e4eb472558ca33f5944f3285b79588de239c7d6f
[linux-2.6.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/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/elf.h>
34 #include <linux/elf-fdpic.h>
35 #include <linux/elfcore.h>
36
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/pgalloc.h>
40
41 typedef char *elf_caddr_t;
42
43 #if 0
44 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
45 #else
46 #define kdebug(fmt, ...) do {} while(0)
47 #endif
48
49 #if 0
50 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
51 #else
52 #define kdcore(fmt, ...) do {} while(0)
53 #endif
54
55 MODULE_LICENSE("GPL");
56
57 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
58 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
59 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
60                               struct mm_struct *, const char *);
61
62 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
63                                    struct elf_fdpic_params *,
64                                    struct elf_fdpic_params *);
65
66 #ifndef CONFIG_MMU
67 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
68                                             unsigned long *);
69 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
70                                                    struct file *,
71                                                    struct mm_struct *);
72 #endif
73
74 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
75                                              struct file *, struct mm_struct *);
76
77 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
78 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *, unsigned long limit);
79 #endif
80
81 static struct linux_binfmt elf_fdpic_format = {
82         .module         = THIS_MODULE,
83         .load_binary    = load_elf_fdpic_binary,
84 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
85         .core_dump      = elf_fdpic_core_dump,
86 #endif
87         .min_coredump   = ELF_EXEC_PAGESIZE,
88 };
89
90 static int __init init_elf_fdpic_binfmt(void)
91 {
92         return register_binfmt(&elf_fdpic_format);
93 }
94
95 static void __exit exit_elf_fdpic_binfmt(void)
96 {
97         unregister_binfmt(&elf_fdpic_format);
98 }
99
100 core_initcall(init_elf_fdpic_binfmt);
101 module_exit(exit_elf_fdpic_binfmt);
102
103 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
104 {
105         if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
106                 return 0;
107         if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
108                 return 0;
109         if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
110                 return 0;
111         if (!file->f_op || !file->f_op->mmap)
112                 return 0;
113         return 1;
114 }
115
116 /*****************************************************************************/
117 /*
118  * read the program headers table into memory
119  */
120 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
121                                  struct file *file)
122 {
123         struct elf32_phdr *phdr;
124         unsigned long size;
125         int retval, loop;
126
127         if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
128                 return -ENOMEM;
129         if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
130                 return -ENOMEM;
131
132         size = params->hdr.e_phnum * sizeof(struct elf_phdr);
133         params->phdrs = kmalloc(size, GFP_KERNEL);
134         if (!params->phdrs)
135                 return -ENOMEM;
136
137         retval = kernel_read(file, params->hdr.e_phoff,
138                              (char *) params->phdrs, size);
139         if (unlikely(retval != size))
140                 return retval < 0 ? retval : -ENOEXEC;
141
142         /* determine stack size for this binary */
143         phdr = params->phdrs;
144         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
145                 if (phdr->p_type != PT_GNU_STACK)
146                         continue;
147
148                 if (phdr->p_flags & PF_X)
149                         params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
150                 else
151                         params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
152
153                 params->stack_size = phdr->p_memsz;
154                 break;
155         }
156
157         return 0;
158 }
159
160 /*****************************************************************************/
161 /*
162  * load an fdpic binary into various bits of memory
163  */
164 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
165                                  struct pt_regs *regs)
166 {
167         struct elf_fdpic_params exec_params, interp_params;
168         struct elf_phdr *phdr;
169         unsigned long stack_size, entryaddr;
170 #ifndef CONFIG_MMU
171         unsigned long fullsize;
172 #endif
173 #ifdef ELF_FDPIC_PLAT_INIT
174         unsigned long dynaddr;
175 #endif
176         struct file *interpreter = NULL; /* to shut gcc up */
177         char *interpreter_name = NULL;
178         int executable_stack;
179         int retval, i;
180
181         kdebug("____ LOAD %d ____", current->pid);
182
183         memset(&exec_params, 0, sizeof(exec_params));
184         memset(&interp_params, 0, sizeof(interp_params));
185
186         exec_params.hdr = *(struct elfhdr *) bprm->buf;
187         exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
188
189         /* check that this is a binary we know how to deal with */
190         retval = -ENOEXEC;
191         if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
192                 goto error;
193
194         /* read the program header table */
195         retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
196         if (retval < 0)
197                 goto error;
198
199         /* scan for a program header that specifies an interpreter */
200         phdr = exec_params.phdrs;
201
202         for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
203                 switch (phdr->p_type) {
204                 case PT_INTERP:
205                         retval = -ENOMEM;
206                         if (phdr->p_filesz > PATH_MAX)
207                                 goto error;
208                         retval = -ENOENT;
209                         if (phdr->p_filesz < 2)
210                                 goto error;
211
212                         /* read the name of the interpreter into memory */
213                         interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
214                         if (!interpreter_name)
215                                 goto error;
216
217                         retval = kernel_read(bprm->file,
218                                              phdr->p_offset,
219                                              interpreter_name,
220                                              phdr->p_filesz);
221                         if (unlikely(retval != phdr->p_filesz)) {
222                                 if (retval >= 0)
223                                         retval = -ENOEXEC;
224                                 goto error;
225                         }
226
227                         retval = -ENOENT;
228                         if (interpreter_name[phdr->p_filesz - 1] != '\0')
229                                 goto error;
230
231                         kdebug("Using ELF interpreter %s", interpreter_name);
232
233                         /* replace the program with the interpreter */
234                         interpreter = open_exec(interpreter_name);
235                         retval = PTR_ERR(interpreter);
236                         if (IS_ERR(interpreter)) {
237                                 interpreter = NULL;
238                                 goto error;
239                         }
240
241                         /*
242                          * If the binary is not readable then enforce
243                          * mm->dumpable = 0 regardless of the interpreter's
244                          * permissions.
