1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2 *
3 * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 * Derived from binfmt_elf.c
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13 #include <linux/module.h>
14
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/param.h>
40 #include <asm/pgalloc.h>
41
42 typedef char *elf_caddr_t;
43
44 #if 0
45 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
46 #else
47 #define kdebug(fmt, ...) do {} while(0)
48 #endif
49
50 #if 0
51 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
52 #else
53 #define kdcore(fmt, ...) do {} while(0)
54 #endif
55
56 MODULE_LICENSE("GPL");
57
58 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
59 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
60 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
61 struct mm_struct *, const char *);
62
63 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
64 struct elf_fdpic_params *,
65 struct elf_fdpic_params *);
66
67 #ifndef CONFIG_MMU
68 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
69 unsigned long *);
70 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
71 struct file *,
72 struct mm_struct *);
73 #endif
74
75 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
76 struct file *, struct mm_struct *);
77
78 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
79 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *, unsigned long limit);
80 #endif
81
82 static struct linux_binfmt elf_fdpic_format = {
83 .module = THIS_MODULE,
84 .load_binary = load_elf_fdpic_binary,
85 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
86 .core_dump = elf_fdpic_core_dump,
87 #endif
88 .min_coredump = ELF_EXEC_PAGESIZE,
89 };
90
91 static int __init init_elf_fdpic_binfmt(void)
92 {
93 return register_binfmt(&elf_fdpic_format);
94 }
95
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 unregister_binfmt(&elf_fdpic_format);
99 }
100
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103
104 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
105 {
106 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 return 0;
108 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 return 0;
110 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
111 return 0;
112 if (!file->f_op || !file->f_op->mmap)
113 return 0;
114 return 1;
115 }
116
117 /*****************************************************************************/
118 /*
119 * read the program headers table into memory
120 */
121 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
122 struct file *file)
123 {
124 struct elf32_phdr *phdr;
125 unsigned long size;
126 int retval, loop;
127
128 if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
129 return -ENOMEM;
130 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
131 return -ENOMEM;
132
133 size = params->hdr.e_phnum * sizeof(struct elf_phdr);
134 params->phdrs = kmalloc(size, GFP_KERNEL);
135 if (!params->phdrs)
136 return -ENOMEM;
137
138 retval = kernel_read(file, params->hdr.e_phoff,
139 (char *) params->phdrs, size);
140 if (unlikely(retval != size))
141 return retval < 0 ? retval : -ENOEXEC;
142
143 /* determine stack size for this binary */
144 phdr = params->phdrs;
145 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
146 if (phdr->p_type != PT_GNU_STACK)
147 continue;
148
149 if (phdr->p_flags & PF_X)
150 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
151 else
152 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
153
154 params->stack_size = phdr->p_memsz;
155 break;
156 }
157
158 return 0;
159 }
160
161 /*****************************************************************************/
162 /*
163 * load an fdpic binary into various bits of memory
164 */
165 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
166 struct pt_regs *regs)
167 {
168 struct elf_fdpic_params exec_params, interp_params;
169 struct elf_phdr *phdr;
170 unsigned long stack_size, entryaddr;
171 #ifdef ELF_FDPIC_PLAT_INIT
172 unsigned long dynaddr;
173 #endif
174 struct file *interpreter = NULL; /* to shut gcc up */
175 char *interpreter_name = NULL;
176 int executable_stack;
177 int retval, i;
178
179 kdebug("____ LOAD %d ____", current->pid);
180
181 memset(&exec_params, 0, sizeof(exec_params));
182 memset(&interp_params, 0, sizeof(interp_params));
183
184 exec_params.hdr = *(struct elfhdr *) bprm->buf;
185 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
186
187 /* check that this is a binary we know how to deal with */
188 retval = -ENOEXEC;
189 if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
190 goto error;
191
192 /* read the program header table */
193 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
194 if (retval < 0)
195 goto error;
196
197 /* scan for a program header that specifies an interpreter */
198 phdr = exec_params.phdrs;
199
200 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
201 switch (phdr->p_type) {
202 case PT_INTERP:
203 retval = -ENOMEM;
204 if (phdr->p_filesz > PATH_MAX)
205 goto error;
206 retval = -ENOENT;
207 if (phdr->p_filesz < 2)
208 goto error;
209
210 /* read the name of the interpreter into memory */
211 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
212 if (!interpreter_name)
213 goto error;
214
215 retval = kernel_read(bprm->file,
216 phdr->p_offset,
217 interpreter_name,
218 phdr->p_filesz);
219 if (unlikely(retval != phdr->p_filesz)) {
220 if (retval >= 0)
221 retval = -ENOEXEC;
222 goto error;
223 }
224
225 retval = -ENOENT;
226 if (interpreter_name[phdr->p_filesz - 1] != '\0')
227 goto error;
228
229 kdebug("Using ELF interpreter %s", interpreter_name);
230
231 /* replace the program with the interpreter */
232 interpreter = open_exec(interpreter_name);
233 retval = PTR_ERR(interpreter);
234 if (IS_ERR(interpreter)) {
235 interpreter = NULL;
236 goto error;
237 }
238
239 /*
240 * If the binary is not readable then enforce
241 * mm->dumpable = 0 regardless of the interpreter's
242 * permissions.
