?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Cleanup handling of NULL value passed for a mount option
[linux/fpc-iii.git] / fs / binfmt_elf_fdpic.c
blob9bd5612a8224fc5c6374d4dcc27ff99a58ee76e5
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2 *
3 * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 * Derived from binfmt_elf.c
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13 #include <linux/module.h>
14
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37 #include <linux/coredump.h>
38
39 #include <asm/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/pgalloc.h>
42 #include <asm/exec.h>
43
44 typedef char *elf_caddr_t;
45
46 #if 0
47 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
48 #else
49 #define kdebug(fmt, ...) do {} while(0)
50 #endif
51
52 #if 0
53 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
54 #else
55 #define kdcore(fmt, ...) do {} while(0)
56 #endif
57
58 MODULE_LICENSE("GPL");
59
60 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
61 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
62 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
63 struct mm_struct *, const char *);
64
65 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
66 struct elf_fdpic_params *,
67 struct elf_fdpic_params *);
68
69 #ifndef CONFIG_MMU
70 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
71 unsigned long *);
72 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
73 struct file *,
74 struct mm_struct *);
75 #endif
76
77 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
78 struct file *, struct mm_struct *);
79
80 #ifdef CONFIG_ELF_CORE
81 static int elf_fdpic_core_dump(struct coredump_params *cprm);
82 #endif
83
84 static struct linux_binfmt elf_fdpic_format = {
85 .module = THIS_MODULE,
86 .load_binary = load_elf_fdpic_binary,
87 #ifdef CONFIG_ELF_CORE
88 .core_dump = elf_fdpic_core_dump,
89 #endif
90 .min_coredump = ELF_EXEC_PAGESIZE,
91 };
92
93 static int __init init_elf_fdpic_binfmt(void)
94 {
95 register_binfmt(&elf_fdpic_format);
96 return 0;
97 }
98
99 static void __exit exit_elf_fdpic_binfmt(void)
100 {
101 unregister_binfmt(&elf_fdpic_format);
102 }
103
104 core_initcall(init_elf_fdpic_binfmt);
105 module_exit(exit_elf_fdpic_binfmt);
106
107 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
108 {
109 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
110 return 0;
111 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
112 return 0;
113 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
114 return 0;
115 if (!file->f_op || !file->f_op->mmap)
116 return 0;
117 return 1;
118 }
119
120 /*****************************************************************************/
121 /*
122 * read the program headers table into memory
123 */
124 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
125 struct file *file)
126 {
127 struct elf32_phdr *phdr;
128 unsigned long size;
129 int retval, loop;
130
131 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
132 return -ENOMEM;
133 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
134 return -ENOMEM;
135
136 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
137 params->phdrs = kmalloc(size, GFP_KERNEL);
138 if (!params->phdrs)
139 return -ENOMEM;
140
141 retval = kernel_read(file, params->hdr.e_phoff,
142 (char *) params->phdrs, size);
143 if (unlikely(retval != size))
144 return retval < 0 ? retval : -ENOEXEC;
145
146 /* determine stack size for this binary */
147 phdr = params->phdrs;
148 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
149 if (phdr->p_type != PT_GNU_STACK)
150 continue;
151
152 if (phdr->p_flags & PF_X)
153 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
154 else
155 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
156
157 params->stack_size = phdr->p_memsz;
158 break;
159 }
160
161 return 0;
162 }
163
164 /*****************************************************************************/
165 /*
166 * load an fdpic binary into various bits of memory
167 */
168 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
169 struct pt_regs *regs)
170 {
171 struct elf_fdpic_params exec_params, interp_params;
172 struct elf_phdr *phdr;
173 unsigned long stack_size, entryaddr;
174 #ifdef ELF_FDPIC_PLAT_INIT
175 unsigned long dynaddr;
176 #endif
177 #ifndef CONFIG_MMU
178 unsigned long stack_prot;
179 #endif
180 struct file *interpreter = NULL; /* to shut gcc up */
181 char *interpreter_name = NULL;
182 int executable_stack;
183 int retval, i;
184
185 kdebug("____ LOAD %d ____", current->pid);
186
187 memset(&exec_params, 0, sizeof(exec_params));
188 memset(&interp_params, 0, sizeof(interp_params));
189
190 exec_params.hdr = *(struct elfhdr *) bprm->buf;
191 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
192
193 /* check that this is a binary we know how to deal with */
194 retval = -ENOEXEC;
195 if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
196 goto error;
197
198 /* read the program header table */
199 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
200 if (retval < 0)
201 goto error;
202
203 /* scan for a program header that specifies an interpreter */
204 phdr = exec_params.phdrs;
205
206 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
207 switch (phdr->p_type) {
208 case PT_INTERP:
209 retval = -ENOMEM;
210 if (phdr->p_filesz > PATH_MAX)
211 goto error;
212 retval = -ENOENT;
213 if (phdr->p_filesz < 2)
214 goto error;
215
216 /* read the name of the interpreter into memory */
217 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
218 if (!interpreter_name)
219 goto error;
220
221 retval = kernel_read(bprm->file,
222 phdr->p_offset,
223 interpreter_name,
224 phdr->p_filesz);
225 if (unlikely(retval != phdr->p_filesz)) {
226 if (retval >= 0)
227 retval = -ENOEXEC;
228 goto error;
229 }
230
231 retval = -ENOENT;
232 if (interpreter_name[phdr->p_filesz - 1] != '\0')
233 goto error;
234
235 kdebug("Using ELF interpreter %s", interpreter_name);
236
237 /* replace the program with the interpreter */
238 interpreter = open_exec(interpreter_name);
239 retval = PTR_ERR(interpreter);
240 if (IS_ERR(interpreter)) {
241 interpreter = NULL;
242 goto error;
243 }
244
245 /*
246 * If the binary is not readable then enforce
247 * mm->dumpable = 0 regardless of the interpreter's
248 * permissions.