245                          */
246                         if (file_permission(interpreter, MAY_READ) < 0)
247                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
248
249                         retval = kernel_read(interpreter, 0, bprm->buf,
250                                              BINPRM_BUF_SIZE);
251                         if (unlikely(retval != BINPRM_BUF_SIZE)) {
252                                 if (retval >= 0)
253                                         retval = -ENOEXEC;
254                                 goto error;
255                         }
256
257                         interp_params.hdr = *((struct elfhdr *) bprm->buf);
258                         break;
259
260                 case PT_LOAD:
261 #ifdef CONFIG_MMU
262                         if (exec_params.load_addr == 0)
263                                 exec_params.load_addr = phdr->p_vaddr;
264 #endif
265                         break;
266                 }
267
268         }
269
270         if (elf_check_const_displacement(&exec_params.hdr))
271                 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
272
273         /* perform insanity checks on the interpreter */
274         if (interpreter_name) {
275                 retval = -ELIBBAD;
276                 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
277                         goto error;
278
279                 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
280
281                 /* read the interpreter's program header table */
282                 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
283                 if (retval < 0)
284                         goto error;
285         }
286
287         stack_size = exec_params.stack_size;
288         if (stack_size < interp_params.stack_size)
289                 stack_size = interp_params.stack_size;
290
291         if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
292                 executable_stack = EXSTACK_ENABLE_X;
293         else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
294                 executable_stack = EXSTACK_DISABLE_X;
295         else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
296                 executable_stack = EXSTACK_ENABLE_X;
297         else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
298                 executable_stack = EXSTACK_DISABLE_X;
299         else
300                 executable_stack = EXSTACK_DEFAULT;
301
302         retval = -ENOEXEC;
303         if (stack_size == 0)
304                 goto error;
305
306         if (elf_check_const_displacement(&interp_params.hdr))
307                 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
308
309         /* flush all traces of the currently running executable */
310         retval = flush_old_exec(bprm);
311         if (retval)
312                 goto error;
313
314         /* there's now no turning back... the old userspace image is dead,
315          * defunct, deceased, etc. after this point we have to exit via
316          * error_kill */
317         set_personality(PER_LINUX_FDPIC);
318         set_binfmt(&elf_fdpic_format);
319
320         current->mm->start_code = 0;
321         current->mm->end_code = 0;
322         current->mm->start_stack = 0;
323         current->mm->start_data = 0;
324         current->mm->end_data = 0;
325         current->mm->context.exec_fdpic_loadmap = 0;
326         current->mm->context.interp_fdpic_loadmap = 0;
327
328         current->flags &= ~PF_FORKNOEXEC;
329
330 #ifdef CONFIG_MMU
331         elf_fdpic_arch_lay_out_mm(&exec_params,
332                                   &interp_params,
333                                   &current->mm->start_stack,
334                                   &current->mm->start_brk);
335
336         retval = setup_arg_pages(bprm, current->mm->start_stack,
337                                  executable_stack);
338         if (retval < 0) {
339                 send_sig(SIGKILL, current, 0);
340                 goto error_kill;
341         }
342 #endif
343
344         /* load the executable and interpreter into memory */
345         retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
346                                     "executable");
347         if (retval < 0)
348                 goto error_kill;
349
350         if (interpreter_name) {
351                 retval = elf_fdpic_map_file(&interp_params, interpreter,
352                                             current->mm, "interpreter");
353                 if (retval < 0) {
354                         printk(KERN_ERR "Unable to load interpreter\n");
355                         goto error_kill;
356                 }
357
358                 allow_write_access(interpreter);
359                 fput(interpreter);
360                 interpreter = NULL;
361         }
362
363 #ifdef CONFIG_MMU
364         if (!current->mm->start_brk)
365                 current->mm->start_brk = current->mm->end_data;
366
367         current->mm->brk = current->mm->start_brk =
368                 PAGE_ALIGN(current->mm->start_brk);
369
370 #else
371         /* create a stack and brk area big enough for everyone
372          * - the brk heap starts at the bottom and works up
373          * - the stack starts at the top and works down
374          */
375         stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
376         if (stack_size < PAGE_SIZE * 2)
377                 stack_size = PAGE_SIZE * 2;
378
379         down_write(&current->mm->mmap_sem);
380         current->mm->start_brk = do_mmap(NULL, 0, stack_size,
381                                          PROT_READ | PROT_WRITE | PROT_EXEC,
382                                          MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
383                                          0);
384
385         if (IS_ERR_VALUE(current->mm->start_brk)) {
386                 up_write(&current->mm->mmap_sem);
387                 retval = current->mm->start_brk;
388                 current->mm->start_brk = 0;
389                 goto error_kill;
390         }
391
392         /* expand the stack mapping to use up the entire allocation granule */
393         fullsize = kobjsize((char *) current->mm->start_brk);
394         if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
395                                     fullsize, 0, 0)))
396                 stack_size = fullsize;
397         up_write(&current->mm->mmap_sem);
398
399         current->mm->brk = current->mm->start_brk;
400         current->mm->context.end_brk = current->mm->start_brk;
401         current->mm->context.end_brk +=
402                 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
403         current->mm->start_stack = current->mm->start_brk + stack_size;
404 #endif
405
406         compute_creds(bprm);
407         current->flags &= ~PF_FORKNOEXEC;
408         if (create_elf_fdpic_tables(bprm, current->mm,
409                                     &exec_params, &interp_params) < 0)
410                 goto error_kill;
411
412         kdebug("- start_code  %lx", current->mm->start_code);
413         kdebug("- end_code    %lx", current->mm->end_code);
414         kdebug("- start_data  %lx", current->mm->start_data);
415         kdebug("- end_data    %lx", current->mm->end_data);
416         kdebug("- start_brk   %lx", current->mm->start_brk);
417         kdebug("- brk         %lx", current->mm->brk);
418         kdebug("- start_stack %lx", current->mm->start_stack);
419
420 #ifdef ELF_FDPIC_PLAT_INIT
421         /*
422          * The ABI may specify that certain registers be set up in special
423          * ways (on i386 %edx is the address of a DT_FINI function, for
424          * example.  This macro performs whatever initialization to
425          * the regs structure is required.
426          */
427         dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
428         ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
429                             dynaddr);
430 #endif
431
432         /* everything is now ready... get the userspace context ready to roll */
433         entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
434         start_thread(regs, entryaddr, current->mm->start_stack);
435
436         retval = 0;
437
438 error:
439         if (interpreter) {
440                 allow_write_access(interpreter);
441                 fput(interpreter);
442         }
443         kfree(interpreter_name);
444         kfree(exec_params.phdrs);
445         kfree(exec_params.loadmap);
446         kfree(interp_params.phdrs);
447         kfree(interp_params.loadmap);
448         return retval;
449
450         /* unrecoverable error - kill the process */
451 error_kill:
452         send_sig(SIGSEGV, current, 0);
453         goto error;
454
455 }
456
457 /*****************************************************************************/
458
459 #ifndef ELF_BASE_PLATFORM
460 /*
461  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
462  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
463  * will be copied to the user stack in the same manner as AT_PLATFORM.
464  */
465 #define ELF_BASE_PLATFORM NULL
466 #endif
467
468 /*
469  * present useful information to the program
470  */
471 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
472                                    struct mm_struct *mm,
473                                    struct elf_fdpic_params *exec_params,
474                                    struct elf_fdpic_params *interp_params)
475 {
476         unsigned long sp, csp, nitems;
477         elf_caddr_t __user *argv, *envp;
478         size_t platform_len = 0, len;
479         char *k_platform, *k_base_platform;
480         char __user *u_platform, *u_base_platform, *p;
481         long hwcap;
482         int loop;
483         int nr; /* reset for each csp adjustment */
484
485         /* we're going to shovel a whole load of stuff onto the stack */
486 #ifdef CONFIG_MMU
487         sp = bprm->p;
488 #else
489         sp = mm->start_stack;
490
491         /* stack the program arguments and environment */
492         if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
493                 return -EFAULT;
494 #endif
495
496         hwcap = ELF_HWCAP;
497
498         /*
499          * If this architecture has a platform capability string, copy it
500          * to userspace.  In some cases (Sparc), this info is impossible
501          * for userspace to get any other way, in others (i386) it is
502          * merely difficult.
