Merge tag 'gfs2-merge-window' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2...
[deliverable/linux.git] / fs / binfmt_elf_fdpic.c
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2 *
3 * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 * Derived from binfmt_elf.c
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13 #include <linux/module.h>
14
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37 #include <linux/coredump.h>
38
39 #include <asm/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/pgalloc.h>
42
43 typedef char *elf_caddr_t;
44
45 #if 0
46 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
47 #else
48 #define kdebug(fmt, ...) do {} while(0)
49 #endif
50
51 #if 0
52 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
53 #else
54 #define kdcore(fmt, ...) do {} while(0)
55 #endif
56
57 MODULE_LICENSE("GPL");
58
59 static int load_elf_fdpic_binary(struct linux_binprm *);
60 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
61 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
62 struct mm_struct *, const char *);
63
64 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
65 struct elf_fdpic_params *,
66 struct elf_fdpic_params *);
67
68 #ifndef CONFIG_MMU
69 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
70 unsigned long *);
71 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
72 struct file *,
73 struct mm_struct *);
74 #endif
75
76 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
77 struct file *, struct mm_struct *);
78
79 #ifdef CONFIG_ELF_CORE
80 static int elf_fdpic_core_dump(struct coredump_params *cprm);
81 #endif
82
83 static struct linux_binfmt elf_fdpic_format = {
84 .module = THIS_MODULE,
85 .load_binary = load_elf_fdpic_binary,
86 #ifdef CONFIG_ELF_CORE
87 .core_dump = elf_fdpic_core_dump,
88 #endif
89 .min_coredump = ELF_EXEC_PAGESIZE,
90 };
91
92 static int __init init_elf_fdpic_binfmt(void)
93 {
94 register_binfmt(&elf_fdpic_format);
95 return 0;
96 }
97
98 static void __exit exit_elf_fdpic_binfmt(void)
99 {
100 unregister_binfmt(&elf_fdpic_format);
101 }
102
103 core_initcall(init_elf_fdpic_binfmt);
104 module_exit(exit_elf_fdpic_binfmt);
105
106 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
107 {
108 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
109 return 0;
110 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
111 return 0;
112 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
113 return 0;
114 if (!file->f_op->mmap)
115 return 0;
116 return 1;
117 }
118
119 /*****************************************************************************/
120 /*
121 * read the program headers table into memory
122 */
123 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
124 struct file *file)
125 {
126 struct elf32_phdr *phdr;
127 unsigned long size;
128 int retval, loop;
129
130 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
131 return -ENOMEM;
132 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
133 return -ENOMEM;
134
135 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
136 params->phdrs = kmalloc(size, GFP_KERNEL);
137 if (!params->phdrs)
138 return -ENOMEM;
139
140 retval = kernel_read(file, params->hdr.e_phoff,
141 (char *) params->phdrs, size);
142 if (unlikely(retval != size))
143 return retval < 0 ? retval : -ENOEXEC;
144
145 /* determine stack size for this binary */
146 phdr = params->phdrs;
147 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
148 if (phdr->p_type != PT_GNU_STACK)
149 continue;
150
151 if (phdr->p_flags & PF_X)
152 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
153 else
154 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
155
156 params->stack_size = phdr->p_memsz;
157 break;
158 }
159
160 return 0;
161 }
162
163 /*****************************************************************************/
164 /*
165 * load an fdpic binary into various bits of memory
166 */
167 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
168 {
169 struct elf_fdpic_params exec_params, interp_params;
170 struct pt_regs *regs = current_pt_regs();
171 struct elf_phdr *phdr;
172 unsigned long stack_size, entryaddr;
173 #ifdef ELF_FDPIC_PLAT_INIT
174 unsigned long dynaddr;
175 #endif
176 #ifndef CONFIG_MMU
177 unsigned long stack_prot;
178 #endif
179 struct file *interpreter = NULL; /* to shut gcc up */
180 char *interpreter_name = NULL;
181 int executable_stack;
182 int retval, i;
183
184 kdebug("____ LOAD %d ____", current->pid);
185
186 memset(&exec_params, 0, sizeof(exec_params));
187 memset(&interp_params, 0, sizeof(interp_params));
188
189 exec_params.hdr = *(struct elfhdr *) bprm->buf;
190 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
191
192 /* check that this is a binary we know how to deal with */
193 retval = -ENOEXEC;
194 if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
195 goto error;
196
197 /* read the program header table */
198 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
199 if (retval < 0)
200 goto error;
201
202 /* scan for a program header that specifies an interpreter */
203 phdr = exec_params.phdrs;
204
205 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
206 switch (phdr->p_type) {
207 case PT_INTERP:
208 retval = -ENOMEM;
209 if (phdr->p_filesz > PATH_MAX)
210 goto error;
211 retval = -ENOENT;
212 if (phdr->p_filesz < 2)
213 goto error;
214
215 /* read the name of the interpreter into memory */
216 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
217 if (!interpreter_name)
218 goto error;
219
220 retval = kernel_read(bprm->file,
221 phdr->p_offset,
222 interpreter_name,
223 phdr->p_filesz);
224 if (unlikely(retval != phdr->p_filesz)) {
225 if (retval >= 0)
226 retval = -ENOEXEC;
227 goto error;
228 }
229
230 retval = -ENOENT;
231 if (interpreter_name[phdr->p_filesz - 1] != '\0')
232 goto error;
233
234 kdebug("Using ELF interpreter %s", interpreter_name);
235
236 /* replace the program with the interpreter */
237 interpreter = open_exec(interpreter_name);
238 retval = PTR_ERR(interpreter);
239 if (IS_ERR(interpreter)) {
240 interpreter = NULL;
241 goto error;
242 }
243
244 /*
245 * If the binary is not readable then enforce
246 * mm->dumpable = 0 regardless of the interpreter's
247 * permissions.
