707 char *src;
708 int ret = 0;
709
710 stop = bprm->p >> PAGE_SHIFT;
711 sp = *_sp;
712
713 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
714 src = kmap(bprm->page[index]);
715 sp -= PAGE_SIZE;
716 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
717 ret = -EFAULT;
718 kunmap(bprm->page[index]);
719 if (ret < 0)
720 goto out;
721 }
722
723 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
724
725 out:
726 return ret;
727 }
728 #endif
729
730 /*****************************************************************************/
731 /*
732 * load the appropriate binary image (executable or interpreter) into memory
733 * - we assume no MMU is available
734 * - if no other PIC bits are set in params->hdr->e_flags
735 * - we assume that the LOADable segments in the binary are independently relocatable
736 * - we assume R/O executable segments are shareable
737 * - else
738 * - we assume the loadable parts of the image to require fixed displacement
739 * - the image is not shareable
740 */
741 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
742 struct file *file,
743 struct mm_struct *mm,
744 const char *what)
745 {
746 struct elf32_fdpic_loadmap *loadmap;
747 #ifdef CONFIG_MMU
748 struct elf32_fdpic_loadseg *mseg;
749 #endif
750 struct elf32_fdpic_loadseg *seg;
751 struct elf32_phdr *phdr;
752 unsigned long load_addr, stop;
753 unsigned nloads, tmp;
754 size_t size;
755 int loop, ret;
756
757 /* allocate a load map table */
758 nloads = 0;
759 for (loop = 0; loop < params->hdr.e_phnum; loop++)
760 if (params->phdrs[loop].p_type == PT_LOAD)
761 nloads++;
762
763 if (nloads == 0)
764 return -ELIBBAD;
765
766 size = sizeof(*loadmap) + nloads * sizeof(*seg);
767 loadmap = kzalloc(size, GFP_KERNEL);
768 if (!loadmap)
769 return -ENOMEM;
770
771 params->loadmap = loadmap;
772
773 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
774 loadmap->nsegs = nloads;
775
776 load_addr = params->load_addr;
777 seg = loadmap->segs;
778
779 /* map the requested LOADs into the memory space */
780 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
781 case ELF_FDPIC_FLAG_CONSTDISP:
782 case ELF_FDPIC_FLAG_CONTIGUOUS:
783 #ifndef CONFIG_MMU
784 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
785 if (ret < 0)
786 return ret;
787 break;
788 #endif
789 default:
790 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
791 if (ret < 0)
792 return ret;
793 break;
794 }
795
796 /* map the entry point */
797 if (params->hdr.e_entry) {
798 seg = loadmap->segs;
799 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
800 if (params->hdr.e_entry >= seg->p_vaddr &&
801 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
802 params->entry_addr =
803 (params->hdr.e_entry - seg->p_vaddr) +
804 seg->addr;
805 break;
806 }
807 }
808 }
809
810 /* determine where the program header table has wound up if mapped */
811 stop = params->hdr.e_phoff;
812 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
813 phdr = params->phdrs;
814
815 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
816 if (phdr->p_type != PT_LOAD)
817 continue;
818
819 if (phdr->p_offset > params->hdr.e_phoff ||
820 phdr->p_offset + phdr->p_filesz < stop)
821 continue;
822
823 seg = loadmap->segs;
824 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
825 if (phdr->p_vaddr >= seg->p_vaddr &&
826 phdr->p_vaddr + phdr->p_filesz <=
827 seg->p_vaddr + seg->p_memsz) {
828 params->ph_addr =
829 (phdr->p_vaddr - seg->p_vaddr) +
830 seg->addr +
831 params->hdr.e_phoff - phdr->p_offset;
832 break;
833 }
834 }
835 break;
836 }
837
838 /* determine where the dynamic section has wound up if there is one */
839 phdr = params->phdrs;
840 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
841 if (phdr->p_type != PT_DYNAMIC)
842 continue;
843
844 seg = loadmap->segs;
845 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