503          */
504         k_platform = ELF_PLATFORM;
505         u_platform = NULL;
506
507         if (k_platform) {
508                 platform_len = strlen(k_platform) + 1;
509                 sp -= platform_len;
510                 u_platform = (char __user *) sp;
511                 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
512                         return -EFAULT;
513         }
514
515         /*
516          * If this architecture has a "base" platform capability
517          * string, copy it to userspace.
518          */
519         k_base_platform = ELF_BASE_PLATFORM;
520         u_base_platform = NULL;
521
522         if (k_base_platform) {
523                 platform_len = strlen(k_base_platform) + 1;
524                 sp -= platform_len;
525                 u_base_platform = (char __user *) sp;
526                 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
527                         return -EFAULT;
528         }
529
530 #if defined(__i386__) && defined(CONFIG_SMP)
531         /* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
532          * by the processes running on the same package. One thing we can do is
533          * to shuffle the initial stack for them.
534          *
535          * the conditionals here are unneeded, but kept in to make the code
536          * behaviour the same as pre change unless we have hyperthreaded
537          * processors. This keeps Mr Marcelo Person happier but should be
538          * removed for 2.5
539          */
540         if (smp_num_siblings > 1)
541                 sp = sp - ((current->pid % 64) << 7);
542 #endif
543
544         sp &= ~7UL;
545
546         /* stack the load map(s) */
547         len = sizeof(struct elf32_fdpic_loadmap);
548         len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
549         sp = (sp - len) & ~7UL;
550         exec_params->map_addr = sp;
551
552         if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
553                 return -EFAULT;
554
555         current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
556
557         if (interp_params->loadmap) {
558                 len = sizeof(struct elf32_fdpic_loadmap);
559                 len += sizeof(struct elf32_fdpic_loadseg) *
560                         interp_params->loadmap->nsegs;
561                 sp = (sp - len) & ~7UL;
562                 interp_params->map_addr = sp;
563
564                 if (copy_to_user((void __user *) sp, interp_params->loadmap,
565                                  len) != 0)
566                         return -EFAULT;
567
568                 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
569         }
570
571         /* force 16 byte _final_ alignment here for generality */
572 #define DLINFO_ITEMS 13
573
574         nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
575                 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
576
577         csp = sp;
578         sp -= nitems * 2 * sizeof(unsigned long);
579         sp -= (bprm->envc + 1) * sizeof(char *);        /* envv[] */
580         sp -= (bprm->argc + 1) * sizeof(char *);        /* argv[] */
581         sp -= 1 * sizeof(unsigned long);                /* argc */
582
583         csp -= sp & 15UL;
584         sp -= sp & 15UL;
585
586         /* put the ELF interpreter info on the stack */
587 #define NEW_AUX_ENT(id, val)                                            \
588         do {                                                            \
589                 struct { unsigned long _id, _val; } __user *ent;        \
590                                                                         \
591                 ent = (void __user *) csp;                              \
592                 __put_user((id), &ent[nr]._id);                         \
593                 __put_user((val), &ent[nr]._val);                       \
594                 nr++;                                                   \
595         } while (0)
596
597         nr = 0;
598         csp -= 2 * sizeof(unsigned long);
599         NEW_AUX_ENT(AT_NULL, 0);
600         if (k_platform) {
601                 nr = 0;
602                 csp -= 2 * sizeof(unsigned long);
603                 NEW_AUX_ENT(AT_PLATFORM,
604                             (elf_addr_t) (unsigned long) u_platform);
605         }
606
607         if (k_base_platform) {
608                 nr = 0;
609                 csp -= 2 * sizeof(unsigned long);
610                 NEW_AUX_ENT(AT_BASE_PLATFORM,
611                             (elf_addr_t) (unsigned long) u_base_platform);
612         }
613
614         nr = 0;
615         csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
616         NEW_AUX_ENT(AT_HWCAP,   hwcap);
617         NEW_AUX_ENT(AT_PAGESZ,  PAGE_SIZE);
618         NEW_AUX_ENT(AT_CLKTCK,  CLOCKS_PER_SEC);
619         NEW_AUX_ENT(AT_PHDR,    exec_params->ph_addr);
620         NEW_AUX_ENT(AT_PHENT,   sizeof(struct elf_phdr));
621         NEW_AUX_ENT(AT_PHNUM,   exec_params->hdr.e_phnum);
622         NEW_AUX_ENT(AT_BASE,    interp_params->elfhdr_addr);
623         NEW_AUX_ENT(AT_FLAGS,   0);
624         NEW_AUX_ENT(AT_ENTRY,   exec_params->entry_addr);
625         NEW_AUX_ENT(AT_UID,     (elf_addr_t) current->uid);
626         NEW_AUX_ENT(AT_EUID,    (elf_addr_t) current->euid);
627         NEW_AUX_ENT(AT_GID,     (elf_addr_t) current->gid);
628         NEW_AUX_ENT(AT_EGID,    (elf_addr_t) current->egid);
629
630 #ifdef ARCH_DLINFO
631         nr = 0;
632         csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
633
634         /* ARCH_DLINFO must come last so platform specific code can enforce
635          * special alignment requirements on the AUXV if necessary (eg. PPC).