243 */
244 if (file_permission(interpreter, MAY_READ) < 0)
245 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
246
247 retval = kernel_read(interpreter, 0, bprm->buf,
248 BINPRM_BUF_SIZE);
249 if (unlikely(retval != BINPRM_BUF_SIZE)) {
250 if (retval >= 0)
251 retval = -ENOEXEC;
252 goto error;
253 }
254
255 interp_params.hdr = *((struct elfhdr *) bprm->buf);
256 break;
257
258 case PT_LOAD:
259 #ifdef CONFIG_MMU
260 if (exec_params.load_addr == 0)
261 exec_params.load_addr = phdr->p_vaddr;
262 #endif
263 break;
264 }
265
266 }
267
268 if (elf_check_const_displacement(&exec_params.hdr))
269 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
270
271 /* perform insanity checks on the interpreter */
272 if (interpreter_name) {
273 retval = -ELIBBAD;
274 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
275 goto error;
276
277 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
278
279 /* read the interpreter's program header table */
280 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
281 if (retval < 0)
282 goto error;
283 }
284
285 stack_size = exec_params.stack_size;
286 if (stack_size < interp_params.stack_size)
287 stack_size = interp_params.stack_size;
288
289 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
290 executable_stack = EXSTACK_ENABLE_X;
291 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
292 executable_stack = EXSTACK_DISABLE_X;
293 else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
294 executable_stack = EXSTACK_ENABLE_X;
295 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
296 executable_stack = EXSTACK_DISABLE_X;
297 else
298 executable_stack = EXSTACK_DEFAULT;
299
300 retval = -ENOEXEC;
301 if (stack_size == 0)
302 goto error;
303
304 if (elf_check_const_displacement(&interp_params.hdr))
305 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
306
307 /* flush all traces of the currently running executable */
308 retval = flush_old_exec(bprm);
309 if (retval)
310 goto error;
311
312 /* there's now no turning back... the old userspace image is dead,
313 * defunct, deceased, etc. after this point we have to exit via
314 * error_kill */
315 set_personality(PER_LINUX_FDPIC);
316
317 setup_new_exec(bprm);
318
319 set_binfmt(&elf_fdpic_format);
320
321 current->mm->start_code = 0;
322 current->mm->end_code = 0;
323 current->mm->start_stack = 0;
324 current->mm->start_data = 0;
325 current->mm->end_data = 0;
326 current->mm->context.exec_fdpic_loadmap = 0;
327 current->mm->context.interp_fdpic_loadmap = 0;
328
329 current->flags &= ~PF_FORKNOEXEC;
330
331 #ifdef CONFIG_MMU
332 elf_fdpic_arch_lay_out_mm(&exec_params,
333 &interp_params,
334 ¤t->mm->start_stack,
335 ¤t->mm->start_brk);
336
337 retval = setup_arg_pages(bprm, current->mm->start_stack,
338 executable_stack);
339 if (retval < 0) {
340 send_sig(SIGKILL, current, 0);
341 goto error_kill;
342 }
343 #endif
344
345 /* load the executable and interpreter into memory */
346 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
347 "executable");
348 if (retval < 0)
349 goto error_kill;
350
351 if (interpreter_name) {
352 retval = elf_fdpic_map_file(&interp_params, interpreter,
353 current->mm, "interpreter");
354 if (retval < 0) {
355 printk(KERN_ERR "Unable to load interpreter\n");
356 goto error_kill;
357 }
358
359 allow_write_access(interpreter);
360 fput(interpreter);
361 interpreter = NULL;
362 }
363
364 #ifdef CONFIG_MMU
365 if (!current->mm->start_brk)
366 current->mm->start_brk = current->mm->end_data;
367
368 current->mm->brk = current->mm->start_brk =
369 PAGE_ALIGN(current->mm->start_brk);
370
371 #else
372 /* create a stack and brk area big enough for everyone
373 * - the brk heap starts at the bottom and works up
374 * - the stack starts at the top and works down
375 */
376 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
377 if (stack_size < PAGE_SIZE * 2)
378 stack_size = PAGE_SIZE * 2;
379
380 down_write(¤t->mm->mmap_sem);
381 current->mm->start_brk = do_mmap(NULL, 0, stack_size,
382 PROT_READ | PROT_WRITE | PROT_EXEC,
383 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
384 0);
385
386 if (IS_ERR_VALUE(current->mm->start_brk)) {
387 up_write(¤t->mm->mmap_sem);
388 retval = current->mm->start_brk;
389 current->mm->start_brk = 0;
390 goto error_kill;
391 }
392
393 up_write(¤t->mm->mmap_sem);
394
395 current->mm->brk = current->mm->start_brk;
396 current->mm->context.end_brk = current->mm->start_brk;
397 current->mm->context.end_brk +=
398 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
399 current->mm->start_stack = current->mm->start_brk + stack_size;
400 #endif
401
402 install_exec_creds(bprm);
403 current->flags &= ~PF_FORKNOEXEC;
404 if (create_elf_fdpic_tables(bprm, current->mm,
405 &exec_params, &interp_params) < 0)
406 goto error_kill;
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 /* unrecoverable error - kill the process */
447 error_kill:
448 send_sig(SIGSEGV, current, 0);
449 goto error;
450
451 }
452
453 /*****************************************************************************/
454
455 #ifndef ELF_BASE_PLATFORM
456 /*
457 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
458 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
459 * will be copied to the user stack in the same manner as AT_PLATFORM.
460 */
461 #define ELF_BASE_PLATFORM NULL
462 #endif
463
464 /*
465 * present useful information to the program by shovelling it onto the new
466 * process's stack
467 */
468 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
469 struct mm_struct *mm,
470 struct elf_fdpic_params *exec_params,
471 struct elf_fdpic_params *interp_params)
472 {
473 const struct cred *cred = current_cred();
474 unsigned long sp, csp, nitems;
475 elf_caddr_t __user *argv, *envp;
476 size_t platform_len = 0, len;
477 char *k_platform, *k_base_platform;
478 char __user *u_platform, *u_base_platform, *p;
479 long hwcap;
480 int loop;
481 int nr; /* reset for each csp adjustment */
482
483 #ifdef CONFIG_MMU
484 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
485 * by the processes running on the same package. One thing we can do is
486 * to shuffle the initial stack for them, so we give the architecture
487 * an opportunity to do so here.
488 */
489 sp = arch_align_stack(bprm->p);
490 #else
491 sp = mm->start_stack;
492
493 /* stack the program arguments and environment */
494 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
495 return -EFAULT;
496 #endif
497
498 hwcap = ELF_HWCAP;
499
500 /*
501 * If this architecture has a platform capability string, copy it
502 * to userspace. In some cases (Sparc), this info is impossible
503 * for userspace to get any other way, in others (i386) it is
504 * merely difficult.
505 */
506 k_platform = ELF_PLATFORM;
507 u_platform = NULL;
508
509 if (k_platform) {
510 platform_len = strlen(k_platform) + 1;
511 sp -= platform_len;
512 u_platform = (char __user *) sp;
513 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
514 return -EFAULT;
515 }
516
517 /*
518 * If this architecture has a "base" platform capability
519 * string, copy it to userspace.