846 if (phdr->p_vaddr >= seg->p_vaddr &&
847 phdr->p_vaddr + phdr->p_memsz <=
848 seg->p_vaddr + seg->p_memsz) {
849 params->dynamic_addr =
850 (phdr->p_vaddr - seg->p_vaddr) +
851 seg->addr;
852
853 /* check the dynamic section contains at least
854 * one item, and that the last item is a NULL
855 * entry */
856 if (phdr->p_memsz == 0 ||
857 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
858 goto dynamic_error;
859
860 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
861 if (((Elf32_Dyn *)
862 params->dynamic_addr)[tmp - 1].d_tag != 0)
863 goto dynamic_error;
864 break;
865 }
866 }
867 break;
868 }
869
870 /* now elide adjacent segments in the load map on MMU linux
871 * - on uClinux the holes between may actually be filled with system
872 * stuff or stuff from other processes
873 */
874 #ifdef CONFIG_MMU
875 nloads = loadmap->nsegs;
876 mseg = loadmap->segs;
877 seg = mseg + 1;
878 for (loop = 1; loop < nloads; loop++) {
879 /* see if we have a candidate for merging */
880 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
881 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
882 if (load_addr == (seg->addr & PAGE_MASK)) {
883 mseg->p_memsz +=
884 load_addr -
885 (mseg->addr + mseg->p_memsz);
886 mseg->p_memsz += seg->addr & ~PAGE_MASK;
887 mseg->p_memsz += seg->p_memsz;
888 loadmap->nsegs--;
889 continue;
890 }
891 }
892
893 mseg++;
894 if (mseg != seg)
895 *mseg = *seg;
896 }
897 #endif
898
899 kdebug("Mapped Object [%s]:", what);
900 kdebug("- elfhdr : %lx", params->elfhdr_addr);
901 kdebug("- entry : %lx", params->entry_addr);
902 kdebug("- PHDR[] : %lx", params->ph_addr);
903 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
904 seg = loadmap->segs;
905 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
906 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
907 loop,
908 seg->addr, seg->addr + seg->p_memsz - 1,
909 seg->p_vaddr, seg->p_memsz);
910
911 return 0;
912
913 dynamic_error:
914 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
915 what, file->f_path.dentry->d_inode->i_ino);
916 return -ELIBBAD;
917 }
918
919 /*****************************************************************************/
920 /*
921 * map a file with constant displacement under uClinux
922 */
923 #ifndef CONFIG_MMU
924 static int elf_fdpic_map_file_constdisp_on_uclinux(
925 struct elf_fdpic_params *params,
926 struct file *file,
927 struct mm_struct *mm)
928 {
929 struct elf32_fdpic_loadseg *seg;
930 struct elf32_phdr *phdr;
931 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
932 loff_t fpos;
933 int loop, ret;
934
935 load_addr = params->load_addr;
936 seg = params->loadmap->segs;
937
938 /* determine the bounds of the contiguous overall allocation we must
939 * make */
940 phdr = params->phdrs;
941 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
942 if (params->phdrs[loop].p_type != PT_LOAD)
943 continue;
944
945 if (base > phdr->p_vaddr)
946 base = phdr->p_vaddr;
947 if (top < phdr->p_vaddr + phdr->p_memsz)
948 top = phdr->p_vaddr + phdr->p_memsz;
949 }
950
951 /* allocate one big anon block for everything */
952 mflags = MAP_PRIVATE;
953 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
954 mflags |= MAP_EXECUTABLE;
955
956 down_write(&mm->mmap_sem);
957 maddr = do_mmap(NULL, load_addr, top - base,
958 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
959 up_write(&mm->mmap_sem);
960 if (IS_ERR_VALUE(maddr))
961 return (int) maddr;
962
963 if (load_addr != 0)
964 load_addr += PAGE_ALIGN(top - base);
965
966 /* and then load the file segments into it */
967 phdr = params->phdrs;
968 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
969 if (params->phdrs[loop].p_type != PT_LOAD)
970 continue;
971
972 fpos = phdr->p_offset;
973
974 seg->addr = maddr + (phdr->p_vaddr - base);
975 seg->p_vaddr = phdr->p_vaddr;