636          */
637         ARCH_DLINFO;
638 #endif
639 #undef NEW_AUX_ENT
640
641         /* allocate room for argv[] and envv[] */
642         csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
643         envp = (elf_caddr_t __user *) csp;
644         csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
645         argv = (elf_caddr_t __user *) csp;
646
647         /* stack argc */
648         csp -= sizeof(unsigned long);
649         __put_user(bprm->argc, (unsigned long __user *) csp);
650
651         BUG_ON(csp != sp);
652
653         /* fill in the argv[] array */
654 #ifdef CONFIG_MMU
655         current->mm->arg_start = bprm->p;
656 #else
657         current->mm->arg_start = current->mm->start_stack -
658                 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
659 #endif
660
661         p = (char __user *) current->mm->arg_start;
662         for (loop = bprm->argc; loop > 0; loop--) {
663                 __put_user((elf_caddr_t) p, argv++);
664                 len = strnlen_user(p, MAX_ARG_STRLEN);
665                 if (!len || len > MAX_ARG_STRLEN)
666                         return -EINVAL;
667                 p += len;
668         }
669         __put_user(NULL, argv);
670         current->mm->arg_end = (unsigned long) p;
671
672         /* fill in the envv[] array */
673         current->mm->env_start = (unsigned long) p;
674         for (loop = bprm->envc; loop > 0; loop--) {
675                 __put_user((elf_caddr_t)(unsigned long) p, envp++);
676                 len = strnlen_user(p, MAX_ARG_STRLEN);
677                 if (!len || len > MAX_ARG_STRLEN)
678                         return -EINVAL;
679                 p += len;
680         }
681         __put_user(NULL, envp);
682         current->mm->env_end = (unsigned long) p;
683
684         mm->start_stack = (unsigned long) sp;
685         return 0;
686 }
687
688 /*****************************************************************************/
689 /*
690  * transfer the program arguments and environment from the holding pages onto
691  * the stack
692  */
693 #ifndef CONFIG_MMU
694 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
695                                             unsigned long *_sp)
696 {
697         unsigned long index, stop, sp;
698         char *src;
699         int ret = 0;
700
701         stop = bprm->p >> PAGE_SHIFT;
702         sp = *_sp;
703
704         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
705                 src = kmap(bprm->page[index]);
706                 sp -= PAGE_SIZE;
707                 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
708                         ret = -EFAULT;
709                 kunmap(bprm->page[index]);
710                 if (ret < 0)
711                         goto out;
712         }
713
714         *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
715
716 out:
717         return ret;
718 }
719 #endif
720
721 /*****************************************************************************/
722 /*
723  * load the appropriate binary image (executable or interpreter) into memory
724  * - we assume no MMU is available
725  * - if no other PIC bits are set in params->hdr->e_flags
726  *   - we assume that the LOADable segments in the binary are independently relocatable
727  *   - we assume R/O executable segments are shareable
728  * - else
729  *   - we assume the loadable parts of the image to require fixed displacement
730  *   - the image is not shareable
731  */
732 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
733                               struct file *file,
734                               struct mm_struct *mm,
735                               const char *what)
736 {
737         struct elf32_fdpic_loadmap *loadmap;
738 #ifdef CONFIG_MMU
739         struct elf32_fdpic_loadseg *mseg;
740 #endif
741         struct elf32_fdpic_loadseg *seg;
742         struct elf32_phdr *phdr;
743         unsigned long load_addr, stop;
744         unsigned nloads, tmp;
745         size_t size;
746         int loop, ret;
747
748         /* allocate a load map table */
749         nloads = 0;
750         for (loop = 0; loop < params->hdr.e_phnum; loop++)
751                 if (params->phdrs[loop].p_type == PT_LOAD)
752                         nloads++;
753
754         if (nloads == 0)
755                 return -ELIBBAD;
756
757         size = sizeof(*loadmap) + nloads * sizeof(*seg);
758         loadmap = kzalloc(size, GFP_KERNEL);
759         if (!loadmap)
760                 return -ENOMEM;
761
762         params->loadmap = loadmap;
763
764         loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
765         loadmap->nsegs = nloads;
766
767         load_addr = params->load_addr;
768         seg = loadmap->segs;
769
770         /* map the requested LOADs into the memory space */
771         switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
772         case ELF_FDPIC_FLAG_CONSTDISP:
773         case ELF_FDPIC_FLAG_CONTIGUOUS:
774 #ifndef CONFIG_MMU
775                 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
776                 if (ret < 0)
777                         return ret;
778                 break;
779 #endif
780         default:
781                 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
782                 if (ret < 0)
783                         return ret;
784                 break;
785         }
786
787         /* map the entry point */
788         if (params->hdr.e_entry) {
789                 seg = loadmap->segs;
790                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
791                         if (params->hdr.e_entry >= seg->p_vaddr &&
792                             params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
793                                 params->entry_addr =
794                                         (params->hdr.e_entry - seg->p_vaddr) +
795                                         seg->addr;
796                                 break;
797                         }
798                 }
799         }
800
801         /* determine where the program header table has wound up if mapped */
802         stop = params->hdr.e_phoff;
803         stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
804         phdr = params->phdrs;
805
806         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
807                 if (phdr->p_type != PT_LOAD)
808                         continue;
809
810                 if (phdr->p_offset > params->hdr.e_phoff ||
811                     phdr->p_offset + phdr->p_filesz < stop)
812                         continue;
813
814                 seg = loadmap->segs;
815                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
816                         if (phdr->p_vaddr >= seg->p_vaddr &&
817                             phdr->p_vaddr + phdr->p_filesz <=
818                             seg->p_vaddr + seg->p_memsz) {
819                                 params->ph_addr =
820                                         (phdr->p_vaddr - seg->p_vaddr) +
821                                         seg->addr +
822                                         params->hdr.e_phoff - phdr->p_offset;
823                                 break;
824                         }
825                 }
826                 break;
827         }
828
829         /* determine where the dynamic section has wound up if there is one */
830         phdr = params->phdrs;
831         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
832                 if (phdr->p_type != PT_DYNAMIC)
833                         continue;
834
835                 seg = loadmap->segs;
836                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
837                         if (phdr->p_vaddr >= seg->p_vaddr &&
838                             phdr->p_vaddr + phdr->p_memsz <=
839                             seg->p_vaddr + seg->p_memsz) {
840                                 params->dynamic_addr =
841                                         (phdr->p_vaddr - seg->p_vaddr) +
842                                         seg->addr;
843
844                                 /* check the dynamic section contains at least
845                                  * one item, and that the last item is a NULL
846                                  * entry */
847                                 if (phdr->p_memsz == 0 ||
848                                     phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
849                                         goto dynamic_error;
850
851                                 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
852                                 if (((Elf32_Dyn *)
853                                      params->dynamic_addr)[tmp - 1].d_tag != 0)
854                                         goto dynamic_error;
855                                 break;
856                         }
857                 }
858                 break;
859         }
860
861         /* now elide adjacent segments in the load map on MMU linux
862          * - on uClinux the holes between may actually be filled with system
863          *   stuff or stuff from other processes
864          */
865 #ifdef CONFIG_MMU
866         nloads = loadmap->nsegs;
867         mseg = loadmap->segs;
868         seg = mseg + 1;
869         for (loop = 1; loop < nloads; loop++) {
870                 /* see if we have a candidate for merging */
871                 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
872                         load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
873                         if (load_addr == (seg->addr & PAGE_MASK)) {
874                                 mseg->p_memsz +=
875                                         load_addr -
876                                         (mseg->addr + mseg->p_memsz);