248 */
249 would_dump(bprm, interpreter);
250
251 retval = kernel_read(interpreter, 0, bprm->buf,
252 BINPRM_BUF_SIZE);
253 if (unlikely(retval != BINPRM_BUF_SIZE)) {
254 if (retval >= 0)
255 retval = -ENOEXEC;
256 goto error;
257 }
258
259 interp_params.hdr = *((struct elfhdr *) bprm->buf);
260 break;
261
262 case PT_LOAD:
263 #ifdef CONFIG_MMU
264 if (exec_params.load_addr == 0)
265 exec_params.load_addr = phdr->p_vaddr;
266 #endif
267 break;
268 }
269
270 }
271
272 if (elf_check_const_displacement(&exec_params.hdr))
273 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
274
275 /* perform insanity checks on the interpreter */
276 if (interpreter_name) {
277 retval = -ELIBBAD;
278 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
279 goto error;
280
281 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
282
283 /* read the interpreter's program header table */
284 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
285 if (retval < 0)
286 goto error;
287 }
288
289 stack_size = exec_params.stack_size;
290 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
291 executable_stack = EXSTACK_ENABLE_X;
292 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
293 executable_stack = EXSTACK_DISABLE_X;
294 else
295 executable_stack = EXSTACK_DEFAULT;
296
297 if (stack_size == 0) {
298 stack_size = interp_params.stack_size;
299 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
300 executable_stack = EXSTACK_ENABLE_X;
301 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
302 executable_stack = EXSTACK_DISABLE_X;
303 else
304 executable_stack = EXSTACK_DEFAULT;
305 }
306
307 retval = -ENOEXEC;
308 if (stack_size == 0)
309 goto error;
310
311 if (elf_check_const_displacement(&interp_params.hdr))
312 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
313
314 /* flush all traces of the currently running executable */
315 retval = flush_old_exec(bprm);
316 if (retval)
317 goto error;
318
319 /* there's now no turning back... the old userspace image is dead,
320 * defunct, deceased, etc.
321 */
322 set_personality(PER_LINUX_FDPIC);
323 if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
324 current->personality |= READ_IMPLIES_EXEC;
325
326 setup_new_exec(bprm);
327
328 set_binfmt(&elf_fdpic_format);
329
330 current->mm->start_code = 0;
331 current->mm->end_code = 0;
332 current->mm->start_stack = 0;
333 current->mm->start_data = 0;
334 current->mm->end_data = 0;
335 current->mm->context.exec_fdpic_loadmap = 0;
336 current->mm->context.interp_fdpic_loadmap = 0;
337
338 #ifdef CONFIG_MMU
339 elf_fdpic_arch_lay_out_mm(&exec_params,
340 &interp_params,
341 &current->mm->start_stack,
342 &current->mm->start_brk);
343
344 retval = setup_arg_pages(bprm, current->mm->start_stack,
345 executable_stack);
346 if (retval < 0)
347 goto error;
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;
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;
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 area and zero-size brk area */
378 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
379 if (stack_size < PAGE_SIZE * 2)
380 stack_size = PAGE_SIZE * 2;
381
382 stack_prot = PROT_READ | PROT_WRITE;
383 if (executable_stack == EXSTACK_ENABLE_X ||
384 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
385 stack_prot |= PROT_EXEC;
386
387 current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
388 MAP_PRIVATE | MAP_ANONYMOUS |
389 MAP_UNINITIALIZED | MAP_GROWSDOWN,
390 0);
391
392 if (IS_ERR_VALUE(current->mm->start_brk)) {
393 retval = current->mm->start_brk;
394 current->mm->start_brk = 0;
395 goto error;
396 }
397
398 current->mm->brk = current->mm->start_brk;
399 current->mm->context.end_brk = current->mm->start_brk;
400 current->mm->start_stack = current->mm->start_brk + stack_size;
401 #endif
402
403 install_exec_creds(bprm);
404 if (create_elf_fdpic_tables(bprm, current->mm,
405 &exec_params, &interp_params) < 0)
406 goto error;
407
408 kdebug("- start_code %lx", current->mm->start_code);
409 kdebug("- end_code %lx", current->mm->end_code);
410 kdebug("- start_data %lx", current->mm->start_data);
411 kdebug("- end_data %lx", current->mm->end_data);
412 kdebug("- start_brk %lx", current->mm->start_brk);
413 kdebug("- brk %lx", current->mm->brk);
414 kdebug("- start_stack %lx", current->mm->start_stack);
415
416 #ifdef ELF_FDPIC_PLAT_INIT
417 /*
418 * The ABI may specify that certain registers be set up in special
419 * ways (on i386 %edx is the address of a DT_FINI function, for
420 * example. This macro performs whatever initialization to
421 * the regs structure is required.
422 */
423 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
424 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
425 dynaddr);
426 #endif
427
428 /* everything is now ready... get the userspace context ready to roll */
429 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
430 start_thread(regs, entryaddr, current->mm->start_stack);
431
432 retval = 0;
433
434 error:
435 if (interpreter) {
436 allow_write_access(interpreter);
437 fput(interpreter);
438 }
439 kfree(interpreter_name);
440 kfree(exec_params.phdrs);
441 kfree(exec_params.loadmap);
442 kfree(interp_params.phdrs);
443 kfree(interp_params.loadmap);
444 return retval;
445 }
446
447 /*****************************************************************************/
448
449 #ifndef ELF_BASE_PLATFORM
450 /*
451 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
452 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
453 * will be copied to the user stack in the same manner as AT_PLATFORM.
454 */
455 #define ELF_BASE_PLATFORM NULL
456 #endif
457
458 /*
459 * present useful information to the program by shovelling it onto the new
460 * process's stack
461 */
462 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
463 struct mm_struct *mm,
464 struct elf_fdpic_params *exec_params,
465 struct elf_fdpic_params *interp_params)
466 {
467 const struct cred *cred = current_cred();
468 unsigned long sp, csp, nitems;
469 elf_caddr_t __user *argv, *envp;
470 size_t platform_len = 0, len;
471 char *k_platform, *k_base_platform;
472 char __user *u_platform, *u_base_platform, *p;
473 int loop;
474 int nr; /* reset for each csp adjustment */
475
476 #ifdef CONFIG_MMU
477 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
478 * by the processes running on the same package. One thing we can do is
479 * to shuffle the initial stack for them, so we give the architecture
480 * an opportunity to do so here.
481 */
482 sp = arch_align_stack(bprm->p);
483 #else
484 sp = mm->start_stack;
485
486 /* stack the program arguments and environment */
487 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
488 return -EFAULT;
489 #endif
490
491 /*
492 * If this architecture has a platform capability string, copy it
493 * to userspace. In some cases (Sparc), this info is impossible
494 * for userspace to get any other way, in others (i386) it is
495 * merely difficult.
496 */
497 k_platform = ELF_PLATFORM;
498 u_platform = NULL;
499
500 if (k_platform) {
501 platform_len = strlen(k_platform) + 1;
502 sp -= platform_len;
503 u_platform = (char __user *) sp;
504 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
505 return -EFAULT;
506 }
507
508 /*
509 * If this architecture has a "base" platform capability
510 * string, copy it to userspace.