976 seg->p_memsz = phdr->p_memsz;
977
978 ret = file->f_op->read(file, (void *) seg->addr,
979 phdr->p_filesz, &fpos);
980 if (ret < 0)
981 return ret;
982
983 /* map the ELF header address if in this segment */
984 if (phdr->p_offset == 0)
985 params->elfhdr_addr = seg->addr;
986
987 /* clear any space allocated but not loaded */
988 if (phdr->p_filesz < phdr->p_memsz) {
989 ret = clear_user((void *) (seg->addr + phdr->p_filesz),
990 phdr->p_memsz - phdr->p_filesz);
991 if (ret)
992 return ret;
993 }
994
995 if (mm) {
996 if (phdr->p_flags & PF_X) {
997 if (!mm->start_code) {
998 mm->start_code = seg->addr;
999 mm->end_code = seg->addr +
1000 phdr->p_memsz;
1001 }
1002 } else if (!mm->start_data) {
1003 mm->start_data = seg->addr;
1004 #ifndef CONFIG_MMU
1005 mm->end_data = seg->addr + phdr->p_memsz;
1006 #endif
1007 }
1008
1009 #ifdef CONFIG_MMU
1010 if (seg->addr + phdr->p_memsz > mm->end_data)
1011 mm->end_data = seg->addr + phdr->p_memsz;
1012 #endif
1013 }
1014
1015 seg++;
1016 }
1017
1018 return 0;
1019 }
1020 #endif
1021
1022 /*****************************************************************************/
1023 /*
1024 * map a binary by direct mmap() of the individual PT_LOAD segments
1025 */
1026 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1027 struct file *file,
1028 struct mm_struct *mm)
1029 {
1030 struct elf32_fdpic_loadseg *seg;
1031 struct elf32_phdr *phdr;
1032 unsigned long load_addr, delta_vaddr;
1033 int loop, dvset, ret;
1034
1035 load_addr = params->load_addr;
1036 delta_vaddr = 0;
1037 dvset = 0;
1038
1039 seg = params->loadmap->segs;
1040
1041 /* deal with each load segment separately */
1042 phdr = params->phdrs;
1043 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1044 unsigned long maddr, disp, excess, excess1;
1045 int prot = 0, flags;
1046
1047 if (phdr->p_type != PT_LOAD)
1048 continue;
1049
1050 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1051 (unsigned long) phdr->p_vaddr,
1052 (unsigned long) phdr->p_offset,
1053 (unsigned long) phdr->p_filesz,
1054 (unsigned long) phdr->p_memsz);
1055
1056 /* determine the mapping parameters */
1057 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1058 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1059 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1060
1061 flags = MAP_PRIVATE | MAP_DENYWRITE;
1062 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1063 flags |= MAP_EXECUTABLE;
1064
1065 maddr = 0;
1066
1067 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1068 case ELF_FDPIC_FLAG_INDEPENDENT:
1069 /* PT_LOADs are independently locatable */
1070 break;
1071
1072 case ELF_FDPIC_FLAG_HONOURVADDR:
1073 /* the specified virtual address must be honoured */
1074 maddr = phdr->p_vaddr;
1075 flags |= MAP_FIXED;
1076 break;
1077
1078 case ELF_FDPIC_FLAG_CONSTDISP:
1079 /* constant displacement
1080 * - can be mapped anywhere, but must be mapped as a
1081 * unit
1082 */
1083 if (!dvset) {
1084 maddr = load_addr;
1085 delta_vaddr = phdr->p_vaddr;
1086 dvset = 1;
1087 } else {
1088 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1089 flags |= MAP_FIXED;
1090 }
1091 break;
1092
1093 case ELF_FDPIC_FLAG_CONTIGUOUS:
1094 /* contiguity handled later */
1095 break;
1096
1097 default:
1098 BUG();
1099 }
1100
1101 maddr &= PAGE_MASK;
1102
1103 /* create the mapping */
1104 disp = phdr->p_vaddr & ~PAGE_MASK;
1105 down_write(&mm->mmap_sem);
1106 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1107 phdr->p_offset - disp);
1108 up_write(&mm->mmap_sem);
1109
1110 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1111 loop, phdr->p_memsz + disp, prot, flags,
1112 phdr->p_offset - disp, maddr);
1113
1114 if (IS_ERR_VALUE(maddr))
1115 return (int) maddr;
1116
1117 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1118 ELF_FDPIC_FLAG_CONTIGUOUS)