877                                 mseg->p_memsz += seg->addr & ~PAGE_MASK;
878                                 mseg->p_memsz += seg->p_memsz;
879                                 loadmap->nsegs--;
880                                 continue;
881                         }
882                 }
883
884                 mseg++;
885                 if (mseg != seg)
886                         *mseg = *seg;
887         }
888 #endif
889
890         kdebug("Mapped Object [%s]:", what);
891         kdebug("- elfhdr   : %lx", params->elfhdr_addr);
892         kdebug("- entry    : %lx", params->entry_addr);
893         kdebug("- PHDR[]   : %lx", params->ph_addr);
894         kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
895         seg = loadmap->segs;
896         for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
897                 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
898                        loop,
899                        seg->addr, seg->addr + seg->p_memsz - 1,
900                        seg->p_vaddr, seg->p_memsz);
901
902         return 0;
903
904 dynamic_error:
905         printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
906                what, file->f_path.dentry->d_inode->i_ino);
907         return -ELIBBAD;
908 }
909
910 /*****************************************************************************/
911 /*
912  * map a file with constant displacement under uClinux
913  */
914 #ifndef CONFIG_MMU
915 static int elf_fdpic_map_file_constdisp_on_uclinux(
916         struct elf_fdpic_params *params,
917         struct file *file,
918         struct mm_struct *mm)
919 {
920         struct elf32_fdpic_loadseg *seg;
921         struct elf32_phdr *phdr;
922         unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
923         loff_t fpos;
924         int loop, ret;
925
926         load_addr = params->load_addr;
927         seg = params->loadmap->segs;
928
929         /* determine the bounds of the contiguous overall allocation we must
930          * make */
931         phdr = params->phdrs;
932         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
933                 if (params->phdrs[loop].p_type != PT_LOAD)
934                         continue;
935
936                 if (base > phdr->p_vaddr)
937                         base = phdr->p_vaddr;
938                 if (top < phdr->p_vaddr + phdr->p_memsz)
939                         top = phdr->p_vaddr + phdr->p_memsz;
940         }
941
942         /* allocate one big anon block for everything */
943         mflags = MAP_PRIVATE;
944         if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
945                 mflags |= MAP_EXECUTABLE;
946
947         down_write(&mm->mmap_sem);
948         maddr = do_mmap(NULL, load_addr, top - base,
949                         PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
950         up_write(&mm->mmap_sem);
951         if (IS_ERR_VALUE(maddr))
952                 return (int) maddr;
953
954         if (load_addr != 0)
955                 load_addr += PAGE_ALIGN(top - base);
956
957         /* and then load the file segments into it */
958         phdr = params->phdrs;
959         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
960                 if (params->phdrs[loop].p_type != PT_LOAD)
961                         continue;
962
963                 fpos = phdr->p_offset;
964
965                 seg->addr = maddr + (phdr->p_vaddr - base);
966                 seg->p_vaddr = phdr->p_vaddr;
967                 seg->p_memsz = phdr->p_memsz;
968
969                 ret = file->f_op->read(file, (void *) seg->addr,
970                                        phdr->p_filesz, &fpos);
971                 if (ret < 0)
972                         return ret;
973
974                 /* map the ELF header address if in this segment */
975                 if (phdr->p_offset == 0)
976                         params->elfhdr_addr = seg->addr;
977
978                 /* clear any space allocated but not loaded */
979                 if (phdr->p_filesz < phdr->p_memsz)
980                         clear_user((void *) (seg->addr + phdr->p_filesz),
981                                    phdr->p_memsz - phdr->p_filesz);
982
983                 if (mm) {
984                         if (phdr->p_flags & PF_X) {
985                                 if (!mm->start_code) {
986                                         mm->start_code = seg->addr;
987                                         mm->end_code = seg->addr +
988                                                 phdr->p_memsz;
989                                 }
990                         } else if (!mm->start_data) {
991                                 mm->start_data = seg->addr;
992 #ifndef CONFIG_MMU
993                                 mm->end_data = seg->addr + phdr->p_memsz;
994 #endif
995                         }
996
997 #ifdef CONFIG_MMU
998                         if (seg->addr + phdr->p_memsz > mm->end_data)
999                                 mm->end_data = seg->addr + phdr->p_memsz;
1000 #endif
1001                 }
1002
1003                 seg++;
1004         }
1005
1006         return 0;
1007 }
1008 #endif
1009
1010 /*****************************************************************************/
1011 /*
1012  * map a binary by direct mmap() of the individual PT_LOAD segments
1013  */
1014 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1015                                              struct file *file,
1016                                              struct mm_struct *mm)
1017 {
1018         struct elf32_fdpic_loadseg *seg;
1019         struct elf32_phdr *phdr;
1020         unsigned long load_addr, delta_vaddr;
1021         int loop, dvset;
1022
1023         load_addr = params->load_addr;
1024         delta_vaddr = 0;
1025         dvset = 0;
1026
1027         seg = params->loadmap->segs;
1028
1029         /* deal with each load segment separately */
1030         phdr = params->phdrs;
1031         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1032                 unsigned long maddr, disp, excess, excess1;
1033                 int prot = 0, flags;
1034
1035                 if (phdr->p_type != PT_LOAD)
1036                         continue;
1037
1038                 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1039                        (unsigned long) phdr->p_vaddr,
1040                        (unsigned long) phdr->p_offset,
1041                        (unsigned long) phdr->p_filesz,
1042                        (unsigned long) phdr->p_memsz);
1043
1044                 /* determine the mapping parameters */
1045                 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1046                 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1047                 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1048
1049                 flags = MAP_PRIVATE | MAP_DENYWRITE;
1050                 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1051                         flags |= MAP_EXECUTABLE;
1052
1053                 maddr = 0;
1054
1055                 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1056                 case ELF_FDPIC_FLAG_INDEPENDENT:
1057                         /* PT_LOADs are independently locatable */
1058                         break;
1059
1060                 case ELF_FDPIC_FLAG_HONOURVADDR:
1061                         /* the specified virtual address must be honoured */
1062                         maddr = phdr->p_vaddr;
1063                         flags |= MAP_FIXED;
1064                         break;
1065
1066                 case ELF_FDPIC_FLAG_CONSTDISP:
1067                         /* constant displacement
1068                          * - can be mapped anywhere, but must be mapped as a
1069                          *   unit
1070                          */
1071                         if (!dvset) {
1072                                 maddr = load_addr;
1073                                 delta_vaddr = phdr->p_vaddr;
1074                                 dvset = 1;
1075                         } else {
1076                                 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1077                                 flags |= MAP_FIXED;
1078                         }
1079                         break;
1080
1081                 case ELF_FDPIC_FLAG_CONTIGUOUS:
1082                         /* contiguity handled later */
1083                         break;
1084
1085                 default:
1086                         BUG();
1087                 }
1088
1089                 maddr &= PAGE_MASK;
1090
1091                 /* create the mapping */
1092                 disp = phdr->p_vaddr & ~PAGE_MASK;
1093                 down_write(&mm->mmap_sem);
1094                 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1095                                 phdr->p_offset - disp);
1096                 up_write(&mm->mmap_sem);
1097
1098                 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1099                        loop, phdr->p_memsz + disp, prot, flags,
1100                        phdr->p_offset - disp, maddr);
1101
1102                 if (IS_ERR_VALUE(maddr))
1103                         return (int) maddr;
1104
1105                 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1106                     ELF_FDPIC_FLAG_CONTIGUOUS)
1107                         load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1108
1109                 seg->addr = maddr + disp;
1110                 seg->p_vaddr = phdr->p_vaddr;
1111                 seg->p_memsz = phdr->p_memsz;
1112
1113                 /* map the ELF header address if in this segment */
1114                 if (phdr->p_offset == 0)
1115                         params->elfhdr_addr = seg->addr;
1116