249 */
250 would_dump(bprm, interpreter);
251
252 retval = kernel_read(interpreter, 0, bprm->buf,
253 BINPRM_BUF_SIZE);
254 if (unlikely(retval != BINPRM_BUF_SIZE)) {
255 if (retval >= 0)
256 retval = -ENOEXEC;
257 goto error;
258 }
259
260 interp_params.hdr = *((struct elfhdr *) bprm->buf);
261 break;
262
263 case PT_LOAD:
264 #ifdef CONFIG_MMU
265 if (exec_params.load_addr == 0)
266 exec_params.load_addr = phdr->p_vaddr;
267 #endif
268 break;
269 }
270
271 }
272
273 if (elf_check_const_displacement(&exec_params.hdr))
274 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
275
276 /* perform insanity checks on the interpreter */
277 if (interpreter_name) {
278 retval = -ELIBBAD;
279 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
280 goto error;
281
282 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
283
284 /* read the interpreter's program header table */
285 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
286 if (retval < 0)
287 goto error;
288 }
289
290 stack_size = exec_params.stack_size;
291 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
292 executable_stack = EXSTACK_ENABLE_X;
293 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
294 executable_stack = EXSTACK_DISABLE_X;
295 else
296 executable_stack = EXSTACK_DEFAULT;
297
298 if (stack_size == 0) {
299 stack_size = interp_params.stack_size;
300 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
301 executable_stack = EXSTACK_ENABLE_X;
302 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
303 executable_stack = EXSTACK_DISABLE_X;
304 else
305 executable_stack = EXSTACK_DEFAULT;
306 }
307
308 retval = -ENOEXEC;
309 if (stack_size == 0)
310 goto error;
311
312 if (elf_check_const_displacement(&interp_params.hdr))
313 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
314
315 /* flush all traces of the currently running executable */
316 retval = flush_old_exec(bprm);
317 if (retval)
318 goto error;
319
320 /* there's now no turning back... the old userspace image is dead,
321 * defunct, deceased, etc. after this point we have to exit via
322 * error_kill */
323 set_personality(PER_LINUX_FDPIC);
324 if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
325 current->personality |= READ_IMPLIES_EXEC;
326
327 setup_new_exec(bprm);
328
329 set_binfmt(&elf_fdpic_format);
330
331 current->mm->start_code = 0;
332 current->mm->end_code = 0;
333 current->mm->start_stack = 0;
334 current->mm->start_data = 0;
335 current->mm->end_data = 0;
336 current->mm->context.exec_fdpic_loadmap = 0;
337 current->mm->context.interp_fdpic_loadmap = 0;
338
339 #ifdef CONFIG_MMU
340 elf_fdpic_arch_lay_out_mm(&exec_params,
341 &interp_params,
342 &current->mm->start_stack,
343 &current->mm->start_brk);
344
345 retval = setup_arg_pages(bprm, current->mm->start_stack,
346 executable_stack);
347 if (retval < 0) {
348 send_sig(SIGKILL, current, 0);
349 goto error_kill;
350 }
351 #endif
352
353 /* load the executable and interpreter into memory */
354 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
355 "executable");
356 if (retval < 0)
357 goto error_kill;
358
359 if (interpreter_name) {
360 retval = elf_fdpic_map_file(&interp_params, interpreter,
361 current->mm, "interpreter");
362 if (retval < 0) {
363 printk(KERN_ERR "Unable to load interpreter\n");
364 goto error_kill;
365 }
366
367 allow_write_access(interpreter);
368 fput(interpreter);
369 interpreter = NULL;
370 }
371
372 #ifdef CONFIG_MMU
373 if (!current->mm->start_brk)
374 current->mm->start_brk = current->mm->end_data;
375
376 current->mm->brk = current->mm->start_brk =
377 PAGE_ALIGN(current->mm->start_brk);
378
379 #else
380 /* create a stack and brk area big enough for everyone
381 * - the brk heap starts at the bottom and works up
382 * - the stack starts at the top and works down
383 */