1119 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1120
1121 seg->addr = maddr + disp;
1122 seg->p_vaddr = phdr->p_vaddr;
1123 seg->p_memsz = phdr->p_memsz;
1124
1125 /* map the ELF header address if in this segment */
1126 if (phdr->p_offset == 0)
1127 params->elfhdr_addr = seg->addr;
1128
1129 /* clear the bit between beginning of mapping and beginning of
1130 * PT_LOAD */
1131 if (prot & PROT_WRITE && disp > 0) {
1132 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1133 ret = clear_user((void __user *) maddr, disp);
1134 if (ret)
1135 return ret;
1136 maddr += disp;
1137 }
1138
1139 /* clear any space allocated but not loaded
1140 * - on uClinux we can just clear the lot
1141 * - on MMU linux we'll get a SIGBUS beyond the last page
1142 * extant in the file
1143 */
1144 excess = phdr->p_memsz - phdr->p_filesz;
1145 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1146
1147 #ifdef CONFIG_MMU
1148 if (excess > excess1) {
1149 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1150 unsigned long xmaddr;
1151
1152 flags |= MAP_FIXED | MAP_ANONYMOUS;
1153 down_write(&mm->mmap_sem);
1154 xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1155 prot, flags, 0);
1156 up_write(&mm->mmap_sem);
1157
1158 kdebug("mmap[%d] <anon>"
1159 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1160 loop, xaddr, excess - excess1, prot, flags,
1161 xmaddr);
1162
1163 if (xmaddr != xaddr)
1164 return -ENOMEM;
1165 }
1166
1167 if (prot & PROT_WRITE && excess1 > 0) {
1168 kdebug("clear[%d] ad=%lx sz=%lx",
1169 loop, maddr + phdr->p_filesz, excess1);
1170 ret = clear_user((void __user *) maddr + phdr->p_filesz,
1171 excess1);
1172 if (ret)
1173 return ret;
1174 }
1175
1176 #else
1177 if (excess > 0) {
1178 kdebug("clear[%d] ad=%lx sz=%lx",
1179 loop, maddr + phdr->p_filesz, excess);
1180 ret = clear_user((void *) maddr + phdr->p_filesz, excess);
1181 if (ret)
1182 return ret;
1183 }
1184 #endif
1185
1186 if (mm) {
1187 if (phdr->p_flags & PF_X) {
1188 if (!mm->start_code) {
1189 mm->start_code = maddr;
1190 mm->end_code = maddr + phdr->p_memsz;
1191 }
1192 } else if (!mm->start_data) {
1193 mm->start_data = maddr;
1194 mm->end_data = maddr + phdr->p_memsz;
1195 }
1196 }
1197
1198 seg++;
1199 }
1200
1201 return 0;
1202 }
1203
1204 /*****************************************************************************/
1205 /*
1206 * ELF-FDPIC core dumper
1207 *
1208 * Modelled on fs/exec.c:aout_core_dump()
1209 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1210 *
1211 * Modelled on fs/binfmt_elf.c core dumper
1212 */
1213 #ifdef CONFIG_ELF_CORE
1214
1215 /*
1216 * These are the only things you should do on a core-file: use only these
1217 * functions to write out all the necessary info.
1218 */
1219 static int dump_write(struct file *file, const void *addr, int nr)
1220 {
1221 return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1222 }
1223
1224 static int dump_seek(struct file *file, loff_t off)
1225 {
1226 if (file->f_op->llseek) {
1227 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1228 return 0;
1229 } else {
1230 file->f_pos = off;
1231 }
1232 return 1;
1233 }
1234
1235 /*
1236 * Decide whether a segment is worth dumping; default is yes to be
1237 * sure (missing info is worse than too much; etc).
1238 * Personally I'd include everything, and use the coredump limit...
1239 *
1240 * I think we should skip something. But I am not sure how. H.J.
1241 */
1242 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1243 {
1244 int dump_ok;
1245
1246 /* Do not dump I/O mapped devices or special mappings */
1247 if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1248 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1249 return 0;
1250 }
1251
1252 /* If we may not read the contents, don't allow us to dump
1253 * them either. "dump_write()" can't handle it anyway.