511 */
512 k_base_platform = ELF_BASE_PLATFORM;
513 u_base_platform = NULL;
514
515 if (k_base_platform) {
516 platform_len = strlen(k_base_platform) + 1;
517 sp -= platform_len;
518 u_base_platform = (char __user *) sp;
519 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
520 return -EFAULT;
521 }
522
523 sp &= ~7UL;
524
525 /* stack the load map(s) */
526 len = sizeof(struct elf32_fdpic_loadmap);
527 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
528 sp = (sp - len) & ~7UL;
529 exec_params->map_addr = sp;
530
531 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
532 return -EFAULT;
533
534 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
535
536 if (interp_params->loadmap) {
537 len = sizeof(struct elf32_fdpic_loadmap);
538 len += sizeof(struct elf32_fdpic_loadseg) *
539 interp_params->loadmap->nsegs;
540 sp = (sp - len) & ~7UL;
541 interp_params->map_addr = sp;
542
543 if (copy_to_user((void __user *) sp, interp_params->loadmap,
544 len) != 0)
545 return -EFAULT;
546
547 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
548 }
549
550 /* force 16 byte _final_ alignment here for generality */
551 #define DLINFO_ITEMS 15
552
553 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
554 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
555
556 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
557 nitems++;
558
559 csp = sp;
560 sp -= nitems * 2 * sizeof(unsigned long);
561 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
562 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
563 sp -= 1 * sizeof(unsigned long); /* argc */
564
565 csp -= sp & 15UL;
566 sp -= sp & 15UL;
567
568 /* put the ELF interpreter info on the stack */
569 #define NEW_AUX_ENT(id, val) \
570 do { \
571 struct { unsigned long _id, _val; } __user *ent; \
572 \
573 ent = (void __user *) csp; \
574 __put_user((id), &ent[nr]._id); \
575 __put_user((val), &ent[nr]._val); \
576 nr++; \
577 } while (0)
578
579 nr = 0;
580 csp -= 2 * sizeof(unsigned long);
581 NEW_AUX_ENT(AT_NULL, 0);
582 if (k_platform) {
583 nr = 0;
584 csp -= 2 * sizeof(unsigned long);
585 NEW_AUX_ENT(AT_PLATFORM,
586 (elf_addr_t) (unsigned long) u_platform);
587 }
588
589 if (k_base_platform) {
590 nr = 0;
591 csp -= 2 * sizeof(unsigned long);
592 NEW_AUX_ENT(AT_BASE_PLATFORM,
593 (elf_addr_t) (unsigned long) u_base_platform);
594 }
595
596 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
597 nr = 0;
598 csp -= 2 * sizeof(unsigned long);
599 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
600 }
601
602 nr = 0;
603 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
604 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
605 #ifdef ELF_HWCAP2
606 NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
607 #endif
608 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
609 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
610 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
611 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
612 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
613 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
614 NEW_AUX_ENT(AT_FLAGS, 0);
615 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
616 NEW_AUX_ENT(AT_UID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
617 NEW_AUX_ENT(AT_EUID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
618 NEW_AUX_ENT(AT_GID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
619 NEW_AUX_ENT(AT_EGID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
620 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
621 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
622
623 #ifdef ARCH_DLINFO
624 nr = 0;
625 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
626
627 /* ARCH_DLINFO must come last so platform specific code can enforce
628 * special alignment requirements on the AUXV if necessary (eg. PPC).
629 */
630 ARCH_DLINFO;
631 #endif
632 #undef NEW_AUX_ENT
633
634 /* allocate room for argv[] and envv[] */
635 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
636 envp = (elf_caddr_t __user *) csp;
637 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
638 argv = (elf_caddr_t __user *) csp;
639
640 /* stack argc */
641 csp -= sizeof(unsigned long);
642 __put_user(bprm->argc, (unsigned long __user *) csp);
643
644 BUG_ON(csp != sp);
645
646 /* fill in the argv[] array */
647 #ifdef CONFIG_MMU
648 current->mm->arg_start = bprm->p;
649 #else
650 current->mm->arg_start = current->mm->start_stack -
651 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
652 #endif
653
654 p = (char __user *) current->mm->arg_start;
655 for (loop = bprm->argc; loop > 0; loop--) {
656 __put_user((elf_caddr_t) p, argv++);
657 len = strnlen_user(p, MAX_ARG_STRLEN);
658 if (!len || len > MAX_ARG_STRLEN)
659 return -EINVAL;
660 p += len;
661 }
662 __put_user(NULL, argv);
663 current->mm->arg_end = (unsigned long) p;
664
665 /* fill in the envv[] array */
666 current->mm->env_start = (unsigned long) p;
667 for (loop = bprm->envc; loop > 0; loop--) {
668 __put_user((elf_caddr_t)(unsigned long) p, envp++);
669 len = strnlen_user(p, MAX_ARG_STRLEN);
670 if (!len || len > MAX_ARG_STRLEN)
671 return -EINVAL;
672 p += len;
673 }
674 __put_user(NULL, envp);
675 current->mm->env_end = (unsigned long) p;
676
677 mm->start_stack = (unsigned long) sp;
678 return 0;
679 }
680
681 /*****************************************************************************/
682 /*
683 * transfer the program arguments and environment from the holding pages onto
684 * the stack
685 */
686 #ifndef CONFIG_MMU
687 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
688 unsigned long *_sp)
689 {
690 unsigned long index, stop, sp;
691 char *src;
692 int ret = 0;
693
694 stop = bprm->p >> PAGE_SHIFT;
695 sp = *_sp;
696
697 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
698 src = kmap(bprm->page[index]);
699 sp -= PAGE_SIZE;
700 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
701 ret = -EFAULT;
702 kunmap(bprm->page[index]);
703 if (ret < 0)
704 goto out;
705 }
706
707 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
708
709 out:
710 return ret;
711 }
712 #endif
713
714 /*****************************************************************************/
715 /*
716 * load the appropriate binary image (executable or interpreter) into memory
717 * - we assume no MMU is available
718 * - if no other PIC bits are set in params->hdr->e_flags
719 * - we assume that the LOADable segments in the binary are independently relocatable