384 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
385 if (stack_size < PAGE_SIZE * 2)
386 stack_size = PAGE_SIZE * 2;
387
388 stack_prot = PROT_READ | PROT_WRITE;
389 if (executable_stack == EXSTACK_ENABLE_X ||
390 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
391 stack_prot |= PROT_EXEC;
392
393 down_write(&current->mm->mmap_sem);
394 current->mm->start_brk = do_mmap(NULL, 0, stack_size, stack_prot,
395 MAP_PRIVATE | MAP_ANONYMOUS |
396 MAP_UNINITIALIZED | MAP_GROWSDOWN,
397 0);
398
399 if (IS_ERR_VALUE(current->mm->start_brk)) {
400 up_write(&current->mm->mmap_sem);
401 retval = current->mm->start_brk;
402 current->mm->start_brk = 0;
403 goto error_kill;
404 }
405
406 up_write(&current->mm->mmap_sem);
407
408 current->mm->brk = current->mm->start_brk;
409 current->mm->context.end_brk = current->mm->start_brk;
410 current->mm->context.end_brk +=
411 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
412 current->mm->start_stack = current->mm->start_brk + stack_size;
413 #endif
414
415 install_exec_creds(bprm);
416 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 if (clear_user((void *) (seg->addr + phdr->p_filesz),
992 phdr->p_memsz - phdr->p_filesz))
993 return -EFAULT;
994 }
995
996 if (mm) {
997 if (phdr->p_flags & PF_X) {
998 if (!mm->start_code) {
999 mm->start_code = seg->addr;
1000 mm->end_code = seg->addr +
1001 phdr->p_memsz;
1002 }
1003 } else if (!mm->start_data) {
1004 mm->start_data = seg->addr;
1005 mm->end_data = seg->addr + phdr->p_memsz;
1006 }
1007 }
1008
1009 seg++;
1010 }
1011
1012 return 0;
1013 }
1014 #endif
1015
1016 /*****************************************************************************/
1017 /*
1018 * map a binary by direct mmap() of the individual PT_LOAD segments
1019 */
1020 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1021 struct file *file,
1022 struct mm_struct *mm)
1023 {
1024 struct elf32_fdpic_loadseg *seg;
1025 struct elf32_phdr *phdr;
1026 unsigned long load_addr, delta_vaddr;
1027 int loop, dvset;
1028
1029 load_addr = params->load_addr;
1030 delta_vaddr = 0;
1031 dvset = 0;
1032
1033 seg = params->loadmap->segs;
1034
1035 /* deal with each load segment separately */
1036 phdr = params->phdrs;
1037 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1038 unsigned long maddr, disp, excess, excess1;
1039 int prot = 0, flags;
1040
1041 if (phdr->p_type != PT_LOAD)
1042 continue;
1043
1044 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1045 (unsigned long) phdr->p_vaddr,
1046 (unsigned long) phdr->p_offset,
1047 (unsigned long) phdr->p_filesz,
1048 (unsigned long) phdr->p_memsz);
1049
1050 /* determine the mapping parameters */
1051 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1052 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1053 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1054
1055 flags = MAP_PRIVATE | MAP_DENYWRITE;
1056 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1057 flags |= MAP_EXECUTABLE;
1058
1059 maddr = 0;
1060
1061 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1062 case ELF_FDPIC_FLAG_INDEPENDENT:
1063 /* PT_LOADs are independently locatable */
1064 break;
1065
1066 case ELF_FDPIC_FLAG_HONOURVADDR:
1067 /* the specified virtual address must be honoured */
1068 maddr = phdr->p_vaddr;
1069 flags |= MAP_FIXED;
1070 break;
1071
1072 case ELF_FDPIC_FLAG_CONSTDISP:
1073 /* constant displacement
1074 * - can be mapped anywhere, but must be mapped as a
1075 * unit
1076 */
1077 if (!dvset) {
1078 maddr = load_addr;
1079 delta_vaddr = phdr->p_vaddr;
1080 dvset = 1;
1081 } else {
1082 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1083 flags |= MAP_FIXED;
1084 }
1085 break;
1086
1087 case ELF_FDPIC_FLAG_CONTIGUOUS:
1088 /* contiguity handled later */
1089 break;
1090
1091 default:
1092 BUG();
1093 }
1094