1117                 /* clear the bit between beginning of mapping and beginning of
1118                  * PT_LOAD */
1119                 if (prot & PROT_WRITE && disp > 0) {
1120                         kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1121                         clear_user((void __user *) maddr, disp);
1122                         maddr += disp;
1123                 }
1124
1125                 /* clear any space allocated but not loaded
1126                  * - on uClinux we can just clear the lot
1127                  * - on MMU linux we'll get a SIGBUS beyond the last page
1128                  *   extant in the file
1129                  */
1130                 excess = phdr->p_memsz - phdr->p_filesz;
1131                 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1132
1133 #ifdef CONFIG_MMU
1134                 if (excess > excess1) {
1135                         unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1136                         unsigned long xmaddr;
1137
1138                         flags |= MAP_FIXED | MAP_ANONYMOUS;
1139                         down_write(&mm->mmap_sem);
1140                         xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1141                                          prot, flags, 0);
1142                         up_write(&mm->mmap_sem);
1143
1144                         kdebug("mmap[%d] <anon>"
1145                                " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1146                                loop, xaddr, excess - excess1, prot, flags,
1147                                xmaddr);
1148
1149                         if (xmaddr != xaddr)
1150                                 return -ENOMEM;
1151                 }
1152
1153                 if (prot & PROT_WRITE && excess1 > 0) {
1154                         kdebug("clear[%d] ad=%lx sz=%lx",
1155                                loop, maddr + phdr->p_filesz, excess1);
1156                         clear_user((void __user *) maddr + phdr->p_filesz,
1157                                    excess1);
1158                 }
1159
1160 #else
1161                 if (excess > 0) {
1162                         kdebug("clear[%d] ad=%lx sz=%lx",
1163                                loop, maddr + phdr->p_filesz, excess);
1164                         clear_user((void *) maddr + phdr->p_filesz, excess);
1165                 }
1166 #endif
1167
1168                 if (mm) {
1169                         if (phdr->p_flags & PF_X) {
1170                                 if (!mm->start_code) {
1171                                         mm->start_code = maddr;
1172                                         mm->end_code = maddr + phdr->p_memsz;
1173                                 }
1174                         } else if (!mm->start_data) {
1175                                 mm->start_data = maddr;
1176                                 mm->end_data = maddr + phdr->p_memsz;
1177                         }
1178                 }
1179
1180                 seg++;
1181         }
1182
1183         return 0;
1184 }
1185
1186 /*****************************************************************************/
1187 /*
1188  * ELF-FDPIC core dumper
1189  *
1190  * Modelled on fs/exec.c:aout_core_dump()
1191  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1192  *
1193  * Modelled on fs/binfmt_elf.c core dumper
1194  */
1195 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1196
1197 /*
1198  * These are the only things you should do on a core-file: use only these
1199  * functions to write out all the necessary info.
1200  */
1201 static int dump_write(struct file *file, const void *addr, int nr)
1202 {
1203         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1204 }
1205
1206 static int dump_seek(struct file *file, loff_t off)
1207 {
1208         if (file->f_op->llseek) {
1209                 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1210                         return 0;
1211         } else {
1212                 file->f_pos = off;
1213         }
1214         return 1;
1215 }
1216
1217 /*
1218  * Decide whether a segment is worth dumping; default is yes to be
1219  * sure (missing info is worse than too much; etc).
1220  * Personally I'd include everything, and use the coredump limit...
1221  *
1222  * I think we should skip something. But I am not sure how. H.J.
1223  */
1224 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1225 {
1226         int dump_ok;
1227
1228         /* Do not dump I/O mapped devices or special mappings */
1229         if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1230                 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1231                 return 0;
1232         }
1233
1234         /* If we may not read the contents, don't allow us to dump
1235          * them either. "dump_write()" can't handle it anyway.
1236          */
1237         if (!(vma->vm_flags & VM_READ)) {
1238                 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1239                 return 0;
1240         }
1241
1242         /* By default, dump shared memory if mapped from an anonymous file. */
1243         if (vma->vm_flags & VM_SHARED) {
1244                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1245                         dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1246                         kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1247                                vma->vm_flags, dump_ok ? "yes" : "no");
1248                         return dump_ok;
1249                 }
1250
1251                 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1252                 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1253                        vma->vm_flags, dump_ok ? "yes" : "no");
1254                 return dump_ok;
1255         }
1256
1257 #ifdef CONFIG_MMU
1258         /* By default, if it hasn't been written to, don't write it out */
1259         if (!vma->anon_vma) {
1260                 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1261                 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1262                        vma->vm_flags, dump_ok ? "yes" : "no");
1263                 return dump_ok;
1264         }
1265 #endif
1266
1267         dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1268         kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1269                dump_ok ? "yes" : "no");
1270         return dump_ok;
1271 }
1272
1273 /* An ELF note in memory */
1274 struct memelfnote
1275 {
1276         const char *name;
1277         int type;
1278         unsigned int datasz;
1279         void *data;
1280 };
1281
1282 static int notesize(struct memelfnote *en)
1283 {
1284         int sz;
1285
1286         sz = sizeof(struct elf_note);
1287         sz += roundup(strlen(en->name) + 1, 4);
1288         sz += roundup(en->datasz, 4);
1289
1290         return sz;
1291 }
1292
1293 /* #define DEBUG */
1294
1295 #define DUMP_WRITE(addr, nr)    \
1296         do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1297 #define DUMP_SEEK(off)  \
1298         do { if (!dump_seek(file, (off))) return 0; } while(0)
1299
1300 static int writenote(struct memelfnote *men, struct file *file)
1301 {
1302         struct elf_note en;
1303
1304         en.n_namesz = strlen(men->name) + 1;
1305         en.n_descsz = men->datasz;
1306         en.n_type = men->type;
1307
1308         DUMP_WRITE(&en, sizeof(en));
1309         DUMP_WRITE(men->name, en.n_namesz);
1310         /* XXX - cast from long long to long to avoid need for libgcc.a */
1311         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1312         DUMP_WRITE(men->data, men->datasz);
1313         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1314
1315         return 1;
1316 }
1317 #undef DUMP_WRITE
1318 #undef DUMP_SEEK
1319
1320 #define DUMP_WRITE(addr, nr)    \
1321         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1322                 goto end_coredump;
1323 #define DUMP_SEEK(off)  \
1324         if (!dump_seek(file, (off))) \
1325                 goto end_coredump;
1326
1327 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1328 {
1329         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1330         elf->e_ident[EI_CLASS] = ELF_CLASS;
1331         elf->e_ident[EI_DATA] = ELF_DATA;
1332         elf->e_ident[EI_VERSION] = EV_CURRENT;
1333         elf->e_ident[EI_OSABI] = ELF_OSABI;
1334         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1335
1336         elf->e_type = ET_CORE;
1337         elf->e_machine = ELF_ARCH;
1338         elf->e_version = EV_CURRENT;
1339         elf->e_entry = 0;
1340         elf->e_phoff = sizeof(struct elfhdr);
1341         elf->e_shoff = 0;
1342         elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1343         elf->e_ehsize = sizeof(struct elfhdr);
1344         elf->e_phentsize = sizeof(struct elf_phdr);
1345         elf->e_phnum = segs;
1346         elf->e_shentsize = 0;
1347         elf->e_shnum = 0;
1348         elf->e_shstrndx = 0;
1349         return;
1350 }
1351
1352 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1353 {
1354         phdr->p_type = PT_NOTE;
1355         phdr->p_offset = offset;
1356         phdr->p_vaddr = 0;
1357         phdr->p_paddr = 0;
1358         phdr->p_filesz = sz;
1359         phdr->p_memsz = 0;
1360         phdr->p_flags = 0;
1361         phdr->p_align = 0;
1362         return;
1363 }
1364
1365 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1366                 unsigned int sz, void *data)
1367 {
1368         note->name = name;
1369         note->type = type;
1370         note->datasz = sz;
1371         note->data = data;
1372         return;
1373 }
1374
1375 /*
1376  * fill up all the fields in prstatus from the given task struct, except
1377  * registers which need to be filled up seperately.