520 */
521 k_base_platform = ELF_BASE_PLATFORM;
522 u_base_platform = NULL;
523
524 if (k_base_platform) {
525 platform_len = strlen(k_base_platform) + 1;
526 sp -= platform_len;
527 u_base_platform = (char __user *) sp;
528 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
529 return -EFAULT;
530 }
531
532 sp &= ~7UL;
533
534 /* stack the load map(s) */
535 len = sizeof(struct elf32_fdpic_loadmap);
536 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
537 sp = (sp - len) & ~7UL;
538 exec_params->map_addr = sp;
539
540 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
541 return -EFAULT;
542
543 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
544
545 if (interp_params->loadmap) {
546 len = sizeof(struct elf32_fdpic_loadmap);
547 len += sizeof(struct elf32_fdpic_loadseg) *
548 interp_params->loadmap->nsegs;
549 sp = (sp - len) & ~7UL;
550 interp_params->map_addr = sp;
551
552 if (copy_to_user((void __user *) sp, interp_params->loadmap,
553 len) != 0)
554 return -EFAULT;
555
556 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
557 }
558
559 /* force 16 byte _final_ alignment here for generality */
560 #define DLINFO_ITEMS 15
561
562 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
563 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
564
565 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
566 nitems++;
567
568 csp = sp;
569 sp -= nitems * 2 * sizeof(unsigned long);
570 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */
571 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */
572 sp -= 1 * sizeof(unsigned long); /* argc */
573
574 csp -= sp & 15UL;
575 sp -= sp & 15UL;
576
577 /* put the ELF interpreter info on the stack */
578 #define NEW_AUX_ENT(id, val) \
579 do { \
580 struct { unsigned long _id, _val; } __user *ent; \
581 \
582 ent = (void __user *) csp; \
583 __put_user((id), &ent[nr]._id); \
584 __put_user((val), &ent[nr]._val); \
585 nr++; \
586 } while (0)
587
588 nr = 0;
589 csp -= 2 * sizeof(unsigned long);
590 NEW_AUX_ENT(AT_NULL, 0);
591 if (k_platform) {
592 nr = 0;
593 csp -= 2 * sizeof(unsigned long);
594 NEW_AUX_ENT(AT_PLATFORM,
595 (elf_addr_t) (unsigned long) u_platform);
596 }
597
598 if (k_base_platform) {
599 nr = 0;
600 csp -= 2 * sizeof(unsigned long);
601 NEW_AUX_ENT(AT_BASE_PLATFORM,
602 (elf_addr_t) (unsigned long) u_base_platform);
603 }
604
605 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
606 nr = 0;
607 csp -= 2 * sizeof(unsigned long);
608 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
609 }
610
611 nr = 0;
612 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
613 NEW_AUX_ENT(AT_HWCAP, hwcap);
614 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE);
615 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
616 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr);
617 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
618 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum);
619 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr);
620 NEW_AUX_ENT(AT_FLAGS, 0);
621 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr);
622 NEW_AUX_ENT(AT_UID, (elf_addr_t) cred->uid);
623 NEW_AUX_ENT(AT_EUID, (elf_addr_t) cred->euid);
624 NEW_AUX_ENT(AT_GID, (elf_addr_t) cred->gid);
625 NEW_AUX_ENT(AT_EGID, (elf_addr_t) cred->egid);
626 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
627 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
628
629 #ifdef ARCH_DLINFO
630 nr = 0;
631 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
632
633 /* ARCH_DLINFO must come last so platform specific code can enforce
634 * special alignment requirements on the AUXV if necessary (eg. PPC).
635 */
636 ARCH_DLINFO;
637 #endif
638 #undef NEW_AUX_ENT
639
640 /* allocate room for argv[] and envv[] */
641 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
642 envp = (elf_caddr_t __user *) csp;
643 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
644 argv = (elf_caddr_t __user *) csp;
645
646 /* stack argc */
647 csp -= sizeof(unsigned long);
648 __put_user(bprm->argc, (unsigned long __user *) csp);
649
650 BUG_ON(csp != sp);
651
652 /* fill in the argv[] array */
653 #ifdef CONFIG_MMU
654 current->mm->arg_start = bprm->p;
655 #else
656 current->mm->arg_start = current->mm->start_stack -
657 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
658 #endif
659
660 p = (char __user *) current->mm->arg_start;
661 for (loop = bprm->argc; loop > 0; loop--) {
662 __put_user((elf_caddr_t) p, argv++);
663 len = strnlen_user(p, MAX_ARG_STRLEN);
664 if (!len || len > MAX_ARG_STRLEN)
665 return -EINVAL;
666 p += len;
667 }
668 __put_user(NULL, argv);
669 current->mm->arg_end = (unsigned long) p;
670
671 /* fill in the envv[] array */
672 current->mm->env_start = (unsigned long) p;
673 for (loop = bprm->envc; loop > 0; loop--) {
674 __put_user((elf_caddr_t)(unsigned long) p, envp++);
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, envp);
681 current->mm->env_end = (unsigned long) p;
682
683 mm->start_stack = (unsigned long) sp;
684 return 0;
685 }
686
687 /*****************************************************************************/
688 /*
689 * transfer the program arguments and environment from the holding pages onto
690 * the stack
691 */
692 #ifndef CONFIG_MMU
693 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
694 unsigned long *_sp)
695 {
696 unsigned long index, stop, sp;
697 char *src;
698 int ret = 0;
699
700 stop = bprm->p >> PAGE_SHIFT;
701 sp = *_sp;
702
703 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
704 src = kmap(bprm->page[index]);
705 sp -= PAGE_SIZE;
706 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
707 ret = -EFAULT;
708 kunmap(bprm->page[index]);
709 if (ret < 0)
710 goto out;
711 }
712
713 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
714
715 out:
716 return ret;
717 }
718 #endif
719
720 /*****************************************************************************/
721 /*
722 * load the appropriate binary image (executable or interpreter) into memory
723 * - we assume no MMU is available
724 * - if no other PIC bits are set in params->hdr->e_flags
725 * - we assume that the LOADable segments in the binary are independently relocatable
726 * - we assume R/O executable segments are shareable
727 * - else
728 * - we assume the loadable parts of the image to require fixed displacement
729 * - the image is not shareable
730 */
731 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
732 struct file *file,
733 struct mm_struct *mm,
734 const char *what)
735 {