1095 maddr &= PAGE_MASK;
1096
1097 /* create the mapping */
1098 disp = phdr->p_vaddr & ~PAGE_MASK;
1099 down_write(&mm->mmap_sem);
1100 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1101 phdr->p_offset - disp);
1102 up_write(&mm->mmap_sem);
1103
1104 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1105 loop, phdr->p_memsz + disp, prot, flags,
1106 phdr->p_offset - disp, maddr);
1107
1108 if (IS_ERR_VALUE(maddr))
1109 return (int) maddr;
1110
1111 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1112 ELF_FDPIC_FLAG_CONTIGUOUS)
1113 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1114
1115 seg->addr = maddr + disp;
1116 seg->p_vaddr = phdr->p_vaddr;
1117 seg->p_memsz = phdr->p_memsz;
1118
1119 /* map the ELF header address if in this segment */
1120 if (phdr->p_offset == 0)
1121 params->elfhdr_addr = seg->addr;
1122
1123 /* clear the bit between beginning of mapping and beginning of
1124 * PT_LOAD */
1125 if (prot & PROT_WRITE && disp > 0) {
1126 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1127 if (clear_user((void __user *) maddr, disp))
1128 return -EFAULT;
1129 maddr += disp;
1130 }
1131
1132 /* clear any space allocated but not loaded
1133 * - on uClinux we can just clear the lot
1134 * - on MMU linux we'll get a SIGBUS beyond the last page
1135 * extant in the file
1136 */
1137 excess = phdr->p_memsz - phdr->p_filesz;
1138 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1139
1140 #ifdef CONFIG_MMU
1141 if (excess > excess1) {
1142 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1143 unsigned long xmaddr;
1144
1145 flags |= MAP_FIXED | MAP_ANONYMOUS;
1146 down_write(&mm->mmap_sem);
1147 xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1148 prot, flags, 0);
1149 up_write(&mm->mmap_sem);
1150
1151 kdebug("mmap[%d] <anon>"
1152 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1153 loop, xaddr, excess - excess1, prot, flags,
1154 xmaddr);
1155
1156 if (xmaddr != xaddr)
1157 return -ENOMEM;
1158 }
1159
1160 if (prot & PROT_WRITE && excess1 > 0) {
1161 kdebug("clear[%d] ad=%lx sz=%lx",
1162 loop, maddr + phdr->p_filesz, excess1);
1163 if (clear_user((void __user *) maddr + phdr->p_filesz,
1164 excess1))
1165 return -EFAULT;
1166 }
1167
1168 #else
1169 if (excess > 0) {
1170 kdebug("clear[%d] ad=%lx sz=%lx",
1171 loop, maddr + phdr->p_filesz, excess);
1172 if (clear_user((void *) maddr + phdr->p_filesz, excess))
1173 return -EFAULT;
1174 }
1175 #endif
1176
1177 if (mm) {
1178 if (phdr->p_flags & PF_X) {
1179 if (!mm->start_code) {
1180 mm->start_code = maddr;
1181 mm->end_code = maddr + phdr->p_memsz;
1182 }
1183 } else if (!mm->start_data) {
1184 mm->start_data = maddr;
1185 mm->end_data = maddr + phdr->p_memsz;
1186 }
1187 }
1188
1189 seg++;
1190 }
1191
1192 return 0;
1193 }
1194
1195 /*****************************************************************************/
1196 /*
1197 * ELF-FDPIC core dumper
1198 *
1199 * Modelled on fs/exec.c:aout_core_dump()
1200 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1201 *
1202 * Modelled on fs/binfmt_elf.c core dumper
1203 */
1204 #ifdef CONFIG_ELF_CORE
1205
1206 /*
1207 * Decide whether a segment is worth dumping; default is yes to be
1208 * sure (missing info is worse than too much; etc).
1209 * Personally I'd include everything, and use the coredump limit...
1210 *
1211 * I think we should skip something. But I am not sure how. H.J.
1212 */
1213 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1214 {
1215 int dump_ok;
1216
1217 /* Do not dump I/O mapped devices or special mappings */
1218 if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1219 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1220 return 0;
1221 }
1222
1223 /* If we may not read the contents, don't allow us to dump
1224 * them either. "dump_write()" can't handle it anyway.