1378  */
1379 static void fill_prstatus(struct elf_prstatus *prstatus,
1380                           struct task_struct *p, long signr)
1381 {
1382         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1383         prstatus->pr_sigpend = p->pending.signal.sig[0];
1384         prstatus->pr_sighold = p->blocked.sig[0];
1385         prstatus->pr_pid = task_pid_vnr(p);
1386         prstatus->pr_ppid = task_pid_vnr(p->parent);
1387         prstatus->pr_pgrp = task_pgrp_vnr(p);
1388         prstatus->pr_sid = task_session_vnr(p);
1389         if (thread_group_leader(p)) {
1390                 /*
1391                  * This is the record for the group leader.  Add in the
1392                  * cumulative times of previous dead threads.  This total
1393                  * won't include the time of each live thread whose state
1394                  * is included in the core dump.  The final total reported
1395                  * to our parent process when it calls wait4 will include
1396                  * those sums as well as the little bit more time it takes
1397                  * this and each other thread to finish dying after the
1398                  * core dump synchronization phase.
1399                  */
1400                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1401                                    &prstatus->pr_utime);
1402                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1403                                    &prstatus->pr_stime);
1404         } else {
1405                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1406                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1407         }
1408         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1409         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1410
1411         prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1412         prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1413 }
1414
1415 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1416                        struct mm_struct *mm)
1417 {
1418         unsigned int i, len;
1419
1420         /* first copy the parameters from user space */
1421         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1422
1423         len = mm->arg_end - mm->arg_start;
1424         if (len >= ELF_PRARGSZ)
1425                 len = ELF_PRARGSZ - 1;
1426         if (copy_from_user(&psinfo->pr_psargs,
1427                            (const char __user *) mm->arg_start, len))
1428                 return -EFAULT;
1429         for (i = 0; i < len; i++)
1430                 if (psinfo->pr_psargs[i] == 0)
1431                         psinfo->pr_psargs[i] = ' ';
1432         psinfo->pr_psargs[len] = 0;
1433
1434         psinfo->pr_pid = task_pid_vnr(p);
1435         psinfo->pr_ppid = task_pid_vnr(p->parent);
1436         psinfo->pr_pgrp = task_pgrp_vnr(p);
1437         psinfo->pr_sid = task_session_vnr(p);
1438
1439         i = p->state ? ffz(~p->state) + 1 : 0;
1440         psinfo->pr_state = i;
1441         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1442         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1443         psinfo->pr_nice = task_nice(p);
1444         psinfo->pr_flag = p->flags;
1445         SET_UID(psinfo->pr_uid, p->uid);
1446         SET_GID(psinfo->pr_gid, p->gid);
1447         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1448
1449         return 0;
1450 }
1451
1452 /* Here is the structure in which status of each thread is captured. */
1453 struct elf_thread_status
1454 {
1455         struct list_head list;
1456         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1457         elf_fpregset_t fpu;             /* NT_PRFPREG */
1458         struct task_struct *thread;
1459 #ifdef ELF_CORE_COPY_XFPREGS
1460         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1461 #endif
1462         struct memelfnote notes[3];
1463         int num_notes;
1464 };
1465
1466 /*
1467  * In order to add the specific thread information for the elf file format,
1468  * we need to keep a linked list of every thread's pr_status and then create
1469  * a single section for them in the final core file.
1470  */
1471 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1472 {
1473         struct task_struct *p = t->thread;
1474         int sz = 0;
1475
1476         t->num_notes = 0;
1477
1478         fill_prstatus(&t->prstatus, p, signr);
1479         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1480
1481         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1482                   &t->prstatus);
1483         t->num_notes++;
1484         sz += notesize(&t->notes[0]);
1485
1486         t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1487         if (t->prstatus.pr_fpvalid) {
1488                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1489                           &t->fpu);
1490                 t->num_notes++;
1491                 sz += notesize(&t->notes[1]);
1492         }
1493
1494 #ifdef ELF_CORE_COPY_XFPREGS
1495         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1496                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1497                           sizeof(t->xfpu), &t->xfpu);
1498                 t->num_notes++;
1499                 sz += notesize(&t->notes[2]);
1500         }
1501 #endif
1502         return sz;
1503 }
1504
1505 /*
1506  * dump the segments for an MMU process
1507  */
1508 #ifdef CONFIG_MMU
1509 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1510                            unsigned long *limit, unsigned long mm_flags)
1511 {
1512         struct vm_area_struct *vma;
1513
1514         for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1515                 unsigned long addr;
1516
1517                 if (!maydump(vma, mm_flags))
1518                         continue;
1519
1520                 for (addr = vma->vm_start;
1521                      addr < vma->vm_end;
1522                      addr += PAGE_SIZE
1523                      ) {
1524                         struct vm_area_struct *vma;
1525                         struct page *page;
1526
1527                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1528                                            &page, &vma) <= 0) {
1529                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1530                         }
1531                         else if (page == ZERO_PAGE(0)) {
1532                                 page_cache_release(page);
1533                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1534                         }
1535                         else {
1536                                 void *kaddr;
1537
1538                                 flush_cache_page(vma, addr, page_to_pfn(page));
1539                                 kaddr = kmap(page);
1540                                 if ((*size += PAGE_SIZE) > *limit ||
1541                                     !dump_write(file, kaddr, PAGE_SIZE)
1542                                     ) {
1543                                         kunmap(page);
1544                                         page_cache_release(page);
1545                                         return -EIO;
1546                                 }
1547                                 kunmap(page);
1548                                 page_cache_release(page);
1549                         }
1550                 }
1551         }
1552
1553         return 0;
1554
1555 end_coredump:
1556         return -EFBIG;
1557 }
1558 #endif
1559
1560 /*
1561  * dump the segments for a NOMMU process
1562  */
1563 #ifndef CONFIG_MMU
1564 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1565                            unsigned long *limit, unsigned long mm_flags)
1566 {
1567         struct vm_list_struct *vml;
1568
1569         for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1570         struct vm_area_struct *vma = vml->vma;
1571
1572                 if (!maydump(vma, mm_flags))
1573                         continue;
1574
1575                 if ((*size += PAGE_SIZE) > *limit)
1576                         return -EFBIG;
1577
1578                 if (!dump_write(file, (void *) vma->vm_start,
1579                                 vma->vm_end - vma->vm_start))
1580                         return -EIO;
1581         }
1582
1583         return 0;
1584 }
1585 #endif
1586
1587 /*
1588  * Actual dumper
1589  *
1590  * This is a two-pass process; first we find the offsets of the bits,
1591  * and then they are actually written out.  If we run out of core limit
1592  * we just truncate.