1254 */
1255 if (!(vma->vm_flags & VM_READ)) {
1256 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1257 return 0;
1258 }
1259
1260 /* By default, dump shared memory if mapped from an anonymous file. */
1261 if (vma->vm_flags & VM_SHARED) {
1262 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1263 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1264 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1265 vma->vm_flags, dump_ok ? "yes" : "no");
1266 return dump_ok;
1267 }
1268
1269 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1270 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1271 vma->vm_flags, dump_ok ? "yes" : "no");
1272 return dump_ok;
1273 }
1274
1275 #ifdef CONFIG_MMU
1276 /* By default, if it hasn't been written to, don't write it out */
1277 if (!vma->anon_vma) {
1278 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1279 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1280 vma->vm_flags, dump_ok ? "yes" : "no");
1281 return dump_ok;
1282 }
1283 #endif
1284
1285 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1286 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1287 dump_ok ? "yes" : "no");
1288 return dump_ok;
1289 }
1290
1291 /* An ELF note in memory */
1292 struct memelfnote
1293 {
1294 const char *name;
1295 int type;
1296 unsigned int datasz;
1297 void *data;
1298 };
1299
1300 static int notesize(struct memelfnote *en)
1301 {
1302 int sz;
1303
1304 sz = sizeof(struct elf_note);
1305 sz += roundup(strlen(en->name) + 1, 4);
1306 sz += roundup(en->datasz, 4);
1307
1308 return sz;
1309 }
1310
1311 /* #define DEBUG */
1312
1313 #define DUMP_WRITE(addr, nr) \
1314 do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1315 #define DUMP_SEEK(off) \
1316 do { if (!dump_seek(file, (off))) return 0; } while(0)
1317
1318 static int writenote(struct memelfnote *men, struct file *file)
1319 {
1320 struct elf_note en;
1321
1322 en.n_namesz = strlen(men->name) + 1;
1323 en.n_descsz = men->datasz;
1324 en.n_type = men->type;
1325
1326 DUMP_WRITE(&en, sizeof(en));
1327 DUMP_WRITE(men->name, en.n_namesz);
1328 /* XXX - cast from long long to long to avoid need for libgcc.a */
1329 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1330 DUMP_WRITE(men->data, men->datasz);
1331 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1332
1333 return 1;
1334 }
1335 #undef DUMP_WRITE
1336 #undef DUMP_SEEK
1337
1338 #define DUMP_WRITE(addr, nr) \
1339 if ((size += (nr)) > cprm->limit || \
1340 !dump_write(cprm->file, (addr), (nr))) \
1341 goto end_coredump;
1342
1343 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1344 {
1345 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1346 elf->e_ident[EI_CLASS] = ELF_CLASS;
1347 elf->e_ident[EI_DATA] = ELF_DATA;
1348 elf->e_ident[EI_VERSION] = EV_CURRENT;
1349 elf->e_ident[EI_OSABI] = ELF_OSABI;
1350 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1351
1352 elf->e_type = ET_CORE;
1353 elf->e_machine = ELF_ARCH;
1354 elf->e_version = EV_CURRENT;
1355 elf->e_entry = 0;
1356 elf->e_phoff = sizeof(struct elfhdr);
1357 elf->e_shoff = 0;
1358 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1359 elf->e_ehsize = sizeof(struct elfhdr);
1360 elf->e_phentsize = sizeof(struct elf_phdr);
1361 elf->e_phnum = segs;
1362 elf->e_shentsize = 0;
1363 elf->e_shnum = 0;
1364 elf->e_shstrndx = 0;
1365 return;
1366 }
1367
1368 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1369 {
1370 phdr->p_type = PT_NOTE;
1371 phdr->p_offset = offset;
1372 phdr->p_vaddr = 0;
1373 phdr->p_paddr = 0;
1374 phdr->p_filesz = sz;
1375 phdr->p_memsz = 0;
1376 phdr->p_flags = 0;
1377 phdr->p_align = 0;
1378 return;
1379 }
1380
1381 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1382 unsigned int sz, void *data)
1383 {
1384 note->name = name;
1385 note->type = type;
1386 note->datasz = sz;
1387 note->data = data;
1388 return;
1389 }
1390
1391 /*
1392 * fill up all the fields in prstatus from the given task struct, except
1393 * registers which need to be filled up seperately.