1225 */
1226 if (!(vma->vm_flags & VM_READ)) {
1227 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1228 return 0;
1229 }
1230
1231 /* By default, dump shared memory if mapped from an anonymous file. */
1232 if (vma->vm_flags & VM_SHARED) {
1233 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1234 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1235 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1236 vma->vm_flags, dump_ok ? "yes" : "no");
1237 return dump_ok;
1238 }
1239
1240 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1241 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1242 vma->vm_flags, dump_ok ? "yes" : "no");
1243 return dump_ok;
1244 }
1245
1246 #ifdef CONFIG_MMU
1247 /* By default, if it hasn't been written to, don't write it out */
1248 if (!vma->anon_vma) {
1249 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1250 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1251 vma->vm_flags, dump_ok ? "yes" : "no");
1252 return dump_ok;
1253 }
1254 #endif
1255
1256 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1257 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1258 dump_ok ? "yes" : "no");
1259 return dump_ok;
1260 }
1261
1262 /* An ELF note in memory */
1263 struct memelfnote
1264 {
1265 const char *name;
1266 int type;
1267 unsigned int datasz;
1268 void *data;
1269 };
1270
1271 static int notesize(struct memelfnote *en)
1272 {
1273 int sz;
1274
1275 sz = sizeof(struct elf_note);
1276 sz += roundup(strlen(en->name) + 1, 4);
1277 sz += roundup(en->datasz, 4);
1278
1279 return sz;
1280 }
1281
1282 /* #define DEBUG */
1283
1284 #define DUMP_WRITE(addr, nr, foffset) \
1285 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1286
1287 static int alignfile(struct file *file, loff_t *foffset)
1288 {
1289 static const char buf[4] = { 0, };
1290 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1291 return 1;
1292 }
1293
1294 static int writenote(struct memelfnote *men, struct file *file,
1295 loff_t *foffset)
1296 {
1297 struct elf_note en;
1298 en.n_namesz = strlen(men->name) + 1;
1299 en.n_descsz = men->datasz;
1300 en.n_type = men->type;
1301
1302 DUMP_WRITE(&en, sizeof(en), foffset);
1303 DUMP_WRITE(men->name, en.n_namesz, foffset);
1304 if (!alignfile(file, foffset))
1305 return 0;
1306 DUMP_WRITE(men->data, men->datasz, foffset);
1307 if (!alignfile(file, foffset))
1308 return 0;
1309
1310 return 1;
1311 }
1312 #undef DUMP_WRITE
1313
1314 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1315 {
1316 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1317 elf->e_ident[EI_CLASS] = ELF_CLASS;
1318 elf->e_ident[EI_DATA] = ELF_DATA;
1319 elf->e_ident[EI_VERSION] = EV_CURRENT;
1320 elf->e_ident[EI_OSABI] = ELF_OSABI;
1321 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1322
1323 elf->e_type = ET_CORE;
1324 elf->e_machine = ELF_ARCH;
1325 elf->e_version = EV_CURRENT;
1326 elf->e_entry = 0;
1327 elf->e_phoff = sizeof(struct elfhdr);
1328 elf->e_shoff = 0;
1329 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1330 elf->e_ehsize = sizeof(struct elfhdr);
1331 elf->e_phentsize = sizeof(struct elf_phdr);
1332 elf->e_phnum = segs;
1333 elf->e_shentsize = 0;
1334 elf->e_shnum = 0;
1335 elf->e_shstrndx = 0;
1336 return;
1337 }
1338
1339 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1340 {
1341 phdr->p_type = PT_NOTE;
1342 phdr->p_offset = offset;
1343 phdr->p_vaddr = 0;
1344 phdr->p_paddr = 0;
1345 phdr->p_filesz = sz;
1346 phdr->p_memsz = 0;
1347 phdr->p_flags = 0;
1348 phdr->p_align = 0;
1349 return;
1350 }
1351
1352 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1353 unsigned int sz, void *data)
1354 {
1355 note->name = name;
1356 note->type = type;
1357 note->datasz = sz;
1358 note->data = data;
1359 return;
1360 }
1361
1362 /*
1363 * fill up all the fields in prstatus from the given task struct, except
1364 * registers which need to be filled up separately.
1365 */
1366 static void fill_prstatus(struct elf_prstatus *prstatus,
1367 struct task_struct *p, long signr)
1368 {
1369 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1370 prstatus->pr_sigpend = p->pending.signal.sig[0];
1371 prstatus->pr_sighold = p->blocked.sig[0];
1372 rcu_read_lock();
1373 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1374 rcu_read_unlock();
1375 prstatus->pr_pid = task_pid_vnr(p);
1376 prstatus->pr_pgrp = task_pgrp_vnr(p);
1377 prstatus->pr_sid = task_session_vnr(p);
1378 if (thread_group_leader(p)) {
1379 struct task_cputime cputime;
1380
1381 /*
1382 * This is the record for the group leader. It shows the
1383 * group-wide total, not its individual thread total.