720 * - we assume R/O executable segments are shareable
721 * - else
722 * - we assume the loadable parts of the image to require fixed displacement
723 * - the image is not shareable
724 */
725 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
726 struct file *file,
727 struct mm_struct *mm,
728 const char *what)
729 {
730 struct elf32_fdpic_loadmap *loadmap;
731 #ifdef CONFIG_MMU
732 struct elf32_fdpic_loadseg *mseg;
733 #endif
734 struct elf32_fdpic_loadseg *seg;
735 struct elf32_phdr *phdr;
736 unsigned long load_addr, stop;
737 unsigned nloads, tmp;
738 size_t size;
739 int loop, ret;
740
741 /* allocate a load map table */
742 nloads = 0;
743 for (loop = 0; loop < params->hdr.e_phnum; loop++)
744 if (params->phdrs[loop].p_type == PT_LOAD)
745 nloads++;
746
747 if (nloads == 0)
748 return -ELIBBAD;
749
750 size = sizeof(*loadmap) + nloads * sizeof(*seg);
751 loadmap = kzalloc(size, GFP_KERNEL);
752 if (!loadmap)
753 return -ENOMEM;
754
755 params->loadmap = loadmap;
756
757 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
758 loadmap->nsegs = nloads;
759
760 load_addr = params->load_addr;
761 seg = loadmap->segs;
762
763 /* map the requested LOADs into the memory space */
764 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
765 case ELF_FDPIC_FLAG_CONSTDISP:
766 case ELF_FDPIC_FLAG_CONTIGUOUS:
767 #ifndef CONFIG_MMU
768 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
769 if (ret < 0)
770 return ret;
771 break;
772 #endif
773 default:
774 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
775 if (ret < 0)
776 return ret;
777 break;
778 }
779
780 /* map the entry point */
781 if (params->hdr.e_entry) {
782 seg = loadmap->segs;
783 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
784 if (params->hdr.e_entry >= seg->p_vaddr &&
785 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
786 params->entry_addr =
787 (params->hdr.e_entry - seg->p_vaddr) +
788 seg->addr;
789 break;
790 }
791 }
792 }
793
794 /* determine where the program header table has wound up if mapped */
795 stop = params->hdr.e_phoff;
796 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
797 phdr = params->phdrs;
798
799 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
800 if (phdr->p_type != PT_LOAD)
801 continue;
802
803 if (phdr->p_offset > params->hdr.e_phoff ||
804 phdr->p_offset + phdr->p_filesz < stop)
805 continue;
806
807 seg = loadmap->segs;
808 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
809 if (phdr->p_vaddr >= seg->p_vaddr &&
810 phdr->p_vaddr + phdr->p_filesz <=
811 seg->p_vaddr + seg->p_memsz) {
812 params->ph_addr =
813 (phdr->p_vaddr - seg->p_vaddr) +
814 seg->addr +
815 params->hdr.e_phoff - phdr->p_offset;
816 break;
817 }
818 }
819 break;
820 }
821
822 /* determine where the dynamic section has wound up if there is one */
823 phdr = params->phdrs;
824 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
825 if (phdr->p_type != PT_DYNAMIC)
826 continue;
827
828 seg = loadmap->segs;
829 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
830 if (phdr->p_vaddr >= seg->p_vaddr &&
831 phdr->p_vaddr + phdr->p_memsz <=
832 seg->p_vaddr + seg->p_memsz) {
833 params->dynamic_addr =
834 (phdr->p_vaddr - seg->p_vaddr) +
835 seg->addr;
836
837 /* check the dynamic section contains at least
838 * one item, and that the last item is a NULL
839 * entry */
840 if (phdr->p_memsz == 0 ||
841 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
842 goto dynamic_error;
843
844 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
845 if (((Elf32_Dyn *)
846 params->dynamic_addr)[tmp - 1].d_tag != 0)
847 goto dynamic_error;
848 break;
849 }
850 }
851 break;
852 }
853
854 /* now elide adjacent segments in the load map on MMU linux
855 * - on uClinux the holes between may actually be filled with system
856 * stuff or stuff from other processes
857 */
858 #ifdef CONFIG_MMU
859 nloads = loadmap->nsegs;
860 mseg = loadmap->segs;
861 seg = mseg + 1;
862 for (loop = 1; loop < nloads; loop++) {
863 /* see if we have a candidate for merging */
864 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
865 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
866 if (load_addr == (seg->addr & PAGE_MASK)) {
867 mseg->p_memsz +=
868 load_addr -
869 (mseg->addr + mseg->p_memsz);
870 mseg->p_memsz += seg->addr & ~PAGE_MASK;
871 mseg->p_memsz += seg->p_memsz;
872 loadmap->nsegs--;
873 continue;
874 }
875 }
876
877 mseg++;
878 if (mseg != seg)
879 *mseg = *seg;
880 }
881 #endif
882
883 kdebug("Mapped Object [%s]:", what);
884 kdebug("- elfhdr : %lx", params->elfhdr_addr);
885 kdebug("- entry : %lx", params->entry_addr);
886 kdebug("- PHDR[] : %lx", params->ph_addr);
887 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
888 seg = loadmap->segs;
889 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
890 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
891 loop,
892 seg->addr, seg->addr + seg->p_memsz - 1,
893 seg->p_vaddr, seg->p_memsz);
894
895 return 0;
896
897 dynamic_error:
898 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
899 what, file_inode(file)->i_ino);
900 return -ELIBBAD;
901 }
902
903 /*****************************************************************************/
904 /*
905 * map a file with constant displacement under uClinux
906 */
907 #ifndef CONFIG_MMU
908 static int elf_fdpic_map_file_constdisp_on_uclinux(
909 struct elf_fdpic_params *params,
910 struct file *file,
911 struct mm_struct *mm)
912 {
913 struct elf32_fdpic_loadseg *seg;
914 struct elf32_phdr *phdr;
915 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
916 int loop, ret;
917
918 load_addr = params->load_addr;
919 seg = params->loadmap->segs;
920
921 /* determine the bounds of the contiguous overall allocation we must
922 * make */
923 phdr = params->phdrs;
924 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
925 if (params->phdrs[loop].p_type != PT_LOAD)
926 continue;
927
928 if (base > phdr->p_vaddr)
929 base = phdr->p_vaddr;
930 if (top < phdr->p_vaddr + phdr->p_memsz)
931 top = phdr->p_vaddr + phdr->p_memsz;
932 }
933
934 /* allocate one big anon block for everything */
935 mflags = MAP_PRIVATE;
936 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
937 mflags |= MAP_EXECUTABLE;
938
939 maddr = vm_mmap(NULL, load_addr, top - base,
940 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
941 if (IS_ERR_VALUE(maddr))
942 return (int) maddr;
943
944 if (load_addr != 0)
945 load_addr += PAGE_ALIGN(top - base);
946
947 /* and then load the file segments into it */
948 phdr = params->phdrs;
949 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
950 if (params->phdrs[loop].p_type != PT_LOAD)
951 continue;
952
953 seg->addr = maddr + (phdr->p_vaddr - base);
954 seg->p_vaddr = phdr->p_vaddr;
955 seg->p_memsz = phdr->p_memsz;
956
957 ret = read_code(file, seg->addr, phdr->p_offset,
958 phdr->p_filesz);
959 if (ret < 0)
960 return ret;
961
962 /* map the ELF header address if in this segment */
963 if (phdr->p_offset == 0)
964 params->elfhdr_addr = seg->addr;
965
966 /* clear any space allocated but not loaded */
967 if (phdr->p_filesz < phdr->p_memsz) {
968 if (clear_user((void *) (seg->addr + phdr->p_filesz),
969 phdr->p_memsz - phdr->p_filesz))
970 return -EFAULT;
971 }
972
973 if (mm) {
974 if (phdr->p_flags & PF_X) {
975 if (!mm->start_code) {
976 mm->start_code = seg->addr;
977 mm->end_code = seg->addr +
978 phdr->p_memsz;
979 }
980 } else if (!mm->start_data) {
981 mm->start_data = seg->addr;
982 mm->end_data = seg->addr + phdr->p_memsz;
983 }
984 }
985
986 seg++;
987 }
988
989 return 0;
990 }
991 #endif
992
993 /*****************************************************************************/
994 /*
995 * map a binary by direct mmap() of the individual PT_LOAD segments
996 */
997 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
998 struct file *file,
999 struct mm_struct *mm)
1000 {
1001 struct elf32_fdpic_loadseg *seg;
1002 struct elf32_phdr *phdr;
1003 unsigned long load_addr, delta_vaddr;
1004 int loop, dvset;
1005
1006 load_addr = params->load_addr;
1007 delta_vaddr = 0;
1008 dvset = 0;
1009
1010 seg = params->loadmap->segs;
1011
1012 /* deal with each load segment separately */
1013 phdr = params->phdrs;
1014 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1015 unsigned long maddr, disp, excess, excess1;
1016 int prot = 0, flags;
1017
1018 if (phdr->p_type != PT_LOAD)
1019 continue;
1020
1021 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1022 (unsigned long) phdr->p_vaddr,
1023 (unsigned long) phdr->p_offset,
1024 (unsigned long) phdr->p_filesz,
1025 (unsigned long) phdr->p_memsz);
1026
1027 /* determine the mapping parameters */
1028 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1029 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1030 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1031
1032 flags = MAP_PRIVATE | MAP_DENYWRITE;
1033 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1034 flags |= MAP_EXECUTABLE;
1035
1036 maddr = 0;
1037
1038 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1039 case ELF_FDPIC_FLAG_INDEPENDENT:
1040 /* PT_LOADs are independently locatable */
1041 break;
1042
1043 case ELF_FDPIC_FLAG_HONOURVADDR:
1044 /* the specified virtual address must be honoured */
1045 maddr = phdr->p_vaddr;
1046 flags |= MAP_FIXED;
1047 break;
1048
1049 case ELF_FDPIC_FLAG_CONSTDISP:
1050 /* constant displacement
1051 * - can be mapped anywhere, but must be mapped as a
1052 * unit
1053 */
1054 if (!dvset) {
1055 maddr = load_addr;
1056 delta_vaddr = phdr->p_vaddr;
1057 dvset = 1;
1058 } else {
1059 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1060 flags |= MAP_FIXED;
1061 }
1062 break;
1063
1064 case ELF_FDPIC_FLAG_CONTIGUOUS:
1065 /* contiguity handled later */
1066 break;
1067
1068 default:
1069 BUG();
1070 }
1071
1072 maddr &= PAGE_MASK;
1073
1074 /* create the mapping */
1075 disp = phdr->p_vaddr & ~PAGE_MASK;
1076 maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1077 phdr->p_offset - disp);
1078
1079 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1080 loop, phdr->p_memsz + disp, prot, flags,
1081 phdr->p_offset - disp, maddr);
1082
1083 if (IS_ERR_VALUE(maddr))
1084 return (int) maddr;
1085
1086 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1087 ELF_FDPIC_FLAG_CONTIGUOUS)
1088 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1089
1090 seg->addr = maddr + disp;
1091 seg->p_vaddr = phdr->p_vaddr;
1092 seg->p_memsz = phdr->p_memsz;
1093
1094 /* map the ELF header address if in this segment */
1095 if (phdr->p_offset == 0)
1096 params->elfhdr_addr = seg->addr;
1097
1098 /* clear the bit between beginning of mapping and beginning of
1099 * PT_LOAD */
1100 if (prot & PROT_WRITE && disp > 0) {
1101 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1102 if (clear_user((void __user *) maddr, disp))
1103 return -EFAULT;
1104 maddr += disp;
1105 }
1106
1107 /* clear any space allocated but not loaded
1108 * - on uClinux we can just clear the lot
1109 * - on MMU linux we'll get a SIGBUS beyond the last page
1110 * extant in the file
1111 */
1112 excess = phdr->p_memsz - phdr->p_filesz;
1113 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1114
1115 #ifdef CONFIG_MMU
1116 if (excess > excess1) {
1117 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1118 unsigned long xmaddr;
1119
1120 flags |= MAP_FIXED | MAP_ANONYMOUS;
1121 xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1122 prot, flags, 0);
1123
1124 kdebug("mmap[%d] <anon>"
1125 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1126 loop, xaddr, excess - excess1, prot, flags,
1127 xmaddr);
1128
1129 if (xmaddr != xaddr)
1130 return -ENOMEM;
1131 }
1132
1133 if (prot & PROT_WRITE && excess1 > 0) {
1134 kdebug("clear[%d] ad=%lx sz=%lx",
1135 loop, maddr + phdr->p_filesz, excess1);
1136 if (clear_user((void __user *) maddr + phdr->p_filesz,
1137 excess1))
1138 return -EFAULT;
1139 }
1140
1141 #else
1142 if (excess > 0) {
1143 kdebug("clear[%d] ad=%lx sz=%lx",
1144 loop, maddr + phdr->p_filesz, excess);
1145 if (clear_user((void *) maddr + phdr->p_filesz, excess))
1146 return -EFAULT;
1147 }
1148 #endif
1149
1150 if (mm) {
1151 if (phdr->p_flags & PF_X) {
1152 if (!mm->start_code) {
1153 mm->start_code = maddr;
1154 mm->end_code = maddr + phdr->p_memsz;
1155 }
1156 } else if (!mm->start_data) {
1157 mm->start_data = maddr;
1158 mm->end_data = maddr + phdr->p_memsz;
1159 }
1160 }
1161
1162 seg++;
1163 }
1164
1165 return 0;
1166 }
1167
1168 /*****************************************************************************/
1169 /*
1170 * ELF-FDPIC core dumper
1171 *
1172 * Modelled on fs/exec.c:aout_core_dump()
1173 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1174 *
1175 * Modelled on fs/binfmt_elf.c core dumper
1176 */
1177 #ifdef CONFIG_ELF_CORE
1178
1179 /*
1180 * Decide whether a segment is worth dumping; default is yes to be
1181 * sure (missing info is worse than too much; etc).
1182 * Personally I'd include everything, and use the coredump limit...
1183 *
1184 * I think we should skip something. But I am not sure how. H.J.
1185 */
1186 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1187 {
1188 int dump_ok;
1189
1190 /* Do not dump I/O mapped devices or special mappings */
1191 if (vma->vm_flags & VM_IO) {
1192 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1193 return 0;
1194 }
1195
1196 /* If we may not read the contents, don't allow us to dump
1197 * them either. "dump_write()" can't handle it anyway.
1198 */
1199 if (!(vma->vm_flags & VM_READ)) {
1200 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1201 return 0;
1202 }
1203
1204 /* By default, dump shared memory if mapped from an anonymous file. */
1205 if (vma->vm_flags & VM_SHARED) {
1206 if (file_inode(vma->vm_file)->i_nlink == 0) {
1207 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1208 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1209 vma->vm_flags, dump_ok ? "yes" : "no");
1210 return dump_ok;
1211 }
1212
1213 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1214 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1215 vma->vm_flags, dump_ok ? "yes" : "no");
1216 return dump_ok;
1217 }
1218
1219 #ifdef CONFIG_MMU
1220 /* By default, if it hasn't been written to, don't write it out */
1221 if (!vma->anon_vma) {
1222 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1223 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1224 vma->vm_flags, dump_ok ? "yes" : "no");
1225 return dump_ok;
1226 }
1227 #endif
1228
1229 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1230 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1231 dump_ok ? "yes" : "no");
1232 return dump_ok;
1233 }
1234
1235 /* An ELF note in memory */
1236 struct memelfnote
1237 {
1238 const char *name;
1239 int type;
1240 unsigned int datasz;
1241 void *data;
1242 };
1243
1244 static int notesize(struct memelfnote *en)
1245 {
1246 int sz;
1247
1248 sz = sizeof(struct elf_note);
1249 sz += roundup(strlen(en->name) + 1, 4);
1250 sz += roundup(en->datasz, 4);
1251
1252 return sz;
1253 }
1254
1255 /* #define DEBUG */
1256
1257 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1258 {
1259 struct elf_note en;
1260 en.n_namesz = strlen(men->name) + 1;
1261 en.n_descsz = men->datasz;
1262 en.n_type = men->type;
1263
1264 return dump_emit(cprm, &en, sizeof(en)) &&
1265 dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1266 dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1267 }
1268
1269 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1270 {
1271 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1272 elf->e_ident[EI_CLASS] = ELF_CLASS;
1273 elf->e_ident[EI_DATA] = ELF_DATA;
1274 elf->e_ident[EI_VERSION] = EV_CURRENT;
1275 elf->e_ident[EI_OSABI] = ELF_OSABI;
1276 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1277
1278 elf->e_type = ET_CORE;
1279 elf->e_machine = ELF_ARCH;
1280 elf->e_version = EV_CURRENT;
1281 elf->e_entry = 0;
1282 elf->e_phoff = sizeof(struct elfhdr);
1283 elf->e_shoff = 0;
1284 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1285 elf->e_ehsize = sizeof(struct elfhdr);
1286 elf->e_phentsize = sizeof(struct elf_phdr);
1287 elf->e_phnum = segs;
1288 elf->e_shentsize = 0;
1289 elf->e_shnum = 0;
1290 elf->e_shstrndx = 0;
1291 return;
1292 }
1293
1294 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1295 {
1296 phdr->p_type = PT_NOTE;
1297 phdr->p_offset = offset;
1298 phdr->p_vaddr = 0;
1299 phdr->p_paddr = 0;
1300 phdr->p_filesz = sz;
1301 phdr->p_memsz = 0;
1302 phdr->p_flags = 0;
1303 phdr->p_align = 0;
1304 return;
1305 }
1306
1307 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1308 unsigned int sz, void *data)
1309 {
1310 note->name = name;
1311 note->type = type;
1312 note->datasz = sz;
1313 note->data = data;
1314 return;
1315 }
1316
1317 /*
1318 * fill up all the fields in prstatus from the given task struct, except
1319 * registers which need to be filled up separately.
1320 */
1321 static void fill_prstatus(struct elf_prstatus *prstatus,
1322 struct task_struct *p, long signr)
1323 {
1324 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1325 prstatus->pr_sigpend = p->pending.signal.sig[0];
1326 prstatus->pr_sighold = p->blocked.sig[0];
1327 rcu_read_lock();
1328 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1329 rcu_read_unlock();
1330 prstatus->pr_pid = task_pid_vnr(p);
1331 prstatus->pr_pgrp = task_pgrp_vnr(p);
1332 prstatus->pr_sid = task_session_vnr(p);
1333 if (thread_group_leader(p)) {
1334 struct task_cputime cputime;
1335
1336 /*
1337 * This is the record for the group leader. It shows the
1338 * group-wide total, not its individual thread total.
1339 */
1340 thread_group_cputime(p, &cputime);
1341 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1342 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1343 } else {
1344 cputime_t utime, stime;
1345
1346 task_cputime(p, &utime, &stime);
1347 cputime_to_timeval(utime, &prstatus->pr_utime);
1348 cputime_to_timeval(stime, &prstatus->pr_stime);
1349 }
1350 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1351 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1352
1353 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1354 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1355 }
1356
1357 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1358 struct mm_struct *mm)
1359 {
1360 const struct cred *cred;
1361 unsigned int i, len;
1362
1363 /* first copy the parameters from user space */
1364 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1365
1366 len = mm->arg_end - mm->arg_start;
1367 if (len >= ELF_PRARGSZ)
1368 len = ELF_PRARGSZ - 1;
1369 if (copy_from_user(&psinfo->pr_psargs,
1370 (const char __user *) mm->arg_start, len))
1371 return -EFAULT;
1372 for (i = 0; i < len; i++)
1373 if (psinfo->pr_psargs[i] == 0)
1374 psinfo->pr_psargs[i] = ' ';
1375 psinfo->pr_psargs[len] = 0;
1376
1377 rcu_read_lock();
1378 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1379 rcu_read_unlock();
1380 psinfo->pr_pid = task_pid_vnr(p);
1381 psinfo->pr_pgrp = task_pgrp_vnr(p);
1382 psinfo->pr_sid = task_session_vnr(p);
1383
1384 i = p->state ? ffz(~p->state) + 1 : 0;
1385 psinfo->pr_state = i;
1386 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1387 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1388 psinfo->pr_nice = task_nice(p);
1389 psinfo->pr_flag = p->flags;
1390 rcu_read_lock();
1391 cred = __task_cred(p);
1392 SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1393 SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1394 rcu_read_unlock();
1395 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1396
1397 return 0;
1398 }
1399
1400 /* Here is the structure in which status of each thread is captured. */
1401 struct elf_thread_status
1402 {
1403 struct list_head list;
1404 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1405 elf_fpregset_t fpu; /* NT_PRFPREG */
1406 struct task_struct *thread;
1407 #ifdef ELF_CORE_COPY_XFPREGS
1408 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1409 #endif
1410 struct memelfnote notes[3];
1411 int num_notes;
1412 };
1413
1414 /*
1415 * In order to add the specific thread information for the elf file format,
1416 * we need to keep a linked list of every thread's pr_status and then create
1417 * a single section for them in the final core file.
1418 */
1419 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1420 {
1421 struct task_struct *p = t->thread;
1422 int sz = 0;
1423
1424 t->num_notes = 0;
1425
1426 fill_prstatus(&t->prstatus, p, signr);
1427 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1428
1429 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1430 &t->prstatus);
1431 t->num_notes++;
1432 sz += notesize(&t->notes[0]);
1433
1434 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1435 if (t->prstatus.pr_fpvalid) {
1436 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1437 &t->fpu);
1438 t->num_notes++;
1439 sz += notesize(&t->notes[1]);
1440 }
1441
1442 #ifdef ELF_CORE_COPY_XFPREGS
1443 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1444 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1445 sizeof(t->xfpu), &t->xfpu);
1446 t->num_notes++;
1447 sz += notesize(&t->notes[2]);
1448 }
1449 #endif
1450 return sz;
1451 }
1452
1453 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1454 elf_addr_t e_shoff, int segs)
1455 {
1456 elf->e_shoff = e_shoff;
1457 elf->e_shentsize = sizeof(*shdr4extnum);
1458 elf->e_shnum = 1;
1459 elf->e_shstrndx = SHN_UNDEF;
1460
1461 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1462
1463 shdr4extnum->sh_type = SHT_NULL;
1464 shdr4extnum->sh_size = elf->e_shnum;
1465 shdr4extnum->sh_link = elf->e_shstrndx;
1466 shdr4extnum->sh_info = segs;
1467 }
1468
1469 /*
1470 * dump the segments for an MMU process
1471 */
1472 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1473 {
1474 struct vm_area_struct *vma;
1475
1476 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1477 unsigned long addr;
1478
1479 if (!maydump(vma, cprm->mm_flags))
1480 continue;
1481
1482 #ifdef CONFIG_MMU
1483 for (addr = vma->vm_start; addr < vma->vm_end;
1484 addr += PAGE_SIZE) {
1485 bool res;
1486 struct page *page = get_dump_page(addr);
1487 if (page) {
1488 void *kaddr = kmap(page);
1489 res = dump_emit(cprm, kaddr, PAGE_SIZE);
1490 kunmap(page);
1491 page_cache_release(page);
1492 } else {
1493 res = dump_skip(cprm, PAGE_SIZE);
1494 }
1495 if (!res)
1496 return false;
1497 }
1498 #else
1499 if (!dump_emit(cprm, (void *) vma->vm_start,
1500 vma->vm_end - vma->vm_start))
1501 return false;
1502 #endif
1503 }
1504 return true;
1505 }
1506
1507 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1508 {
1509 struct vm_area_struct *vma;
1510 size_t size = 0;
1511
1512 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1513 if (maydump(vma, mm_flags))
1514 size += vma->vm_end - vma->vm_start;
1515 return size;
1516 }
1517
1518 /*
1519 * Actual dumper
1520 *
1521 * This is a two-pass process; first we find the offsets of the bits,
1522 * and then they are actually written out. If we run out of core limit
1523 * we just truncate.
1524 */
1525 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1526 {
1527 #define NUM_NOTES 6
1528 int has_dumped = 0;
1529 mm_segment_t fs;
1530 int segs;
1531 int i;
1532 struct vm_area_struct *vma;
1533 struct elfhdr *elf = NULL;
1534 loff_t offset = 0, dataoff;
1535 int numnote;
1536 struct memelfnote *notes = NULL;
1537 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1538 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1539 LIST_HEAD(thread_list);
1540 struct list_head *t;
1541 elf_fpregset_t *fpu = NULL;
1542 #ifdef ELF_CORE_COPY_XFPREGS
1543 elf_fpxregset_t *xfpu = NULL;
1544 #endif
1545 int thread_status_size = 0;
1546 elf_addr_t *auxv;
1547 struct elf_phdr *phdr4note = NULL;
1548 struct elf_shdr *shdr4extnum = NULL;
1549 Elf_Half e_phnum;
1550 elf_addr_t e_shoff;
1551 struct core_thread *ct;
1552 struct elf_thread_status *tmp;
1553
1554 /*
1555 * We no longer stop all VM operations.
1556 *
1557 * This is because those proceses that could possibly change map_count
1558 * or the mmap / vma pages are now blocked in do_exit on current
1559 * finishing this core dump.
1560 *
1561 * Only ptrace can touch these memory addresses, but it doesn't change
1562 * the map_count or the pages allocated. So no possibility of crashing
1563 * exists while dumping the mm->vm_next areas to the core file.
1564 */
1565
1566 /* alloc memory for large data structures: too large to be on stack */
1567 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1568 if (!elf)
1569 goto cleanup;
1570 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1571 if (!prstatus)
1572 goto cleanup;
1573 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1574 if (!psinfo)
1575 goto cleanup;
1576 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1577 if (!notes)
1578 goto cleanup;
1579 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1580 if (!fpu)
1581 goto cleanup;
1582 #ifdef ELF_CORE_COPY_XFPREGS
1583 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1584 if (!xfpu)
1585 goto cleanup;
1586 #endif
1587
1588 for (ct = current->mm->core_state->dumper.next;
1589 ct; ct = ct->next) {
1590 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1591 if (!tmp)
1592 goto cleanup;
1593
1594 tmp->thread = ct->task;
1595 list_add(&tmp->list, &thread_list);
1596 }
1597
1598 list_for_each(t, &thread_list) {
1599 struct elf_thread_status *tmp;
1600 int sz;
1601
1602 tmp = list_entry(t, struct elf_thread_status, list);
1603 sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1604 thread_status_size += sz;
1605 }
1606
1607 /* now collect the dump for the current */
1608 fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1609 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1610
1611 segs = current->mm->map_count;
1612 segs += elf_core_extra_phdrs();
1613
1614 /* for notes section */
1615 segs++;
1616
1617 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1618 * this, kernel supports extended numbering. Have a look at
1619 * include/linux/elf.h for further information. */
1620 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1621
1622 /* Set up header */
1623 fill_elf_fdpic_header(elf, e_phnum);
1624
1625 has_dumped = 1;
1626 /*
1627 * Set up the notes in similar form to SVR4 core dumps made
1628 * with info from their /proc.
1629 */
1630
1631 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1632 fill_psinfo(psinfo, current->group_leader, current->mm);
1633 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1634
1635 numnote = 2;
1636
1637 auxv = (elf_addr_t *) current->mm->saved_auxv;
1638
1639 i = 0;
1640 do
1641 i += 2;
1642 while (auxv[i - 2] != AT_NULL);
1643 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1644 i * sizeof(elf_addr_t), auxv);
1645
1646 /* Try to dump the FPU. */
1647 if ((prstatus->pr_fpvalid =
1648 elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1649 fill_note(notes + numnote++,
1650 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1651 #ifdef ELF_CORE_COPY_XFPREGS
1652 if (elf_core_copy_task_xfpregs(current, xfpu))
1653 fill_note(notes + numnote++,
1654 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1655 #endif
1656
1657 fs = get_fs();
1658 set_fs(KERNEL_DS);
1659
1660 offset += sizeof(*elf); /* Elf header */
1661 offset += segs * sizeof(struct elf_phdr); /* Program headers */
1662
1663 /* Write notes phdr entry */
1664 {
1665 int sz = 0;
1666
1667 for (i = 0; i < numnote; i++)
1668 sz += notesize(notes + i);
1669
1670 sz += thread_status_size;
1671
1672 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1673 if (!phdr4note)
1674 goto end_coredump;
1675
1676 fill_elf_note_phdr(phdr4note, sz, offset);
1677 offset += sz;
1678 }
1679
1680 /* Page-align dumped data */
1681 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1682
1683 offset += elf_core_vma_data_size(cprm->mm_flags);
1684 offset += elf_core_extra_data_size();
1685 e_shoff = offset;
1686
1687 if (e_phnum == PN_XNUM) {
1688 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1689 if (!shdr4extnum)
1690 goto end_coredump;
1691 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1692 }
1693
1694 offset = dataoff;
1695
1696 if (!dump_emit(cprm, elf, sizeof(*elf)))
1697 goto end_coredump;
1698
1699 if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1700 goto end_coredump;
1701
1702 /* write program headers for segments dump */
1703 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1704 struct elf_phdr phdr;
1705 size_t sz;
1706
1707 sz = vma->vm_end - vma->vm_start;
1708
1709 phdr.p_type = PT_LOAD;
1710 phdr.p_offset = offset;
1711 phdr.p_vaddr = vma->vm_start;
1712 phdr.p_paddr = 0;
1713 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1714 phdr.p_memsz = sz;
1715 offset += phdr.p_filesz;
1716 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1717 if (vma->vm_flags & VM_WRITE)
1718 phdr.p_flags |= PF_W;
1719 if (vma->vm_flags & VM_EXEC)
1720 phdr.p_flags |= PF_X;
1721 phdr.p_align = ELF_EXEC_PAGESIZE;
1722
1723 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1724 goto end_coredump;
1725 }
1726
1727 if (!elf_core_write_extra_phdrs(cprm, offset))
1728 goto end_coredump;
1729
1730 /* write out the notes section */
1731 for (i = 0; i < numnote; i++)
1732 if (!writenote(notes + i, cprm))
1733 goto end_coredump;
1734
1735 /* write out the thread status notes section */
1736 list_for_each(t, &thread_list) {
1737 struct elf_thread_status *tmp =
1738 list_entry(t, struct elf_thread_status, list);
1739
1740 for (i = 0; i < tmp->num_notes; i++)
1741 if (!writenote(&tmp->notes[i], cprm))
1742 goto end_coredump;
1743 }
1744
1745 if (!dump_skip(cprm, dataoff - cprm->written))
1746 goto end_coredump;
1747
1748 if (!elf_fdpic_dump_segments(cprm))
1749 goto end_coredump;
1750
1751 if (!elf_core_write_extra_data(cprm))
1752 goto end_coredump;
1753
1754 if (e_phnum == PN_XNUM) {
1755 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1756 goto end_coredump;
1757 }
1758
1759 if (cprm->file->f_pos != offset) {
1760 /* Sanity check */
1761 printk(KERN_WARNING
1762 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1763 cprm->file->f_pos, offset);
1764 }
1765
1766 end_coredump:
1767 set_fs(fs);
1768
1769 cleanup:
1770 while (!list_empty(&thread_list)) {
1771 struct list_head *tmp = thread_list.next;
1772 list_del(tmp);
1773 kfree(list_entry(tmp, struct elf_thread_status, list));
1774 }
1775 kfree(phdr4note);
1776 kfree(elf);
1777 kfree(prstatus);
1778 kfree(psinfo);
1779 kfree(notes);
1780 kfree(fpu);
1781 kfree(shdr4extnum);
1782 #ifdef ELF_CORE_COPY_XFPREGS
1783 kfree(xfpu);
1784 #endif
1785 return has_dumped;
1786 #undef NUM_NOTES
1787 }
1788
1789 #endif /* CONFIG_ELF_CORE */
This page took 0.099943 seconds and 6 git commands to generate.