1394 */
1395 static void fill_prstatus(struct elf_prstatus *prstatus,
1396 struct task_struct *p, long signr)
1397 {
1398 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1399 prstatus->pr_sigpend = p->pending.signal.sig[0];
1400 prstatus->pr_sighold = p->blocked.sig[0];
1401 rcu_read_lock();
1402 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1403 rcu_read_unlock();
1404 prstatus->pr_pid = task_pid_vnr(p);
1405 prstatus->pr_pgrp = task_pgrp_vnr(p);
1406 prstatus->pr_sid = task_session_vnr(p);
1407 if (thread_group_leader(p)) {
1408 struct task_cputime cputime;
1409
1410 /*
1411 * This is the record for the group leader. It shows the
1412 * group-wide total, not its individual thread total.
1413 */
1414 thread_group_cputime(p, &cputime);
1415 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1416 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1417 } else {
1418 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1419 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1420 }
1421 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1422 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1423
1424 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1425 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1426 }
1427
1428 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1429 struct mm_struct *mm)
1430 {
1431 const struct cred *cred;
1432 unsigned int i, len;
1433
1434 /* first copy the parameters from user space */
1435 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1436
1437 len = mm->arg_end - mm->arg_start;
1438 if (len >= ELF_PRARGSZ)
1439 len = ELF_PRARGSZ - 1;
1440 if (copy_from_user(&psinfo->pr_psargs,
1441 (const char __user *) mm->arg_start, len))
1442 return -EFAULT;
1443 for (i = 0; i < len; i++)
1444 if (psinfo->pr_psargs[i] == 0)
1445 psinfo->pr_psargs[i] = ' ';
1446 psinfo->pr_psargs[len] = 0;
1447
1448 rcu_read_lock();
1449 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1450 rcu_read_unlock();
1451 psinfo->pr_pid = task_pid_vnr(p);
1452 psinfo->pr_pgrp = task_pgrp_vnr(p);
1453 psinfo->pr_sid = task_session_vnr(p);
1454
1455 i = p->state ? ffz(~p->state) + 1 : 0;
1456 psinfo->pr_state = i;
1457 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1458 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1459 psinfo->pr_nice = task_nice(p);
1460 psinfo->pr_flag = p->flags;
1461 rcu_read_lock();
1462 cred = __task_cred(p);
1463 SET_UID(psinfo->pr_uid, cred->uid);
1464 SET_GID(psinfo->pr_gid, cred->gid);
1465 rcu_read_unlock();
1466 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1467
1468 return 0;
1469 }
1470
1471 /* Here is the structure in which status of each thread is captured. */
1472 struct elf_thread_status
1473 {
1474 struct list_head list;
1475 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1476 elf_fpregset_t fpu; /* NT_PRFPREG */
1477 struct task_struct *thread;
1478 #ifdef ELF_CORE_COPY_XFPREGS
1479 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1480 #endif
1481 struct memelfnote notes[3];
1482 int num_notes;
1483 };
1484
1485 /*
1486 * In order to add the specific thread information for the elf file format,
1487 * we need to keep a linked list of every thread's pr_status and then create
1488 * a single section for them in the final core file.