1384 */
1385 thread_group_cputime(p, &cputime);
1386 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1387 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1388 } else {
1389 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1390 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1391 }
1392 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1393 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1394
1395 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1396 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1397 }
1398
1399 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1400 struct mm_struct *mm)
1401 {
1402 const struct cred *cred;
1403 unsigned int i, len;
1404
1405 /* first copy the parameters from user space */
1406 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1407
1408 len = mm->arg_end - mm->arg_start;
1409 if (len >= ELF_PRARGSZ)
1410 len = ELF_PRARGSZ - 1;
1411 if (copy_from_user(&psinfo->pr_psargs,
1412 (const char __user *) mm->arg_start, len))
1413 return -EFAULT;
1414 for (i = 0; i < len; i++)
1415 if (psinfo->pr_psargs[i] == 0)
1416 psinfo->pr_psargs[i] = ' ';
1417 psinfo->pr_psargs[len] = 0;
1418
1419 rcu_read_lock();
1420 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1421 rcu_read_unlock();
1422 psinfo->pr_pid = task_pid_vnr(p);
1423 psinfo->pr_pgrp = task_pgrp_vnr(p);
1424 psinfo->pr_sid = task_session_vnr(p);
1425
1426 i = p->state ? ffz(~p->state) + 1 : 0;
1427 psinfo->pr_state = i;
1428 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1429 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1430 psinfo->pr_nice = task_nice(p);
1431 psinfo->pr_flag = p->flags;
1432 rcu_read_lock();
1433 cred = __task_cred(p);
1434 SET_UID(psinfo->pr_uid, cred->uid);
1435 SET_GID(psinfo->pr_gid, cred->gid);
1436 rcu_read_unlock();
1437 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1438
1439 return 0;
1440 }
1441
1442 /* Here is the structure in which status of each thread is captured. */
1443 struct elf_thread_status
1444 {
1445 struct list_head list;
1446 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1447 elf_fpregset_t fpu; /* NT_PRFPREG */
1448 struct task_struct *thread;
1449 #ifdef ELF_CORE_COPY_XFPREGS
1450 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1451 #endif
1452 struct memelfnote notes[3];
1453 int num_notes;
1454 };
1455
1456 /*
1457 * In order to add the specific thread information for the elf file format,
1458 * we need to keep a linked list of every thread's pr_status and then create
1459 * a single section for them in the final core file.
1460 */
1461 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1462 {
1463 struct task_struct *p = t->thread;
1464 int sz = 0;
1465
1466 t->num_notes = 0;
1467
1468 fill_prstatus(&t->prstatus, p, signr);
1469 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1470
1471 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1472 &t->prstatus);
1473 t->num_notes++;
1474 sz += notesize(&t->notes[0]);
1475
1476 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1477 if (t->prstatus.pr_fpvalid) {
1478 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1479 &t->fpu);
1480 t->num_notes++;
1481 sz += notesize(&t->notes[1]);
1482 }
1483
1484 #ifdef ELF_CORE_COPY_XFPREGS
1485 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1486 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1487 sizeof(t->xfpu), &t->xfpu);
1488 t->num_notes++;
1489 sz += notesize(&t->notes[2]);
1490 }
1491 #endif
1492 return sz;
1493 }
1494
1495 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1496 elf_addr_t e_shoff, int segs)
1497 {
1498 elf->e_shoff = e_shoff;
1499 elf->e_shentsize = sizeof(*shdr4extnum);
1500 elf->e_shnum = 1;
1501 elf->e_shstrndx = SHN_UNDEF;
1502
1503 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1504
1505 shdr4extnum->sh_type = SHT_NULL;
1506 shdr4extnum->sh_size = elf->e_shnum;
1507 shdr4extnum->sh_link = elf->e_shstrndx;
1508 shdr4extnum->sh_info = segs;
1509 }
1510
1511 /*
1512 * dump the segments for an MMU process
1513 */
1514 #ifdef CONFIG_MMU
1515 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1516 unsigned long *limit, unsigned long mm_flags)
1517 {
1518 struct vm_area_struct *vma;
1519 int err = 0;
1520
1521 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1522 unsigned long addr;
1523
1524 if (!maydump(vma, mm_flags))
1525 continue;
1526
1527 for (addr = vma->vm_start; addr < vma->vm_end;
1528 addr += PAGE_SIZE) {
1529 struct page *page = get_dump_page(addr);
1530 if (page) {
1531 void *kaddr = kmap(page);
1532 *size += PAGE_SIZE;
1533 if (*size > *limit)
1534 err = -EFBIG;
1535 else if (!dump_write(file, kaddr, PAGE_SIZE))
1536 err = -EIO;
1537 kunmap(page);
1538 page_cache_release(page);
1539 } else if (!dump_seek(file, PAGE_SIZE))
1540 err = -EFBIG;
1541 if (err)
1542 goto out;
1543 }
1544 }
1545 out:
1546 return err;
1547 }
1548 #endif
1549
1550 /*
1551 * dump the segments for a NOMMU process
1552 */
1553 #ifndef CONFIG_MMU
1554 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1555 unsigned long *limit, unsigned long mm_flags)
1556 {
1557 struct vm_area_struct *vma;
1558
1559 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1560 if (!maydump(vma, mm_flags))
1561 continue;
1562
1563 if ((*size += PAGE_SIZE) > *limit)
1564 return -EFBIG;
1565
1566 if (!dump_write(file, (void *) vma->vm_start,
1567 vma->vm_end - vma->vm_start))
1568 return -EIO;
1569 }
1570
1571 return 0;
1572 }
1573 #endif
1574
1575 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1576 {
1577 struct vm_area_struct *vma;
1578 size_t size = 0;
1579
1580 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1581 if (maydump(vma, mm_flags))
1582 size += vma->vm_end - vma->vm_start;
1583 return size;
1584 }
1585
1586 /*
1587 * Actual dumper
1588 *
1589 * This is a two-pass process; first we find the offsets of the bits,
1590 * and then they are actually written out. If we run out of core limit
1591 * we just truncate.