736 struct elf32_fdpic_loadmap *loadmap;
737 #ifdef CONFIG_MMU
738 struct elf32_fdpic_loadseg *mseg;
739 #endif
740 struct elf32_fdpic_loadseg *seg;
741 struct elf32_phdr *phdr;
742 unsigned long load_addr, stop;
743 unsigned nloads, tmp;
744 size_t size;
745 int loop, ret;
746
747 /* allocate a load map table */
748 nloads = 0;
749 for (loop = 0; loop < params->hdr.e_phnum; loop++)
750 if (params->phdrs[loop].p_type == PT_LOAD)
751 nloads++;
752
753 if (nloads == 0)
754 return -ELIBBAD;
755
756 size = sizeof(*loadmap) + nloads * sizeof(*seg);
757 loadmap = kzalloc(size, GFP_KERNEL);
758 if (!loadmap)
759 return -ENOMEM;
760
761 params->loadmap = loadmap;
762
763 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
764 loadmap->nsegs = nloads;
765
766 load_addr = params->load_addr;
767 seg = loadmap->segs;
768
769 /* map the requested LOADs into the memory space */
770 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
771 case ELF_FDPIC_FLAG_CONSTDISP:
772 case ELF_FDPIC_FLAG_CONTIGUOUS:
773 #ifndef CONFIG_MMU
774 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
775 if (ret < 0)
776 return ret;
777 break;
778 #endif
779 default:
780 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
781 if (ret < 0)
782 return ret;
783 break;
784 }
785
786 /* map the entry point */
787 if (params->hdr.e_entry) {
788 seg = loadmap->segs;
789 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
790 if (params->hdr.e_entry >= seg->p_vaddr &&
791 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
792 params->entry_addr =
793 (params->hdr.e_entry - seg->p_vaddr) +
794 seg->addr;
795 break;
796 }
797 }
798 }
799
800 /* determine where the program header table has wound up if mapped */
801 stop = params->hdr.e_phoff;
802 stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
803 phdr = params->phdrs;
804
805 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
806 if (phdr->p_type != PT_LOAD)
807 continue;
808
809 if (phdr->p_offset > params->hdr.e_phoff ||
810 phdr->p_offset + phdr->p_filesz < stop)
811 continue;
812
813 seg = loadmap->segs;
814 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
815 if (phdr->p_vaddr >= seg->p_vaddr &&
816 phdr->p_vaddr + phdr->p_filesz <=
817 seg->p_vaddr + seg->p_memsz) {
818 params->ph_addr =
819 (phdr->p_vaddr - seg->p_vaddr) +
820 seg->addr +
821 params->hdr.e_phoff - phdr->p_offset;
822 break;
823 }
824 }
825 break;
826 }
827
828 /* determine where the dynamic section has wound up if there is one */
829 phdr = params->phdrs;
830 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
831 if (phdr->p_type != PT_DYNAMIC)
832 continue;
833
834 seg = loadmap->segs;
835 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
836 if (phdr->p_vaddr >= seg->p_vaddr &&
837 phdr->p_vaddr + phdr->p_memsz <=
838 seg->p_vaddr + seg->p_memsz) {
839 params->dynamic_addr =
840 (phdr->p_vaddr - seg->p_vaddr) +
841 seg->addr;
842
843 /* check the dynamic section contains at least
844 * one item, and that the last item is a NULL
845 * entry */
846 if (phdr->p_memsz == 0 ||
847 phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
848 goto dynamic_error;
849
850 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
851 if (((Elf32_Dyn *)
852 params->dynamic_addr)[tmp - 1].d_tag != 0)
853 goto dynamic_error;
854 break;
855 }
856 }
857 break;
858 }
859
860 /* now elide adjacent segments in the load map on MMU linux
861 * - on uClinux the holes between may actually be filled with system
862 * stuff or stuff from other processes
863 */
864 #ifdef CONFIG_MMU
865 nloads = loadmap->nsegs;
866 mseg = loadmap->segs;
867 seg = mseg + 1;
868 for (loop = 1; loop < nloads; loop++) {
869 /* see if we have a candidate for merging */
870 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
871 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
872 if (load_addr == (seg->addr & PAGE_MASK)) {
873 mseg->p_memsz +=
874 load_addr -
875 (mseg->addr + mseg->p_memsz);
876 mseg->p_memsz += seg->addr & ~PAGE_MASK;
877 mseg->p_memsz += seg->p_memsz;
878 loadmap->nsegs--;
879 continue;
880 }
881 }
882
883 mseg++;
884 if (mseg != seg)
885 *mseg = *seg;
886 }
887 #endif
888
889 kdebug("Mapped Object [%s]:", what);
890 kdebug("- elfhdr : %lx", params->elfhdr_addr);
891 kdebug("- entry : %lx", params->entry_addr);
892 kdebug("- PHDR[] : %lx", params->ph_addr);
893 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
894 seg = loadmap->segs;
895 for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
896 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
897 loop,
898 seg->addr, seg->addr + seg->p_memsz - 1,
899 seg->p_vaddr, seg->p_memsz);
900
901 return 0;
902
903 dynamic_error:
904 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
905 what, file->f_path.dentry->d_inode->i_ino);
906 return -ELIBBAD;
907 }
908
909 /*****************************************************************************/
910 /*
911 * map a file with constant displacement under uClinux
912 */
913 #ifndef CONFIG_MMU
914 static int elf_fdpic_map_file_constdisp_on_uclinux(
915 struct elf_fdpic_params *params,
916 struct file *file,
917 struct mm_struct *mm)
918 {
919 struct elf32_fdpic_loadseg *seg;
920 struct elf32_phdr *phdr;
921 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
922 loff_t fpos;
923 int loop, ret;
924
925 load_addr = params->load_addr;
926 seg = params->loadmap->segs;
927
928 /* determine the bounds of the contiguous overall allocation we must
929 * make */
930 phdr = params->phdrs;
931 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
932 if (params->phdrs[loop].p_type != PT_LOAD)
933 continue;
934
935 if (base > phdr->p_vaddr)
936 base = phdr->p_vaddr;
937 if (top < phdr->p_vaddr + phdr->p_memsz)
938 top = phdr->p_vaddr + phdr->p_memsz;
939 }
940
941 /* allocate one big anon block for everything */
942 mflags = MAP_PRIVATE;
943 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
944 mflags |= MAP_EXECUTABLE;
945
946 down_write(&mm->mmap_sem);
947 maddr = do_mmap(NULL, load_addr, top - base,
948 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
949 up_write(&mm->mmap_sem);
950 if (IS_ERR_VALUE(maddr))
951 return (int) maddr;
952
953 if (load_addr != 0)
954 load_addr += PAGE_ALIGN(top - base);
955
956 /* and then load the file segments into it */
957 phdr = params->phdrs;
958 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
959 if (params->phdrs[loop].p_type != PT_LOAD)
960 continue;
961
962 fpos = phdr->p_offset;
963
964 seg->addr = maddr + (phdr->p_vaddr - base);
965 seg->p_vaddr = phdr->p_vaddr;
966 seg->p_memsz = phdr->p_memsz;