1593  */
1594 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1595                                struct file *file, unsigned long limit)
1596 {
1597 #define NUM_NOTES       6
1598         int has_dumped = 0;
1599         mm_segment_t fs;
1600         int segs;
1601         size_t size = 0;
1602         int i;
1603         struct vm_area_struct *vma;
1604         struct elfhdr *elf = NULL;
1605         loff_t offset = 0, dataoff;
1606         int numnote;
1607         struct memelfnote *notes = NULL;
1608         struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1609         struct elf_prpsinfo *psinfo = NULL;     /* NT_PRPSINFO */
1610         LIST_HEAD(thread_list);
1611         struct list_head *t;
1612         elf_fpregset_t *fpu = NULL;
1613 #ifdef ELF_CORE_COPY_XFPREGS
1614         elf_fpxregset_t *xfpu = NULL;
1615 #endif
1616         int thread_status_size = 0;
1617 #ifndef CONFIG_MMU
1618         struct vm_list_struct *vml;
1619 #endif
1620         elf_addr_t *auxv;
1621         unsigned long mm_flags;
1622
1623         /*
1624          * We no longer stop all VM operations.
1625          *
1626          * This is because those proceses that could possibly change map_count
1627          * or the mmap / vma pages are now blocked in do_exit on current
1628          * finishing this core dump.
1629          *
1630          * Only ptrace can touch these memory addresses, but it doesn't change
1631          * the map_count or the pages allocated. So no possibility of crashing
1632          * exists while dumping the mm->vm_next areas to the core file.
1633          */
1634
1635         /* alloc memory for large data structures: too large to be on stack */
1636         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1637         if (!elf)
1638                 goto cleanup;
1639         prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1640         if (!prstatus)
1641                 goto cleanup;
1642         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1643         if (!psinfo)
1644                 goto cleanup;
1645         notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1646         if (!notes)
1647                 goto cleanup;
1648         fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1649         if (!fpu)
1650                 goto cleanup;
1651 #ifdef ELF_CORE_COPY_XFPREGS
1652         xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1653         if (!xfpu)
1654                 goto cleanup;
1655 #endif
1656
1657         if (signr) {
1658                 struct core_thread *ct;
1659                 struct elf_thread_status *tmp;
1660
1661                 for (ct = current->mm->core_state->dumper.next;
1662                                                 ct; ct = ct->next) {
1663                         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1664                         if (!tmp)
1665                                 goto cleanup;
1666
1667                         tmp->thread = ct->task;
1668                         list_add(&tmp->list, &thread_list);
1669                 }
1670
1671                 list_for_each(t, &thread_list) {
1672                         struct elf_thread_status *tmp;
1673                         int sz;
1674
1675                         tmp = list_entry(t, struct elf_thread_status, list);
1676                         sz = elf_dump_thread_status(signr, tmp);
1677                         thread_status_size += sz;
1678                 }
1679         }
1680
1681         /* now collect the dump for the current */
1682         fill_prstatus(prstatus, current, signr);
1683         elf_core_copy_regs(&prstatus->pr_reg, regs);
1684
1685 #ifdef CONFIG_MMU
1686         segs = current->mm->map_count;
1687 #else
1688         segs = 0;
1689         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1690             segs++;
1691 #endif
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(&notes[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 (
1767 #ifdef CONFIG_MMU
1768                 vma = current->mm->mmap; vma; vma = vma->vm_next
1769 #else
1770                         vml = current->mm->context.vmlist; vml; vml = vml->next
1771 #endif
1772              ) {
1773                 struct elf_phdr phdr;
1774                 size_t sz;
1775
1776 #ifndef CONFIG_MMU
1777                 vma = vml->vma;
1778 #endif
1779
1780                 sz = vma->vm_end - vma->vm_start;
1781
1782                 phdr.p_type = PT_LOAD;
1783                 phdr.p_offset = offset;
1784                 phdr.p_vaddr = vma->vm_start;
1785                 phdr.p_paddr = 0;
1786                 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1787                 phdr.p_memsz = sz;
1788                 offset += phdr.p_filesz;
1789                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1790                 if (vma->vm_flags & VM_WRITE)
1791                         phdr.p_flags |= PF_W;
1792                 if (vma->vm_flags & VM_EXEC)
1793                         phdr.p_flags |= PF_X;
1794                 phdr.p_align = ELF_EXEC_PAGESIZE;
1795
1796                 DUMP_WRITE(&phdr, sizeof(phdr));
1797         }
1798
1799 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1800         ELF_CORE_WRITE_EXTRA_PHDRS;
1801 #endif
1802
1803         /* write out the notes section */
1804         for (i = 0; i < numnote; i++)
1805                 if (!writenote(notes + i, file))
1806                         goto end_coredump;
1807
1808         /* write out the thread status notes section */
1809         list_for_each(t, &thread_list) {
1810                 struct elf_thread_status *tmp =
1811                                 list_entry(t, struct elf_thread_status, list);
1812
1813                 for (i = 0; i < tmp->num_notes; i++)
1814                         if (!writenote(&tmp->notes[i], file))
1815                                 goto end_coredump;
1816         }
1817
1818         DUMP_SEEK(dataoff);
1819
1820         if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1821                 goto end_coredump;
1822
1823 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1824         ELF_CORE_WRITE_EXTRA_DATA;
1825 #endif
1826
1827         if (file->f_pos != offset) {
1828                 /* Sanity check */
1829                 printk(KERN_WARNING
1830                        "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1831                        file->f_pos, offset);
1832         }
1833
1834 end_coredump:
1835         set_fs(fs);
1836
1837 cleanup:
1838         while (!list_empty(&thread_list)) {
1839                 struct list_head *tmp = thread_list.next;
1840                 list_del(tmp);
1841                 kfree(list_entry(tmp, struct elf_thread_status, list));
1842         }
1843
1844         kfree(elf);
1845         kfree(prstatus);
1846         kfree(psinfo);
1847         kfree(notes);
1848         kfree(fpu);
1849 #ifdef ELF_CORE_COPY_XFPREGS
1850         kfree(xfpu);
1851 #endif
1852         return has_dumped;
1853 #undef NUM_NOTES
1854 }
1855
1856 #endif          /* USE_ELF_CORE_DUMP */