1592 */
1593 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1594 {
1595 #define NUM_NOTES 6
1596 int has_dumped = 0;
1597 mm_segment_t fs;
1598 int segs;
1599 size_t size = 0;
1600 int i;
1601 struct vm_area_struct *vma;
1602 struct elfhdr *elf = NULL;
1603 loff_t offset = 0, dataoff, foffset;
1604 int numnote;
1605 struct memelfnote *notes = NULL;
1606 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1607 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1608 LIST_HEAD(thread_list);
1609 struct list_head *t;
1610 elf_fpregset_t *fpu = NULL;
1611 #ifdef ELF_CORE_COPY_XFPREGS
1612 elf_fpxregset_t *xfpu = NULL;
1613 #endif
1614 int thread_status_size = 0;
1615 elf_addr_t *auxv;
1616 struct elf_phdr *phdr4note = NULL;
1617 struct elf_shdr *shdr4extnum = NULL;
1618 Elf_Half e_phnum;
1619 elf_addr_t e_shoff;
1620
1621 /*
1622 * We no longer stop all VM operations.
1623 *
1624 * This is because those proceses that could possibly change map_count
1625 * or the mmap / vma pages are now blocked in do_exit on current
1626 * finishing this core dump.
1627 *
1628 * Only ptrace can touch these memory addresses, but it doesn't change
1629 * the map_count or the pages allocated. So no possibility of crashing
1630 * exists while dumping the mm->vm_next areas to the core file.
1631 */
1632
1633 /* alloc memory for large data structures: too large to be on stack */
1634 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1635 if (!elf)
1636 goto cleanup;
1637 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1638 if (!prstatus)
1639 goto cleanup;
1640 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1641 if (!psinfo)
1642 goto cleanup;
1643 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1644 if (!notes)
1645 goto cleanup;
1646 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1647 if (!fpu)
1648 goto cleanup;
1649 #ifdef ELF_CORE_COPY_XFPREGS
1650 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1651 if (!xfpu)
1652 goto cleanup;
1653 #endif
1654
1655 if (cprm->signr) {
1656 struct core_thread *ct;
1657 struct elf_thread_status *tmp;
1658
1659 for (ct = current->mm->core_state->dumper.next;
1660 ct; ct = ct->next) {
1661 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1662 if (!tmp)
1663 goto cleanup;
1664
1665 tmp->thread = ct->task;
1666 list_add(&tmp->list, &thread_list);
1667 }
1668
1669 list_for_each(t, &thread_list) {
1670 struct elf_thread_status *tmp;
1671 int sz;
1672
1673 tmp = list_entry(t, struct elf_thread_status, list);
1674 sz = elf_dump_thread_status(cprm->signr, tmp);
1675 thread_status_size += sz;
1676 }
1677 }
1678
1679 /* now collect the dump for the current */
1680 fill_prstatus(prstatus, current, cprm->signr);
1681 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1682
1683 segs = current->mm->map_count;
1684 segs += elf_core_extra_phdrs();
1685
1686 /* for notes section */
1687 segs++;
1688
1689 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1690 * this, kernel supports extended numbering. Have a look at
1691 * include/linux/elf.h for further information. */
1692 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1693
1694 /* Set up header */
1695 fill_elf_fdpic_header(elf, e_phnum);
1696
1697 has_dumped = 1;
1698 current->flags |= PF_DUMPCORE;
1699
1700 /*
1701 * Set up the notes in similar form to SVR4 core dumps made
1702 * with info from their /proc.