1489 */
1490 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1491 {
1492 struct task_struct *p = t->thread;
1493 int sz = 0;
1494
1495 t->num_notes = 0;
1496
1497 fill_prstatus(&t->prstatus, p, signr);
1498 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1499
1500 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1501 &t->prstatus);
1502 t->num_notes++;
1503 sz += notesize(&t->notes[0]);
1504
1505 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1506 if (t->prstatus.pr_fpvalid) {
1507 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1508 &t->fpu);
1509 t->num_notes++;
1510 sz += notesize(&t->notes[1]);
1511 }
1512
1513 #ifdef ELF_CORE_COPY_XFPREGS
1514 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1515 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1516 sizeof(t->xfpu), &t->xfpu);
1517 t->num_notes++;
1518 sz += notesize(&t->notes[2]);
1519 }
1520 #endif
1521 return sz;
1522 }
1523
1524 /*
1525 * dump the segments for an MMU process
1526 */
1527 #ifdef CONFIG_MMU
1528 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1529 unsigned long *limit, unsigned long mm_flags)
1530 {
1531 struct vm_area_struct *vma;
1532 int err = 0;
1533
1534 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1535 unsigned long addr;
1536
1537 if (!maydump(vma, mm_flags))
1538 continue;
1539
1540 for (addr = vma->vm_start; addr < vma->vm_end;
1541 addr += PAGE_SIZE) {
1542 struct page *page = get_dump_page(addr);
1543 if (page) {
1544 void *kaddr = kmap(page);
1545 *size += PAGE_SIZE;
1546 if (*size > *limit)
1547 err = -EFBIG;
1548 else if (!dump_write(file, kaddr, PAGE_SIZE))
1549 err = -EIO;
1550 kunmap(page);
1551 page_cache_release(page);
1552 } else if (!dump_seek(file, file->f_pos + PAGE_SIZE))
1553 err = -EFBIG;
1554 if (err)
1555 goto out;
1556 }
1557 }
1558 out:
1559 return err;
1560 }
1561 #endif
1562
1563 /*
1564 * dump the segments for a NOMMU process
1565 */
1566 #ifndef CONFIG_MMU
1567 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1568 unsigned long *limit, unsigned long mm_flags)
1569 {
1570 struct vm_area_struct *vma;
1571
1572 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1573 if (!maydump(vma, mm_flags))
1574 continue;
1575
1576 if ((*size += PAGE_SIZE) > *limit)
1577 return -EFBIG;
1578
1579 if (!dump_write(file, (void *) vma->vm_start,
1580 vma->vm_end - vma->vm_start))
1581 return -EIO;
1582 }
1583
1584 return 0;
1585 }
1586 #endif
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;
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 unsigned long mm_flags;
1619
1620 /*
1621 * We no longer stop all VM operations.
1622 *
1623 * This is because those proceses that could possibly change map_count
1624 * or the mmap / vma pages are now blocked in do_exit on current
1625 * finishing this core dump.
1626 *
1627 * Only ptrace can touch these memory addresses, but it doesn't change
1628 * the map_count or the pages allocated. So no possibility of crashing
1629 * exists while dumping the mm->vm_next areas to the core file.
1630 */
1631
1632 /* alloc memory for large data structures: too large to be on stack */
1633 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1634 if (!elf)
1635 goto cleanup;
1636 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1637 if (!prstatus)
1638 goto cleanup;
1639 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1640 if (!psinfo)
1641 goto cleanup;
1642 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1643 if (!notes)
1644 goto cleanup;
1645 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1646 if (!fpu)
1647 goto cleanup;
1648 #ifdef ELF_CORE_COPY_XFPREGS
1649 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1650 if (!xfpu)
1651 goto cleanup;
1652 #endif
1653
1654 if (cprm->signr) {
1655 struct core_thread *ct;
1656 struct elf_thread_status *tmp;
1657
1658 for (ct = current->mm->core_state->dumper.next;
1659 ct; ct = ct->next) {
1660 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1661 if (!tmp)
1662 goto cleanup;
1663
1664 tmp->thread = ct->task;
1665 list_add(&tmp->list, &thread_list);
1666 }
1667
1668 list_for_each(t, &thread_list) {
1669 struct elf_thread_status *tmp;
1670 int sz;
1671
1672 tmp = list_entry(t, struct elf_thread_status, list);
1673 sz = elf_dump_thread_status(cprm->signr, tmp);
1674 thread_status_size += sz;
1675 }
1676 }
1677
1678 /* now collect the dump for the current */
1679 fill_prstatus(prstatus, current, cprm->signr);
1680 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1681
1682 segs = current->mm->map_count;
1683 #ifdef ELF_CORE_EXTRA_PHDRS
1684 segs += ELF_CORE_EXTRA_PHDRS;
1685 #endif
1686
1687 /* Set up header */
1688 fill_elf_fdpic_header(elf, segs + 1); /* including notes section */
1689
1690 has_dumped = 1;
1691 current->flags |= PF_DUMPCORE;
1692
1693 /*
1694 * Set up the notes in similar form to SVR4 core dumps made
1695 * with info from their /proc.