967
968 ret = file->f_op->read(file, (void *) seg->addr,
969 phdr->p_filesz, &fpos);
970 if (ret < 0)
971 return ret;
972
973 /* map the ELF header address if in this segment */
974 if (phdr->p_offset == 0)
975 params->elfhdr_addr = seg->addr;
976
977 /* clear any space allocated but not loaded */
978 if (phdr->p_filesz < phdr->p_memsz) {
979 ret = clear_user((void *) (seg->addr + phdr->p_filesz),
980 phdr->p_memsz - phdr->p_filesz);
981 if (ret)
982 return ret;
983 }
984
985 if (mm) {
986 if (phdr->p_flags & PF_X) {
987 if (!mm->start_code) {
988 mm->start_code = seg->addr;
989 mm->end_code = seg->addr +
990 phdr->p_memsz;
991 }
992 } else if (!mm->start_data) {
993 mm->start_data = seg->addr;
994 #ifndef CONFIG_MMU
995 mm->end_data = seg->addr + phdr->p_memsz;
996 #endif
997 }
998
999 #ifdef CONFIG_MMU
1000 if (seg->addr + phdr->p_memsz > mm->end_data)
1001 mm->end_data = seg->addr + phdr->p_memsz;
1002 #endif
1003 }
1004
1005 seg++;
1006 }
1007
1008 return 0;
1009 }
1010 #endif
1011
1012 /*****************************************************************************/
1013 /*
1014 * map a binary by direct mmap() of the individual PT_LOAD segments
1015 */
1016 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1017 struct file *file,
1018 struct mm_struct *mm)
1019 {
1020 struct elf32_fdpic_loadseg *seg;
1021 struct elf32_phdr *phdr;
1022 unsigned long load_addr, delta_vaddr;
1023 int loop, dvset, ret;
1024
1025 load_addr = params->load_addr;
1026 delta_vaddr = 0;
1027 dvset = 0;
1028
1029 seg = params->loadmap->segs;
1030
1031 /* deal with each load segment separately */
1032 phdr = params->phdrs;
1033 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1034 unsigned long maddr, disp, excess, excess1;
1035 int prot = 0, flags;
1036
1037 if (phdr->p_type != PT_LOAD)
1038 continue;
1039
1040 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1041 (unsigned long) phdr->p_vaddr,
1042 (unsigned long) phdr->p_offset,
1043 (unsigned long) phdr->p_filesz,
1044 (unsigned long) phdr->p_memsz);
1045
1046 /* determine the mapping parameters */
1047 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1048 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1049 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1050
1051 flags = MAP_PRIVATE | MAP_DENYWRITE;
1052 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1053 flags |= MAP_EXECUTABLE;
1054
1055 maddr = 0;
1056
1057 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1058 case ELF_FDPIC_FLAG_INDEPENDENT:
1059 /* PT_LOADs are independently locatable */
1060 break;
1061
1062 case ELF_FDPIC_FLAG_HONOURVADDR:
1063 /* the specified virtual address must be honoured */
1064 maddr = phdr->p_vaddr;
1065 flags |= MAP_FIXED;
1066 break;
1067
1068 case ELF_FDPIC_FLAG_CONSTDISP:
1069 /* constant displacement
1070 * - can be mapped anywhere, but must be mapped as a
1071 * unit
1072 */
1073 if (!dvset) {
1074 maddr = load_addr;
1075 delta_vaddr = phdr->p_vaddr;
1076 dvset = 1;
1077 } else {
1078 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1079 flags |= MAP_FIXED;
1080 }
1081 break;
1082
1083 case ELF_FDPIC_FLAG_CONTIGUOUS:
1084 /* contiguity handled later */
1085 break;
1086
1087 default:
1088 BUG();
1089 }
1090
1091 maddr &= PAGE_MASK;
1092
1093 /* create the mapping */
1094 disp = phdr->p_vaddr & ~PAGE_MASK;
1095 down_write(&mm->mmap_sem);
1096 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1097 phdr->p_offset - disp);
1098 up_write(&mm->mmap_sem);
1099
1100 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1101 loop, phdr->p_memsz + disp, prot, flags,
1102 phdr->p_offset - disp, maddr);
1103
1104 if (IS_ERR_VALUE(maddr))
1105 return (int) maddr;
1106
1107 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1108 ELF_FDPIC_FLAG_CONTIGUOUS)
1109 load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1110
1111 seg->addr = maddr + disp;
1112 seg->p_vaddr = phdr->p_vaddr;
1113 seg->p_memsz = phdr->p_memsz;
1114
1115 /* map the ELF header address if in this segment */
1116 if (phdr->p_offset == 0)
1117 params->elfhdr_addr = seg->addr;
1118
1119 /* clear the bit between beginning of mapping and beginning of
1120 * PT_LOAD */
1121 if (prot & PROT_WRITE && disp > 0) {
1122 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1123 ret = clear_user((void __user *) maddr, disp);
1124 if (ret)
1125 return ret;
1126 maddr += disp;
1127 }
1128
1129 /* clear any space allocated but not loaded
1130 * - on uClinux we can just clear the lot
1131 * - on MMU linux we'll get a SIGBUS beyond the last page
1132 * extant in the file
1133 */
1134 excess = phdr->p_memsz - phdr->p_filesz;
1135 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1136
1137 #ifdef CONFIG_MMU
1138 if (excess > excess1) {
1139 unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1140 unsigned long xmaddr;
1141
1142 flags |= MAP_FIXED | MAP_ANONYMOUS;
1143 down_write(&mm->mmap_sem);
1144 xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1145 prot, flags, 0);
1146 up_write(&mm->mmap_sem);
1147
1148 kdebug("mmap[%d] <anon>"
1149 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1150 loop, xaddr, excess - excess1, prot, flags,
1151 xmaddr);
1152
1153 if (xmaddr != xaddr)
1154 return -ENOMEM;
1155 }
1156
1157 if (prot & PROT_WRITE && excess1 > 0) {
1158 kdebug("clear[%d] ad=%lx sz=%lx",
1159 loop, maddr + phdr->p_filesz, excess1);
1160 ret = clear_user((void __user *) maddr + phdr->p_filesz,
1161 excess1);
1162 if (ret)
1163 return ret;
1164 }
1165
1166 #else
1167 if (excess > 0) {
1168 kdebug("clear[%d] ad=%lx sz=%lx",
1169 loop, maddr + phdr->p_filesz, excess);
1170 ret = clear_user((void *) maddr + phdr->p_filesz, excess);
1171 if (ret)
1172 return ret;
1173 }
1174 #endif
1175
1176 if (mm) {
1177 if (phdr->p_flags & PF_X) {
1178 if (!mm->start_code) {
1179 mm->start_code = maddr;
1180 mm->end_code = maddr + phdr->p_memsz;
1181 }
1182 } else if (!mm->start_data) {
1183 mm->start_data = maddr;
1184 mm->end_data = maddr + phdr->p_memsz;
1185 }
1186 }
1187
1188 seg++;
1189 }
1190
1191 return 0;
1192 }
1193
1194 /*****************************************************************************/
1195 /*
1196 * ELF-FDPIC core dumper
1197 *
1198 * Modelled on fs/exec.c:aout_core_dump()
1199 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1200 *
1201 * Modelled on fs/binfmt_elf.c core dumper
1202 */
1203 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1204
1205 /*
1206 * These are the only things you should do on a core-file: use only these
1207 * functions to write out all the necessary info.
1208 */
1209 static int dump_write(struct file *file, const void *addr, int nr)
1210 {
1211 return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1212 }
1213
1214 static int dump_seek(struct file *file, loff_t off)
1215 {
1216 if (file->f_op->llseek) {
1217 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1218 return 0;
1219 } else {
1220 file->f_pos = off;
1221 }
1222 return 1;
1223 }
1224
1225 /*
1226 * Decide whether a segment is worth dumping; default is yes to be
1227 * sure (missing info is worse than too much; etc).
1228 * Personally I'd include everything, and use the coredump limit...
1229 *
1230 * I think we should skip something. But I am not sure how. H.J.
1231 */
1232 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1233 {
1234 int dump_ok;
1235
1236 /* Do not dump I/O mapped devices or special mappings */
1237 if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1238 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1239 return 0;
1240 }
1241
1242 /* If we may not read the contents, don't allow us to dump
1243 * them either. "dump_write()" can't handle it anyway.
1244 */
1245 if (!(vma->vm_flags & VM_READ)) {
1246 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1247 return 0;
1248 }
1249
1250 /* By default, dump shared memory if mapped from an anonymous file. */
1251 if (vma->vm_flags & VM_SHARED) {
1252 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1253 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1254 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1255 vma->vm_flags, dump_ok ? "yes" : "no");
1256 return dump_ok;
1257 }
1258
1259 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1260 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1261 vma->vm_flags, dump_ok ? "yes" : "no");
1262 return dump_ok;
1263 }
1264
1265 #ifdef CONFIG_MMU
1266 /* By default, if it hasn't been written to, don't write it out */
1267 if (!vma->anon_vma) {
1268 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1269 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1270 vma->vm_flags, dump_ok ? "yes" : "no");
1271 return dump_ok;
1272 }
1273 #endif
1274
1275 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1276 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1277 dump_ok ? "yes" : "no");
1278 return dump_ok;
1279 }
1280
1281 /* An ELF note in memory */
1282 struct memelfnote
1283 {
1284 const char *name;
1285 int type;
1286 unsigned int datasz;
1287 void *data;
1288 };
1289
1290 static int notesize(struct memelfnote *en)
1291 {
1292 int sz;
1293
1294 sz = sizeof(struct elf_note);
1295 sz += roundup(strlen(en->name) + 1, 4);
1296 sz += roundup(en->datasz, 4);
1297
1298 return sz;
1299 }
1300
1301 /* #define DEBUG */
1302
1303 #define DUMP_WRITE(addr, nr) \
1304 do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1305 #define DUMP_SEEK(off) \
1306 do { if (!dump_seek(file, (off))) return 0; } while(0)
1307
1308 static int writenote(struct memelfnote *men, struct file *file)
1309 {
1310 struct elf_note en;
1311
1312 en.n_namesz = strlen(men->name) + 1;
1313 en.n_descsz = men->datasz;
1314 en.n_type = men->type;
1315
1316 DUMP_WRITE(&en, sizeof(en));
1317 DUMP_WRITE(men->name, en.n_namesz);
1318 /* XXX - cast from long long to long to avoid need for libgcc.a */
1319 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1320 DUMP_WRITE(men->data, men->datasz);
1321 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */
1322
1323 return 1;
1324 }
1325 #undef DUMP_WRITE
1326 #undef DUMP_SEEK
1327
1328 #define DUMP_WRITE(addr, nr) \
1329 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1330 goto end_coredump;
1331 #define DUMP_SEEK(off) \
1332 if (!dump_seek(file, (off))) \
1333 goto end_coredump;
1334
1335 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1336 {
1337 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1338 elf->e_ident[EI_CLASS] = ELF_CLASS;
1339 elf->e_ident[EI_DATA] = ELF_DATA;
1340 elf->e_ident[EI_VERSION] = EV_CURRENT;
1341 elf->e_ident[EI_OSABI] = ELF_OSABI;
1342 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1343
1344 elf->e_type = ET_CORE;
1345 elf->e_machine = ELF_ARCH;
1346 elf->e_version = EV_CURRENT;
1347 elf->e_entry = 0;
1348 elf->e_phoff = sizeof(struct elfhdr);
1349 elf->e_shoff = 0;
1350 elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1351 elf->e_ehsize = sizeof(struct elfhdr);
1352 elf->e_phentsize = sizeof(struct elf_phdr);
1353 elf->e_phnum = segs;
1354 elf->e_shentsize = 0;
1355 elf->e_shnum = 0;
1356 elf->e_shstrndx = 0;
1357 return;
1358 }
1359
1360 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1361 {
1362 phdr->p_type = PT_NOTE;
1363 phdr->p_offset = offset;
1364 phdr->p_vaddr = 0;
1365 phdr->p_paddr = 0;
1366 phdr->p_filesz = sz;
1367 phdr->p_memsz = 0;
1368 phdr->p_flags = 0;
1369 phdr->p_align = 0;
1370 return;
1371 }
1372
1373 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1374 unsigned int sz, void *data)
1375 {
1376 note->name = name;
1377 note->type = type;
1378 note->datasz = sz;
1379 note->data = data;
1380 return;
1381 }
1382
1383 /*
1384 * fill up all the fields in prstatus from the given task struct, except
1385 * registers which need to be filled up seperately.
1386 */
1387 static void fill_prstatus(struct elf_prstatus *prstatus,
1388 struct task_struct *p, long signr)
1389 {
1390 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1391 prstatus->pr_sigpend = p->pending.signal.sig[0];
1392 prstatus->pr_sighold = p->blocked.sig[0];
1393 rcu_read_lock();
1394 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1395 rcu_read_unlock();
1396 prstatus->pr_pid = task_pid_vnr(p);
1397 prstatus->pr_pgrp = task_pgrp_vnr(p);
1398 prstatus->pr_sid = task_session_vnr(p);
1399 if (thread_group_leader(p)) {
1400 struct task_cputime cputime;
1401
1402 /*
1403 * This is the record for the group leader. It shows the
1404 * group-wide total, not its individual thread total.
1405 */
1406 thread_group_cputime(p, &cputime);
1407 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1408 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1409 } else {
1410 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1411 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1412 }
1413 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1414 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1415
1416 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1417 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1418 }
1419
1420 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1421 struct mm_struct *mm)
1422 {
1423 const struct cred *cred;
1424 unsigned int i, len;
1425
1426 /* first copy the parameters from user space */
1427 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1428
1429 len = mm->arg_end - mm->arg_start;
1430 if (len >= ELF_PRARGSZ)
1431 len = ELF_PRARGSZ - 1;
1432 if (copy_from_user(&psinfo->pr_psargs,
1433 (const char __user *) mm->arg_start, len))
1434 return -EFAULT;
1435 for (i = 0; i < len; i++)
1436 if (psinfo->pr_psargs[i] == 0)
1437 psinfo->pr_psargs[i] = ' ';
1438 psinfo->pr_psargs[len] = 0;
1439
1440 rcu_read_lock();
1441 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1442 rcu_read_unlock();
1443 psinfo->pr_pid = task_pid_vnr(p);
1444 psinfo->pr_pgrp = task_pgrp_vnr(p);
1445 psinfo->pr_sid = task_session_vnr(p);
1446
1447 i = p->state ? ffz(~p->state) + 1 : 0;
1448 psinfo->pr_state = i;
1449 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1450 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1451 psinfo->pr_nice = task_nice(p);
1452 psinfo->pr_flag = p->flags;
1453 rcu_read_lock();
1454 cred = __task_cred(p);
1455 SET_UID(psinfo->pr_uid, cred->uid);
1456 SET_GID(psinfo->pr_gid, cred->gid);
1457 rcu_read_unlock();
1458 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1459
1460 return 0;
1461 }
1462
1463 /* Here is the structure in which status of each thread is captured. */
1464 struct elf_thread_status
1465 {
1466 struct list_head list;
1467 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1468 elf_fpregset_t fpu; /* NT_PRFPREG */
1469 struct task_struct *thread;
1470 #ifdef ELF_CORE_COPY_XFPREGS
1471 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1472 #endif
1473 struct memelfnote notes[3];
1474 int num_notes;
1475 };
1476
1477 /*
1478 * In order to add the specific thread information for the elf file format,
1479 * we need to keep a linked list of every thread's pr_status and then create
1480 * a single section for them in the final core file.
1481 */
1482 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1483 {
1484 struct task_struct *p = t->thread;
1485 int sz = 0;
1486
1487 t->num_notes = 0;
1488
1489 fill_prstatus(&t->prstatus, p, signr);
1490 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1491
1492 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1493 &t->prstatus);
1494 t->num_notes++;
1495 sz += notesize(&t->notes[0]);
1496
1497 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1498 if (t->prstatus.pr_fpvalid) {
1499 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1500 &t->fpu);
1501 t->num_notes++;
1502 sz += notesize(&t->notes[1]);
1503 }
1504
1505 #ifdef ELF_CORE_COPY_XFPREGS
1506 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1507 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1508 sizeof(t->xfpu), &t->xfpu);
1509 t->num_notes++;
1510 sz += notesize(&t->notes[2]);
1511 }
1512 #endif
1513 return sz;
1514 }
1515
1516 /*
1517 * dump the segments for an MMU process
1518 */
1519 #ifdef CONFIG_MMU
1520 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1521 unsigned long *limit, unsigned long mm_flags)
1522 {
1523 struct vm_area_struct *vma;
1524
1525 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1526 unsigned long addr;
1527
1528 if (!maydump(vma, mm_flags))
1529 continue;
1530
1531 for (addr = vma->vm_start;
1532 addr < vma->vm_end;
1533 addr += PAGE_SIZE
1534 ) {
1535 struct vm_area_struct *vma;
1536 struct page *page;
1537
1538 if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1539 &page, &vma) <= 0) {
1540 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1541 }
1542 else if (page == ZERO_PAGE(0)) {
1543 page_cache_release(page);
1544 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1545 }
1546 else {
1547 void *kaddr;
1548
1549 flush_cache_page(vma, addr, page_to_pfn(page));
1550 kaddr = kmap(page);
1551 if ((*size += PAGE_SIZE) > *limit ||
1552 !dump_write(file, kaddr, PAGE_SIZE)
1553 ) {
1554 kunmap(page);
1555 page_cache_release(page);
1556 return -EIO;
1557 }
1558 kunmap(page);
1559 page_cache_release(page);
1560 }
1561 }
1562 }
1563
1564 return 0;
1565
1566 end_coredump:
1567 return -EFBIG;
1568 }
1569 #endif
1570
1571 /*
1572 * dump the segments for a NOMMU process
1573 */
1574 #ifndef CONFIG_MMU
1575 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1576 unsigned long *limit, unsigned long mm_flags)
1577 {
1578 struct vm_area_struct *vma;
1579
1580 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1581 if (!maydump(vma, mm_flags))
1582 continue;
1583
1584 if ((*size += PAGE_SIZE) > *limit)
1585 return -EFBIG;
1586
1587 if (!dump_write(file, (void *) vma->vm_start,
1588 vma->vm_end - vma->vm_start))
1589 return -EIO;
1590 }
1591
1592 return 0;
1593 }
1594 #endif
1595
1596 /*
1597 * Actual dumper
1598 *
1599 * This is a two-pass process; first we find the offsets of the bits,
1600 * and then they are actually written out. If we run out of core limit
1601 * we just truncate.
1602 */
1603 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1604 struct file *file, unsigned long limit)
1605 {
1606 #define NUM_NOTES 6
1607 int has_dumped = 0;
1608 mm_segment_t fs;
1609 int segs;
1610 size_t size = 0;
1611 int i;
1612 struct vm_area_struct *vma;
1613 struct elfhdr *elf = NULL;
1614 loff_t offset = 0, dataoff;
1615 int numnote;
1616 struct memelfnote *notes = NULL;
1617 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1618 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1619 LIST_HEAD(thread_list);
1620 struct list_head *t;
1621 elf_fpregset_t *fpu = NULL;
1622 #ifdef ELF_CORE_COPY_XFPREGS
1623 elf_fpxregset_t *xfpu = NULL;
1624 #endif
1625 int thread_status_size = 0;
1626 elf_addr_t *auxv;
1627 unsigned long mm_flags;
1628
1629 /*
1630 * We no longer stop all VM operations.
1631 *
1632 * This is because those proceses that could possibly change map_count
1633 * or the mmap / vma pages are now blocked in do_exit on current
1634 * finishing this core dump.
1635 *
1636 * Only ptrace can touch these memory addresses, but it doesn't change
1637 * the map_count or the pages allocated. So no possibility of crashing
1638 * exists while dumping the mm->vm_next areas to the core file.
1639 */
1640
1641 /* alloc memory for large data structures: too large to be on stack */
1642 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1643 if (!elf)
1644 goto cleanup;
1645 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1646 if (!prstatus)
1647 goto cleanup;
1648 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1649 if (!psinfo)
1650 goto cleanup;
1651 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1652 if (!notes)
1653 goto cleanup;
1654 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1655 if (!fpu)
1656 goto cleanup;
1657 #ifdef ELF_CORE_COPY_XFPREGS
1658 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1659 if (!xfpu)
1660 goto cleanup;
1661 #endif
1662
1663 if (signr) {
1664 struct core_thread *ct;
1665 struct elf_thread_status *tmp;
1666
1667 for (ct = current->mm->core_state->dumper.next;
1668 ct; ct = ct->next) {
1669 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1670 if (!tmp)
1671 goto cleanup;
1672
1673 tmp->thread = ct->task;
1674 list_add(&tmp->list, &thread_list);
1675 }
1676
1677 list_for_each(t, &thread_list) {
1678 struct elf_thread_status *tmp;
1679 int sz;
1680
1681 tmp = list_entry(t, struct elf_thread_status, list);
1682 sz = elf_dump_thread_status(signr, tmp);
1683 thread_status_size += sz;
1684 }
1685 }
1686
1687 /* now collect the dump for the current */
1688 fill_prstatus(prstatus, current, signr);
1689 elf_core_copy_regs(&prstatus->pr_reg, regs);
1690
1691 segs = current->mm->map_count;
1692 #ifdef ELF_CORE_EXTRA_PHDRS
1693 segs += ELF_CORE_EXTRA_PHDRS;
1694 #endif
1695
1696 /* Set up header */
1697 fill_elf_fdpic_header(elf, segs + 1); /* including notes section */
1698
1699 has_dumped = 1;
1700 current->flags |= PF_DUMPCORE;
1701
1702 /*
1703 * Set up the notes in similar form to SVR4 core dumps made
1704 * with info from their /proc.
1705 */
1706
1707 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1708 fill_psinfo(psinfo, current->group_leader, current->mm);
1709 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1710
1711 numnote = 2;
1712
1713 auxv = (elf_addr_t *) current->mm->saved_auxv;
1714
1715 i = 0;
1716 do
1717 i += 2;
1718 while (auxv[i - 2] != AT_NULL);
1719 fill_note(¬es[numnote++], "CORE", NT_AUXV,
1720 i * sizeof(elf_addr_t), auxv);
1721
1722 /* Try to dump the FPU. */
1723 if ((prstatus->pr_fpvalid =
1724 elf_core_copy_task_fpregs(current, regs, fpu)))
1725 fill_note(notes + numnote++,
1726 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1727 #ifdef ELF_CORE_COPY_XFPREGS
1728 if (elf_core_copy_task_xfpregs(current, xfpu))
1729 fill_note(notes + numnote++,
1730 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1731 #endif
1732
1733 fs = get_fs();
1734 set_fs(KERNEL_DS);
1735
1736 DUMP_WRITE(elf, sizeof(*elf));
1737 offset += sizeof(*elf); /* Elf header */
1738 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers */
1739
1740 /* Write notes phdr entry */
1741 {
1742 struct elf_phdr phdr;
1743 int sz = 0;
1744
1745 for (i = 0; i < numnote; i++)
1746 sz += notesize(notes + i);
1747
1748 sz += thread_status_size;
1749
1750 fill_elf_note_phdr(&phdr, sz, offset);
1751 offset += sz;
1752 DUMP_WRITE(&phdr, sizeof(phdr));
1753 }
1754
1755 /* Page-align dumped data */
1756 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1757
1758 /*
1759 * We must use the same mm->flags while dumping core to avoid
1760 * inconsistency between the program headers and bodies, otherwise an
1761 * unusable core file can be generated.
1762 */
1763 mm_flags = current->mm->flags;
1764
1765 /* write program headers for segments dump */
1766 for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1767 struct elf_phdr phdr;
1768 size_t sz;
1769
1770 sz = vma->vm_end - vma->vm_start;
1771
1772 phdr.p_type = PT_LOAD;
1773 phdr.p_offset = offset;
1774 phdr.p_vaddr = vma->vm_start;
1775 phdr.p_paddr = 0;
1776 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1777 phdr.p_memsz = sz;
1778 offset += phdr.p_filesz;
1779 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1780 if (vma->vm_flags & VM_WRITE)
1781 phdr.p_flags |= PF_W;
1782 if (vma->vm_flags & VM_EXEC)
1783 phdr.p_flags |= PF_X;
1784 phdr.p_align = ELF_EXEC_PAGESIZE;
1785
1786 DUMP_WRITE(&phdr, sizeof(phdr));
1787 }
1788
1789 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1790 ELF_CORE_WRITE_EXTRA_PHDRS;
1791 #endif
1792
1793 /* write out the notes section */
1794 for (i = 0; i < numnote; i++)
1795 if (!writenote(notes + i, file))
1796 goto end_coredump;
1797
1798 /* write out the thread status notes section */
1799 list_for_each(t, &thread_list) {
1800 struct elf_thread_status *tmp =
1801 list_entry(t, struct elf_thread_status, list);
1802
1803 for (i = 0; i < tmp->num_notes; i++)
1804 if (!writenote(&tmp->notes[i], file))
1805 goto end_coredump;
1806 }
1807
1808 DUMP_SEEK(dataoff);
1809
1810 if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1811 goto end_coredump;
1812
1813 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1814 ELF_CORE_WRITE_EXTRA_DATA;
1815 #endif
1816
1817 if (file->f_pos != offset) {
1818 /* Sanity check */
1819 printk(KERN_WARNING
1820 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1821 file->f_pos, offset);
1822 }
1823
1824 end_coredump:
1825 set_fs(fs);
1826
1827 cleanup:
1828 while (!list_empty(&thread_list)) {
1829 struct list_head *tmp = thread_list.next;
1830 list_del(tmp);
1831 kfree(list_entry(tmp, struct elf_thread_status, list));
1832 }
1833
1834 kfree(elf);
1835 kfree(prstatus);
1836 kfree(psinfo);
1837 kfree(notes);
1838 kfree(fpu);
1839 #ifdef ELF_CORE_COPY_XFPREGS
1840 kfree(xfpu);
1841 #endif
1842 return has_dumped;
1843 #undef NUM_NOTES
1844 }
1845
1846 #endif /* USE_ELF_CORE_DUMP */
1847
|
This page was automatically generated by the
LXR engine.
|