1703 */
1704
1705 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1706 fill_psinfo(psinfo, current->group_leader, current->mm);
1707 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1708
1709 numnote = 2;
1710
1711 auxv = (elf_addr_t *) current->mm->saved_auxv;
1712
1713 i = 0;
1714 do
1715 i += 2;
1716 while (auxv[i - 2] != AT_NULL);
1717 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1718 i * sizeof(elf_addr_t), auxv);
1719
1720 /* Try to dump the FPU. */
1721 if ((prstatus->pr_fpvalid =
1722 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1723 fill_note(notes + numnote++,
1724 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1725 #ifdef ELF_CORE_COPY_XFPREGS
1726 if (elf_core_copy_task_xfpregs(current, xfpu))
1727 fill_note(notes + numnote++,
1728 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1729 #endif
1730
1731 fs = get_fs();
1732 set_fs(KERNEL_DS);
1733
1734 offset += sizeof(*elf); /* Elf header */
1735 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1736 foffset = offset;
1737
1738 /* Write notes phdr entry */
1739 {
1740 int sz = 0;
1741
1742 for (i = 0; i < numnote; i++)
1743 sz += notesize(notes + i);
1744
1745 sz += thread_status_size;
1746
1747 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1748 if (!phdr4note)
1749 goto end_coredump;
1750
1751 fill_elf_note_phdr(phdr4note, sz, offset);
1752 offset += sz;
1753 }
1754
1755 /* Page-align dumped data */
1756 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1757
1758 offset += elf_core_vma_data_size(cprm->mm_flags);
1759 offset += elf_core_extra_data_size();
1760 e_shoff = offset;
1761
1762 if (e_phnum == PN_XNUM) {
1763 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1764 if (!shdr4extnum)
1765 goto end_coredump;
1766 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1767 }
1768
1769 offset = dataoff;
1770
1771 size += sizeof(*elf);
1772 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1773 goto end_coredump;
1774
1775 size += sizeof(*phdr4note);
1776 if (size > cprm->limit
1777 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1778 goto end_coredump;
1779
1780 /* write program headers for segments dump */
1781 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1782 struct elf_phdr phdr;
1783 size_t sz;
1784
1785 sz = vma->vm_end - vma->vm_start;
1786
1787 phdr.p_type = PT_LOAD;
1788 phdr.p_offset = offset;
1789 phdr.p_vaddr = vma->vm_start;
1790 phdr.p_paddr = 0;
1791 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1792 phdr.p_memsz = sz;
1793 offset += phdr.p_filesz;
1794 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1795 if (vma->vm_flags & VM_WRITE)
1796 phdr.p_flags |= PF_W;
1797 if (vma->vm_flags & VM_EXEC)
1798 phdr.p_flags |= PF_X;
1799 phdr.p_align = ELF_EXEC_PAGESIZE;
1800
1801 size += sizeof(phdr);
1802 if (size > cprm->limit
1803 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
1804 goto end_coredump;
1805 }
1806
1807 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
1808 goto end_coredump;
1809
1810 /* write out the notes section */
1811 for (i = 0; i < numnote; i++)
1812 if (!writenote(notes + i, cprm->file, &foffset))
1813 goto end_coredump;
1814
1815 /* write out the thread status notes section */
1816 list_for_each(t, &thread_list) {
1817 struct elf_thread_status *tmp =
1818 list_entry(t, struct elf_thread_status, list);
1819
1820 for (i = 0; i < tmp->num_notes; i++)
1821 if (!writenote(&tmp->notes[i], cprm->file, &foffset))
1822 goto end_coredump;
1823 }
1824
1825 if (!dump_seek(cprm->file, dataoff - foffset))
1826 goto end_coredump;
1827
1828 if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit,
1829 cprm->mm_flags) < 0)
1830 goto end_coredump;
1831
1832 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
1833 goto end_coredump;
1834
1835 if (e_phnum == PN_XNUM) {
1836 size += sizeof(*shdr4extnum);
1837 if (size > cprm->limit
1838 || !dump_write(cprm->file, shdr4extnum,
1839 sizeof(*shdr4extnum)))
1840 goto end_coredump;
1841 }
1842
1843 if (cprm->file->f_pos != offset) {
1844 /* Sanity check */
1845 printk(KERN_WARNING
1846 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1847 cprm->file->f_pos, offset);
1848 }
1849
1850 end_coredump:
1851 set_fs(fs);
1852
1853 cleanup:
1854 while (!list_empty(&thread_list)) {
1855 struct list_head *tmp = thread_list.next;
1856 list_del(tmp);
1857 kfree(list_entry(tmp, struct elf_thread_status, list));
1858 }
1859 kfree(phdr4note);
1860 kfree(elf);
1861 kfree(prstatus);
1862 kfree(psinfo);
1863 kfree(notes);
1864 kfree(fpu);
1865 kfree(shdr4extnum);
1866 #ifdef ELF_CORE_COPY_XFPREGS
1867 kfree(xfpu);
1868 #endif
1869 return has_dumped;
1870 #undef NUM_NOTES
1871 }
1872
1873 #endif /* CONFIG_ELF_CORE */
Linux with added FPC-III support