1696 */
1697
1698 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1699 fill_psinfo(psinfo, current->group_leader, current->mm);
1700 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1701
1702 numnote = 2;
1703
1704 auxv = (elf_addr_t *) current->mm->saved_auxv;
1705
1706 i = 0;
1707 do
1708 i += 2;
1709 while (auxv[i - 2] != AT_NULL);
1710 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1711 i * sizeof(elf_addr_t), auxv);
1712
1713 /* Try to dump the FPU. */
1714 if ((prstatus->pr_fpvalid =
1715 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1716 fill_note(notes + numnote++,
1717 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1718 #ifdef ELF_CORE_COPY_XFPREGS
1719 if (elf_core_copy_task_xfpregs(current, xfpu))
1720 fill_note(notes + numnote++,
1721 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1722 #endif
1723
1724 fs = get_fs();
1725 set_fs(KERNEL_DS);
1726
1727 DUMP_WRITE(elf, sizeof(*elf));
1728 offset += sizeof(*elf); /* Elf header */
1729 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers */
1730
1731 /* Write notes phdr entry */
1732 {
1733 struct elf_phdr phdr;
1734 int sz = 0;
1735
1736 for (i = 0; i < numnote; i++)
1737 sz += notesize(notes + i);
1738
1739 sz += thread_status_size;
1740
1741 fill_elf_note_phdr(&phdr, sz, offset);
1742 offset += sz;
1743 DUMP_WRITE(&phdr, sizeof(phdr));
1744 }
1745
1746 /* Page-align dumped data */
1747 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1748
1749 /*
1750 * We must use the same mm->flags while dumping core to avoid
1751 * inconsistency between the program headers and bodies, otherwise an
1752 * unusable core file can be generated.
1753 */
1754 mm_flags = current->mm->flags;
1755
1756 /* write program headers for segments dump */
1757 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1758 struct elf_phdr phdr;
1759 size_t sz;
1760
1761 sz = vma->vm_end - vma->vm_start;
1762
1763 phdr.p_type = PT_LOAD;
1764 phdr.p_offset = offset;
1765 phdr.p_vaddr = vma->vm_start;
1766 phdr.p_paddr = 0;
1767 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1768 phdr.p_memsz = sz;
1769 offset += phdr.p_filesz;
1770 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1771 if (vma->vm_flags & VM_WRITE)
1772 phdr.p_flags |= PF_W;
1773 if (vma->vm_flags & VM_EXEC)
1774 phdr.p_flags |= PF_X;
1775 phdr.p_align = ELF_EXEC_PAGESIZE;
1776
1777 DUMP_WRITE(&phdr, sizeof(phdr));
1778 }
1779
1780 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1781 ELF_CORE_WRITE_EXTRA_PHDRS;
1782 #endif
1783
1784 /* write out the notes section */
1785 for (i = 0; i < numnote; i++)
1786 if (!writenote(notes + i, cprm->file))
1787 goto end_coredump;
1788
1789 /* write out the thread status notes section */
1790 list_for_each(t, &thread_list) {
1791 struct elf_thread_status *tmp =
1792 list_entry(t, struct elf_thread_status, list);
1793
1794 for (i = 0; i < tmp->num_notes; i++)
1795 if (!writenote(&tmp->notes[i], cprm->file))
1796 goto end_coredump;
1797 }
1798
1799 if (!dump_seek(cprm->file, dataoff))
1800 goto end_coredump;
1801
1802 if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit,
1803 mm_flags) < 0)
1804 goto end_coredump;
1805
1806 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1807 ELF_CORE_WRITE_EXTRA_DATA;
1808 #endif
1809
1810 if (cprm->file->f_pos != offset) {
1811 /* Sanity check */
1812 printk(KERN_WARNING
1813 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1814 cprm->file->f_pos, offset);
1815 }
1816
1817 end_coredump:
1818 set_fs(fs);
1819
1820 cleanup:
1821 while (!list_empty(&thread_list)) {
1822 struct list_head *tmp = thread_list.next;
1823 list_del(tmp);
1824 kfree(list_entry(tmp, struct elf_thread_status, list));
1825 }
1826
1827 kfree(elf);
1828 kfree(prstatus);
1829 kfree(psinfo);
1830 kfree(notes);
1831 kfree(fpu);
1832 #ifdef ELF_CORE_COPY_XFPREGS
1833 kfree(xfpu);
1834 #endif
1835 return has_dumped;
1836 #undef NUM_NOTES
1837 }
1838
1839 #endif /* CONFIG_ELF_CORE */
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree